Investigating Genetic Characteristics of Chinese Holstein Cow’s Milk Somatic Cell Score by Genetic Parameter Estimation and Genome-Wide Association

Abstract

The quality and safety of milk is challenged by cow mastitis, and the value of somatic cell score (SCS) in milk is closely related to the occurrence of mastitis. This study aimed to analyze the genetic characteristics of SCS across the first three parities in Chinese Holstein cattle, as well as to investigate potential candidate genes and biological processes that may play a potential role in the progress of cow mastitis. In this respect, we evaluated genetic parameters and conducted a genome-wide association study based on the test-day records of SCS for Chinese Holstein cows; we also validated key candidate genes using a quantitative reverse transcription PCR (RT-qPCR) experiment in primary bovine mammary epithelial cells (bMECs). The heritability of the SCS 305-day performance in milk varied between 0.07 and 0.24, and decreased with increasing parity. As the time interval grew larger, the genetic and permanent environmental correlations with the number of days in milk (DIM) weakened. Six significant single-nucleotide polymorphisms (SNPs) were identified in the association analysis, one of which was located within the exonic region of CD44. This exon-associated SNP may modify the activity of the protein encoded by the CD44. A total of 32 genes within the two hundred kilobase (kb) range of significant SNPs were detected, and these genes were markedly enriched in eight Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways and 22 biological processes, mainly participating in the progress of transmembrane transport, inflammatory factor regulation, cellular responses, the Toll-like receptor signaling pathway, and the MAPK signaling pathway. Nine genes, including the PKD2, KCNAB1, SLC35A4, SPP1, IBSP, CD14, CD44, MAPK10, and ABCG2 genes, were selected as candidate genes that could have critical functions in cow mastitis. These findings can serve as a foundation for molecular breeding and as valuable data for reducing the incidence of mastitis of Chinese Holstein cattle at the molecular level.

1. Introduction

The quality and abundance of milk produced have significant impacts on the profitability of a dairy farm. Bovine mastitis is one of the most common inflammatory illnesses around the world, with a significant financial impact on milk production and quality issues with dairy products. According to reports, mastitis occurs in 25–60% of bovines worldwide [1,2], and the overall positive rate of subclinical mastitis in dairy herds of China reached 37.7% during the period of 2012–2021, which was typically higher than other countries [3]. In addition, other researchers have found that poor lactation yield and high levels of milk somatic cells are the primary markers reported for voluntary culling in dairy cows [4], and incidences of mastitis tremendously decrease net milk farm revenue and milk’s value for industry. Therefore, research on the prevention and treatment of mastitis is extremely meaningful for China.

However, due to the low heritability of mastitis, indirect selection using the value of milk somatic cells as a biomarker is more effective in reducing the incidence rate of cow mastitis than traditional phenotype-based selection, because the response indicators of milk somatic cells, such as somatic cell count (SCC) and SCS, have considerably higher heritability than mastitis [5,6]. In addition, clinical mastitis and milk somatic cells are known to have a substantial hereditary association; this helps to explain why selection programs to develop mastitis resistance have historically incorporated test-day milk somatic cells response indicator—SCS. Utilizing data gathered regularly during milk-recording methods is also a tactic to enhance genetic selection for disease resistance [7]. Studies have also demonstrated that the expression levels of genes involved in mastitis susceptibility may be different during lactation and between parities [8]. Strucken found that some locus alterations during lactation in German Holstein–Friesian cows accounted for the variation [9]. First-parity cows’ mammary glands are still growing; hence, the mastitis-susceptibility gene expression in this parity can differ from subsequent lactations, which may lead to changes in genetic variance of SCS. The random regression test-day model, which is widely used in dairy cow breeding, has been recognized as an effective method for genetic evaluation of traits based on test day records [10,11], and it was used in this study to estimate the genetic parameters of SCS.

Deeper comprehension of the genetic parameters and biological mechanisms behind SCS at different lactation stages is crucial to promote more efficient genetic breeding of dairy cows, and a genome-wide association study (GWAS) is a powerful method to encourage the marker-assisted selection of mastitis-related phenotypes based on SNPs [12]. In addition, bMECs are an effective model to study the mechanism of bacterial infection in mammalian glands [13,14], and the addition of lipopolysaccharides (LPSs) to bMECs can well simulate the invasion of Gram-negative bacteria since LPSs are often regarded as a primary inducer of mastitis [15]. In this study, we estimate the magnitude and trajectory of genetic parameters for SCS with DIM at each parity in Chinese Holstein cows and perform a GWAS to identify candidate genes associated with SCS. Meanwhile, LPS-challenged bMECs were used to further identify the expression of these GWAS-detected genes, which was also a way to verify the reliability of GWAS. We anticipate that the results may contribute to understanding the genetic architecture and underlying biology of mastitis and provide theoretical guidance on genetic selection for susceptibility to mastitis in Chinese dairy cows.

