Unravelling novel and pleiotropic genes for cannon bone circumference and bone mineral density in Yorkshire pigs

Abstract

Leg weakness is a prevalent health condition in pig farms. The augmentation of cannon bone circumference and bone mineral density can effectively improve limb strength in pigs and alleviate leg weakness. This study measured forelimb cannon bone circumference (fCBC) and rear limb cannon bone circumference (rCBC) using an inelastic tapeline and rear limb metatarsal area bone mineral density (raBMD) using a dual-energy X-ray absorptiometry bone density scanner. The samples of Yorkshire castrated boars were genotyped using a 50K single-nucleotide polymorphism (SNP) array. The SNP-chip data were imputed to the level of whole-genome sequencing data (iWGS). This study used iWGS data to perform genome-wide association studies and identified novel significant SNPs associated with fCBC on SSC6, SSC12, and SSC13, rCBC on SSC12 and SSC14, and raBMD on SSC7. Based on the high phenotypic and genetic correlations between CBC and raBMD, multi-trait meta-analysis was performed to identify pleiotropic SNPs. A significant potential pleiotropic quantitative trait locus (QTL) regulating both CBC and raBMD was identified on SSC15. Bayes fine mapping was used to establish the confidence intervals for these novel QTLs with the most refined confidence interval narrowed down to 56 kb (15.11 to 15.17 Mb on SSC12 for fCBC). Furthermore, the confidence interval for the potential pleiotropic QTL on SSC15 in the meta-analysis was narrowed down to 7.45 kb (137.55 to137.56 Mb on SSC15). Based on the biological functions of genes, the following genes were identified as novel regulatory candidates for different phenotypes: DDX42, MYSM1, FTSJ3, and MECOM for fCBC; SMURF2, and STC1 for rCBC; RGMA for raBMD. Additionally, RAMP1, which was determined to be located 23.68 kb upstream of the confidence interval of the QTL on SSC15 in the meta-analysis, was identified as a potential pleiotropic candidate gene regulating both CBC and raBMD. These findings offered valuable insights for identifying pathogenic genes and elucidating the genetic mechanisms underlying CBC and BMD.

This study aimed to identify novel and pleiotropic genes regulating cannon bone circumference (CBC) and metatarsal area bone mineral density (raBMD) using genome-wide association studies, and demonstrated that RAMP1 was the potential pleiotropic candidate gene regulating CBC and raBMD, providing valuable insights for elucidating the genetic mechanisms underlying limb growth and bone mineral accumulation.

Introduction

The high consumer demand for pork meat and the focus on improved production efficiency have increased the expectations of the growth performance and meat quality of pigs. However, the enhanced focus on productivity has contributed to the increased incidence of limb and hoof-related issues in pigs. In several breeding farms, limb and hoof diseases have resulted in economic losses, adversely affecting animal welfare. Approximately 20% to 50% of boars (Webb et al. 1983; Van Steenbergen 1989) and 20% to 40% of sows are prematurely culled owing to limb and hoof-related issues (Kirk et al. 2005). Previous studies have demonstrated that the within-herd prevalence of sow lameness is high, ranging from 8.8% to 16.9% (Heinonen et al. 2013). The etiological factors for limb and hoof weakness and injuries in pigs include nutritional imbalances, osteoporosis, excessive bodyweight, and suboptimal husbandry practices. The augmentation of cannon bone circumference (CBC) and bone mineral density (BMD) can enhance weight-bearing capacity and alleviate limb and hoof weakness (Jiang et al. 2020).

In pigs, four metacarpal or metatarsal bones with surrounding muscle and skin predominantly contribute to forelimb CBC (fCBC) and rear limb CBC (rCBC). The lateral growth of long bones is achieved through intraosseous osteogenesis (Karsenty and Wagner 2002; Montoya-Sanhueza and Chinsamy 2017). This process is concurrent with the accumulation of bone minerals. Hence, CBC may be genetically correlated to areal BMD (aBMD; Mokhtari-Dizaji et al. 2007). The physiological processes that increase CBC and BMD are complex and have not been elucidated. aBMD, which is defined as the mineral content within a unit area of the bone, can accurately predict the risk of fractures (Chevalley et al. 1991). CBC and BMD are complex traits with high heritability (Uemoto et al. 2008; Gong et al. 2019; Nan et al. 2020; Li et al. 2022). However, limited studies have focused on CBC and BMD in pigs. HMGA1, which is correlated with other physical traits, has been identified as a potential gene regulating CBC (Okumura et al. 2013; Ji et al. 2017; Gong et al. 2019). Thus, the identification of quantitative trait locus (QTL) and causal genes in the pig genome that regulate CBC and BMD is an area of active research.

Currently, both quantitative ultrasound and dual-energy X-ray absorptiometry (DXA) are used for measuring BMD in pigs. However, DXA provides density readings with increased resolution, improving both accuracy and precision. DXA can scan various body regions and can be modified based on the scanning area to calculate the corresponding regional BMD (Schreiweis et al. 2005). In contrast to alternative BMD measurement methods, DXA is associated with enhanced accuracy, precision, and adaptability.

This study used a DXA bone density scanner to measure the rear limb metatarsal area BMD (raBMD) of pigs (second and third metatarsal bones). After integrating BMD data with CBC data, QTLs and candidate genes regulating both CBC and raBMD were identified using genome-wide association studies (GWAS). These findings can be harnessed to expedite the selection for improved limb strength in pigs.

