Multi-omics dissection of the genetic regulation underlying fatty acid composition in cattle

Abstract

Background

Fatty acid composition is a complex trait governed by a polygenic architecture. Although numerous genetic variants associated with fatty acids have been identified through genome-wide association studies (GWAS), the underlying regulatory mechanisms remain incompletely understood. In this study, we employed a multi-omics approach to deeply dissect the genetic regulation of fatty acids in cattle.

Results

Our integrated multi-strategy GWAS framework identifies 118,697 candidate variants spanning 4,234 distinct genes associated with fatty acid compositions. Parallel expression quantitative trait locus analyses across 227 muscle and 116 adipose tissue samples reveal 673,478 significant variant-gene pairs. By integrating multi-omics datasets, including DNA methylation, histone modifications, and chromatin accessibility, we systematically annotate the regulatory landscape of fatty acid associated variants. This comprehensive analysis identifies genes, such as SCD and ACAA2, as key regulators of fatty acid synthesis and metabolism. We further characterize a regulatory variant, chr26:21,274,156, located within the same topologically associating domain as the SCD promoter and enhancer elements, potentially facilitating a 3D chromatin-mediated cis-regulation within the locus. Functional validation through dual-luciferase reporter assays and CRISPR/Cas9-mediated perturbation confirm that this variant enhances transcriptional activity and directly modulates fatty acid profiles.

Conclusions

By integrating multi-omics profiles, this study provides new insights into the genetic regulation of fatty acid synthesis and metabolism. Our findings highlight target genes, regulatory elements, and functional variants that influence fatty acids, offering valuable targets for the genetic improvement of meat quality in livestock.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13059-025-03885-z.

Background

Beef is a primary source of animal-derived nutrition, providing essential nutrients for human health, such as high-quality protein, unsaturated fatty acids, minerals, and various vitamins [1]. Intramuscular fat deposition, which contributes to meat tenderness and flavor, is a desirable characteristic of meat quality. Fatty acids play a critical role in determining meat quality, influencing factors such as flavor, tenderness, and taste, thereby impacting consumers’ sensory experiences and preferences [2]. Research indicates that diets rich in monounsaturated fatty acids (MUFA) and polyunsaturated fatty acids (PUFA), both found in meat, may help reduce the risk of cardiovascular disease and diabetes while enhancing overall human health [3].

Recent advances in sequencing technology have facilitated the genetic dissection of complex traits in mammals, especially for humans and mice [4–6]. International consortia, such as the Functional ANnoTation Of Mammalian genome 5 (FANTOM5) [7] and the Encyclopedia of DNA Elements (ENCODE) project [8], have played pivotal roles in identifying functional genomic elements and uncovering transcriptional regulation across various cell and tissue types. Likewise, initiatives such as the Genotype-Tissue Expression (GTEx) project [9] and the NIH Roadmap Epigenomics Mapping Consortium [10] have been instrumental in studying the interplay between genetic variation [11, 12], gene expression, and epigenetic modifications across diverse cell and tissue types in the context of human health and disease [13, 14].

In farm animals, the Functional Annotation of ANimal Genomes (FAANG) Consortium was established to map functional elements in domesticated species by integrating data on RNA expression, DNA methylation, histone modification, chromatin accessibility, and protein-DNA interaction [15–18]. Furthermore, the Farm Animal Genotype-Tissue Expression (FarmGTEx) Consortium has provided a comprehensive public resource for discovering tissue-specific genetic regulatory variants and predicting molecular phenotype in farm animals, enabling comparative genomics studies across species [19, 20]. The integration of multi-layer genomic data has expedited the identification of candidate variants and deepened our understanding of the genetic regulation of economically important traits in livestock [21, 22]. Furthermore, coupling advanced genomic selection methods with functional annotation has promised to improve the selection of superior genotypes and enhance genetic gains in breeding programs [17, 23, 24].

Fatty acid traits are influenced by complex polygenic inheritance patterns [25]. Key genes involved in fatty acid biosynthesis and metabolism, such as stearoyl-coA desaturase (SCD), fatty acid synthase (FASN), and elongation of very long chain fatty acids protein 6 (ELOVL6), have been reported across multiple species [26–31]. While genome-wide association studies (GWAS) have revealed genetic variants and candidate genes associated with fatty acid traits in farm animals [32–34], multi-omics approaches aimed at deciphering the regulatory mechanisms governing fat deposition and lipid metabolism have been limited, primarily to studies in chickens and pigs [22, 35]. In cattle, the functional characterization of regulatory elements (REs) and their genetic control over fatty acid traits remains largely unexplored.

In this study, we first integrated multi-strategy GWAS signals and cis-eQTL mapping to refine the associated loci for fatty acid traits. We then employed an integrative multi-omics framework to systematically investigate the functional regulation of fatty acids. Our approach combined chromatin accessibility profiling (ATAC-seq), histone modification (ChIP-seq), transcriptome profiling, DNA methylation, and Hi-C data to identify REs, including promoters, enhancers, silencers, and CTCF-binding regions. By linking functional noncoding variants to gene expression and fatty acid composition-associated loci, we investigated potential regulatory mechanisms. Finally, we experimentally validated selected genes and candidate variants to characterize their roles in fatty acid metabolism. This integrative analysis reveals how genetic variation shapes regulatory landscapes influencing lipid metabolism and identifies molecular targets for improving fatty acid profiles in cattle.

Results

Data summary and overview of the study

Summary statistics of the multi-omics datasets and a schematic overview of the study are provided in Additional file 1: Fig. S1. Transcriptome sequencing of muscle and adipose tissues from 343 samples yielded 7.98 billion high-quality reads, ranging from 5.61 to 10.99 Gb per sample (Additional file 2: Table S1). The individuals were genotyped using Bovine HD SNP array and then imputed into the sequence SNPs according to our previous study [36]. After quality control (QC), 12,102,431 SNPs were retained for eQTL analysis. We also generated 108.93 Gb of ATAC-seq clean reads from the ten samples (see next part for sample grouping) of muscle tissues (Additional file 2: Table S2), with an average of 48 million effective reads per sample, ranging from 43 to 56 million. Key quality metrics included normalized strand coefficient (NSC) values greater than 1.50, relative strand correlation (RSC) values exceeding 1.62, and average fraction of reads in peaks (FRiP) scores of 0.26 (ranging from 0.16 to 0.34). We also generated single-nucleotide resolution methylation profiles from the same ten muscle samples using whole-genome bisulfite sequencing (WGBS), obtaining 971.60 Gb of clean data (Additional file 2: Table S3). The average methylation percentages for CpG, CHG, and CHH were 72.99%, 0.33%, and 0.33%, respectively.

Fatty acid compositions measurement and sample grouping

To evaluate variation in fatty acid composition across muscle samples, we performed principal component analysis (PCA) on 29 fatty acid traits. The PCA revealed substantial diversity in fatty acid profiles, with the top five principal components (PCs) explaining ~ 90% of the total variance (Additional file 1: Fig. S2a). Based on the top two PCs, we selected ten muscle samples and categorized them into high (n = 5) and low (n = 5) fatty acid groups (Additional file 2: Table S4). Significant differences were observed between the two groups for three individual fatty acid compositions (C16:0, C18:1n9c, C18:0) and five fatty acid categories (MUFA, UFA, SFA, n9, TFA) (Additional file 2: Table S5). Also, we found that muscle samples were clearly divided into the high and low groups (Additional file 1: Fig. S2b). The top two PCs accounted for 82% of the total variance. Hierarchical clustering analysis further revealed clustering patterns consistent with those observed in the PCA (Additional file 1: Fig. S2c).

