Genome-wide unraveling SNP pairwise epistatic effects associated with sheep body weight

Abstract

Epistatic effects are an important part of the genetic effect of complex traits in livestock. In this study, we used 218 synthetic ewes from the Xinjiang Academy of Agricultural Reclamation in China to identify interacting paired with genome-wide single nucleotide polymorphisms (SNPs) associated with birth weight, weaning weight, and one-yearling weight. We detected 2 and 66 SNP-SNP interactions of sheep birth weight and weaning weight, respectively. No significant epistatic interaction of one-year-old body weight was detected. The genetic interaction of sheep body weight is dynamic and time-dependent. Most significant interactions of weaning body weight contributed 1% or higher. In the weaning weight trait, 66 significant SNP pairs consisted of 98 single SNPs covering 23 chromosomes, 5 of which were nonsynonymous SNPs (nsSNPs), resulting in single amino acid substitution. We found that genes that interact with transcription factors (TFs) are target genes for the corresponding TFs. Four epitron networks affecting weaning weight, including subnetworks of HIVEP3 and BACH2 transcription factors, constructed using significant SNP pairs, were also analyzed and annotated. These results suggest that transcription factors may play an important role in explaining epistatic effects. It provides a new idea to study the genetic mechanism of weight developing.

Introduction

Small ruminants are of obvious importance to human society. They can provide animal products, such as meat, milk, fur, to improve the quality of human life. Therefore, it is of decisive significance by increasing productivity and reproductive performance of ewes through breeding. QTLs mapping is an important means in sheep genetic improvement by revealing relationship between phenotype and genotype.^1–3 A major challenge in drawing genotype-phenotype maps is to identify the majority of genetic variations in DNA sequences that cause phenotypic variation in quantitative traits that are influenced by many genetic and environmental factors. To date, hundreds of genetic variants and candidate genes affecting body weight have been identified by linkage analysis and association studies. Epistasis interactions between loci are also known to play an essential role in revealing the full extent of genetic architecture in complex traits. For example, the epistatic variation in the growth traits during the differential development stages in Adani goats ranged from 10% to 20% of phenotypic variation.⁴ Thus, knowledge of epistatic effects would improve the prediction of response to artificial selection in sheep.

Strategies such as biological knowledge-driven and data-driven approaches using genome-wide genetic markers were developed to explore quantitative trait interactions. As a result, epistatic interactions were identified for many traits. Raadsma et al.⁵ detected pairwise epistasis effects of QTLs associated with sheep skin and fiber pigmentation on chromosomes 2 and 19. Hepp et al.⁶ found that the MC1R-ASIP gene interaction involved wool color. Following GWAS methods, genome-wide interaction studies (GWIS) investigated SNP pairwise interactions associated with complex traits, focusing on the whole genome level. Jiao et al.⁷ used GWIS to demonstrate that the interaction of EXOC4 and TOB1-related pathways may contribute to the development of obesity. Kramer et al.⁸ found epistatic interactions for 44 fatty acid traits in a population of Angus beef cattle based on GWIS. Banerjee et al.⁹ determined the genome-wide epistatic variants that affected feed efficiency traits in Duroc and Landrace purebreds. These studies showed that the GWIS strategy is an effective approach to discovering new genetic factors and candidate genes and providing additional biological insights.

As we known, sheep body weight was influenced by polygenes. So far, many candidate genes affect sheep weight at different developmental stages, such as MSTN, CAST, and GHR genes.^10–12 However, the impact of non-additive genetic variation on sheep weight remains unknown. To better understand the genetic mechanisms of body weight, this study aimed to identify genome-wide epistatic interactions in sheep that could explain additional genetic variation.

Materials and methods

Animal population

A total of 218 ewes (a composite line bred from Australian Suffolk sheep, Chinese Hu sheep, and Chinese Kazakh sheep) were collected from the Xinjiang Academy of Agricultural and Reclamation Science in China. All ewes have the same racial composition. The detailed information of these ewes was previously described.¹³ We carefully recorded weights at three stages: birth body weight (BIRTH_WT), weaning body weight (WEAN_WT), and yearling body weight (BW).