Materials and Methods

Ethical approval and consent to participate

All animal experiments were performed according to the Guidelines for the Care and Use of Laboratory Animals prepared by the Institutional Animal Welfare and Ethics Committee of Nanjing Agricultural University, Nanjing, China [certification no: SYXK (Su) 2022-0031].

Animals and phenotypic data

Yorkshire castrated boars (n = 413) of similar age were selected from the same farm of CP (Xuzhou) Food Co., Ltd and slaughtered in 5 batches. The carcass weight (CW) of each pig was recorded during slaughter. The mean CW was 99.8 (± 0.6) kg. The ear tissue samples were collected from all Yorkshire pigs (n = 413). The left forelimb and rear limb of slaughtered pigs were collected to measure fCBC, rCBC, and raBMD. CBC is the circumference of the middle part of the metacarpal bone or metatarsal bone of pigs. raBMD is the combined area BMD of the second metatarsal bone and third metatarsal bones of pigs measured using a DXA bone densitometer (Fig. 1). Complete information on slaughter batch, CW, fCBC, rCBC, and raBMD was recorded for all pigs. The fCBC data of 1 pig and the rCBC and raBMD data of 2 pigs were missing.

Figure 1.

The scanning image of the bone mineral density of second and third metatarsal bones in pigs was obtained using a dual-energy X-ray absorptiometry (DXA) bone density scanner. The rectangle represents the scanning area, while the bone mineral content and area of the irregular box were used to calculate bone mineral density.

Genotyping and quality control

Genomic DNA was extracted from ear tissue samples using the Megi universal nucleic acid extraction kit. Quantification and qualification of DNA were performed using a NanoDrop 2000 based on the following criteria: OD260/280 value = 1.7 to 2.1; concentration > 50 ng/μL. Genotyping was performed using the Compass Porcine 50K Plus Breeding Beadchip (Tianjing, China) containing 57,466 single-nucleotide polymorphisms (SNPs) across the whole genome. The physical positions of all SNPs were updated to the Sus scrofa 11.1 build (Sscrofa11.1) of the pig reference genome. Quality control was performed using PLINK v1.9 software (Chang et al. 2015). Data with SNP call rates < 90% were removed. SNPs with genotype-missing rates > 0.1 and minor allele frequency (MAF) < 0.05 were removed. Only SNPs on autosomal chromosomes were retained. After quality control, 413 individuals with 36,582 eligible SNPs were used for further analysis.

Genetic parameters estimation

Variance components and heritability were estimated using the single-trait model and multiple traits model in DMU (Version 6, release 5.5) software (Madsen et al., 2014). The statistical single-trait model for estimating variance components was as follows:

$${\displaystyle \begin{array}{l}{\displaystyle y=Xa+g+e,}\end{array}}$$ (1)

where y represents the vector of phenotype, a is the vector of fixed effects (fixed effects include CW and slaughter batch), g is the vector of random additive effect, e is the vector of random residual effects, and X is the incidence matrix. Random effects are assumed to follow a normal distribution and to be independent of each other. This study assumed that g ~ N (0, G${\sigma }_A^2$) and e ~ N (0, I${\sigma }_e^2$), where G is the matrix of additive genetic relationships derived from genome-wide markers of SNP-chip data using invgmatrix software (Vanraden, 2008). The mathematical equation for constructing a genomic relationship matrix was as follows:

$${\displaystyle \begin{array}{l}{\displaystyle G=\frac{Z{Z}^{\prime }}{\sum 2p_i{q}_i},}\end{array}}$$

where Z is the numeric coded genotype matrix with a dimension of n by m, n is the number of individuals and m is the number of markers. The codes of the ith column in the M matrix are (0 − 2pi) for genotype A₁A₁, (1 − 2pi) for genotype A₁A₂, and $(1-2pi)$ for genotype A₂A₂. $q_i$ and $p_i$ are the frequencies of allele 1 (A₁) and allele 2 (A₂) at locus i, respectively (Su et al. 2012). I is the identity matrix, ${\sigma }_A^2$ is the additive genetic variance, and ${\sigma }_e^2$ is the residual variance. The phenotypic variance ${\sigma }_p^2$ was defined as the sum of ${\sigma }_A^2$ and ${\sigma }_e^2$. Heritability $h^2$ was defined as the ratio of ${\sigma }_A^2$ and ${\sigma }_p^2$.

The statistical model of multiple traits for estimating variance components was as follows:

$${\displaystyle \begin{array}{l}{\displaystyle \left[\begin{array}{l}{y}_1{y}_2\end{array}\right]=\left[\begin{array}{lll}{X}_1& {\displaystyle 00}& {\displaystyle X_2}\end{array}\right]\left[\begin{array}{l}{a}_1{a}_2\end{array}\right]+\left[\begin{array}{l}{g}_1{g}_2\end{array}\right]+\left[\begin{array}{l}{e}_1{e}_2\end{array}\right],}\end{array}}$$ (2)

