Generating the Nachman Panel

Wild house mice were caught in 2012–2013 from five geographic locations (Fig 1). Animals were transported to UC Berkeley, and mice from each geographic region were randomly paired to create inbred lines which were propagated through brother-sister mating for at least 10 generations. Initially, ~10 independent lines were established from each location. No attempts were made to rescue lines that exhibited infertility due to inbreeding depression and half of the initiated lines eventually became extinct. At the time of writing, 25 of the initiated lines remain in Dr. Nachman’s strain holdings at UC Berkeley.

A subset of 20 strains were selected for importation to the JAX Repository (S2 Table). Of these strains, 11 were successfully rederived via in vitro fertilization (IVF) and embryo transfer to a pseudopregnant dam and integrated into production breeding colonies (Fig 1). Reasons for rederivation failure were complex and variable, ranging from poor breeding, failure to recover sufficient numbers of oocytes for IVF, no live births following multiple embryo transfers, and accidental strain contamination (S2 Table). One successfully imported strain (GAIA/NachJ) was subsequently terminated due to poor breeding performance.

Breeding performance

Many wild-derived inbred strains breed poorly, a consideration that limits their practical utility in biomedical research. Our strategic propagation of only the best performing inbreeding lineages derived from each wild-caught founder pair should ensure that resulting inbred strains are vigorous breeders. Indeed, the inbred Nachman lines breed reliably across two independent animal facilities (Fig 2). Fewer than 25% of established matings are non-productive and average weaned litter sizes for most strains are between 3–6 pups.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Distribution of weaned litter sizes across 10 strains in the Nachman WDIS panel.

Sizes of litters born at JAX and in the Nachman Lab at UC Berkeley are presented as standard box plots, with box width defining the inter-quartile range and the thick line bisecting each box denoting the median. Litter sizes are jittered for ease of visualization.

https://doi.org/10.1371/journal.pgen.1011228.g002

While comprehensive diallele crosses have not yet been performed, there are no signs of F1 sterility or reduced fertility in the few F1 hybrids between independent Nachman strains that have been generated to date (S3 Table). To the contrary, testis weight and sperm density measurements in F1 hybrids exceed fertility metrics quantified in the inbred parental lines and classical inbred strains, revealing hybrid vigor (S1 Fig and S3 Table). Efforts to more comprehensively profile reproductive traits in the inbred Nachman strains and their derivative F1 hybrids are on-going, and we expect our future results to strengthen empirical support for the early trends reported here. We note that these preliminary findings fall in contrast to observations from diallele crosses between the inbred founder strains of the DO and CC [34], which include strains from reproductively isolated subspecies. Many incipient inbred CC strains were lost due to male infertility [35] and many surviving strains exhibit poor breeding performance, male infertility, and extreme sex ratio distortion [36]. Thus, strains in the Nachman WDIS panel comprise a shelf-stable mouse diversity resource with limited potential to uncover genetic incompatibilities in experimental crosses between Nachman strains or between Nachman strains and CI strains.

Cytogenetic characterization of Nachman Strain Karyotypes

Wild M. m. domesticus populations harbor frequent Robertsonian chromosomal translocations that give rise to considerable karyotypic diversity across Europe [37]. Hybrid mice from crosses between different karyotypic races exhibit reduced fitness owing to altered chromosome dynamics and inefficient chromosome segregation during meiosis [37–40]. To confirm the absence of large-scale karyotypic alterations or changes in chromosome number among strains, we generated spermatoctye cell spreads from each inbred Nachman line and assessed karyotype and meiotic chromosome pairing by fluorescent microscopy (S2 Fig). These cytogenetic analyses are on-going, but all Nachman strains evaluated to date exhibit the standard 2n = 40 all-acrocentric karyotype (7 tested strains: EDMB, GAIC, SARA, SARB, MANB, MANF, and TUCB).

Genomic diversity in the Nachman Panel

To assess the extent of genomic diversity in this new strain resource, we sequenced the whole genomes of a representative male from each strain to moderate coverage using PacBio HiFi sequencing technology (~10x coverage per strain; S4 Table). Average read lengths exceeded 10kb for all strains (range: 10.217–14.957kb), with >96.67% of sequenced bases exceeding a quality score of 30 (S4 Table). Sequenced reads were mapped to the GRCm39 reference genome assembly [41] and subject to single nucleotide variant (SNV) calling using DeepVariant [42, 43]. Each strain harbors between 4.73–6.54 million fixed SNV differences relative to the C57BL/6J-based reference, with >2.75M SNVs distinguishing any pair of Nachman strains (Fig 3A; corresponding to a minimum of ~1 SNP every ~1 kb). Although sequenced mice have undergone a modest number of inbreeding generations, within strain heterozygosity is low (π < 0.0006 versus ~0.0017 in wild-caught M. m. domesticus) and aligns with theoretical expectations for the number of inbreeding generations (S3 Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Genetic diversity in the Nachman strains.

(A) Heatmap displaying the number of pairwise SNP differences between Nachman lines. Counts derive from the autosomal genome fraction only and exclude unplaced contigs. (B) PC analysis of autosomal genetic diversity partitions the Nachman lines by sample location and isolates Nachman strains from the CI mouse strains. (C) Maximum likelihood phylogenetic tree constructed from SNPs on chr19. The CI strains form a single clade nested within the diversity sampled by the Nachman strains.

https://doi.org/10.1371/journal.pgen.1011228.g003

We performed joint SNV calling with the Nachman lines and 51 CI mouse strains previously sequenced by the Mouse Genomes Project (44). Of the 16,071,877 autosomal SNVs observed in the Nachman strains, 6,836,536 variants (42.5%) are absent from CI strains [44]. We constructed a maximum likelihood phylogenetic tree from SNPs on chr19 to assess relationships between existing mouse strains and strains in this new wild-derived inbred strain resource. The CI strains present as a single clade nested within the diversity sampled in the Nachman panel (Fig 3C). Branch lengths are notably longer for the WDIS than for the CI strains, reflecting the greater diversity in the former. These findings are confirmed by a principal component (PC) analysis, which reveals a single cluster of points corresponding to the CI strains (Fig 3B), with the Nachman strains and other WDIS well-separated in PC coordinate space. However, we acknowledge that differences in the technologies used to sequence the Nachman strains (PacBio HiFi) and other strains (Illumina) could partially contribute to the isolation of the Nachman strains along these PC axes. Nachman strains derived from a common locale are more genetically similar than strains from divergent locations, although strains from Gainesville, Edmonton, and Saratoga Springs are minimally separated along PC1-PC5 (41.53% of the variance). PC dimensions 6–8 provide separation across these geographic sample locations (S4 Fig).

