Gene-by-environment Interactions and Adaptive Body Size Variation in Mice From the Americas

Abstract

The relationship between genotype and phenotype is often mediated by the environment. Moreover, gene-by-environment (GxE) interactions can contribute to variation in phenotypes and fitness. In the last 500 yr, house mice have invaded the Americas. Despite their short residence time, there is evidence of rapid climate adaptation, including shifts in body size and aspects of metabolism with latitude. Previous selection scans have identified candidate genes for metabolic adaptation. However, environmental variation in diet as well as GxE interactions likely impact body mass variation in wild populations. Here, we investigated the role of the environment and GxE interactions in shaping adaptive phenotypic variation. Using new locally adapted inbred strains from North and South America, we evaluated response to a high-fat diet, finding that sex, strain, diet, and the interaction between strain and diet contributed significantly to variation in body size. We also found that the transcriptional response to diet is largely strain-specific, indicating that GxE interactions affecting gene expression are pervasive. Next, we used crosses between strains from contrasting climates to characterize gene expression regulatory divergence on a standard diet and on a high-fat diet. We found that gene regulatory divergence is often condition-specific, particularly for trans-acting changes. Finally, we found evidence for lineage-specific selection on cis-regulatory variation involved in diverse processes, including lipid metabolism. Overlap with scans for selection identified candidate genes for environmental adaptation with diet-specific effects. Together, our results underscore the importance of environmental variation and GxE interactions in shaping adaptive variation in complex traits.

Introduction

Understanding how populations vary in genotype, phenotype, and fitness in association with the local environment is critical to understanding the process of adaptation. One enduring challenge is integrating the influence of the environment on the generation of and selection for phenotypic variation. While environmental factors can act as selective agents favoring particular phenotypes, they can also mediate the relationship between genotype and phenotype. A single genotype may produce different phenotypes across different environments, known as phenotypic plasticity (Scheiner 1993; West-Eberhard 2003; Pigliucci 2005). These plastic responses can allow individuals to rapidly respond to environmental variability (Baythavong 2011; Diamond and Martin 2021). However, phenotypic plasticity itself can vary substantially among individuals and populations and be the target of selection (Lande 2009; Sommer 2020). When the environment influences the relationship between genotype and phenotype, this is known as a genotype- or gene-by-environment (GxE) interaction (Scheiner 1993; Josephs 2018). GxE interactions can result in phenotypic differences between individuals that are only observed under specific conditions, making them difficult to study in many natural populations (Pigliucci 2005). However, the context-dependent nature of GxE interactions also means these interactions can play an important role in shaping the relationship between environment and organismal fitness during local adaptation, particularly when GxE interactions impact the expression of adaptive phenotypes (Price et al. 2003; Ghalambor et al. 2007; Corl et al. 2018; Campbell-Staton et al. 2021; Chen et al. 2023b; Bautista et al. 2024; Jacobs et al. 2024; Siddiq et al. 2024).

House mice (Mus musculus domesticus) have proven to be a valuable system for studying local adaptation. Alongside humans, in the last ∼500 yr, house mice have colonized a wide range of habitats and climates in the Americas (Phifer-Rixey and Nachman 2015; Agwamba and Nachman 2022). Populations across latitudinal gradients show significant evidence of local adaptation at several complex traits (Lynch 1992; Phifer-Rixey et al. 2018; Ferris et al. 2021; Gutiérrez-Guerrero et al. 2024). One of the most striking examples is body mass. Past studies have identified variation in body size in house mice correlated with latitude (Lynch 1992; Phifer-Rixey et al. 2018; Suzuki et al. 2020), consistent with the ecogeographic pattern known as Bergman's rule (Bergmann 1848). In mammals, the covariance of body size and latitude is generally attributed to thermoregulatory adaptations to differences in temperature (Ashton et al. 2000). At higher latitudes, where individuals must contend with colder climates, larger body sizes can diminish heat loss through lower surface area to volume ratios. For example, mice from the equatorial city Manaus in Brazil weigh nearly 37% less on average than mice from New York state in the United States (Suzuki et al. 2020). These differences in body mass persist in the lab after multiple generations, indicating they have a genetic basis (Phifer-Rixey et al. 2018; Dumont et al. 2024). Genome-wide scans for selection and gene expression analyses have identified a number of candidate genes for climate adaptation, including genes with previously described roles in metabolism and body size variation (Phifer-Rixey et al. 2018; Mack et al. 2018; Ferris et al. 2021; Ballinger et al. 2023; Gutiérrez-Guerrero et al. 2024; Durkin et al. 2024). Candidate variants for environmental adaptation identified are predominantly non-coding, suggesting that changes in gene expression regulation play an important role in driving adaptation in this system (Mack et al. 2018; Phifer-Rixey et al. 2018). For example, cis-regulatory variation at genes Adam17 and Bcat2 was significantly associated with clinal body mass variation in wild mice in North America (Mack et al. 2018). Previous studies have also suggested an important role for plastic responses in phenotypic and gene expression variation in this system (Bittner et al. 2021; Ballinger and Nachman 2022; Ballinger et al. 2023). For example, the temperature at which mice are reared has been found to affect aspects of morphology and gene expression variation (Bittner et al. 2021; Ballinger and Nachman 2022; Ballinger et al. 2023). When mice from Amazonas, Brazil and Saratoga Springs, New York, USA were reared under cold temperatures, extremities (i.e. tail and ear length) were shorter than when mice were reared under warm conditions, consistent with Allen's Rule (Ballinger and Nachman 2022). Genotype-by-temperature interactions affecting gene expression were also common (affecting 5% to 10% of genes surveyed) (Ballinger et al. 2023).

In humans, genetics, diet, and GxE interactions have all been shown to contribute to variation in body mass and composition (Sulc et al. 2020; Bachmann et al. 2022; Jung et al. 2023) with GxE interactions playing a major role in metabolic syndrome and obesity risk (Loos and Yeo 2022; Kim et al. 2024). A now well-described example comes from the fat mass and obesity-associated (FTO) locus (Speliotes et al. 2010). Interactions between FTO risk alleles and diet (e.g. fatty foods) are associated with increased obesity risk (Phillips et al. 2012). The interaction between genetic and environmental factors related to diet can also play an important role in local adaptation. Variation in food abundance or nutritional value can exert strong selection pressure. For example, there is evidence that Greenlandic Inuits have adapted to lipid-rich diets (Fumagalli et al. 2015). Selection on genetic variants related to fat metabolism may have facilitated tolerance of diets high in omega-3 polyunsaturated fatty acids from seafood and variation in those loci is associated with differences in body weight and height.

Similarly, differences in access to food resources may have exerted selection pressure as house mice colonized new climates in the Americas. Studies in wild mice indicate diet can vary drastically by location and season and that survival and breeding can be affected by food availability (Pennycuik 1972; Bomford 1987a, b; Goertz et al. 2019; Polito et al. 2022). Consequently, climatic and geographic variation affecting diet may contribute to local adaptation in house mice. While little is known about the role of diet and GxE interactions in metabolic and body size variation in wild mice, studies of classical inbred strains of house mice have been critical to our understanding of the genetic architecture of body size (e.g. Ingalls et al. 1950; Bennett 1961; Hummel et al. 1966; Chen et al. 1996; Cheverud et al. 1996; Keightley et al. 1996; Brockmann et al. 1998; Corva and Medrano 2001; Cheverud 2005; Kemper et al. 2012). Because of clinical relevance, the impact of diet on body size, physiology, and metabolism in mice has been intensively investigated for decades (e.g. Fenton and Dowling 1953; Lemonnier et al. 1975; Hariri and Thibault 2010; de Moura E Dias et al. 2021). In addition to sex, genetic background has consistently been shown to contribute to variation in response to high-fat diet in laboratory strains (e.g. Nishikawa et al. 2007; Parks et al. 2013; Attie et al. 2017; de Conti et al. 2020; de Moura E Dias et al. 2021; Roy et al. 2021; Bachmann et al. 2022), thus GxE effects are expected to be common and may be important in body size adaptation in wild populations.