the meaning of each parameter ($y_i$, $a_i$, $g_i$, $X_i$) is same as model 1. This study assumed that $\left[\begin{array}{l}{g}_1{g}_2\end{array}\right]$ ~ N$(0,G\otimes \left[\begin{array}{ccc}{\sigma }_{A1}^2& {\displaystyle {\sigma }_{A1A2}{\sigma }_{A1A2}}& {\displaystyle {\sigma }_{A2}^2}\end{array}\right])$ and $\left[\begin{array}{l}{e}_1{e}_2\end{array}\right]$ ~ N$(0,I\otimes \left[\begin{array}{ccc}{\sigma }_{e1}^2& {\displaystyle {\sigma }_{e1e2}{\sigma }_{e1e2}}& {\displaystyle {\sigma }_{e2}^2}\end{array}\right])$, where ${\sigma }_{A1A2}$ and ${\sigma }_{e1e2}$ is the additive genetic covariance and residual covariance of the two traits, respectively. The definition of the rest of the parameters (G, I, ${\sigma }_{Ai}^2$, ${\sigma }_{ei}^2$) was the same as that in model 1 (where i represents one of the three traits fCBC, rCBC and raBMD). The genetic correlation ($r_G$) between traits was defined as follows:

$${\displaystyle \begin{array}{l}{\displaystyle r_G=\frac{{\sigma }_{A1A2}}{\sqrt{{\sigma }_{A1}^2\times {\sigma }_{A2}^2}},}\end{array}}$$

Meanwhile, the phenotypic correlation ($r_P$) between traits was defined as follows:

$${\displaystyle \begin{array}{l}{\displaystyle r_P=\frac{{\sigma }_{A1A2}+{\sigma }_{e1e2}}{\sqrt{{\sigma }_{A1}^2+{\sigma }_{e1}^2}\times \sqrt{{\sigma }_{A2}^2+{\sigma }_{e2}^2}}.}\end{array}}$$

Imputation

The whole-genome resequencing data of 1662 pigs were used as the reference panel, which includes the resequencing data of 1,602 pigs of multiple breeds from the PigGTEX project (Teng et al. 2024) and 60 Suhuai pigs (Liu et al. 2023). Haplotype phasing of the reference panel (Browning and Browning 2007) and the imputation of the SNP-chip data to the whole-genome density (Browning et al. 2018) were performed using the beagle software (version 5.2) with the default parameters. SNPs with dosage R-squared (DR2) < 0.9 and MAF < 0.05 were removed. After quality control, 413 individuals with 5,752,897 SNPs were retained for further analysis.

Genome-wide association studies

A univariate linear mixed model implemented in LDAK v5.2 software (Speed et al. 2012) was used to test the single-marker associations between each phenotype trait and the whole-genome imputed variants and, is described as follows:

$${\displaystyle \begin{array}{l}{\displaystyle y=Xa+w_i{b}_i+k+e,}\end{array}}$$ (3)

where the definitions ofy, a, X, and e$e$ are the same as those in model 1. b_i represents the substitution effect for the fitted SNP_i, w_i is the vector of genotype coefficients at SNP_i with elements of 0, 1, or 2 representing the number of a particular allele at SNP_i that each individual carries, k is the random additive effect following N (0, K${\sigma }_A^2$) distribution, K is the genomic kinship matrix calculated with the LDAK-Thin algorithm under the default parameters (prune threshold = 0.98, window size = 100 kb) (Wang et al. 2021a). This model assigns weights to each SNP based on linkage disequilibrium (LD) to calculate the kinship matrix. This method had strong power and efficiency for identifying trait-related SNPs and QTLs in pigs (Speed et al. 2012).

In the traditional model calculation process, the SNP appears in both fixed and random effects, which will lead to double-fitting and the loss of statistical power. However, each candidate SNP tested does not appear in fixed and random effects simultaneously, which requires significant computational resources. Therefore, the method of ‘leave-one-chromosome-out’ (Yang et al. 2014) was performed to address this problem. The genome-wide significance threshold was set to 0.05/N based on the Bonferroni correction method, and 1/N was set as the suggestive threshold (where N represents the number of filtered SNPs (36,582) in SNP-chip data). For iWGS data, due to the overly stringent genome-wide significance threshold associated with Bonferroni correction, N represents the number of effective SNPs (142,973), which was estimated with PLINK v1.9 software using the ‘--indep-pairwise 50 5 0.2’ parameter (Wang et al. 2021b). The proportion of phenotypic variance explained by the SNP additive effects was calculated as follows: $2p(1-p){\beta }^2/{\sigma }_P^2$, where p is the MAF of the SNP, β is the estimate of allele substitution effect, and ${\sigma }_P^2$ represents the phenotypic variance. The Manhattan plots and quantile–quantile (Q–Q) plots were drawn using the CMplot R package (Yin et al. 2021). To accurately dissect independent signals, the most significant SNP within each region was used as a fixed effect in the respective GWAS, enabling conditional analysis on each chromosome.

Multi-trait meta-analysis chi-squared statistic

The multi-trait meta-analysis (Bolormaa et al. 2014) was performed to identify the significance level of SNPs in multiple traits. The multiple-trait test statistic is approximately chi-square-distributed with the number of traits tested as the number of degrees of freedom. This statistic tests a null hypothesis stating that the SNP does not affect any of the traits. For each SNP, the multi-trait statistic was calculated as follows:

Multi-trait ${\chi }^2={t^{\prime }}_i{V}^{-1}{t}_i$, (4)

where t_i is a 3 × 1 vector of the signed statistic of SNP_i in single-trait GWAS for the 3 traits, which is the effect size divided by the standard error, t_i is a transpose of the vector t_i (1 × 3), and $V^{-1}$ is an inverse of the 3 × 3 correlation matrix. The correlation between the two traits is the correlation over the 5,752,897 estimated SNP effects (signed t-values) of the two traits. The method of determining significant thresholds in the meta-analyses was consistent with that of single-trait GWAS using iWGS data.