SNP genotyping and quality control

Genomic DNA was extracted from sheep ear tissue.¹⁴ All samples were genotyped using the Affymetrix Ovis600K Genotyping BeadChip kit (Compass Biotechnology Co., Ltd, in China). Data quality control was performed using PLINK (V1.90)¹⁵ as previously described, including (1) minor allele frequency (MAF) ≥$$5% SNPs, (2) SNP call rate ≥$\hspace{0.17em}$95%, (3) individual call rate ≥$$90%, and (4) SNPs mapping to X chromosomes and autosomes were evaluated. For SNP pairs to be tested, the sample size for each of the two SNP genotype combinations had to be ≥$\hspace{0.17em}$5.

Beagle (V5.3)¹⁶ was used for SNP genotype imputation. The number of independent SNP is calculated using HAPLOVIEW v4.2.¹⁷

Principal component analysis (PCA)

Using PLINK, a PCA was constructed from the data of the Ovis600K SNP chip to identify the population structure. Then, each eigenvalue and its corresponding eigenvector were calculated. The optimal number of PCs was selected using the Scree test according to Cattell’s law.¹⁸

Genome-wide interaction study

The GWIS statistical model for each stage weight of sheep in this study was as follows.

$$\mathrm{y}=\mu +\mathit{\text{year}}+(\mathit{\text{BIRTH}}\_\mathit{\text{WT}})+\sum \mathrm{\text{PCs}}+{\mathit{\text{SNP}}}_1+{\mathit{\text{SNP}}}_2+{\mathit{\text{SNP}}}_1\mathrm{\hspace{0.17em}}\mathrm{*}\mathrm{\hspace{0.17em}}{\mathit{\text{SNP}}}_2+\mathrm{e}$$ (1)

where y is the phenotypic value (birth weight, weaning weight, yearling weight). The year is the year effect. PCs are the optimal number of principal component effects. BIRTH_WT is the birth weight effect and covariates when analyzing the weaning weight and yearling weight traits. SNP₁ and SNP₂ are genetic effects at a single SNP locus. SNP₁ * SNP₂ is the interaction effect at two SNP, and e is the random residual.

PLINK was used to identify genome-wide SNP-SNP interactions, dividing the GWIS model into two tiers: a linear model of phenotypes to correct for the year and BIRTH_WT effects (see Equation (2)) and an ‘epistasis’ parameter in PLINK to set the SNP-SNP interaction using regression model (see Equation (3)).

$$\mathrm{y}=\mathrm{\text{year}}+\left(\mathrm{\text{bw}}\right)+\mathrm{\text{PCA}}+\mathrm{y}\prime$$ (2)

$$\mathrm{y}\prime ={\mathit{\text{SNP}}}_1+{\mathit{\text{SNP}}}_2+{\mathit{\text{SNP}}}_1\mathrm{\hspace{0.17em}}\mathrm{*}\mathrm{\hspace{0.17em}}{\mathit{\text{SNP}}}_2+\mathrm{e}$$ (3)

Bonferroni correction was used to adjust multiple comparison tests for detected SNP-SNP interactions. In total, 120,818 independent SNPs were identified. For each trait, the threshold p value is $p={0.05/\mathrm{C}}_{120,818}^2=6.85\mathrm{E}-12,$ where n is the number of independent SNPs.

The contribution rate of significant SNP-SNP interactions to phenotypic variation was calculated by Equation (4).

$$C=\frac{{\mathrm{\text{SS}}}_{\mathrm{\text{SNP}}1\mathrm{\hspace{0.17em}*\hspace{0.17em}}\mathrm{\text{SNP}}2}}{\text{Phenotypic variance}}\times 100\mathrm{\%}$$ (4)

where ${\mathrm{\text{SS}}}_{\mathrm{\text{SNP}}1\times \mathrm{\text{SNP}}2}$ is the variance of the significant SNP-SNP interactive effect.