We next annotated variants according to likely functional impact. Variant counts for different functional categories are provided in S5 Table. Overall, the Nachman strain panel harbors 1,976 SNPs predicted to be highly deleterious, with 1,319 of these variants not observed in classical inbred laboratory strains. These highly deleterious variants are enriched in genes with biological roles in sensory perception and G protein-coupled receptor signaling (S5 Table). The Nachman strains also harbor predicted loss-of-function alleles at genes implicated in human diseases. For example, the TUCA/NachJ and TUCB/NachJ lines harbor a premature stop variant in Eif2b1 that is a predicted target of nonsense-mediated decay (NMD; chr5:124716942). Eif2b1 encodes a subunit of the eukaryotic translation initiation factor eIF2B, which is essential for protein synthesis. Mutations in this gene have been linked to leukoencephalopathy with vanishing white matter and ovarian failure in humans [45,46]. Similarly, a predicted NMD variant in Yeats2 (chr16:20028820, stop-gain mutation) is present in GAIC/NachJ, TUCA/NachJ, and TUCB/NachJ. Mutations in this gene are causally associated with myoclonic epilepsy in humans [47]. Strains MANB/NachJ, MANF/NachJ, TUCA/NachJ, and TUCB/NachJ carry a predicted loss-of-function splice-donor variant in Ifnar1 (chr16:91302893). Humans with mutations in this gene have immunodeficiency-106 and exhibit hyperinflammatory responses to some vaccines [48].

Taken together, our analyses indicate that the Nachman strains harbor considerable genetic diversity that is not captured in existing inbred mouse strain panels. A subset of this variation is likely functional, establishing the prediction of broad phenotypic diversity among these strains and their phenotypic divergence from CI mouse strains.

Structural diversity in the Nachman Panel

Structural variants (SVs) are important contributors to phenotypic diversity, including disease risk and incidence. Prior work has suggested that house mouse genomes are burdened by higher rates of structural mutation than human genomes [49,50], leading us to posit the presence of abundant, potentially functional SVs within the Nachman strain genomes.

Using an ensemble approach to minimize false positive SV calls (see Methods), we identified 274,177 autosomal structural variants across the 11 sequenced Nachman strain genomes (139,931 deletions, 133,058 insertions, 710 duplications, and 478 inversions). Of the 139,931 deletions discovered in the Nachman strains, 80,222 were not previously detected in a diverse panel of inbred strains (57.3%; requiring 75% reciprocal overlap [49]). Similarly, 60.6% of insertions ascertained in the Nachman panel are not present in the current mouse SV catalog [49] (n = 80,568 unique insertions in the Nachman lines). Nachman strain genomes each contain 13.5 Mb—18.3 Mb of sequence that is absent in the mm39 C57BL/6J-based reference genome (Fig 4A), with 23.5 Mb—30.6 Mb of sequence in the reference genome absent from any given Nachman strain (Fig 4B). The modest genome coverage of the Nachman lines may lead to an appreciable number of missed SVs (i.e., false negatives) in these genomes, implying that SVs potentially have a greater collective impact on these genomes than suggested by the numbers presented here.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4.

Numbers of bases impacted by (A) insertion and (B) deletion events in each Nachman strain relative to the mm39 reference sequence.

https://doi.org/10.1371/journal.pgen.1011228.g004

Many SVs in the Nachman strain genomes are potentially functional. Approximately 46.2% of SVs overlap RefSeq genes (126,741 SVs overlapping 18,504 unique genes), with 89,757 SVs overlapping the annotated coding regions of 12,209 unique genes. An additional 933 SVs lead to the ablation of whole transcripts (S6 Table). The majority of these transcript ablating SVs impact predicted genes or pseudogenes, although several olfactory receptors, vomeronasal receptors, and immunoglobulins harbor loss-of-function structural mutations.

Recent work has suggested that a majority of SVs in mouse genomes are due to transposable element (TE) activity [49]. We theorized that many SVs in the Nachman strains are likewise mediated by TE-activity in the genome. Indeed, the size distribution of insertion and deletion calls in the Nachman strain call set is consistent with key contributions from TEs (S5 Fig), with peaks corresponding to the average length of SINEs and LINEs. To more formally assess this possibility, we annotated all insertion and deletion SV calls using repeatMasker (see Methods). Overall, we identify 113.7 Mb of SV-associated sequence in the Nachman strains comprised of TE-derived sequence (57.01Mb deletion, 56.70Mb insertion), corresponding to 78.1% of SV-impacted bases (145.76Mb). SINE B2 MM1A repeats are the most abundant TEs within SVs (15,169 polymorphic MM1A elements in the Nachman panel; S7 Table), consistent with prior reports of active SINE B2 transposition in house mouse genomes [49,51]. RLTR10, IAPEz, and L1MdGf elements are also highly abundant in SVs identified in the Nachman lines (S7 Table), consistent with the young age of these elements and their active mobilization in mouse genomes [49,51,52]. Overall, our results suggest that TEs have substantially contributed to the landscape of SVs in the inbred Nachman strains and, by extension, the wild mouse populations from which these strains derive.

Strain subspecies ancestry

An earlier investigation of wild-caught mice from Tucson, Arizona uncovered evidence of introgression from M. m. castaneus [30]. We used a sample of 134 wild-caught mouse genomes from M. m castaneus, M. m. domesticus, and the outgroup M. spretus to evaluate genomic evidence of possible M. m. castaneus introgression in the TUCA/NachJ and TUCB/NachJ lines (S8 Table). Patterson’s D is significantly non-zero for both strains (D_TUCA = 0.213 and D_TUCB = 0.192; P < 2.3 x 10⁻¹⁶ for both strains), with estimated M. m. castaneus admixture proportions (f₄ ratio) of 13.4% and 11.3%, respectively. D is also significantly greater than zero for SARB/NachJ and SARC/NachJ (P < 0.05; S9 Table). However, the estimated admixture proportion from M. m. castaneus is <1% in these strains and may be attributable to the incomplete sampling of ancestral wild mouse diversity. We conclude that the genomes of both Tucson lines, but not strains from other locations, harbor significant M. m. castaneus ancestry. We also investigated possible introgression from M. m. musculus and found that all Nachman strains have <1% admixture from M. m. musculus, indicating no significant or recent introgression from this subspecies (S9 Table).

