Multi‐Omics Integration Identifies ABCD4 as a Key Regulator of Backfat Thickness and Lipid Deposition in Pietrain Pigs

ABSTRACT

The commercial pork production sector prioritizes genetic improvements in lean meat percentage to enhance profitability and meet consumer preferences. The Pietrain pig, a premier terminal sire breed renowned for its exceptional muscularity and leanness, serves as an ideal model to decipher the genetic underpinnings of these traits. This study investigated two key measures of leanness, backfat thickness and loin muscle depth, in two distinct Pietrain populations to elucidate the genetic architecture underlying these traits. We estimated genetic parameters and performed a meta‐analysis of genome‐wide association studies, identifying ABCD4, LTBP2, NUMB, and SLC30A9 as candidate genes. To further investigate these associations, we integrated information on molecular quantitative trait loci from the PigGTEx project and single‐cell transcriptomic resources. This integrative approach prioritized ABCD4 as a key candidate gene regulating backfat thickness. Functional validation in 3T3‐L1 preadipocytes revealed a novel dual regulatory role for ABCD4: its knockdown suppressed cell proliferation while simultaneously stimulating adipogenic differentiation, as demonstrated by the upregulation of key markers. Our findings positioned ABCD4 as a critical modulator of fat deposition, likely through its influence on the core adipogenic transcriptional network. By establishing an analytical framework that integrates large‐scale sequencing data from Pietrain pigs with functional validation, our study addresses a key gap in understanding the genetic basis of leanness and provides novel insights for precision breeding.

1. Introduction

The commercial focus of the modern pork industry centers on meeting consumer demand for high‐lean‐meat pork while ensuring producer profitability (Zhou et al. 2021). This dual objective directly drives commercial pig breeding toward superior carcass lean percentage, optimal growth efficiency, and acceptable meat quality. However, a single breed rarely excels in both reproductive performance and carcass traits. Consequently, modern production widely adopts multi‐breed crossbreeding systems to integrate complementary strengths through heterosis (Tang et al. 2023). Within typical three‐way crosses, the terminal sire contributes 50% of the genetic composition and plays a critical role in maximizing lean meat yield, improving growth performance, and enhancing feed efficiency in commercial pig offspring (Aymerich et al. 2019). Originating from Belgium, the Pietrain pig is characterized by an exceptionally high genetic lean meat percentage, reported to reach 66.7% (Heidaritabar et al. 2023). This trait has established it as a globally recognized terminal sire breed of choice within commercial pork production systems. Consequently, to meet the industry's need for selecting even leaner Pietrain terminal boars that enhance offspring leanness, it is essential to elucidate the genetic architecture of this breed‐specific trait and screen for effective genetic markers, thereby optimizing breeding strategies.

In practical breeding programs, direct measurement of lean meat percentage in live animals poses a significant challenge in breeding practice (Alempijevic et al. 2021). As a result, easily measurable ultrasonography‐derived traits such as backfat thickness (BFT) and loin muscle depth (LMD) are widely employed as reliable proxy indicators in both breeding applications and genetic studies (Ding et al. 2022). Due to its extreme phenotypic values, the Pietrain pig facilitates the detection of large‐effect genetic variants that might be diluted in outbred populations. However, previous large‐scale genetic studies in Pietrain pigs have primarily relied on single nucleotide polymorphisms (SNP) array data, which provide limited genomic resolution (Reyer et al. 2024). Consequently, to achieve a more comprehensive mapping of the variants underlying these extreme traits, genome‐wide association studies (GWAS) based on whole‐genome sequencing (WGS) are required.

Genetic heterogeneity across populations, arising from differences in genetic structure, allele frequencies, and environmental exposures, can introduce false positives in genome‐wide association studies (GWAS). To mitigate this and obtain more reliable and generalizable estimates of genetic effects for complex traits, meta‐GWAS has been widely adopted in pig genomics. For instance, one meta‐GWAS analyzing semen quality traits across five pig breeds demonstrated its utility in improving the accuracy of genomic selection (Cheng et al. 2024). Similarly, by integrating data from the Pig Genotype‐Tissue Expression (PigGTEx) project and the Functional Annotation of Animal Genomes (FAANG) project, a separate meta‐GWAS inferred quantitative trait loci for 139 complex traits, thereby advancing the study of regulatory mechanisms linking genetic variation to phenotypic outcomes (Xu et al. 2025).

While meta‐GWAS has successfully identified numerous trait‐associated loci, a critical challenge remains: the majority of these signals map to non‐coding regions, complicating the identification of causal variants and the elucidation of their mechanistic roles. To address this, transcriptome‐wide association studies (TWAS) and expression quantitative trait locus (eQTL) colocalization analyses leverage gene expression data to test associations between genetically predicted gene expression and complex traits, and to evaluate whether GWAS and eQTL signals are driven by the same underlying causal variant, thereby strengthening evidence for specific gene‐trait relationships (Cai et al. 2023). Within the PigGTEx project, the expression of ABCD4 in both the small intestine and brain was successfully identified as significantly associated with BFT (Teng et al. 2024). While these methods represent a substantial advance, their findings are constrained by tissue‐level resolution and, as in silico studies, necessitate direct experimental validation to define gene function. Tissues, however, are complex ecosystems composed of heterogeneous cell types, each with distinct functions. A trait‐associated signal identified in a bulk tissue analysis cannot distinguish which specific cell type. This blind spot fundamentally limits our ability to formulate precise mechanistic hypotheses and design targeted functional experiments. To achieve cellular‐resolution insight into the genetic architecture of BFT and LMD, we turned to single‐cell RNA sequencing (scRNA‐seq) data. We applied the single‐cell Disease Relevance Score (scDRS) framework to move beyond tissue‐level associations and quantitatively link the polygenic signatures of these traits to the transcriptional programs of individual cell types (Zhang et al. 2022).

The genetic architecture of BFT and LMD in Pietrain pigs remains incompletely characterized, limiting their optimized application in breeding programs. To address this, we investigated these traits in two genetically divergent Pietrain populations (Figure 1). We first estimated the genetic parameters for both traits and conducted a meta‐analysis of GWAS to identify associated genomic regions. To further elucidate the underlying mechanisms, we integrated multi‐omics data, including gene expression datasets from the PigGTEx project and single‐cell transcriptomic resources from multiple porcine tissues. For the functional validation of candidate genes, we employed the in vitro 3T3‐L1 preadipocyte model. This classic model was selected due to its well‐defined differentiation protocol and high reproducibility, which allowed for an efficient initial assessment of gene function in adipogenesis, complementary to our genetic mapping results. By synthesizing findings from association mapping and multi‐omics insights, this study aims to pinpoint robust candidate loci for precision breeding, thereby contributing to enhanced production efficiency and meat quality in swine breeding.