Gene annotation analysis and functional enrichment analysis

The candidate genes closest to the significant SNPs were filtered within the upstream and downstream 200 kb of the SNPs via the Ovis aries (oarv3.1) genome assembly. The functional annotation of candidate genes were used the David database (https://david.ncifcrf.gov/),¹⁹ including Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses.

Functional prediction of non-synonymous single nucleotide polymorphisms

Some significant non-synonymous single nucleotide polymorphisms (nsSNPs) alter the type of amino acids translated. NsSNPs function was predicted by the online tools PredictSNP²⁰ and Mupro.²¹ A score of 0 to 1 or −1 to 0 in PredictSNP classified nsSNPs as deleterious or neutral, respectively. The free energy change (DDG) values were calculated using MUpro. The DDG values between −0.05 and 0.05 were considered neutral; the values > 0.05, or < −0.05 were considered to increase or decrease protein stability, respectively.

Prediction of transcription factor target genes

The position weight matrix (PWM) of sheep transcription factors was downloaded from the cisbp database.²² The biomaRt package²³ in the R language was used to extract 2 kb sequences of upstream genes. The TFBSTools package²⁴ was used to predict transcription factor target genes with a threshold relscore value of 0.7.

Results

SNPs marker information

A total of 479,470 SNPs on autosomal and X chromosomes of sheep were included in this study (see Table 1). The average distance between adjacent markers was 5.40 kb. According to the results of the HAPLOVIEW, 120,818 separate SNPs were obtained. Based on scree plot analysis (see Fig. 1), the first 26 components were selected to correct for population stratification in the study.

Table 1.

Distribution of SNPs on each chromosome in sheep.

chromosome	Chromosome length (megabytes)	SNP number after quality control	Average distance (kilobytes)	Number of independent single nucleotide species
1	275.61	50,735	5.43	12,930
2	248.99	45,827	5.43	11,248
3	224.28	42,835	5.24	10,442
4	119.26	21,753	5.48	5642
5	107.90	19,989	5.40	4911
6	117.03	20,549	5.70	5405
7	100.08	18,795	5.32	4763
8	90.70	16,466	5.51	4423
9	94.73	17,130	5.53	4327
10	86.45	15,183	5.69	3829
11	62.25	12,809	4.86	3223
12	79.10	15,282	5.18	3853
13	83.08	16,298	5.10	3984
14	62.72	12,905	4.86	3318
15	80.92	15,213	5.32	3786
16	71.72	13,023	5.51	3477
17	72.29	13,178	5.49	3529
18	68.60	13,062	5.25	3316
19	60.46	11,946	5.06	2959
20	51.18	9926	5.16	2642
21	50.07	9532	5.25	2601
22	50.83	9623	5.28	2398
23	62.33	11,385	5.47	2989
24	42.03	8693	4.83	2209
25	45.37	8682	5.23	2281
26	44.08	8329	5.29	2341
X	135.44	20,322	6.66	3992
total	2587.5	479,470	5.40	120,818

Figure 1.

Gravel diagram of sheep population based on SNP.

Genome-wide statistical SNP pairwise interactions

In this study, PLINK was performed to identify genome-wide SNP-SNP interactions associated with body weight at different stages. The p value profiles (Q-Q plots) for each trait are shown in Fig. 2. Lambda values for all traits are closed to 1. There were 2 and 66 SNP-SNP interactions for birth weight and weaning weight, respectively (see Fig. 2, Supplementary Table S1, and Supplementary Table S2), with no replicated significant pairs. No significant epistatic interactions were detected for yearling weight. The p value for the interaction effect between Affx281004149 on OAR13 and Affx281155091 on OAR2 was the most significant effect associated with WEAN_WT, reaching 3.58E − 14.

Figure 2.

The Q-Q plots and the network diagrams for the genome-wide interaction study. (A–C) The Q-Q plots of birth weight, weaning body weight, and yearling body weight, respectively. (D–E) The network diagrams of birth weight and weaning body weight, respectively.

