Systems Genetics of Metabolism: The Use of the BXD Murine Reference Panel for Multiscalar Integration of Traits

The BXD Family Has Wide Variation in Metabolic Traits

Metabolic traits vary to a remarkable degree among the BXD strains, with every trait showing significant variation (Figure 2A). The least variable trait, basal body temperature, varied significantly (from 35.8 ± 0.3 °C in BXD34 to 38.9 ± 0.2 °C in BXD95) despite an overall variation of less than 1.1-fold. The most variable trait, alkaline phosphatase (ALPL) protein activity level, varied 6.5-fold between BXD31 (44 ± 11 U/l) and BXD90 (287 ± 18 U/l). For the majority of traits, the distribution of phenotypes is continuous and approximately normal in both sexes.

Open in a separate window

Figure 2

Strain and Sex Influences Differ Widely between Parameters

(A) Metabolic phenotype variation for all cohorts.

(B) Traits with higher variation were more likely to have significant differences between males and females; 41% of traits are significantly different (traits on the x axis lower than rank 22).

(C) Example of a sex-independent trait that is mostly determined by environment: the ventricular shortening fraction.

(D) Example of a sex-independent trait that is mostly determined by genetics: the mean cell volume of red blood cells.

(E) Example of a sex-influenced trait that is explained by a mix of genetic and environmental factors: body weight at 18 weeks. Bar graphs and X-Y plots are expressed as mean + SEM.

As expected, sex differences are a significant source of variance (Champy et al., 2004), with the more variable traits more likely to exhibit significant sexual dimorphism (Figure 2B). Traits can be divided into four categories in terms of the significance of variation and correlations between sexes. Nearly 60% of traits had no significant sex differences. These traits fall into two categories: those without tight male-female correlation (e.g., shortening fraction in Figure 2C) and those with it (e.g., red blood cell volume in Figure 2D). The first category, with no significant sex differences and without strong correlation between males and females, indicates phenotypes where variation cannot be reliably attributed to genetic factors (i.e., variation is due to environmental or technical influence) or that there is a strong sex-by-strain interaction effect. In the case of the shortening fraction, we suspect the former factor as the cause of the low heritability (h² = 0.5) of this trait (Figure 2C). In contrast, the traits with strong correlations between sexes indicate that the genetic variability is not affected greatly by the environment, and thus the slope of the correlation is close to 1 (Figure 2D).

The remaining 40% of traits exhibit significant differences between males and females, and again with or without strong male-female correlation. The most striking example is body weight, with a dramatic difference between males and females, but without consistency across strains: e.g., BXD75 males are the leanest, whereas the BXD75 females are in the middle range of body weights (Figure 2E). Lastly, a few traits had both significant sex-associated differences and strong genetic correlation—one of them being discussed in Figure 4.

Open in a separate window

Figure 4

Reducing the ALPL QTL to Its Causative Mutation between DBA and C57 Alleles

(A) Expression variation of serum ALPL activity and mRNA levels from liver transcriptome data. Strains with the D allele (red) have significantly more ALPL protein and Alpl mRNA than strains with the B allele (white) for all comparisons (p < 0.005). Males have significantly less ALPL protein than females, but there is no difference in mRNA levels.

(B) Correlation between male and female Alpl mRNA and ALPL protein levels demonstrate high trait heritability (Figure 1D).

(C) QTL map for liver Alpl mRNA and blood ALPL levels in males and females. All four QTLs (male and female traits are mapped separately) mapped to the same location and crossed the significance threshold of p < 0.05, indicated by the red horizontal line. The Alpl gene is located below the rectangular peak at right.

(D) 3D model of ALPL from Phyre2, based on the crystal structure of the placental form of ALP (ALPP). Enlarged on the right are the structure of the amino acids flanking and including the two missense mutations (noted in green).

(E) Forty-four amino acid excerpt of sequence comparison of orthologs of ALPL, the amino acid sequence of the region modeled in (D) is shown here. The full protein sequence is highly conserved in mammals (>90% homology) and moderately from mammals to bacteria (>30% homology).

