A GWAS was performed using 1156 DNA samples that were selected from more than 27,500 Japanese Black steers. The carcass weight of the selected samples had higher ratios of both extremes than in the collected samples, but was normally distributed (Additional file 1). The 39,011 SNP on autosomes that fulfilled the quality control criteria were used for the association analysis. The analysis was performed using EMMAX software 13, which is based on a linear mixed model approach using a genetic relationship matrix estimated by high-density SNP genotypes to model the correlation between the phenotypes of the sample subjects. The approach can adjust for both population stratification and relatedness between samples 14. The genomic inflation factor (λ_GC) was 1.008 in this analysis, indicating that the adjustment of the sample structure of our samples was appropriate. The quantile-quantile (Q-Q) plot with p < 10^-3 revealed large deviations from the distribution under the null hypothesis, indicating that robust associations were obtained (Additional file 2). Bonferroni-corrected genome-wide significant associations (p < 1.28 × 10^-6) were obtained for 19 SNP on 3 chromosomes: 12 on BTA 6, 3 on BTA 8, and 4 on BTA 14 (Figure 1; Additional file 3).

The strongest association was detected with Hapmap46986-BTA-34282 (p = 1.03 × 10^-12) located at 23.5 Mb on BTA 14 (Btau4.0; Figure 2A; Table 1). This and three other significant SNP located from 23.5 to 26.2 Mb on BTA 14 were in moderate linkage disequilibrium (LD) with linkage disequilibrium coefficients (r²) ranging from 0.23 to 0.79. The second strongest association was obtained with Hapmap26308-BTC-057761 (p = 9.65 × 10^-12) on BTA 6, which had 11 other significant SNP in the interval (Figure 2B; Table 1). Eleven SNP, including Hapmap26308-BTC-057761, located within 3.5 Mb from 35.7 to 39.2 Mb and were in strong LD with r² ranging from 0.65 to 1.00. A residual SNP was 7.9 Mb from them, but in strong LD with the 11 SNP (r² = 0.62-0.73). An extended LD of this region was also observed in a previous study 3, which is probably due to a specific (local) lineage of the Q allele. On BTA 8, 3 SNP locating from 88.7 Mb to 89.4 Mb reached genome-wide significance (Figure 2C). The most significant SNP was BTA-52694-no-rs (p = 4.24 x 10^-9; Table 1). In this case, the LD among the 3 SNP was relatively weak with r² ranging from 0.14 to 0.35.

¹⁾A, SNP on the BovineSNP50 BeadChip that showed the strongest association on a chromosome.

¹⁾B, Critical SNP or a haplotype in the associated regions.

We then performed conditioned analyses to examine the independence of the SNP associations. Genotypes of the most associated SNP on each chromosome, Hapmap26308-BTC-057761 on BTA 6, BTA-52694-no-rs on BTA 8, and Hapmap46986-BTA-34282 on BTA 14, were individually included as a covariate in the mixed model. After conditioning, the associations of the surrounding SNP disappeared (Figure 2A, B), indicating that a single QTL was present in the respective regions. In the case of BTA 8, the p value of Hapmap40466-BTA-82123 decreased from 0.216 to 1.31 × 10^-6, a nearly genome-wide significance level, after conditioning on BTA-52694-no-rs, while BTA-52694-no-rs became much more significant (p = 2.17 × 10^-14) after conditioning on Hapmap40466-BTA-82123. When conditioned on both BTA-52694-no-rs and Hapmap40466-BTA-82123, the associations of the surrounding SNP disappeared (Figure 2C). This result suggested that a certain haplotype comprising the two SNP represented the QTL on BTA 8 and that a SNP in strong LD with the causative variation of the QTL was not included in the BovineSNP50 BeadChip. When conditioned on the four SNP described above, the Q-Q plot did not deviate from the distribution under the null hypothesis, indicating that no other QTL were detected (Additional file 2).

