Identification of Candidate Variants Associated with Milk Production, Health and Reproductive Traits for Holstein Cows in Southern China

Abstract

Genome-wide association studies (GWAS) have been a successful tool for identifying quantitative trait loci (QTL) for economically important traits in dairy cows. However, the availability of QTLs linked to phenotypic traits is limited in the literature. In this study, we used GWAS, haplotype association, and fine-mapping analyses to identify candidate variants associated with milk production, health, and reproductive traits in 380 Chinese Holstein cattle from Southern China using whole-genome sequence data. GWAS identified 91 genome-wide significant signals that were annotated to 63 genes associated with milk production, health, and reproductive traits in dairy cattle. Haplotype association analysis further revealed that eight GWAS signals within three QTLs were associated with milk production and health traits of cows. Fine-mapping analysis revealed that 3 GWAS signals (6_92530313_G_A, 10_17185230_G_A, and 10_17209112_T_G) were the potential causal variants. Several candidate genes, including ANKS1B, IL17RD, CNOT6L, AOC1, and TLE3, have been confirmed to be associated with milk production, health, and reproductive traits in dairy cows. These findings significantly contribute to unraveling the genetic basis of economically important traits in Holstein cattle.

1. Introduction

Chinese Holstein cattle are the foremost dairy breed propelling high-quality advancements in the dairy industry in China. They have become the predominant breed within the country’s dairy cattle population, comprising over 85% of the national herd [1]. According to our knowledge, Chinese Holstein cows are predominantly found in the northern regions of China, with comparatively fewer numbers present in the southern areas. High humidity and high temperature are the most prominent climatic features in southern China. High temperature and humidity are known to be the main cause of heat stress [2]. Heat stress in dairy cows is a serious issue that can negatively impact milk production [3], health [4], and reproduction [5]. For example, Rejeb et al. (2016) observed a decrease of 5.6 kg per day in cows during the spring and summer seasons [6]. Becker et al. (2020) emphasize that heat stress heightens the risk of mastitis in dairy cows by suppressing their immune response [7]. According to Bernabucci et al. (2010), conception rates typically decrease by approximately 24% during the summer months [8]. Obviously, the high humidity and elevated temperatures in southern China are not favorable for breeding Chinese Holstein cows. In this regard, there is a need to breed dairy cattle populations for improved heat tolerance. To address this issue, a deeper understanding of the genetic architecture of these traits related to these issues is crucial. This is because it can not only comprehensively improve the production performance of dairy cattle but also optimize breeding strategies, accelerating the genetic progress of these traits at the population level.

Genome-wide association studies (GWASs) have identified thousands of genetic variants linked to complex traits in cattle [9,10,11] and innovative traits, such as heat tolerance [12], methane emissions [13], and feed efficiency [14]. This approach has facilitated the understanding of the genetic mechanisms underlying complex traits and accelerated their genetic improvement [15]. To date, this approach has also been widely applied in the field of genetic analysis of economically important traits in livestock and poultry, with numerous published GWASs on dairy cattle [16,17,18], buffalo [19,20,21], horses [22,23,24], sheep [25,26,27], goats [28,29,30], pigs [31,32,33], and chickens [34,35], which has promoted the progress of genetic research in these species. Nonetheless, the genetic variants identified in existing GWASs have poor reproducibility and lack validation, and the fact that these loci are located in non-coding regions leads to small genetic effects, thereby affecting the further application of these loci in genetic breeding programs.

Numerous genetic variants identified by GWAS are not causal and are simply in linkage disequilibrium (LD) with the true causal variants. In general, genetic variants can be divided into three main groups [36]: the first group is the causal variants, which are responsible for the association signal and have a biological effect on the phenotype; the second group is the variants that are statistically associated with the phenotype due to LD with a causal variant; and the third group is the variants that are not statistically associated with the phenotype and are not causal. Undoubtedly, it is crucial to further precisely identify the causal loci within the GWAS signals. To address this issue, a method known as “fine-mapping” has been developed, which seeks to prioritize potential causal variants for subsequent functional studies. Currently, fine-mapping is a prevalent post-GWAS analysis technique in which the posterior probabilities of causality (PPC) are assigned to the candidate variants and genes. Importantly, the application of this approach to complex traits for single-variant resolution has begun in human studies [37,38]. Meanwhile, a large number of fine-mapping analysis software [39,40,41] have emerged one after another, accelerating the application of this method in causal loci. For example, FINEMAP implements GWAS summary statistics to explore a set of the most important causal configurations of regions using a shotgun stochastic search algorithm [42]. However, these fine-mapping methods are not easily applicable to large-scale GWAS and fine-mapping studies in cattle populations, because the data availability and high LD levels in livestock populations have restricted the utility of fine-mapping in GWAS loci. Fortunately, Jiang et al. developed a fast Bayesian fine-mapping method that efficiently integrates functional annotations with fine mapping [43]. Importantly, functional enrichment estimated by this approach enriches causal effects, whereas traditional stratified LD score regression estimates the enrichment of heritability [44].

In this study, therefore, we performed the GWAS analysis to identify candidate variants associated with milk production, health, and reproductive traits in Chinese Holstein cattle from Southern China. Subsequently, we integrate the haplotype association and fine-mapping approaches to identify the key candidate genes within GWAS signals. Our research enhances the understanding of the genetic framework underlying economically important traits in dairy cattle.

