Discovery and refinement of muscle weight QTLs in B6 × D2 advanced intercross mice

Abstract

The genes underlying variation in skeletal muscle mass are poorly understood. Although many quantitative trait loci (QTLs) have been mapped in crosses of mouse strains, the limited resolution inherent in these conventional studies has made it difficult to reliably pinpoint the causal genetic variants. The accumulated recombination events in an advanced intercross line (AIL), in which mice from two inbred strains are mated at random for several generations, can improve mapping resolution. We demonstrate these advancements in mapping QTLs for hindlimb muscle weights in an AIL (n = 832) of the C57BL/6J (B6) and DBA/2J (D2) strains, generations F8–F13. We mapped muscle weight QTLs using the high-density MegaMUGA SNP panel. The QTLs highlight the shared genetic architecture of four hindlimb muscles and suggest that the genetic contributions to muscle variation are substantially different in males and females, at least in the B6D2 lineage. Out of the 15 muscle weight QTLs identified in the AIL, nine overlapped the genomic regions discovered in an earlier B6D2 F2 intercross. Mapping resolution, however, was substantially improved in our study to a median QTL interval of 12.5 Mb. Subsequent sequence analysis of the QTL regions revealed 20 genes with nonsense or potentially damaging missense mutations. Further refinement of the muscle weight QTLs using additional functional information, such as gene expression differences between alleles, will be important for discerning the causal genes.

muscle strength varies greatly among individuals of similar age (55). Resistance training and immobilization studies have demonstrated a robust positive association between strength and muscle mass (41, 57). Fiber number and their size are the main factors that contribute to muscle mass, and these factors are highly variable; for example, a more than twofold difference in fiber content has been reported among young adults (29, 40). The number of fibers in mammals is set soon after birth (48), whereas fiber size changes substantially during growth (20) and is partly determined by physical activity (40). Variation in muscle mass might affect susceptibility to impairment of vital functions conferred by muscle tissue, such as locomotion, respiration, thermoregulation, maintenance of glucose homeostasis, and protection of bones and vital organs when muscle is lost due to aging or disease. We know that approximately half of the variation in muscle strength can be explained by genetic factors (55). However, genes underlying this variation remain poorly understood.

In this paper, we study muscle mass in crosses of inbred mouse strains to better characterize the genetic factors underlying variation in skeletal muscle. Muscle mass of inbred mouse strains can differ as much as 10-fold, partly as a consequence of variation in fiber size and number (26, 36). Several studies of mouse models have been conducted over the past decade, with the aim of mapping genes that underlie observed variation in muscle mass. Some studies have examined crosses of phenotypically contrasting strains (44), strains selected for divergent growth, such as DUI and DBA/2 (7) and LG/J and SM/J (33), and strains selected for growth (24). Other studies have focused on the classical laboratory strains C57BL/6J and DBA/2J (30–32). These efforts have uncovered many quantitative trait loci (QTL) for muscle mass. However, because these studies were all carried out in an F2 intercross (7, 30–32, 44) or in early generations of an advanced intercross (24), poor mapping resolution, owing to a limited amount of recombination, has hindered identification of the causal genes (47).

Accumulation of recombination events in an advanced intercross line (AIL) population permits greater precision in QTL mapping, while retaining alleles at high frequencies in the population (17, 47). For example, one study mapped muscle weight QTLs to chromosomal regions at a median width of 3.7 Mb using the F34 generation of an AIL derived from the LG/J and SM/J strains (33). By comparison, QTLs in the F2 population span 30–40 Mb (56). AIL populations present additional challenges for QTL mapping because of the varying levels of genetic sharing among individuals (13). Fortunately, in the past few years robust statistical methods have been developed to tackle this issue (12, 22, 23, 37, 38, 64–66).

A recently developed AIL between the C57BL/6J and DBA/2J strains (51) presents an attractive opportunity to follow up and refine the genetic factors affecting muscle mass in mice of a different genetic background than the LG/J × SM/J AIL. In this study, we mapped loci underlying muscle weight variation in an AIL that was maintained for 13 generations, beginning with crosses of the B6 and D2 strains. In addition to mapping QTLs that overlap with loci identified in existing QTL mapping studies on muscle weight, we identified new muscle weight QTLs. We narrowed the chromosomal position of the QTL within each locus to a median interval of 12.5 Mb.

MATERIALS AND METHODS

All experiments on B6D2 AIL mice were carried out in accordance with National Institutes of Health (NIH) Guidelines for Care and Use of Laboratory Animals. Studies were approved by University of Chicago's Institutional Animal Care and Use Committee (IACUC) for the F8 mouse cohort and by the IACUC of the Pennsylvania State University for the F9–F13 mice.

We have made available the full code and data for reproducing the steps of our analyses at http://github.com/pcarbo/b6d2muscle.

Animals

All mice in the AIL are descendants of the C57BL/6J (B6) and DBA/2J (D2) strains. The AIL was initiated by crossing B6 females and D2 males, then intercrossing the F1 progeny. This AIL was previously used to investigate behavioral traits and has been described previously (50, 51).

A pseudorandom mating scheme was used to generate the F8 subpopulation and separately the F9–F13 crosses. In each generation, we selected matings that produce the smallest average inbreeding coefficients in the offspring. (The R code to generate these matings is available at http://github.com/pcarbo/breedail.) The F8 mice were bred in Chicago and housed at 12:12 h light-dark cycle, with two to five same-sex animals per cage and with ad libitum access to chow (Harlan 2918) and water. The F9–F13 mice were bred at Penn State and housed at 12:12 h light-dark cycle, with one to four same-sex animals per cage and with ad libitum access to chow (Purina LabDiet 5001) and water. Note that the F8 mice phenotyped in this study are not parents of the F9 mice; additional litters from the F8 generation served as parents of the F9 generation.

Equal numbers of males and females were chosen from the F8 subpopulation; in our analyses we included 212 males and 213 females, after discarding problematic samples, as explained below. The F8 mice were killed at 14 wk of age, and one hindlimb was removed and stored at −80°C for measuring muscle weights.

The F9–F13 mice were also divided to maintain a balance of males and females; one male and one female were drawn at random from each litter to produce this sample. We drew 77, 91, 73, 65, and 101 mice from generations F9 through F13, respectively. In total, we included in our analyses 205 males and 202 females from the F9–F13 generations, after discarding problematic samples. Animals were killed at the age of 29 wk, after which one hindlimb was removed. Hindlimbs were phenotyped in the same way as hindlimbs from the F8 mice (see below for details).

Unlike some other AIL studies, we did not record the relationships among the mice in a pedigree. Instead, we used genetic marker data to estimate familial relationships; marker-based estimates of relatedness are often accurate provided genetic variation is ascertained at an adequate density (60).