FIGURE 1.

Structural diagram of the methodological framework.

2. Materials and Methods

2.1. Animals, Phenotypes, and Genotypes

The Pietrain pigs evaluated in this study were provided by Scigene Biotechnology Co. Ltd., originating from a foundation herd introduced from Choice Genetics in 2014. Over the following years, this ancestral population underwent independent selection and breeding at two geographically distinct farms, leading to genetic divergence. In total, 1320 individuals were sampled, comprising 248 pigs from the Chizhou, Anhui farm (PAH) and 1072 pigs from the Guigang, Guangxi farm (PGX). Specifically, the animals in the PAH and PGX populations were born during 2017–2023 and 2019–2023, respectively.

In this study, two lean meat‐related traits were considered, including backfat thickness (BFT) and loin muscle depth (LMD). Both traits, along with body weight, were measured on live pigs prior to slaughter at approximately 24 weeks of age. BFT and LMD were extracted from ultrasound images captured between the 3rd and 4th last ribs of each pig using BioSoft Toolbox (v2.6.0.1), and the phenotypic data were subsequently adjusted to a standard body weight of 115 kg. The formula used for BFT adjustment is as follows:

$$\mathrm{BFT}={\mathrm{BFT}}_{\mathrm{m}}\times \frac{A}{A+B\times \left(W-115\right)}$$ (1)

where ${\mathrm{BFT}}_{\mathrm{m}}$ is the actual measured BFT value, $W$ is the actual measured body weight, and $A$ and $B$ are correction constants ($A$ = 9.97 and $B$ = 0.07 for boars; $A$ = 9.89 and $B$ = 0.07 for sows). The correction formula for LMD is as follows:

$$\mathrm{LMD}={\mathrm{LMD}}_{\mathrm{m}}\times \frac{C}{C+D\times \left(W-115\right)}$$ (2)

where ${\mathrm{LMD}}_{\mathrm{m}}$ is the actual measured LMD value, $W$ is the actual measured body weight, and $C$ and $D$ are correction constants ($C$ = 61.01 and $D$ = 0.27 for boars; $C$ = 65.11 and $D$ = 0.3 for sows).

A small piece of ear tissue was collected from live pigs using a sterile ear notcher and immediately placed into an Eppendorf tube containing 70% ethanol. Genomic DNA was extracted from the ear tissue using the phenol‐chloroform extraction method at 4°C. The purity and integrity of the extracted DNA were assessed by agarose gel electrophoresis and ultraviolet spectrophotometry. Sequencing was performed on the BGI platform (BGISEQ, DNBSEQ‐T7, PE 150 model, 3G/pigs). After quality control with fastp v0.20.0 (Chen et al. 2018), reads were mapped to Sscrofa11.1 with GTX v2.1.5 (Xing et al. 2017) and PCR duplicates were removed. Genotypes were imputed to the haplotype reference panel of the PHARPv3 panel (Wang, Zhang, et al. 2022) using GLIMPSE2 (Rubinacci et al. 2021), which included 42 402 636 loci. After imputation, variants were filtered with an information quality score > 0.3. Autosomal variants were then filtered using PLINK v1.90 (Purcell et al. 2007) based on a minor allele frequency (MAF) threshold of 0.05 (‐‐maf 0.05) and Hardy–Weinberg equilibrium (‐‐hwe 1e‐6). Ultimately, 8 556 704 and 8 421 891 autosomal SNPs of PAH and PGX, respectively, were retained for further analysis.

2.2. Genetic Parameters Estimation and Pre‐Correction of Phenotypes

We used a bivariate linear mixed model (LMM) to estimate genetic parameters using HIBLUP v1.4.0 (Yin et al. 2023). The model is as follows:

$$\mathbf{y}\mathbf{=}\mathbf{X\beta }\mathbf{+}\mathbf{Z\gamma }+\mathit{e}$$ (3)

where $\mathbf{y}$ is a vector of the phenotypic values of BFT and LMD; $\mathbf{\beta }$ is the fixed effect vector comprising sex, birth year, birth season, herd, and parity; $\mathbf{\gamma }$ is the vector of additive genetic random effects, following a multivariate normal distribution N(0, $\mathbf{G}\mathbf{⨂}{\mathbf{\sum }}_{\mathbf{\gamma }}$), where ${\mathbf{\sum }}_{\mathbf{\gamma }}$ is the genetic covariance matrix, and $\mathbf{G}$ is the genomic kinship matrix constructed based on the Van Raden method (VanRaden 2008) from the pruned SNP markers of the PLINK software (‐‐indep‐pairwise 50 5 0.5). The formula was as follows:

$$\mathbf{G}\mathbf{=}\frac{{\mathbf{WW}}^{\prime }}{2{\sum }_{i=1}^m{p}_i\left(1-p_i\right)}$$ (4)

where $\mathbf{W}$ is the matrix of genotypes encoded as $0–2p_i$, $1–2p_i$, $2–2p_i$, and $p_i$ is the allele frequency of the $i$th marker. In Formula (3), $\mathbf{e}$ is the vector of residual random effects following N (0, $\mathbf{I}{\mathbf{⨂}\mathbf{\sum }}_{\mathbf{e}}$), where ${\mathbf{\sum }}_{\mathbf{e}}$ is the residual covariance, $\mathbf{I}$ is the identity matrix and ⨂ is the Kronecker product of matrices; $\mathbf{X}$ and $\mathbf{Z}$ are the incidence matrices to the fixed effects and random effects, respectively. The covariance matrices were as follows:

$${\sum }_{\mathrm{\gamma }}=\left[\begin{array}{cc}{\sigma }_{{\mathrm{\gamma }}_1}^2& {\sigma }_{{\mathrm{\gamma }}_1}{\sigma }_{{\mathrm{\gamma }}_2}\\ {\sigma }_{{\mathrm{\gamma }}_1}{\sigma }_{{\mathrm{\gamma }}_2}& {\sigma }_{{\mathrm{\gamma }}_2}^2\end{array}\right]$$ (5)