Bayes fine mapping

To generate a comprehensive list of candidate genes within the target QTLs, the Bayesian framework CAVIARBF (Chen et al. 2015) was used to determine the QTL confidence interval in iWGS data. For each signal, all significant SNPs (P < 0.01) within a 2 Mb window (± 1.0 Mb relative to the lead SNP) were considered as candidate variant sets (Maller et al. 2012). The marginal association statistics and LD patterns of the candidate variant set were used to calculate the smallest variant set that included causal variants with 95% posterior probability using CAVIARBF v0.2.1 software.

Function of candidate genes

To study the biological functions of the candidate genes within the QTL confidence interval, the Ensembl BioMart tool (https://asia.ensembl.org/index.html; release 110-July 2023) was used to identify genes within the QTL confidence interval. PubMed (https://pubmed.ncbi.nlm.nih.gov/) was used to search the literature for functional annotation of the candidate genes.

Results

Descriptive statistics and heritability estimates

The maximum, minimum, mean, standard error, and coefficient of variation values for each parameter of CBC and BMD are presented in Table 1.

Table 1.

Descriptive statistics for fCBC, rCBC, and raBMD

Trait	N1	Min2	Max3	Mean4 ± SE5	CV6(%)
fCBC, cm	412	15.9	21.0	18.35 ± 0.04	4.06
rCBC, cm	411	16.1	21.3	18.50 ± 0.04	4.29
raBMD, g/cm2	411	0.925	1.583	1.195 ± 0.005	8.52

¹Number of individuals with phenotypic records.

²Minimum of phenotype.

³Maximum of phenotype.

⁴Mean of phenotype.

⁵Standard error.

⁶Coefficient of variation.

The heritability values of CBC and BMD and the phenotypic and genetic correlation coefficients are summarized in Table 2. The heritability values for fCBC and rCBC were 0.52 (± 0.09) and 0.47 (± 0.09), respectively, indicating that CBC is strongly influenced by genetic factors. In contrast, raBMD exhibited a decreased heritability value of 0.16 (±0.07). The phenotypic and genetic correlation coefficients among these three parameters were significantly high (>0.33 and > 0.83, respectively).

Table 2.

Genetic parameters¹ for fCBC, rCBC, and raBMD

Trait	fCBC	rCBC	raBMD
fCBC	0.52 ± 0.09	0.86 ± 0.07	0.86 ± 0.14
rCBC	0.65 ± 0.02	0.47 ± 0.09	0.83 ± 0.15
raBMD	0.33 ± 0.02	0.40 ± 0.02	0.16 ± 0.07

¹Genetic correlation coefficients are listed in the upper triangle, phenotypic correlation coefficients are listed in the lower triangle, and heritability values are listed in the diagonal. All P values of the correlation coefficients are lower than 0.01.

Genome-wide association studies

No individual was clearly separated from the population in the result of principal component analysis, suggesting there was no population stratification phenomenon, and the neighbor-joining tree reveals the population is divided into 12 pedigrees (Supplementary Fig. S1).

To evaluate the accuracy of the imputation, 5% of SNPs were randomly removed and imputed again in SNP-chip data. The allele consistency rate and correlation of these 5% SNPs were used as the criteria for evaluating the accuracy of the imputation. The average consistency rate and correlation were 98.46% and 98.67%, respectively (Supplementary Fig. S2), after filtering based on the following criteria: MAF > 0.05 and DR² > 0.9. GWAS was performed using SNP-chip and iWGS data for fCBC, rCBC, and raBMD. The Manhattan plots for the traits are presented in Fig. 2 and the corresponding Q–Q plots are presented in Supplementary Fig. S3.

Figure 2.

Manhattan plots of genome-wide association studies for cannon bone circumference (CBC) and rear limb metatarsal area bone mineral density (raBMD) traits. The results in (a), (c) and (e) were based on single nucleotide polymorphism (SNP)-chip data analysis, while those in (b), (d), and (f) were based on imputed whole-genome sequencing (iWGS) data analysis. Negative log10 P-values of SNPs (y-axis) were plotted against their corresponding genomic positions (x-axis). The horizontal solid and dashed lines represent the genome-wide significance (P = 3.50E–07) and suggestive thresholds (P = 6.99E–06), respectively.

Based on SNP-chip data, 81 SNPs distributed on SSC6, SSC12, and SSC13 were significantly associated with fCBC. Of these, 15 SNPs exceeded the genome-wide significance threshold. The most significant SNP (CNC10120304) on SSC12 explained 7.63% of the phenotypic variance, while the most significant SNPs on SSC6 (CNC10063057) and SSC13 (CNC10132307) explained 5.31% and 6.96% of the phenotypic variance, respectively (Table 3). For rCBC, 10 significant SNPs were identified on SSC12, SSC14, and SSC15. The most significant SNP (CNC10120304) on SSC12 was consistent with the results of fCBC analysis, explaining 6.27% of the phenotypic variance. Meanwhile, the most significant SNPs on SSC14 (CNC10140182) and SSC15 (CNCB10011072) explained 5.25% and 6.16% of the phenotypic variance, respectively (Table 3). However, no SNPs associated with raBMD were identified using GWAS with SNP-chip data.

Table 3.