2. Materials and Methods

2.1. Ethical Statement

The guidelines of the Institutional Animal Care and Use Committee of the School of Yangzhou University Animal Experiments Ethics Committee (License Number: SYXK (Su) IACUC 2020-0910) were used as the implementation standard in the sample collection and data preparation processes of this study, and no animals were subjected to psychological fear or physical abuse.

2.2. Animal, Phenotype, and Genotype Data

In the present study, a total of 149,065 test-day records of 15,216 Holstein cows from four herds in the Jiangsu province of China were collected. These records spanned from 2010 to 2019 and were all generated at the Jiangsu Provincial Dairy Cow Performance Testing Center (Nanjing, China). To approximately follow a normal distribution, SCC values were converted to somatic cell score (SCS) values through the calculation of (log₂ (SCC/100,000) +3) [16]. The following quality control measures were conducted on these test-day records: (1) records were kept of cows ranging from parity 1 to parity 3; (2) records were kept for milk production ranging from 5 kg to 80 kg, for DIM stages between 5 and 305, and for SCS records ranging from 0 to 9; (3) records were removed if recorded fewer than 6 times in a parity; and (4) records were removed for cows with first-calving ages of less than 22 months or more than 36 months. Finally, 86,281 test-day records of 8580 cows, which were the offspring of 504 bulls, remained for genetic parameter estimation and phenotypic correction, and these individuals’ ancestries were tracked back to at least three generations. The criterion for an uninfected mammary gland is an SCC of less than 100,000 cells/mL (SCS < 2) according to the International Dairy Federation [17], therefore, the healthy cows and cows with mastitis were all included for data collection. The descriptive statistics of the SCS values are shown in

Table 1. Descriptive statistics and estimated heritability of SCS records of Holsteins of different parities.

Parity	N	RN	Mean	SE	Median	Min	Max	Skew	Kurtosis	Heritability (SE)
1	5127	37,179	2.57	0.01	2.26	0.67	9.00	0.75	0.76	0.24 (0.01)
2	3777	27,432	2.83	0.01	2.68	0.68	9.00	0.58	0.25	0.14 (0.03)
3	2982	21,670	2.88	0.02	2.68	0.54	9.00	0.61	0.21	0.07 (0.02)
1–3	8580	86,281	2.70	0.01	2.49	0.54	9.00	0.69	0.50	0.18 (0.01)

RN: number of records; N: number of animals.

.

Hair samples of 999 cows from the above 8580 cows were collected and genotyped using GGPBovine 100K SNP Chips (Neogen, Michigan, USA) [18]. SNPs were retained if the minor allele frequency (MAF), the call rate, and the p-value of the Hardy–Weinberg equilibrium (HWE) were higher than 5%, 90%, and 1.0 × 10⁻⁶, respectively. Additionally, individuals with genotype call rates of less than 95% were excluded from the analysis. After quality control, 984 individuals and 87,598 SNPs of 30 chromosomes (29 autosomes and 1 X chromosome) remained for GWAS.

2.3. Genetic Parameter Estimation

The variance components of SCS in parity 1, parity 2, and parity 3, as well as parities 1-3, were estimated using a random regression test-day model. The model is described by the following formula:

$$\begin{array}{c}{y}_{ijklmn}=HT{D}_i+P{A}_j+FC{A}_k+{\displaystyle \sum _{n=0}^6}{b}_{ln}{L}_n\left({\omega }_t\right)+{\displaystyle \sum _{n=0}^2}\left(a_{mn}{L}_n\left({\omega }_t\right)+p_{mn}{L}_n\left({\omega }_t\right)\right)+e_{ijklmn},\end{array}$$

(1)

where $y_{ijklmn}$ is the value of the SCS, ranging from 0 to 9, and $HT{D}_i$, $P{A}_j$, and $FC{A}_k$ are three fixed effects in the model, representing the ith herd test-day, the jth parity, and the kth level of the first-calving age, respectively. First-calving age was categorized as one of the following four levels: level 1 included first-calving ages ≤ 23 months; level 2 included the range of 24 ≤ age ≤ 27; level 3 included the range of 28 ≤ age ≤ 31; and level 4 included first-calving ages ≥ 32 [19].$b_{ln}$ is the lth fixed regression coefficient corresponding to the nth Legendre polynomial;$a_{mn}$ and $p_{mn}$ are the additive genetic effect and the permanent environmental effect of the $m$th cow at the nth random regression, respectively, and they are considered as the random regression effects in the model; $L_n\left({\omega }_t\right)$ is the $n$th covariate of the Legendre polynomial at day $t$ in milk (DIM_t); ${\omega }_t$ is the standardized time at DIM_t; and $e_{ijklmn}$ is the random residual assumed to be homogeneous throughout lactation. To ensure the accuracy of the estimate, the model had to meet the convergence requirements reported by Madsen [20]. After converting to the three-column format required for DMU operation, a total of 21,855 pedigree records were used to construct the kinship matrices of the individuals. The variance components, as well as the genetic parameters of SCS, were estimated using the following equations:

$${\sigma }_{at}^2={L^{\prime }}_t\widehat{G}{L}_t,$$

(2)

$${\sigma }_{pet}^2={L^{\prime }}_t\widehat{P}{L}_t,$$

(3)

$${\sigma }_{a\left(t1,t2\right)}^2={L^{\prime }}_{t1}\widehat{G}{L}_{t2},$$

(4)

$$r_{a\left(t1,t2\right)}=\frac{{\sigma }_{a\left(t1,t2\right)}^2}{\sqrt{{\sigma }_{a\left(t1\right)}^2*{\sigma }_{a\left(t2\right)}^2}},$$

(5)

$$h_t^2=\frac{{\sigma }_{at}^2}{\sqrt{{\sigma }_{at}^2+{\sigma }_{pet}^2+{\sigma }_{et}^2}},$$

(6)

$$h_T^2={\displaystyle \sum }_{t1=5}^{305}{\displaystyle \sum }_{t2=5}^{305}{\sigma }_{a\left(t1,t2\right)}^2/({\displaystyle \sum }_{t1=5}^{305}{\displaystyle \sum }_{t2=5}^{305}({\sigma }_{a\left(t1,t2\right)}^2+{\sigma }_{pe\left(t1,t2\right)}^2)+{\displaystyle \sum }_5^{305}{\sigma }_e^2,$$

(7)

where ${\sigma }_{at}^2$ and ${\sigma }_{pet}^2$ are the genetic variance and the permanent environmental variance at DIMt, respectively; ${\sigma }_{a\left(t1,t2\right)}^2$ is the genetic covariance between DIMt₁ and DIMt₂; $L_t$ is a vector of the Legendre polynomial at DIMt;$r_{a\left(t1,t2\right)}$ is the genetic correlation between DIMt₁ and DIMt₂; $\widehat{G}$ and $\widehat{P}$ are the corresponding (co)variance matrices for the random regression coefficients of the genetic and permanent environmental effects, respectively$; h_t^2$ is the heritability at DIMt of SCS; $h_T^2$ is the heritability of the 305-day performance of SCS; and${\sigma }_e^2$ is the residual variance.

After removing the fixed effects, the EBV of each animal was adjusted to a 305-day performance, and then the deregressed proof (DRP) was calculated as the response variable for the association study to remove the parent average effects. The DRP was obtained according to the method provided by Garrick [21], and the distributions and the correlations of the DRPs are shown in Figure S1. The 305-day performance of EBV was adjusted as follows:

$$y_{305d}={\displaystyle \sum }_{t=5}^{305}{\displaystyle \sum }_{n=0}^3{a}_{mn}{L}_n\left({\omega }_t\right),$$

(8)

where $y_{305d}$ is the 305-day performance of SCS, and $a_{mn}$ stays the same as in Formula (1).

2.4. Analysis of Principal Components

A principal component analysis was carried out using the Plink command-line program (v1.90) [22] to illustrate the structure of our study population, and the twstats programme in EIGENSTRAT software was used to identify the statistically significant principal components [20]. At the same time, a parallel coordinates analysis, a cluster analysis, and an identity-by-state (IBS) analysis were performed with the SNPRelate package in R language (v 4.2.1) [23].

2.5. Genome-Wide Association Analysis

A GWAS was employed using the fixed and random model circulating probability unification (FarmCPU) approach [24]. The FarmCPU process contained two models: one was a fixed effects model, and the other was a random effects model. The fixed effects model identified SNPs above the threshold as pseudo-quantitative-trait nucleotides (QTNs), which were then substantiated by the random effects model. Before the number of QTNs in the fixed effects model could remain constant, the fixed effects and random effects models were iteratively conducted [25]. The first 5 highest principal components (PCs) were handled as covariates in the fixed effects model to reduce the effect of population genetic background. The fixed effects model was as follows:

$$y_{DRP}=X{b}_X+M_i{b}_i+S_j{d}_j+e,$$

(9)

where $y_{DRP}$ is the vector of DRP; $b_X$ is the effect value for the principal components, with $X$ as the corresponding design matrix; $b_t$is the fixed effect of the $i$th pseudo-QTN, which is determined and enhanced with each iteration; $M_i$ is the matrix generated from the genotype; $S_j$ is the effect value of the $j$th SNP’s genotype; $d_j$ is the corresponding effect value of the $j$th SNP; and $e$ is the residual effect value. The random effects model was as follows:

$$y_{DRP}=u+e,$$

(10)

where $y_{DRP}$ and $e$stay the same as in Formula (9); and $u$ represents the vector generated from the total genetic effect of population. The assumption is that $u$ satisfies a normal distribution with a mean of 0 and a variance of K${\sigma }_u^2$, where K is the kinship matrix built by the QTNs generated from the fixed effects model in each cycle, and ${\sigma }_u^2$ is the unidentified genetic variation [26]. The explained phenotypic variations (EPV) of the SNPs were measured using the following equation:

$$EP{V}_n=\frac{2*MA{F}_n*\left(1-MA{F}_n\right)*{\beta }_n^2}{{\sigma }_{y_{DRP}}^2}$$