Multiple strategy GWAS identified candidate variants and genes associated with fatty acids

We performed single-trait GWAS on 29 fatty acid traits using 12,102,431 imputed variants. We totally identified 45,185 significant variant-trait associations for 18 traits at a genome-wide level (FDR ≤ 0.05). Of these, 3,591 unique candidate genes were identified within ± 50 kb of significant SNPs. Particularly, a candidate variant (chr24:49,396,936) for C22:5 was located 45.50 kb upstream of ACAA2. Additionally, a pleiotropic variant (chr26:21,255,813) for C20:0 and C20:4 was identified within the PKD2L1-SCD locus. Seven non-coding candidate variants associated with C14:0 and C14:1 were also located within the CCDC57-FASN locus. Furthermore, we performed a multi-trait GWAS based on summary statistics, identifying 74,093 potential pleiotropic variants at a genome-wide significant level (FDR ≤ 0.05) (Fig. 1a). Specifically, we identified the PKD2L1-SCD and CCDC57-FASN loci harbored 27 and 15 significant variants, respectively. Then, we performed gene-based GWAS using 26,101 protein-coding genes and identified 1,733 genes associated with the 29 fatty acids (Additional file 2: Table S6). Notably, the FASN gene was associated with C14:1 (FDR = 2.08 × 10^–6) and C14:0 (FDR = 2.29 × 10^–10) (Fig. 1b), while SCD was associated with C18:0 (FDR = 2.20 × 10^–5) (Additional file 2: Table S6).

Fig. 1.

Multiple strategy GWAS and transcriptome analyses for fatty acid traits. a-b Manhattan plots for -log₁₀ (P-values) for genes associated with fatty acids using single trait GWAS (a) and gene-based GWAS analyses (b). c-d Cattle QTLdb annotation of cis-eQTLs in muscle (c) and adipose (d) tissues, respectively. e Heatmap showing relatively high correlation between 12 modules and 16 fatty acid traits. Boxes contain Pearson correlation coefficients and associated P values. Red indicates positive correlation. Blue indicates negative correlation. f-h Examples of colocalization between cis-eQTLs of the ACSL1 gene in muscle and GWAS loci of C20:5 in cattle. The top colocalized SNP (rs43167612) is the top GWAS signal of C20:5 (f) and the second top cis-eQTL of ACSL1 (g). The Pearson correlation between P-values from ACSL1cis-eQTLs in muscle and GWAS of C20:5 (h).

Transcriptomics and colocalization reveal key variant-gene pairs for fatty acid compositions

To explore regulatory mechanisms, we carried out eQTL analysis and identified 672,779 cis-eQTLs in muscle tissues (Additional file 2: Table S7) and 1,607 cis-eQTLs in adipose tissues (Additional file 2: Table S8). Using the GALLO package, we observed that 47.53% and 37.38% of cis-eQTLs were enriched for meat and carcass traits in muscle and adipose tissues, respectively (Fig. 1c-d). Furthermore, we detected 5,050 cis-eQTL genes (eGenes) in muscle tissues (Additional file 2: Table S7) and 293 eGenes in adipose tissues (Additional file 2: Table S8). Among the eGenes, ACAA2, THEM4, HADH, and PPT1 were implicated in the fatty acid elongation process, SCD, FADS1, ACOT2, and ACOX3 were associated with biosynthesis of unsaturated fatty acids, and ACADL, ACADS, ACSL1, and EHHADH were related to fatty acid metabolism and fatty acid degradation. Expression levels of ACOT2, ACADL, and ACSL1 were significantly higher in the high fatty acid group compared to the low group (P < 0.05). Notably, the integration of GWAS and eQTL results revealed 650 shared genes, including key candidate genes associated with fatty acid traits, such as SCD and ACAA2.

To characterize transcriptional regulatory networks, we performed Weighted Gene Co-expression Network Analysis (WGCNA) and identified six modules highly correlated with 16 traits (Fig. 1e). Subsequent functional annotation of these modules identified 79 high-confidence candidate genes implicated in fatty acid biosynthesis and regulation (Additional file 2: Table S9), 12 of which (e.g., ACAA2, ACADS, and ACSF3) overlapped with eGenes. Comparative analysis showed significant upregulation of three regulatory genes (SREBF1, ABHD4, and TMEM120A) in high fatty acid group relative to low group.

Finally, we applied Bayesian colocalization (COLOC) and summary-based Mendelian randomization (SMR) to investigate causal relationships between GWAS loci and gene expression. We identified 260 GWAS loci across 29 fatty acid traits that colocalized with 221 unique genes. For example, rs43167612 showed high colocalization (PP.H4 = 1) with both ACSL1 expression and C20:5 levels (Fig. 1f-h). GWAS signals for C18:0 colocalized with cis-eQTLs of HIST1H1E, NCAPG, DYNC2I1, MED28, and SLIT2 (Additional File 2: Table S10). rs136385169, a variant for C18:2, colocalized with BIN1 expression (PP.H4 = 1) (Additional File 1: Fig. S3a-c). Furthermore, SMR analysis revealed 31 significant variant-gene-trait associations (Additional File 2: Table S11). Importantly, 95% (19/20) of the unique variants identified by SMR were also supported by COLOC, reinforcing their regulatory roles in fatty acid traits.

Chromatin accessibility landscape and functional variants/genes identification for fatty acids

We detected a total of 57,931 chromatin accessibility peaks shared between the high and low fatty acid groups. These peaks collectively span ~115 Mb, covering 4.28% of the genome (Additional file 2: Table S12). About 16% of these peaks were located in promoter regions (Fig. 2a), with strong enrichment around transcription start sites (TSS). Moreover, we assessed the distribution of peaks relative to the TSS of the nearest gene and found peak enrichment predominantly within the 10–100 kb range, followed by the 0–1 kb range (Fig. 2b).

Fig. 2.

Comprehansive analysis of open chromatin peaks and identification of fatty acid related genes. a Percentages of overlap between peaks and genomic features for muscle samples. b Percentage of peaks upstream and downstream from the TSS of their nearest genes. c Profile of peaks relative to TSS between high and low specific groups. d The top ten pathways for genes enriched in FACA peaks. e–f Expression regulation patterns of THBS1 (e) and INSIG1 (f) from multi-omics data, including RNA-seq, WGBS, and ATAC-seq. g The top eight motifs identified based on the FACA peaks

Comparative analysis of chromatin accessibility was conducted using two approaches. First, we identified group-specific peaks (GSPs), which were unique to either the high or low fatty acid groups. A total of 51,309 GSPs were detected, including 27,880 in the high fatty acid group and 23,429 in the low fatty acid group. Notably, GSPs from both groups were enriched around TSS regions, with high group-specific peaks also predominantly located within ± 3 kb of the TSS (Fig. 2c). Second, we identified differentially open chromatin regions (DOCRs), which exhibited significant differences in accessibility between the two groups. In total, 2,750 DOCRs were identified, including 1,479 upregulated and 1,271 downregulated regions in the high fatty acid group. By merging GSPs and DOCRs, we defined fatty acid related chromatin accessibility (FACA) regions as potential functional REs.

Next, we implemented an integrative annotation framework combining ATAC-seq and ChIP-seq data from muscle tissues. This annotation highlighted key lipid metabolism genes (FADS2, SCD, ACOX3, ACAA1, INSIG1, THBS1, ACAA2, etc.) were enriched in pathways such as adipocytokine signaling, glycerolipid metabolism, and PPAR signaling (Fig. 2d). Additionally, INSIG1 and THBS1 showed elevated chromatin accessibility and gene expression levels in the high fatty acid group (Fig. 2e-f). Motif analysis of peak signals revealed significant enrichment of transcription factor binding motifs for regulators such as PHF8, NRF1, STAT1, IRF1, etc., in the FACA regions, implicating these lipid metabolism regulators in the observed phenotypic differences (Fig. 2g). Remarkably, we identified a group-specific chromatin accessibility peak (i.e., peak_78080) located in an intron of the SCD gene in the high-specific fatty acid group.

To identify potentially functional variants for fatty acids, we systematically classified these FACA regions into distinct REs based on genomic features, using ChIP-seq data as a previous study [21]. Specifically, we identified 1,268 promoters, 1,596 enhancers, 1,962 silencers, 3,520 CTCF-enriched regions, and 37,299 low-signal regions (Fig. 3a). Promoter and enhancer regions exhibited stronger RE enrichments (Fig. 3b). We also identified 39 super-enhancers (SEs) and 35 super-silencers (SSs) (Fig. 3c-d), of which 22 SEs and 14 SSs overlapped with FACA peaks. REs in these regions showed significantly higher regulatory activity than REs in typical regions (Fig. 3e-f). To functionally characterize GWAS variants for fatty acids, we found 111, 115, 163, and 216 in FACA-promoter, -enhancer, -silencer, and -CTCF regions, as well as 1 and 2 variants in SEs and SSs, respectively (Fig. 3g). In addition, we identified 2,429, 2,562, 3,082, and 3,716 cis-eQTLs within FACA-promoter, -enhancer, -silencer, and -CTCF regions, along with 39 and 175 cis-eQTLs in SE and SS regions, respectively.