GWAS results for fCBC, rCBC, and raBMD based on SNP-chip and iWGS data

Trait	Dataset	Chr1	Position	Top SNP	P-value of the top SNP2	Var3	Nearest gene to the top SNP
fCBC	SNP-chip data	6	164317121	CNC10063057	1.3581E-05	5.31%	ENSSSCG00000020872
		12	13826063	CNC10120304	1.353E-07	7.63%	PITPNC1
		13	107044070	CNC10132307	1.5581E-07	6.96%	GOLIM4
	iWGS data	6	154155607	rs81320554	2.3742E-06	5.97%	OMA1
		12	15155264	rs332162515	1.2994E-09	10.10%	DDX42
		13	106951317	rs334440543	5.8284E-08	7.34%	GOLIM4
rCBC	SNP-chip data	12	13826063	CNC10120304	1.9703E-06	6.27%	PITPNC1
		14	8134833	CNC10140182	5.8319E-06	5.25%	STC1
		15	122663668	CNCB10011072	6.8297E-06	6.16%	EPHA4
	iWGS data	12	13805191	rs340361371	5.7701E-08	7.87%	PITPNC1
		14	8130847	rs330628419	3.1579E-06	5.32%	STC1
			82001659	rs343413376	1.0972E-06	5.55%	MBL1
raBMD	iWGS data	7	86861210	rs345479532	1.6569E-06	6.54%	SLCO3A1

¹ Sus scrofa chromosome, the same as below.

² P-value according to the Wald test.

³Phenotypic variation explained by the top SNP.

Based on iWGS data, 3206 significant SNPs associated with fCBC were detected on SSC6, SSC12, and SSC13 (Supplementary Table S1). The most significant SNPs on SSC6 (rs81320554), SSC12 (rs332162515), and SSC13 (rs334440543) explained 5.97%, 10.10%, and 7.34% of the phenotypic variance, respectively. Meanwhile, the most significant SNP on SSC12 explained 7.87% of the phenotypic variance (Table 3). For rCBC, 524 significant SNPs were identified on SSC12 and SSC14 (Supplementary Table S2). The most significant SNP (rs340361371) on SSC12 explained 7.87% of the phenotypic variance. Two significant QTLs were detected on SSC14. The most significant SNPs rs330628419 and rs343413376 explained 5.32% and 5.55% of the phenotypic variance, respectively (Table 3). Additionally, 14 significant SNPs associated with raBMD were identified on SSC7 (Supplementary Table S3). The most significant SNP, which was located in SLCO3A1, explained 6.54% of the phenotypic variance (Table 3).

In the conditional analysis, the top SNP within each significant region was incorporated as a fixed effect into the GWAS model. No SNPs exceeded the significance threshold (Supplementary Figs. S4 to S6), indicating the presence of high LD among significant SNPs within these regions. Only one QTL was present in these significant regions.

Multi-trait meta-analysis

As this study demonstrated a high genetic correlation between CBC and raBMD, meta-analysis of the three traits was performed to detect pleiotropic SNPs. In total, 546 significant SNPs on SSC12, SSC13, and SSC15 were identified (Fig. 3, Supplementary Table S4). The significant signal on SSC15 (Supplementary Fig. S7), which was not detected in the single-trait analysis, suggested that this QTL is significantly associated with both CBC and raBMD.

Figure 3.

Manhattan plots of multi-trait meta-analysis between forelimb cannon bone circumference (fCBC), rear limb cannon bone circumference (rCBC), and rear limb metatarsal area bone mineral density (raBMD) traits based on imputed whole-genome sequencing (iWGS) data. Negative log10 P-values of single-nucleotide polymorphisms (SNPs) (y-axis) were plotted against their corresponding genomic positions (x-axis). The horizontal solid and dashed lines represent the genome-wide significance (P = 3.50E–07) and suggestive thresholds (P = 6.99E–06), respectively.

Bayes fine mapping

For each signal, the ability of each variant to explain the observed signal within a 2 Mb window (±1.0 Mb relative to the original lead SNP) was computed, and the smallest set of variants that included the causal variant with 95% probability (95% credible set) was obtained. For fCBC, the QTL confidence intervals were determined to be 153.32 to 154.99, 15.11 to 15.17, and 106.37 to 107.76 Mb on SSC6, SSC12, and SSC13, respectively (Fig. 4). BFM indicated that the QTL confidence intervals for rCBC were 13.75 to 14.80 Mb on SSC12 and 7.81 to 9.12 and 81.05 to 82.97 Mb on SSC14 (Fig. 5). For raBMD, the QTL confidence interval on SSC7 was 85.86 to 87.82 Mb (Fig. 6). The results of the meta-analysis were subjected to BFM (Table 5). The QTL confidence interval on SSC15 was determined to be 137.55 to 137.56 Mb (Fig. 7), which was located 23.68 kb downstream of RAMP1. On SSC12, the posterior probability of the variant rs332162515 located on the DDX42 exceeded 0.95 (Fig. 7). Due to the low marker density and potential biases in iWGS data, only this variant was considered as a promising SNP. However, this result still forms the basis for precise regional mapping and candidate gene selection. The confidence interval on SSC13 was 107.09 to 107.32 Mb (Supplementary Fig. S8), which is narrower than the QTL confidence interval identified in the single-trait GWAS results.

Figure 4.

Bayes fine mapping of the quantitative trait loci (QTLs) regulating forelimb cannon bone circumference (fCBC) on chromosomes 6 (a), 12 (b), and 13 (c). The x-axis represents the physical positions of single nucleotide polymorphisms (SNPs) in the genome, while the y-axis indicates posterior probability.

Figure 5.