(11)

where $EP{V}_n$ is the phenotypic variation explained by the $n$th SNP; $MA{F}_n$ is the minimum allele frequency of the $n$th SNP; ${\beta }_n$ is regression coefficient for the association analysis of the $n$th SNP; and ${\sigma }_{y_{DRP}}^2$ is the variance in the DRP.

2.6. Function Annotation Analysis of Candidate Genes

Genes were recognized as potential factors associated with SCS if they were within a 200 kb range of significant SNPs [25,27]. A gene ontology (GO) assessment and a Kyoto Encyclopedia of Genes and Genomes (KEGG) evaluation were carried out using the clusterProfiler package and the GOplot package in R (v 4.2.1) [28] to further comprehend the biological roles of these genes.

2.7. Cell Culture and Lipopolysaccharide (LPS) Treatment

To understand the changes in expression of GWAS-detected candidate genes during the bacterial infection in mammalian glands, and to further evaluate the reliability of the GWAS results, we conducted a PCR experiment on the normal bMECs and LPS-challeged bMECs. Culturing and LPS treatment of bMECs were carried out as described in previous research [29]. Briefly, bMECs were separated and refined from the mammary gland tissue of one healthy, mid-lactating Holstein cow (~175 DIM) with a daily milk yield of 30.26 ± 3.1 kg and no clinical diseases occurring before sampling. This cow did not belong to the GWAS study population. Cells from the 4th to 6th passages were used to conduct the experiment. Cells were evenly dispersed in 12-well plates (approximately 1 × 10⁵ per well) and cultured overnight in complete medium (90% DMEM/F-12, 31330095, Gibco, CA; 10% heat-inactivated fetal bovine serum, Gibco, CA). Finally, cells were cultured and maintained in a humidified 5% CO₂ incubator with a temperature of 37 °C. The LPS (L2630, Sigma, St. Louis, MO, USA) used in the study was obtained from Escherichia coli (E. coli) with serotype O111:B4 (https://www.sigmaaldrich.cn/CN/zh/product/sigma/l2630, accessed on 17 January 2023), and the LPS inflammatory model was validated and improved according to our previous study [30]. In brief, complete medium without antibiotics was re-added to the wells, followed by three washes with PBS after 24 h of incubation. Then, 6 wells were challenged with 2 μg/mL LPS, and another 6 wells were mixed with the same concentration of PBS as the control group (negative control group, NC) [30].

2.8. Extraction of RNA and Quantitative Real-Time PCR Analysis

An RNA isolator (R401-01, Vazyme, Nanjing, China) was used to obtain total RNA following the company’s protocol, and the quality of the isolated RNA was guaranteed (1.8 < OD260/OD280 < 2.0) by spectrophotometry (Nanodrop ND-1000, Thermo Fisher, Waltham, MA, USA). Then, the RNA was reverse-transcribed into cDNA with a PrimeScript™ RT reagent Kit (RR037Q, Takara, Kusatsu, Japan) and purified using a purification kit (Axygen, Tewksbury, MA, USA). qPCR was conducted with a Bio-Rad CFX96 real-time PCR detection system (Bio-Rad, Hercules, CA, USA) with a One Step PrimeScript™ RT-PCR Kit (RR064A, Takara, Japan) according to the company’s protocol. The GAPDH, RPS9, and ACTB genes, which have shown high stability for normalizing gene expression in mammary samples and bMECs in previous reports [31,32], were selected as internal control genes to calibrate the expression levels of the target genes. All the primers (Table S3) were synthesized with Premier 6.0 software (Premier Biosoft International, Palo Alto, CA, USA), and the 2^−△△Ct approach was used to estimate the relative quantifications of the target genes (with average 98.7% amplification efficiency in this study). The procedures were repeated three times, with three replicates in each experiment [33].

2.9. Statistical Analysis

In the association analysis, the Bonferroni correction method was applied to decrease the rate of false positives during hypothesis testing. The threshold for significant SNP detection was 5.70 × 10⁻⁷ (0.05/87598) [34]. The gene expression data were presented as means ± standard errors, and Student’s t-test was performed in R (v 4.2.1) to compare the differences between the LPS group and the NC group. We considered statistical differences significant between groups if the p-values were less than 0.05 in the qRT-PCR analysis.