Fig. 3.

Chromatin accessibility for the annotation and identification of fatty acid related functional variants. a Number of annotated REs in FACA peaks. b Comparative analysis of chromatin marks in FACA peaks with violin plots showing activity signals of the data distribution. c-d Venn plots depicting the identification of SE (c) and SS (d) within FACA between the high and low FA groups. e–f Comparative analysis of chromatin marks in SEs (e) and SSs (f) vs. typical counterparts with violin plots showing activity signals of the data distribution. P values were obtained using an unpaired two-tailed Student’s t-test. g Number of annotated functional variants across different REs

Characterization of DNA methylation regulatory underlying fatty acid metabolism

To systematically characterize the epigenetic regulatory patterns, we detected 52,733 computationally predicted CpG islands (cCpGIs) and 28,732 experimentally supported CpGIs (eCpGIs) in muscle tissue (Additional file 2: Table S13). These eCpGIs were widely distributed across chromosomes and were particularly enriched in gene-dense regions (Fig. 4a). Furthermore, we identified 16,126 eCpGI-promoter region pairs, comprising 12,473 promoter regions and 14,394 eCpGIs (Additional file 1: Fig. S4a). The number of hypomethylated region (HMR) varied across chromosomes, with the highest number found on Bos taurus autosome 19 (BTA19) (Additional file 1: Fig. S4b). In addition, both the count and methylation levels of HMRs were higher in gene bodies compared to promoter regions (Additional file 1: Fig. S4c-d).

Fig. 4.

Integration analysis of DNA methylation and gene expression levels for fatty acid related candidate genes. a The distribution of eCpG islands across the chromosome. Blue to red indicates increasing gene density. b Significant positive correlation between the DMC66931310 methylation levels and TRAF6 expression (P < 0.05). c Significant negative correlation between the DMC1771450 methylation levels and ECI1expression (P < 0.05). d Regulatory features of the region overlapping DMC1771450, which was correlated with ECI1 expression. e Regulatory patterns of the region around DMC21288670, which was located within AGMO. The red lines represent the DMC position

To delineate epigenetic regulatory mechanisms underlying fatty acid metabolism, we conducted genome-wide comparative methylation profiling through DMC and DMR analyses between high and low fatty acid groups. We identified 23,804 differentially methylated cytosines (DMCs), including 12,690 hypermethylated and 11,114 hypomethylated sites, as well as 3,417 differentially methylated regions (DMRs), comprising 1,842 hypermethylated and 1,575 hypomethylated regions (Additional file 2: Table S14). These DMCs and DMRs mapped to 4,233 and 1,863 genes, respectively. Notably, hypermethylated genes were enriched in pathways such as fat digestion and absorption (e.g., FABP1, NPC1L1, PLPP3, PPARG) and lipid particle formation (ANXA2, TRAF6, RAB18, RAP1B), whereas hypomethylated genes were mainly linked to fatty acid biosynthesis (e.g., ACACB, ELOVL6, SCD, AGMO, HSD17B12) and fatty acid oxidation (e.g., ACAA2, ACOX1, ECI1). We also annotated 8 cis-eQTLs in DMR-promoters, 8 in DMR-enhancers, 8 in DMR-silencers, and 14 in DMR-CTCF regions, respectively (Additional file 2: Table S15).

Interestingly, two DMCs (DMC66931310 and DMC66961735) in the TRAF6 showed significant positive correlations with its expression (P < 0.05), suggesting regulatory interactions between DNA methylation and gene expression (Fig. 4b, Additional file 1: Fig. S5a). In contrast, negative correlations (P < 0.05) were observed between ECI1 and DMC1771450 (Fig. 4c) and between PLPP3 and DMC90074149 (Additional file 1: Fig. S5b). Notably, DMC1771450 was located near a CTCF peak, suggesting its potential involvement in transcriptional regulation (Fig. 4d). We also identified three genes (SCD, AGMO, and ACAA2) through GWAS that exhibited epigenetic regulation according to peak signals from DMC. For example, DMC23751544 was located within AGMO and was flanked by H3K27ac, H3K4me3, and CTCF peaks, suggesting a potential regulatory role in gene expression through epigenetic modifications (Fig. 4e).

Integrative multi-omics analysis identified two key candidate genes

By integrating GWAS, RNA-seq, WGBS, and ATAC-seq datasets, we identified 68 candidate genes involved in fatty acid metabolism. Among them, SCD and ACAA2 emerged as the most promising candidates (Fig. 5a). Functional annotation revealed significant enrichment in fatty acid related biological processes, including β-oxidation, lipid biosynthesis, and cellular lipid metabolic regulation (Fig. 5b).

Fig. 5.

Comprehensive analysis of the SCD gene expression regulation from multiple databases and experimental validation. a Number of shared and unique genes by GWAS, RNA-seq, WGBS, and ATAC-seq analyses. b Shared candidate genes of fatty acids involved in biological processes. c Expression regulation pattern of SCD in multi-omics data. The red box represents the promoter region of the SCD, and the red line indicates the position of the variant chr20:21,274,156. d SCD gene expression profile across 51 bovine tissues. e mRNA levels of SCD in adipocytes transfected with si-SCD−1, si-SCD−2, si-SCD−3, and si-NC. f The mRNA expression of fatty acid biosynthesis genes after transfection with si-NC and si-SCD. g Preadipocyte fatty acid composition changes after transfection with si-NC and si-SCD. h Preadipocyte triglyceride (TAG) content alteration after transfection with si-NC and si-SCD. qPCR and TAG data were presented as mean ± SD (n = 3), derived from three biological replicates (independent experiments) with three technical replicates per biological replicate for detection reproducibility. Fatty acid composition analyses were conducted across three independent biological experiments. * P < 0.05, ** P < 0.01, *** P < 0.001

We further explored regulatory mechanisms using integrative multi-omics analysis of SCD and ACAA2, including transcriptomics, chromatin accessibility, methylation, histone modifications (H3K4me1, H3K4me3, H3K27ac, H3K27me3), and CTCF binding sites, as reported previously [21]. Notably, we found open chromatin signals at the promoter regions of SCD and ACAA2, with predominantly hypomethylated levels. Additionally, H3K4me3 was enriched at the upstream promoter region of SCD (Fig. 5c), while several histone modification peaks (H3K4me3, H3K27ac) were enriched upstream of ACAA2 (Additional file 1: Fig. S6a).

We next examined tissue-specific expression profiles of SCD and ACAA2 across 51 tissues, as described in our previous analysis [37]. SCD was highly expressed in kidney fat, mesentery fat, subcutaneous fat, and heart fat (Fig. 5d), while ACAA2 was highly expressed in the liver (Additional file 1: Fig. S6b). SCD expression in various tissues was further validated using the CattleGTEx dataset [19], which confirmed high expression in adipose and intramuscular fat (Additional file 1: Fig. S6c). Comparative analysis across species and breeds revealed significantly higher SCD expression in Dzo cattle (cattle-yak hybrids) (Additional file 1: Fig. S6d), suggesting its diverse expression pattern among species. To evaluate the functional conservation of candidate genes between cattle and humans, we performed an Phenome-wide association studies (PheWAS) analysis (https://atlas.ctglab.nl/) on human orthologs across multiple traits. Our analysis revealed that SCD exhibited significant associations with metabolic traits, particularly stearate (C18:0, P = 3.48 × 10⁻⁴) and myristoleate (C14:1n5; P = 0.04) levels (Additional file 1: Fig. S6e). These findings support conserved roles of SCD and ACAA2 in fatty acid metabolism across mammals.