Bayes fine mapping of the quantitative trait loci (QTLs) regulating rear limb cannon bone circumference (rCBC) on chromosomes 12 (a) and 14 (b) (c). The x-axis represents the physical positions of single nucleotide polymorphisms (SNPs) in the genome, while the y-axis indicates posterior probability.

Figure 6.

Bayes fine mapping of the quantitative trait loci (QTLs) regulating rear limb metatarsal area bone mineral density (raBMD) on chromosome 7. The x-axis represents the physical positions of single nucleotide polymorphisms (SNPs) in the genome, while the y-axis indicates posterior probability.

Table 5.

The significant SNPs and corresponding QTL confidence intervals through multi-trait meta-analysis using iWGS data

Chr	QTL region, Mb	Top SNP	Position of the top SNPs, bp	P-value of the top SNP	Nearest gene of QTL	Distance1
12	15.16	rs332162515	15155264	1.4263E-09	DDX42	In
13	107.09 to 107.32	rs334440543	106951317	1.10695E-06	GOLIM4	Up
15	137.55 to 137.56	rs698895836	137552871	1.12275E-06	RAMP1	Up

¹The relationship between QTL confidence intervals and nearest gene. Down represents the nearest gene in downstream of the QTL confidence intervals. Up represents the nearest gene in upstream of the QTL confidence intervals. In represents the nearest gene in the QTL confidence intervals.

Figure 7.

Bayes fine mapping of the quantitative trait loci (QTLs) applying results of multi-trait meta-analysis on chromosomes 12 (a) and 15 (b). The x-axis represents the physical positions of single nucleotide polymorphisms (SNPs) in the genome, while the y-axis indicates posterior probability.

Candidate genes

In total, 62 protein-coding genes were located within the QTL confidence intervals determined using BFM in the three single-trait GWAS (Table 4). Functional annotation based on literature review revealed the candidate genes regulating different phenotypes were as follows: fCBC: MYSM1, FTSJ3, and MECOM; rCBC: SMURF2 and STC1; raBMD: RGMA. For multi-trait meta-analysis (Table 5), DDX42 on SSC12 (which includes the promising SNP determined by BFM) and RAMP1 on SSC15 (the protein-coding gene closest to the QTL confidence interval determined by BFM) were identified as functional candidate genes that may potentially exhibit pleiotropic effects.

Table 4.

Protein coding genes for fCBC, rCBC, and raBMD in the QTL regions identified through Bayes fine mapping

Trait	Chr	QTL region (Mb)	Genes1
fCBC	6	153.32 to 154.99	JUN, MYSM1 , TACSTD2, OMA1, ENSSSCG00000054142, DAB1
	12	15.11 to 15.17	FTSJ3 , DDX42 , CCDC47
	13	106.37 to 107.76	WDR49, PDCD10, SERPINI1, ENSSSCG00000040950, GOLIM4, ENSSSCG00000035525, ENSSSCG00000062384, MECOM
rCBC	12	13.75 to 14.80	PITPNC1, NOL11, BPTF, C17orf58, KPNA2, SMURF2 , CEP95, DDX5, POLG2, MILR1, PECAM1, TEX2
	14	7.81 to 9.12	NKX2-6, STC1 , ADAM28, ADAMDEC1, ADAM7, ENSSSCG00000057307, ENSSSCG00000060373, ENSSSCG00000009648, NEFL
		81.05 to 82.97	ZMIZ1, PPIF, ZCCHC24, ANXA11, PLAC9, TMEM254, MBL1, SFTPD, SFTPA1, MAT1A, DYDC1, ENSSSCG00000045226, DYDC2, PRXL2A, TSPAN14, SH2D4B, ENSSSCG00000010343, ENSSSCG00000051331
raBMD	7	85.86 to 87.82	RGMA , CHD2, FAM174B, ST8SIA2, SLCO3A1, SV2B

¹The genes indicated in bold were identified as potential functional candidate genes for the corresponding trait under analysis based on literature support.

Discussion

Heritability estimates

CBC is classified as a moderately to highly heritable trait with heritability values ranging from 0.46 to 0.77 (Uemoto et al. 2008; Gong et al. 2019; Li et al. 2022). In this study, the heritability estimates for fCBC and rCBC were 0.52 and 0.47, respectively, which were consistent with the previously reported values. Research on BMD has traditionally focused on humans and mice. In humans, most phenotypes related to BMD exhibit high heritability values, typically ranging from 0.5 to 0.85 (Smith et al. 1973; Gueguen et al. 1995; Slemenda et al. 1996). Similarly, in mice, the heritability values of BMD are estimated to be in the range of 0.6 to 0.7 (Beamer et al. 1999; Li et al. 2001). However, in pigs, research has predominantly focused on leg weakness and leg scores (Laenoi et al. 2011, 2012; Guo et al. 2013). Nan et al. measured BMD in 212 Landrace and 537 Yorkshire pigs using an ultrasound bone densitometer and reported estimated heritability values of 0.21 and 0.31, respectively (Nan et al. 2020). In this study, the heritability value of raBMD measured using DXA was 0.16. The variations observed may be attributed to differences in measurement methods, population size, and feeding. This study, for the first time, performed a simultaneous genetic parameter estimation for both CBC and raBMD and reported a strong and significant genetic correlation between them.