Phenotypes

After hindlimbs from the F8 and F9–F13 animals were defrosted, tibialis anterior (TA), extensor digitorum longus (EDL), gastrocnemius (gastroc), and soleus muscles were dissected under a dissection microscope and weighed to a precision of 0.1 mg on a balance (Pioneer, Ohaus).

Systematic differences in collection of phenotype measurements among the different generations of the F9–F13 subpopulation are not to be expected. However, discrepancies are expected between the F8 and F9–F13 mice as they were maintained at different facilities, fed different diets, and their muscle weights were measured at different ages. We accounted for these systematic differences between the F8 and F9–F13 cohorts in our analyses, as we explain below.

Genotypes

F8 mouse genotypes.

The F8 AIL samples were genotyped using the Illumina Mouse Medium Density Linkage Panel (Illumina, San Diego, CA). We called genotypes at 900 single nucleotide polymorphisms (SNPs) on the X and autosomal chromosomes after discarding suspect SNPs, as described below. Although we did not map muscle weight loci on the X chromosome (for reasons given below), the X chromosome is still useful for our analysis, including assessing sample quality. SNPs and genotyping procedures for the F8 mice are also described in previous publications (50, 51). The median distance between pairs of adjacent SNPs is ∼2 Mb, and the maximum inter-SNP interval on any one chromosome is 11.3 Mb, except for chromosomes 8 and 9, which have maximum inter-SNP intervals of 19.7 and 18.6, respectively. Of the 900 SNPs, 95% correspond to SNPs that are polymorphic between the B6 and D2 strains according to data retrieved from the Center for Genome Dynamics Mouse SNP Database (http://cgd.jax.org/cgdsnpdb/build37/data/imputedsnps).

Removal of low-quality SNPs and genotypes in F8 mice.

Two markers in the F8 SNP panel were removed (rs3695988 on chromosome 1 and rs3699086 on chromosome 7) because their genotypes did not fit the linkage disequilibrium pattern of nearby markers, or because they had unusually small or large allele frequencies. Two more markers were discarded (rs13481297 on chromosome 12 and rs13482729 on chromosome 15) because their genotype frequencies deviated substantially from what we would expect under Hardy-Weinberg equilibrium.

Genotypes called with lower quality scores were discarded and treated as missing. Overall, 107 individual genotypes (0.03%) in the 425 F8 mice were treated as missing.

F9–F13 mouse genotypes.

In the F9–F13 cohort, SNPs were genotyped using the Mega Mouse Universal Genotyping Array (MegaMUGA) (http://csbio.unc.edu/CCstatus), which provided genotypes at 20,691 SNPs on the X and autosomal chromosomes, after taking quality control steps to remove poor-quality SNPs. The MegaMUGA genotyping platform was developed for the Collaborative Cross and is based on the Illumina Infinium genotyping assay (61). The median interval between adjacent pairs of SNPs is 42.6 kb. The maximum inter-SNP interval on any individual chromosome ranges between 3.4 and 8.8 Mb, except for chromosomes 8, 9, and 10, where the maximum inter-SNP intervals are 13.8, 16.4, and 15.8, respectively. A total of ∼35,000 individual genotypes (0.4%) in these SNPs are missing, and at most 2,020 genotypes (10%) are missing in any one mouse.

Removal of low-quality SNPs and genotypes in F9–F13 mice.

After inspecting the genotypes, we removed several hundred additional SNPs because 1) their frequencies deviated substantially from Hardy-Weinberg equilibrium; 2) they had unusually small or unusually large allele frequencies; or 3) because the SNPs were not correlated with nearby SNPs. About 30 SNPs were inversely correlated with adjacent SNPs, suggesting that allele labels were reversed. In these cases, we fixed the allele labels.

Genotype imputation.

To analyze the combined data, we estimated the genotypes of SNPs that were not directly genotyped in both F8 and F9–F13 cohorts. Among the SNPs on autosomal chromosomes genotyped in the two cohorts, 36 are common to both, so the combined data set has 20,403 + 857 − 36 = 21,224 SNPs; we added 857 − 36 = 821 SNPs with missing genotypes to the F9–F13 SNP panel, and we added 20,403 − 36 = 20,367 SNPs with missing genotypes to the F8 SNP panel.

Imputing missing genotypes in AILs is relatively straightforward, as all alleles (and haplotypes) are common in the population and originate from one of the two inbred progenitors (in this case, B6 or D2). For all unavailable genotypes, we estimated their probabilities based on correlation patterns with available genotypes, using recombination models suitable for autosomal chromosomes in advanced intercrosses (17). We estimated genotype probabilities separately in each filial generation, since each generation exhibits a different pattern of recombination. Estimation of the missing genotypes requires genetic distance estimates at all markers. We used the genetic map constructed from a large, heterogeneous reference panel (16). These data were retrieved from the Jackson Laboratories' Mouse Map Converter website (http://cgd.jax.org/mousemapconverter).

Our ability to accurately impute genotypes in an AIL ultimately depends on the availability of genotypes at nearby, correlated SNPs. When no nearby SNPs are genotyped, the genotype estimation will be poor quality; that is, the genotype will be estimated with high uncertainty. In the F9–F13 mice, since we have genotype data at a relatively high density, only 67 (<0.001%) of genotypes have a maximum genotype probability <50%. In the F8 mice, even though we have a much sparser set of SNPs genotyped using the Illumina array, this SNP panel is for the most part sufficient to accurately recover genotypes at the other SNPs; only 0.08% of the genotypes have a maximum genotype probability <50%. Since we take into account uncertainty in the genotypes when mapping QTLs, any genotypes imputed with low confidence should contribute little to support for association at a given SNP.

Removal of low-quality samples.

We took several steps to identify and remove problematic samples. We discarded three F8 mice from further analysis because of inconsistencies in their genotypes: one male had a large number of heterozygous genotypes on the X chromosome, and two other mice had the same genotypes at all SNPs, but different phenotype measurements.

In the F9–F13 population, two mice had identical genotypes, but different phenotype measurements. Three males had a large number of heterozygous genotypes on the X chromosome, suggesting that sex was recorded incorrectly or the samples were mixed up. Two pairs expected to be siblings based on our records had unusually low kinship coefficients, suggesting that they were probably not siblings and indicating possible sample mix-up. We excluded all these samples from our analyses. Two mice were discarded because a large proportion of their genotypes were missing. In a block of 14 samples from the F11 cross, pairwise relatedness estimates did not align with recorded sibling pairs, and males and females appeared to be switched based on inspecting X chromosome genotypes, indicating likely sample mix-ups. All 14 of these samples were excluded from our analyses. Finally, several males had a small number of heterozygous genotypes on the X chromosome. These were probably genotyping errors or misrecorded genotypes, so these SNPs were treated as missing. These mice were also excluded from our analyses.

