Metabolite Genome-Wide Association Study (mGWAS) and Gene-Metabolite Interaction Network Analysis Reveal Potential Biomarkers for Feed Efficiency in Pigs

Abstract

Metabolites represent the ultimate response of biological systems, so metabolomics is considered the link between genotypes and phenotypes. Feed efficiency is one of the most important phenotypes in sustainable pig production and is the main breeding goal trait. We utilized metabolic and genomic datasets from a total of 108 pigs from our own previously published studies that involved 59 Duroc and 49 Landrace pigs with data on feed efficiency (residual feed intake (RFI)), genotype (PorcineSNP80 BeadChip) data, and metabolomic data (45 final metabolite datasets derived from LC-MS system). Utilizing these datasets, our main aim was to identify genetic variants (single-nucleotide polymorphisms (SNPs)) that affect 45 different metabolite concentrations in plasma collected at the start and end of the performance testing of pigs categorized as high or low in their feed efficiency (based on RFI values). Genome-wide significant genetic variants could be then used as potential genetic or biomarkers in breeding programs for feed efficiency. The other objective was to reveal the biochemical mechanisms underlying genetic variation for pigs' feed efficiency. In order to achieve these objectives, we firstly conducted a metabolite genome-wide association study (mGWAS) based on mixed linear models and found 152 genome-wide significant SNPs (p-value < 1.06 × 10^-6) in association with 17 metabolites that included 90 significant SNPs annotated to 52 genes. On chromosome one alone, 51 significant SNPs associated with isovalerylcarnitine and propionylcarnitine were found to be in strong linkage disequilibrium (LD). SNPs in strong LD annotated to FBXL4, and CCNC consisted of two haplotype blocks where three SNPs (ALGA0004000, ALGA0004041, and ALGA0004042) were in the intron regions of FBXL4 and CCNC. The interaction network revealed that CCNC and FBXL4 were linked by the hub gene N6AMT1 that was associated with isovalerylcarnitine and propionylcarnitine. Moreover, three metabolites (i.e., isovalerylcarnitine, propionylcarnitine, and pyruvic acid) were clustered in one group based on the low-high RFI pigs. This study performed a comprehensive metabolite-based genome-wide association study (GWAS) analysis for pigs with differences in feed efficiency and provided significant metabolites for which there is significant genetic variation as well as biological interaction networks. The identified metabolite genetic variants, genes, and networks in high versus low feed efficient pigs could be considered as potential genetic or biomarkers for feed efficiency.

Keywords: GWAS; RFI; SNPs; biomarkers; feed efficiency; metabolomics; pathways; pigs.

Figure 1

(A) Partial least squares-discriminant analysis (PLS-DA) of 45 metabolites. Note: D/L with first/second indicates the sampling time of Duroc/Landrace pigs. (B) Metabolic pathways for 45 metabolites using Homo sapiens as the library. Note: The size and color of the circles for each pathway indicate the matched metabolite ratio and the −log (p-value), respectively.

Figure 2

Manhattan plots of genome-wide association for (A) isovalerylcarnitine and (B) propionylcarnitine. Note: Y-axis indicates the log₁₀ (p-value). Blue dotted and red solid lines indicate the genome-wide threshold of 0.05 and 0.01 after Bonferroni multiple testing, respectively. The three tracks indicate the metabolites from first sampling time, second sampling time, and combined two sampling times from outside to inside.

Figure 3

Linkage disequilibrium (LD) pattern for 53 significant SNPs on chromosome one. Note: the solid line triangle indicates LD. One square indicates LD level (r²) between two SNPs, and the squares are colored by the D’ & LLOR standard scheme. D’ & LLOR standard scheme is that red indicates LLOR > 2, D’ = 1; pink indicates LLOR > 2, D < 1; blue indicates LLOR < 2, D’ = 1; and white indicates LLOR < 2, D’ < 1. LLOR is the logarithm of likelihood odds ratio and the reliable index to measure D’.

Figure 4

Gene pathway, metabolite cluster, and the interaction network: (A) Pathway for significant SNP-related genes. (B) Network for the genes in which significant SNPs were annotated to gene components. (C) Heatmap cluster for the metabolites that were associated with significant SNPs annotated to gene components. (D) Metabolites (i.e., aspartic acid, isovalerylcarnitine, propionylcarnitine, and pyruvic acid) values in high and low residual feed intake (RFI) pigs associated with the genes in which significant SNPs annotate to gene components. Note: The high RFI pigs and low RFI pigs were from left and right parts of all the pigs (n = 108) with one SD of actual RFI values.

Figure 5

Distribution of actual RFI values of Duroc (n = 59) and Landrace (n = 49) pigs.

Figure 6

Overall analysis workflow.

Figure 7

Statistical description of 45 metabolites from combined two sampling times. Note: The red solid circle indicates the limit of detection (LOD) relative value of each metabolite. LOD refers to the lowest value of a metabolite that the LC-MS method can detect. M1 to M45 indicate the metabolites of 1-hexadecyl-sn-glycero-3-phosphocholine, 1-myristoyl-sn-glycero-3-phosphocholine, (3-Carboxypropyl)trimethylammonium, 4-aminobenzoic acid, 5-methyl-5,6-dihydrouracils, acetaminophen, acetylcarnitine, alanine, arginine, aspartic acid, benzoic acid, carnitine, citrulline, cotinine, creatinine, cytidine, disaccharide, glutamic acid, guanine, guanosine, hypoxanthine, indoleacrylic acid, inosine, isoleucine, isoleucyl proline, isovalerylcarnitine, lactic acid, leucyl methionine, lysoPC(16:0), manNAc, methionine, monosaccharide, nicotine amide, ornithine, phenylalanine, proline, propionylcarnitine, pyruvic acid, riboflavine, sorbitol, thiamine, threonine, thymidine, uridine, and xanthine, respectively.