$${\sum }_{\mathrm{e}}=\left[\begin{array}{cc}{\mathrm{\sigma }}_{{\mathrm{e}}_1}^2& {\mathrm{\sigma }}_{{\mathrm{e}}_1}{\mathrm{\sigma }}_{{\mathrm{e}}_2}\\ {\mathrm{\sigma }}_{{\mathrm{e}}_1}{\mathrm{\sigma }}_{{\mathrm{e}}_2}& {\mathrm{\sigma }}_{{\mathrm{e}}_2}^2\end{array}\right]$$ (6)

where ${\sigma }_{{\mathrm{\gamma }}_n}^2$ and ${\sigma }_{{\mathrm{e}}_n}^2$ are the additive variance and random residual variance for trait $n$ (BFT and LMD, respectively). The variance components were estimated using the average information restricted maximum likelihood (AI‐REML) algorithm implemented in HIBLUP v1.4.0. The heritability for each trait was estimated using the following formula:

$$h^2={\widehat{\mathrm{\sigma }}}_{\mathrm{\gamma }}^2/{\widehat{\mathrm{\sigma }}}_{\mathrm{p}}^2$$ (7)

where $h^2$ is the estimation of heritability and ${\widehat{\sigma }}_{\mathbf{p}}^{\mathbf{2}}$ is the phenotypic variance of each trait. The genetic correlations ($r_{\mathbf{g}}$) and phenotypic correlations ($r_{\mathbf{p}}$) were calculated following the equations below:

$$r_{\mathrm{g}}=\frac{{{\sigma }_{\mathrm{\gamma }}}_1{{\sigma }_{\mathrm{\gamma }}}_2}{\sqrt{{\sigma }_{{\mathrm{\gamma }}_1}^2{\sigma }_{{\mathrm{\gamma }}_2}^2}};r_{\mathrm{p}}=\frac{{\mathrm{\sigma }}_{{\mathrm{\gamma }}_1{\mathrm{\gamma }}_2}+{\mathrm{\sigma }}_{{\mathrm{e}}_1{\mathrm{e}}_2}}{\sqrt{\left({\mathrm{\sigma }}_{{\mathrm{\gamma }}_1}^2+{\mathrm{\sigma }}_{{\mathrm{e}}_1}^2\right)\left({\mathrm{\sigma }}_{{\mathrm{\gamma }}_2}^2+{\mathrm{\sigma }}_{{\mathrm{e}}_2}^2\right)}}$$ (8)

2.3. Population Structure Analysis, GWAS, and Meta‐Analysis

To clarify the population structure of the two groups, principal component analysis (PCA) was performed using PLINK v1.90.

A univariate linear mixed model was used to calculate the association between genetic markers and two traits for the two populations in GEMMA (Zhou and Stephens 2012). The model is as follows:

$${\mathbf{y}}_{\mathbf{c}}=\mathbf{1}\mu +\mathbf{Xg}+\mathbf{u}+\mathbf{\epsilon }$$ (9)

where ${\mathbf{y}}_{\mathrm{c}}$ is vector of corrected phenotype, calculated as the sum of the estimated additive genetic effects and the individual residual of the $i$th individual (${\mathrm{y}}_{\mathrm{i}}=\widehat{\mathrm{\gamma }}+\widehat{\mathrm{e}}$) using Equation (3); $\mathrm{\mu }$ is the overall mean of corrected phenotypes; $\mathbf{X}$ is the matrix of genotype coded in 0, 1, 2; $\mathbf{g}$ is the fixed additive effect estimate of the SNP; $\mathbf{u}$ is a vector of random polygenic effect, following the multivariate normal distribution MVN_n(0, $\lambda {\tau }^{-1}\mathbf{K}$), where is $\lambda$ the ratio between ${\widehat{\sigma }}_{\mathbf{\gamma }}^{\mathbf{2}}$ and ${\widehat{\sigma }}_{\mathbf{e}}^{\mathbf{2}}$, ${\tau }^{-1}$ represents the variance of the residual, and $\mathbf{K}$ is a kinship matrix constructed based on the SNP pruned by PLINK v1.90 (‘‐‐indep‐pairwise 50 5 0.5’); $\mathbf{\epsilon }$ is the residual vector that follows the multivariate normal distribution MVN_n(0, ${\tau }^{-1}{\mathbf{I}}_{\mathbf{n}}$), where ${\mathbf{I}}_{\mathbf{n}}$ is an identity matrix.

To further investigate the loci influencing these traits and mitigate the impact of population stratification, we conducted a meta‐GWAS analysis using a fixed‐effect model based on GWAS summary statistics from two populations, as provided by METASOFT (Han and Eskin 2011), which included beta values and standard errors (SEs). The significance threshold was calculated as 0.05/N, where N denotes the number of pruned SNPs; hence, a Bonferroni‐corrected threshold of 1.33 × 10⁻⁷ (α = 0.05) was applied for both BFT (N = 376 051) and LMD (N = 376 901). The proportion of variance in phenotype explained (PVE) of significant SNPs was estimated as follows:

$$\mathrm{PVE}=\frac{2{\widehat{\beta }}^2\mathrm{MAF}\left(1-\mathrm{MAF}\right)}{2{\widehat{\beta }}^2\mathrm{MAF}\left(1-\mathrm{MAF}\right)+{\left(\mathrm{se}\left(\widehat{\beta }\right)\right)}^{2}2n·\mathrm{MAF}\left(1-\mathrm{MAF}\right)}$$ (10)

where $\widehat{\beta }$ is the effect size of each SNP estimated by meta‐analysis, MAF is the minor allele frequency, $\mathrm{se}\left(\widehat{\beta }\right)$ is the standard error of $\widehat{\beta }$, and $n$ is the final sample size with valid phenotypic records included in the analysis ($n$ = 1284 and 1283 for BFT and LMD, respectively).

2.4. SNP Annotation and Enrichments

Significant SNPs within a 50 kb upstream and downstream of the loci were mapped to genes, using the GALLO v1.3 package (Fonseca et al. 2020), with reference to the pig Gene Transfer Format (GTF, Release 111) files. Subsequently, we identified the nearest gene to each SNP locus as the candidate gene. Furthermore, we compared the genetic differences of traits among breeds with reference to data from PigBiobank (https://pigbiobank.farmgtex.org/).