We used coat color recorded in the F9–13 mice to flag other potential sample mix-ups. The B6 strain has a black coat, and the D2 strain has a dilute brown coat (d allele). We obtained very strong evidence for SNPs correlated with coat color on chromosome 9, 74.4–75.7 Mb. This region overlaps Myo5a, and variants of this gene underlie the dilute (d) phenotype (21). The d allele is recessive, so mice with B6/B6 and heterozygous genotypes are black, and mice with D2/D2 genotypes are dilute brown. Two animals were recorded as black but had D2/D2 genotypes, and one mouse was recorded as brown but had a B6/D2 genotype. These inconsistencies indicated possible sample mix-ups, and these samples were discarded as well.

SNP genotype and allele frequencies.

In the F8 sample, the frequency of the D2 allele at the 857 SNPs on autosomal chromosomes ranged between 0.24 and 0.76. The average frequency was 0.504, indicating little unintended selection or genetic drift (see Fig. 1, A and B). As expected, allele frequencies at SNPs in the F9–F13 generations exhibited a slightly greater range (0.21–0.80). The average frequency of 0.504 again suggests no selection, intended or unintended, toward any one of the strains. Figure 1, C and D, shows the genotype frequencies for all SNPs in the F8 and F9–F13 populations; for the most part, the observed proportions of homozygous and heterozygous genotypes do not deviate substantially from expectations for a randomly mating population (Hardy-Weinberg equilibrium).

Fig. 1.

Allele and genotype frequencies in F8 and F9–F13 advanced intercross line (AIL) samples. A: distribution of DBA/2J (D2) allele frequencies in F8 mouse cohort across the 857 single nucleotide polymorphisms (SNPs) genotyped on autosomal chromosomes. B: distribution of D2 allele frequencies in F9–F13 cohort across the 20,403 SNPs on autosomal chromosomes. C: “ternary plot” of genotype frequencies for all 857 SNPs genotyped in F8 mice on autosomal chromosomes. D: ternary plot of genotype frequencies for all SNPs genotyped in F9–F13 mice on autosomal chromosomes. In the ternary plots, each point corresponds to a SNP. The triangle axes labeled “B6/B6”, “B6/D2,” and “D2/D2” correspond to SNPs in which all genotypes are homozygous B6, heterozygous B6/D2 and homozygous D2/D2, respectively.

All SNP information and genomic positions are based on NCBI release 37 of the Mouse Genome Assembly and build 128 of the dbSNP database.

QTL Mapping

The amount of the genome that is shared can vary considerably among mice in advanced generations of an intercross line. These varied levels of relatedness can confound tests for phenotype-genotype association (5, 13, 23). There is potential to map QTLs more reliably if we are able to correct for genetic sharing in tests for association (5, 52). Several popular approaches to modeling relatedness in association tests are based on linear mixed models (LMMs) (22, 23, 37, 38, 64–66). We used the QTLRel mixed model framework, which was originally developed for mapping QTLs in AILs (13).

LMM.

Like most approaches to QTL mapping, QTLRel models the phenotype of individual i, which we denote by y_i, as a linear combination of the genotype at a given SNP, additional covariates (e.g., cohort, sex), represented as vector z_i, and a residual (“noise”) term, ϵ_i. Here we use i to index individual mice, and j to index SNPs. Following standard practice in QTL analysis of mouse intercrosses, we included two terms in the linear regression for each SNP j: the “additive genotype” g_ij (allele count) and the “dominance genotype” d_ij (0 = heterozygous, 1 = homozygous). What distinguishes the mixed model from a standard linear regression is the inclusion of an additional “polygenic” term u_i with a correlation structure that is intended to capture genome-wide sharing for all pairs of individuals:

$$y_i=\mu +z_i^T{\beta }_z+g_{ij}{\beta }_{g,j}+d_{ij}{\beta }_{d,j}+u_i+{ϵ}_i$$

This is the LMM. The expression for the covariance matrix of the polygenic effect is given in Ref. 13.

Covariates.

In all our analyses, we included sex as a covariate. To analyze the combined data from the F8 and F9–F13 cohorts, we included a binary covariate that indicates whether the sample originates from the F8 or F9–F13 cohort. The mice in these two cohorts were phenotyped at considerably different ages (Table 2) and were raised in different environments (University of Chicago, Penn State) and fed different diets, so this covariate serves as a proxy for these factors; see Table 2 and Fig. 2 for phenotype distributions in the F8 and F9–F13 mice. Sex and cohort account for some differences in body weight among the mice. We could further correct for body weight by including it in the linear model. However, doing so could also remove informative variation in muscle weight, since muscle weight accounts for a substantial proportion of total body weight. For this reason, we chose not to include body weight as a covariate. We did not include age as a covariate because it had a negligible effect on muscle weight within each of the cohorts.

Table 2.

Muscle weight in B6D2 AIL mice

Cohort	Sex	Age, day	TA, mg	EDL, mg	Gastroc, mg	Soleus, mg
F8	F (n = 209)	96 ± 7	37.0 ± 3.2	7.2 ± 0.7	92.6 ± 7.4	5.7 ± 0.8
	M (n = 212)	97 ± 8	48.3 ± 4.7	9.5 ± 1.1	128.7 ± 13.1	7.3 ± 1.0
F9–F13	F (n = 181)	204 ± 7	41.2 ± 4.2	8.0 ± 0.8	97.6 ± 9.1	6.4 ± 0.9
	M (n = 187)	204 ± 7	46.7 ± 4.4	9.0 ± 0.9	116.5 ± 10.2	7.4 ± 1.0

Sample sizes (n) correspond those used to map quantitative trait loci (QTLs) for muscle weights. Table entries in phenotype columns give the sample means ± SD. See Fig. 3 to compare the muscle weight distributions in these samples. AIL, advanced intercross line; TA, tibialis anterior; EDL, extensor digitorum longus; gastroc, gastrocnemius; F, female; M, male.

Fig. 2.

Distribution of muscle weights in males and females. Distribution of muscle weights measured in F8 females (top row), F9–F13 females (2nd row from top), F8 males (3rd row from top), and F9–F13 males (bottom row). The vertical axis in each plot gives the number of samples.

Model with sex-specific genetic contributions.

Since muscle weights differ considerably in males and females, we hypothesized that some of the genetic contributions to muscle weight also differ between males and females. Therefore, we also tested for “sex-specific” QTLs with a regression model that includes additional terms to allow for different SNP effects in males and females:

$$y_i=\mu +z_i^T{\beta }_z+\left(g_{ij}{\beta }_{g,j}+d_{ij}{\beta }_{d,j}\right)z_{\text{sex},i}+\left(g_{ij}{\beta }_{g,j}^{\ast }+d_{ij}{\beta }_{d,j}^{\ast }\right)\left(1-z_{\text{sex},i}\right)+u_i+{ϵ}_i$$

We encode sex (z_sex,i) as 0 for females and 1 for males, so that β_g,j and β_d,j are the additive and dominance effects of SNP j in males only, and ${\beta }_{\mathrm{g,j}}^{*}$ and ${\beta }_{\mathrm{d,j}}^{*}$ are the effects in females only. Although this model splits the data by sex, one advantage of integrating the male and female-specific SNP effects into a single model is that we can assess significance of sex-specific QTLs by comparing the logarithm of the likelihood ratio (the LOD score) from models with and without sex-specific effects (see below).