We next estimated admixture statistics in windows of 5000 informative SNPs (2500 slide) across the genomes of TUCA/NachJ and TUCB/NachJ, focusing on the 5% of windows with the highest f_dM values to identify regions of likely M. m. castaneus introgression. We utilize f_dM, rather than the related D statistic, as the variance in D can be quite large when applied to small windows [53,54]. Introgressed regions are overwhelmingly unique to either TUCA/NachJ or TUCB/NachJ (S10 and S11 Tables and S6 Fig), consistent with the high genomic divergence between these lines (Fig 3). For strain TUCA/NachJ, genes in introgressed regions are enriched for biological processes related to regulation of lipopolysaccharide-mediated signaling, zymogen activation, and sensory perception of taste and smell (S12 Table). For strain TUCB/NachJ, genes in regions of M. m. castaneus introgression are enriched for biological processes related to testosterone biosynthesis, keratinization, and multiple aspects of immunity (S13 Table). Overall, introgressed regions are short, implying that events are not recent (S6 Fig).

Relationship to wild mouse diversity

To contextualize the variation present in the inbred Nachman strains with that observed in wild M. m. domesticus, we created a joint variant callset featuring the 11 inbred Nachman strains and 97 publicly available wild M. m. domesticus samples from multiple populations (Iran, France, Germany, and the Eastern United States). PC analysis on this Nachman-wild mouse call set reveals genetic clustering of the Nachman lines with samples from Europe and the US, with samples from Iran isolated along PC1 (13.9% of the variance; Fig 5A). These trends are recapitulated by a maximum likelihood phylogenetic tree of these samples, which places the Iranian samples as ancestral to the Nachman strains and wild-caught mice from Europe and the United States (Fig 5C). Our findings are consistent with previous population genetic analyses of wild mice, which suggest that M. m. domesticus from the Indo-Iranian valley harbor elevated genetic diversity [11,55] and point to the European origin of North American house mice [56].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Genetic diversity in Nachman strains and wild-caught M. m. domesticus.

(A) Principal component analysis performed on autosomal variants segregating in wild-caught M. m. domesticus mice from multiple populations and inbred Nachman strains. PC1 separates wild-caught mice from Iran and all other mice. PC2 stratifies mice from within Iran. (B) Excluding mice from Iran provides increased granularity to detect differences across other M. m. domesticus populations. PC2 isolates the Nachman strains from Tucson, Arizona from other strains and populations. (C) Maximum likelihood tree constructed from biallelic chr19 SNPs. Strains are color-coded according to the legends in A and B.

https://doi.org/10.1371/journal.pgen.1011228.g005

We excluded the Iranian samples and repeated the PCA to evaluate how the genetic diversity captured in the Nachman samples compares to the genetic variation in contemporary M. m. domesticus from Europe and the Eastern US (Fig 5B). PC1 (12.7% of the variance) stratifies mice from wild-caught M. m. domesticus populations, with Nachman lines falling at intermediate positions along this PC. PC2 (10.2% of the variance) isolates the two Nachman strains from Tucson, likely reflecting the presence of M. m. castaneus admixture in these lines. While the Nachman lines introduce significant new variation into laboratory strain collections, the 11 inbred strains in this panel sample only a subset of the genetic variation present in wild M. m. domesticus populations.

Nachman wild-derived inbred strains capture extensive phenotypic diversity

We subjected mice from the majority of JAX imported Nachman lines to a 19-week phenotyping pipeline to profile strain variation in multiple metabolic, neurobehavioral, physiological, morphological and biochemical traits (Fig 6). Males and females from 9 strains were phenotyped across 16 cohorts of age-matched animals (+/-6 days). Cohorts were comprised of mice from multiple strains, and the majority of phenotyping cohorts included C57BL/6J control mice to permit post-hoc detection of potential batch effects. The strain composition of each cohort is provided in S14 Table. Overall, we collected 1119 phenotype measures from each animal, although many trait values are highly correlated and therefore not independent (S15 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Schematic of the phenotyping pipeline.

Mice were transferred to the Center for Biometric Analysis (CBA) at JAX at 4 weeks of age. Nuclear magnetic resonance (NMR) was used to assess body composition at 4, 8, 12, 16, and 19 weeks. From weeks 9–13, animals were subject to a series of neurobehavioral testing paradigms, including light-dark test, open field, spontaneous alternation with Y-maze, and voluntary wheel running. From weeks 16–17, animals were subject to 5-day indirect calorimetry trials and intraperitoneal glucose tolerance testing (GTT). At week 18, mice underwent an unconscious EKG to assess cardiac rhythm and function. Mice exited the phenotyping pipeline at 19 weeks. Blood and plasma samples were used to assess multiple biochemical traits and multiple organs were harvested and weighed. A subset of dissected tissues were frozen for future molecular analysis and others were paraffin embedded for future histological study.

https://doi.org/10.1371/journal.pgen.1011228.g006

More than 90% of the 1119 surveyed phenotypes differ among Nachman strains, with 86.7% and 80.4% of phenotypes differing significantly in comparisons involving only females or males, respectively (Kruskal-Wallis test, P<0.05; S16 Table). We obtain qualitatively identical results using both one-way ANOVA (S17 Table) and a linear modelling approach with strain treated as a random factor (73.8% of models including strain as a random effect variable provide significantly better model fit than a reduced model excluding strain; see Methods; S18 and S19 Tables). Approximately 25% (n = 288) of phenotypes exhibit significant differences between males and females and a significant strain-by-sex effect is observed for 21.5% (n = 235) of surveyed measures (two-way fixed effects ANOVA, P < 0.05; S20 Table). Similar trends are again observed using a linear mixed effects model comparison strategy: 26.6% and 8.5% of models including sex and interaction terms, respectively, provide improved fit compared to simpler models that exclude these effects (S18 and S19 Tables). Thus, phenotypic variation is ubiquitous across the Nachman strain panel, and many traits vary between sexes in a strain-dependent manner.

The wild-caught mice used to establish the Nachman lines were subject to unique, location-dependent selective pressures in their native environments. Consistent with this legacy of environmental adaptation, we observe systematic phenotype differences across strains as a function of geographic origin (S21 and S22 Tables). Location is a significant factor accounting for variation in 85.16% of phenotypes (Kruskal-Wallis test, P<0.05; n = 930/1092; S21 Table). We obtain similar results when fitting linear mixed effect models: 61.2% of models that include geographic sample origin as a random effect provide a better fit compared to models excluding this term (S23 Table). For the majority of surveyed traits, strains derived from a common location are more phenotypically similar to each other than strains from distinct locations, echoing trends in sequence-level similarity among strains (Fig 3).