2.5. TWAS and Colocalization With eQTLs

To investigate potential associations between the studied traits and cis‐regulated gene expression across 34 tissues, we input data including chromosome, position, SNP ID, effect alleles, non‐effect alleles, p‐value, and beta coefficient to the FarmGTEx TWAS‐Server (Zhang et al. 2025) and applied the FUSION BLUP method (Gusev et al. 2016) to predict gene expression‐trait correlations. The threshold was set to 0.05/N, where N denotes the number of all genes detected in a TWAS.

To assess evidence for a shared genetic basis between trait associations and gene expression, we obtained gene expression data from PigGTEx (https://piggtex.farmgtex.org/) and performed colocalization analysis using coloc v5.2.3 (Giambartolomei et al. 2014). This assessed whether significant meta‐GWAS signals and eQTL share a common causal variant. Genes within ±1 Mb of meta‐GWAS lead variants were extracted for colocalization with corresponding eGenes, with colocalization defined by a posterior probability H4 (PP.H4) > 0.8.

2.6. scRNA‐Sequencing Data Processing and scDRS Analysis

We obtained a public scRNA‐seq dataset comprising 23 domestic pig tissues as detailed (Wang, Ding, et al. 2022). Scanpy v1.9.5 (Wolf et al. 2018) was employed for filtering based on specific criteria: cells with over 200 genes and genes expressed in over 3 cells were retained; cells with 200–7500 genes and mitochondrial gene expression at 10% were retained. The dataset underwent normalization, identification of highly variable genes, and UMAP visualization using the top 10 principal components. Clustering and annotation of UMAP data were performed using Leiden. Gene expression data were imputed and denoised using the MAGIC algorithm to address technical dropouts.

To link trait heritability to specific cell types, MAGMA (de Leeuw, et al. 2015) was used to annotate SNP association signals from GWAS within a 10 kb window surrounding each gene. The association strength between genes and phenotypes was calculated, and the top 1000 genes with the strongest associations with each phenotype were extracted to generate a gene set file. Gene scores related to cell‐level traits were quantified using the compute‐score function in scDRS v1.0.2 (Zhang et al. 2022).

2.7. ABCD4 Gene Function Verification

2.7.1. Cell Culture and Adipogenic Differentiation Induction

A mouse embryonic fibroblast (preadipocytes) cell line (3T3‐L1) was purchased from Procell Life Science & Technology Company (Wuhan, China) and subsequently cultured in complete growth medium (high‐glucose DMEM medium supplemented with 10% calf serum and 1% penicillin–streptomycin) in a CO₂ incubator. Cells were digested with 0.25% trypsin and passaged after attaining 80%–90% confluence. The 3T3‐L1 preadipocytes were cultured in the aforementioned complete growth medium for 2 days to achieve contact inhibition and initiate adipocyte differentiation. On Day 0, 3T3‐L1 preadipocytes were stimulated with IMA (high‐glucose DMEM supplemented with 10% FBS, 0.5 mM IBMX, 1 μM dexamethasone, 10 μg/mL insulin). After 2 days (on Day 2), the medium was replaced with IMB (high‐glucose DMEM supplemented with 10% FBS and 10 μg/mL insulin). Following a further 1‐day incubation (on Day 3), the medium was changed back to IMA, initiating one cycle of the A/B alternation. This cycle was repeated until the endpoint of differentiation on Day 8.

2.7.2. SiRNA Transfection Assay

Small‐interfering RNAs (siRNAs) targeting ABCD4 were synthesized by Sangon Biotech (Shanghai, China). The sequences were as follows: sense strand, 5′‐AGAAUGUCUUAAUGUUCAU‐dTdT‐3′; antisense strand, 5′‐AUGAACAUUAAGACAUUCU‐dTdT‐3′. Cells seeded in 12‐well plates were transfected after attaining 70%–80% confluence. Briefly, siRNA or a control vector was mixed with Lipofectamine 2000 (Thermo Fisher Scientific, USA) in high‐glucose DMEM, incubated at room temperature for 5 min, and added drop‐wise to the culture medium. Following transfection and incubation, cells were harvested for CCK‐8 assay, Oil Red O staining, triglyceride (TG) quantification, and gene expression analysis.

2.7.3. CCK‐8 Assay

For cell viability, cells were seeded in 96‐well plates. After transfection with siRNA at concentrations of 30, 60, and 90 nM for 24 h, CCK‐8 detection reagent (Solarbio, Beijing, China) was added. After incubation at 37°C for 1 h, absorbance at 450 nm was measured using a microplate reader (BioTek, Vermont, USA).

2.7.4. Oil Red O Staining (ORO)

Culture medium was gently removed, and cells were washed with phosphate‐buffered saline (PBS). Cells were fixed with 4% paraformaldehyde (Servicebio, G1101) fixative for 10 min at room temperature. After PBS rinsing, cells were covered with 60% isopropanol for 20 s. The solution was discarded, and ORO working solution n (Beyotime Biotechnology, C0158S) was added for 10 min. Following removal of the staining solution, cells were sequentially rinsed with 60% isopropanol and PBS. The stained lipid droplets can be visualized and captured under bright field on an inverted microscope. To determine the lipid content in 3T3‐L1 adipocytes, cells were lysed in isopropanol, and the absorbance of the eluate at 510 nm was measured using a microplate reader (BioTek, Vermont, USA).

2.7.5. TG Biochemical Index Detection

Cellular TG levels were measured using a commercially available kit (Shanghai Hengyuan 401 Biotechnology, L‐740‐SH) at an absorbance of 505 nm according to the manufacturer's protocol.

2.7.6. Total RNA Extraction and Quantitative Real‐Time PCR

Total RNA was extracted from cells using the EasyPure Fast Cell RNA Kit (TransGen Biotech, Beijing, China), and was reverse‐transcribed into cDNA using a reverse transcription kit (Vazyme, China). Quantitative polymerase chain reaction (qPCR) was then performed using the SYBR Green PCR mix kit (Vazyme, China) in a QIAquant96 2plex system (Qiagen, Hilden, Germany). PCR cycling parameters were as follows: 95°C for 3 min, 40 cycles of 95°C for 5 s, and 60°C for 30 s with melt curve analysis. Primers (Table S1) were designed using Primer 5.0 and synthesized by Sangon Biotech (Shanghai) Co. Ltd. (Shanghai, China). Relative expression levels were calculated using the 2^−∆Ct method using β‐actin as the housekeeping control. At least three biological replicates were performed for each sample.