There were two and six SNPs pair-wise interaction on the same chromosome associated with BIRTH_WT and WEAN_WT, respectively, which is also referred to as the intrachromosomal epistasis effect. Other SNP-SNP interactions belonged to the inter-chromosomal epistasis effect, accounting for 88.24%. The location of each SNP is shown in Supplementary Table S3. Interestingly, for SNP pair-wise interactions associated with WEAN_WT, we found several SNPs (located in the same LD region or gene) were interacted with the same locus.

The contribution rates of the two SNP-SNP interactions to BIRTH_WT were 0.18% and 0.07%, respectively. The contribution rates of significant epistatic SNP pairs associated with WEAN_WT ranged from 0.18% to 6.72%. However, this is likely an overestimate.

Annotation analysis of SNP loci

Significant SNP-SNP interactions associated with BIRTH_WT included four SNPs. Of these SNPs, Affx280965374 was located in the fourth exon of the CNTLN (Centlein) gene. For the WEAN_TW trait, 66 significant SNPs pairs consisted of 98 single SNPs covering 23 chromosomes (see Fig. 2), and 5 SNPs were non-synonymous single nucleotide polymorphisms (nsSNPs) leading to a single amino acid substitution.

To predict the functional effects of nsSNPs on body weight in different stages of sheep, the effects of mutations on protein function and stability were analyzed using Predictsnp and MUpro software (see Table 2), respectively. The results showed that the mutations of all nsSNPs could reduce the stability of the corresponding proteins. Affx280920580 and Affx281060312 were classified as neutral mutations according to the PredictSNP score, while the other nsSNPs were shown to belong to deleterious mutations.

Table 2.

Predicted function of nsSNPs loci associated with weight traits.

Traits	OAR	SNP name	Position (Mb)	Gene ID (mutation position)	Gene name	Mutation	PredictSNPScore	MUpro DDG(kj/mol)
BIRTH_WT	2	Affx280965374	85.10	ENSOARG00000014084 (4th Exon)	CNTLN	K109W	0.76	−0.47
WEAN_WT	3	Affx280854322	146.42	ENSOARG00000019913 (2th Exon)	–	C1118W	0.87	−0.87
WEAN_WT	4	Affx280920580	9.31	ENSOARG00000017010 (30th Exon)	AKAP9	E2119Q	−0.6	−0.7
WEAN_WT	4	Affx280891687	30.37	ENSOARG00000010841 (13th Exon)	DNAH11	L718H	0.52	−1.99
WEAN_WT	10	Affx280762335	69.62	ENSOARG00000000821 (11th Exon)	GPR180	I398M	0.72	−0.48
WEAN_WT	11	Affx281060312	8.79	ENSOARG00000008940 (11th Exon)	LPO	T585R	−0.63	−0.97

In total, all significant SNP interactions associated with BIRTH_WT and WEAN_WT were located on 2 and 65 candidate genes, respectively, including ABCC4, CRYL1, GPR180, and OLFM4 genes (see Table S3). Functional enrichment analysis using David tools showed that these candidate genes were found to be enriched in the ATP-binding pathway (molecular function, GO: 0005524, p value = 6.80E − 04).

Among these candidate genes, eight transcription factors were found, such as HIVEP3, BACH2, FOXB1, and GLI3. The PWM matrices of these TFs were collected and their target genes were predicted (see Table 3). The results showed that the genes interacting with TFs were the target genes of the corresponding TFs, with relscore values greater than 0.70.

Table 3.

Interactions between transcription factors associated with weaning weight and their target genes.