Gene expression can be a powerful lens for studying GxE interactions and their role in adaptation (e.g. Kita and Fraser 2016; He et al. 2021; Hämälä et al. 2022). Gene expression is sensitive to the environment and can evolve rapidly in response to environmental pressures. As an intermediate phenotype, gene expression can also be used to help meet the challenge of connecting genetic variation to variation in complex traits. Plasticity in gene expression may be regulated by changes in cis, where mutations alter local regulatory element activity, or via changes in trans, where mutations may be far from the gene of interest (Wittkopp and Kalay 2011; Signor and Nuzhdin 2018). In general, cis-regulatory changes are expected to be more modular and less pleiotropic relative to changes in trans, which may affect the expression of many downstream genes (Stern and Orgogozo 2008; Signor and Nuzhdin 2018). This expectation has also led to the prediction that cis-regulatory mutations may be driving forces in adaptive evolution (Wray 2007; Stern and Orgogozo 2008; Signor and Nuzhdin 2018). Where many studies now support an important role for cis-regulatory mutations in gene expression divergence (Wittkopp et al. 2008; Gordon and Ruvinsky 2012; Coolon et al. 2014; Metzger et al. 2017) and adaptation (e.g. Tishkoff et al. 2007; Chan et al. 2010; Fraser et al. 2010; Fraser 2013; O’Brown et al. 2015), the relationship between gene regulation and expression plasticity is less understood. While some studies have found that cis-acting variation is the primary basis for variation in plasticity (Lovell et al. 2016; Zhang et al. 2023), others have identified a larger role for trans-acting variation (Li et al. 2006; Chen et al. 2015; Lovell et al. 2018; Ballinger et al. 2023).

Here, our goal was to investigate the impact of genetic and environmental variation and GxE interactions on variation in body size and gene expression in the context of adaptive evolution. We leveraged new wild-derived inbred strains of house mice collected from different climates in North and South America and subjected them to both a standard and a high-fat diet, including both male and female mice. First, we found that variation in body mass among strains is influenced by diet, strain, and strain-by-diet interactions, supporting an important role for the environment and GxE in phenotypic variation. Next, using RNA-seq of liver tissue, we found that strain, sex, and diet drive variation in gene expression. Importantly, differences among the sexes highlight the need to consider sex-specific processes in adaptive evolution of body size. Using gene co-expression network analysis, we identified candidate gene sets associated with body mass variation and response to high-fat diet. Then, using crosses between strains from divergent climates (New York and Florida, New York and Amazonas, Brazil), we identified gene regulatory differences between strains, as well as cis-by-diet and trans-by-diet interactions affecting expression divergence. Importantly, we show evidence for lineage-specific selection on cis-regulatory changes related to metabolic processes between mice from divergent climates. Finally, comparing our expression results to selection scans, we identified candidate genes which may contribute to variation in body mass among the populations. In particular, we identify candidates with cis-by-diet interactions in our analysis with known roles in diet-induced metabolic phenotypes. Overall, our results underscore the importance of interactions between environmental and phenotypic variation during adaptive evolution.

Results and Discussion

Variation in Phenotypic Responses to a High-Fat Diet

Largely consistent with Bergmann's Rule, wild and lab-born mice from different populations in the Americas vary in traits relating to body size, with larger mice at higher latitudes (Phifer-Rixey et al. 2018; Dumont et al. 2024). Body size is a complex trait, dependent on many genes and the environment (e.g. Kemper et al. 2012; Loos and Yeo 2022). Among environmental variables, diet is well-established as a factor that contributes significantly to variation in body weight (e.g. Wang and Liao 2012; Loos and Yeo 2022). While house mice are human commensals, they are opportunistic omnivores, and diet, estimated via carbon (δ¹³C) and nitrogen (δ¹⁵N) stable isotopes, has been shown to vary among wild populations in the Americas (Suzuki et al. 2020) (Kruskal–Wallis, δ¹⁵N P = 7.96 × 10⁻⁵, δ¹³C P = 6.60 × 10⁻⁴; supplementary table S1 and fig. S1, Supplementary Material online). Moreover, populations at different latitudes are expected to differ in seasonal resource availability, making metabolic response to diet critical to fitness.

Response to diet is known to vary among classical inbred strains (Nishikawa et al. 2007; Fujisaka et al. 2018; Siersbæk et al. 2020; Bachmann et al. 2022). Here, our goal was to determine if new locally adapted strains from different climates also show GxE interactions affecting body size. Specifically, we asked whether 8 new wild-derived strains from the Americas varied in aspects of body size and growth on a regular fat breeder diet (hereafter, “regular” for simplicity) and a high-fat diet over 12 wk postweaning. These strains are derived from 5 localities in the Americas: Manaus, Amazonas, Brazil (MANB), Gainesville, Florida, USA (GAIB), Tucson, Arizona, USA (TUCC), Saratoga Springs, New York, USA (SARA, SARB, SARC), Edmonton, Alberta, Canada (EDME, EDMF). There are currently only a modest number of commercially available wild-derived inbred strains. Laboratory mice capture only a fraction of the genetic variation found in wild mouse populations, and many were developed through intense artificial selection for traits of interest as well as for breeding ability and docility (Beck et al. 2000; Yang et al. 2007, 2011; Dumont et al. 2024). Consequently, the genetic basis of many complex traits of interest in lab mice may differ from that of wild populations (e.g. Dumont et al. 2024). A recent study on these new strains from the Americas found that they harbored millions of genetic variants absent from current laboratory models as well as phenotypic variation across many complex traits of interest for evolutionary biology and biomedicine (e.g. biochemical, neurobehavioral, physiological, morphological, and metabolic traits) (Dumont et al. 2024). The strains chosen for this study represent as much climatic variation as possible, including multiple strains from the same location when available.

Because house mice are sexually dimorphic (e.g. Slábová et al. 2010; Ruff et al. 2017), with male mice typically larger than female mice (as seen in our data, Fig. 1, supplementary fig. S2 and tables S2 to S5, Supplementary Material online), we analyzed the morphological data for each sex separately. As expected, we found that the covariate in all analyses (either wean weight or week 12 body length), contributed significantly to variation in all measures (Table 1). We also found that strain contributed significantly to variation in all measures, with the exception of body length in females (Table 1). Tukey's tests suggest that strains from higher latitude locations tend to be larger than those from lower latitude locations (e.g. Saratoga Springs, New York, USA strains (SARA, SARB) and Edmonton, Alberta, Canada strains (EDME, EDMF) versus MANB and Gainesville, Florida, USA (GAIB) strains) (supplementary tables S6 and S7, Supplementary Material online), consistent with previous data from wild and lab-reared individuals (Lynch 1992; Phifer-Rixey et al. 2018; Suzuki et al. 2020; Dumont et al. 2024). A notable exception is SARC, a strain from Saratoga Springs, New York, USA, which is smaller than other high latitude strains. This result is not unexpected. Body size is not a fixed trait and varies within populations. Each wild-derived strain represents a subset of variation sampled from a larger population. SARC showed similar results in a previous study, with body sizes closer to strains from lower latitudes (Dumont et al. 2024). Similarly, males from the Gainesville, FL strain in our study (GAIB) are smaller on average than MANB males from Amazonas, Brazil, despite mice from Amazonas typically being smaller (Suzuki et al. 2020; Dumont et al. 2024). Overall, differences among strains are largely consistent with results from wild populations and early generations of inbreeding (Phifer-Rixey et al. 2018; Suzuki et al. 2019) as well as previous phenotyping of rederived inbred strains (Dumont et al. 2024).

Fig. 1.

Variation in body size and response to diet among strains from the Americas. (a) The new wild derived inbred strains originate from locations in the Americas. (b) In female experimental mice, strain and the interaction between strain and diet contributed to variation in body weight. (c) In male mice, strain, diet, and their interaction contributed to variation in body weight. Bars in (b) and (c) are colored by locality in (a).

Table 1.

Results of analyses of body size in mice fed either a regular or high-fat diet for 12 wk postweaning

Response variable	Sex	Predictor	Sums of squares	df	F	P-value
Body Weight (g)	M	Strain	223.757	7	7.543	1.65 × 10−7
		Diet	51.405	1	12.130	0.001
		Wean Weight	203.416	1	48.001	2.34 × 10−10
		Strain:Diet	79.979	7	2.696	0.013
		Residuals	504.289	119	–	–
	F	Strain	111.262	7	8.219	3.46 × 10−6
		Diet	6.205	1	3.209	0.076
		Wean Weight	157.666	1	81.533	3.11 × 10−15
		Strain:Diet	42.617	7	3.148	0.004
		Residuals	235.921	122	–	–
Body length (mm)	M	Strain	431.397	7	3.283	0.003
		Diet	5.462	1	0.291	0.59
		Wean Weight	516.472	1	27.516	6.90 × 10−7
		Strain:Diet	101.002	7	0.769	0.615
		Residuals	2252.372	120	–	–
	F	Strain	220.139	7	1.844	0.085
		Diet	52.740	1	3.093	0.081
		Wean Weight	204.323	1	11.981	0.001
		Strain:Diet	126.333	7	1.058	0.39
		Residuals	2097.542	123	–	–
Body weight/length	M	Strain	0.017	7	6.640	1.24 × 10−6
		Diet	0.005	1	13.706	3.25 × 10−4
		Wean Weight	0.011	1	31.405	1.38 × 10−7
		Strain:Diet	0.006	7	2.368	0.027
		Residuals	0.043	119	–	–
	F	Strain	0.011	7	7.027	4.88 × 10−7
		Diet	0.000	1	0.680	0.411
		Wean Weight	0.013	1	58.084	6.03 × 10−12
		Strain:Diet	0.005	7	3.029	0.006
		Residuals	0.028	122	–	–
BMI (kg/m2)	M	Strain	1.254	7	3.421	0.002
		Diet	0.451	1	8.603	0.004
		Wean Weight	0.349	1	6.663	0.011
		Strain:Diet	0.441	7	1.204	0.306
		Residuals	6.232	119	–	–
	F	Strain	1.225	7	4.078	4.71 × 10−4
		Diet	0.002	1	0.053	0.819
		Wean Weight	0.937	1	21.837	7.713 × 10−6
		Strain:Diet	0.684	7	2.276	0.033
		Residuals	5.234	122	–	–
Growth rate	M	Strain	1.597	7	9.115	5.54 × 10−9
		Diet	0.347	1	13.871	3.007 × 10−4
		W12 Length	0.572	1	22.863	5.028 × 10−6
		Strain:Diet	0.433	7	2.470	0.021
		Residuals	2.978	119	–	–
	F	Strain	0.739	7	7.383	2.19 × 10−7
		Diet	0.028	1	1.991	0.16
		W12 Length	0.071	1	4.949	0.028
		Strain:Diet	0.273	7	2.724	0.012
		Residuals	1.745	122	–	–