Marker-based estimates of genetic sharing.

We used the genotypes of the genetic markers to define the n × n covariance matrix of the polygenic effect u_i, where n is the number of samples. While marker-based and pedigree-based estimates of genetic sharing yield QTL mapping results that broadly agree in some settings (14), there are at least two benefits to using SNPs to estimate relatedness: one, genetic markers often yield more precise estimates of genetic sharing; two, the computation is much less intensive, as algorithms for computing identity coefficients from a pedigree scale poorly to studies with large numbers of individuals. The general expression for the covariance matrix for an inbred population is derived in Abney et al. (1). Since the additive component typically makes a much larger contribution to the covariance matrix than the remaining components in an outbred population, we assumed that the contribution of the other components is small, and we retained only the additive component. (This is a common, if implicit, assumption made by many other QTL mapping approaches based on linear mixed models.) Therefore, the entries of the n × n covariance matrix are σ_ij = 2Φ_ijσ_a² = R_ijσ_a², where σ_a² is a model parameter that will be estimated from the data, and Φ_ij is the kinship coefficient for pair of individuals i and j; that is, the probability that a pair of alleles chosen randomly from individuals i and j at the same autosomal locus are identical-by-descent (IBD) (27). We obtained the final estimate of R_ij = 2Φ_ij by taking the average allele sharing over all SNPs on autosomal chromosomes. We excluded from this average SNPs on the same chromosome as the candidate SNP to avoid the loss of power due to the candidate marker being included in multiple components of the mixed model; this issue is often called “proximal contamination” (38, 64). Therefore, in the QTL mapping we used a slightly different polygenic covariance matrix for each chromosome.

For a given SNP, the estimate of R_ij = 2Φ_ij is simply the number of alleles that share the same state: 0 if the genotypes of individuals i and j are homozygous and different; 2 if both genotypes are homozygous and the same; and 1 in all other cases. (In an AIL, this is equivalent to the number of alleles that are IBD since all alleles originate from the two inbred founders.) To account for uncertainty in the genotype estimates whenever the genotypes are missing, we compute the expected number of shared alleles:

$$E\left[2{\Phi }_{ij}\right]=2p_{AA}^{\left(i\right)}{p}_{AA}^{\left(j\right)}+2p_{BB}^{\left(i\right)}{p}_{BB}^{\left(j\right)}+p_{AB}^{\left(i\right)}+p_{AB}^{\left(j\right)}-p_{AB}^{\left(i\right)}{p}_{AB}^{\left(j\right)},$$

where p_AA⁽ⁱ⁾, p_AB⁽ⁱ⁾, and p_BB⁽ⁱ⁾ are the probabilities that individual i has genotype AA, AB, and BB, respectively, at the given SNP. Note that other marker-based estimates of genetic sharing are sometimes used, based on alternative derivations of the polygenic covariance matrix.

Figure 3 shows the distribution of marker-based pairwise relatedness estimates for the F8 and F9–F13 cohorts. The diagonal entries R_ii of the relatedness matrix (Fig. 3, left) are roughly centered at 1.5, as expected, although we see a slight excess of values >1.5. This suggests mating in the AIL exhibits a slight amount of inbreeding, which is not surprising. No individuals have inbreeding coefficients large enough to suggest that they are offspring of highly related parents. Figure 3, right, shows the extent to which genetic sharing varies among pairs of individuals in each of the cohorts. (Note that marker-based pairwise relatedness estimates between the F8 mice and mice from subsequent generations are not shown, but these are computed for our analyses since we need to account for relationships between these mice as well.) The off-diagonal entries R_ij (Fig. 3, right) would be 1 on average if mating were completely random, so the histograms suggest that the mice are slightly more related than expected by random mating. Reassuringly, we observed similar overall genetic sharing within the F8 and F9–F13 cohorts.

Fig. 3.

Marker-based estimates of pairwise relatedness. These plots show the distribution of the relatedness coefficient, or 2 times the kinship coefficient, for all B6 × D2 AIL mice. Diagonal entries of the n × n pairwise relatedness matrix (corresponding to inbreeding coefficients) are shown at left, and off-diagonal entries (i ≠ j) are shown at right. Pairwise relatedness estimates are obtained by calculating the number of alleles at each SNP that share the same state between the pair of individuals, then averaging over all available SNPs on autosomal chromosomes (see materials and methods for details).

Assessing support for association.

For each SNP, we reported the LOD score, or the base-10 logarithm of the likelihood ratio, as a measure of support for association with the phenotype. (Note that we did not assess support for QTLs between markers using interval mapping.) The likelihood ratio statistic was calculated in QTLRel by fitting two models to the data, a model with additive and dominance effects included for the given SNP, and another model assuming that no markers have an effect on the phenotype (the null model). Estimating the parameters for each of these models separately for each SNP is highly computationally intensive, so in practice QTLRel fits the LMM to the data by estimating the additive and dominance effects of the SNP (and, when applicable, additional sex-specific effects) while fixing the remaining parameters to their values estimated under the null model, up to a constant of proportionality, which is then re-estimated from the data. This is similar to the strategy used in the program EMMAX (22). Recently, methods have been developed to efficiently compute the test statistic at each marker without this approximation, yielding improved power to detect associations in certain circumstances (37, 66).

Determining significance of LOD scores.

To determine whether or not an LOD score constitutes “significant” support for an association between genotype and phenotype, we calculated a significance threshold by estimating the distribution of maximum LOD scores under the null hypothesis, and we took the threshold to be the 100(1 − α)th percentile of maximum LOD scores, with α = 0.05. A common approach to estimating the null distribution is to randomly permute the phenotype samples while keeping the genotypes the same. However, such a permutation procedure fails to account for differing relationships among the AIL mice and hence may yield incorrect significance thresholds and poor control of type 1 error (13, 14). Therefore, we compared significance thresholds by two different methods for permutation-based tests: 1) the method of Abney et al. (2) that accounts for the covariance structure in the AIL when permuting the data; and 2) the standard permutation procedure that assumes exchangeability of the samples (i.e., all mice are assumed to be equally related). Since allowing for an arbitrary covariance matrix dramatically increases the computational cost of the permutation-based test, we compared the two methods in the F8 mice, which exhibit broadly similar familial relationships to the F9–F13 mice (compare the distributions of kinship coefficients in Fig. 3). For each of the four muscle weights (TA, EDL, gastroc, and soleus), the significance thresholds calculated using both methods were both close to 4; the difference between the two thresholds was at most 0.15 for any one muscle weight; and the thresholds were not consistently larger or smaller in one method or the other. Therefore, we concluded that the standard permutation-based test that assumes independence of the samples would yield adequate estimates of significance thresholds. In all our analyses, we generated 1,000 permutation replicates to estimate the null distribution of the test statistic, following the recommendation of Churchill and Doerge (15).