2. Materials and Methods

2.1. Animals and Phenotypes

A total of 380 Chinese Holstein cattle with 1038 lactation records were obtained from two farms (Hubei Jinxu Agriculture Development Co., Ltd. and Wuhan Guangmin Ecological Demonstration Dairy Cow Farm Co., Ltd., Wuhan, China) in Hubei Province, Southern China. Geographical coordinates of the two farms ranged from 29°01′53″ N to 33°6′47″ N and 108°21′42″ E to 116°07′50″ E. Blood samples were collected from the jugular vein using sterilized vacutainer tubes coated with EDTA as an anticoagulant. Phenotypic records of all dairy cows were collected between 2015 and 2023. Phenotypic records included five milk production traits: 305-day milk yield (MY305), fat percentage (FP), protein percentage (PP), peak milk yield (PM), and days to peak milk yield (PD); one reproductive trait: calving interval (CI), and one health trait: somatic cell score (SCS). SCS = log base 2 (SCC/100,000) + 3, where SCC is somatic cells per milliliter. PD refers to the lactation day when milk production is highest in lactation stage determination, with units in days. The pedigree data comprised 568 animals across three generations.

2.2. Deregressed Estimated Breeding Values

The deregressed estimated breeding value (dEBV) for each cow was calculated using a single-trait animal model implemented in the “blupADC (v1.0.7)” R package [45]. First, a linear model was used to evaluate all available components for significant impacts, and only significant factors were retained for inclusion in the single-trait animal models. The fixed effects of the milk production and health traits included herd–year–season (HYS; 55 classes) and parity (4 levels). The fixed effects of the reproductive traits included parity (4 levels), calving year (CY; 8 levels), and calving season (CS). The calving season was divided into four levels (spring: 3–5, summer: 6–8, autumn: 9–11, and winter: 12–2). In addition, individual additive genetics and permanent environment were included in the model as random effects. The analyzed model of the phenotypic traits was as follows:

$$y=Xb+Z_{1}a+Z_{2}p+e$$

(1)

where y is the phenotypic record of milk production, health or reproductive traits; X is a matrix associated with fixed effects; b is a fixed vector including HYS, CY, CS, and/or parity; $Z_1$ is a matrix associated with the individual additive genetic effects; a is an additive genetic effect $\left[a~\left(0,A{\sigma }_a^2\right)\right]$ and A is the pedigree-based relationships; $Z_2$ is a matrix related to the animals’ permanent environmental effects of individual animal $\left[p~\left(0,I{\sigma }_p^2\right)\right]$; and e is the residual error $\left[e~\left(0,I{\sigma }_e^2\right)\right]$ and I is the identity matrices. The ${\sigma }_a^2$, ${\sigma }_p^2$, and ${\sigma }_e^2$ represent the additive genetic, permanent environmental, and residual error variances, respectively. Heritability calculations for each trait were performed using the following formula: $h^2={\displaystyle \frac{{\sigma }_a^2}{{\sigma }_a^2+{\sigma }_p^2+{\sigma }_e^2}}$. The reliability of EBV for each trait was calculated using the following formula: $EBV reliabilities=1-{\displaystyle \frac{prediction error variance}{additive genetic variance of the trait}}$. EBV reliabilities were used to calculate dEBVs using the method described by Garrick et al. [46]. The dEBVs were used as pseudo-phenotypes for genome-wide association analysis.

2.3. Genotyping and Quality Control

Genomic DNA was isolated from whole blood samples using the TIANamp Blood Genomic DNA Kit (TIANGEN, Beijing, China) following the manufacturer’s instructions. All cows were genotyped using whole-genome sequencing at the 10× depth on the Illumina HiSeq 2500 platform. SNP calling was performed using the Genome Analysis Toolkit v4.2 (GATK) [47] with the default parameters. The Hereford cattle genome (assembly version: ARS-UCD2.0) was used as the reference genome. Next, SNP imputation was performed using Beagle ver5.4 [48] software, with a multi-breed cattle reference panel with 61.8 million SNPs as the reference [49]. The Beagle R-squared value was more than 0.8. Subsequently, all imputed SNPs for quality control (QC) were performed using PLINK v1.9 [50] with the following criteria: (1) individuals with a call rate ≥ 97%; (2) SNP call rate ≥ 95%; (3) minor allele frequency (MAF) ≥ 0.05; (4) Hardy–Weinberg equilibrium (HWE) p-value ≥ 10⁻⁶; and (5) SNP with unknown Bos taurus chromosomes (BTA) were excluded. After QC, 380 dairy cows and 1,030,098 SNPs successfully passed QC and were used for further analyses.

2.4. Estimation of Population Structure and Linkage Disequilibrium

To explore the population structure of the Chinese Holstein cattle used in this study, principal component analysis (PCA) and LD analyses were performed. PLINK v1.9 [50] was used to conduct PCA in the studied populations. PopLDdecay ver3.43 [51] was used to investigate the LD decay distance across the cattle genomes.

