Genome-wide identifications of regulatory variants influencing milk fat percentage in Chinese Holstein cow Authors: Changliang Gong, Yaran Zhang, Chunhong Yang, Zhihua Ju, Wenhao Liu, Qiang Jiang, Xiuge Wang, Yao Xiao, Xiaochao Wei, Yaping Gao, Lingling Wang, Yundong Gao, Jinpeng Wang, Jinming Huang Published: 2025 ABSTRACT AbstractBackgroundMilk fat is a critical economic trait in Holstein cattle; however, the identification of genomic loci associated with milk fat content remains limited, constraining progress in genetic improvement and overall dairy production efficiency. A comprehensive understanding of the genetic and regulatory mechanisms underlying milk fat synthesis is therefore essential for advancing dairy breeding and productivity.ResultsWe conducted a genome-wide association study (GWAS) using a linear mixed model (LMM) on the genomic estimated breeding values (GEBVs) of milk fat percentage (MFP) in 2,709 Chinese Holstein cows. Population structure and relatedness were controlled using a genomic relationship matrix (GRM), and multiple testing was corrected by applying the False Discovery Rate (FDR) method. A total of 104 significant SNPs (FDR < 0.05) were identified, including 32 SNPs clustered in the DGAT1 region on BTA14 (1.3–2.3 Mb, Bos_taurus_UMD_3.1.1). Favorable alleles for each SNP were identified. Among the significant variants, eight SNPs located within the exons of SPAG1, SLC15A5, PLEKHA5, and HERC6 were found, including four missense mutations predicted to increase protein stability according to I-Mutant and MUpro analyses. Validation in an independent population of 99 cows confirmed associations between MFP and three intergenic SNPs: BovineHD1400001112, BovineHD1400000698, and BovineHD0500026460, all located in predicted transcriptional regulatory regions. Dual-luciferase assays demonstrated that these SNPs affected transcriptional activity, with BovineHD1400001112 influencing the binding of SP1, a core regulator of milk fat synthesis. qRT-PCR in peripheral blood (T/T vs. C/C) showed differential expression of FASN and ACACA.ConclusionsThese findings support the notion that BovineHD1400001112 may act as a regulatory variant influencing milk fat synthesis through modulation of key transcription factors and nearby genes, providing insights into potential targets for genetic improvement in Holstein cattle.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12864-025-12449-1.Keywords: Milk fat, Regulatory variant, Transcription factor, Cattle Background MFP is one of the most important economic traits in dairy cattle breeding [1–3]. In Holstein cattle, genetic improvement of milk fat content has long been a primary objective of breeding programs, resulting in substantial genetic gains over recent decades. Whole-genome selection has played a pivotal role in accelerating this progress by enabling more accurate and efficient selection decisions. The advent of high-density SNP chips, containing over 50,000 markers (50 K), has further facilitated the widespread adoption of genomic selection in major dairy-producing countries, including the United States, Canada, the United Kingdom, Ireland, New Zealand, Australia, France, the Netherlands, and Germany [4, 5]. GWAS have become indispensable tools for identifying genetic loci underlying complex traits such as MFP, and the SNPs identified through these studies have markedly improved the accuracy of genomic selection. Notable progress has been made in elucidating the genetic architecture of MFP through GWAS. Cross-population integrative analyses have revealed that approximately 50% of the SNPs significantly associated with MFP are clustered on bovine chromosome 14 [6]. Among the most consistently reported candidate genes is Diacylglycerol O-acyltransferase 1 (DGAT1), which encodes a key enzyme involved in triglyceride synthesis and accounts for a substantial proportion of the genetic variation in MFP [7]. Other genes, such as Stearoyl-CoA desaturase 1 (SCD1) on chromosome 6, LPIN1 on chromosome 19, and Growth hormone receptor (GHR) on chromosome 20, have also been repeatedly validated by independent studies as being significantly associated with the milk fat synthesis pathway [8–10]. Most GWAS investigating milk fat and related production traits in Holstein cattle have been conducted within Western breeding programs—particularly in Europe, North America, and Australia—using large national cohorts that have strongly shaped our current understanding of the trait’s genetic architecture [11, 12]. These datasets reflect unique breeding histories, management systems, and predominantly temperate environments that differ considerably from the diverse production settings in China. Consequently, both the effect sizes and even the associated loci identified in Western populations may not be directly transferable to Chinese Holsteins [13]. Although several GWAS have been performed in Chinese Holstein populations, these studies are relatively limited in number and often smaller in scale. Moreover, Chinese substantial climatic variation (e.g., hot–humid southern regions versus colder northern regions) and heterogeneous management practices could contribute to