Experimental validation of SCD and ACAA2 for fatty acid regulation

We investigated SCD expression dynamics during bovine preadipocyte differentiation using a 6-day induction model (Additional file 1: Fig. S7a). SCD expression showed biphasic regulation: initial downregulation on day 1, transient upregulation on day 2, followed by progressive decline with extended differentiation (Additional file 1: Fig. S7b). All three synthesized SCD-specific siRNAs effectively suppressed SCD expression (Fig. 5e), with the most potent construct selected for downstream functional studies. SCD knockdown significantly upregulated three fatty acid biosynthesis genes including FASN, SREBF1, and ACACA (Fig. 5f), suggesting compensatory metabolic responses. Lipidomic profiling revealed the altered fatty acid composition in si-SCD-treated cells, with increased C14:1 (tetradecenoic acid), C17:0 (margaric acid), and C20:0 (arachidic acid), and decreased C20:2 (eicosadienoic acid) levels relative to controls (Fig. 5g). TAG quantification also revealed a significant decrease (P < 0.001) following SCD knockdown (Fig. 5h), corroborated by Oil Red O staining, which showed fewer lipid droplets in adipocytes (Additional File 1: Fig. S7c). Similarly, ACAA2 expression decreased rapidly during preadipocyte differentiation and remained at a lower expression level compared to control groups (Additional File 1: Fig. S7d). To further investigate ACAA2 function, three distinct ACAA2-specific siRNAs were synthesized and transfected into preadipocytes (Additional File 1: Fig. S7e). Post-interference analysis revealed significant downregulation of five fatty acid biosynthesis-related genes including SCD5, FASN, SREBF1, ACACA, and CAMKK2 (Additional File 1: Fig. S7f), implicating their roles in fatty acid biosynthesis and metabolic regulation. Similar to the SCD perturbation results, ACAA2 knockdown led to significant reductions in the levels of C8:0 (octanoic acid) and C10:0 (decanoic acid) compared to controls (Additional File 1: Fig. S7g). Similarly, TAG content was significantly decreased (P < 0.05) following ACAA2 knockdown (Additional File 1: Fig. S7h). Collectively, these molecular assay results reinforce the critical roles of SCD and ACAA2 in fatty acid metabolism.

Multi-omics dissection of regulatory variants near SCD

To explore regulatory variants influencing fatty acids, we performed a region-based GWAS on variants within 1 Mb upstream and downstream of SCD, identifying 37 suggestive variant-trait associations (FDR < 0.1). Of these, two variants (chr26:21,255,813 and chr26:21,274,156) were located 7,914 bp and 10,429 bp from the TSS of SCD, respectively. Notably, the regulatory variant chr26:21,274,156 was located at ~ 6 kb downstream of a muscle cis-eQTL region (Additional file 1: Fig. S8a-c), and was in high LD with two eVariants (chr26:21,267,520, chr26:21,268,154) (Additional file 1: Fig. S8d). We also performed a preliminary mediation analysis assessing the relationships among the variant, SCD gene expression and fatty acid compositions. Although the results showed that the average causal mediation effect of the chr26:21,274,156 variation on fatty acid levels was not significantly mediated by SCD expression (P value > 0.05, Additional file 2: Table S16). Both the average direct effect and total effect were significant for ten fatty acid traits (P value < 0.05). We further found that genotypes CC (chr26:21,267,520) and TT (chr26:21,268,154) were significantly associated with higher SCD expression (Additional file 1: Fig. S8e-f). Moreover, these eVariants, chr26:21,274,156, and the SCD gene (including its enhancer and promoter regions) were all within the same TAD derived from the Hi-C data (Fig. 6a). Notably, chr26:21,274,156 lies within an accessible chromatin region (chr26:21,271,904–21,274,367) in muscle tissue (Fig. 6a), located 455 bp downstream of an H3K27ac peak, while the presence of an eCpGI and HMR in the SCD promoter further implies regulatory activity. We also found the TC genotype of chr26:21,274,156 was significantly associated with C16:1 content compared to the CC genotype (Additional file 1: Fig. S8g). Additionally, TF-binding prediction in the SCD promoter revealed factors including E2F6, GABPA, ELF5, and ELF3. Motif disruption analysis showed that the C allele strongly disrupted nine TF motifs, especially within the ELF family, which is involved in fatty acid metabolism (Additional file 1: Fig. S9a). Our results also revealed that regions surrounding the variant had low DNA methylation and high chromatin accessibility within the PKD2L1-SCD TAD, as well as increased SCD expression in high fatty acid samples, suggesting a cis-regulatory mechanism.

Fig. 6.

Molecular validation of the candidate variant at chr26:21,274,156. a Expression regulation pattern of chr26:21,274,156 from multi-omics data, including RNA-seq, WGBS, ATAC-seq, ChIP-seq, and Hi-C. b Dual-luciferase assay results showing the enhancer activity in the region surrounding the SNP with different genotypes affecting the enhancement effect. c-f RT-qPCR analysis of SCD expression (c), Western blotting detection of SCD protein expression (d), Relative intracellular triglyceride levels (e), Oil Red O staining assessing lipid droplet content (f) between edited and control groups in 293T cells. P values were obtained using an unpaired two-tailed Student’s t-test. Dual-luciferase, qPCR and TAG data were presented as mean ± SD (n = 3), derived from three biological replicates (independent experiments) with three technical replicates per biological replicatefor detection reproducibility. Oil Red O staining analyses were conducted across three independent biological experiments. * P < 0.05, ** P < 0.01, *** P < 0.001

Functional validation of SCD candidate variant at chr26:21,274,156

We hypothesized that chr26:21,274,156 enhances SCD expression by interacting with its promoter via TAD structure. To test this, we performed allele-specific luciferase reporter assays using a 1,001 bp fragment containing either the T (wild-type) or C (mutant) allele cloned into a PGL4.23 (Luc2-minP) vector (Fig. 6b). Both constructs enhanced transcriptional activity relative to the control, while the T allele showed significantly greater activity (P < 0.05) (Fig. 6b), supporting its role as an allele-specific enhancer. To validate the function of surrounding regions on SCD expression, we first mapped this variant in bovine ARS-UCD1.2 to the human hg38. We then performed CRISPR/Cas9-mediated editing in 293T cells (Human embryonic kidney 293 cells, Laboratory Storage) using two sgRNAs. Sequencing showed high editing efficiency (sgRNA-1: 70.6%, sgRNA-2: 93.2%, sgRNA-1/2: 85.6%) (Additional file 1: Fig. S9b). Compared with the control group, edited cells had significantly reduced SCD mRNA (sgRNA-1, sgRNA-2, sgRNA-1/2, all P-values < 0.001) (Fig. 6c), SCD protein (sgRNA-1, P-value = 0.029; sgRNA-2, P-value = 0.003; sgRNA-1/2, P-value = 0.013) (Fig. 6d, Additional file 1: Fig. S10), and TAG content (all P-values < 0.001) (Fig. 6e). Oil Red O staining further confirmed a significant reduction in lipid droplet contents in the edited cells (sgRNA-1, P-value = 0.003; sgRNA-2 and sgRNA-1/2, P-value < 0.001) (Fig. 6f, Additional file 1: Fig. S9c). These results demonstrate that the chr26:21,274,156 region may enhance SCD transcription through allele-specific enhancer activity within a conserved regulatory domain.

Discussion

Cattle provide high-quality animal-derived nutrients, including bioavailable proteins and essential fatty acids, which play a key role in meat quality and human health [3, 38–40]. In this study, we employed an integrative multi-omics strategy—combining ATAC-seq, ChIP-seq, RNA-seq, WGBS, Hi-C, and genetic analyses—to dissect the regulatory mechanisms of fatty acid metabolism in cattle. We first integrated multi-strategy GWAS and cis-eQTL mapping to refine loci associated with fatty acid composition, and then identified critical variants and target genes for functional validation. This framework enables scalable discovery of causal variants and informs precision breeding.

Chromatin accessibility and histone modification profiles revealed active REs [41], including promoters, enhancers, and super-enhancers, enriched near lipid-related genes such as FADS2, SCD, ACOX3, ACAA1, and ACAA2. Key lipid metabolism transcription factors, including STAT1 and IRF1, were predicted to bind these regions [42, 43]. Consistent with prior studies in pigs and chickens [22, 44], REs showed strong activity in bovine muscle, highlighting their tissue-specific functions. We annotated > 200 variants within FACA regions, offering mechanistic insight into the genetic regulation of fatty acids.