3. Results

3.1. Genetic Parameter Evaluation

The heritability of the 305-day SCS values for the parities varied between 0.07 and 0.24. It was highest in the first parity, decreasing in a downward trend to the third parity with corresponding values of 0.24, 0.14, and 0.07, respectively (Table 1). Figure 1 displays the variance components and heritability of SCS at various DIM stages in the three parities.

The curves of the phenotypic and genetic parameters of SCS within lactations of different parities, particularly the curves of additive genetic variance, permanent environment ratios, and heritability, as functions of DIM within each parity, are shown in Figure 1A–C, respectively. It was found that the additive genetic variances (${\sigma }_a^2$) for the first parity were higher than those in the second parity. In addition, the additive genetic variances (${\sigma }_a^2$) through the first three parities all showed a downward trend during the early DIM stage (5–100 days), starting high (0.2), then declining to reach the lowest value (<0.15) during the mid-early lactation period (5–100 days), and then increasing gradually through the mid-lactation period (100–200 days) until the end of the DIM at 305 days for parity 1. The values of ${\sigma }_a^2$ decreased (<0.05) until the end of the lactation period (200–305 days) for the second and third parities (Figure 1A). The variations in the permanent environmental variance (${\sigma }_{pe}^2$) of the first parity, second parity, and third parity during various DIM stages were similar. All of these parities displayed declining tendencies, which peaked in the early DIM stage (5–100 days of lactation), steadily increased during mid-lactation (100–200 days), and then decreased gradually to the lowest value at the end of the DIM at 200–305 days (Figure 1B). The value of ${\sigma }_{pe}^2$ for parity 3 was higher (>1.0) than those of parity 1 and parity 2 (>0.5), which was roughly reversed from the trajectories of additive genetic variance and heritability between days 5 and 305 of the lactation period across the first and third parities. The heritability of SCS during different DIM stages across parity 1, parity 2, and parity 3 had a similar trend to that of additive genetic variance. The heritability of SCS started high (<0.06) at the beginning of the DIM (5–100 days), decreased until the middle of the early lactation period, and increased again through mid-lactation (100–200 days) until the end of the DIM (200–305 days) for parity 1. However, for parity 2, it decreased gradually through mid-lactation (100–200 days) to the end of the late DIM period (200–305 days). The heritability of parity 3 was lower compared to the heritability of parity 1; it started high (0.02), decreased gradually through the early DIM stage (5–100 days), increased during mid-lactation (100–200 days), and decreased to attain its minimum value at the end of the late lactation period (200–305 days, Figure 1C).

3.2. Genetic and Permanent Environmental Correlations

The genetic correlations of SCS with DIM in the parities are represented in Figure 2. As the DIM interval extended, the genetic correlations of SCS decreased for the first, second, and third parities, with the lowest genetic associations occurring between the start of the DIM period (5–100 days) and the completion of late lactation (200–305 days; Figure 2). Within each lactation stage, as opposed to different lactation stages, the genetic correlations of SCS with DIM for parity 1 and parities 1–3 were stronger (Figure 2A,D). However, for parity 2 and parity 3, the genetic correlations all presented at a relatively high level in all of the DIM stages between the mid-lactation period (100–200 days) and the late lactation period (200–305 days), as shown in Figure 2B, C.

The permanent environmental correlations of the SCC with different DIM stages during lactation are displayed in Figure S2. The permanent environmental correlation between any two DIM stages during lactation was above 0.5 for parities 1 and 1–3 (Figure S2A,D). However, for parity 2 and parity 3, the permanent environmental correlations in the early lactation period and late lactation period were relatively low (<0.3, Figure S2B,C).

3.3. Population Structure Analysis

The individuals in this study were divided into several groups of varying size, as seen in Figure 3. The population stratification level was calculated using the top three principal components (PCs). The three greatest PCs, which accounted for 11.8%, 9.2%, and 7.3% of the variance, respectively, explained 28.3% of the variation when combined (Figure 3A–C), but only the first two principal components played significant roles (p < 0.05) in population stratification (

Table 2. Statistical information for the first eight principal components of population variation.

PC	Variance (%)	Twstat	p-Value
1	11.822	3.613	0.001
2	9.811	2.831	0.002
3	7.303	0.552	0.088
4	6.393	0.699	0.073
5	5.511	0.587	0.084
6	4.553	−0.366	0.244
7	4.302	0.800	0.064
8	3.503	−0.168	0.200

PC: principal component; variance: proportion of population variation; Twstst: F-value of Tracy–Widom distribution.

). After the fifth principal component, each of the following principal components could separate the population effectively, and the top five PCs were therefore included as covariates in the fixed effects model for the association analysis (GWAS) to prevent false positives caused by group stratification (Figure 3D).

We also used a population cluster analysis and an IBS analysis based on the genotype data for all the individuals. The results showed the cattle population’s genetic structure and estimated the ancestry proportions to assign individuals to clusters of four farms. The farms were shown to have a large number of clusters, although the clusters were very close to each other in the observed data (Figure 4A). The results of the IBS analysis also showed a relatively similar genetic background among the individuals in the farms (Figure 4B).