population-specific genetic signals for milk fat percentage [14, 15]. Therefore, conducting a GWAS specifically focused on Chinese Holstein cattle is essential to identify genetic variants with direct relevance to domestic selection and to support the development of China-specific genomic breeding indices. Furthermore, many previous studies have primarily focused on statistical associations between SNPs and phenotypic traits, with limited efforts toward experimental validation. Consequently, the biological mechanisms through which these candidate SNPs may influence milk fat synthesis, such as potential effects on transcription factor binding and gene expression, remain largely unexplored. Addressing these gaps is important for confirming the relevance of known loci in Chinese Holstein cows and for identifying novel SNPs that could be leveraged in breeding programs aimed at improving milk fat content and dairy production efficiency. In this study, we genotyped 2,709 Chinese Holstein cows from 11 farms using the Illumina BovineSNP50 chip and obtained their GEBVs for MFP from the Council on Dairy Cattle Breeding of America (CDCB). GWAS analysis identified SNPs significantly associated with this trait, which were further validated in a Holstein population selected for extreme MFP. Selected intergenic SNPs were assessed for potential transcriptional regulatory activity using dual-luciferase reporter assays, and qRT-PCR was performed to detect differential expression of genes potentially regulated by the corresponding transcription factors. While these experiments provide evidence consistent with regulatory effects, the exact molecular mechanisms underlying these associations remain to be elucidated. This study identifies SNPs associated with MFP in Chinese Holstein cattle and offers preliminary insights into their possible regulatory roles, which may inform future genomic selection and molecular breeding strategies. Methods Ethics statement The processes of collection of samples and traits data were performed according to the regulation of the Ministry of Agriculture of the People’s Republic of China and the Animal Ethics Committee of Shandong Academy of Agricultural Sciences. Samples for GWAS analysis A total of 2,709 six-month-old Holstein cows from 11 farms of China were selected and sampled from March to August 2023 for GWAS on GEBVs of MFP trait (the samples information was listed in Supplementary Table 1). Blood samples were collected from the tail vein using a sterile lancet, placed in a 5 mL vacuum blood collection tube, and anticoagulant (EDTA) was added. After collection, samples were transferred to the laboratory for storage at 4 °C and subsequently sent to Neogen Biotechnology (Shanghai, China) Co., Ltd. for DNA extraction and genotyping using Illumina BovineSNP50 chip (based on the reference genome Bos_taurus_UMD_3.1.1). SNP quality control and principal component analysis (PCA) on genotypes Data quality control was performed to remove SNPs and individuals with an insufficient genotyping quality. SNPs located on the sex chromosome or those with unknown positions were excluded from subsequent analysis. The SNPs with call rates less than 95% or with minor allele frequency (MAF) less than 0.05 and SNPs with genotypes not in accordance with the Hardy–Weinberg equilibrium (P > 10− 6) were also eliminated. Since the SNP positions in the BovineSNP50 chip data are based on the Bos taurus UMD3.1.1 reference genome, and some alleles are reported on the negative strand, we converted all chip-derived alleles to correspond to the reference alleles of the Bos taurus ARS-UCD1.2 genome to ensure consistency with the subsequent resequencing data. Quality control and PCA were performed using PLINK v1.9 [16]. GWAS and LD for MFP GEMMA v0.98.5 software was used to detect the correlation between SNP loci and phenotype by univariate LMM, which included fixed effect and random effect, and the specific model was as follows: where y is the phenotypic value; a is the population mean; b is a fixed effect, including farm effect, year of birth and season of birth effect; x is the SNP genotype indicator variable, taken as 0, 1, or 2, respectively. g is the residual polygenic effect which follows the normal distribution N(0,Kσg2), where K represents the kinship matrix that describes genetic relatedness among individuals and σg2 is the variance of the genetic effect, indicating the contribution of genetic factors to phenotypic variation. e is the random residual effect which follows the normal distribution N(0,Iσe2), where I is an identity matrix, indicating the assumption that environmental effects are independent across individuals and σe2 is the variance of the environmental effect, indicating the contribution of environmental factors or residual error to phenotypic variation. To avoid false positives, the results of the association analysis were corrected using the FDR method. SNPs with an FDR of less than 0.05 were identified as significant SNPs. The genomic inflation factor (λ_GC) was calculated to assess potential population stratification, defined as the median of the observed chi-squared test statistics divided by the expected median