We prioritized SCD and ACAA2 through integrative analysis of GWAS, cis-eQTLs, chromatin accessibility, and transcriptomic profiles. As DNA methylation contributes to regulatory complexity, we selectively investigated hypo- and hyper-methylated promoters overlapping key regulatory loci. With promoter hypomethylation near SCD and ACAA2 supporting their active expression, methylation analyses also help to refine candidate REs in conjunction with chromatin and expression data. SCD, encoding a stearoyl-CoA desaturase that converts saturated to monounsaturated fatty acids, showed high expression in adipose and muscle, co-localizing with open chromatin and active histone marks [45]. Similarly, ACAA2, involved in fatty acid degradation, displayed strong regulatory signals and association with fatty acid composition [46, 47]. These findings were consistent with public datasets (CattleGTEx) and literature across livestock species [19, 20, 48].

Previous studies have reported multiple variants within the SCD promoter associated with fatty acid traits in livestock [49, 50], but functional validation has been limited due to the lack of available multi-omics tools. In this study, we identified a novel variant (chr26:21,274,156) near the TSS of SCD, located approximately 6 kb downstream of a muscle eQTL region and in high LD with two other eVariants. This variant, along with SCD and its cis-eQTL signal (chr26:21,267,520–21,268,154), is located within a single TAD, suggesting coordinated regulation of SCD expression by these REs. Importantly, chr26:21,274,156 overlaps with regions marked by H3K27ac and open chromatin in high fatty acid animals, supporting that SCD expression is influenced by histone modifications, consistent with SCD’s key role in fatty acid synthesis and lipid metabolism [22].

The lack of significance of the average causal mediation effect in our mediation analysis may be due to limited sample size or complex regulatory mechanisms [51, 52]. The significant average direct effect and total effect results suggest that chr26:21,274,156 may affect fatty acid traits through alternative pathways, such as epigenetic marks or other regulatory mechanisms. Additionally, the dual-luciferase and CRISPR/Cas9 assays confirmed that this variant functions as a cis regulatory element within the PKD2L1-SCD locus, potentially modulating SCD transcription. We also functionally validated SCD and ACAA2 using siRNA knockdown and triglyceride assays, further supporting their regulatory roles in adipocyte differentiation and fatty acid metabolism.

Future research could leverage additional omics data from various developmental stages and tissue types, as well as long-read sequencing and single-cell RNA-seq technology [53, 54], to further elucidate the molecular drivers and regulatory basis of fatty acid metabolism. This holistic approach holds promise for advancing our understanding of fatty acid regulation in human health and improving genetic improvement programs in farm animals.

Conclusions

Our study establishes a multi-omics framework for identifying functional variants and regulatory elements underlying fatty acid metabolism in cattle. By integrating GWAS, cis-eQTL, and epigenomic data, we pinpointed and validated key genes like SCD and ACAA2 and their candidate regulatory variants. These findings advance functional genomics in livestock, provide biological priors for genomic prediction, and offer insights to improve meat quality through precision breeding.

Methods

Sample collection and fatty acid composition measurement

A total of 227 muscle and 116 adipose tissues were collected from Huaxi cattle, a breed derived from Chinese Simmental beef cattle. All animals were raised at Inner Mongolia Oaks Co., Ltd. in Ulgai, Xilingol League, Inner Mongolia, China. To minimize the influences of feeding strategies and environmental conditions, these cattle were fattened at Jingxin Xufa Agricultural Development Co., Ltd, and transferred to Inner Mongolia Zhongao Food Co., Ltd. for slaughter at approximately 24 months of age. The longissimus dorsi (12th-13th ribs) and subcutaneous adipose tissues were collected and preserved in RNAlater (Qiagen, Hilden, Germany) and snap-frozen in liquid nitrogen, while blood samples were collected for DNA extraction. The tissue processing and measurement of fatty acid composition were described in our previous study [55]. The fatty acid composition was identified by comparing the retention time of methyl esters of the samples with the GB/T 5009.168–2016 and commercial standard for fatty acids Supelco TM Component FAME Mix (cat 18919, Supelco, Bellefonte, PA, USA). The fatty acid composition was measured based on a gravimetric basis (g/100 g) and quantified as the weight percentage of total fatty acids.

DNA extraction, genotyping, and quality control

Genomic DNA was isolated from the blood samples using the standard phenol–chloroform extraction method and quantified with a NanoDrop-2000 spectrophotometer (Thermo Fisher Scientific, Inc, Waltham, MA). These individuals were then genotyped using the Illumina BovineHD (770 K) BeadChip. QC was performed using PLINK (v1.90), with the parameter (a threshold for minor allele frequency of 5%). Moreover, individuals with larger than 10% missing genotypes were excluded. A total of 643,456 SNPs were used for imputation. We then imputed the filtered SNPs on autosomes to the whole genome sequence level using Beagle 5.4 with the same setting, as described in our previous studies [36]. Finally, we retained 12,102,431 imputed variants for GWAS and eQTL mapping.

RNA library preparation and sequencing analysis

Total RNA extraction and RNA library construction were performed as described in our previous study [37]. The raw sequence data were generated using the Illumina NovaSeq 6000 system employing a paired-end strategy. The RNA library construction and sequencing were completed by Beijing Novogene Technology Co., Ltd. Raw sequencing reads were processed using FASTP (v0.21.0) [56] to remove reads containing poly-Ns, trim adapters, and filter low-quality reads. Clean reads were aligned to the ARS-UCD1.2 reference genome using HISAT2 (v2.1.0) [57], and alignment files were converted to BAM format and sorted using SAMtools (v1.9) [58]. Read counts were obtained with FeatureCounts (v1.5.2) [59] Gene expression levels were quantified using both read counts and FPKM (fragments per kilobase of transcript per million mapped reads).

Multi-strategy GWAS

We analyzed 29 fatty acid traits, including saturated, monounsaturated, and polyunsaturated fatty acids. Each trait was expressed as a percentage of total fatty acids, as previously described [28]. Genotype data from Niu et al. [36] were used and lifted over from UMD3.1 to ARS-UCD1.2 coordinates using liftOver. A total of 12,102,431 imputed variants were retained. Here, we performed multi-strategy GWASs for fatty acid traits, including sinlgle-trait GWAS, multi-trait GWAS, gene-based GWAS, and region-based GWAS.

Single-trait and multi-trait GWAS

SNP-based single-trait GWAS was performed using the GCTA-MLMA module (v1.93.1). Multi-trait association tests were conducted using GWAS summary statistics and chi-square-based meta-analyses, as described previously [36]. Significant variants were selected based on FDR ≤ 0.05. Candidate genes were defined within a ± 50 kb window around significant SNPs on the ARS-UCD 1.2 reference genome.

Gene-based GWAS

GWAS was conducted using MAGMA (v1.07), utilizing protein-coding gene annotations from ARS-UCD1.2.100.gtf (Ensembl) [60]. Three models—principal component regression, SNP-wise mean chi-square, and SNP-wise top chi-square—were used. A gene-wide significance threshold was set at P < 3.82 × 10⁻⁵.

Region-based GWAS

To reduce multiple testing bias, region-based GWAS was conducted using GCTA-MLMA, focusing on 2-Mb windows (± 1 Mb from TSS) for SCD and FASN. Candidate variants were selected using an FDR < 0.1 [61].

eQTL detection

Cis-eQTL mapping was conducted using FastQTL (v2.184) [62]. Variants within ± 1 Mb of each gene’s TSS were considered. The adaptive permutation mode (–permute 1000 10,000) was used to obtain beta distribution-extrapolated empirical P-values. Genes with significant eQTLs (FDR < 0.05) were defined as eGenes. In addition, nominal mode was run to evaluate variant-gene pairs, significant associations were defined by nominal P-values below the gene-level threshold.