2.5. GWAS Analysis

The GWAS for each trait was performed using the fixed and random model circulating probability unification (FarmCPU) method implemented in the rMVP package [52]. The kinship matrix (K) was included in the model to control for confounding effects due to the shared ancestry. In addition to the kinship matrix, the first four principal components were included in the model to minimize the potential for spurious associations due to population stratification. The fixed-effects model is as follows:

$$y=X{b}_x+M_t{b}_t+s_j{d}_j+e$$

(2)

where y is the individual’s dEBVs; X is the matrix of fixed effect for the four PCs; $M_t$ is the genotyped matrix of t pseudo quantitative traits nucleotides (QTNs), initiated as an empty set; $b_x$ and $b_t$ are the corresponding effects of X and $M_t$, respectively; $s_j$ is the genotype of the j marker; $d_j$ is the corresponding effect; e is the residual effect $\left[e~\left(0,I{\sigma }_e^2\right)\right]$. The random-effects model is as follows:

$$y=u+e$$

(3)

where y and e stay the same as in the fixed effect model; u is the genetic effect of the individual and u~N (0, K${\sigma }_u^2$), in which K is the kinship matrix derived from the pseudo-QTNs. The Bonferroni correction was used to control for false positives resulting from multiple comparisons. The significance threshold was defined as 0.05/N, where N is the number of SNPs tested (n = 1,030,098). The genomic inflation factor λ was calculated by using the GWAS results.

2.6. Functional Annotation for GWAS Signals

SNP position information from the WGS data was based on Bos taurus UCD version 2.0. The ARS-UCD2.0 bovine reference genome was downloaded from the NCBI Genome website (https://www.ncbi.nlm.nih.gov/datasets/genome/ (accessed on 1 June 2025)). Considering that 50 Kb is the decay distance of LD analysis, a window size of 50 Kb was used to define the boundary range upstream and downstream of the GWAS signals. Candidate genes associated with GWAS signals were extracted from the genome gff3 files using bedtools ver.2.31.0 software [53]. KOBAS [54] was used for Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses of candidate genes identified by GWAS.

2.7. Haplotype Association Analysis

This study employed haplotype association analysis to identify QTLs associated with specific phenotypic traits. These GWAS signals were chosen from a 50 Kb window around potential genomic hotspots and were used to construct haplotype blocks. The geneHapR [55] package was used for haplotype identification, statistics, phenotype association, and result visualization analysis. Genomic regions containing haplotype blocks related to phenotypic traits (p < 0.05) were identified as candidate QTLs.

2.8. Fine-Mapping Analysis

A fine-mapping procedure using a Bayesian approach was used to identify potential candidate genes and their causal loci. This method was performed by using the BFMAP v.1 software [43], which included three steps: forward selection to add independent signals in the additive Bayesian model, repositioning signals, and generating credible variant sets for each signal. The causality of SNP loci was assessed by computing the p-value of causality for independent association signals within candidate QTL regions and the posterior probability of causality (PPC) for each variant. More information about the BFMAP algorithm and its procedure can be found on the GitHub website (https://github.com/jiang18/bfmap (accessed on 1 June 2025)). The p-value < 0.0001 and PPC > 0.5 were defined as the threshold for selecting the causal loci.

3. Results

3.1. Descriptive Statistics

In this study,

Table 1. Statistical description of study traits in Chinese Holstein cattle population.

Traits	Mean	SD	PV	AV	PEV	RV	h2	SE (h2)	Reliability
Milk production traits
MY305 (kg)	9754.66	2733.97	4,393,615.80	1,522,274	3.00 × 10−5	2,871,341.80	0.35	0.16	0.49
FP (%)	4.31	0.55	0.28	0.15	0	0.14	0.52	0.19	0.51
PP (%)	3.33	0.24	0.06	0.02	0.004	0.03	0.39	0.16	0.42
PM (kg)	45.02	14.33	95.51	18.71	3.66	73.14	0.20	0.12	0.36
PD (days)	88.38	55.90	2629.12	324.73	0	2304.45	0.12	0.07	0.30
Health trait
SCS	2.92	0.01	0.0001	1.00 × 10−5	0	9.00 × 10−5	0.14	0.09	0.26
Reproductive trait
CI (days)	413.86	99.43	9937.97	736.11	0	9201.85	0.07	0.06	0.20

Note. SD = Standard deviation; PV = Phenotypic Variance, AV = Additive Variance, PEV = Permanent Variance, RV = Residual Variance; h² = heritability of traits; SE = Standard error for heritability; MY305 = 305-day milk yield; FP = Fat percentage; PP = Protein percentage; PM = peak milk yield; PD = days to peak milk yield; SCS = Somatic cell score; CI = Calving interval.

presents the descriptive statistical results for the milk production, health, and reproductive traits in Chinese Holstein cattle. Overall, the phenotypic values of each trait were normally distributed (Figure S1). The cows in the present study had a higher 305-day milk yield (9754.66 ± 2733.97 kg) than that reported for German Holstein cattle (9690 ± 1580.00 kg) [56] and Xinjiang Brown cattle (4126.49 ± 1405.71 kg) [57]. The fat and protein percentages in the present study were 4.31% and 3.33%, respectively, which are similar to those of Brazilian Holstein [58], New Zealand cows [59], and Thai dairy cattle [60] breeds. The peak milk yield during the total lactation period was 45.02 ± 14.33 kg. The mean PD was 88.38 days, with a standard deviation of 55.90 days. Heritabilities were moderate for most milk production traits, except for FP, ranging from 0.12 to 0.52, which is similar to the findings of other studies on dairy and beef cattle [61,62,63].

SCS is known to be an indicator for predicting udder health, and a smaller SCS value indicates a lower risk of mastitis [64]. In our population, the mean SCS value was 2.92, with a standard deviation of 0.01 and heritability of 0.14.