under the null hypothesis. Linkage disequilibrium (LD) analysis was conducted using Haploview 4.2 [17]. Target SNP genotypes were extracted from the quality-controlled PLINK binary files (.bed/.bim/.fam) and converted into Haploview compatible .ped and .info formats with PLINK 1.9. Pairwise LD coefficients (r²) were calculated and displayed as heatmaps using a white-to-red gradient (white represents r² = 0 and red represents r² = 1). LD blocks were defined according to the algorithm of Gabriel et al. [18]. Functional enrichment analysis Candidate genes were searched within the ± 200 kb region flanking the significant SNPs using the BioMart tool [19]. Then, the set of candidate genes was analyzed using the gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses using the Omicshare platform (https://www.omicshare.com/tools/, accessed on 27 May 2024). Validation population for significant SNPs A total of 1,000 lactating Holstein cows, mostly in their 2nd to 4th lactation, were selected as the validation population from a farm located in Shandong Province, China. The MFP of each dairy cattle was obtained according to the Agricultural Industry Standard entitled “Technical specification of production performance test for dairy cattle”, which is issued by Ministry of Agriculture and Rural Affairs of the People’s Republic of China. Milk samples were collected monthly within the first 10 months after calving. MFP for each sample was measured using the Bentley milk component analyzer, and the individual average MFP was calculated for each cow. Subsequently, cows with extreme phenotypes at both ends of the distribution were selected to form two experimental groups: the High MFP group and the Low MFP group. These groups comprised a total of 99 cows, which formed the validation population independent from the discovery cohort (Supplementary Table 2). Genotypes of the three replicated intergenic SNPs were quality-controlled, with MAF > 0.05 and call rates > 98%; no genotype imputation was applied. The blood samples were collected from these 99 cows and used for whole-genome resequencing. Whole-genome resequencing and data analysis Genomic DNA was extracted from 99 cattle with extreme phenotypes in the validation cohort. Bioinformatics analysis of whole-genome resequencing data was performed as follows: Sequencing libraries were constructed using the BGI standard protocol (BGI-Shenzhen, China) and subjected to paired-end sequencing (150 bp read length, PE150) on the BGI-Seq platform, achieving an average depth of 10× coverage. First, the short reads generated from sequencing were aligned to the reference genome ARS-UCD1.2 with bwa-mem algorithm from the BWA v7.3.0 [20] with default setting. Next, the resulting Binary Alignment/Map (BAM) files were processed using Samtools v1.3.1 [21] for sorting, marking of duplicate reads, and other quality control steps. Finally, Bcftools v1.3.1 [22] was employed to identify SNPs and perform genotype calling. Student’s t-test was used to investigate the association relationship between significant SNPs from the aforementioned GEBV and MFP in the validation population. Transcription factor binding site prediction To investigate the potential regulatory functions of the three candidate SNPs in the validation population, transcription factor binding site (TFBS) prediction analysis was performed using the JASPAR database (https://jaspar.elixir.no/). Structural impact prediction The prediction of changes in protein stability upon amino acid substitutions was conducted utilizing I-Mutant 2.0 [17] and MUPro [18]. Negative ΔΔG values indicate decreased stability, whereas positive values suggest increased stability. AlphaFold3 [23] was employed to predict the three-dimensional structures of amino acid sequences corresponding to the different allelic variants. PyMOL [24] was used to visualize the protein data bank files generated by AlphaFold3 and to calculate the root mean square deviation (RMSD), which measures the average displacement of backbone atoms between two aligned structures. Luciferase reporter assays Genomic fragments encompassing ± 50 bp flanking each SNP locus were amplified via high-fidelity PCR (Platinum SuperFi II DNA Polymerase, TaKaRa), followed by KpnI/XhoI double digestion and directional cloning into the multiple cloning site upstream of the firefly luciferase gene in the pGL3-Promoter vector. All recombinant constructs were validated by bidirectional Sanger sequencing with Q30 scores exceeding 95%. For transient transfection assays, HEK293T cells were seeded in 24-well plates at 1 × 105 cells/well and co-transfected using Lipofectamine™ 3000 (Thermo Fisher, L3000008) with 500 ng test plasmid (reference/mutant variants) and 50 ng pRL-TK normalization plasmid. Cells were harvested 48 h post-transfection, and dual-luciferase activity was quantified using the Dual-Luciferase® Reporter Assay System (Promega, E1910) on a GloMax Navigator microplate reader, with firefly luciferase signals normalized to Renilla values. Firefly luciferase activity was normalized to Renilla signals. Each construct was assayed in three biological replicates, each with three technical replicates. Statistical