Following a similar protocol for the model with sex-specific SNP effects, we found that LOD scores of some SNPs only exceeded the significance threshold (the 95th percentile of the null distribution of LOD scores) after allowing for different effects in males and females. However, this on its own does not correctly assess the support for a sex-specific QTL; we must also compare the evidence for models with and without separate effects in males and females and ask whether the sex-specific model yields a significant increase in the LOD score. To determine whether the increase in the LOD score is significant, we followed the permutation-based test procedure described in Broman and Sen (9) for estimating the null distribution of LOD score differences. We then declared a sex-specific QTL as “significant” only if both the LOD score and the increase in the LOD score exceeded the 95th percentile of the respective null distributions estimated via the permutation-based tests.

QTL regions.

The standard method for estimating QTL intervals in F2 intercrosses is the “LOD support interval” (e.g., Ref. 9). However, no standard procedure exists for advanced intercrosses, and in practice the best interval for each QTL depends on a number of factors, including the QTL effect size (43). To be consistent with previous QTL mapping procedures in AILs (13), we assigned the region containing the QTL (the “QTL region”) to be the interval bounded by SNPs within 2 LOD of the maximum LOD score. This 2-LOD support interval corresponds to slightly more uncertainty in the location of the QTL than the region given by the 1.5-LOD or 1.8-LOD support intervals (18, 43).

X chromosome.

Although we have called genotypes on the X chromosome, we restricted our attention to mapping QTLs on autosomal chromosomes because no methods exist for correctly analyzing the X chromosome in advanced intercross populations. Special considerations for the X chromosome in AILs include: 1) separate models for males and females (9, 63); 2) additional variance components in the LMM for males, females, plus a third variance component for “cross-sex” relatedness (49); 3) estimates of recombinant proportions for the X chromosome (8); and 4) separate estimates of significance thresholds for the X chromosome. Methods have been developed specifically for mapping QTLs on the X chromosome in other populations (e.g., F2 intercrosses), but these methods do not apply to AILs.

Bioinformatics Databases

We used the PolyPhen-2 web-based tool (53) to predict the possible effects of amino acid substitution on the function of a protein. These predictions are based on multiple sequence alignments and functional and structural characterization of the substitution site. Genes with the strongest prediction, “Probably Damaging,” in the HumDiv model are reported.

The Mouse Phenome Database (42) and Mouse Genomes Project (Sanger Institute) database were used to screen potentially causative SNPs in the positional candidate genes and compare alleles among the strains.

RESULTS

We assessed support for muscle weight QTLs at 21,224 candidate SNPs on autosomal chromosomes using data combined from two separate AIL cohorts: 1) mice from the F8 generation of a B6 × D2 AIL bred at the University of Chicago; 2) the F9–F13 generation of an AIL bred at Penn State (Table 1). We restricted our analysis to genetic variants on autosomal chromosomes for reasons given in materials and methods. Since muscle weights were measured in the F8 and F9–F13 mice at different ages and in different environments (Table 2), to analyze the combined data we included an additional binary covariate that indicates whether the sample came from the F8 or F9–F13 cohort and acts as a proxy for these factors (that is, age and environment). As expected, muscles in males were consistently larger than those in females, so we always included sex as a covariate to adjust for sex differences. It is important to point out that we did not include body weight as a covariate; even though controlling for a complex trait such as body weight can improve power to detect QTLs for some traits, in this case skeletal muscle makes a substantial contribution to overall body weight [roughly one-third of body weight is due to muscle mass (10)], so correcting for body weight could have the effect of concealing muscle weight QTLs, particularly those that are not muscle specific.

Table 1.

Summary of mice and genotype data used in study

Cohort	Breeding Location	Males	Females	Genotyping Platform	SNPs
F8	U. Chicago	212	213	Mouse MD Linkage	857
F9–F13	Penn State	205	202	MegaMUGA	20,403
Total		417	415		21,224

See materials and methods for more details about the data included in the study. Given single nucleotide polymorphism (SNP) totals are for autosomal chromosomes only. Note that 36 SNPs are common to both genotyping platforms.

All candidate SNPs were polymorphic in B6 × D2 crosses, and their alleles were at high frequencies in the AIL (Fig. 1). Even though genetic variation was ascertained with different SNP panels in the F8 and F9–F13 cohorts (36 SNPs are common to both cohorts), imputing genotypes of the nonoverlapping SNPs is straightforward in an AIL and yielded high-quality genotype estimates (see materials and methods).

Before presenting the main results of the QTL mapping, we first illustrate the benefits of combining the F8 and F9–F13 populations for mapping gastroc muscle weight QTLs.

In the analyses of gastroc muscle in the F8 mice alone, we identified a QTL on chromosome 2 and another QTL on chromosome 13 with borderline support at the significance threshold of α = 0.05 (Fig. 4, top). Allowing for different effects in males and females (gray curve in Fig. 4) revealed two additional QTLs on chromosomes 4 and 5. (Note that the threshold for declaring a significant LOD score is higher when allowing for different effects in males and females, owing to 2 extra degrees of freedom in the model.) In the F9–F13 mice alone (Fig. 4, middle), we obtained moderate support for a region overlapping the same QTL region on chromosome 2, but evidence from this cohort alone did not constitute a significant QTL. We also identified two other QTLs on chromosomes 5 and 7 (again at α = 0.05) that were not found in the F8 cross. Thus, separate analysis of the F8 and F9–F13 mice revealed no muscle QTLs common to both cohorts.

Fig. 4.

Genome-wide scans for gastrocnemius muscle weight (gastroc) in the F8 sample alone (top), in the F9–F13 mice alone (middle), and in the combined cohort (bottom). The vertical axis shows support for a quantitative trait locus (QTL) at each SNP, as measured by the logarithm of odds (LOD) score, or the base-10 logarithm of likelihood ratio. LOD scores are shown at 857 SNPs in the F8 mice, 20,403 SNPs in the F9–F13 mice, and 21,224 SNPs in the combined sample. SNPs are shown by base-pair position on each chromosome. Autosomal chromosomes are shown in alternating shades of gray. LOD scores are calculated with a linear mixed model that accounts for differences in genetic sharing among the advanced intercross mice, in which we used SNP genotypes to estimate genetic sharing. Each panel shows 2 genome-wide scans, one using a model that allows for different effects on the phenotype in males and females (in gray), and one using a model that does not allow for sex-specific effects (in black). Each of the dotted lines indicates the threshold for declaring a significant QTL; the significance threshold is defined as the 95th percentile (α = 0.05) of the likelihood ratio statistic (LOD score) under the null distribution. Note that the threshold is higher for the model that allows for different effects in males and females; this is expected given the additional degrees of freedom.