2.7.7. Statistical Analysis

All statistical analyses were conducted with GraphPad Prism 9.5.1 (GraphPad Software, San Diego, CA, USA). Data are presented as mean ± standard error of the mean (SEM) from at least three independent biological replicates. For analyzing relative gene expression levels via qPCR, lipid droplet accumulation via ORO, and intracellular TG content between the control and siRNA‐transfected groups, an unpaired two‐tailed Student's t‐test was applied. For evaluating cell viability in the CCK‐8 assay, one‐way ANOVA followed by Dunnett's post hoc test was performed to compare the treatment groups against the control. Significance levels are indicated as *p < 0.05, **p < 0.01, and ***p < 0.001.

3. Results

3.1. Data Summary and Population Stratification

The PAH population exhibited an average genome sequencing depth of 2.6× and a mean coverage of 69%, while the PGX population averaged 2.3× depth with 71% coverage. Following genotype imputation using the PHARP reference genome haplotype panel, both populations contained 42 400 723 genetic variants. After quality control, the PAH and PGX datasets comprised 8 556 704 and 8 421 891 SNPs, respectively. Of these, 7 796 775 SNPs were shared between the two populations, while 759 929 and 625 112 SNPs were specific to the PAH and PGX groups, respectively. The SNP density maps are presented in Figure S1a.

PCA revealed significant genetic differentiation between the populations, with principal component 1 (PC1) explaining 3.07% of the variation and PC2 accounting for 1.88% as shown in Figure 2a. Given this pronounced population stratification, we conducted GWAS for each population separately and combined the results using a meta‐analysis approach to identify consistent association signals across these genetically distinct Pietrain groups.

FIGURE 2.

Population structure and colocalized loci for BFT and LMD. (a) Principal component analysis (PCA; PC1 and PC2) was performed on 1320 samples based on the integrated variant set. (b, c) Manhattan plots from the meta‐GWAS for BFT and LMD. The x‐axis represents the chromosomes, and the y‐axis shows the association significance as ‐log₁₀(P) value. Genome‐wide significance thresholds are set at 1.33 × 10⁻⁷ for BFT and LMD. (d, e) Distribution of BFT phenotypic values across different genotypes of the SNP rs326218005 (annotated to ABCD4 in the meta‐GWAS) and the eQTL rs1109589390 (colocalized with ABCD4). “ns” denotes not significant. The n in parentheses below the genotype on the horizontal axis represents the sample size for each genotype. *p < 0.05, **p < 0.01, ***p < 0.001. (f, i) COLOC results for ABCD4. The top left panel shows the distribution of SNPs within a 1 Mb flanking region of the lead SNP from the BFT meta‐GWAS. The bottom left panel shows the corresponding eQTL signals. The purple square marks the lead colocalized SNP. The right‐side legend indicates the linkage disequilibrium (LD) between neighboring SNPs and the lead SNP using a color gradient. (j–m) The expression levels of ABCD4 in the brain, frontal cortex, muscle, and jejunum of individuals with three genotypes (AA, AG, and GG) in the PigGTEx project, respectively. The n in parentheses below the genotype on the horizontal axis represents the sample size for each genotype. (n) Venn diagram illustrating genes significantly associated with BFT identified via meta‐GWAS, TWAS, and eQTL colocalization. The overlapping gene present across all three analyses is ABCD4.

3.2. Phenotypic Descriptive Statistics and Genetic Parameter Estimates

Table 1 summarizes the phenotypic characteristics of BFT and LMD in the PAH and PGX populations. Compared with the PAH population, the PGX population exhibited lower BFT and greater LMD. Between traits, BFT showed a higher coefficient of variation than LMD. Notable differences in heritability were observed: heritability estimates for BFT were 0.50 ± 0.20 and 0.53 ± 0.05 in the PAH and PGX populations, respectively, whereas those for LMD were 0.38 ± 0.17 and 0.39 ± 0.05, indicating moderate‐to‐high heritability for both traits. Genetic and phenotypic correlations between BFT and LMD revealed distinct patterns between the populations. Within the PAH population, a strong negative genetic correlation existed ($r_g$ = −0.41), although their phenotypic correlation was negligible ($r_p$ = −0.04). In contrast, both genetic and phenotypic correlations between these traits were weak in the PGX population ($r_g$ = −0.11, $r_p$ = 0.04).

TABLE 1.

Phenotypic statistics and genetic parameters.

Population	Trait	Number	Max	Min	Mean ± SD	CV (%)	h 2 ± SE
PAH	BFT	242	14.26	5.44	9.09 ± 1.62	17.82	0.50 ± 0.20
	LMD	241	79.6	47.2	61.26 ± 6.1	9.96	0.38 ± 0.17
PGX	BFT	1042	17.12	4.97	8.84 ± 1.61	18.21	0.53 ± 0.05
	LMD	1042	82.6	47.4	64.49 ± 5.61	8.7	0.39 ± 0.05

Abbreviations: BFT, backfat thickness; h ², heritability; LMD, loin muscle depth; Max, maximum; Min, minimum.

3.3. Genetic Loci Associated With BFT and LMD

The meta‐analysis identified 734 significant loci distributed across five chromosomal regions associated with BFT (Table S2), corresponding to 36 annotated protein‐coding genes (Table 2). This genetic architecture suggests that BFT in Pietrain pigs is influenced by a limited number of loci with major effects against a polygenic background. As illustrated in Figure 2b, a prominent SNP cluster spanning approximately 2.6 Mb was identified on Sus scrofa chromosome (SSC7). The lead variant in this region, rs326218005, explained 2.90% of the phenotypic variance and was associated with significantly higher BFT values in GG (n = 166) homozygotes compared to AA (n = 521) and AG (n = 597) individuals (Figure 2d). Adjacent loci on SSC7 encompass several genes with previously implicated roles in adipogenesis, obesity, fatty acid metabolism, and meat quality, including ABCD4, NUMB, and VRTN. A comparative analysis between our Pietrain results and the PigBiobank meta‐analysis of Duroc, Landrace, and Yorkshire breeds revealed overlapping BFT regulatory regions on SSC7. Meanwhile, SSC13 exhibited the highest density of significant SNPs, distributed across a 9.5 Mb region. For LMD, a total of 70 significant SNPs across four chromosomes were mapped to seven protein‐coding genes (Tables 3 and S3), with the strongest associations concentrated within a 1.26 Mb interval (131.35–132.61 Mb) on SSC14 (Figure 2c). The peak variant on SSC14, rs324172246, accounted for 2.40% of the phenotypic variance.