Because most reproductive traits are difficult to measure and vary across environmental conditions, only CI was selected in this study. The average value for this trait was 413.86 days, with a standard deviation of 99.43 days, indicating moderate variability. A low heritability (0.07) for CI was found, which is consistent with the findings of other studies on dairy cattle [65,66,67].

3.2. Population Stratification and LD Decay Analysis

Principal component analysis was used to explore the population structure of the dairy cattle population. The results showed that our population could be divided into four groups. The plot of the eigenvalues demonstrated that the first two PCs explained 28.3% of the total variability (Figure 1A). Furthermore, the square of the correlation coefficient between markers at the two loci (r²) was used to estimate the LD. As expected, the r² value in LD tended to decrease with increasing marker distance between pairwise SNPs, with a rapidly declining trend observed over the first 50 Kb. In the Holstein cattle population under investigation, LD decayed to an r² value of 0.2 at approximately 50 Kb in the studied population (Figure 1B).

3.3. GWAS for Milk Production Traits

The GWAS of milk production traits revealed 71 significant SNPs distributed across 20 chromosomes (

Table 2. GWAS signals for milk production traits in the Chinese Holstein cattle.

Traits	SNP	CHR	Position (bp)	p-Value	FDR	Nearest Gene	Distance (Kb)
MY305	1_72919372_T_G	1	72919372	1.54 × 10−10	3.97 × 10−5	TMEM44	within
MY305	5_12384753_C_G	5	12384753	7.67 × 10−14	7.89 × 10−8	TMTC2	within
MY305	5_102787852_T_A	5	102787852	3.32 × 10−8	4.08 × 10−3	CD163L1	within
MY305	7_136097_G_A	7	136097	4.72 × 10−10	9.73 × 10−5	LOC101907627	16.19
MY305	7_3235359_G_C	7	3235359	3.57 × 10−8	4.08 × 10−3	LOC104969038	within
MY305	13_25815175_G_A	13	25815175	2.62 × 10−9	4.50 × 10−4	GPR158	within
MY305	17_63091807_C_T	17	63091807	3.33 × 10−11	1.71 × 10−5	SPPL3	within
MY305	18_43016754_A_G	18	43016754	1.32 × 10−10	3.97 × 10−5	DPY19L3	23.5
MY305	21_20072203_C_G	21	20072203	3.17 × 10−8	4.08 × 10−3	LOC132343335	2.99
FP	3_15686591_C_T	3	15686591	3.71 × 10−11	9.97 × 10−6	PMVK	within
FP	5_9957392_A_C	5	9957392	3.87 × 10−11	9.97 × 10−6	OTOGL	29.17
FP	5_63598729_G_A	5	63598729	2.63 × 10−15	2.71 × 10−9	ANKS1B	within
FP	5_63599285_G_T	5	63599285	1.11 × 10−12	5.71 × 10−7	ANKS1B	within
FP	9_49700710_C_T	9	49700710	3.38 × 10−9	3.87 × 10−4	SIM1	141.51
FP	18_36490909_C_T	18	36490909	2.33 × 10−9	3.12 × 10−4	NIP7	within
FP	21_33857755_C_G	21	33857755	4.24 × 10−9	4.37 × 10−4	LMAN1L	within
FP	21_33914609_G_T	21	33914609	1.30 × 10−8	1.22 × 10−3	CYP1A2	8.50
FP	21_33936823_G_T	21	33936823	2.33 × 10−9	3.12 × 10−4	CYP1A2	6.75
FP	29_49287451_G_A	29	49287451	9.09 × 10−10	1.87 × 10−4	TSPAN32	5.21
FP	29_50418489_T_C	29	50418489	2.43 × 10−9	3.12 × 10−4	AP2A2	within
PP	1_72709215_G_A	1	72709215	3.45 × 10−8	4.18 × 10−3	LOC132346096	93.63
PP	6_55498889_T_C	6	55498889	4.26 × 10−11	1.46 × 10−5	ARAP2	within
PP	6_84194728_C_A	6	84194728	4.14 × 10−11	1.46 × 10−5	LOC100138004	within
PP	10_41071426_T_C	10	41071426	2.38 × 10−10	6.12 × 10−4	MDGA2	323.43
PP	16_36517381_G_A	16	36517381	1.19 × 10−9	2.45 × 10−4	DPT	183.56
PP	17_71197319_G_A	17	71197319	1.97 × 10−8	2.89 × 10−3	ZNF70	14.25
PP	24_690958_G_A	24	690958	1.59 × 10−9	2.73 × 10−4	CTDP1	within
PP	X_49143795_A_C	X	49143795	3.65 × 10−8	4.18 × 10−3	PCDH19	154.91
PP	X_54208034_T_G	X	54208034	6.80 × 10−12	7.01 × 10−6	IL1RAPL2	within
PM	3_15509105_C_A	3	15509105	1.54 × 10−8	5.27 × 10−4	EFNA3	3.28
PM	3_15583935_G_T	3	15583935	3.50 × 10−11	2.77 ×10−6	ZBTB7B	within