To visualize the extent of phenotypic diversity across the Nachman panel, we performed a PC analysis of all surveyed phenotypes. The spatial distribution of strains along the first two PCs (26.4% of variance) affirms model-based findings above, revealing loose clusters based on strain geography and differences between the sexes (S7 Fig). While strains are more tightly clustered in the genotypic PCA than the phenotypic PCA, there is significant similarity between the two PC matrices (Similarity of Matrices Index = 0.709; P < 0.001; S7 Fig). This correspondence with the genotypic PC matrix would seem to suggest that the surveyed phenotypes are subject to polygenic control in the Nachman strains.

Below, we highlight results from each phenotype testing paradigm, present estimates of broad sense heritability, and compare the phenotypic variance among the Nachman samples to that present across classical inbred mouse strain panels. We then present results from correlation analyses that cut across multiple phenotype domains. Comprehensive phenotype data are available in S24 Table. Strain-level and Strain x Sex-level phenotype means are provided in S25 Table. Broad-sense heritability estimates are provided in S17 Table. Results from non-parametric Kruskal-Wallis tests of strain effects on each phenotype are provided in S16 Table. Two-way ANOVA results (Phenotype ~ Strain * Sex) are presented in S20 Table. Results from linear mixed effects model fitting are provided in S18 and S19 Tables. Results from Kruskall-Wallis tests, one-way ANOVA, and linear mixed effects models testing geographic origin as an explanatory variable (as opposed to strain) are provided in S21, S22, and S23 Tables, respectively.

Body composition analysis by NMR.

We assessed multiple metrics of body composition (body weight, fat mass, lean mass, water mass) via NMR at 5 timepoints between 4 and 19 weeks of age. As expected, all mice gained weight during this period, although the percentage increase in body mass differed among strains across this 15-week interval (range: 28.1% SARC males to 95.92% in GAIC/NachJ males). Males trended toward larger body mass than females (average sex dimorphism across strains at 19 weeks: 4.45 g; Fig 7), and all Nachman lines are smaller than C57BL/6J control mice. The broad sense heritability of body weight at 19 weeks is ~0.8 in both males and females (S17 Table), indicating that significant proportion of strain variation in body mass is attributable to underlying genetic differences between strains.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Changes in body composition over time as assessed by NMR.

Points correspond to strain-level means and data are partitioned by sex to highlight the degree of sex-dimorphism in body composition. Strains are color-coded by geographic origin, with strains from a common location distinguished by line type.

https://doi.org/10.1371/journal.pgen.1011228.g007

We similarly document significant strain and sex differences in body fat mass, lean mass, and water mass (S16, S17, and S20 Tables and Fig 7). Body composition phenotypes are highly correlated over time within strains (average Spearman’s rho = 0.669; S26 Table), implying that increased body mass is driven by coordinated changes in fat, lean, and water mass rather than isolated changes in one compositional component. Body composition measures are also correlated across strains at individual timepoints (all Spearman’s rho > 0.3; S7 Fig), consistent with the observation of near parallel growth trajectories across strains (Fig 7).

Home cage wheel running.

At week 13, mice were subjected to a 3-day home cage wheel running assay to assess overall activity and circadian rhythms. Strains vary 6.8-fold in total distance run over the 60h trial, from a low of 9,112 m in MANE/NachJ to 62,019 m in SARC/NachJ (S25 Table). Mice are nocturnal and, as expected, activity was highest during nighttime hours for all strains (Wilcoxon Rank Sum Test, P < 0.0005; Fig 8A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 8. Strain variation in voluntary exercise and circadian rhythm.

(A) The average number of wheel revolutions per strain over a 60-hour trial period. (B) Hour-to-hour change in wheel running activity (number of wheel revolutions). Abrupt changes in activity correspond to transitions between sleep-wake cycles. In both figures, strains are color-coded by geographic origin, with strains from the same location further distinguished by line type.

https://doi.org/10.1371/journal.pgen.1011228.g008

The Nachman strains derive from wild-caught mice exposed to distinct photoperiods in the wild. We sought to determine whether strains differ in the timing of their peak wheel running as a function of geographic origin. To this end, we computed the change in the number of wheel revolutions per hour across the full trial, defining the transition between day and night as the point exhibiting the most extreme difference in activity level. All strains commence nighttime levels of activity at ~18:00h, but the onset of daytime rest periods is considerably more variable, with strain differences in the duration of nighttime exercise and daytime activity. For example, mice from Manaus, Brazil exhibit a shorter duration and reduced levels of nighttime activity compared to strains from Saratoga Springs, New York, with mice from the latter strain remaining active during much of the day (Fig 8B).

Indirect calorimetry.

Mice were subject to continuous, high-definition respiratory monitoring in Promethion Core cages for 5 days, allowing estimation of energy expenditure, activity levels, and food consumption. We observe significant strain effects on all surveyed phenotypes (P < 0.01), with approximately half of the metabolic traits captured in this assay also exhibiting significant differences between the sexes (S16, S17, and S20 Tables). Both sexes of all strains were most active, exhibited highest energy expenditure, and consumed the most food during the dark cycle (Fig 9A–9C; Paired Wilcoxon Signed Rank Test, all P < 10⁻¹⁰). Most metabolic traits are highly heritable (S17 Table), nominating the Nachman lines as excellent models for genetic studies of metabolism. Intriguingly, despite their close genetic affinity, the two strains from Gainesville, Florida exhibited the lowest (GAIC/NachJ; 7.53 kcal/24h) and highest (GAIA/NachJ; 9.98 kcal/24h) total energy expenditure, with C57BL/6J controls presenting an intermediate phenotype (9.67 kcal/24h (Fig 9D).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 9. Strain differences in metabolic phenotypes quantified by continuous respiratory monitoring.

(A) All strains exhibit higher activity during the night, although the extent of the nighttime activity bias varies across strains. (B) Similarly, all strains expend more energy and (C) consume more food at night, but the ratio of day:night energy expenditure and food intake varies across strains. (D) Total energy expenditure varies significantly across strains, with the two strains from Gainesville, Florida (GAIC/NachJ and GAIA/NachJ) delimiting the extremes.

https://doi.org/10.1371/journal.pgen.1011228.g009

Glucose tolerance test.