Genome-wide association studies

GWAS is a powerful method for identifying associations between genetic variations and traits (Tam et al. 2019). However, the Illumina array platforms, which are extensively used, primarily comprise SNPs linked to common genetic variations. These arrays encompass a fraction of the genetic diversity landscape, which encompasses over 300 million common and rare variants (Sherry et al. 2001). Whole-genome sequencing is costly and technically challenging, especially for analyzing large sample sizes. Genotype imputation is the most widely used approach to compensate for the lack of comprehensive genomic coverage by genotyping arrays (Naj 2019). This study performed single-trait GWAS using SNP-chip data and iWGS data to identify SNPs significantly associated with fCBC on SSC6, SSC12, and SSC13. For rCBC, the significant SNPs were identified on SSC12, SSC14, and SSC15. QTLs associated with CBC have been previously reported on SSC6, SSC12, SSC13, and SSC14 in pigs. The candidate genes, such as HSF5 on SSC12 and TEX14 on SSC13, were preliminarily identified as potential regulators of CBC (Liu et al. 2021; Li et al. 2022). However, these findings were not consistent with the results of this study, which may be due to genetic differences among populations. Furthermore, a strong and significant QTL associated with CBC was previously identified near 30.6 Mb on SSC7. This QTL contains a highly promising candidate gene HMGA1 (Okumura et al. 2013; Ji et al. 2017; Gong et al. 2019). Similarly, QTLs related to CBC have also been reported on SSC1, SSC2, SSC3, SSC4, SSC10, SSC15, SSC16, and SSC18 in pigs (Soma et al. 2011; Zhou et al. 2019; Liu et al. 2021; Li et al. 2022). Additionally, this study identified significant SNPs associated with raBMD on SSC7 using iWGS data with GWAS. Limited studies have focused on BMD in pigs. These studies have used diverse methods and measurement sites for BMD assessment. QTLs potentially correlated with porcine distal femur aBMD measured using DXA have been reported at 127 and 106 cM on SSC4 and SSC11, respectively (Mao et al. 2008). One study used the DXA method and detected significant QTLs influencing porcine whole-body aBMD on SSC6 and SSC12. However, no candidate genes related to BMD were detected within these regions (Rothammer et al. 2014). Single-marker GWAS and meta GWAS of Yorkshire and Landrace pig BMD, which was assessed using an ultrasound bone densitometer, revealed a significant signal on SSC6. In this region, CNR2, ZBTB40, and LIN28A were identified as the candidate genes due to their previously reported correlation with human BMD (Nan et al. 2020). However, this study measured aBMD in the metatarsal bones using DXA. Despite the absence of significant signals in GWAS using SNP-chip data, a novel QTL associated with raBMD on SSC7 was identified using iWGS data analysis.

For the proportion of phenotypic variance explained by the SNP additive effects, the presence of the Beavis effect must be considered (Xu 2003). Due to limitations in population size, the effect sizes of the most significant SNPs may be overestimated, leading to a certain degree of overestimation in the phenotypic variance explained by these SNPs. However, this does not compromise the reliability of the results.

Multi-trait meta-analysis

Multi-trait meta-analysis enhances the power to detect QTL, especially among correlated traits (Liu et al. 2017; Fang and Pausch 2019). In this study, the high positive genetic correlations between traits were leveraged, and multi-trait meta-analysis was performed to identify SNPs with pleiotropic effects. A novel QTL on SSC15, which was undetected in a single-trait GWAS, revealed the possibility of pleiotropic effects of this region on CBC and raBMD. The lack of significant signals in the single-trait GWAS for this region may be attributed to its relatively modest effect. Previous studies have identified several QTLs that were not detected in single-trait GWAS using multivariate methods, such as multi-trait meta-analysis (Noskova et al. 2023). Additionally, there have been reports of associations detected in single-trait GWAS that may vanish in multi-trait analyses (Pausch et al. 2016; Turley et al. 2018).

Bayes fine mapping

The intricate LD patterns within genomic regions can lead to potential misdirection when evaluating correlations with a single SNP. Furthermore, the reliance only on pairwise LD patterns or even haplotype blocks for fine mapping of complex traits is constrained by factors beyond recombination (Schaid et al. 2018). BFM combines genomic LD information with marginal association statistics (Z-scores) from a candidate variant set to identify the minimal set of variants with high confidence that includes causal mutation (Li et al. 2021). This study refined the QTL confidence intervals for key significant regions identified in single-trait GWAS using BFM, narrowing down the QTL confidence interval on SSC12 associated with fCBC to 56 kb (15.11 to 15.17 Mb). Additionally, BFM was applied to the results of the meta-analysis. For the significant QTLs on SSC12 and SSC13 previously identified in single-trait analyses, the QTL confidence intervals were further refined, particularly on SSC12. The previously identified large-effect region was narrowed down to a single promising SNP (rs332162515). A previous study on the effector genes associated with human coronary artery calcification used the Bayesian methods to identify functional mutations with a posterior probability of 1. The confidence intervals for the remaining significant QTLs were narrowed down to a region within 9.5 kb (Kavousi et al. 2023).