Analyses included data from all strains with models of the form: Response Variable ∼ Strain * Diet + Covariate.

In male mice, diet contributed significantly to variation in body weight measures (Table 1). The same was not true in female mice (Table 1). However, in both male and female mice, Tukey's tests provide evidence that mice on the high-fat diet are significantly heavier (supplementary table S8, Supplementary Material online). Importantly, in all tests of measures that incorporate body weight in female mice, the interaction between strain and diet contributed significantly to variation (Table 1). Although results were less extreme, that was also true in male mice with the exception of BMI (Table 1). Significant contributions to variation from the strain × diet interaction provide evidence that there is variation among the strains in phenotypic response to diet. As with past studies (Phifer-Rixey et al. 2018), there was no evidence of differences among strains in food intake (supplementary table S9, Supplementary Material online) suggesting that differences in body weight among these strains are not driven by differences in food consumption. There was also little evidence that diet contributed to differences in food intake, although the strain × diet interaction did contribute to variation in food intake in week 5 in female mice (supplementary table S9, Supplementary Material online). Consistent with studies of classical inbred and recombinant inbred strains (e.g. Roy et al. 2021; Bachmann et al. 2022), these data demonstrate both that the environment (i.e. diet) impacts body weight and that there is genetic variation in plasticity in response to diet among these strains (Fig. 1, Table 1; supplementary table S3, Supplementary Material online).

Extensive Gene Expression Divergence Between Strains

Importantly, variation in response to diet among these locally adapted strains provides an opportunity to investigate the genetic underpinnings of plasticity. Plastic responses are often driven by change in gene regulation. To examine regulatory differences related to metabolic evolution and local adaptation, we analyzed gene expression differences across a subset of the phenotyped strains (MANB, GAIB, SARA, SARB, EDME). Results of analyses of body size with this subset of strains were similar to results with the full set (supplementary fig. S3 and table S10, Supplementary Material online). To examine the impact of environmental variation on gene expression, liver tissue from each strain was collected from mice on both the regular and high-fat diet. Collections from each biological replicate were completed at 12 wk postweaning and mRNA was extracted and sequenced (for details, see Methods).

Strain and sex were the primary drivers of expression variation. Principal component (PC) analysis of gene-wise mRNA abundance separated samples by sex on PC1 and strain on PC2 (27% and 21% of variance, respectively) (Fig. 2a). Strains clustered on PC2 consistent with genetic distance (Dumont et al. 2024). Reflecting their shared geographic origin, the 2 New York strains (SARA and SARB) clustered closely on PC2. Within sexes and strains, diet treatment was also found to be a major source of variation (supplementary figs. S4 and S5, Supplementary Material online).

Fig. 2.

Gene expression differences across populations and diet treatments. (a) PC analysis of gene-wise mRNA abundance. Population is indicated by color, where sex and diet are designated by shape. (b) Pairwise overlap between strains in expression responses to diet. In each square is the number of genes with differential expression between a high-fat and regular diet shared between 2 strains. Boxes are colored by percent overlap. See also supplementary figs. S8 and S9, Supplementary Material online for overlap across sets. (c) Overlap in differentially expressed genes between diet treatments in males and females. Venn diagrams are scaled within strains.

A total of 22,497 genes could be examined for differential expression between strains. Pairwise comparisons revealed extensive gene expression differences between strains. Between 2,389 and 6,335 genes showed significant differences in expression between at least 2 genotypes under either diet (DESeq2 Wald test) (supplementary table S11, Supplementary Material online). These data provide strong evidence of divergence in gene expression among strains and between male and female mice.

Gene Expression Responses to Diet are Strain- and Diet-specific Among North and South American Mice

Given variation in the phenotypic response to diet and extensive variation in gene expression among the strains, we next investigated plasticity in gene expression. To understand whether the transcriptional response to diet is shared or strain-specific, we compared expression between diet treatments for each strain. Across all genotypes, we identified 2,163 genes with a significant transcriptional response to diet, representing approximately 10% of genes surveyed (41 to 1,303 genes per line/sex; DESeq2 Wald test, FDR < 0.05). Genes with changes in expression in response to diet were significantly enriched for GO terms associated with aspects of metabolism (e.g. cholesterol biosynthetic process, triglyceride metabolic process, regulation of fatty acid metabolic process) as well as genes with mutant phenotypes that affected lipid levels (MP:0001547, Benjamini-Hochberg adjusted P-value (q) = 3.19 × 10⁻⁸), triglyceride levels (MP:0011969, q = 7.4 × 10⁻⁸), and cholesterol levels (MP:0003947, q = 0.003).

Diet effects were largely sex- and strain-specific rather than shared (Fig. 2b), indicating GxE interactions are pervasive. Over 80% of diet-induced changes in gene expression were observed in only 1 genotype (1,749 genes). Female mice from the SARA line showed the greatest transcriptional response to diet based on the number of genes with differential expression (1,303) and average fold change per gene between diets (|log₂ (high-fat/regular)|; Kruskal-Wallis, P = 5.12 × 10⁻¹⁷⁴, ad-hoc Wilcoxon signed-rank test, SARA females versus other genotypes, P < 2.38 × 10⁻⁷⁵). The greater transcriptional response in SARA mirrored what was observed at the organismal level, with SARA mice gaining the most weight on a high-fat diet, especially female mice (supplementary table S3, Supplementary Material online). Genes with diet-induced changes in expression specific to the SARA line were enriched for ontogeny terms related to lipid metabolism and gene expression (e.g. lipid metabolic process, q = 2.2 × 10⁻³; gene expression, q = 1.04 × 10⁻⁵; rRNA processing, q = 3.45 × 10⁻²) relative to other genes with diet-induced expression changes. Between the 2 strains derived from the New York population (SARA and SARB), we also found transcriptional response to diet to be highly divergent. Only 7.5% and 14% of genes with transcriptional responses to diet were shared between the 2 lines for females and males, respectively.

Only 1 gene, Gstm2, showed expression changes in response to high-fat diet in both sexes across all strains. Although basal expression varied across strains (supplementary fig. S6, Supplementary Material online), Gstm2 was consistently downregulated on a high-fat diet. Gstm2 plays an important role in modulating hepatic fat, and its deficiency is associated with aggravated insulin resistance, hepatic steatosis, and inflammation on a highfat diet (Jin et al. 2022; Lan et al. 2022). Eight genes showed significant responses to diet in males across all lines (Celf2, Cyp2c29, Cyp2c55, Gm32342, Gstm1, Gstm2, Gstm2-ps1, Rnd2) and 4 genes showed diet responses across all lines for females (Gstm3, Ctse, Mgrn1, Gstm2). Other genes with diet responses across multiple genotypes included genes with known roles in metabolism (e.g. Apoa1, Apoe, Fabp5, Irs2, Scp2, Srebf2), and responsiveness to diet in classical inbred mouse strains (e.g. Cyp3a11, Scd1) (Ntambi et al. 2002; Sun et al. 2024). These data provide strong evidence for GxE interactions affecting gene expression, including many diet-responsive genes with functional ties to metabolism.