Differential gene expression analysis

To investigate the expression changes associated with fatty acid composition, we selected ten half-sibling individuals from the same slaughter batch with similar genetic backgrounds. Based on extreme fatty acid composition (identified via PCA), samples were grouped into high and low fatty acid groups. Fatty acid content differences were assessed using descriptive statistics and paired t-tests. To detect intrinsic repeatability and outliers of samples, we used the read count values (no-normalized) as the input for DESeq2 (v1.32.0). The read counts were normalized using the rlog function. The samples were clustered and visualized by using the plotPCA function in DESeq2. Hierarchical clustering was implemented using the Pheatmap package (v1.0.12). Differentially expressed genes (DEGs) were identified with DESeq2 based on |log₂FC| and P < 0.05. Volcano plots were generated using the ggpubr package (v0.2.2).

Weighted gene co-expression network analysis (WGCNA)

WGCNA was performed to construct co-expression networks from muscle transcriptome data [63]. Genes were included based on two criteria: (1) expression level > 1 in at least 25% of individuals, and (2) coefficient of variation (CV) between 0.1 and 1. Hierarchical clustering of samples was conducted using hclust function. The soft-threshold power β was determined based on the scale-free topology criterion, with β = 7 selected as the optimal value. Module-trait correlations were assessed using Spearman correlation between modules and 24 fatty acid compositions, plus 11 fatty acid groups. Functional enrichment of module genes was performed using DAVID (v6.8), applying Bonferroni correction for multiple testing [64].

COLOC and SMR analysis

Colocalization and Mendelian randomization analysis between eQTL and GWAS loci for 29 fatty acid traits. To identify potential causal genetic variants linking eQTLs and fatty acid traits, we integrated colocalization and Mendelian randomization analyses to provide mechanistic insights into genetic architecture of fatty acids. First, using COLOC (v5.1.0) [65], we performed colocalization analysis between muscle eQTLs and suggestive GWAS loci (P < 10⁻⁴). We retained trait-gene pairs with a posterior probability of colocalization model H4 (PP.H4) > 0.8, indicating a high-confidence colocalization. Subsequently, we employed SMR (v1.03) to perform Mendelian randomization analysis [66], treating eQTLs as instrumental variables to infer causal effects of gene expression on fatty acid traits. Gene–trait associations were considered significant when the adjusted P_SMR < 0.05 and P_GWAS < 1 × 10⁻⁴.

ATAC-seq library preparation and sequencing

ATAC-seq libraries were constructed according to Halstead et al. [67]. Transposition reactions were performed using the transposase MIX from the Nextera DNA library preparation kit (Illumina, CA, USA), which contained transposase and two equal-mole adapters (adapter 1 and adapter 2), and were then incubated at 37 °C for 30 min. Transposed DNA was subsequently purified with the MinElute PCR Purification Kit (Qiagen, Hilden, Germany). The eluted DNA was amplified with custom-synthesized index primers using real-time PCR. After PCR amplification, libraries were purified with AMPure beads (Agencourt, Beckman Coulter Genomics, Beverly, MA, USA) and quantified by a Qubit 2.0 fluorometer (Thermo Fisher Scientific, Inc., Waltham, MA, USA). Clustering of the index-coded samples was performed on a cBot Cluster Generation System using TruSeq PE Cluster Kit v3-cBot-HS (Illumina, CA, USA) according to the manufacturer’s instructions. After cluster generation, the library preparations were sequenced on an Illumina HiSeq X Ten platform, generating 150 bp paired-end reads.

ATAC-seq chromatin accessibility analysis

Raw sequence reads were initially assessed for quality using FastQC (v0.11.5). Subsequently, the raw reads were filtered using FASTP (v0.21.0) [56]. The strict filtration criteria were to remove (1) reads with more than 40% of the total reads with a base quality value of less than 15; (2) reads with N more than 6 base; (3) adaptor reads; (4) reads shorter than 50 bp after trimming. Clean reads were then aligned to ARS-UCD1.2 using Bowtie2 (v2.2.7) [68] with default settings. BAM files were sorted using SAMtools (v1.9) [58] software. Peak calling samples were performed using Genrich (https://github.com/jsh58/Genrich) with -m 30, -j (ATAC-seq mode), -r (remove PCR duplicates), -e MT (to exclude mitochondrial chromosomes), -q 0.05 (FDR-adjusted p-value). Annotation of ATAC-seq peaks was performed with TxDb.Btaurus.UCSC.bosTau9.refGene and org.Bt.eg.db using the R package ChIPseeker (v1.6.7) [69]. Briefly, the ChIPseeker covplot function was employed to calculate and visualize the coverage of peak regions across chromosomes. The profile of peaks binding to transcription start regions (defined as ± 3 kb of TSS) was obtained, and peaks mapped to TSS regions were aligned using the getTagMatrix function in ChIPseeker. Then, heatmaps of peak profiles around TSS were generated using the tagHeatmap function, while peak distribution profiles were produced using the plotAvProf function, which generated 0.95 confidence intervals via the bootstrap method. Peak annotation to functional categories, including promoter, 5’UTR, 3’UTR, exon, intron, downstream, and distal intergenic, were performed by the annotatePeak function in ChIPseeker. The distance to TSS of the nearest gene was obtained using the org.Bt.eg.db database as a reference. The plotDisttoTSS function in ChIPseeker was used to calculate the percentage of peaks upstream and downstream from the TSS of the nearest genes and visualize the distribution.

GSP were identified using bedtools (v2.29.2) intersect -v [70]. Gene Ontology (GO) enrichment was performed using clusterProfiler [71] with the following parameters: OrgDb = org.Bt.eg.db, fun = “enrichGO,” ont = “ALL,” pAdjustMethod = “BH,” pvalueCutoff = 0.05. To identify enriched transcription factor binding sites (TFBSs) within group-specific peaks, the GSP regions were converted to human coordinates (hg19) using UCSC liftOver, yielding 23,443 (high group) and 19,977 (low group) human regions [72]. Motif analysis was done using i-cisTarget (v6.0), and motifs were ranked by Normalized Enrichment Score (NES). The NES for a given motif was computed as the AUC value of the motif minus the mean of all AUCs for all motifs, divided by the standard deviation of all AUCs. DOCRs were identified between the high and low groups using Diffbind (v3.2.7) packages, with a P value less than 0.05. GO and pathway enrichment of the DOCRs were performed using clusterProfiler [71]. SumGSE was used to perform GWAS signal enrichment analysis for DMRs and DOCRs in both high and low fatty acid groups [73]. Enrichment of eQTL in functional elements was analyzed using GAT (v1.3.6) [74].

WGBS library preparation and sequencing

Genomic DNA for muscle tissue was isolated using the QIAamp DNA Mini Kit (Qiagen, Hilden, Germany). DNA purity and integrity were assessed using agarose gel electrophoresis, and the OD 260/280 ratio was measured with a Nanodrop spectrophotometer (Nanodrop, Santa Clara, CA). Additionally, DNA concentration was accurately quantified using a Qubit 2.0 fluorometer (Thermo Fisher Scientific, Inc., Waltham, MA, USA). Genomic DNA (3 μg) was sheared to a size of 200–300 bp using a Covaris S220 (Covaris, Inc., Woburn, MA, USA), followed by terminal repairing and ligating adaptors. The DNA bisulfite conversion was performed using the EZ DNA Methylation Gold Kit (Zymo Research, Irvine, CA, USA). Libraries were PCR-amplified and assessed for quality, integrity, and fragment size using an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA) and qPCR (iCycler, BioRad Laboratories, Hercules, CA, USA). WGBS libraries were sequenced using the Illumina Hiseq X Ten platform with 150-bp paired-end reads by Novogene (Novogene, Beijing, China).

Identification of methylcytosine

Raw reads from WGBS were filtered using FASTP (v0.21.0) [56]. The filtration criteria were as follows after removing: (1) the adaptor sequences; (2) reads containing N bases exceeding 10% of the total read length; and (3) reads with low-quality base values exceeding 50% of the total read length. Bismark genome preparation (v0.23.0) [75] was used to build the index of the ARS-UCD1.2 reference genome. Clean reads were aligned to the reference genome using Bismark (v0.23.0) [75] with default parameters. After removing the duplication reads, methyl cytosine (mCs) information was extracted using the bismark_methylation_extractor and summarized using the Bismark2summary. Methylated sites were identified with a binomial test based on the counts of methylated cytosines (mC), total cytosines (mC + umC), and the non-conversion rate (r). Sites with FDR-corrected P value < 0.05 were considered methylated. Methylation level of the sequence was calculated as the ratio of methylated reads to the total number of reads mapping to each site.