TABLE 2.

Summary of independent genomic loci associated with BFT.

Chr	Genomic region (Mb)	No. of sig. SNPs	Lead SNP	Top p	Candidate genes
1	163.05–163.67	3	1_163360584	5.69E‐8	ATP8B1, IGDCC3, DENND4A
1	205.23–222.25	17	1_205245991	1.07E‐8	SH3GL2, PIP5K1B
5	96.56–98.53	37	5_96567460	2.61E‐8	TMTC2
7	96.61–99.26	395	7_99232411	5.87E‐10	NUMB, BBOF1, LIN52, ABCD4, VRTN, SYNDIG1L, LTBP2, FLVCR2, TTLL5, TGFB3
12	83.84–84.42	8	12_8384361	2.84E‐08	—
13	83.53–91.16	274	13_86965563	3.90E‐9	PCOLCE2, CHST2, SLC9A9, DIPK2A, PLSCR1, PLSCR5, ZIC1, CPB1, HPS3, CPHL1, TM4SF1, WWTR1, SERP1

TABLE 3.

Summary of independent genomic loci associated with LMD.

Chr	Genomic region (Mb)	No. of sig. SNPs	Lead SNP	Top p	Candidate genes
2	17.82	1	2_17825553	6.09E‐08	CD82
7	35.29	2	7_35397754	9.42E‐08	SH3GL2, PIP5K1B
7	60.93	1	7_60932587	8.37E‐08	CELF6
7	77.68–78.41	4	7_77688479	8.72E‐08	RAB2B, SUPT16H, PNP
14	131.34–131.40	60	14_131348605	9.09E‐09	ATE1, NSMCE4A

3.4. TWAS and eQTL Co‐Localization

Utilizing expression quantitative trait loci data from 34 tissues, TWAS identified a total of 79 genes significantly associated with the two traits (Figure S1c and Table S4). Among these, 66 genes showed significant associations with BFT, 8 of which were specifically associated with expression in muscle tissue. The most significant TWAS signal was observed for the COMMD2 gene in testis (p = 6.34 × 10⁻¹¹). Notably, the ABCD4 gene, located on SSC7, exhibited significant associations between its predicted expression levels and BFT in both blood (p = 1.25 × 10⁻⁷) and brain (p = 5.46 × 10⁻⁷). Furthermore, 16 genes demonstrated significant associations with LMD. The most significant association for LMD was found for the EAPP gene in muscle tissue (p = 2.64 × 10⁻⁹), also located on SSC7. Expression of the gene ENSSSCG00000043238 was significantly associated with LMD in three tissues: jejunum (p = 3.00 × 10⁻⁸), blastocyst (p = 3.33 × 10⁻⁸), and small intestine (p = 1.53 × 10⁻⁷).

COLOC analysis revealed shared causal variants between BFT and gene expression for two genes, with the SNP rs1109589390 (SSC7:97603637) showing strong evidence of colocalization with the BFT association signal (Figure 2f–i). To investigate the regulatory role of rs1109589390, we queried the PigGTEx database and identified tissue‐specific effects on ABCD4 expression. Specifically, the G allele was associated with reduced expression in brain (AA = 135, AG = 178, GG = 106; G allele frequency = 0.465394; p = 5.02 × 10 ⁻¹³ ), frontal cortex (AA = 22, AG = 26, GG = 27; allele frequency = 0.533333; p = 2.77 × 10⁻⁶), and muscle (AA = 402, AG = 556, GG = 363; allele frequency = 0.485238; p = 3.16 × 10⁻⁹), but with increased expression in jejunum (AA = 33, AG = 34, GG = 8; allele frequency = 0.333333; p = 1.25 × 10⁻⁷). Consistent with this regulatory pattern in the brain, frontal cortex, and muscle tissue, genotype–phenotype analysis in our study population showed that individuals carrying the G allele (AG and GG genotypes) exhibited significantly higher BFT values than AA homozygotes (AA: n = 521, AG: n = 597, GG: n = 166; G allele frequency = 0.3618) as shown in Figure 2j–m. The ABCD4 gene was consistently implicated across meta‐GWAS, TWAS, and eQTL colocalization analyses, supporting its potential mechanistic role in regulating BFT in Pietrain pigs (Figure 2n). Additionally, SNP rs331837789 (SSC7:97330015) was colocalized with both muscle ENTPD5 expression and BFT (Figure S1d), further highlighting the complex genetic regulation of fat deposition on SSC7.

3.5. Cell‐Type‐Specific Enrichment of Trait‐Associated Genes

In our study of BFT and LMD, we analyzed single‐cell sequencing data from several tissues, including subcutaneous adipose, visceral adipose, muscle, liver, and brain regions involved in colocalization. Specific cell populations within both adipose tissues, as well as in muscle, brain, frontal lobe, and parietal lobe, demonstrated significant associations with BFT (Figures 3a and S1e). The liver, being central to lipid metabolism, displayed a notable abundance of cells and considerable heterogeneity that correlated with both traits, particularly among cholangiocytes, hepatic stellate cells, and hepatocytes. Within muscle tissue, cell types associated with muscle development and adipogenesis, such as fibro‐adipogenic progenitors (FAPs), muscle stem cells (MuSCs), and adipocytes, showed strong associations with BFT. These findings suggest a potential link between BFT and the adipogenic differentiation of FAPs, as well as functional changes in MuSCs, processes that may ultimately influence intramuscular fat deposition.

FIGURE 3.

Gene‐based single‐cell enrichment results and effects of ABCD4 on the proliferation and adipogenic differentiation of 3T3‐L1 cells. (a) Heatmap depicting the strength of association between various cell types in subcutaneous adipose tissue, visceral adipose tissue, liver, and muscle with the trait of interest. Color intensity represents the proportion of significantly associated cells within each cell type. The presence of a square indicates a significant association between the cell type and the trait at the population level. The presence of a cross indicates significant heterogeneity in the cell‐trait association among individual cells within that cell type. (b) Violin plot showing the gene expression levels of ABCD4 across different cell types within subcutaneous adipose tissue. The first UMAP visualization displays the distinct cell clusters identified in subcutaneous adipose tissue. The second UMAP illustrates the expression pattern of ABCD4 across these cellular populations. (c) Violin plots displaying the expression levels of ABCD4 in different cell types from visceral adipose tissue, liver, and muscle. (d) Expression levels of ABCD4 at different time points during adipogenic differentiation induced in 3T3‐L1 cells. (e) CCK‐8 assay performed 24 h after ABCD4 knockdown. (f) mRNA expression of genes related to the promoters and inhibitors (Cyclin D, Cyclin E, CDK4, and P21). (g) Oil Red O staining of lipid droplets and absorbance at 510 nm in ABCD4‐inhibited 3T3‐L1 cells after differentiation for 8 days. (h) Intercellular TG contents in ABCD4‐inhibited 3T3‐L1 cells after differentiation for 8 days. (i, j) mRNA expression of adipogenesis‐related genes (PPARγ, C/EBPα, ADIPOQ, FABP4, DGAT, LPL) after differentiation for 8 days. Data are presented as mean ± SEM. Significance was determined by Student's t‐test: *p < 0.05, **p < 0.01, and ***p < 0.001.