OAR	TF gene	OAR	Target gene	p Valuea	Beta	Stat.	Contribution rate (%)	Opt. genotypea	Relscore value
1	HIVEP3	3	ABCC4L	2.67E − 13	15.58	53.45	5.45	CC-GG	0.917
1	HIVEP3	10	GPR180	6.56E − 12	14.80	47.17	4.02	CC-CC	0.897
1	HIVEP3	10	ENSOARG00000000905	1.99E − 13	15.70	54.03	5.19	CC-GG	0.756
1	HIVEP3	10	ABCC4	1.46E − 13	15.95	54.64	2.54	CC-TT	0.870
1	HIVEP3	10	ENSOARG00000001021	5.01E − 12	14.85	47.70	4.42	CC-AA	0.802
4	SP4	2	LOC101120225	3.21E − 12	12.32	48.57	2.80	GA-TG	0.841
4	GLI3	4	DNAH11	6.22E − 12	−7.84	47.27	2.57	CT-AA	0.738
7	FOXB1	10	OLFM4	1.31E − 12	−7.49	50.32	3.76	CC-AG	0.829
8	BACH2	3	NBAS	4.25E − 12	19.12	48.02	3.30	GG-AA	0.796
8	BACH2	4	AKAP9	4.11E − 12	−12.4	48.08	1.10	AG-GA	0.705
8	BACH2	10	CRYL1	7.93E − 13	14.64	51.31	0.89	TT-AA	0.734
18	HMG20A	26	DLGAP2	4.01E − 12	12.53	48.13	0.18	AG-AG	0.970
23	ZNF407	5	HSPA4	4.47E − 12	−8.81	47.92	4.07	TT-GG	0.886

^aThe most significant SNP pairwise interaction in TF-targets.

Epistasis subnetworks associated with weaning weight traits

This study constructed four epistatic subnetworks affecting weaning weight using significant SNP pairs, including epistasis subnetwork associated with the HIVEP3 transcription factor, epistasis subnetwork on BACH2 transcription factor, epistatic subnetwork between OAR2:231.22 Mb and OAR20:30.05–30.07 Mb, and intrachromosomal epistasis subnetwork of OAR18.

The HIVEP3 gene is a transcription factor and plays a regulatory role in adult bone formation. Target gene prediction analysis indicated that the HIVEP3 gene might regulate the ABCC4L gene and a series of genes, including GPR180, and ABCC4 gene in the region of OAR10:69.61–70.31 Mb. According to the GWIS for weaning weight traits, the Affx280781222 SNP located 119.17 kb downstream of the HIVEP3 gene interacted with seven SNPs in the OAR10:69.61–70.31 Mb region and the Affx281264807 SNP in the intron of the ABCC4L gene, respectively (see Fig. 3 and Table 3). Of these epistatic effects, the most significant interaction was that between Affx280781222 and Affx280752933 (on the intron of the ABCC4 gene), with a contribution rate to weaning weight traits of 4.72%. The results showed that sheep with the CC genotype of the Affx280781222 SNP and the TT genotype of the Affx280752933 SNP had the highest weaning weight (p = 2.5E − 2).

Figure 3.

HIVEP3 transcription factors and associated epistasis subnetworks. (A) Epistasis subnetwork diagram. (B) Regulatory relationships between the HIVEP3 transcription factor and its target genes. (C) Distribution of weaning weight-differential genotype combinations of Affx280781222–Affx280752933 SNP interaction.

The TF encoded by the BACH2 gene is an essential regulator of the human immune system and is believed to have played an essential role in the maintenance of regulatory T cell function and B cell maturation. Some studies have also reported that this gene is associated with intramuscular fat content and milk fatty acids in cattle. CRYL1, AKAP9, and NBAS genes were predicted as target genes. In this study, we found that three SNPs located in the first intron of this gene interact with the Affx280811908 SNP in close proximity to the CRYL1 gene. In addition, SNP-SNP epistatic effects were identified between the Affx281247773 SNP located in the fourth intron of the BACH2 gene and four SNPs of the AKAP9 gene and one SNP of the NBAS gene, respectively (see Fig. 4 and Table 3). In this subnetwork, the Affx280864943–Affx280811908 interaction was detected as the most significant SNP-SNP interaction, with a contribution rate to weaning weight traits of 0.89%. The optimal genotype combination was TT-AA associated with weaning weight (p = 3.5E − 2). The contribution rate of Affx281247773–Affx122821168 corresponding to the BACH2-NBAS gene interaction had the highest weaning weight in this subnetwork at 3.30%, and the AA-GG combination genotype has the highest weaning weight (p = 1.3E − 2).