In the combined sample (Fig. 4, bottom), we confirmed the QTL on chromosome 2 that was initially identified in the F8 and the QTL on chromosome 7 that was initially identified in the F9–F13 mice. In addition, three more QTLs (chromosomes 6 and 9, and a sex-specific QTL on chromosome 8) only became significant once we combined the data sets.

Across all four muscles, using a stringent criterion for significance, we established strong evidence for 15 independent genetic factors on chromosomes 1, 2, 4, 5, 6, 7, 8, 9, 12, 14, 17, and 18 contributing to variation in the four muscles; see Fig. 5 and Table 3, and for expanded results on TA, EDL, and soleus in the F8 and F9–F13 populations, see Fig. 6, and Table 4. Individually these QTLs explain a small fraction of variation in muscle weight (0.9–3.2% proportion of variance explained).

Fig. 5.

Genome-wide scans for muscle weight QTLs in the combined F8–F13 B6D2 population. Evidence for association with the quantitative trait is shown for 21,224 SNPs on autosomal chromosomes. SNPs are plotted along the horizontal axis by chromosome and then by base-pair position. Each panel shows 2 genome-wide scans: the gray curve is from a model that allows for different genetic effects in males and females (sex-specific QTLs); the black curve is from a model in which the effect on the phenotype is the same in males and females. Dotted lines indicate significance thresholds for the genome-wide scans, taken to be the 95th percentile (α = 0.05) of the test statistic (the LOD score) under the null distribution. For further explanation, see Fig. 4.

Table 3.

Genetic architecture of muscle weights identified in the B6 × D2 advanced intercross

Chr.	QTL Region, Mb	Top SNP	Trait	2-LOD Interval Width, Mb	LOD	LOD +Sex Int.	PVE, %	Mean Trait (mg) B6/Het/D2	Candidate Gene(s)	Ref. No.
	170.074–177.699		TA	7.6	3.44	5.63*	1.3	38.0/39.8/40.5 (F) 47.4/47.4/49.4 (M)
1	170.074–176.172	rs31218323	EDL	6.1	4.96	9.75	2.4	7.32/7.68/8.04 (F) 9.30/9.14/9.72 (M)
	173.277–174.375		Gastroc	1.1	2.81	7.96	1.4	93/96/99 (F) 124/121/127 (M)
2	113.258–134.288	rs4138562	EDL	21.0	5.15	5.52	1.3	8.54/8.43/8.13	A430105I19Rik, Mga, Zscan29, Fbn1,Cep152, Anapc1, Zc3h6, Chchd5	31
	112.804–127.880		Gastroc	15.1	6.51	6.78	1.2	111/109/106
4	95.451–103.355	rs28094289	EDL	7.9	6.75	7.89	1.7	8.64/8.30/8.37
5	112.037–120.143	rs45922215	soleus	8.1	6.43	6.53	2.1	6.96/6.67/6.42	Myo18b	31, 32
	3.321–16.011		TA	12.7	1.47	4.95	1.2	38.7/38.8/39.4 (F) 48.6/47.8/46.2 (M)
6	3.984–16.011	rs30074418	EDL	12.0	2.05	5.83	1.7	7.41/7.52/7.64 (F) 9.46/9.30/9.02 (M)
	5.381–12.098		Gastroc	6.7	2.67	6.94	1.2	95/95/95 (F) 127/124/119 (M)
	34.599–43.777		TA	9.2	6.12	9.93	2.3	39.7/38.5/39.3 (F) 48.6/47.7/45.4 (M)		31
6	34.599–43.777	rs47723713	EDL	9.2	1.56	6.29	1.6	7.50/7.45/7.78 (F) 9.47/9.24/8.98 (M)
	34.145–43.777		Gastroc	9.6	4.82	8.24	1.5	97/94/95 (F) 127/123/118 (M)
	4.690–17.790		TA	13.1	1.75	4.57	1.1	39.5/39.2/39.0 (F) 46.1/47.9/47.8 (M)		32
7	4.690–17.261	rs31137734	EDL	12.8	2.99	5.68*	1.4	7.58/7.57/7.50 (F) 8.91/9.28/9.40 (M)	Ppp1r12c
	4.690–17.790		Gastroc	13.1	5.35	8.74	1.6	94/96/96 (F) 117/123/126 (M)
	121.231–130.356		TA	9.1	0.12	5.12	1.2	38.4/39.1/40.0 (F) 48.1/47.5/46.4 (M)		31
8	126.207–130.354	rs33444220	EDL	4.1	0.62	5.97	1.5	7.33/7.54/7.86 (F) 9.31/9.29/9.02 (M)	Pard3
	126.207–130.354		Gastroc	4.1	0.25	6.76	1.2	93/95/97 (F) 124/123/119 (M)
9	13.390–42.323		TA	28.3	4.23	5.32	1.0	44.4/43.4/42.6	Prdm10, Srpr, Phldb1	31
	34.763–49.608	rs6413270	EDL	14.8	5.30	6.37	1.4	8.67/8.43/8.23
	106.324–111.539	rs13480407	TA	5.2	6.70	7.57	1.6	44.8/44.0/42.4	Dhx30, Nbeal2, Lrrc2	30
9	106.229–111.860		EDL	5.4	3.66	4.36	0.9	8.48/8.55/8.27
	106.466–111.838		gastroc	5.4	5.01	6.31	0.9	113/111/107
12	28.926–53.911	rs29138639	EDL	25.0	4.85	5.33	1.2	8.31/8.40/8.73	Zfp277
17	51.905–55.432	rs6272475	soleus	3.5	4.44	4.65	1.5	6.58/6.86/6.58		32

Columns from left to right are: 1) chromosome; 2) QTL region based on 2-logarithm of odds (LOD) support interval; 3) SNP in region with largest LOD score; 4) relevant trait(s); 5) width of the QTL region; 6) LOD score of top SNP in region using a model that does not differentiate effects of SNPs in males and females; 7) LOD score of top SNP in region using a model that allows for different effects in males and females; 8) proportion of variance (in muscle weight) explained (PVE) by top SNP; 9) average phenotype value corresponding to homozygous B6, heterozygous (Het) and homozygous D2 genotypes—for sex-specific effects, the average phenotypes are shown separately for females (F) in 1st row, and males (M) in 2nd row; 10) candidate gene with stop gain (underlined) or damaging SNPs; 11) references for overlapping QTL in B6D2 F2 intercross. In the “LOD + Sex Int.” column, LOD score in boldface and italics indicates QTL is only identified after allowing for different effects in males and females or that support for a QTL increases significantly after allowing for sex-specific effects; more precisely, LOD score exceeds significance threshold at α = 0.05, and difference between LOD scores with and without sex-specific effects exceeds significance threshold at the same percentile. In the “LOD” column, boldface + italics means LOD score exceeds significance threshold at α = 0.05, and sex-specific LOD score does not satisfy our criteria for significance. Two LOD scores are marked with an asterisk (

^*) because they do not meet the α = 0.05 threshold but are just below it. In both cases, they exceed the 90th percentile (α = 0.10) of the null distribution. All SNP information and genomic positions are based on NCBI release 37 of the Mouse Genome Assembly and build 128 of the dbSNP database.