All Nachman lines show an attenuated response to intraperitoneal injection of a controlled concentration of glucose relative to C57BL/6J (Fig 10). Peak blood glucose concentrations are higher in males than females for all strains, with exceptionally pronounced dimorphisms in SARB/NachJ and GAIC/NachJ (range of sex dimorphism across strains at 15 minutes post injection: 5.29–91 mg/dL). Glucose response curves are significantly different for all pairwise strain comparisons involving SARC/NachJ males and females (Permutation test P < 0.005, S27 Table and Fig 10), emphasizing the unique response profile of this strain. The area under the glucose response curve varies 2.4-fold across strains, with strains GAIC/NachJ and C57BL/6J delimiting the lower and upper extremes, respectively (S25 Table). Overall, results from glucose tolerance testing nominate several Nachman strains (most notably, SARC/NachJ) as potential models of Type II diabetes resistance.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 10. Glucose response curves over a 2-hour window.

Females and males were analyzed separately. Mice were fasted for 4 hours prior to administration of a fixed (w/v) amount of glucose by intraperitoneal injection. Blood glucose levels were assessed at 15, 30, 60, 90, and 120 minutes post injection. In both panels, strains are color-coded by geographic origin, with strains from a common location further distinguished by line type. Vertical bars denote +/- 1 SD around the strain mean.

https://doi.org/10.1371/journal.pgen.1011228.g010

Clinical chemistry.

We assessed levels of numerous blood-based clinical markers after subjecting mice to a 4 hr fast at 19 weeks (Fig 11). Nachman strains vary in immune cell populations, blood lipid profiles, measures of liver function, and both platelet and red blood cell composition (S16 and S17 Tables). For example, total cholesterol levels range 2.7-fold among strains, with TUCB/NachJ and GAIC/NachJ mice defining the extremes (71.2 mg/dL– 195 mg/dL; S25 Table). Males from all strains have higher total cholesterol values than their female conspecifics (average dimorphism = 34.3 mg/dL), a trend echoed in levels of HDL cholesterol (Fig 11A and 11B). Nachman lines show exceptionally high variability in platelet counts (range of strain means: 754–1725 x10³ cells/uL) and platelet volume (range: 6.39–9.58 fL), with lines from Manaus, Brazil presenting high numbers of low-volume platelets, and strains from Saratoga Springs having low numbers of high-volume platelets (Fig 11C and 11D and S21 and S22 Tables). Multiple immune cell populations, including monocytes, lymphocytes, neutrophils, and eosinophils, exhibit heritable variation across the Nachman strain panel (S17 Table), observations that may underlie recent reports of strain differences in pathogen response [58].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 11.

Boxplots revealing the extent of strain and sex variation in (A) total cholesterol, (B) HDL cholesterol, (C) platelet count, and (D) platelet volume.

https://doi.org/10.1371/journal.pgen.1011228.g011

Integrated analysis of Nachman and classical inbred strain phenotypes expands the variance observed in classical inbred strains

We intersected phenotype data from the Nachman lines with existing strain survey datasets deposited on the Mouse Phenome Database (see Methods; [59]). The Nachman animals expand the range of phenotype values observed across inbred strain panels for several phenotypes, including total platelet counts, percent fat mass, and hematocrit levels (S11 Fig).

The inclusion of C57BL/6J mice as controls in our phenotyping cohorts provides a common touchpoint between our data and previously published datasets. While we find that C57BL/6J trait values are generally stable across our phenotyped cohorts (S28 Table), C57BL/6J trait values differ significantly between our data and legacy datasets (S29 and S30 Tables). The inconsistency in trait values could owe to differences in animal age at the time of phenotyping, differences in housing conditions, animal handling by different technicians and research staff, differences in phenotyping protocols or equipment, or other vagaries of the experimental environment. Regardless of the source of this phenotypic variance, it poses major obstacle to phenotype data integration with the Nachman strains and underscores the need for caution in the interpretation of absolute differences in trait values across independently collected datasets.

Strain-level trait correlations across phenotype assays

Significant strain-level phenotype correlations may reveal traits regulated through shared genetic pathways. As expected, measures of body composition, body size, and organ weights are positively correlated, suggesting a general pattern of isometric growth (S8 Fig). Additionally, measures of total activity in indirect calorimetry and open field assays are positively correlated (S15 Table). Total serum bilirubin levels are negatively correlated with several phenotypes assessed from the spontaneous alternation assay, including total number of arm entries (rho = -0.97; P = 0.00017) and the number of spontaneous alternations (rho = -0. 83, P = 0.0083), but suggestively positively correlated with the percent alternation (rho = 0.62; P = 0.086). This latter result supports published reports of a negative association between serum bilirubin concentration and cognitive impairment in schizophrenia in humans [60], tentatively extending this correlation to mice. Leptin levels are positively correlated with fat mass at each timepoint assessed by NMR (4, 8, 12, 16, and 19 weeks; rho > 0.66; P < 0.06) and gonadal fat pad weight (rho = 0.63; P = 0.076), and are negatively correlated with food consumption per unit body weight (rho = -0.85; P = 0.0061). These results reinforce the well-established link between leptin secretion and satiety sensing. The duration of the QRS interval on an unconscious EKG is negatively correlated with percent lean mass, but positively correlated with fat mass across strains (S15 Table). These findings lend independent support to a recently published association between QRS duration and BMI in humans with no overt cardiac disease [61]. A comprehensive summary of all pairwise trait correlations is provided in S15 Table. Sex-specific trait correlations are provided in S31 and S32 Tables.

Ethics statement

All animal procedures were conducted in compliance with animal care protocols approved by The Jackson Laboratory’s Animal Care and Use Committee (Animal Use Summary # 18070) and the University of California Berkeley Animal Care and Use Committee (AUP R361-0514 and AUP-2016-03-8548).

Inbred strain development

Wild house mice were caught in 2012–2013 from five geographic locations (Fig 1) using Sherman live traps baited with oats. To avoid catching closely related individuals, each mouse was trapped at least 500m from every other mouse. Mice were transported to UC Berkeley and held in quarantine during pathogen testing. Mice from each geographic region were randomly paired to create inbred lines which were propagated through brother-sister mating for at least 10 generations. Initially, 10 independent lines were established from each location. No attempts were made to rescue lines that exhibited infertility due to inbreeding depression, ensuring that surviving lines likely harbor low deleterious mutation loads. The founders of these strains were prepared as museum specimens and have been deposited in the collections of the UC Berkeley Museum of Vertebrate Zoology.