(F) Phylogenetic tree of Alpl. The gene has orthologs in all known species, notably including invertebrates. Bar graphs and X-Y plots are expressed as mean+SEM. Related to Figure S2.

QTLs Link Phenotypes to Causal Loci

The data sets generated consist of both general blood parameters, such as hematocrit and iron levels, and physical phenotypes, such as heart rate and oxygen consumption. Suggestive or significant QTLs were detected for nearly 40% of all phenotypes (Table S1). Traits with the highest heritability and the most significant QTLs usually mapped to a single locus. These traits are exemplified by the tissue-nonspecific (liver/bone/kidney) ALPL (also commonly known as TNSALP) protein activity, a trait for which 58% of the variance is explained by the single peak QTL (Figure 3A). Many other traits, including most classical phenotypes, map to several suggestive or significant loci. In these cases, oligogenic or polygenic networks of unlinked loci influence trait variance. One good example is basal oxygen consumption (VO₂), a trait for which three loci independently contribute to phenotypic variation—21% for the most significant locus on Chr1, 18% for the locus on Chr4, and 8% for the locus on Chr11 (Figure 3B).

Open in a separate window

Figure 3

Analysis of Peak Trait QTLs

(A) The genetic variation attributable to a QTL is calculated by the correlation between the trait values (here, ALPL levels) and the genotype at the peak location. ALPL has only one QTL, which explains 58% of the genetic variation.

(B) The genetic variation attributable to each of the three QTLs mapped to oxygen consumption. Each QTL has a much smaller effect, but together account for a large amount of variation. Generalized linear modeling or ANOVA (used here) is necessary to calculate the variation independently attributable to each QTL for traits which map to multiple QTLs.

(C) The LOD score, corrected p value, and QTL peak location are given along with the trait for the top 10 distinct phenotypes. Of note, this is not all of the significant or suggestive peaks for each trait; only the most significant peak is listed. One positional candidate is given for traits when a literature analysis of every gene under each peak has yielded a linked result. Candidates not discussed in the text include the link between white blood cells and Irak-M (Xie et al., 2007), Pten and heart rate (Zu et al., 2011), and RER in females and Bckdh (Joshi et al., 2006). Systolic BP LOD score is for analysis as described in (Koutnikova et al., 2009). A list of all 54 traits with suggestive or significant peak QTLs is available in Table S1.

Genes located within QTLs were examined for known links to function by using PubMed and GeneRIF (Mitchell et al., 2003). All but one of the top ten QTLs contained at least one strong positional and functional candidate (Figure 3C). In some cases, the positional candidate is well established as the causative gene behind the QTL. This is best illustrated by hematocrit levels, which map to a locus containing the hemoglobin gene (Popp, 1962). In other cases, the candidates are linked more tentatively. For example, the QTL for bone surface area contains the steroid 5 alpha-reductase 1 (Srd5a1) gene, which has been linked to bone mass in a recent study on Srd5a1 knockout mice (Windahl et al., 2011).

Genetic Polymorphisms Underlying Variation in ALPL Levels Contribute to Changes in Bone and Vitamin B6 Homeostasis