significance was assessed by two-tailed unpaired Student’s t-test. Quantitative real-time PCR (qRT-PCR) validation To validate the regulatory impact of the SNP BovineHD1400001112, we selected six individuals with homozygous genotypes (three C/C and three T/T) (Supplementary Table 3) from the 99 validated population and measured the relative mRNA expression levels of nearby genes as well as target genes of the corresponding transcription factors in blood. Peripheral blood samples were collected from these animals for RNA extraction. Total RNA was isolated using the TRIzol reagent (TaKaRa, Dalian, China) according to the manufacturer’s instructions. RNA quality and concentration were assessed by spectrophotometry, and samples with an A260/A280 ratio between 1.9 and 2.1 were used for subsequent analysis. First-strand cDNA was synthesized from 1 µg of total RNA using the PrimeScript™ RT reagent Kit with gDNA Eraser (RR047A, TaKaRa, Dalian, China). Quantitative real-time PCR (qRT-PCR) was performed using the TB Green® Premix Ex Taq™ II (RR820A, TaKaRa, Dalian, China) on a LightCycler 480 system (Roche, Mannheim, Germany) under a two-step amplification protocol following the manufacturer’s guidelines. Primers used for bovine genes, which were synthesized by Beijing Tsingke Biotech Co, Ltd., are listed in Supplementary Table 4. Relative mRNA expression levels were normalized to β-actin and quantified using the 2−△△Ct method. Each sample was analyzed in three technical replicates, and differences between genotypes (C/C vs. T/T) were evaluated using a two-tailed unpaired Student’s t-test. Results PCA and GEBV analyses of GWAS population A total of 44,242 SNPs were used for GWAS analysis after quality control and the average genotyping rate was 99%. PCA clustering showed that the dairy cattle were largely grouped according to their geographic origin (Fig. 1A), indicating subtle population stratification likely due to regional breeding and management practices. To account for this structure, the first five principal components (PCs) were included as covariates in the GWAS linear mixed model. The histogram plot showed that the GEBVs of MFP largely conform to a normal distribution (Supplementary Fig. 1). The basic descriptive statistics of GEBV of MFP for 2709 Holstein cattle are listed in Supplementary Table 5. Genome-wide analysis of MFP in Holstein cattle. A PCA clustering of dairy cattle populations based on high-density genotyping data. Manhattan (B) and quantile–quantile (C) plots of genome-wide association results for MFP in Holstein cattle. In the Manhattan plot, chromosomal positions (Chr 1–29) are shown on the x-axis and statistical significance (− log10 P) on the y-axis, with regional SNP density indicated in the legend. The QQ plot displays the observed versus expected − log10 P values. D LD structure of the DGAT1 region (1.3–2.3 Mb) on BTA14.The plot shows the LD pattern among genome-wide significant SNPs associated with MFP within the 1.3–2.3 Mb interval on BTA14, which encompasses the DGAT1 locus (Bos_taurus_UMD_3.1.1). Each diamond represents a SNP, and the color scale indicates the pairwise LD (R²) values between SNPs, with darker colors representing stronger LD GWAS on GEBVs of MFP GWAS analysis identified 104 genome-wide significant SNPs associated with GEBV for MFP (Fig. 1). The majority of these SNPs are located on chromosomes 5, 6, and 14 (Supplementary Table 6). Eight SNPs are located in the exon regions of PLEKHA5, MGST1, SLC15A5, HERC6, SPAG1, OSR2, FBXO43, and SLC52A2, of which four SNPs are missense mutations in PLEKHA5, SLC15A5, SPAG1, and HERC6. Fifteen SNPs are mapped to the intronic regions of ZBTB7B, MGST1, TRAPPC9, and MAPK15, while the remaining 81 SNPs are located in intergenic regions. Moreover, the enrichment of association signals within the 1.3–2.3 Mb (Bos_taurus_UMD_3.1.1) interval on BTA14, which harbors the well-established DGAT1 locus, further supports the reliability of our experimental design and validates the robustness of the detected associations (Fig. 1D). GO and KEGG analyses A total of 140 genes were identified within 200 kb upstream and downstream of the 104 significant SNPs. GO analysis detected 54 significantly enriched terms (P < 0.05), including biological processes related to substance transport, stress response, and RNA regulation (Fig. 2). Importantly, several enriched terms were directly linked to lipid metabolism, such as monoglyceride biosynthesis, acetyl-CoA/fatty acyl-CoA transport, and the mevalonate pathway, all of which contribute to the provision and mobilization of precursors required for milk fat synthesis [25–27]. GO enrichment of GWAS candidate genes for MFP. Bars are colored by GO domain—Biological Process (red), Cellular Component (blue), and Molecular Function (green). The x-axis shows − log10(pvalue), indicating enrichment significance (larger values denote stronger enrichment). “Gene count” is the number of genes associated with each GO term KEGG pathway analysis identified four significantly enriched pathway (P < 0.05), including metabolic pathways, nicotinate and nicotinamide metabolism, glutathione metabolism, and inositol phosphate metabolism (Table 1). The metabolic pathways contained key lipid-associated genes, including DGAT1 and HMGCS1, which are essential for triglyceride synthesis and vesicle stability, respectively [28]. Enrichment of nicotinate and nicotinamide metabolism (ko00760) involved NNT, a gene that helps sustain NADPH production, thereby supporting de novo fatty acid synthesis. Meanwhile, glutathione metabolism (ko00480), involving GST and OPLAH, suggests a role for redox homeostasis in maintaining efficient lipid secretion. Together, these pathways highlight the involvement of both precursor supply and cellular metabolic balance in regulating milk fat biosynthesis. Table 1Significant KEGG pathways and associated genes identified in the GWAS of MFPTerm IDDescriptionP-ValueGeneko01100Metabolic pathways0.014PYCR3, NNT, PNCB, SIAT4A, GST, GPT, PIK3C2, OPLAH, DERA, HMGCS1, COX6C, GAA1, PLCZ, DGAT1ko00760Nicotinate and nicotinamide metabolism0.015NNT, PNCBko00480Glutathione metabolism0.03GST, OPLAHko00562Inositol phosphate metabolism0.048PIK3C2, PLCZ Significant KEGG pathways and associated genes identified in the GWAS of MFP Predicted effects of missense SNPs on protein structural stability and conformation The effects of four missense SNPs on protein structural stability were evaluated using the I-Mutant 2.0 and MUPro (Table 2). We found that the favorable alleles associated with higher MFP at chr14:66328304 (located in SPAG1) and chr5:94055934 (located in SLC15A5) exhibited higher protein stability compared with the alternative alleles, whereas the favorable alleles at BovineHD0500025863 (located in PLEKHA5) and BovineHD4100004526 (located in HERC6) showed lower stability relative to the alternative alleles. Table 2The effects of missense SNPs on protein structural stability predicted with I-Mutant 2.0 and MUProSNPsGene SymbolFavorableallele for high MFPGenotypesAmino acidsΔΔG(I-Mutant2.0)ΔΔG (Mupro)Favorableallelethe other alleleFavorableallelethe other allelechr14:66328304 SPAG1 GGgcAgcGS-1.09-0.802chr5:94055934 SLC15A5 GGgtAgtGS-1.15-0.449BovineHD0500025863 PLEKHA5 AtTg (minus strand)tCg(minus strand)LS0.160.031BovineHD4100004526 HERC6 GgCt (minus strand)gTt(minus strand)AV1.511.90The alleles are the reference alleles based on the assembly of ARS-UCD1.2. ΔΔG = ΔG_preferred − ΔG_alternative, where ΔG_preferred and ΔG_alternative denote the folding free energies of the proteins encoded by the preferred and alternative alleles, respectively The effects of missense SNPs on protein structural stability predicted with I-Mutant 2.0 and MUPro The alleles are the reference alleles based on the assembly of ARS-UCD1.2. ΔΔG = ΔG_preferred − ΔG_alternative, where ΔG_preferred and ΔG_alternative denote the folding free energies of the proteins encoded by the preferred and alternative alleles, respectively AlphaFold-based analysis of protein conformations indicated that three missense SNPs had minimal effects on protein structure (RMSD < 1), whereas BovineHD0500025863 caused a substantial conformational change (RMSD = 20.849, Fig. 3). AlphaFold3-based structural comparison of protein conformations derived from the two allelic amino acid sequences of missense SNPs in four genes. A Conformations of SPAG1 carrying chr14:66328304 variant. B Conformations of HERC6 carrying BovineHD4100004526 variant. C Conformations of PLEKHA5 carrying BovineHD0500025863 variant. D Conformations of SLC15A5 carrying chr5:94055934 variant. Conformations corresponding to the favorable alleles are shown in cyan, while those corresponding to the alternative alleles are shown in green. Amino acids altered by the missense SNPs are labeled at their positions and highlighted in red. Lower RMSD values indicate minimal deviation and structural similarity, whereas higher values suggest substantial conformational changes that could affect protein stability or function Resequencing of cattle with extreme MFP phenotypes Resequencing of 99 validation cows generated approximately 2,700 GB of raw reads, achieving 10× coverage for each cattle genome. After filtering low-quality reads and SNPs, a total of 2,168,550 high-quality biallelic SNPs were retained. Among the 104 candidate SNPs initially detected by GEBV-based GWAS, only 33 were found to be polymorphic in the validation population (Supplementary Table 7). The reduced number of polymorphic loci may be attributed to the sequencing coverage and/or limited allelic diversity within the sampled population. Association analysis of these 33 polymorphic SNPs with MFP showed that 30 loci did not exhibit significant effects, while only three intergenic SNPs—BovineHD0500026460 (Chr5:92772499), BovineHD1400000698 (Chr14:2390445), and BovineHD1400001112 (Chr14:3768615), remained significantly associated with MFP in the validation cohort (Table 3). Table 3Significant SNPs associated with MFP in the validation populationSNP IDSNP positionFavorable allele for high MFPGenotypes and corresponding MFP (%)P-ValueBovineHD0500026460chr5:92772499CT/T (3.78 ± 0.06)C/T (4.05 ± 0.11)0.002BovineHD1400000698chr14:2390445GA/A (3.74 ± 0.05)G/A (4.03 ± 0.11)0.016BovineHD1400001112chr14:3768615CT/T (3.70 ± 0.08)C/T (3.91 ± 0.07)0.009The MFP is reported as mean ± SD. SD, standard deviation. The alleles presented here have been aligned to the bovine reference genome version ARS-UCD1.2 Significant SNPs associated with MFP in the validation population The MFP is reported as mean ± SD. SD, standard deviation. The alleles presented here have been aligned to the bovine reference genome version ARS-UCD1.2 In silico TF binding prediction and dual-luciferase assays reveal allele-specific transcriptional effects Three SNPs (BovineHD0500026460, BovineHD1400000698, and BovineHD1400001112) were predicted to affect the binding sites of MEF2A, MTF1, and SP1, respectively, according to the JASPAR database (Fig. 4A-C). The results demonstrated that all three SNPs exhibited significant allele-dependent effects on transcriptional activity (Fig. 4D–F). For BovineHD0500026460 (T > C), the C allele—which was associated with higher MFP in GWAS—increased promoter activity by 21.4% ± 2.1% compared to the T allele (P = 0.0022). In contrast, for BovineHD1400000698 (A > G), the G allele—also linked to elevated milk fat—reduced luciferase activity by 45.0% ± 3.8% relative to the A allele (P = 0.0159). For BovineHD1400001112 (C > T), the C allele significantly increased transcriptional activity by 47.0% ± 4.2% compared with the T allele (P = 0.0096). Identification and functional validation of significant SNP loci through transcription factor binding predictions and dual-luciferase reporter assays. A The motif of MEF2A binding sites and alleles of BovineHD0500026460 (in red color). B The motif of MTF1 binding sites and alleles of BovineHD1400000698 (in red color). C The motif of SP1 binding sites and alleles of BovineHD1400001112 (in red color). D Relative luciferase activity ratio at locus BovineHD0500026460. E Relative luciferase activity ratio at locus BovineHD1400000698. F Relative luciferase activity ratio at locus BovineHD1400001112. The data are shown as the mean ± SD. * P < 0.05, **P < 0.01 Mutation at BovineHD1400001112 altered expression pattern of key genes participated in milk fat metabolism Given that SP1 has been reported to participate in the regulation of milk fat metabolism, we examined the relative mRNA expression of its target genes and interacting genes, along with genes neighboring BovineHD1400001112, in samples with different genotypes of this SNP (Fig. 5). Relative mRNA expression of candidate genes affected by the SNP BovineHD1400001112, measured by qRT-PCR. Data are presented as mean ± SD. For relative quantification, the CC genotype served as the reference (calibrator), and all expression values were normalized to the CC group (set to 1.0). Statistical significance was defined as P < 0.05 and P < 0.01 Among the seven target and interacting genes (CREBBP, EP300, TP53, RELA, FASN, ACACA, and DGAT1), RELA was significantly downregulated in samples carrying the homozygous preferred allele (CC genotype), whereas FASN and ACACA were significantly upregulated in samples with the homozygous preferred allele. In addition, among the genes neighboring BovineHD1400001112, PTP4A3 and TRAPPC9 were markedly upregulated in samples with the homozygous preferred allele. Discussion Milk fat percentage is a pivotal economic trait that determines milk quality and strongly influences pricing. For example, each 0.1% increase in MFP is estimated to boost the raw milk premium by approximately 2.3%, which underscores its central role in guiding selection strategies in global dairy breeding programs [29, 30]. While major genes like DGAT1 and FASN have been identified as critical contributors to milk fat synthesis, the underlying regulatory variants and molecular mechanisms governing MFP remain incompletely characterized. Furthermore, Holstein populations across different regions have likely evolved distinct regulatory networks, shaped by unique genetic backgrounds and adaptive pressures [31, 32]. This regional variability underscores the need to further dissect the genetic and regulatory basis of MFP. Notably, 32 of the 58 significant SNPs located on chromosome BTA14 were enriched in the 1.3–2.3 Mb region, a genomic interval that harbors DGAT1, a well-established key gene in milk fat synthesis. DGAT1 plays a key role in intestinal fat absorption, lipoprotein assembly, and triglyceride metabolism [33, 34]. The K232A missense mutation of DGAT1 has been shown to be significantly associated with changes in milk fat content. Furthermore, the direction of this effect is consistent with findings from cross-species studies, providing evidence that reinforces the evolutionary conservation of DGAT1’s function in lipid metabolism [7]. Beyond BTA14, two loci on BTA5, MGST1 (BovineHD0500026619) and SLC15A5 (BovineHD0500026737) were found to activate the mTORC1 signaling pathway: MGST1 acts via glutathione metabolism, while SLC15A5 acts via oligopeptide transport. This observation aligns with the conserved regulatory mechanisms of lipid synthesis enzymes such as FASN and SCD1, and recent studies further support the universality of the mTORC1 pathway in mediating lipid metabolism. To gain deeper insights into MFP regulation, we performed GO and KEGG pathway analyses, which revealed that candidate genes for MFP were