Mouse breeding, rederivation, and husbandry

Mice from 20 incipient inbred strains were imported from the Nachman Laboratory mouse colony at UC Berkeley to the Jackson Laboratory Importation Facility (S2 Table). Strain colonies were expansion bred to obtain cohorts of >20 breeding-aged females for oocyte collection. To expedite breeding, strains were mated with some allowance for breeding between generations, a departure from strict sib-sib inbreeding.

Females were superovulated according to standard procedures [72] and harvested oocytes were in vitro fertilized with sperm from conspecific males [73]. Embryos were cultured to the 2-cell or 8-cell stage before implantation into pseudopregnant QSi5/IanmTaftJ x C57BL/6J (JAX strains 027001 x 000664) F1 recipient dams of high health status. Live-born pups were transferred to the Dumont Laboratory colony where they were maintained at intermediate health status by strict sib-sib mating. A representative male and female from each successfully imported strain was genotyped on JAX’s standard 54-SNP panel to assure expected strain identity and safeguard against contamination with known laboratory stocks.

Mice were housed under SPF conditions in sterile plastic caging and subject to biweekly cage changes. All animals were fed 6% sterilized rodent chow (Lab Diet ® Formulation 5K0G) and provided with acidified water ad libitium. Cages were supplied with aspen bedding material and enriched with nestlets, crinkle paper nesting material, red igloos, and cedar blocks for gnawing.

Cytogenetic chromosome analysis

Spermatocyte cell spreads were prepared from testis tissue of males aged >8 weeks and immunostained as previously described [74]. Antibodies used were a polyclonal antibody against mouse SYCP3 (1:100 dilution; Novus Biologicals, cat. # NB300-231) and human anti-centromere protein (1:100 dilution; Antibodies Incorporated, cat. # 15–235). Cells were then imaged on a Leica DM6B upright fluorescent microscope equipped with fluorescent filters, LED illumination, and a cooled monochrome Leica DFC7000 GT 2.8 megapixel digital camera. A minimum of 20 cells were captured per strain. Fluorescent intensity and background signal were manually adjusted, and the number of chromosomes counted using ImageJ (v 1.53k).

Whole genome sequencing

DNA extraction, library preparation, quality control, and sequencing were performed by the Genome Technologies Scientific Service at The Jackson Laboratory. An initial survey of short-read Illumina whole genome sequences from three strains (SARA/NachJ, SARB/NachJ, MANB/NachJ) pointed to high levels of structural variation relative to the mm39 reference genome (S12 Fig). This discovery motivated our use of long-read PacBio HiFi technology for all subsequent sequencing.

High molecular weight DNA was isolated from spleen tissue of a single male from each of the 11 imported Nachman wild-derived inbred strains using the Wizard DNA Purification Kit (Promega) or the Monarch HMW DNA kit (NEB) according to manufacturer’s instructions. DNA concentration and quality were assessed using the Nanodrop 2000 spectrophotometer (Thermo Scientific), the Qubit 3.0 dsDNA BR Assay (Thermo Scientific), and the Genomic DNA ScreenTape Analysis Assay (Agilent Technologies). DNA quality from all samples was assessed to be high (260/280 > 1.79 and 260/280 < 1.86, 260/230 > 1.99) and suitable for input for PacBio HiFi library construction.

PacBio HiFi libraries were constructed for each sample using the SMRTbell Express Template Prep Kit 2.0 (Pacific Biosciences) according to the manufacturer’s protocols. Briefly, the protocol entails shearing DNA using a g-TUBE device (Covaris), ligating PacBio specific barcoded adapters, and size selection on the Blue Pippin (Sage Science). The quality and concentration of the library were assessed using the Femto Pulse Genomic DNA 165 kb Kit (Agilent Technologies) and Qubit dsDNA HS Assay (ThermoFisher), respectively, according to the manufacturers’ instructions. The resultant library for each strain was sequenced on a single SMRT cell on the Sequel II platform (Pacific Biosciences) using a 30-hour movie time.

Read mapping and variant calling

Read quality and library coverage were assessed using fastp [75]. Individual SMRT cells yielded between 22.04 and 38.04 Gb of unique sequence data, with an average read length of 12.9kb (S4 Table). Reads were then mapped to the GRCm39 reference sequence using minimap2 invoking the “HIFI” preset. Per sample single nucleotide calling was performed using DeepVariant (v1.2.0) under the “PACBIO” model [43]. Per sample gVCF files were then merged using glnexus (v1.2.7) under the DeepVariantWGS configuration to produce a joint call set [42]. Sites with missing data, genotype quality <30, and indels were subsequently filtered using bcftools (v 0.1.19; [76]). We further eliminated sites with heterozygous calls as these sites are potentially enriched for false positives given our modest sequencing coverage. We refer to this call set as the “Nachman only” call set.

Variants in the Nachman strain panel were compared to those previously discovered in laboratory inbred strains in order to assess the extent to which this new resource captures novel genetic diversity. We downloaded GRCm39-mapped bam files for 51 inbred strains from the European Nucleotide Archive (PRJEB47108). Per sample variant calling was performed using the “WGS” model in DeepVariant (v1.2.0). Joint variant calling of laboratory strain genomes and the 11 inbred Nachman strains was carried out using glnexus (1.2.7) in DeepVariantWGS mode. As above, indels and variants with genotype quality <30 were removed using bcftools (v. 0.1.19). Sites with >10% missing data and heterozygous calls were also excluded. We refer to this callset as the “Sanger-Nachman” call set.

Identical methodology was adopted to generate a joint call set for the Nachman strains and 97 wild-caught M. m. domesticus, 22 M. m. musculus, 30 M. m. castaneus, and 7 M. spretus. Sample accession numbers for the wild mouse genome sequences used are provided in S8 Table. We refer to this call set as the “wild-Nachman” call set. Again, sites with >10% missing data, genotype quality <30, and indels were excluded using bcftools (v 0.1.19).

Basic statistics were computed on each call set using bcftools stats. Call sets were partitioned and intersected using bcftools view and and bcftools isec, respectively. Variant effects were predicted using the Variant Effect Predictor (Ensembl release 109.3) using the GRCm39 Mus musculus assembly [77]. GO Enrichment Analyses were performed using the enrichment analysis tool on the GO Consortium Website (http://geneontology.org/), with the entire set of annotated genes in Mus musculus as background [78,79]. Enrichment was determined by a Fisher’s Exact test with False Discovery Rate of P < 0.05.