In our half-sib analyses, we previously detected carcass weight QTL on BTA 14 (CW-1) 121516, BTA 6 (CW-2) 13, and recently on BTA 8 (CW-3; Additional file 4). As for CW-2, the SNP encoding NCAPG Ile442Met has been identified as a candidate causative variation 3. The SNP was located at 38.2 Mb and was in strong LD with Hapmap26308-BTC-057761 (r² = 0.97), indicating that the observed association corresponded to CW-2 (Figure 2B; Table 1). On BTA 8, we detected a carcass weight QTL around 90 cM in two half-sib families (Additional file 4), whose sires are father and son and shared a superior Q haplotype (data not shown). Because the Q haplotype of the sires had an A-allele for BTA-52694-no-rs and a C-allele for Hapmap40466-BTA-82123, the haplotypes comprising the two SNP were estimated for the GWAS samples and the association of the Q haplotype was examined (Table 1). The Q haplotype was highly associated with carcass weight (p = 8.13 × 10^-13), indicating that the association detected on BTA 8 corresponded to CW-3. In contrast, the 1.1-Mb critical region for CW-12 was 2.3-Mb telomeric from the most strongly associated SNP on BTA 14 in the GWAS. The details are discussed below.

The CW-1 QTL was originally mapped between BMS1941 and INRA094 (24.7-30.9 Mb on Btau4.0) using a Japanese Black half-sib family 15, and narrowed down between DIK7012 and DIK7020 (25.8-26.7 Mb on Btau4.0) by LD mapping 2. A subsequent haplotype analysis showed that the six sires segregating a carcass weight QTL around the CW-1 locus shared the same Q haplotype between DIK7013 and NRKM-040 (26.0-26.7 Mb on Btau4.0) with the CW-1-segregating sire 1, indicating that all seven sires harbored the CW-1 QTL. The fact that many sires harbored the CW-1 QTL suggested that the QTL was prevalant in the Japanese Black population. Despite this, the most significant SNP on BTA 14 in the GWAS (Hapmap46986-BTA-3428 at 23.5 Mb on Btau4.0) was slightly outside of the confidence interval (CI) of CW-1 and the conditioned analysis of the GWAS showed only one significant QTL on BTA 14. Thus, we performed haplotype analysis and LD mapping using high-density microsatellites developed around Hapmap46986-BTA-34282, to examine if the observed association was derived from CW-1, and if so, which location CW-1 lies in. Haplotypes were examined using the three sires that segregated the CW-1 QTL and harbored the same Q but different q (inferior) haplotypes between DIK7013 and NRKM-0401. All three sires shared the same Q haplotype spanning more than 1.1 Mb around Hapmap46986-BTA-34282 (Additional file 5). Therefore, the two identity-by-state (IBS) regions located 2.4-Mb apart on the Q haplotypes of the CW-1-segregating sires, which implied that the observed association was consistent with CW-1. The sires also shared the same q alleles at the two consecutive microsatellites close to Hapmap46986-BTA-34282 (Additional file 5).

Linkage disequilibrium mapping was performed using 142 cattle with heavy carcass weight and 145 cattle with light carcass weight that were selected from the GWAS samples and did not include half-sibs in either group. Also, in the selected samples, Hapmap46986-BTA-34282 had the strongest association on BTA 14 (Figure 3upper). The frequency of haplotypes comprising two contiguous microsatellites was estimated in each group and the difference in haplotype frequencies between the groups was tested by Fisher’s exact test using 2xn or 2x2 contingency tables (Figure 3middle). A stronger association was detected around Hapmap46986-BTA-34282 than the previously identified CW-1 region in both tests using 2xn and 2x2 contingency tables, respectively, indicating that CW-1 did not locate within the previously identified region, but located between DIK7104 and DIK7113 (22.8-23.7 Mb on Btau4.0; 24.6-25.4 Mb on UMD3.0). The revised CW-1 region included both RPS20 and PLAG1-CHCHD7, which were previously identified as a locus for body size 9 and stature 11, respectively.