Among all genes within the scDRS‐derived trait‐associated gene sets, ABCD4 demonstrated the strongest association with BFT. We therefore focused on characterizing the cell‐specific expression patterns of ABCD4. This gene was broadly expressed across multiple cell types in subcutaneous adipose (Figure 3b), with particularly high expression in beige adipocytes. A similar pattern of heterogeneous expression across cell types was observed in visceral adipose, liver, muscle, brain, and multiple brain regions (Figures 3c and S1e). These results suggest that ABCD4 may be involved in adipogenesis. We propose ABCD4 as a key candidate gene for further verification through in vitro experiments.

3.6. Interference With ABCD4 Inhibits 3T3‐L1 Cell Proliferation and Promotes Cell Differentiation

As illustrated, the expression level of ABCD4 decreased on the second day of 3T3‐L1 cell differentiation induction and gradually increased from the fourth to the sixth day, suggesting that ABCD4 may play a role in the later stages of differentiation (Figure 3d). Given that the proliferation and differentiation of adipocytes influence lipid accumulation, we introduced three different concentrations of ABCD4‐targeting siRNA to assess its impact on cell proliferation. CCK‐8 assay results indicated that cell viability decreased progressively with increasing siRNA concentration. A significant reduction in cell viability was observed at a concentration of 90 nM (Figure 3e). Furthermore, qPCR analysis demonstrated that knockdown of ABCD4 significantly downregulated key promoters of the cell cycle, including Cyclin D, Cyclin E, and CDK4, while the level of the cell cycle inhibitor P21 remained unchanged (Figure 3f). This specific suppression of positive cell cycle regulators suggests that the down regulation of ABCD4 impedes cell proliferation.

To elucidate the role of ABCD4 in adipogenesis, we knocked down its expression in 3T3‐L1 preadipocytes using siRNA throughout the 8‐day differentiation period. Transfection with ABCD4 siRNA resulted in a significant reduction of ABCD4 expression compared to the negative control (Figure 3i). As illustrated in Figure 3g, this knockdown promoted adipogenic differentiation, as evidenced by a marked increase in lipid droplet accumulation visualized under microscopy and quantified by ORO staining. Consistent with this morphological change, intracellular TG levels were significantly elevated in ABCD4‐knockdown cells (Figure 3h). At the molecular level, the expression of key adipogenic markers was upregulated, including the master regulators PPARγ and C/EBPα, as well as functional genes such as FABP4, ADIPOQ, LPL, and DGAT (Figure 3j). These findings collectively demonstrate that ABCD4 knockdown enhances lipid accumulation and adipogenesis in 3T3‐L1 cells.

4. Discussion

This study aims to elucidate the genetic architecture underlying BFT and LMD to enhance breeding efficiency and commercial profitability in pigs. Utilizing low‐coverage sequencing and genome‐wide imputation, we constructed a high‐density SNP dataset, significantly improving our capacity to pinpoint causal variants and identify candidate genes. We evaluated genetic parameters in a large‐scale purebred Pietrain population and performed a meta‐analysis based on GWAS summary statistics from two genetically distinct subpopulations. Subsequent post‐GWAS analyses were conducted to prioritize candidate genes, followed by functional investigation in vitro of a key candidate gene for BFT, ABCD4, using mouse 3T3‐L1 preadipocytes to assess its role in adipogenesis.

The two Pietrain populations examined in this study originated from farms in two distinct regions of China. Although derived from a common ancestral background, they exhibited clear genetic divergence in PCA as a result of decades of separate selective breeding. BFT and LMD exhibited moderate to high heritability estimates, indicating that a substantial proportion of phenotypic variation is attributable to additive genetic effects. This supports their suitability for genetic mapping via GWAS. Notably, when adjusted to 115 kg body weight, the Pietrain populations in this study displayed markedly lower BFT values, averaging 9.09 and 8.84 mm, compared to reported values for Duroc (14.8–28 mm), Landrace (14.1 mm), Large White (14.5 mm) (Ogawa et al. 2023), and Pietrain‐cross commercial pigs (11.16–26.32 mm) (Van Wijk et al. 2005). Heritability estimates for BFT were consistent with those previously reported. Meanwhile, LMD measurements in the Pietrain populations, averaging 61.26 and 64.49 mm, were higher than those observed in Duroc (34.4–67.2 mm) and Pietrain‐cross commercial pigs (59.3 mm) (Bergamaschi et al. 2020). The heritability estimate for LMD was slightly higher than that reported by Van Wijk et al. (2005). Current understanding of the genetic parameters for carcass traits in Pietrain pigs remains limited, and direct comparisons across studies are challenging due to inconsistencies in measurement age and weight protocols (Lenoir et al. 2022). The considerable difference in genetic correlations between the two populations may be partly explained by pronounced disparities in sample sizes. Collectively, these findings underscore the distinct genetic advantages of Pietrain pigs in lean meat yield and muscular development, reinforcing their established role as terminal sires for enhancing meat production efficiency in commercial swine operations.

The meta‐GWAS analysis further identified genetic loci specifically associated with BFT and LMD in the Pietrain population, with the abundance of significant signals reflecting the pleiotropy and complexity underlying the regulation of these traits. For BFT, overlapping genetic regions on SSC1 and SSC7 were consistent with those reported in a previous meta‐analysis of Duroc, Landrace, and Large White breeds from the PigBiobank database, strengthening the credibility of these findings. Among the genes mapped within these regions, several genes, including SH3GL2 (El Nagar et al. 2023), NUMB (Wang et al. 2024), LIN52 (Zhao et al. 2020), and LTBP2, have previously been shown to be associated with fat deposition, body fat distribution, or hypertriglyceridemia in poultry and livestock. For the LMD trait, we also enriched some genetic signals at SSC7 and SSC14. However, the interpretation of the results is further complicated by the unusually long linkage disequilibrium blocks in the SSC7 region, likely resulting from extended haplotype homozygosity due to targeted selection. To clarify these associations and mitigate false‐positive risks, future studies should prioritize high‐resolution fine‐mapping in independent cohorts with larger sample sizes.