Genome sequence analysis

SNP variation in the Sanger-Nachman call set was summarized via principal component analysis. We first eliminated six strains of non-M. m. domesticus ancestry from the call set (CAST/EiJ, CZECHII/EiJ, JF1/MsJ, MOLF/EiJ, PWK/PhJ, and SPRET/EiJ) and removed fixed variants using the view command in bcftools (v 0.1.19), invoking the flags -q 0.01 -s ^CAST_EiJ,CZECHII_EiJ,JF1_MsJ,MOLF_EiJ,PWK_PhJ,SPRET_EiJ. Variants were greedily thinned to include only those with r² > 0.2 using PLINK (v1.90b6.18; [80]):

plink --vcf $VCF \

--double-id --allow-extra-chr \

--set-missing-var-ids @:# \

--vcf-half-call missing \

--indep-pairwise 50 50 0.2 --out $THINNED_VARIANTS

Principle component analysis was performed on the pruned set of variants using the–pca command in plink:

plink --vcf $VCF \

--double-id --allow-extra-chr \

--set-missing-var-ids @:# \

--vcf-half-call missing \

--extract $THINNED_VARIANTS \

--make-bed --pca --out $PCA_RESULTS

A maximum likelihood phylogenetic tree was constructed from the LD-thinned SNPs on chr19 in the Sanger-Nachman call set using phyml (version 3.3; [81]). We focus on chr19 for the sake of computational efficiency. We first created a fasta format file from the thinned chr19 SNPs using a custom perl script. The output file was then converted to phylip format using the SeqIO.write function in the BioPython (v1.43) library. A maximum likelihood tree was constructed under a GTR model of nucleotide evolution, with nucleotide frequencies computed from the empirical data, and with the transition/transversion ratio, proportion of invariant sites, and gamma distribution of rate classes estimated via maximum likelihood. The executed command was:

phyml --input $ALN \

--datatype nt \

--bootstrap -1 \

--model GTR \

-f e -t e \

--pinv e --alpha e \

--r_seed 12345

Identical methodology was used to perform PC analysis and phylogeny construction on the Nachman-wild call set. Prior to analysis, we first filtered the vcf file to include only wild M. m. domesticus samples, recognizing that PCA results would be dominated by subspecies-level diversity if samples from M. m. musculus and M. m. castaneus were retained. We also further subset the Nachman-wild call set to exclude M. m. domesticus samples from Iran.

Introgression analysis

Nachman strain genomes were scanned for potential signatures of introgression from M. m. castaneus and M. m. musculus using the allele sharing methods implemented in Dsuite (v. 0.5) [82]. Wild M. m. domesticus and M. m. castaneus (or M. m. musculus) genomes in the wild-Nachman call set were utilized as P1 and P3, respectively, with M. spretus genomes specified as the outgroup taxon (S8 Table). Dsuite Dtrio was run separately on each chromosome and with each inbred Nachman strain individually profiled as P2. Chromosome-level statistics were integrated into genome-wide estimates using Dcombine. To pinpoint specific sites of introgression between M. m. castanus and each focal strain, a windowed analysis was performed using the Dinvestigate command in Dsuite with a window size of 5000 informative sites and 2500 site slide. Output statistics were plotted in RStudio (2022.02.0, Build 443). We focus on sites with an excess of derived alleles shared between each of our profiled Nachman lines and M. m. castaneus and extracted the 5% of windows with the highest f_dM estimates consistent with potential introgression from M. m. castaneus. These outlier windows were then intersected with Ensembl genes (v. 109; Mus musculus GRCm39) using bedtools isec (v. 2.28.0; [83]) and subjected to a GO Enrichment Analysis using the online enrichment analysis tool on the GO Consortium Website (http://geneontology.org/) [78,79]. We specified the entire set of annotated genes in Mus musculus as background to identify specific biological processes enriched in putative regions of M. m. castaneus introgression. Enrichment was determined by a Fisher’s Exact test with False Discovery Rate of P < 0.05.

Structural variant discovery and calling

We identified SVs in the 11 Nachman wild-derived inbred strain genomes using both pbsv (https://github.com/PacificBiosciences/pbsv) and sniffles2 (v. 2.0.7; [84]). pbsv was first run on each sample in discover mode to identify read signatures consistent with possible SVs. SVs where then called and samples jointly genotyped by executing pbsv in call mode. Tandem repeats in the GRCm39 assembly were identified using the findTandemRepeats.py script (https://github.com/PacificBiosciences/pbsv/commit/bcec7d382f3ea40158ed9cca3c5fef9686a76641) and supplied when executing sniffles2 to improve the accuracy of calls in repetitive regions. Per sample SV calls generated by sniffles2 were merged and filtered to include only autosomal calls using bcftools merge (v. 0.1.19). Calls with close or overlapping breakpoints across samples were collapsed using truvari (v4.0.0; [85]), with the following parameters specified: -pctsize 0.75 –pctovl 0.5 –pctseq 0.7 -s 20 -S 10000000 -k common—chain. We then intersected pbsv and sniffles2 SV calls using truvari bench to produce a higher confidence call set. We used the pbsv callset as the “truth” set and invoked the following command line parameters: -pctsize 0.75 –pctovl 0.5 –pctseq 0.7 –dup-to-ins–passonly -sizemin 20. SVs were annotated using the Ensembl Variant Effect Predictor (release 109.3) and gene model annotations from the GRCm39 assembly.

Deletions and insertion sequences in this intersection call set were converted to fasta format and input to RepeatMasker (v 4.1.5). The following repeatMasker command was executed:

RepeatMasker \

-e hmmer

-q \

-species “mus musculus” \

-lcambig \

-nocut \

-div 50 \

-no_id \

-dir $OUTDIR \

-trf_prgm $PATH_TO_TRF/trf409.linux64 \

-hmmer_dir $PATH_TO_HMMER/hmmer-3.3.1/src \

$FASTA

The number and base coverage of each family and class of transposable element within SVs was calculated from the alignment file output of RepeatMasker using the parseRM.pl script available at: https://github.com/4ureliek/Parsing-RepeatMasker-Outputs/blob/master.

Lastly, SV calls were intersected with a previously published SV call set for a diverse set of inbred mouse strains [49]. Shared SVs (>75% reciprocal overlap) were identified using truvari bench, with the Nachman callset featured as the “truth” set and invoking the same command line parameters as above.