Stronger Transcriptional Responses to High-Fat Diet in Female Mice

As house mice are typically sexually dimorphic (Schulte-Hostedde 2008; Slábová et al. 2010; Ruff et al. 2017) and past studies with laboratory strains have shown differences in response to high-fat diet among the sexes (Ingvorsen et al. 2017; Casimiro et al. 2021; Bachmann et al. 2022), we examined sex-specific responses to diet affecting gene expression (Fig. 2c). Over twice as many genes were found to show diet-based changes in gene expression in females compared with males (1,862 genes and 571 genes, respectively, FDR < 0.05). However, 1,107 genes with female-specific diet responses were observed only in the SARA line. Excluding SARA, a greater number of expression changes were still observed for females, but this difference was more modest (717 versus 548 genes). The magnitude of expression difference in response to diet per gene based on fold change was also higher for females than males in all strains except MANB (|log₂ fold change (High-fat/Regular fat)|; Wilcoxon signed-rank test: MANB P-value = 0.79, other P-values = <0.0069). Increases in growth rate and weight gain on the high-fat diet are also higher for female mice in some strains (MANB, GAIB, SARA) but not others (SARB, EDME, supplementary table S3, Supplementary Material online). The greatest average expression difference between the sexes was observed for SARA mice, followed by the EDME mice (mean |log2| difference SARA = 0.087, EDME = 0.02). Greater sexual dimorphism in SARA is consistent with female SARA mice having the largest phenotypic and transcriptional response to diet.

Next, we used DESeq2 to identify significant interactions between sex and diet in individual strains (sex-by-diet effects). Forty-three genes were found to have significant sex-by-diet interactions. All but 1 gene with sex-by-diet interactions were strain specific (42/43 genes, FDR < 0.05). Genes with significant sex-by-diet interactions in the liver were enriched for metabolic process terms (thyroid hormone metabolic process, q = 8.75 × 10⁻⁴; lipid metabolic process, q = 1.42 × 10⁻²; ethanol metabolic process, q = 2.59 × 10⁻⁴). As in studies of laboratory strains (Bachmann et al. 2022), there is strong evidence of sexual dimorphism in the transcriptional response to diet in these strains. Moreover, genes that show sex-by-diet interactions demonstrate the importance of considering sex-specific metabolic processes.

Strain-specificity of High-Fat Diet Responses in Wild-derived and Classical Inbred Strains

It is clear that the morphological response to high-fat diet is highly dependent on genetic background, both from this study and studies of classical inbred and recombinant inbred strains of mice (e.g. Roy et al. 2021; Bachmann et al. 2022). To compare our results to established laboratory strains, we re-analyzed liver expression data from Bachmann et al. (2022) for 6 classical inbred lines (NZO/HILtJ, C57BL6/J, DBA2/J, A/J, 129S1/SvlmJ, NOD/ShiLtJ) and 3 wild-derived inbred lines (WSB/EiJ, PWK/PhJ, CAST/EiJ) subjected to either a regular or high-fat diet (see Methods). In this dataset, between 33 and 2,447 genes were differentially expressed between diet treatments for each genotype (0.02% to 14% of genes surveyed) (see Methods). As in our dataset, more differentially expressed genes were identified in female mice (5,864 versus 3,026 genes) (Bachmann et al. 2022).

Approximately 54% of genes with diet-based differences in expression in our dataset were also identified in at least one other classical or wild-derived inbred strain (1,171 genes). This degree of overlap was observed despite differences in time spent on a high-fat diet and age across experiments (see Methods). Comparing across all strains, 8,359 genes showed significant differences in expression associated with diet. Importantly, though, approximately 58% of these were observed in only 1 genotype. Overall, both classical and wild-derived inbred lines show high levels of strain-specificity in their transcriptional response to diet (supplementary figs. S7 to S9, Supplementary Material online), emphasizing the importance of GxE interactions in both evolutionary and biomedical contexts. A notable example of this is Gstm2. Gstm2 has been studied in C57BL for its role in protecting mice from excess fat accumulation on high-fat diets (Jin et al. 2022; Lan et al. 2022). Several studies have found that Gstm2 is upregulated in male C57BL6 in response to high-fat diet (Wang et al. 2016; He et al. 2020; Jin et al. 2022). While we also found upregulation of Gstm2 in C57BL6/J males on a high-fat diet (log₂ fold change (high-fat/regular fat) = 0.91, q = 0.016) using data from Bachmann et al. (2022), significant upregulation of Gstm2 was not observed in the other classical inbred strains. Moreover, this gene was found to be consistently downregulated in the strains from the Americas (supplementary fig. S6, Supplementary Material online) and in CAST/EiJ males from Bachmann et al. (2022) (log₂ fold change (high-fat/regular fat) = −0.89, q = 0.036).

Gene Co-expression Analysis Identifies Genes Associated With Variation in Metabolic Traits

As a proximal phenotype, gene expression can help connect genetic variation to variation in organismal level traits. To explore such connections, we next performed a weighted gene co-expression network analysis (WGCNA) (Langfelder and Horvath 2008). WGCNA was used to identify groups of genes with highly correlated expression, called co-expression modules, for each diet treatment and sex separately. Co-expression modules were then tested for correlations with metabolic phenotypes (body weight, body length, BMI, growth rate, and food intake for weeks 1 and 9). Metabolic phenotypes, except for measures of food intake, were non-independent and correlated across individuals (supplementary table S12, Supplementary Material online). We identified several modules associated with trait variation in each group (supplementary table S13, Supplementary Material online). Between 9 and 17 co-expression modules were associated with body mass in each comparison, the majority of which were also significantly associated with other metabolic traits. In contrast, few co-expression modules were associated with food intake measures (2 to 4 modules per diet and genotype combination) and these modules were typically not significantly associated with other metabolic phenotypes.

To identify common modules across diet treatments, we performed a consensus network analysis (full dataset available in Supplementary material online). In a consensus network analysis, co-expression modules are identified across treatments (63 in males, 32 in females). We saw relatively high preservation of gene co-expression modules across the 2 diets (Density value (D) = 0.92 in females, 0.83 in males) (see Methods). In the consensus network, we identified co-expression modules associated with weight (7 and 14 modules in females and males, respectively), BMI (2 and 3), body length (4 and 15), growth (5 and 11), and food intake (1 and 0) (supplementary figs. S10 and S11, Supplementary Material online). To identify modules with different associations with metabolic traits in regular versus high-fat diets, we compared the preservation between module and trait associations (i.e. weight, BMI) across these sets. Overall, the relationship between traits and modules was highly preserved across diets (supplementary figs. S12 and S13, Supplementary Material online). This result suggests that, while at finer resolution transcriptional response to diet is highly strain-specific, at the level of pathways, associations between phenotype and expression are largely shared.

Nevertheless, we did identify module-trait relationships with lower preservation, where a module was correlated with trait variation under only 1 diet. For example, in males the “yellowgreen” module was associated with body mass only under a high-fat diet (high-fat, corr = 0.69, P = 4 × 10⁻⁹; regular fat, corr = −0.16, P = 0.4, preservation = 0.57). Genes in this module with strong associations with body weight under the high-fat diet include Insulin induced gene 2 (Insig2). Genetic variation at this gene has been associated with metabolic traits in humans (Hotta et al. 2008; Talbert et al. 2009; Kaulfers et al. 2015). In females, the “steelblue” module was associated with body mass variation only under a high-fat diet (high-fat, Pearson's correlation [corr] = 0.63, P = 2 × 10⁻⁴; regular fat, corr = 0.21, P = 0.3, preservation = 0.79). Genes in this module were enriched for ontology terms related to cell cycle (e.g. cell cycle, q = 1.29 × 10⁻⁴¹, cell division, q = 7.66 × 10⁻³⁸). These modules point to promising candidate genes that may mediate response to high-fat diet.

Gene Regulatory Divergence Between Mice From Divergent Climates

Understanding the genetic architecture of gene expression is critical to understanding the evolution of gene regulation (Signor and Nuzhdin 2018). Changes in gene expression can be a result of local regulatory divergence (cis) and/or distant regulatory changes (trans). To examine regulatory divergence in locally adapted house mouse populations, we crossed strains originating from a cold temperate region (SARA, SARB) and tropical/subtropical localities (GAIB, MANB) to create F1 progeny (GAIB × SARA, SARB × MANB; Fig. 3a). These progeny were subjected to the same experimental diet manipulations as described above. Body size measures were collected (supplementary table S3, Supplementary Material online, supplementary figs. S14 and S15, Supplementary Material online) and liver tissue was collected for mRNA sequencing. Expression measurements in F1s can be used to discriminate between cis- and trans-acting changes underlying gene expression differences (Fig. 3b) (Wittkopp et al. 2004). As F1 progeny inherit a copy of each chromosome from both parents, alleles from both parents are present in the same cellular environment. Consequently, differences in expression between the 2 parental alleles in an F1 (i.e. ASE, allele-specific expression) are indicative of a cis-regulatory change (Cowles et al. 2002; Wittkopp et al. 2004). In contrast, when differences in expression are observed between parents but not in F1s, we can infer that expression differences are due to trans-acting changes (Fig. 3b) (Wittkopp et al. 2004).