Serum ALPL levels are strongly variable among BXD strains, with females having significantly higher expression (but less variation) than males (Figures 4A and S2A). Expression of Alpl mRNA in the liver is also highly variable but does not differ by sex (Figures 4A and S2B) (Gatti et al., 2007). Interestingly, both Alpl mRNA and ALPL protein activity are highly correlated by strain, regardless of the sex of the animal (Figure 4B), suggesting strong genetic modulation and a sex effect shared by all strains. Data for ALPL activity in males and females map to a common QTL: an 11 Mb region of Chr4 (Figure 4C) that has already been reported in an F2 cross between C57BL/6 and DBA/2 mice (Foreman et al., 2005). Alpl mRNA also map to the same region (Figure 4C). This QTL harbors ~100 genes, including the Alpl gene itself. We hypothesized that differences in ALPL activity are primarily and perhaps almost exclusively caused by one or more allelic differences in the cognate Alpl gene, i.e that this is a so-called cis-regulated expression QTL (cis eQTL). Rapid identification of all sequence variants in the region was performed by using a mouse SNP database (www.genenetwork.org). Of 11 exonic SNPs that differentiate B and D haplotypes of Alpl, nine cause synonymous mutations and two cause missense mutations in exon 9, resulting in amino acid changes of R318Q and L324P (Figure S2C). We modeled the impact of these mutations on ALPL protein structure by using Phyre2 (www.sbg.bio.ic.ac.uk/phyre2/) (Figure 4D). Although these mutations are not located in the active site of ALPL (around S110) (UniProt Consortium, 2012), they are predicted to change the β sheet/α-helix structure (R318Q and L324P are in green in Figure 4D, showing the described excerpt of ALPL ± 21 amino acids). Remarkably, mouse and human ALPL protein sequences share 93%/97% identity/homology. In particular, the region harboring the two SNPs is highly conserved in vertebrates—notably proline 324 is invariant in other vertebrates, except in C57BL/6 where it has mutated into leucine (Figure 4E, same excerpt as in 4D). Moreover, the high degree of conservation of the ALPL protein is also revealed by phylogenetic tree analysis—human ALPL has 77%/87% sequence identity/homology with zebrafish ALPL, 44%/61% with D. melanogaster, and 32%/45% with E. coli (Figure 4E). ALPL orthologs separate into three clusters: vertebrates, insects, and bacteria (Figure 4F). Given this high genetic conservation, it is likely ALPL has functions in vertebrates that are wholly unrelated to bone structure.

In humans, mutations in the ALPL gene can cause varying degrees of hypophosphatasia (Mornet, 2008), a disorder nearly always associated with the accumulation of phosphoethanolamine (PEA) in the urine, extracellular pyridoxal-5′-phosphate (PLP) in the blood, and extracellular inorganic pyrophosphate (PP_i) in urine and blood. Downstream of these biomarkers are many pathways involved in bone remodeling (Whyte, 2000), amino acid metabolism, and neurotransmitter regulation (Lehninger et al., 2008). Therefore, we examined patterns of segregation of phenotypes among BXDs by testing several strains with either the B allele (C57BL/6, BXD48, BXD65, BXD68, BXD75, and BXD103) or the D allele of Alpl (DBA/2, BXD89, BXD90, and BXD95). We first reconfirmed that ALPL enzymatic activity was different among strains carrying the alleles (Figure 5A). As predicted, extracellular PLP levels were strongly negatively correlated with ALPL in both sexes, supporting the hypothesis that the BXDs are a good model for hypophosphatasia (Figure 5B). Furthermore, ALPL levels did not correlate with levels of calcium or phosphate (P_i) (Figure S3A), consistent with studies in humans differentiating hypophosphatasia from rickets and osteomalacia (Whyte, 2000). Femur bone structure evaluated via micro computed tomography (microCT) indicated decreased bone area and volume in females with the B allele of Alpl (Figure 5C), with no difference in overall animal weight between cohorts (p = 0.76). The same tendency was also noted for males (Figure S3B). Together these data suggest that the BXD strains are a useful model for hypophosphatasia.

Open in a separate window

Figure 5

Clinical Phenotyping of Mice with Extreme Alkaline Phosphatase Values

(A) ALPL levels in all individuals in a further in-depth study in male and female mice for individuals from the 11 strains described in the text. The effect observed on serum ALPL levels shown in Figure 4A was highly reproducible in this independent study, even given a significant difference in the age of the animals in the two studies (19 weeks versus ~46 weeks).

(B) Extracellular PLP levels strongly negatively correlated with ALPL levels in males and females. No correlations were observed between ALPL and extracellular PL, but the ratio of extracellular PLP/PL was also strongly negatively correlated with ALPL.