Our integrated post‐GWAS analyses establish ABCD4 as a key regulator of BFT. TWAS revealed significant associations between ABCD4 expression in both blood and brain tissues and BFT. Further strengthening this link, eQTL colocalization analyses identified the downstream SNP rs1109589390 as a significant regulator of ABCD4 expression. This finding aligns with the established pig genetic regulatory map, which has previously reported colocalization between BFT GWAS signals and ABCD4 cis‐eQTLs in tissues including the brain and small intestine (Teng et al. 2024). A pivotal observation emerged from the apparent discrepancy between the genetic and functional genomic data: while the G allele of rs1109589390 is associated with increased BFT, we found no evidence that it acts as a cis‐eQTL for ABCD4 in adipose tissue. This paradox may be explained by the tissue‐specific regulatory nature of rs1109589390, which exhibits opposing directional effects on ABCD4 expression in tissues such as the brain, frontal cortex, and muscle compared to the jejunum. This suggests that the SNPs' influence on fat deposition may involve systemic mechanisms affecting whole‐body energy metabolism rather than direct local regulation within adipocytes (Laumen et al. 2008). However, it is worth noting that the sample size for the GG genotype in the jejunum was relatively limited. Therefore, while the statistical significance is high, this tissue‐specific reversal should be interpreted with caution and warrants further validation in a larger independent cohort. Supporting a direct role for ABCD4 in adipose biology, single‐cell RNA sequencing demonstrated high ABCD4 expression in subcutaneous beige adipocytes. To functionally validate this, we performed ABCD4 knockdown in 3T3‐L1 cells, which significantly enhanced adipogenesis. The concordance between the BFT‐increasing G allele and the pro‐adipogenic effect of ABCD4 knockdown reveals a compelling functional link, yet the precise molecular nature of the G allele remains to be fully elucidated. Our study firmly positions ABCD4 as a key regulator of fat deposition and nominates rs1109589390 as a tissue‐specific variant, providing a foundation for future studies to directly interrogate its functional consequences.

While this integrated genomic study provides novel insights into the genetic architecture of BFT and LMD in Pietrain pigs, several limitations should be acknowledged. First, our meta‐GWAS analysis, while powerful, may be susceptible to false positive associations. These could stem from residual population stratification between the two subpopulations or from extended linkage disequilibrium blocks specific to the Pietrain breed, a potential signature of its intense selective breeding history. Second, our transcriptomic analyses, including eQTL colocalization and single‐cell data interpretation, relied heavily on publicly available datasets from other breeds or tissues. Although this approach allowed for the identification of regulatory mechanisms potentially shared across different pig breeds, the lack of a dedicated, breed‐specific multi‐tissue transcriptome resource for Pietrain pigs limits the precision of our functional inferences. This limitation is particularly relevant for interpreting the absent colocalization signal for ABCD4 in adipose tissue, underscoring the critical influence of genetic background and tissue context on regulatory variant mapping. Finally, although the functional validation of ABCD4 in mouse 3T3‐L1 preadipocytes demonstrated a pro‐adipogenic role, the translational relevance to porcine biology requires caution due to inherent species differences in adipocyte physiology and gene regulatory networks.

To address these limitations, future work should prioritize several avenues. High‐resolution fine‐mapping in larger, independent Pietrain cohorts is essential to distinguish true causal variants from spurious associations linked by long‐range LD. Furthermore, generating a comprehensive Pietrain‐specific transcriptome atlas across key developmental stages and metabolic tissues would greatly enhance the discovery of breed‐relevant regulatory mechanisms and improve the colocalization power for traits like BFT. Ultimately, direct functional validation using porcine primary preadipocytes or the creation of pig models with targeted genetic manipulations will be the gold standard for conclusively establishing the physiological roles of ABCD4 and other candidate genes identified in this study.

5. Conclusion

By integrating multi‐omics and statistical genetics approaches, this study systematically dissected the genetic basis of BFT and LMD in Pietrain pigs. We identified a large number of significantly associated genetic loci and candidate genes, revealing a complex genetic architecture and pleiotropic regulatory patterns underlying these traits. Notably, ABCD4 was pinpointed as a key candidate gene for BFT regulation through convergent evidence from genetic association, expression colocalization, and functional validation in mouse 3T3‐L1 cell models, where it significantly influenced lipid accumulation. These findings provide important theoretical foundations and genetic targets for molecular breeding of Pietrain pigs and enhance our understanding of the genetic mechanisms controlling fat deposition and muscle development in swine.

Author Contributions

Quanjun Zhan: writing – original draft, data curation, formal analysis, validation, visualization. Hong Xie: formal analysis, writing – review and editing, resources. He Han: formal analysis. Zhenyang Zhang: data curation and formal analysis. Ran Wei: formal analysis. Haixia Li: resources and validation. Wei Zhao: data curation. Bo Feng: formal analysis and visualization. Xiaoliang Hou: resources. Jianlan Wang: data curation. Yongqi He: investigation and data curation. Yan Fu: methodology, and writing – review and editing. Xiaoling Guo: supervision and writing – review and editing. Yuchun Pan: writing – review and editing, conceptualization, resources, supervision, project administration. Chong Cao: methodology, validation, writing – review and editing, and supervision. Zhe Zhang: conceptualization, methodology, supervision, writing – review and editing, funding acquisition, project administration. All authors have read and agreed to the published version of the manuscript.

Funding

This work was supported by the National Natural Science Foundation of China (32272832) and Zhejiang Science and Technology Major Program on Agricultural New Variety Breeding (Grant 2021C02068).

Ethics Statement

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: SNP density, quantile‐quantile plots of meta‐GWAS, TWAS results, COLOC results for ENTPD5 and sc‐RNA data analysis results.

Table S1: Primer sets for quantitative real‐time PCR.

Table S2: Significant loci linked to BFT.

Table S3: Significant loci linked to LMD.

Table S4: The result of TWAS.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.