Fig. 3.

Gene expression regulatory divergence across strains and diet treatments. (a) Mice from different localities were crossed to produce F1 progeny. Parental strains and F1 progeny were raised on either a regular or high-fat diet. (b) A schematic showing how regulatory differences can be inferred via allele-specific expression. (c) Gene expression regulatory divergence between Florida (GAIB) and New York (SARA) (left) and New York (SARB) and Brazil (MANB) (right) was characterized through comparisons of allele-specific expression in F1 progeny and parental gene expression for female mice. Relative expression between strains (New York versus Florida and New York versus Brazil) is plotted against relative allelic expression in F1s. Each point indicates a gene colored by their regulatory classification. Regulatory divergence due to only cis changes will fall on the diagonal x = y line (red dots). Genes with no difference due to cis (trans-only divergence) will fall along the horizontal x = 0 line (purple dots). (d) Comparison of gene regulatory divergence between female mice on high-fat (“HIGH”) and regular (“REG”) diets. Comparison across diets revealed extensive environmental effects on gene regulation. For results in male mice, see supplementary fig. S17, Supplementary Material online.

Gene regulatory divergence was examined in 6,951 and 5,716 genes between SARB and MANB and GAIB and SARA, respectively (Fig. 3c) (see Methods). Between SARB and MANB, cis changes were found to be predominant over trans changes, consistent with previous intraspecific studies of house mice (Ballinger et al. 2023) (Fig. 3c, supplementary table S14, Supplementary Material online). In contrast, between GAIB and SARA, trans changes were typically found to be predominant (Fig. 3c; supplementary table S15, Supplementary Material online). As cis-regulatory changes have been found to accumulate with genetic distance in other systems (Coolon et al. 2014; Metzger et al. 2017), this difference may reflect greater divergence between the populations from different continents (SARB and MANB) versus populations both from North America (SARA and GAIB) (Gutiérrez-Guerrero et al. 2024). Effect sizes of cis changes were also found to be smaller on average between SARA and GAIB versus SARB and MANB, again potentially reflecting lower genetic divergence between these populations (Wilcoxon signed-rank tests, P-values < 0.01; Fig. 2a). In summary, both cis and trans changes contribute to regulatory variation among strains, but the relative contribution to gene expression differences varies between strains.

Gene Regulatory Divergence Between Populations is Diet-dependent

The environment can induce changes in gene expression via cis- or trans-acting mechanisms (Fig. 3b). The regulatory architecture of plasticity can affect how it evolves as well as the adaptive potential of populations (van Gestel and Weissing 2016; Kovuri et al. 2023). Trans regulators can modify the expression of hundreds of downstream genes and enable rapid responses to environmental perturbations. In contrast, cis-regulatory changes have local impacts and are expected to be less pleiotropic on average (Signor and Nuzhdin 2018). Therefore, either may be favored in the face of new environmental challenges. To understand the relationship between diet and gene regulatory divergence in this system, we compared our results between individuals fed a regular versus High-fat diet. Effect sizes of cis and trans effects on gene expression were correlated across diets, though more so for cis changes than trans changes (cis, Spearman's rho = 0.76 to 0.77; trans, rho = 0.56 to 0.67) (supplementary fig. S16, Supplementary Material online). However, we found that for many genes, gene regulatory divergence between strains was often dependent on diet. Between 32% and 45% of genes showed a different regulatory categorization between diet treatments for each comparison (Fig. 3d, supplementary fig. S17, Supplementary Material online). This suggests that gene regulatory divergence between strains is often context-specific. The greatest differences in regulatory categorization were observed between females from the GAIB and SARA cross, likely reflecting the marked response of SARA females to diet (see above, Fig. 2c). This cross also showed the lowest correlation of trans effect sizes across diets (Spearman's rho = 0.56), suggesting trans changes play an important role in high-fat diet response in SARA females. A greater proportion of genes showed evidence of regulatory divergence in cis and/or trans on a regular diet in most comparisons (Fisher's exact tests, SARB × MANB females, P = 0.25, all other comparisons P < 0.0003), suggesting a greater conservation of regulatory architecture between strains on a high-fat diet. Our results add to the growing body of studies demonstrating that gene regulatory divergence can differ substantially based on the environment in which expression is measured (Tirosh et al. 2009; Chen et al. 2015; Verta and Jones 2019; Ballinger et al. 2023).

Next, we asked about the relative contribution of cis versus trans changes to plastic responses. We found that cis-regulatory changes between strains were more robust to diet, whereas trans divergence was more responsive to diet differences. Gene expression differences ascribed to cis changes were more stable across diet treatments than trans changes in most comparisons (Fisher's exact tests, GAIB × SARA Females P = 0.07, all other crosses P < 0.0001) (Fig. 3d, supplementary fig. S17, Supplementary Material online). Effect size differences also reflected this, with diet having a smaller effect on cis compared with trans divergence between strains (Wilcoxon signed-rank tests, P-values < 7.27 × 10⁻¹¹³) (supplementary fig. S16, Supplementary Material online). The robustness of cis-acting changes compared with trans-acting changes to environmental variation has also been observed in other systems (Li et al. 2006; Smith and Kruglyak 2008; Tirosh et al. 2009; Chen et al. 2015). Therefore, evidence suggests that trans regulation may play a more important role in the evolution of environmentally induced changes in gene expression.

Cis-by-diet Interactions are Common and Related to Metabolism

Given evidence of diet dependance in gene regulation, we next asked whether cis-regulatory changes were context dependent. To identify specific genes with evidence for cis-by-diet effects on expression (Fig. 3b), we compared ASE ratios in F1 progeny across diet treatments (i.e. GAIB/SARA allele in high-fat versus GAIB/SARA allele in regular fat and MANB/SARB allele in high-fat versus MANB/SARB allele in regular fat). We identified 425 genes with a cis-by-diet interaction (see Methods, FDR < 0.05) (supplementary fig. S18, Supplementary Material online). The majority of cis-by-diet interactions were observed in only 1 cross or sex (∼81%), with only 3 genes showing cis-by-diet effects in all comparisons (Serpinc1, Fmo5, B2m). Consistent with a potential role for these changes in modulating responses to diet between strains, genes with cis-by-diet interactions were enriched for metabolic GO terms (e.g. fatty acid metabolic process, cellular response to cholesterol, regulation of bile acid secretion; FDR < 0.05) and phenotypes (e.g. abnormal triglyceride level, abnormal phospholipid level, abnormal food intake; FDR < 0.05) compared with all genes with ASE.

On average, cis-by-diet interactions were of small or modest effect that resulted in a change in the magnitude of ASE between lines (supplementary fig. S18, Supplementary Material online). However, we also identified cases where cis changes were specific to 1 diet (122 genes). We consider these cases of diet-induced ASE. Many of these genes have previously been linked to diet-induced metabolic phenotypes in either humans or mice (Supplementary online data). Genes with cis-by-diet interactions were also identified in gene co-expression modules associated with metabolic phenotypes. These genes are promising candidates for mediating strain-specific diet responses (Supplementary online data). For example, protein sterol carrier protein-2 (Scp2) is a hub gene in the module with the strongest positive association with body weight in males on a high-fat diet (green module, corr = 0.83). Scp2 was found to have a cis-by-diet interaction in the SARB × MANB cross and is significantly correlated with BMI and body weight variation in males in our data (regular diet, corr = 0.6, P = 0.0006; high-fat diet, corr =0.83, P = 8.73 × 10⁻⁸). Scp2 is involved in regulating lipid metabolism in the liver (Xu et al. 2022: 2) and has been implicated in nonalcoholic fatty liver disease (Xu et al. 2022). These data demonstrate that cis-by-diet interactions are common and that while they are typically small in magnitude, many are functionally related to metabolism, suggesting they may play an important role in response to diet.

Evidence for Polygenic Selection on Cis-regulatory Divergence Between Mice From Divergent Climates

Previous work in this system suggests cis-regulatory changes in particular may play an important role in environmental adaptation (Mack et al. 2018; Phifer-Rixey et al. 2018; Ballinger et al. 2023; Durkin et al. 2024). Gene expression adaptation via cis-regulatory mutations can be highly polygenic, involving many independent changes at functionally related genes (Bullard et al. 2010; Fraser et al. 2011, 2012). To test for evidence of selection on cis-regulatory changes between mice from divergent climates, we applied a gene-set test of selection based on a sign test framework (Bullard et al. 2010; Fraser et al. 2011). For a given pathway or biological function, we expect cis-regulatory changes to be unbiased with respect to their directionality (e.g. an equal number of cis changes upregulate SARB and MANB alleles). If, however, cis-regulatory changes act in the same direction more than expected, this can indicate lineage-specific selection on the cis-regulation of this gene set (Fig. 4a) (Bullard et al. 2010; Fraser et al. 2011; Fraser 2011). Applying the sign test to Gene Ontology (GO) gene sets in each cross, we identified multiple gene sets with evidence for biased directionally (full list in supplementary table S16, Supplementary Material online) (see Methods).