CpG island and HMR identification

CGs with at least 10 × coverage were used for the identification of CpG islands (CpGIs) and HMRs analysis. For CpGIs detection, we used CpGIScan with default parameters (length ≥ 500 bp, GC content ≥ 55%, and ObsCpG/ExpCpG ratios ≥ 0.65). The methylation level of the predicted CpG island (cCpGI) was calculated, and those with a methylation level < 0.3 in at least one sample were defined as experimentally supported CpGIs (eCpGIs). HMR was identified with a 10 kb window size following the methpipe (v5.0.0) manual with the default parameters.

Differential methylation analysis

Differential methylation analysis was performed as described by Zhou et al. [76]. DMCs and DMRs were identified using the R package methylkit (v1.18.0) with the calculateDiffMeth function. DMCs and DMRs were defined as having a methylation difference > 30% and a q value < 0.05, and an average methylation difference > 30% and a q value < 0.05, using a window size = 500 bp and step size = 500 bp, respectively. DMCs/DMRs were divided into two groups: hyper-methylated (up) DMCs/DMRs and hypo-methylated (down) DMCs/DMRs. The distribution of hyper-methylated and hypo-methylated cytosine/regions per chromosome was visualized using the diffMethPerChr function. DMCs and DMRs were annotated based on the bosTau9_Refseq.bed file (https://genome.ucsc.edu/) using the genomation (v1.24.0) package. Percentages of DMCs and DMRs in functional components were calculated, including promoters (defined as ± 1000 bp around TSS), introns, exons, intergenic regions, CpG islands, and CpG shore regions (defined as 2000 bp flanking regions on upstream and downstream of a given CpG island). Distances to the nearest TSS were obtained using the getAssociationWithTSS function. Functional annotation and pathway enrichment of genes for DMCs and DMRs were performed using DAVID (v6.8).

Detection of H3K27ac and H3K4me3 signals

Muscle tissues from three males (approximately 24 months old) were used for ChIP experiments with antibodies against H3K27ac (Abcam ab4729) and H3K4me3 (Abcam ab85850). DNA degradation and contamination were monitored on agarose gels, while DNA purity was assessed using the NanoPhotometer® spectrophotometer (IMPLEN, CA, USA). DNA concentration was measured using Qubit® DNA Assay Kit in Qubit® 3.0 Fluorometer (Life Technologies, CA, USA). The purified DNA was used for ChIP-seq library preparation, which was constructed by Novogene Corporation (Novogene, Beijing, China). Subsequently, the libraries underwent paired-end sequencing on the Illumina platform (Illumina, CA, USA). Library quality was evaluated using the Agilent Bioanalyzer 2100 system. Quality control and reads mapping to the reference genome were consistent with our “ATAC-seq chromatin accessibility analysis” workflow. After mapping reads to the reference genome, we used the Genrich software (https://github.com/jsh58/Genrich) to identify regions of IP enrichment over the background with setting “-v -r -e MT,X -m 30 -q 0.01”. A q-value threshold of 0.01 was applied to all datasets.

REs annotation and functional intersection analysis

We systematically annotated REs by integrating histone modification ChIP-seq datasets (H3K27ac, H3K27me3, H3K4me1, H3K4me3) and CTCF binding profiles, following a modified approach adapted from previous report [21]. FACAs, DMCs, and DMRs were classified into five major RE categories: promoters, enhancers, silencers, CTCF-enriched regions, and low-signal regions. SEs and SSs were identified using the ROSE algorithm based on H3K27ac and H3K27me3 signals, respectively. We used BEDTools to intersect fatty acid trait-associated GWAS variants and cis-eQTLs with annotated REs. Promoter-associated variants were defined as those within − 4 kb to + 2 kb of TSS, while variants located within 10 kb of other REs (enhancers, silencers, CTCF, SEs, and SSs) were assigned to the nearest gene.

Tissue-specific validation of fatty acid genes

To explore the tissue-specific expression patterns of genes for fatty acids, we retrieved gene expression profiles across 51 bovine tissue types [37]. We focused on candidate genes SCD and ACAA2, and evaluated their expression across tissues based on our prior analyses. Expression patterns were further validated using CattleGTEx [19].

PheWAS analysis

PheWAS were used to detect known genetic associations and identify potential novel associations across a wide spectrum of phenotypes, as described in humans [77]. We first obtained human orthologs for bovine fatty acid candidate genes using the BLAST (https://blast.ncbi.nlm.nih.gov/Blast.cgi). To test whether these orthologous genes showed similar associations with a wide range of phenotypes in humans, we performed the PheWAS analysis on human orthologues using data from https://atlas.ctglab.nl/. Only GWAS traits with Bonferroni-corrected P value < 0.05 were displayed in the PheWAS plots.

Mediation analysis

We performed causal mediation analyses using the mediation R package [78]. This approach assesses how a mediator M (gene expression) influences the relationship between an independent variable X (genetics) and a dependent variable Y (fatty acid traits). We estimated the average causal mediation effect (ACME), average direct effect (ADE), and total effect using 1,000 simulation draws, and fitted a normal distribution to the empirical estimates. Statistical significance was assessed at a P-value threshold of 0.05.

Identification of TAD using Hi-C data

Hi-C libraries were constructed from two male muscle tissues (~ 24 months old) according to previous studies [79]. Briefly, samples were cross-linked with 1% formaldehyde for 10 min at room temperature and quenched with 0.125 M final concentration glycine for 5 min. The cross-linked cells were subsequently lysed, and endogenous nuclease was inactivated with 0.3% SDS. Chromatin DNA was digested by 100U MboI (NEB, Ipswich, MA, USA), labeled with biotin-14-dCTP (Invitrogen, Carlsbad, CA, USA), and then ligated by 50U T4 DNA ligase (NEB, Ipswich, MA, USA). After reversing cross-links, the ligated DNA was extracted through QIAamp DNA Mini Kit (Qiagen, Hilden, Germany) according to the manufacturer’s instructions. The purified DNA was sheared to 300- to 500- bp fragments, blunt-end repaired, A-tailed, and ligated with sequencing adapters, followed by purification through biotin-streptavidin–mediated pull-down and PCR amplification. Finally, Hi-C libraries were quantified and sequenced on the MGI-seq platform (Beijing Genomics Institute, Shenzhen, China). Clean reads were mapped on the reference genome (ARS-UCD1.2) using HiC-Pro (v3.1.0) [80] pipeline with the following parameters: the GENOME_FRAGMENT bed file of restriction fragments was generated by digest_genome.py with the parameter -r GATC, and the LIGATION_SITE was set as GATCGATC. Data from the two samples were merged and processed to generate Hi-C contact matrices at resolutions of 5 kb, 10 kb, 20 kb, 40 kb, 150 kb, 500 kb, and 1 Mb. We applied HiCExplorer (v3.7.5) [81] to convert the file format, build a Hi-C contact matrix with 10 kb resolution, and identify TAD using the hicFindTAD tool with an FDR threshold of less than 0.01.

Cell culture

Preadipocytes were purified from the subcutaneous adipose tissue of newborn calves [82]. Cells were seeded in culture dishes containing DMEM/F12 (Gibco, 11330032) supplemented with 10% FBS (Gibco, 16000044) and 1% penicillin–streptomycin (Gibco, A1113803). Cultures were maintained in a humidified atmosphere with 5% CO₂ at 37 °C. Differentiation into adipocytes was induced following previously described methods [83]. 293T cell line was obtained from the China Center for Type Culture Collection and cultured in DMEM medium (Gibco, 11965092) containing 10% FBS and 1% penicillin–streptomycin in an incubator at 37 ℃ with 5% CO₂. When the cells reached approximately 90% confluence, passage was performed with trypsin (Thermo, 25200072). All cells were regularly tested for mycoplasma contamination using PCR-based assays in our laboratory, with all results confirmed negative prior to use.