Animal phenotyping

We developed a phenotyping pipeline loosely modeled on that used by the KOMP2 Project (https://www.mousephenotype.org/impress/index), with modifications to account for the overall health and wildness of the Nachman strains (Fig 6). Phenotypes broadly profile disease-relevant neurobehavioral, physiological, metabolic, biochemical, and morphological trait variation.

Mice were organized into 16 phenotyping cohorts ranging in size from 6–16 animals (mean = 13.4; S14 Table). With the exception of cohort 13, each cohort was composed of animals from multiple strains (N = 2–5 strains) and mice of both sexes. Cohort 13 included only animals from strain GAIA/NachJ. Ten of the 16 cohorts included C57BL/6J mice as controls (n = 2–5 mice per cohort), providing a means for ensuring stability of phenotype measurements across cohorts. A total of 215 mice (108 female, 107 male) were phenotyped from 9 of the 11 Nachman lines in two waves. The first wave included 5 cohorts that were phenotyped between August 2020 and November 2020. The second wave included the remaining 11 cohorts which were phenotyped between June 2021 and October 2021. Strain EDMB/NachJ was not imported in time for inclusion in this phenotyping effort.

All live animal phenotyping was performed by trained staff in the Center for Biometric Analysis at The Jackson Laboratory (JAX). Animals were ~4 weeks (± 6 days) at study intake and exited the phenotyping pipeline at 19 weeks (± 6 days). Terminal collections and tissue harvests were performed by the Necropsy Scientific Service at JAX, with samples subsequently transferred to the Clinical Chemistry and Histopathological Sciences Scientific Services at JAX.

Phenotype data integration, statistical analysis, and broad sense heritability estimation

All phenotype data were wrangled into a single file in RStudio (v. 2022.02.0 Build 443) using the dplyr (v. 1.1.1), tidyverse (v. 2.0.0), data.table (v. 1.14.8), readxl (v. 1.4.2), and lubridate (v. 1.9.2) packages (S24 Table). Categorical traits collected as part of the frailty analysis (n = 27) showed little variation among strains (S25 Table) and were excluded from statistical analyses. All remaining phenotypes were continuous. We tested for overall strain effects on each phenotype using non-parametric Kruskall Wallis (S16 Table) and one-way ANOVA tests (phenotype value ~ strain; S17 Table). Additionally, we tested for an effect of geographic origin on variation in each phenotype using Kruskall-Wallis (S21 Table) and one-way ANOVA tests (phenotype value ~ sample locale; S22 Table). C57BL/6J controls were excluded from these analyses to allow specific focus on variation across the Nachman strains. Although the ANOVA assumption of normality does not hold for many analyzed traits, there is strong agreement of P-values estimated by these two statistical methods (Spearman’s Rho = 0.929, P < 2.2 x 10⁻¹⁶). Given that our objective is to flag phenotypes that likely vary across the Nachman panel (rather than make robust claims about specific phenotypes), reported P-values are not corrected for multiple testing.

PC analysis was performed on the set of traits from each phenotype assay using the pca command within the pcaMethods R package [87]. We distilled trait values into per strain, per sex means and normalized the resulting values to have unit variance centered on zero. We then identified the number of PCs required to explain 90% of the variance associated with each phenotype assay and combined these into a single matrix. PC analysis was performed on this combined matrix to summarize multidimensional trait variation across strains and sexes (S7 Fig). To compare the phenotypic PC matrix with the genotypic PC matrix, phenotypic PC values were averaged across both sexes within a strain. Matrix similarity was assessed by the similarity of matrices index using the SMI function call within the MatrixCorrelation R package [88]. Significance of the resulting SMI value was determined by 1000 permutations.

Broad sense heritability was estimated from one-way ANOVA output by the intraclass correlation method [89]: where MSB and MSW are the mean squares between and within strains, respectively. Heritability estimates were computed from one-way ANOVA tests run on each sex separately, as well as both sexes combined (S17 Table). Two-way ANOVA was also performed to test the impact of strain, sex, and their interaction on the variability observed in each trait (phenotype value ~ strain * sex; S20 Table). Spearman’s rank correlate was used to quantify trait correlations at the strain, strain × sex, and individual levels.

In parallel to this approach, we also fit a series of nested linear mixed effect models to assess the impact of strain, sex, and strain × sex interaction on each phenotype. Specifically, for each trait t, we fit the following series of models using the lmer command within the lme4 (version 1.1–32) R package: (1) (2) (3) (4)

Sex was treated as a fixed effect, with strain, cohort, and the interaction between strain and sex treated as random effects. Nested models were compared by a log likelihood test using the anova function call in R. Specifically, models 1 and 2 were compared to assess whether inclusion of strain as a random effect significantly improved model fit. Models 1 and 3 were compared to assess whether inclusion of sex significantly improved model fit, and models 1 and 4 were compared to determine whether the interaction between strain and sex meaningfully improved model fit. Owing to the modest amount of data and the complexity of these models, many models were singular, necessitating caution in the interpretation of results from this model fitting approach. Of the 1092 traits analyzed under this mixed model framework, only 358 provided non-singular model fits (S18 Table).

Many phenotypes are highly correlated with body weight (S15 Table), motivating us to also consider body weight-adjusted trait values in our models. We fit the residuals from a simple linear regression of each trait value on body weight at 18 weeks in each of the models above. This had a similar impact on the number of singular model fits (1083 analyzed traits, with 462 with non-singular fits across models 1–4; S19 Table). We excluded body weight traits at all assessed timepoints, accounting for the reduced number of phenotypes analyzed under this framework.

To explore the role of geographic sample location on observed variation across strains, we fit linear mixed effect models analogous to those in Eqs (1) and (2), exchanging the random factor “Strain” for “Location”.

(5)

(6)

Model comparisons were performed via likelihood ratio test to determine whether inclusion of location as a random effect significantly improved model fit. Phenotype values were first regressed against body weight at 18 weeks to eliminate the correlation with this trait. As above, the majority of phenotypes provided singular fits (S23 Table).

Comparisons with legacy phenotype data

We accessed the CGDpheno1 and CGDpheno3 datasets from the Mouse Phenome Database [4]. These strain survey datasets were prioritized because: (1) they share multiple phenotypes in common with those measured in the Nachman strains, (2) they were generated within the CBA at JAX, providing methodological consistency in data collection and similarity in housing conditions, and (3) animals were not subject to any experimental treatments. Public datasets were combined with the Nachman data using base R commands and dplyr (v. 1.1.1), with attention paid to consistency of measurement units. Where needed, values were converted onto common unit scales using the appropriate conversion factors or mathematical manipulations.