Fig. 4.

Evidence for polygenic cis-regulatory evolution. (a) Testing for polygenic cis-regulatory evolution (adapted from (Fraser et al. 2011)). Four functionally related and unlinked genes are shown for the MANB and SARB strain. Expression of each gene is denoted by the number of squiggly gray lines. An “X” indicates an independent cis-regulatory mutation that alters expression at a given gene. Under neutrality, we expect an equal number of cis-regulatory changes affecting expression to increase and to decrease expression, as shown in the top panel. However, if in comparing MANB and SARB lines, we see that all cis-regulatory mutations increase expression of the SARB alleles (as in the bottom panel), this would be consistent with lineage-specific selection for altered expression of this entire gene set. (b) Effect directions of ASE for genes related to cholesterol homeostasis (top) and fatty acid biosynthetic process (bottom). We observed a significant bias of upregulation of SARB-alleles (blue left-hand bars) versus. MANB-alleles (orange right-hand bars) for genes associated with these processes. ***P < 0.001, ***P < 0.01, N.S. = Not significant.

In the SARB × MANB cross, multiple GO gene sets related to metabolism (fatty acid biosynthetic process, Benjamini-Hochberg adjusted P (q) = 0.0058; cholesterol homeostasis, q = 0.079; retinol metabolic process, q = 0.079) showed significantly biased upregulation of New York (SARB) alleles compared with Brazil (MANB) alleles (Fig. 4b, supplementary table S16, Supplementary Material online and Supplementary online data). Genes associated with metabolism ontogeny terms include important regulators of body weight and fat accumulation (e.g. Fads2, Acsm1, Acsm3, Scd1) (Stoffel et al. 2014; Olga et al. 2021; Cazarin and Altman 2024). Polygenic selection on cis-regulatory variation related to metabolism is consistent with phenotypic differences in body mass and blood chemistry observed between mice from divergent climates (Phifer-Rixey et al. 2018; Gutiérrez-Guerrero et al. 2024). Other gene sets with evidence for biased directionality included terms related to immunity (e.g. adaptive immune response, q = 0.0058; antibacterial humoral response, q = 0.008) and cellular processes (e.g. regulation of apoptotic process, q = 6.21 × 10⁻⁵). In the GAIB × SARA cross, only 1 GO term (response to ethanol, q = 0.0046) was significant after correction for multiple testing. Importantly, these data provide evidence of lineage-specific selection on cis-regulatory changes especially related to metabolism.

Overlap Between Signatures of Natural Selection and Gene Expression Divergence

Another way to test for evidence of selection related to expression divergence is to investigate overlap between our results and genomic signatures of selection on nucleotide variation in populations of wild house mice in the Americas. To identify genomic signatures of environmental adaptation, we used previously generated population genomic data for individuals collected from populations in North and South America along a latitudinal transect (134 individuals) (Phifer-Rixey et al. 2018; Gutiérrez-Guerrero et al. 2024). Variants associated with latitude were identified using a latent factor mixed model (LFMM) (Frichot et al. 2013) accounting for population structure (as previously described Phifer-Rixey et al. 2018; Gutiérrez-Guerrero et al. 2024). Genomic outliers for associations with latitude for North American populations and South American populations were retained at a z-score > |3| for overlap analysis with the expression data (see Methods).

Differentially expressed genes between the Brazil (MANB) and New York, USA (SARA, SARB) strains overlapped outliers for genome-environment associations in North and South America (223 and 2,502 genes, respectively). Next, we focused on overlap between outliers and genes with evidence of cis-regulatory divergence based on ASE. We focus on these genes as ASE is indicative of local genetic variation affecting gene expression. We identified 549 ASE genes overlapping outlier variants. ASE genes containing genomic outliers were enriched for ontogeny terms related to metabolism (e.g. long-chain fatty acid metabolic process, q = 7.49 × 10⁻³; unsaturated fatty acid metabolic process, q = 4.51 × 10⁻²) and gene expression regulation (e.g. regulation of transcription by RNA polymerase II, q = 5.36 × 10⁻⁶) over the background of genes that could be tested for ASE. Included in this set are genes identified in our ASE sign test (see above, Supplementary online data). For example, several genes associated with the ontogeny term for “fatty acid biosynthetic process” also overlapped outlier loci (e.g. Acsm1, Elovl5, Elovl2) as did genes associated with “cholesterol homeostasis” (e.g. Lipc, Ttc39b, Nr1h4). In comparisons between New York and Florida strains, 246 genes with evidence for differential expression overlapped genomic outlier regions in North America, of which 59 were identified as having a cis-component underlying regulatory divergence. We found no enrichment for this set.

Highlighting the importance of examining gene expression under multiple environmental contexts, 110 genes overlapping outlier loci also showed cis-by-diet interactions. This set included genes with previously described gene-by-diet effects on metabolism (Table 2) and was also enriched for metabolic GO terms (e.g. fatty acid metabolic process, 6.29 × 10⁻³; lipid metabolic process 2.30 × 10⁻²) relative to other ASE genes overlapping genomic outliers (Supplementary online data). Differential gene expression and ASE can help make connections between candidates from genome scans and adaptive phenotypic variation (e.g. Jones et al. 2012; Mack et al. 2018, 2023a; Phifer-Rixey et al. 2018; Verta and Jones 2019; Ballinger et al. 2023). Incorporating environmental effects further ties candidates to phenotype by identifying induced differences in gene expression and cis-by-environment interactions. Therefore, this type of approach can be useful in investigating the genetic basis of adaptation in complex traits.

Table 2.

Outliers with cis-by-diet effects and diet-dependent effects on metabolism

Gene	Cis-by-diet effects observeda		Gene-by-diet effects on metabolism	Reference
Fmo5	GAIBxSARA SARBxMANB	F,M F,M	Knockouts (KOs) are protected against weight gain and reduced insulin sensitivity on a high-fat diet (HFD)	Scott et al. (2017)
G6pc1	SARBxMANB	F	KOs are resistant to developing obesity and diabetes on a HFD/high sucrose diet	Abdul-Wahed et al. (2014)
Agtr1a	SARBxMANB	M	KOs show attenuation of diet-induced body weight gain and adiposity on a HFD	Kouyama et al. (2005)
Ces1d	SARBxMANB	F,M	Adipose tissue-specific KOs gain more weight and show increased fat on a HFD	Li et al. (2022)
C3	GAIBxSARA	M	KOs affect metabolic traits (e.g. energy expenditure, glucose oxidation, fatty acid oxidation) on a HFD	Roy et al. (2008)
Elovl5	GAIBxSARA SARBxMANB	M F	Elevated activity attenuates hyperglycemia associated with HFD	Tripathy et al. (2010)
Rgs5	SARBxMANB	M	KOs exacerbated metabolic dysfunction and inflammatory state on a HFD	Deng et al. (2012)
Hmgcs2	SARBxMANB	F	KOs and overexpression are associated with diet-dependent metabolic dysfunction	Asif et al. (2022)
Sesn2	SARBxMANB	M	KOs exacerbate HFD-induced metabolic disorders	Zhang et al. (2024)
Ildr2	GAIBxSARA SARBxMANB	F,M M	KOs protects against obesity on a HFD	Chandra et al. (2022)
Enpp1	SARBxMANB	M	Overexpression in adipose tissue is associated with fatty liver and metabolic abnormalities on a HFD	Pan et al. (2011)
Pzp	SARBxMANB	M	–	–
Cth	SARBxMANB	F	Skeletal muscle KOs associated with obesity and insulin resistance on a HFD	Lu et al. (2024)
Cdh1	SARBxMANB	F	Conditional KOs show aggregated hepatic steatosis on a HFD	Chen et al. (2023a)

^aCross and sex in which cis-by-diet effect was observed.

Conclusions

The evolution of complex traits is inextricably tied to the environment. In this study, we used an experimental manipulation of diet in a new set of wild-derived inbred mouse strains from divergent climates to investigate the role of the environment and GxE interactions in shaping adaptive variation in complex traits. We found that there were significant effects of sex, strain, diet, and strain-by-diet interactions on body size and gene expression. Moreover, transcriptional response to diet was largely strain-specific, making GxE interactions the rule rather than the exception.