Candidate genes

The functional information of all protein-encoding genes within the QTL confidence intervals was retrieved to identify functional candidate genes associated with CBC and raBMD. Both CBC and BMD are considered growth-related traits and are closely related to skeletal growth and development. For fCBC, mesenchymal stem cells (MSCs) within the bone marrow serve as precursor cells for osteoblasts, chondrocytes, adipocytes, and other cell types involved in the processes of bone formation and resorption (Jiang et al. 2002; Gregory et al. 2005; Baksh et al. 2007; Herdrich et al. 2008; Mumme et al. 2012). MYSM1 plays a crucial role in the maintenance and differentiation fate determination of MSCs (Li et al. 2016). Mysm1-deficient mice exhibit aberrant hindlimb and tail development (Jiang et al. 2011; Gatzka et al. 2015). FTSJ3 is reported to be correlated with growth-related traits, such as bodyweight and growth rate in chickens (Seifi et al. 2021), as well as with body size in dogs (Abrams et al. 2020). Additionally, CBC exhibits a significant phenotypic and genetic correlation with growth-related traits (Wang et al. 2014). Mecom was associated with limb bud formation and development during the embryonic period in mice. Evi-1, a protein encoded by MECOM, is upregulated in limb bud mesenchyme during embryonic cartilage formation before ossification (Perkins et al. 1991). Moreover, the loss of Mecom in mice is associated with spinal deformities and decreased bone mass (Juneja et al. 2014). Therefore, MYSM1 on SSC6, FTSJ3 on SSC12, and MECOM on SSC13 are promising candidate genes regulating fCBC. Both rCBC and fCBC were associated with major signals within the same genomic region on SSC12. However, BFM revealed that different lead SNPs resulted in distinct QTL confidence intervals. Consequently, within the QTL confidence interval on SSC12, SMURF2 was identified as the candidate gene regulating rCBC. The process of cortical bone thickening in the skeleton is primarily driven by bone deposition by osteoblasts within the periosteum and bone resorption by osteoclasts (Montoya-Sanhueza and Chinsamy 2017). Several studies have indicated that SMURF2 can regulate bone homeostasis by modulating the activity of osteoblasts and osteoclasts (Wu et al. 2008; Huang et al. 2016; Xu et al. 2017; Zhang et al. 2021). Additionally, STC1 can regulate blood calcium levels by increasing renal phosphate reabsorption. In vitro studies have demonstrated that STC1 can reduce calcium flux in the intestines of rats and pigs (Olsen et al. 1996; Madsen et al. 1998). Furthermore, studies involving transgenic mice overexpressing Stc1 have reported stunted growth and decreased bone size (Varghese et al. 2002). Hence, STC1 on SSC14 is a potential candidate gene. For raBMD, RGMA was the most promising candidate. RGMA serves as an auxiliary receptor for bone morphogenetic proteins, which belong to the transforming growth factor-beta superfamily of growth factors (Babitt et al. 2005; Xia et al. 2007; Severyn et al. 2009; Tian and Liu 2013). Previous studies have reported that RGMA is one of the top candidate genes strongly associated with trabecular bone thickness (Levy et al. 2015). The augmentation of bone formation and the suppression of bone resorption effectively increase trabecular bone volume and thickness in mice (Thongchote et al. 2014). The increase in trabecular bone thickness is considered to be one of the factors contributing to increased BMD. In the multi-trait meta-analysis, the promising SNP was located on SSC12 within the genomic region of DDX42. This suggests that DDX42 may have a significant effect or pleiotropic effect on CBC or raBMD. DDX42 encodes a member of the Asp-Glu-Ala-Asp (DEAD) box protein family. Members of this protein family are putative RNA helicases and are involved in various cellular processes regulating RNA secondary structure (Uhlmann-Schiffler et al. 2006). The role of DDX42 in skeletal growth and development has not been previously reported., and its functional impact on CBC and BMD warrants further investigation. Additionally, RAMP1 on SSC15 is a potential pleiotropic candidate gene. The overexpression of RAMP1 promotes the expression of calcitonin receptor-like receptor (Appelt et al. 2020; Xu et al. 2020). Cells overexpressing RAMP1 form numerous well-defined mineralized nodules and exhibit typical calcium deposits in the bone matrix structure (Zhao et al. 2013). In follow-up studies, we plan to expand the sample size to comprehensively identify genes associated with CBC and BMD. Additionally, validation will be conducted at both the cellular and organism levels.

Conclusions

This study revealed a strong correlation between CBC and raBMD. A combination of GWAS and BFM was used to identify 7 novel QTLs associated with CBC or BMD. Additionally, multi-trait meta-analysis was used to identify an additional QTL on SSC15 that may exert pleiotropic effects on both CBC and BMD. This study identified the following genes regulating different phenotypes: fCBC: DDX42, MYSM1, FTSJ3, and MECOM; rCBC: SMURF2 and STC1; raBMD: RGMA. RAMP1 was identified as a potential pleiotropic candidate gene regulating both CBC and raBMD. These findings improved our understanding of the factors regulating CBC and BMD, offering valuable insights for exploring pathogenic genes and elucidating the genetic mechanisms underlying the regulation of CBC and BMD.

Glossary

Abbreviations

BMD

bone mineral density

BFM

Bayes fine mapping

CBC

cannon bone circumference

CW

carcass weight

DXA

dual-energy X-ray absorptiometry

fCBC

fore limb cannon bone circumference

GWAS

genome-wide association studies

iWGS

imputed whole-genome sequence

LD

linkage disequilibrium

MSC

mesenchymal stem cell

QTL

quantitative trait locus

raBMD

rear limb metatarsal area bone mineral density

rCBC

rear limb cannon bone circumference

Data Availability

The phenotype data (https://doi.org/10.6084/m9.figshare.24204411.v1) and SNP-chip data (https://doi.org/10.6084/m9.figshare.24204171.v1) used in this study are deposited in the figshare repository.

Conflict Interest Statement

The authors declare that they have no competing interests.