Figure 4.

BACH2 transcription factors and associated epistasis subnetworks. (A) Epistasis subnetwork diagram. (B) Regulatory relationships between the BACH2 transcription factor and its target genes. (C) Distribution of weaning weight-differential genotype combinations of Affx281247773–Affx122821168 SNP interaction.

The interaction between two SNPs on OAR2:231.22 Mb and three SNPs on OAR20:30.05-30.07 Mb was significantly associated with the weaning weight trait (see Fig. 5), explaining 2.80% of the phenotypic variation. This result indicates that an interaction occurs between the two regions on OAR2 and OAR20, corresponding to the SP4-LOC101120225 gene interaction (see Fig. 5). The most significant SNP pairwise interaction was Affx280800964–Affx122817026, whose optimal genotype was TG-GA (p = 2.8E − 2).

Figure 5.

Epistatic subnetwork between OAR2 and OAR20 and within OAR18. (A–B) Diagrams of the two epistasis subnetworks, respectively. (C–D) Distribution of weaning weight-differential genotype combinations in the most significant interaction of the two epistasis subnetworks.

In addition, OAR18 has a gap of approximately 58 Mb between genes, confirming the effect of intrachromosomal epistasis: three SNPs in the fourth intron of GABRG3 (OAR18: 4.18 Mb region) interacted significantly with the Affx280901120 marker (on OAR18: 62.23 Mb region 129.07 kb downstream of the ZNF674 gene) (see Fig. 5). The most significant interaction was between Affx280940339 and Affx280901120, explaining 4.63% of the phenotypic variation. Individuals with the AA-TG genotype had higher weaning weight than other genotype combinations (p = 1.12E − 5).

Discussion

GWAS is an effective method for searching for candidate genes associated with quantitative traits. However, these genetic variants and candidate genes explain only part of the phenotypic variance, leading to missing heritability. For example, the heritability of human height is reported to be about 0.80 by classical estimation methods, but at least 40 loci are associated with height, explaining about 5% of the phenotypic variance according to a study of tens of thousands of individuals.²⁵ The vast majority of heritability remains unresolved. Much of the literature concurs, and much of the missing heritability can be attributed to genetic interactions.^26–28 A GWAS using the same animal population as the present study detected several SNPs loci for the WEAN_WT trait but could explain only about 20% of the phenotypic variation (Without publication). In this study, we found that the majority of significant interactions for WEAN_WT contributed more than 1.00%. This result suggests that genetic interactions are likely to make a significant contribution, although these values may be overestimated.

Body weight trait is a classical quantitative trait with medium to high heritability. Body weight at each developmental stage from birth to adulthood is also an important breeding indicator. Few studies have detected SNP pairwise interaction loci associated with body weight in the chicken, bovine and human genomes. For example, Yuan et al.²⁹ showed significant gene-gene interaction effects on the body weight phenotypes from birth to 15 weeks of age in chickens. Li and Li³⁰ identified 67, 4, and 2 significant SNP-SNP interactions associated with weight at birth, 1 and 3 weeks of age in broilers, respectively. Reverter et al.³¹ identified SNPs with buffering epistasis effects for body weight at yearling in beef cattle. For the human genome, Dong et al.³² and Sadeghi et al.⁴ identified CAND2 and EXOC4 as novel susceptibility genes for obesity and body mass index, respectively, through genome-wide interaction association studies. To the best of our knowledge, this study was the first to detect genome-wide SNP-SNP interactions affecting body weight at different stages in sheep. These results may help to elucidate novel candidate genes and potential biological mechanisms of these genes involved in body growth and development. However, caution should be exercised in interpreting these results, as the potential mechanisms of epistatic interactions between these genes will need to be tested in future studies.