3.4. Genome-Wide Association Study for SCS

In this study, the cutoff for choosing significant SNPs in the GWAS analysis was 5.9 × 10⁻⁷ (0.05/84407), as previously indicated. The six SNPs of rs135806474, rs41256968, rs41566683, rs109267271, rs134115197, and rs109756462 were located on chromosomes 6, 15, 1, 7, 1, and 6, respectively. These SNPs were identified to be related to SCS. Among the genes closest to the SNPs were SPP1 (rs135806474, within intron), CD44 (rs41256968, within exon), EPHA3 (rs41566683, within intron), CD14 (rs109267271, 8146 bp upstream of CD14), TIPARP (rs134115197, 24,264 bp downstream of TIPARP), and MAPK10 (rs109756462, within intron), respectively (

Table 3. Details on the detected significant SNPs and their closest genes.

SNP	CHR	Position	Nearest Gene	Distance (kb)	MAF	EVP	p-Value
rs135806474	6	36725781	SPP1	within (intron)	0.251174	2.46%	1.26 × 10−9
rs41256968	15	65736591	CD44	within (exon)	0.224178	2.17%	1.50 × 10−8
rs41566683	1	37299219	EPHA3	within (intron)	0.113850	2.06%	1.15 × 10−7
rs109267271	7	51754749	CD14	−8146	0.409976	1.17%	1.22 × 10−7
rs134115197	1	110980155	TIPARP	+24,264	0.433099	1.25%	1.40 × 10−7
rs109756462	6	101037292	MAPK10	within (intron)	0.397887	1.26%	4.33 × 10−7

CHR: chromosome; MAF: minor allele frequency; EVP: explained phenotypic variation. Positive and negative values suggest that the SNPs reside downstream and upstream of the genes, respectively.

, Figure 5A). The descriptive statistics for these six candidate SNPs are shown in Table 3.

To guarantee the reliability of the association analysis between SCS and the variations, a quantile–quantile (QQ) plot was generated from the p-values of mutations. The great majority of the variants stayed close to the predicted p-value, proving the accuracy of the GWAS technique in this study (Figure 5B).

3.5. Annotation and Enrichment Analysis of Candidate Genes

In order to fully comprehend the molecular basis of these mutations, the KEGG and GO analyses were performed on the nearest genes and the genes within 200 kb of the significant SNPs In total, 32 such candidate genes were obtained in total (Table S1). Then, according to the principle of KEGG analysis, the probability and significance of the pathways that were enriched by candidate genes were calculated using a cumulative hypergeometric distribution model [28]. We found that these genes were significantly enriched in the regulation of transmembrane transport signaling, cytokine production, and cellular responses to iron pathways (

Table 4. KEGG pathway information of the nearest genes and the genes within 200 kb of the significant SNPs.

Pathway	Description	Gene Name	p-Value
bta04512	ECM–receptor interaction	SPP1, IBSP, CD44	0.0003
bta04620	Toll-like receptor signaling pathway	SPP1, CD14, MAPK10	0.0016
bta04510	Focal adhesion	SPP1, IBSP, MAPK10	0.0028
bta00970	Aminoacyl-tRNA biosynthesis	HARS1, HARS2	0.0050
bta05169	Epstein–Barr virus infection	CD44, MAPK10	0.0178
bta05417	Lipids and atherosclerosis	CD14, MAPK10	0.0387
bta04215	Multiple species of apoptosis	MAPK10	0.0408
bta04010	MAPK signaling pathway	CD14, MAPK10	0.0475

, p < 0.05), such as ECM–receptor interaction, the Toll-like receptor signaling pathway, focal adhesion, aminoacyl-tRNA biosynthesis, Epstein–Barr virus infection, lipids and atherosclerosis, multiple species of apoptosis, and the MAPK signaling pathway (Table 4).

In total, 162 GO terms were significantly enriched (Table S3). The genes that were only enriched in one GO term were eliminated, and eight genes enriched in 22 GO terms were maintained (Figure 6). The majority of these took part in transmembrane transport, tissue development, and inflammatory factor regulation, as well as cellular responses, such as potassium ion transmembrane transport, transmembrane transport, the regulation of transmembrane transport, the regulation of interferon gamma production, the positive regulation of tumor necrosis factor superfamily cytokine production, the regulation of potassium ion transport, protein hetero-oligomerization, the positive regulation of tumor necrosis factor production, the regulation of cation channel activity, biomineral tissue development, lamellipodium organization, tissue development, and cellular responses to calcium ions. Eight of the thirty-two genes were identified in the aforementioned 22 biological processes (Figure 6).