PM	3_15686591_C_T	3	15686591	1.43 × 10−28	1.47 × 10−22	PMVK	within
PM	3_15826776_C_T	3	15826776	4.93 × 10−24	2.54 × 10−18	KCNN3	within
PM	3_25219743_G_C	3	25219743	9.88 × 10−9	3.63 × 10−4	SPAG17	within
PM	5_56801630_G_A	5	56801630	1.73 × 10−12	1.98 × 10−7	BAZ2A	within
PM	6_57627037_C_T	6	57627037	3.56 × 10−9	1.89 × 10−4	TBC1D1	198.39
PM	6_113000819_G_A	6	113000819	4.54 × 10−9	2.11 × 10−4	LOC107132586	within
PM	6_113005937_G_A	6	113005937	6.73 × 10−9	2.57 × 10−4	LOC107132586	within
PM	6_113008264_C_T	6	113008264	1.55 × 10−16	3.98 × 10−11	LOC107132586	within
PM	6_113031755_A_G	6	113031755	3.03 × 10−10	1.95 × 10−5	LOC107132586	2.36
PM	6_113032815_G_A	6	113032815	1.97 × 10−11	1.84 × 10−6	LOC107132586	3.42
PM	6_113032851_G_A	6	113032851	3.87 × 10−9	1.89 × 10−4	LOC107132586	3.46
PM	6_113032867_C_A	6	113032867	2.27 × 10−11	1.95 × 10−6	LOC107132586	3.47
PM	6_113033403_C_T	6	113033403	4.92 × 10−9	2.11 × 10−4	LOC107132586	4.01
PM	6_113033404_T_C	6	113033404	4.92 × 10−9	2.11 × 10−4	LOC107132586	4.08
PM	11_87592639_G_A	11	87592639	5.51 × 10−15	1.13 × 10−9	GRHL1	within
PM	21_56854166_T_G	21	56854166	3.87 × 10−9	1.89 × 10−4	CPSF2	8.50
PM	21_56858110_T_A	21	56858110	1.24 × 10−14	1.83 × 10−9	CPSF2	12.50
PM	21_56860640_G_T	21	56860640	7.60 × 10−11	5.59 × 10−6	CPSF2	15.03
PM	21_56871715_G_A	21	56871715	2.18 × 10−10	1.49 × 10−5	CPSF2	26.10
PM	22_43984285_G_T	22	43984285	5.60 × 10−10	3.39 × 10−5	IL17RD	within
PM	22_43988976_G_A	22	43988976	4.92 × 10−13	6.34 × 10−8	IL17RD	within
PM	22_43989144_G_A	22	43989144	5.39 × 10−9	2.14 × 10−4	IL17RD	within
PM	22_43990890_G_A	22	43990890	5.39 × 10−9	2.14 × 10−4	IL17RD	within
PM	23_5521491_G_T	23	5521491	2.23 × 10−8	7.19 × 10−4	FAM83B	within
PM	25_6321969_A_G	25	6321969	1.07 × 10−8	3.79 × 10−4	RBFOX1	within
PM	26_397054_G_A	26	397054	2.12 × 10−8	7.06 × 10−4	OR5D18	4.73
PM	26_556311_G_T	26	556311	2.05 × 10−19	7.04 × 10−14	UBE2D1	110.44
PM	26_638379_C_T	26	638379	9.98 × 10−15	1.71 × 10−9	UBE2D1	28.37
PM	26_803096_C_T	26	803096	3.58 × 10−9	1.89 × 10−4	IPMK	within
PM	29_45659217_C_T	29	45659217	2.11 × 10−12	2.18 × 10−7	KMT5B	6.47
PD	6_60140889_C_T	6	60140889	4.38 × 10−8	0.004	UCHL1	6.69
PD	6_60141280_T_C	6	60141280	3.87 × 10−8	0.004	UCHL1	6.30
PD	6_92530313_G_A	6	92530313	1.56 × 10−8	0.004	CNOT6L	2.43
PD	6_92534492_G_A	6	92534492	9.82 × 10−9	0.004	CNOT6L	6.61
PD	6_93430541_C_T	6	93430541	3.34 × 10−8	0.004	ANXA3	57.16
PD	6_93782255_G_A	6	93782255	4.22 × 10−8	0.004	PAQR3	104.48
PD	6_99689795_T_C	6	99689795	3.26 × 10−8	0.004	WDFY3	within
PD	6_99724549_T_C	6	99724549	2.96 × 10−8	0.004	WDFY3	within
PD	6_99766367_G_A	6	99766367	2.49 × 10−8	0.004	WDFY3	within
PD	6_99766578_T_G	6	99766578	1.99 × 10−8	0.004	WDFY3	within
PD	22_12760734_A_G	22	12760734	3.41 × 10−8	0.004	LOC101903734	within

SNP = single nucleotide polymorphism; CHR = chromosome; MY305 = 305-day milk yield, FP = Fat percentage, PP = Protein percentage, PM= Peak milk yield, PD = day to peak milk yield, FDR = False discovery rate.

; Figure S2). Among them, 9, 12, 8, 32, and 11 GWAS signals were detected for MY305, FP, PP, PM, and PD traits, respectively. These GWAS signals were mainly located on BTA6, BTA21, and BTA22. In total, 49 candidate genes were annotated with these signals based on the bovine reference genome (version: ARS-UCD2.0) gff3. In total, 36 significant SNPs were identified in 27 candidate genes. GO enrichment analysis revealed that the majority of candidate genes were enriched in the integral component of the membrane, nucleoplasm, and plasma membranes (Figure S3). KEGG enrichment analysis showed that two genes, LMAN1L and UBE2D1, were enriched in the protein processing in the endoplasmic reticulum pathway.