(C) Bone volume and surface area were significantly decreased in animals with the B allele of Alpl, consistent with the hypothesis of hypophosphatasia in the strains with the B allele.

(D) Scheme summarizing the multiple roles of ALPL in metabolism. Before entering tissues, pyridoxal-5′-phosphate (PLP) in the plasma must be dephosphorylated by ALPL into pyridoxal (PL), which can traverse cell membranes. PL is converted back into PLP by the pyridoxine kinase (PDXK) in the target tissues, such as liver or brain. PLP is a cofactor in many enzymatic reactions. Circulating ALPL also converts pyrophosphate (PP_i) into inorganic phosphate (P_i). Although PP_i inhibits bone mineralization, P_i and calcium (Ca²⁺) stimulate it. Low levels of ALPL activity disrupt proper bone development and remodeling via this mechanism.

Related to Figure S3.

It is very plausible that the differential regulation of PLP (Figure 5B) by ALPL drives further phenotypic changes (Figure 5D). PLP acts as a cofactor in many important biochemical reactions: principally, it is a required prosthetic group for all aminotransferases, suggesting a critical role in amino acid metabolism (Lehninger et al., 2008), and it is also an essential coenzyme for biosynthesis of heme (Astner et al., 2005). Confirmation of this link between ALPL over PLP to amino acid and hematological parameters warrants future work.

Wide Variation in Glucose Response Identifies Genetic Factors of Diabetes

Analysis of complex metabolic phenotypes also yielded QTLs with promising candidates. One such phenotype is glucose response during intraperitoneal glucose tolerance test (IPGTT), which records glycaemia levels before and 15, 30, 60 and 120 min after glucose injection and the overall glucose excursion curve (AUC). The AUC is significantly higher in males (Figure 6A) but correlates between the sexes (Figure 6B and Table S2A). The AUC is also linked to physical parameters such as body weight and composition in males, but surprisingly, not in females (Figure 6B and Tables S2B and S2C), perhaps partly explained by the fact that fewer female strains were studied. These results show that variation in glucose response is significantly determined by genetics and partly sex independent. Consistent with these observations, mapping QTLs for glucose levels during IPGTT revealed both common and distinct loci in both sexes (Figure S4).

Open in a separate window

Figure 6

Analysis of Response to an Intraperitoneal Glucose Tolerance Test

(A) Variation of the AUC of the glucose levels from 0 to 120 min.

(B) Network of glucose tolerance response determinants in males and females. Full details are given in Table S2.

(C) Multiple and single QTL heat map for glucose levels at each time point. Loci switch according to the feeding status of the mice. Fasting glucose (t = 0 min) maps appear only on Chr7. The early response to glucose injection (t = 15 min) maps on Chr2 only, whereas the rest of the time points also map on Chr1 and Chr9. Candidates under each QTL were selected according to existing literature showing their link with type 2 or type 1 diabetes. Further details are given in Figure S4.

(D) Network built around overall glucose response by using gene expression in the liver. Among the top 500 liver mRNA correlates with AUC, two were located on Chr1 QTL (Arpc5 and BC003331) and one on Chr9 QTL (Rab6b) (yellow boxes). One gene had a transQTL on the Chr1 QTL (Ppp1r3b), four had transQTLs on the Chr2 QTL (Sfrs6, Clic4, Slco4a1 and Mtf2), and one had a transQTL on the Chr9 QTL (Gip) (yellow boxes). Light blue boxes are attributed to the five genes that best correlated with AUC (Wapal, Rin3, Omt2b, Stradb and Myc). Bold dark blue lines represent −1 < r < −0.7, light blue lines −0.7 < r < −0.5, light orange lines 0.5 < r < 0.7 and bold red lines 0.7 < r < 1.

Related to Figure S4 and Table S2.