3.6. Expressions of Candidate Genes in LPS-Challenged bMECs

The relative expressions of candidate genes in the LPS-challenged bMECs was tested. PKD2, SLC35A4, SPP1, CD44, KCNAB1, and ABCG2 gene expressions were validated using the RT-qPCR technique. The results revealed significant differences between the LPS-treated group and the control group for the PKD2, SLC35A4, SPP1, CD44, KCNAB1, and ABCG2 genes, as their expression was increased significantly in response to LPS inducement (Figure 7).

4. Discussion

The quantity and quality of milk produced are important determinants of the profitability of dairy businesses. Studies have revealed that genes associated with SCS may be expressed differentially throughout lactation and across parities [9]. To investigate the trajectories of genetic parameters throughout the lactation period, this study used the test-day model (TDM), a model that can increase the accuracy of EBV estimates [11], to fit the test-day records of SCS. The trajectories of additive genetic effects and heritability with DIM in the first parity of Chinese Holsteins were similar to those of Canadian Holsteins [35], as shown in Figure 1C. They decreased in the early lactation period and attained the lowest levels throughout lactation; then, they showed a small rise and subsequent fall in the mid-lactation period and at the beginning of the late lactation period, respectively, and increased until the end of lactation. It was also found that, throughout the whole lactation period, the heritability of SCS was greater in the first parity than it was in the second and third parities (Figure 1C). The third parity had the smallest heritability and was more influenced by the environment throughout the lactation period (Figure 1B,C), which revealed that the effect of the automatic milking system on the mammary glands did not go away with the end of the lactation stage, but continued into the next parity [36]. The additive effect and heritability of SCS showed decreasing trends with parity (Figure 1A,C, and Table 1), similar to that estimated for SCS from a study in US dairy cattle [37]. Said study also gave a suggestion for the prevention and control of mastitis in dairy cows, that is, breeding could be considered for mastitis for dairy cows in the first parity, but for later parities more consideration should be given to the impact of the environment.

The genetic correlations of SCS and DIM for the first, second, and third parities, as well as parities 1–3, showed downward trends as the DIM interval grew longer (Figure 2), which is comparable to studies on milk-related traits in dairy cows [17,38]. This indicates that the genetic structures of milk-related traits and SCS are similar [39,40]. We also found that the genetic correlations of SCS and DIM within each lactation stage were greater than those in different lactation stages (Figure 2), which revealed that a few missing data records within a single lactation stage might be adequately compensated for by the records of adjacent DIMs within the same stage, thus not affecting the genetic evaluation of an individual [41]. However, the start of the DIM (5–100 days) and the termination of the late lactation period (200–305 days) had the weakest genetic connections and were sometimes even negative (Figure 2A,C). We speculated that it is unreliable to speculate on the genetic performances of dairy cows based only on the performance of SCS in a single lactation stage.

Population stratification is a crucial cause of confusion since it can result in GWAS false positives due to systematic ancestry differences [42]. In many published studies, the first two principal components generated from genotype data can explain at least 60% of variance [43], but this percentage is reduced in extended families with many closely related paternal siblings, and more false-positive SNPs might be generated in an association analysis [44]. This study showed that only the first two PCs had a significant effect on group stratification (Table 2), and the results of the cluster and IBS analyses also indicated relatively close kinship among the individuals, which might be the reason why only 21.0% (11.8% and 9.2%, respectively) of the variation with the top two PCs was explained (Figure 3). Since polygenic effects were considered in the random effects model with the FarmCPU method [24], the majority of the mutations remained consistent with the predicted p-value (Figure 5B), demonstrating that population stratification was well-managed [45].

In our study, genes detected within 200 kb of the significant SNPs were found to be associated with SCS. Among these, rs135806474 and rs109756462 were located on chromosome 6, and the nearest genes within the exonic genomic region were SPP1 and MAPK10, respectively. The rs41256968 SNP was located on chromosome 15, and the nearest gene within the genome region was CD44. Interestingly, through cell surface receptors, such as immunological ligand CD44, SPP1 is a phosphoglycoprotein-encoded gene with cell-binding capabilities that controls numerous functions [46]. SPP1 was also reported to be associated with mastitis and to the best of our knowledge, milk somatic cells had a positive correlation with mastitis [47], and was used as a common mark of resistance to mastitis [5]. The activity of OPN can be induced by a range of factors, such as lipopolysaccharides, tumor necrosis factor-alpha (TNF-a), interleukin-1 beta (IL-1 beta), interleukin-10 (IL-10), and interferon-gamma (IFN-γ) [48]. In addition, the rs41566683 and rs134115197 SNPs were located on chromosome 1, and CD14 was the nearest gene to the rs41566683 SNP within the exonic region of the genome (Table 3 and Table S1). This gene was associated with SCS in the current study, and it has also been reported that TLR4 and CD14 were among genes that regulated mastitis resistance [49].