Our results showed that only a few SNP markers were significant in a GWAS study (Not published) among the significantly interacting SNPs. This result suggested that both genes may have at the same time a weak (or even no) marginal effect but an important effect through their interaction. Moore³³ deemed that complex interactions may be important while the individual gene effects are weak. Kotti et al.³⁴ provided a new locus-by-locus detecting strategy based on the classical TDT, which allows the detection of the involvement of two genes without individual effect or with weak marginal effect. Winham et al.³⁵ developed the SNPInterForest tool to identify interactions between SNPs without marginal effects by improving the random forest framework. Wu et al.³⁶ reported that the majority of the significantly interacting SNPs showed no marginal association with human psoriasis. These results indicated that interactions between genes with weak marginal effect, should not be ignored. If only testing interaction among genetic variants with significant marginal would lose the majority of interactions.

We detected SNP-SNP interactions for a variety of growth traits in sheep. The results showed that genetic epistasis effects varied by time period, with dramatically more SNP pairwise interactions associated with WEAN_WT than with BW. Epistasis is an important contributor to the genetic variance of growth and, as Carlborg et al.³⁷ showed, has the greatest effect on early growth. Yuan et al.²⁹ and Li and Li³⁰ analyzed epistasis associated with different stages of weight traits in different chicken breeds. They draw similar conclusions that there may be different genetic regulations for early and late growth and that epistasis is more important for early growth than for late growth.

According to known gene functions, several candidate genes were associated with body weight at different developmental stages in other species, including GLI3, ACSL3, GABRG3, and OLFM4 genes (see Table 4). The GLI3 gene was associated with multi-stage weight in Nanyang cattle,³⁸ Yancheng chicken,³⁹ and broilers.⁴⁰ Polymorphisms in the ACSL3 gene were associated with growth traits in Dezhou donkeys⁴¹ and were a novel direct LXR (liver X receptor) target gene with a regulatory role in fatty acid metabolism in human placental trophoblast cells.⁴² Methylation within the GABRG3 gene was associated with human birth weight. Deyssenroth et al.⁴³ observed that the OLFM4 gene was associated with childhood overweight.⁴⁴

Table 4.

Important candidate genes related to body weight.

Gene name	Location (OAR:start–end, Mb)	Full name	Gene function
HIVEP3	1:15.83–15.89	HIVEP zinc finger 3	Regulated the osteochondrogenesis.54
SLC35D1	1:42.30–42.34	Solute carrier family 35 member D1	Is a lethal skeletal dysplasia.55,56
ACSL3	2:224.07–224.10	Acyl-CoA synthetase long chain Family member 3	Associated with growth traits in Dezhou donkey;41 participated in fatty acid metabolism in human placental trophoblast cells.42
AKAP9	4:9.18–9.35	A-kinase anchoring protein 9	Governed preadipocyte differentiation through the PKA signaling.47
DNAH11	4:30.33–30.68	Dynein axonemal heavy chain 11	Associated with serum lipid levels at a genome-wide significance level.49
GLI3	4:79.03–79.13	GLI family zinc finger 3	Related to the multi-stages body weight of Nanyang cattle,38 Yancheng chicken,39 and broilers.40
BACH2	8:47.19–47.48	BTB domain and CNC homolog 2	As a key regulators of the bovine milk fatty acid metabolism.50
OLFM4	10:11.00–11.03	Olfactomedin 4	Associated with childhood overweight.44
CRYL1	10:36.12–36.18	Crystallin lambda 1	Significantly different in skeletal muscle from the control group and obese group (fed a high high-fat diet) in rabbits.45
GPC6	10:68.81–68.81	Glypican 6	Stimulated Hedgehog signaling to promote the growth of developing long bones;52 associated with intramuscular fat in the pig longissimus dorsi muscle.53
GPR180	10:69.59–69.62	G protein-coupled receptor 180	Participated in TGFβ signaling that promotes thermogenic adipocyte function.48
ABCC4	10:69.97–70.13	ATP-binding cassette transporter family class C4	Affected body weight and adipose tissue in mouse.51
GABRG3	18:3.77–4.24	Gamma-aminobutyric acid type A Receptor subunit gamma3	Associated with birth weight.43
AKAP6	18:41.63–42.12	A-kinase anchoring protein 6	Played an important role in skeletal myotubes.46