Chemical synthesis of siRNA and transfection

siRNAs targeting bovine SCD (si-SCD-1, si-SCD-2, si-SCD-3) and ACAA2 (si-ACAA2-1, si-ACAA2-2, si-ACAA2-3), as well as a negative control siRNA (si-NC), were synthesized by RiboBio (RiboBio, Guangzhou, China). Sequences are listed in Additional file 2: Table S17. To knock down SCD or ACAA2, cells at 80–90% confluence were transfected with siRNAs or si-NC (100 nM) using the Lipofectamine™ 3000 transfection kit (Invitrogen, L3000015).

RNA extraction and RT-qPCR analysis

Total RNA was extracted using TRIzol (Invitrogen, 15596026CN) following standard protocols. Then, cDNA was synthesized from 1 μg of total RNA using the PrimeScript RT Reagent Kit with gDNA Eraser (Perfect Real Time) (TaKaRa Biotech Co. Ltd, Tokyo, Japan) according to the manufacturer’s instructions. The primers for qPCR analysis were summarized in Additional file 2: Table S18 and synthesized by Sangon Biotech (Sangon Biotech, Shanghai, China). mRNA expression was assessed using the KAPA SYBR® FAST qPCR Master Mix (2X) Kit (Roche, KK4601) on a QuantStudio 7 Flex Real-Time PCR system (Thermo Fisher Scientific, CA, USA). The relative abundance of target mRNAs was calculated using the 2^−ΔΔCt method.

Gas chromatography analysis of fatty acids

Cells cultured in 60 mm wells were transfected with si-SCD, si-ACAA2, or si-NC at 80–90% confluence. After 48 h, cells were collected, and fatty acid extraction and analysis were performed as previously described [84]. Relative proportions of fatty acids were determined as percentages of the total peak area.

Triglyceride assay and Oil Red O staining

Preadipocytes and edited-293T cells were cultured in 6-well plates. At ~ 90% confluence, cells were washed three times with 1 × PBS and harvested. The triglyceride content was quantified using an enzymatic triglyceride assay kit (Applygen, E1003-125) following the manufacturer’s instructions. The total triglyceride content was normalized to protein concentration (µg/mg). For Oil Red O staining, when reaching approximately 90% confluence, the adipocytes and edited-293T cells were washed three times with 1 × PBS and then fixed for 10 min with 4% paraformaldehyde. Then, cells were stained using a Modified Oil Red O Staining Kit (Beyotime, C0158S) according to the manufacturer’s instructions.

Dual luciferase reporter assays

Dual luciferase assays were carried out in 293T cells. A 1,001 bp enhancer region of SCD (including 500 base pairs upstream and downstream of the SNP), containing either the wild-type T or the mutated C alleles, was synthesized by GeneCarer, following purification and digested with KpnI and HindIII restriction enzymes before inserted into PGL4.23(Luc2-minP) vectors (Promega, E8411). For dual-luciferase reporter assays, 293T cells were plated in 12-well plates and transfected with the recombinant plasmids using Lipofectamine™ 3000 when the cells reached approximately 70% confluence. A total of 1 μg of plasmid was transfected into the cells at a 20:1 ratio of pGL4.23 to pRL-TK. After 48 h, cells were washed with PBS and lysed with passive lysis buffer according to the manufacturer’s protocol (Promega, E1910). Firefly and Renilla luciferase activities were measured in white-bottom 96-well plates using an automated luminometer (Infinite M200 PRO, TECAN). Reporter activity (Firefly) was normalized to the intrinsic Renilla luciferase activity. All experiments were performed using 293T cells in at least four independent replicates.

Gene-edited cell construction and screening

Functional analysis of the SNP region was conducted in 293T cells using CRISPR technology. The human SNP locus chr10:100,359,678 (hg38) was identified as homologous to chr26:21,274,156 in cattle (ARS-UCD1.2) using hgLiftOver with default parameters. The sgRNAs (sgRNA-1, sgRNA-2) were designed to target the candidate human SNP using the CRISPR design tool (https://crispor.gi.ucsc.edu/) (Additional file 2: Table S19). These sgRNAs were then cloned into the px459 vector (Addgene, 48139). For the knockout experiments, when 293T cells seeded in 6-well plates reached approximately 80% confluence, they were transfected with the recombinant plasmids (px459 sgRNA-1, sgRNA-2, or sgRNA-1/2) using Lipofectamine™ 3000. After 24 h, the medium was replaced with a full medium containing puromycin, and then the entire medium was changed every 24 h. After 5 days of incubation, the cells were transferred to a 10 cm dish for extended culture. The editing efficiency of the target region was then assessed and analyzed by sequencing (BGI, Shenzhen, China), and the validated cells were used for subsequent experiments.

Western blotting

Protein concentrations were evaluated using a BCA kit (Beyotime, P0010). Proteins were separated by 4%−12% SurePAGE gels and then transferred to a nitrocellulose membrane. Membranes were incubated overnight at 4 °C with primary antibodies, followed by incubation with the secondary antibodies at 37 °C for 1 h. SCD protein was detected using the monoclonal antibody A9A11-R (1:2000, HUABIO), and ACTIN was detected with the polyclonal antibody ab197345 (1:2000, Abcam). Protein bands were visualized using ECL western blotting detection reagent (Beyotime, P0018S).

Statistical analysis

Statistical analyses were conducted using GraphPad Prism 9.0.2. P-values were calculated using one-way analysis of variance (ANOVA), Dunnett’s multiple comparison test, and Student’s t-test (for two groups). Asterisks denote different significance levels (*P < 0.05, **P < 0.01, and ***P < 0.001). Unless otherwise stated, n = 3 independent biological replicates were performed for each experiment.

Authors’ contributions

L.Y.X., J.Y.L., and T.L.Z. conceived and designed the study. X.G., Y.C., L.P.Z., H.J.G., T.F., and T.Y.G. collected the samples and gave access to the tissue samples. T.L.Z. and Q.H.N. performed the computational analysis. T.Z.W., J.Y.W., B.Z., and Z.Z.W. performed the experiments. L.Y.X., T.L.Z., and G.E.L. wrote the manuscript. The authors read and approved the final manuscript.

Funding

This study was supported by the National Key R&D Program of China (2022YFF1000600) and National Natural Science Foundation of China (31972554 and 32272835). LYX was supported by the Elite Youth Program at Chinese Academy of Agricultural Sciences. The project was also partly supported by the Agricultural Science and Technology Innovation Program in Chinese Academy of Agricultural Sciences (ASTIP-IAS-TS-16 and ASTIP-IAS03) and National Beef Cattle Industrial Technology System (CARS-37) for the data analysis and interpretation of the study. GEL was supported in part by AFRI grant numbers 2019-67015-29321 and 2021-67015-33409 from the USDA National Institute of Food and Agriculture (NIFA, Kansas City, MO, USA).

Data availability

The WGBS and ATAC-seq datasets generated for this study have been submitted to the National Genomics Data Center (NGDC) (https://bigd.big.ac.cn/) under accession numbers PRJCA017331 [85]. The RNA-seq datasets derived from muscle tissues in this study have been deposited in the National Genomics Data Center (NGDC) under the accession number PRJCA017824 [86]. The ChIP-seq and Hi-C datasets generated for this study have been submitted to the NGDC under accession number PRJCA041184 [87]. The RNA-seq datasets derived from adipose tissues in this study have been deposited in the NCBI SRA database (https://www.ncbi.nlm.nih.gov/sra) under the accession number PRJNA846691 [88]. All chromatin marks for muscle data in this study were downloaded from the NCBI Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/) under accession number GSE158430 [89].

Declarations

Ethics approval and consent to participate

All the animal procedures were performed strictly according to the guidelines proposed by the China Council on Animal Care and the Ministry of Agriculture, People’s Republic of China, and in compliance with the Animal Research: Reporting In Vivo Experiments (ARRIVE) guidelines. Tissue samples from beef cattle were collected with the approval of the Ethics Committee of the Science Research Department of the Institute of Animal Science, Chinese Academy of Agricultural Sciences, under IAS2020-48.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.