In this study, KEGG and GO analyses were conducted on the nearest genes, as well as on genes within 200 kb of the SNPs. A total of 32 genes were retrieved to further understand the biological processes of the six candidate SNPs associated with SCS (Table S1). According to the KEGG analysis, these genes were significantly enriched in the regulation of transmembrane transport signaling, cytokine production, and cellular response to iron pathways (Table 4, p < 0.05), such as ECM–receptor interaction (SPP1, IBSP, and CD44), the Toll-like receptor signaling pathway (SPP1, CD14, and MAPK10), focal adhesion, aminoacyl-tRNA biosynthesis, Epstein–Barr virus infection, lipids and atherosclerosis, multiple species of apoptosis (MAPK10), and the MAPK signaling pathway (CD14 and MAPK10). The CD14 gene found in the present study was associated with SCS, and it has also been reported that resistance to mastitis is controlled by multiple genes, including TLR4, CD14, and CXCR2 [13]. We also identified 22 GO terms enriched by eight genes (Figure 6), and some of them are reported to play key roles in organism immune response. The Toll-like receptor signaling pathway is crucial for initiating innate immune responses and identifying pathogen-associated molecular patterns [50]. The Toll-like receptor signaling pathway is mainly represented by CD14 [51]. The NF-κB and MAPK signaling pathways regulate cytokine and chemokine expressions during inflammation [52]. In response to bacterial toxins, MAPKs play a key role in activating the expressions encoding a wide range of cytokines and chemokines. In addition, the MAPK pathway regulates important cellular processes, including cell proliferation, differentiation, migration, senescence, and apoptosis [53]. These SNPs and pathways could be used as potential candidates for Holstein breeding program selection.

The LPS of E. coli was used to challenge the bMECs for investigating the inflammatory responses of cow mastitis in the present study. LPS is the component of the outer membranes of highly vital Gram-negative bacteria, and one of the most common mastitis-causing pathogens is E. coli, which infects the mammary gland during parturition and early lactation of dairy cow [54,55]. The alveolar epithelial secretory cells in the mammary gland can not only synthesize milk, but also serve as the first line of defense against pathogen invasion [56]. Therefore, bMECs challenged with E. coli LPS have been considered an important model for cow mastitis research [57,58]. Meanwhile, the challenge of mammary cells with LPS stimulates their innate immunity response, including Toll-like receptor 4 (TLR4), NF-κB, and mitogen-activated protein kinase (MAPK) signaling [59,60]. The expression of genes associated with inflammation and inflammasomes are also activated, such as IL1, IL6, IL1B, TNF, CXCL8, and NLRP3. In our previous study, we used the expression of TNF, IL1B, IL6 and CXCL8 as the standard, and found that treatment of bMECs with E. coli LPS at a concentration of 2 μg/mL for 6 h could achieve the desired inflammatory response without affecting cell viability [30]. However, it may be necessary to identify the optimal challenge conditions for the inflammatory responses caused by other pathogenic bacteria.

The relative expressions of potential candidate genes were validated in LPS-challenged bMECs using the RT-qPCR technique. The results revealed a significant difference between the LPS-treated group and the control group. Interestingly, the expression of PKD2, SLC35A4, SPP1, CD44, KCNAB1, and ABCG2 was significantly upregulated in response to LPS (Figure 7). SPP1, IBSP, and CD44, in particular, were related to the Toll-like receptor signaling pathway cascade. Additionally, CD14 and MAPK10 were related to the MAPK signaling pathway and multiple species of apoptosis, indicating that the Toll-like receptor signaling pathway played a significant part in the development of inflammation by recognizing molecular patterns linked with pathogens and triggering innate immune responses. The MAPK pathway has also been involved in the development of inflammation in bovine-related studies [61,62], and so we believe these genes and pathways may perform important regulatory functions in the occurrence of dairy cow mastitis.

5. Conclusions

In the current research, we evaluated the genetic parameters of SCS across the first three parities and carried out a GWAS on these characteristics. Throughout 305 days of milking, the heritabilities, permanent environmental variances, and additive genetic variations of these parities all underwent continual change. The heritability of milk SCS between 5 and 305 DIM for the parities varied between 0.07 and 0.24, and the genetic correlations between various DIM periods for the first, second, and third parities exhibited declining tendencies as the DIM interval expanded. A total of six SNPs were shown to be closely related to SCS in the association analysis, and four of them were located in the exonic regions of their corresponding genes. A total of nine candidate genes, including PKD2, KCNAB1, SLC35A4, SPP1, IBSP, CD14, CD44, MAPK10, and ABCG2, were selected as key genes that could be involved in the occurrence and regulation of dairy cow mastitis. However, the regulation mechanism of SNPs on these genes’ expression and the specific functions of these genes in the occurrence of mastitis are worthy of further investigation. The present study is helpful in understanding the genetic architecture of SCS as an indicator for mastitis resistance across parities in dairy cows, thus providing theoretical guidance for dairy cow genome breeding in China.