Using crosses between strains from differing climates, we were able to further strengthen ties between gene regulation, the environment, and adaptation. We identified both cis- and trans-acting changes that contributed to gene expression differences between strains from tropical and temperate regions of the Americas. Importantly, we found that gene regulatory divergence is highly dependent on environmental context, particularly for trans-acting changes. Higher sensitivity of trans divergence to diet suggests a greater role for trans regulation in gene expression plasticity (Li et al. 2006; Smith and Kruglyak 2008; Tirosh et al. 2009; Lovell et al. 2018; Ballinger et al. 2023; Chen et al. 2023b). In contrast, associations between metabolic phenotypes and gene co-expression modules were largely robust to changes in diet, highlighting the exceptions as modules containing candidate genes for phenotypic response to diet. We were further able to connect local variants to plasticity in gene expression and phenotype via identification of genes with cis-by-diet interactions. Those genes were also found in co-expression modules associated with variation in body mass related phenotypes and as a group enriched for metabolic GO terms. Notably, ASE sign tests provide evidence of selection on cis-regulatory changes related to metabolism, consistent with the differences in body mass and blood chemistry among these populations (Phifer-Rixey et al. 2018; Dumont et al. 2024). Overlap among outliers from selection scans and differentially expressed genes also helps connect candidate genes to adaptive variation in phenotype, in particular those with evidence of cis-regulatory divergence and functional ties to metabolism. Outlier loci with cis-by-diet interactions are strong candidates for adaptive plasticity.

More generally, our results underscore the value of incorporating environmental variation in evolutionary and biomedical studies of complex traits. Moreover, they add to growing evidence that genetic variation is crucial to studies of response to high-fat diet, a trait of considerable biomedical significance (e.g. Roy et al. 2021; Bachmann et al. 2022). Most of the strains included in this study are now available commercially, making them potentially useful for the study of diet-induced obesity (Dumont et al. 2024). Sex-specific transcriptional responses to diet also point to the importance of explicitly considering sex both in experimental design and in investigating underlying biological processes. Overall, this study represents a step forward in our understanding of adaptive body size variation in this system and demonstrates the importance of environmental variation and GxE interactions in disentangling the processes that underlie adaptive variation in complex traits.

Methods

Experimental Design

To evaluate variation in response to a high-fat diet, we examined 8 newly derived strains of M. m. domesticus. Five were derived from populations in the United States: TUCC (Tucson, AZ), GAIB (Gainesville, FL, USA) and SARA, SARB, and SARC (Saratoga Springs, NY, USA). The remainder were from Canada, EDME and EDMF (Edmonton, Alberta) and Brazil, MANB (Manaus, Amazonas). The strains were generated via brother-sister pairing (Phifer-Rixey et al. 2018; Dumont et al. 2024) for at least 12 generations. Importantly, these mice descend directly from wild-caught progenitors and were not rederived, a process which would be expected to alter their gut microbiome. However, several of these strains have now been rederived and are available commercially (Dumont et al. 2024). Mice were kept in common laboratory conditions at approximately 22 °C with a 14/10 light/dark cycle throughout the experiment. After weaning at ∼24 d, mice were housed singly with enrichment and provided with water ad libitum as well as 1 of 2 diets: high-fat (TestDiet BLUE 58Y1, 34.9% fat) or a breeder chow (LabDiet Extruded Picolab select mouse diet SVSM 50 IF/9F, 8% fat). At weaning and every week thereafter, animals were weighed (SPX222 Scout Analytical Balance, Ohaus, Parsippany, NJ, USA). Body length and tail length were measured via ruler at 1, 5, and 9 wk. In addition, at those time points, ∼120 grams of food were weighed, placed in a clean cage, and collected and re-weighed 24 h later (SPX222 Scout Analytical Balance, Ohaus). From these data, the amount of food consumed per hour was calculated. In some cases, food was weighed slightly earlier or later and, in those cases, the rate calculations were corrected for the difference. In addition, many mice ground food excessively, resulting in food in the cage that was clearly not consumed. In these cases, food intake data were not included in the analysis. After 12 wk, animals were euthanized. Body weight, body length, tail length, hindlimb length, and ear length were measured (SPX123 Scout Analytical Balance, Ohaus) and liver tissue was immediately collected for gene expression analyses. Livers were stored in RNAlater at 4 °C for ∼24 h and then transferred to a −80 °C freezer. All work was performed with approval from the Monmouth University Institutional Animal Care and Use Committee.

RNA was extracted from frozen liver tissues using the Qiagen RNeasy + Mini kits with tissue disruption via bead beating (Beadruptor 12, Omni International, Kennesaw, GA, USA). RNA samples were quantified (Qubit 2.0 Fluorometer; ThermoFisher Scientific, Waltham, MA, USA) and integrity was verified (4200 TapeStation; Agilent Technologies, Palo Alto, CA, USA). RNA quality was high (RIN_average = 9.90, sd = 0.25) and strand-specific libraries were prepared (Azenta, Inc., South Plainfield, NJ) using NEBNext Ultra II RNA Library prep kits (New England Biolabs, Ipswich, MA, USA). Sequencing libraries were validated (Agilent TapeStation) and quantified via Qubit as well as by quantitative PCR (KAPA Biosystems, Wilmington, MA, USA). For the initial run, the sequencing libraries were multiplexed and clustered onto 11 flow cells. After clustering, flow cells were loaded onto the Illumina HiSeq instrument and sequenced using a 2× 150 bp Paired End (PE) configuration. Raw sequence data (.bcl files) were converted into fastq files and de-multiplexed using Illumina bcl2fastq 2.20 software. One mismatch was allowed for index sequence identification. To augment these data, 2 additional lanes were sequenced using the same approach on the NovaSeq platform (2 × 150 bp PE). Reads sequenced per sample are available in supplementary table S17, Supplementary Material online (median = 62,289,663 PE reads).

Analysis of Morphological Data

To test whether aspects of body size varied among strains, we used linear models including relevant factors, interactions, and covariates (Table 1, supplementary tables S4 to S10, Supplementary Material online). The models were implemented in R via the aov function and evaluated via the Anova function choosing the Type III option (Chambers et al. 1992; Fox 2015). Tukey's tests for each factor were used to further clarify differences. Tests were implemented in R using the TukeyHSD function (Miller 1981; Yandell 1997). Because these strains are new, we also collected testis weights. Strain contributed significantly to variation in testis weight (supplementary table S9, Supplementary Material online), but this was largely driven by differences among EDMF males and the other strains (supplementary table S6, Supplementary Material online). Despite differences in design and conditions, for strains also reported in Dumont et al. 2024; MANB, SARA, SARB, SARC), average values for testis weight were broadly similar between the studies.

Read Mapping and Parental Expression Analysis

Reads were mapped to the Mus musculus reference genome (GRCm39) using STAR (v 2.7.11.a) (Dobin et al. 2013) with the Ensembl GRCm39 annotation. Reads overlapping exons were counted for gene-wise quantification using HTSeq (Anders et al. 2015). Raw read count data was transformed using variance stabilizing transformation to assess transcriptome-wide expression patterns via PCA. DESeq2 was used to assess differential expression by fitting a generalized linear model following a negative binomial distribution (Love et al. 2014). Gene expression was compared between lines and between diet treatments using a model that included population of origin, sex, and diet (Wald Test). A Benjamini–Hochberg multiple test correction was used on P-values. We considered genes with an FDR < 0.05 to be differentially expressed between comparisons.

Identifying Variants for ASE

To quantify ASE, we mapped reads from each F1 individual to the mouse reference genome with STAR. We utilized the Genome Analysis ToolKit (McKenna et al. 2010) to identify variants for measures of ASE. First, duplicates were marked with the Picard tool MarkDuplicates. Read groups were added with AddOrReplaceReadGroups. SplitNCigarReads was then used to split reads that contain Ns in their cigar string (e.g. spanning splice events). HaplotypeCaller and GenotypeGVCFs were utilized for joint genotyping across parental and F1 samples. SNP calls were filtered with VariantFiltration (QD < 2.0; QUAL < 30.0; FS > 200; ReadPosRankSum < −20.0). Variants were included for downstream analysis if all genotyped F1s were called as heterozygous and parental lines were assigned alternative homozygous calls. We used these calls to create a high-quality list of informative variants for ASE.

To identify reads overlapping informative sites, we mapped F1 reads to the mouse genome using the WASP implementation in STAR to mitigate mapping bias associated with the reference allele (van de Geijn et al. 2015; Asiimwe and Alexander 2024). WASP eliminates reads with potential for mapping bias by flagging reference-biased reads. First, reads containing SNPs are identified. Then, WASP simulates reads with alternative alleles at that locus and remaps reads to the reference. Reads that do not map to the same location are flagged (van de Geijn et al. 2015). Reads overlapping informative variant calls that passed WASP filtering were then retained to estimate ASE. HTSeq-count (Anders et al. 2015) was used to count reads associated with each gene for each parental allele separately.