All genes under the four significant or suggestive QTLs for female glucose response on Chrs1, 2, 7 and 9 (Figure 6C) were investigated. The locus on Chr1 was associated with late glucose response, with glycaemia at 120 min giving the highest LOD score (Figure 6C). Within this locus, two genes are associated with type 2 diabetes susceptibility in humans: cytosolic phospholipase A2 group IV gene (Pla2g4a) and prostaglandinendoperoxide synthase 2 (Ptgs2, previously known as Cox2), both involved in prostaglandin metabolism (Konheim and Wolford, 2003; Wolford et al., 2003). A second QTL for early glucose response mapped on Chr2 (Figure 6C). Several genes known to be involved in insulin secretion or type 1 diabetes susceptibility are located under this peak, including paired box 6 (Pax6), Cd44, and catalase (Cat) (Chen et al., 2005; Jiang et al., 2011;Wen et al., 2009). When blood glucose is measured from mice under fasting conditions, only a single locus is mapped on Chr7 (Figure 6C). Although there are 91 genes under the locus, no gene was previously linked to fasting glucose levels in mice or in humans. Finally, the global AUC is strongly associated with the QTL located on Chr9 (Figure 6C). This region contains 31 genes, including one that has also been linked with insulin secretion: plasma membrane-related Ca²⁺-ATPase-1 (Atp2c1) (Mitchell et al., 2004).

Functional analysis of the three QTLs related to glucose response on Chr1, 2, and 9 uncovered six genes known to be involved in glucose regulation. However, these QTLs contain altogether 131 genes, many of which are not yet characterized. Therefore, we highlighted potential new candidates by finding the strongest mRNA correlates in liver (Figure 6D). From the list of the 500 best correlates, we selected candidates from three categories: (1) genes located under the QTLs on Chr1, 2, and 9; (2) genes that have trans-QTLs mapping to these loci; and (3) the five top gene correlates with glucose response. The most significant matches were integrated into the network (Figure 6D). Protein phosphatase 1 regulatory subunit 3B (Ppp1r3b), a protein that regulates hepatic glycogen synthesis (Kelsall et al., 2009), Myc, which plays a role in hepatic glucose uptake and utilization (Riu et al., 1996), STE20-related kinase adaptor beta (Stradb), and Rab6b, correlated exclusively with AUC (Figure 6D). In contrast, the rest of the candidates were entangled in an intricate network revolving around the AUC (Mtf2, Slco4a1, BC030417, Srfs6, Gip, Wapal, Clic4, Arpc5, Rin3, Omt2b, and BC003331) (Figure 6D). Among these genes, Sfrs6 and Gip have also been linked with glucose homeostasis (Fossey et al., 2001;Lyssenko et al., 2011).

General Study Plan and Phenotyping tests

The general outline of the phenotyping study is described in the results section and summarized in Figure 1C. All tests were carried out according to rigorous standard operating procedures (SOP) established and validated within the EUMODIC EMPReSS program (Champy et al., 2004; Champy et al., 2008).

We would like to thank Laurent Pouilly, Tania Sorg, Hongbo Zhang, and Ulrike Kettenberger for technical assistance with various aspects of mouse phenotyping. Discussions with Thomas Vogt (Merck Research Laboratories), Carmen Argmann and Sander Houten (University of Amsterdam), Vincent Mooser (CHUV), Stephan Morgenthaler (EPFL), Dominique Pioletti (EPFL), Bernard Thorens (UNIL), Paul Franken (UNIL), and Ioannis Xenarios (UNIL) and the team members of the Auwerx lab are acknowledged. R.H.H. is supported by a Rubicon fellowship of the NWO. J.A. is the Nestlé Chair in Energy Metabolism and the J.A./K.S. laboratory is supported by grants of the Ecole Polytechnique Fédérale de Lausanne, the EU Ideas program (ERC-2008-AdG-23118), the Velux Stiftung, the Swiss National Science Foundation (31003A-124713, 31003A-125487, and the Sinergia grant CSRII3-136201). R.W.W. and GeneNetwork are supported by NIH (P20-DA 21131 UO1AA13499 and U01AA14425) and the UT Center for Integrative and Translational Genomics.