significantly enriched in core metabolic and regulatory networks. Among these, metabolic pathways emerged as a central hub, encompassing most candidate genes, including DGAT1, a rate-limiting enzyme gene for milk fat synthesis. Consistent with previous studies, DGAT1 polymorphisms strongly correlate with milk fat content, further validating its central role in this hub. Further dissection of specific pathways uncovered additional regulatory links: In the nicotinate and nicotinamide metabolism pathway, the genes NNT and PNCB may influence fatty acid oxidation and lipid droplet formation by regulating NAD+/NADH homeostasis. This in turn, has been shown to regulate cellular redox equilibrium [35]. Meanwhile, in the glutathione metabolism pathway, GST and OPLAH have been shown to play a crucial role in mitigating oxidative stress in mammary cells. Oxidative stress, which is induced by intensive lipid synthesis, is prevented by the sustained activation of the antioxidant defense system. Notably, the enrichment of the inositol phosphate metabolism pathway suggests a role for genes like PIK3C2 and PLCZ in coordinating lipid synthesis and secretion with hormonal signals (e.g., insulin, prolactin). These genes act via phosphoinositide signaling, which mediates Ca²⁺ dynamics and membrane trafficking to link hormonal cues to lipid-related cellular processes. Collectively, the synergistic interactions among these pathways reveal a dual regulatory network: one layer focused on metabolic execution (e.g., DGAT1-driven triglyceride synthesis) and the other on dynamic regulation (e.g., NAD + homeostasis, redox balance, and hormonal responses). This network provides novel insights into the multi-layered mechanisms that govern milk fat synthesis. In this study, we assessed the impact of four missense SNPs on protein structural stability using I-Mutant 2.0, MUPro, and AlphaFold-based analyses. Our results showed that the favorable alleles at chr14_66328304 (SPAG1) and chr5_94055934 (SLC15A5) were associated with increased protein stability, whereas the favorable alleles at BovineHD0500025863 (PLEKHA5) and BovineHD4100004526 (HERC6) exhibited reduced stability. Notably, AlphaFold-based structural predictions indicated that BovineHD0500025863 induced a substantial conformational change (RMSD = 20.849), whereas the other three SNPs had minimal effects (RMSD < 1). We further validated the association of three intergenic SNPs, including BovineHD0500026460 (Chr5:92772499), BovineHD1400000698 (Chr14:2390445), and BovineHD1400001112 (Chr14:3768615) with MFP in a Chinese Holstein population. These SNPs were polymorphic in the validation cohort and remained significantly associated with MFP, underscoring their robustness across Holstein populations. Their identification as markers for MFP highlights their potential utility in genomic selection programs aimed at improving milk fat content. While the exact functional roles of these intergenic regions remain to be elucidated, their consistent association with MFP suggests that they may contain regulatory elements influencing genes involved in lipid metabolism. Further functional studies are warranted to clarify the mechanisms by which these SNPs affect milk fat synthesis and to assess their applicability in breeding strategies for enhanced milk quality. Integrative analyses combining in silico TF binding prediction and dual-luciferase reporter assays demonstrated that BovineHD0500026460, BovineHD1400000698, and BovineHD1400001112 exert allele-specific effects on gene expression. Computational prediction indicated that these SNPs alter the binding affinity of MEF2A, MTF1, and SP1, respectively, potentially modulating transcriptional regulation. The favorable alleles (C at BovineHD0500026460 and C at BovineHD1400001112) enhanced transcriptional activity by creating or stabilizing TF binding sites, while the G allele at BovineHD1400000698 suppressed TF binding, yet all three variants elevated milk fat percentage, suggesting divergent regulatory mechanisms converging on lipid accumulation pathways. These results highlight the importance of integrating computational and experimental approaches to identify functional regulatory variants and provide mechanistic insights into the genetic control of milk fat traits. SP1 has been previously implicated in milk fat metabolism; for example, Zhu et al. reported that SP1 upregulation in goat mammary epithelial cells increased PPARγ expression and decreased LXRα levels, thereby modulating the transcription of fatty acid metabolism–related genes and promoting lipid droplet formation. In our study, among SP1 target and interacting genes (CREBBP, EP300, TP53, RELA, FASN, ACACA, and DGAT1), RELA was significantly downregulated in samples carrying the homozygous preferred allele (CC genotype), whereas FASN and ACACA were significantly upregulated. These genes play critical roles in de novo fatty acid synthesis [36]. Furthermore, among genes neighboring BovineHD1400001112, PTP4A3 and TRAPPC9 were markedly upregulated in samples with the homozygous preferred allele, suggesting that these genes may influence milk fat synthesis indirectly through intracellular signaling or vesicular transport in mammary epithelial cells [37, 38]. Our findings indicate that BovineHD1400001112 functions as a regulatory variant modulating milk fat synthesis by affecting SP1 and its downstream targets. The favorable alleles identified through GWAS were successfully replicated in an independent validation cohort, thereby robustly validating these genetic signals. Notably, using GEBVs computed under the U.S. national genetic evaluation system, we identified 104 significant loci; in an independent re-sequenced cohort of 99 Chinese Holsteins with direct MFP phenotype, 33 of these loci were polymorphic, yet only three remained significant. This suggests that U.S.