To identify genes with evidence for ASE, reads that mapped preferentially to 1 parental allele in F1 progeny were compared with DESeq2 (Ballinger et al. 2023; Mack et al. 2023a, 2023b). Reads were fit to a model with allele, sample, and sex (Wald test, FDR < 0.05). For F1 samples, DESeq2 library size factor normalization was disabled as read counts came from the same sequencing library (setting: SizeFactors = 1). Previously published code is available on Github (github.com/malballinger/BallingerMack_PNAS_2023) for these analyses.

Characterizing Gene Regulatory Divergence

Expression in pure strains (“parental”) and F1s was compared with identify cis and trans divergence at each gene. Genes were partitioned into different regulatory categories by comparing allelic expression in F1s, expression differences in the parentals, and a comparison between allelic and parental ratios (McManus et al. 2010). In contrast, genes that do not show evidence of regulatory divergence based on differences between alleles in F1s or expression between parents are inferred to have conserved regulatory architecture between strains. To equalize power across comparisons, we randomly dropped individuals to maintain an equal number of parental and allelic samples in each comparison. Parental reads were then downsampled to match that of F1 libraries and all libraries were downsampled to equalize power across replicates (Coolon et al. 2014; Mack et al. 2016; Ballinger et al. 2023). Parental and allele-specific F1 counts for each replicate were summed and binomial exact tests were used to identify cis and trans divergence, as in previous studies (see below for method comparison) (Coolon et al. 2014; Lemmon et al. 2014; Mack et al. 2016; Ballinger et al. 2023). Fisher's exact tests were used to compare allelic and parental ratios. The resulting P-values from binomial and Fisher's exact tests were corrected following the Benjamini–Hochberg method and considered significant at an FDR < 0.05. Genes were then sorted into regulatory categories using previously described criteria (McManus et al. 2010; Goncalves et al. 2012; Coolon et al. 2014; Mack et al. 2016; Ballinger et al. 2023). In brief, genes were sorted as follows:

• Cis only: (1) a significant difference in expression between populations, (2) a significant difference between alleles in F1s, (3) no significant difference between parental and allelic ratios.

• Cis + trans: (1) a significant difference in expression between populations, (2) a significant difference between alleles in F1s, (3) a significant difference between parental and allelic ratios.

• Trans only: (1) a significant difference in expression between populations, (2) no significant difference between alleles in F1s, (3) a significant difference between parental and allelic ratios.

• Conserved or Ambiguous: All other patterns.

The downsampling and pooling of replicates was chosen as the approach to categorize genes based on cis and trans divergence due to comparisons with unequal read depth and replication (Coolon et al. 2014). As in previous studies (Ballinger et al. 2023), we saw congruence between the use of binomial tests and DESeq2 for identifying cis-regulatory divergence via ASE (DESeq2 Wald test, see above). P-values from DESeq2 and binomial tests were found to be highly correlated (Spearman's rho > 0.79, P-values < 2.2 × 10⁻¹⁶ for all pairwise comparisons).

Comparing Liver Expression Responses to Diet Between Classical and Wild-derived Inbred Strains

Previously published RNA-seq data from Bachmann et al. (2022) for 9 strains was downloaded from the GEO expression database under the accession GEO:GSE182668. Reads were mapped to the M. musculus reference genome (GRCm39) and reads overlapping exons were counted as described above. Count data was then analyzed in DESeq2 using a model incorporating sex, strain, and diet, as described above. We note that differences in experimental approach and other potential batch or technical effects (e.g. sample preparation, sequencing chemistry) likely affect overlap between genes identified between our studies. Experimental differences between the 2 studies include the following: (1) time and age at which mice are put on a high-fat diet (8 to 21 wk in Bachmann et al. (2022)), (2) age at which livers are collected (>21 wk versus ∼15 wk).

Weighted Gene Co-expression Analysis

To identify co-expression gene sets and associate expression variation with phenotypic variation across strains, we used WGCNA. This approach has been used previously to associate gene expression variation with variation in complex traits and adaptive phenotypes (e.g. Velotta et al. 2016, 2020; Campbell-Staton et al. 2018). We carried out a WGCNA on normalized expression data following WGCNA protocols (Langfelder and Horvath 2008, 2012). We filtered parental data for genes with low expression across samples (>20 reads per sample in at least 4 samples), resulting in 15,622 genes for analysis. The phenotypes of growth rate, weight, body length, and BMI were correlated (see supplementary table S12, Supplementary Material online, Pearson's Correlation 0.39 to 0.84). However, as correlations between traits varied, we explored the relationship between traits and co-expression modules individually.

We constructed a gene co-expression network, represented by an adjacency matrix, which denotes co-expression similarity between pairs of genes among different individuals. Modules were identified using unsupervised clustering. Dissimilarity between clusters is measured based on the topological overlap and defined by cutting branches off the dendrogram (Zhang and Horvath 2005; Langfelder and Horvath 2008). Soft-thresholding power was chosen based on the pickSoftThreshold function in WGCNA to achieve an approximately scale free topology. Expression modules were inferred for regular and high-fat treatment and male and female mice separately, and then consensus modules were created to identify co-expression patterns shared across groups. We chose a minimum module size of 30 genes. To associate modules with traits of interest, we then tested for correlations between eigengenes (the first PC of a module) and each trait (Pearson correlation). In order to explore the preservation of module trait relationships in regular and high-fat diet treatments, we analyzed the differential eigengene network (supplementary figs. S12 and S13, Supplementary Material online). Plots of differential analysis were created with the command plotEigengeneNetworks. The density (D) of the eigengene network is defined as the average scaled connectivity. Larger values of D (closer to 1) are indicative of stronger correlation preservation between all pairs of eigengenes across the high-fat and regular networks. Gene module membership and trait associations from our consensus WGCNA analysis can be found on FigShare (https://figshare.com/s/721a12ffbe1bfa3b921c).

ASE Sign Test

GO categories were obtained through Ensembl annotations for mice (GRCm39). We restricted our analysis to GO Biological Process terms. For each cohort separately (e.g. cross [GAIBxSARA or SARBxMANB], diet-treatment [high-fat, regular] and sex [males, females]), genes with significant evidence for ASE were divided based on which allele was upregulated (SARA/B or MANB/GAIB). GO Biological Process terms with fewer than 10 members in a cohort were excluded from the analysis. A Fisher's exact test was used to identify lineage-specific bias on each set (Fraser et al. 2011; Mack et al. 2016, 2023a).

P-values for each gene set in a cohort (sex, diet) were combined using Fisher's method to obtain a single P-value for each gene set in a cross (GAIBxSARA or SARBxMANB), as previously described (Fraser et al. 2011; Mack et al. 2023a). FDR was estimated in 2 ways, via (1) the Benjamini-Hochberg Procedure implemented in R with p.adjust, and (2) gene category assignments permutations (Fraser et al. 2011; Artieri et al. 2017; Mack et al. 2023a). For each cohort, gene-GO category assignments were randomly shuffled and the test was repeated 1,000 times. P-values were determined by asking how often a result of equal or greater significance was observed in the permuted set (equivalent to a GO category-specific FDR (Artieri et al. 2017)). Gene sets for which significant biased directionality in both tests are reported in supplementary table S16, Supplementary Material online.

Overlaps With Selection Scans

LFMM (Frichot et al. 2013) software was used to identify associations between genetic variants and latitude as previously described (Phifer-Rixey et al. 2018; Gutiérrez-Guerrero et al. 2024). In brief, LFMM was run separately for individuals from east coast populations in the United States (Phifer-Rixey et al. 2018) and for South American populations (burn-in = 100,000 and 500,000 iterations) (Gutiérrez-Guerrero et al. 2024) along 2 latitudinal transects. K values of 2, and 3 were used for North America and South America, respectively. For each run, the genomic inflation factor was estimated (λ) and P-values were corrected for false-discovery rate. We use outliers at a |z-score| threshold of >3 to examine overlap with expression data.

Enrichment Analyses

GO and pathway enrichment were performed with PANTHER (Thomas et al. 2003). Phenotype enrichment analyses were performed with ModPhea (Weng and Liao 2017).

Supplementary Material

Supplementary material is available at Molecular Biology and Evolution online.

Funding

Research was supported by internal funding from Monmouth University and Drexel University. M.P.R. is supported by NSF Division of Environmental Biology Award #2332998. N.P.L., T.C.L., S.A.V., and L.A.C. were supported by the summer research program at Monmouth University. K.L.M. is funded by the National Institute of General Medical Sciences of the National Institutes of Health under Award Number P30 GM145499 and R35GM154966. This work used Expanse at the San Deigo Supercomputer Center through allocation BIO230113 from the Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support (ACCESS) program, which is supported by grants #2138259, #2138286, #2138307, #2137603, and #2138296 from the National Science Foundation.

Data Availability

All raw sequence data is available online through NCBI Bioproject (PRJNA1164275). Supplemental online datasets are available through FigShare (https://figshare.com/s/721a12ffbe1bfa3b921c).