To further detect QTLs associated with milk production traits, we used a 50 Kb window around the potential genomic hotspot to determine the LD relationships (Figure 2). Two identified SNPs (5_63598729_G_A and 5_63599285_G_T) for FP traits were located on BTA5 at 63 Mb in the ANKS1B gene, and within the detected window from 63.579 to 63.623 Mb, three blocks (Table S1) were recognized (Figure 2A). We noted that the two identified SNPs (5_63598729_G_A and 5_63599285_G_T) were located on FP_QTL3 block, which harbors 18 SNPs, but no significant haplotypes were observed (Figure 2B). For PM traits, we identified five blocks (Table S2). Importantly, four identified SNPs, including 22_43984285_G_T, 22_43988976_G_A, 22_43989144_G_A, and 22_43990890_G_A, were located on the PM_QTL4 block in the IL17RD gene. This block spanned 14.102 Kb and involved eight SNP loci (Figure 2C). Interestingly, the haplotype association analysis revealed that the H001 (TTGGGTGA) genotype in PM_QTL4 was significantly (p < 0.001) associated with PM traits in cows (Figure 2D), indicating that this haplotype was the dominant genotype affecting PM in dairy cows. Furthermore, four blocks (Table S3) were found for the PD traits, and two identified SNPs (6_92530313_G_A and 6_92534492_G_A) were harbored in PD_QTL2 block (Figure 2E). This region, spanning 4.18 Kb, consisted of only two SNPs and was significantly (p < 0.05) associated with PD traits (Figure 2F). Notably, cows with the H001 (GG) haplotype had a lower EBV value of PD than those with the H002 (AA) haplotype, indicating that the H001 haplotype is the dominant genotype for selecting PD in dairy cows. These results suggest that the six SNP loci, including 22_43984285_G_T, 22_43988976_G_A, 22_43989144_G_A, 22_43990890_G_A, 6_92530313_G_A and 6_92534492_G_A, can be candidate markers associated with milk production traits in cattle.

3.4. GWAS for Health Traits

A single-trait GWAS was conducted for SCS traits, which led to the identification of 10 significant SNPs across eight chromosomes (

Table 3. GWAS signals for health trait in the Chinese Holstein cattle.

Traits	SNP	CHR	Position (bp)	p-Value	FDR	Nearest Gene	Distance (Kb)
SCS	4_113485958_C_T	4	113485958	9.85 × 10−10	2.54 × 10−4	AOC1	22.70
SCS	4_113489752_G_A	4	113489752	1.28 × 10−8	1.32 × 10−3	AOC1	26.50
SCS	4_113489803_G_A	4	113489803	8.06 × 10−9	9.22 × 10−4	AOC1	26.55
SCS	6_94098782_G_T	6	94098782	2.26 × 10−9	3.87 × 10−4	GK2	16.99
SCS	8_15818198_G_T	8	15818198	1.92 × 10−9	3.87 × 10−4	LINGO2	within
SCS	9_56683722_C_T	9	56683722	5.68 × 10−9	7.31 × 10−4	EPHA7	67.57
SCS	15_19068193_A_G	15	19068193	3.26 × 10−12	3.36 × 10−6	C15H11orf87	261.28
SCS	17_16535314_G_A	17	16535314	4.58 × 10−9	6.74 × 10−4	ZNF330	within
SCS	21_69672126_G_A	21	69672126	2.54 × 10−10	8.74 × 10−5	PACS2	within
SCS	29_50103110_C_A	29	50103110	2.88 × 10−11	1.49 × 10−5	BRSK2	1.95

SNP = single nucleotide polymorphism; CHR = chromosome; SCS= Somatic cell score, FDR = False discovery rate.

; Figure S4). Eight candidate genes were annotated for these significant SNPs, including AOC1, GK2, LINGO2, EPHA7, C15H11orf87, ZNF330, PACS2, and BRSK2. Furthermore, the majority of identified SNPs were found to be clustered into BTA4, and we observed that 13 blocks (Table S4) were found in the detected window from 113.465 to 113.511 Mb (Figure 3A). Two identified SNPs (4_113489752_G_A and 4_113489803_G_A) were located in the SCS_QTL5 block. This region, spanning 3.075 Kb, consisted of 20 SNPs and was significantly (p < 0.05) associated with SCS (Figure 3B). Notably, cows with H002 (AGGGCGGGATGGAACGTTGC) in SCS_QTL5 exhibited a significantly (p < 0.05) higher SCS EBV than those with other haplotypes, indicating that this haplotype may have influenced mastitis in cows. These results suggest that both 4_113489752_G_A and 4_113489803_G_A are candidate markers that affect the health traits in cattle.

3.5. GWAS for Reproductive Traits

The GWAS of CI revealed 10 significant SNPs distributed across seven chromosomes (

Table 4. GWAS signals for reproductive trait in the Chinese Holstein cattle.