Fig. 6.

Genome-wide scans for tibialis anterior (TA, A), extensor digitorum longus (EDL, B), and soleus (C) muscle weights in the F8 sample alone (top rows), in the F9–F13 mice alone (middle rows), and in the combined cohort (bottom rows). For full details about this figure, see legend for Fig. 4.

Table 4.

Genetic architecture of muscle weight in the F8 and F9–F13 generations of the advanced intercross of B6 and D2 strains

						LOD			PVE, %			Mean trait (mg) B6/Het/D2
	Chr.	QTL Region, Mb	Top SNP	2-LOD Size	F8	F9–F13	Comb.	F8, %	F9–F13	Comb., %	F8	F9–F13	Comb.	Candidate Gene(s)
GAST	5	5.43–25.3	rs3714258	19.9	6.57		2.49	1.3		0.4	113/110/107		110/109/106
TA	17	24.54–31.15	rs49469183	6.6		4.97	3.70		3.2	0.9		41.7/42.5/43.3	43.7/43.4/42.3	Abca3
SOL	18	25.47–66.63	rs3676196	41.2	4.25		2.29	2.1		0.8	6.79/6.57/6.24		6.89/6.71/6.57	Bin1

These QTLs exceed genome-wide significance threshold at α = 0.05 in either F8 or F9–F13 cohorts of the B6D2 AIL but not combined (Comb.). LOD score at the QTL, percentage of variance explained (PVE), homozygous B6, heterozygous (Het) and homozygous D2 genotypes presented for the F8, F9–13 and the combined population. The 2-LOD drop off confidence interval is based on Build 37. Underlined candidate genes carry nonsense mutations, otherwise missense mutations with potentially damaging effects (Probably Damaging based on HumDiv model in PolyPhen-2).

We identified a large number of these QTLs only after allowing for differences in genetic effects between males and females. In other cases, the support for these QTLs increased significantly under a regression model with sex-specific effects (these are the QTLs with bold and italic LOD scores in the “LOD + sex int” column in Table 3). As we elaborate below, three QTLs predominantly affected males, and alleles at the QTL on chromosome 8 exerted opposite effects on male and female muscles. We emphasize that the LOD scores reported in Table 3 do not, on their own, constitute compelling evidence for sex specificity; we must also check whether the model allowing for separate SNP effects in males and females yields a significant increase in the LOD score over the model assuming the same effects in males and females. We estimated significance thresholds for the increase in the LOD score to be between 2.37 and 2.68 for the four muscle weights, and we only reported sex-specific QTLs when the increase in the LOD score exceeded that threshold (see materials and methods for a more detailed explanation). We found, unexpectedly, that muscles of males in the F8 cohort tended to be larger than those from F9–F13 males (who are older), whereas the females showed the reverse trend: muscles in the younger, F8 females tended to be smaller than those from the older F9–F13 females; see Table 2 and Fig. 2. This raised the additional concern whether these differences and other differences between the two cohorts might confound tests for sex-specific association. However, this concern is alleviated by the fact that we matched the proportions of males and females both in the F8 and F9–F13 samples (see Table 1). In summary, our findings suggest that the genetic contributions to muscle variation are substantially different in males and females, at least in B6 × D2 mice.

Note that in analyses of the combined data we obtained strong support for a QTL on chromosome 11 in TA and EDL muscles with LOD score as large as 7.54 (see Fig. 5), but this strong association signal was isolated to a single SNP, and the genotypes at this SNP were estimated with high uncertainty. Due to these concerns, we did not report this SNP in our final results.

Our QTL findings (Table 3) also indicate widespread shared genetic architecture, or pleiotropy, of the muscles we investigated. Weights of the different muscles were all positively correlated, in both males and females, and in the F8 and F9–F13 cohorts, with the soleus muscle exhibiting weakest correlations with other muscles: r = 0.51–0.73 for soleus and another muscle, and r = 0.65–0.85 for all other combinations of muscles (Table 5). Correspondingly, soleus exhibited the least number of shared QTLs with other muscles measured in this study. Reporting only QTLs that reach thresholds for significance probably underestimates the genetic factors that are common to multiple muscles, so in our results (Table 3) we also included loci for some muscle weights that showed moderate support for containing a QTL, and for which another muscle had a significant QTL at the same SNP. Note that two of the sex-specific QTLs given in Table 3, indicated by asterisks, did not quite meet our criteria for significance, but given that we obtain significant support for sex-specific SNP effects in other muscles at the same SNP, this strongly suggests sex-specific effects in these cases as well.

Table 5.

Correlations among muscles weight measurements, separately in F and M, and separately in F8 and F9–F13 mice

	Cohort	EDL F/M	Gastroc F/M	Soleus F/M
TA	F8	0.73/0.83	0.78/0.85	0.68/0.72
	F9–F13	0.65/0.70	0.73/0.73	0.51/0.45
EDL	F8		0.73/0.82	0.65/0.71
	F9–F13		0.72/0.66	0.53/0.50
Gastroc	F8			0.70/0.73
	F9–F13			0.57/0.56
Soleus	F8
	F9–F13

Due to the accumulated recombinations in the AIL, we were able to narrow the candidate genetic variant within 2-LOD support intervals to regions as small as 3.5 Mb; overall the QTL regions range 3.5 and 41.6 Mb, with a median of 12.5 Mb (Table 3). Subsequent inspection of these regions identified promising functional candidate genes in some of these regions (see below).

Among the 15 QTLs identified, the B6 and D2 alleles were associated with an increased muscle mass in eight QTLs (chromosomes 2, 4, 5, 9, 17, and 18) and two QTLs (chromosomes 1 and 12), respectively. For the three QTLs that showed evidence of being male specific, the B6 allele was associated with increased muscle mass in two QTLs (chromosome 6) and the D2 allele in one (chromosome 7). For one sex-specific QTL on chromosome 8, the B6 allele was associated with decreased muscle mass in females. Finally, heterozygous animals exhibited the largest muscle weight, implicating heterosis at one QTL (soleus) on chromosome 17.

Candidate Genes

We filtered genes within the QTL regions of the AILs for nonsynonymous SNPs and stop gain or loss mutations between the B6 and D2 strains. There were eight stop gain mutations found in seven genes. Six of those genes are expressed in skeletal muscle (34). In one of them, A430105I19Rik, a premature stop occurs in the B6 strain, and in the D2 strain for the remaining genes (Tables 3 and 4). In addition, there were >1,900 nonsynonymous SNPs. Over 280 genes harboring these SNPs are expressed in muscle tissue. Using PolyPhen-2 (53), we estimated if a substitution of amino acids could affect protein function and found 14 genes with such substitutions (see Tables 3 and 4).