-trained effect estimates may not fully translate to Chinese Holsteins due to differences in genetic background and production environments. Mechanistically, limited transferability is expected due to (i) population-specific allele frequencies and linkage disequilibrium/phase that alter marker–QTL tagging; (ii) genotype-by-environment interactions under divergent heat–humidity and management conditions; and (iii) differences in array ascertainment, imputation reference panels, and GEBV scaling—further compounded by limited power in the validation cohort (n = 99).Together, these findings argue for a China-specific genetic evaluation framework and reference population to improve portability of marker effects and maximize selection response—complementing, rather than replacing, international evaluation practices (e.g., MACE) with domestic recalibration and training. Conclusions In summary, our study integrates GWAS, protein structural analysis, functional validation, and gene expression profiling to identify genetic variants influencing milk fat percentage in Chinese Holstein cattle. We confirmed that BovineHD1400001112 functions as a regulatory variant that modulates milk fat synthesis by affecting SP1 binding and the expression of key lipid metabolism genes such as FASN and ACACA. Additionally, missense SNPs in SPAG1, SLC15A5, PLEKHA5, and HERC6 were shown to influence protein stability, though their direct roles in milk fat regulation remain to be elucidated. Our findings provide novel insights into the genetic and regulatory architecture of milk fat traits and offer valuable molecular markers and potential targets for genomic selection aimed at improving dairy production efficiency. Supplementary Information Supplementary Material 1. Acknowledgements The authors would like to thank all those who assisted us during this project for essential guidance and technical support. Abbreviations Bos taurus autosome Council on Dairy Cattle Breeding of America Cyclic Adenosine Monophosphate False Discovery Rate Fragments Per Kilobase of Transcript Per Million Mapped Reads Genomic Estimated Breeding Value Genome-wide Association Study Integrated Stress Response Kyoto Encyclopedia of Genes and Genomes Linear Mixed Model Minor Allele Frequency Milk Fat Percentage Next-Generation Sequencing Principal Component Analysis Quantitative Trait Locus Root Mean Square Deviation Single Nucleotide Polymorphism Transcription Factor Transcription Factor Binding Site Authors’ contributions JW and JH conceptualized the study. CG, YZ, and CY carried out the methodology and sequencing. Data analysis was performed by ZJ, YX, QJ, WL, XGW, XCW, and YG. Writing of the manuscript was done by CG and JW, and review & editing of the manuscript was carried out by JW, LW, YG and JH. JH was involved in supervision of the research. JW, YZ, YX and JH acquired funding. Funding This research was funded by the National Natural Science Foundation of China (32202648), the Key Research and Development Program of Shandong (2021LZGC011 and 2025LZGC006), the Natural Science Foundation of Shandong (ZR2024MC139), the National Key Research and Development Program of China (2021YFF1000703). Data availability The genotype datasets generated during the current study are available in the European Variation Archive (EVA) repository with the accession number PRJEB90543. The estimated genomic breeding values for MFP are provided in the Supplementary Information. Declarations Ethics approval and consent to participate Institutional Review Board Statement: This research has been reviewed and approved by the Institutional Animal Care and Use Committee of Shandong Academy of Agricultural Sciences (Approval number: SAAS-2022-G06). The experiment was conducted following the regulations and guidelines established by this committee. Informed consent was obtained from all animal owners to include all animals in this study. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Changliang Gong, Yaran Zhang and Chunhong Yang contributed equally to this work. Contributor Information Jinpeng Wang, Email: wangjinpeng0225@163.com. Jinming Huang, Email: huangjinm@sina.com. References Associated Data Supplementary Materials Supplementary Material 1. Data Availability Statement The genotype datasets generated during the current study are available in the European Variation Archive (EVA) repository with the accession number PRJEB90543. The estimated genomic breeding values for MFP are provided in the Supplementary Information.