Traits	SNP	CHR	Position (bp)	p-Value	FDR	Nearest Gene	Distance (Kb)
CI	3_15494215_C_A	3	15494215	3.64 × 10−8	3.75 ×10−3	EFNA3	18.17
CI	7_1588817_A_C	7	1588817	1.36 × 10−8	1.75 × 10−3	MAML1	within
CI	10_17178126_G_A	10	17178126	4.94 × 10−11	8.47 × 10−6	LOC132346336	98.85
CI	10_17181283_C_T	10	17181283	4.94 × 10−11	8.47 × 10−6	LOC132346336	95.69
CI	10_17185002_G_A	10	17185002	4.94 × 10−11	8.47 × 10−6	LOC132346336	91.97
CI	10_17185230_G_A	10	17185230	1.19 × 10−12	6.14 × 10−7	LOC132346336	91.74
CI	13_16401282_T_C	13	16401282	2.24 × 10−8	2.56 × 10−3	SFMBT2	within
CI	14_9758134_C_T	14	9758134	4.09 × 10−14	4.21 × 10−8	ADCY8	within
CI	18_10045332_G_A	18	10045332	2.89 × 10−9	4.25 × 10−4	CDH13	within
CI	X_77899540_G_A	X	77899540	2.04 × 10−11	7.01 × 10−6	PHKA1	21.95

SNP = single nucleotide polymorphism; CHR = chromosome; CI = Calving interval, FDR = False discovery rate.

; Figure S5). Seven candidate genes were annotated: EFNA3, MAML1, TLE3, SFMBT2, ADCY8, CDH13, and PHKA1. We observed that the majority of identified SNPs were clustered into BTA10 and that five blocks (Table S5) were found in the detected window from 17.158 to 17.197 Mb (Figure 4A). Notably, the four identified SNPs were located in two distinct blocks (Figure 4A). Two identified SNPs (10_17178126_G_A and 10_17181283_C_T) were harbored in the CI_QTL4 block, which spanned 6.408 Kb and involved 33 SNPs. The other two identified SNPs (10_17185002_G_A and 10_17185230_G_A) were located in the CI_QTL5 block, spanning 13.319 Kb and covering 35 SNPs. The haplotype association analysis revealed that cows with H001 and H002 haplotypes had a higher CI EBV than those with H003; however, no significant difference (p > 0.05) was observed (Figure 4B).

3.6. Fine-Mapping Analyses

To detect the causal loci associated with phenotypic traits, loci from topSNP and its surrounding 1 Mb were selected and used for fine-mapping analysis. The PPC values for individual variants and genes were calculated and used to identify potential causal loci (

Table 5. List of potential causal loci associated with milk production, health, and reproductive traits in Chinese Holstein cattle.

Traits	Nearest Gene	Distance (Kb)	Type	SNP	CHR	POS	GWAS p-Value	Casualty p-Value	PPC
PD	CNOT6L	2.43	Protein coding	6_92530313_G_A	6	92530313	1.56 × 10−8	1.33 × 10−8	0.885
PD	CNOT6L	6.61	Protein coding	6_92534492_G_A	6	92534492	9.82 × 10−9	1.08 × 10−7	0.115
SCS	KCNH2	36.167	Protein coding	4_113490019_G_A	4	113490019	1.28 × 10−5	5.15 × 10−6	0.059
SCS	KCNH2	35.336	Protein coding	4_113490850_A_G	4	113490850	1.28 × 10−5	5.15 × 10−6	0.059
SCS	KCNH2	33.921	Protein coding	4_113492265_T_C	4	113492265	1.28 × 10−5	5.15 × 10−6	0.059
SCS	SLC25A13	within	Protein coding	4_113485958_C_T	4	113485958	9.85 × 10−10	1.25 × 10−5	0.023
CI	LOC132346336	91.74	lncRNA	a 10_17185230_G_A	10	17185230	1.19 × 10−12	2.34 × 10−12	0.979
CI	LOC132346336	67.86	lncRNA	10_17209112_T_G	10	17209112	1.36 × 10−5	7.36 × 10−5	0.963

Note: ^a represents TopSNP; SNP = single nucleotide polymorphism; CHR = chromosome; POS = position; PPC = Posterior probability of causality.

). The results showed that three SNPs, including 6_92530313_G_A (PPC = 0.885), 10_17185230_G_A (PPC = 0.979), and 10_17209112_T_G (PPC = 0.963), were successfully fine-mapped for PD and CI traits, respectively. Among them, the SNP 10_17185230_G_A was the topSNP and had the highest PPC value. The regional association plot with the LD heatmap for CNOT6L and LOC132346336 genes is shown in Figure S6A,B, respectively. These results suggest that three SNPs were causal loci affecting PD or CI traits.

4. Discussion

In the present study, we performed a powerful GWAS analysis of milk production, health, and reproductive traits in Chinese Holstein cattle. Both haplotype association and Bayesian fine-mapping approaches were used to further examine the GWAS signals to identify potential candidate genes and causal loci. Finally, we present a list of candidate genes or loci involved in cattle health and production that can be readily explored in future functional-validation studies.

4.1. Milk Production Traits

In our GWAS, 71 SNPs were associated with the milk production traits. A total of 49 genes were annotated in these signals. Several genes, including ANKS1B [68], TSPAN32 [69], IL1RAPL2 [70], and TBC1D1 [71], have been associated with milk production traits in dairy cows. In addition, the SIM1 gene is known to be associated with human obesity and obesity-related traits [72], suggesting that SIM1 may be a candidate gene affecting the FP trait. The ARAP2 gene has been associated with fatty acid composition in sheep [73].

We also focused on several QTLs related to FP, PM, and PD, which were mapped to BTA5, BTA22, and BTA6, respectively. For the FP trait, we observed one QTL consisting of two identified SNPs located within the ANKS1B gene, spanning 44.60 Kb. Evidence has demonstrated that ANKS1B is a candidate gene affecting milk composition traits [68], emphasizing the presence of six QTLs that play a role in the milk C14 index trait. Importantly, our result also supports ANKS1B as a candidate gene affecting FP.