DISCUSSION

In this study, we replicated, expanded, and refined the genetic factors for muscle mass in the B6D2 lineage. In total, we mapped 15 muscle weight QTLs to regions with a median width of 12.5 Mb (ranging between 3.5 and 41.6 Mb). Our results also point to the importance of sex-specific genetic factors in muscle development.

We investigated a panel of different muscles to address the possibility of distinct factors contributing to the variation in the size of different muscles. The substantial QTL overlap between the TA, EDL, and gastroc muscles and the absence of a shared genetic basis with soleus muscles align well with our broad understanding of the development of these muscles. Adult skeletal muscles in mice consist of a mix of different types of muscle fibers: type 1, type 2A, type 2X, and type 2B. There are developmental (46), functional, and morphological differences between the fiber types (54). Varying content of the fibers and the proportion of different fiber types determine functional and morphological properties of the muscle. The majority of mouse appendicular muscles, including quadriceps femoris, TA, EDL, gastroc, and plantaris are dominated by various proportions of type 2B, 2X, and 2A fibers (3, 6), whereas soleus muscle primarily consists of type 1, type 2A, and type 2X fibers (3, 6). It is plausible that this difference in fiber types between soleus and other muscles is an underlying cause for consistently weaker muscle weight correlations involving the soleus (Table 5). There is the potential to discover shared genetic factors for muscle weights more systematically by jointly analyzing all muscle data, similar to methods developed to detect gene expression QTLs common to multiple tissues (19).

The difference in ages of the F8 (14 wk) and F9–F13 (29 wk) populations could have played a role in which genes contributed to variation in these muscles. Therefore, it is important to consider how that might have affected detection of the genetic loci. Muscle mass is a function of the number and size of muscle fibers. The former is determined prenatally (48), whereas the size undergoes a rapid growth in length and the radius of the fiber during the first four postnatal weeks (20), followed by a slower growth phase until adult size is reached by ∼13 wk of age (62). Muscle weight in B6D2 F2 mice, however, continued to increase until 71 wk of age, although we do not know how muscle fibers were affected (30, 31). It is plausible that different factors would affect muscle mass at different ages. In support of this notion, it has been demonstrated that the genetic architecture of body weight can vary during the period of early postnatal growth; some QTLs were detected only during the first few postnatal weeks, whereas other QTLs emerged only in weeks 8–10 (59). Our studies on muscle weight in the populations of markedly differing ages (29, 71, and 114 wk of age) of B6D2 F2 mice also indicated that the genetic architecture could vary substantially (30–32). Thus, age differences can at least partially explain a limited overlap of the genetic architecture between the F8 and F9–F13 AILs. Nevertheless, combining both sets of samples, while accounting for possible confounding factors between the F8 and F9–F13 cohorts, allowed us to identify three QTLs that did not meet significance thresholds in separate analyses.

Many of the findings for these muscle traits align with results of previous QTL mapping studies in crosses of B6 and D2 mice and in crosses of other lab strains. Nine out of 15 QTLs identified in the B6D2 AILs overlap with genomic regions reported in earlier studies with B6D2 F2 intercrosses and BXD RI strains (30–32). In addition, these nine QTL also largely replicate the allelic effects reported earlier (with exception of QTLs on chromosome 8 and 17, where no sex specificity or heterosis, respectively, was detected earlier; Table 3). One more QTL, chromosome 12, overlaps with the QTL detected in the cross between the DUi and DBA/2 strains (7). Convergence of different studies on the same genomic regions strengthens the evidence for capturing genuine genetic architecture of skeletal muscle in this species. However, an important improvement by the present study is smaller QTL support interval compared with the F2s.

Sex-specific effects were detected in approximately one-third of identified QTLs. A larger muscle mass in males compared with females has been well established. Different level of androgens between males and females plays an important role. For instance, testosterone in serum is 10–20 times higher in males than females (58) and could significantly contribute to sex dimorphism in muscle mass. In mice, the number of fibers is similar across sex, but fibers are larger in males (11). Also, at least in soleus muscle, there is a shift toward oxidative, type 1, fibers in females compared with males (11). This shift can further contribute to the sex difference in weight because type 1 fibers are smaller in cross-sectional area (CSA) compared with glycolytic, type 2, fibers in one of the parental strains, DBA/2J (26). Testosterone is associated with the shift from the oxidative toward glycolytic fibers and with an increase in CSA of the fibers, particularly in the muscles dominated by the glycolytic fibers. However, it does not affect the fiber number (58). Thus, if a causative gene is involved in determination of the number of fibers, both sexes are expected to respond similarly. However, in the instances where elevated level of testosterone mediates manifestation of the allelic effect, the response might differ between males and females. Collectively, these findings highlight the importance of allowing for sex-specific effects in QTL mapping.

We did not find any established muscle development genes within the QTL regions we identified (Table 3). We also did not identify large contributions to muscle weight variation within loci that harbor known muscle growth modifiers such as myostatin (45), follistatin (28), PKB/Akt (4), and Igf-1 and its receptor (39). This lack of overlap with myostatin and another genes is consistent with our previous experience in an advanced intercross of the LG/J and SM/J strains (33, 35). It is unlikely that this is due to low power to detect QTLs considering that more than half of the muscle QTLs presented in this report overlap with previously mapped muscle weight loci in B6D2 lineage. A more plausible explanation is that there are no variants between the B6 and D2 strains that induce large changes in function or expression of these genes. Thus, the results suggest that a substantial fraction of variation in muscle mass in these mice is caused by genes with poorly understood roles in skeletal muscle. There are 20 genes within the QTL with premature stop codon or nonsynonymous SNPs in evolutionary conserved regions (Table 3). However, noncoding changes are also likely to be important contributors to the phenotypic differences. Indeed, there are extensive expression differences between the B6 and D2 strains; these differences can be observed in ∼40% of genes expressed in various tissues (25). (Note that skeletal muscle was not one of the tissues included in that study.) In the future, it will be important to determine the identity of differentially expressed genes in the muscles of B6 and D2 strains. This would aid nomination of candidate genes, particularly for those QTLs where no genes could be prioritized based on coding sequence annotations.

In summary, we replicated and identified novel muscle weight QTLs in the B6D2 lineage, and we demonstrated that mapping in AIL can markedly improve resolution of the QTL regions.

GRANTS

Research reported in this publication was supported by NIH Grant AR-056280 (D. A. Blizard). This work was supported by NIH Grant DA-021336 (A. A. Palmer). P. Carbonetto was supported by a cross-disciplinary postdoctoral fellowship from the Human Frontiers Science Program.

DISCLOSURES

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

AUTHOR CONTRIBUTIONS

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