Genes involved in tissue formation such as muscle, fat, and bone may play an important role in body weight. Based on the literature, candidate genes such as CRYL1, GPR180, GPR180, and HIVEP3 may be associated with body weight. The CRYL1 gene was significantly expression different in skeletal muscle of control and obese (fed a high-fat diet) rabbits. Li et al.⁴⁵ and Becker et al.⁴⁶ suggested that the AKAP6 gene played an important role in skeletal myotubes. The AKAP9 gene governed preadipocyte differentiation via PKA signaling. Zhang et al.⁴⁷ and Balazova et al.⁴⁸ showed that the GPR180 gene was a component of TGFβ signaling that promotes thermogenic adipocyte function. Aulchenko et al.⁴⁹ established the DNAH11 gene associated with serum lipid levels at genome-wide significance levels. Pegolo et al.⁵⁰ found the BACH2 gene as an important regulator of bovine milk fatty acid metabolism. Donepudi et al.⁵¹ found that the ABCC4 gene affected body weight and adipose tissue in mice. The GPC6 gene stimulated hedgehog signaling and promoted the growth of developing long bones.⁵² The GPC6 gene was also found to be associated with intramuscular fat in the longissimus dorsi muscle of pigs.⁵³ The HIVEP3 gene regulated osteochondrogenesis.⁵⁴ The loss-of-function mutation in the SLC35D1 gene caused a lethal skeletal dysplasia.^55,56

Conclusions

In conclusion, this is the first genome-wide interaction study in sheep to identify epistatic factors affecting body weight. Two and 66 SNP-SNP interactions were significant for birth weight and weaning weight, respectively. The contribution rate of most of the significant interactions to weaning weight was greater than 1%. Four epistatic subnetworks affecting weaning weight constructed using significant SNP pairs were also analyzed and annotated, including subnetworks on HIVEP3 and BACH2 transcription factors. These results provide a new reference for the study of genetic mechanisms of sheep body weight.

Funding Statement

This work was supported by Young and middle-aged scientific and technological innovation leading talent plan of the Xinjiang Production and Construction Corps (2019CB019), the Guide Project of State Key Laboratory of Sheep Genetic Improvement and Healthy Production (SKLSGIHP2016A01), Major Scientific and Technological Project of the Xinjiang Production and Construction Corps (2017AA006), the Open Project of State Key Laboratory of Sheep Genetic Improvement and Healthy Production (MYSKLKF202001), and National Natural Science Foundation of China (32070571).

Ethical approval

This study was approved by the Experimental Animal Care and Use Committee of the Xinjiang Academy of Agricultural and Reclamation Sciences (Shihezi, China, approval number: XJNKKXY-AEP-039, January 22 2012). All procedures and animal collections were also approved by the Northeast Agricultural University (Harbin, China) Animal Care and Treatment Committee (IACUCNEAU20150616). The guidelines for the Care and Use of Laboratory Animals were carefully adhered to.

Author contributions

Hua Yang, Zhipeng Wang, Peng Zuo, and Chaoxin Zhang conceived the study. Chaoxin Zhang, Yonglin Yang, and Qian Yu were involved in the acquisition of data, and Chaoxin Zhang, Yupeng Gao, Lijunyi Zhao, and Jiaxu Guo performed all data analysis. Zhipeng Wang, Chaoxin Zhang, Peng Zuo, and Hua Yang drafted the manuscript, and Yupeng Gao, Lijunyi Zhao, Jiaxu Guo, and Yunna Li contributed to the writing and editing. All authors read and approved the final manuscript.

Disclosure statement

The authors declare no conflict of interest.

Data availability statement

The variation data reported in this article have been deposited in the Genome Variation Map (GVM) in Big Data Center, Beijing Institute of Genomics (BIG), and Chinese Academy of Sciences, under accession numbers GVM000068 that are publicly accessible at https://bigd.big.ac.cn/gvm/getProjectDetail?project= GVM000068. The Bioproject accession number is PRJCA002639.