We identified one noteworthy QTL on BTA22 that was associated with PM in this study. This QTL contained four identified SNPs (22_43984285_G_T, 22_43988976_G_A, 22_43989144_G_A, and 22_43990890_G_A) and spanned 14.102 Kb, involving 8 SNP loci. Haplotype association analysis also supported the association of these identified SNPs with PM traits. Notably, this QTL was mapped to the IL17RD gene, which encodes a component of the interleukin-17 receptor signaling complex. The IL17RD is a single-pass transmembrane protein that has been implicated in the regulation of various signaling pathways, receptor tyrosine kinase signaling cascades, and proinflammatory signaling pathways including Interleukin-17A (IL17A), Toll-like receptors (TLR) and Interleukin-1α (IL1α) in vertebrate species [74]. Several studies have demonstrated that IL17A is associated with milk production traits and somatic cell scores in dairy cows [75,76]. Investigation revealed that IL-17A, known for its essential role in mucosal immunity against a variety of extracellular and some intracellular pathogens, also plays a part in protecting the mammary gland, which is a non-mucosal tissue with a secretory epithelium [77,78]. The results outlined in the preceding hypothesis indicate that IL17RD is crucial in the generation of peak milk yield.

Furthermore, we found that one QTL region consisting of two identified SNPs (6_92530313_G_A and 6_92534492_G_A) was associated with PD in dairy cows. More importantly, haplotype association analysis also supported this point, while fine-mapping analysis confirmed that the SNP 6_92530313_G_A could be defined as causal loci affecting PD traits. For this QTL, we observed that CNOT6L was a neighboring gene. Evidence has shown that CNOT6L plays a vital role in the positive regulation of cell proliferation [79]. It is well known that days to peak milk yield (PD) refers to the day when the peak milk yield occurs after calving [80]. Mammary epithelial cells are responsible for milk production in the bovine mammary glands. In other words, the arrival of the lactation peak day indicated a significant increase in the number of mammary epithelial cells. In this regard, we speculate that CNOT6L may be involved in the regulation of mammary epithelial cell proliferation in dairy cows, thereby affecting milk performance in dairy cows. Further functional validation is required to confirm this hypothesis. In summary, these results suggest that CNOT6L could be a candidate gene affecting PD traits in dairy cattle.

4.2. Health Trait

By annotating gene functions, eight candidate genes were annotated within the GWAS signals, including AOC1, GK2, LINGO2, EPHA7, C15H11orf87, ZNF330, PACS2, and BRSK2. Among these, the GK2 gene was reported to be associated with bovine tuberculosis susceptibility [81], and the EPHA7 gene was associated with the calf sucking reflex trait [82]. Moreover, we observed a genomic region on BTA4 consisting of three significant GWAS signals (4_113485958_C_T, 4_113489752_G_A, and 4_113489803_G_A). Haplotype association analysis further demonstrated that two identified SNPs (4_113489752_G_A and 4_113489803_G_A) were harbored in the SCS_QTL5 block and were significantly (p < 0.05) associated with SCS. The AOC1 is the neighboring gene in the QTL region. Evidence has shown that AOC1 is a copper-containing amine oxidase that plays a crucial role in immune response [83]. It is well known that the immune response is involved in the entire process of mastitis onset [84]. Therefore, identifying genes related to the immune response will help in understanding mastitis pathogenesis. Therefore, we inferred that AOC1 may play an important role in mastitis development. The detailed functions of this gene require further experimental verification.

4.3. Reproductive Trait

We identified one noteworthy candidate gene, LOC132346336, which is associated with CI traits on BTA10. This region spanned 13.319 Kb and harbored 35 SNPs in total. Two identified SNPs (10_17185002_G_A and 10_17185230_G_A) were located in this region of the genome. Notably, TopSNP 10_17185230_G_A and neighboring SNP 10_17209112_T_G were confirmed as a causal locus affecting CI traits using fine-mapping analysis. However, the potential role of this causal locus still needs further functional validation. Furthermore, LOC132346336 is an lncRNA gene, and its function remains unknown at this time. Therefore, the potential link between LOC132346336 and the calving interval still requires further verification. Nonetheless, our findings indicate that this gene is a candidate gene influencing the calving interval.

Overall, this study also has limitations, primarily due to a relatively small sample size. The reason for this is that there are only three large-scale Chinese Holstein dairy farms (herds > 100) in Hubei Province, one of which is a newly built farm, and the group’s production performance testing has only just been incorporated into the Dairy Herds Improvement project, resulting in fewer valid phenotypic data. Next, we will continue to increase the size of the study population to improve the statistical power of the population GWAS study. Nevertheless, our results are reliable because we strictly followed the GWAS analysis pipeline, and the analysis results also corroborate this. Furthermore, we believe that the validation of results for fine-mapping and haplotype analysis remains a priority, as this is necessary for the further use of these candidate variants in breeding practices.

5. Conclusions

In summary, we performed a GWAS analysis of milk production, health, and reproductive traits in dairy cows and identified some candidate variants related to these phenotypic traits. Using haplotype association analysis, we observed that some GWAS signals were affected by strong linkage disequilibrium and identified several dominant haplotypes affecting milk production and health traits of the cows. Furthermore, we identified several GWAS signals belonging to the causal variants. Our findings significantly contribute to unraveling the genetic basis of milk production, health, and reproductive traits in Holstein cattle.