To identify the causative variation of CW-1, targeted resequencing was performed. The three sires segregating the CW-1 QTL, three Q-homozygous sires, and a q-homozygous steer were subjected to sequence capture using a NimbleGen custom array followed by resequencing using Illumina GAIIx (Illumina, San Diego, CA). The Q-homozygous sires were determined as those with at least one Q haplotype within the revised CW-1 region and homozygous Q alleles at Hapmap46986-BTA-34282 (Additional file 5), and did not segregate a carcass weight QTL on BTA 14 in the half-sib analyses using more than 255 offspring per family. The respective GAIIx-reads (40-bp paired-ends) were aligned against both Btau4.0 and UMD3.1 as reference genome sequences, because UMD3.1 includes the largest number of gene annotations among the existing bovine genome assemblies, while information about sequence conservation across species is available on Btau4.0 17. The mean sequence depth of the targeted region was estimated to be 26 to 59 per animal. On UMD3.1, 711 to 1309 putative sequence variations were identified in the CW-1-heterozygous sires. The obtained putative sequence variations were filtered by: (1) heterozygous in the three sires segregating the CW-1 QTL, (2) homozygous in the three Q- and the q-homozygous animals, and (3) opposite alleles between the Q- and the q-homozygous animals, resulting in 211 candidate QTN (197 SNP and 14 Indel) on UMD3.1. There were 4 exonic variations, including 1 non-synonymous SNP, while 1 synonymous SNP and 5 SNP in the intergenic regions were within highly conserved segments in mammals (PhastCons score >0.9 on Btau4.0 17). We checked the sequence coverage of the coding regions and the highly conserved regions and confirmed that no other candidate QTN was present in them except for the PLAG1CHCHD7 intergenic region (see Methods). The same variations (1 SNP and 1 variable number of tandem repeat) as those identified by Karim et al. 11 were detected in the PLAG1CHCHD7 intergenic region of the CW-1 heterozygous sires, while the SNP in a putative polyadenylation signal of RPS209 was homozygous in one of the CW-1 heterozygous sires (Sire J in Additional file 5). Thus, a total of 11 sequence variations were examined for an association with carcass weight (Figure 3lower; Additional file 6). The strongest association was obtained for the variations between PLAG1 and CHCHD7, strongly suggesting that CW-1 was identical to the QTL for bovine stature identified by Karim et al. 11.

We compared haplotypes around the QTN for bovine stature between the Japanese Black sires and the F1 (Holstein x Jersey) sires reported by Karim et al. 11. The Q haplotype of the Japanese Black sires had a mosaic structure of the Q and q haplotypes of the F1 sires, and shared the same Q alleles only at the QTN for stature, FJX_PLAPROTRI and FJX_PLAPROSNP (Additional file 7). This observation strongly supports the idea that a crossbreed comparison is a useful way to pinpoint the causative variation for a QTL in bovine 3 as well as in canine 18.

Allele frequencies and effect size of the three QTL were examined using FJX_PLAPROTRI (CW-1), NCAPG c.1326T > G (CW-2), and a haplotype comprising Hapmap40466-BTA-82123 and BTA-52694-no-rs (CW-3) (Table 1). The Q allele frequency was high for CW-1 (75.4%), but low for CW-2 (20.0%) and CW-3 (11.5%), while allele substitution effects of CW-2 and CW-3 (35.3 and 35.0 kg/allele) were larger than that of CW-1 (28.4 kg/allele). Thus, CW-2 and CW-3 can be used to efficiently increase carcass weight in the Japanese Black population.

The fraction of phenotypic variance explained by the empirically estimated relatedness matrix is calculated in EMMAX as pseudoheritability: h 2 = σ a 2 / σ a 2 + σ e 2 . The value was 0.636 and decreased to 0.423 after conditioning on the 3 QTL (FJX_PLAPROTRI, NCAPG c.1326T > G and a haplotype comprising BTA-52694-no-rs and Hapmap40466-BTA-82123), suggesting that the 3 QTL account for one-third [=(0.636-0.423)/0.636] of the total genetic variance of the GWAS samples.

This research was undertaken with the approval of the Shirakawa Institute of Animal Genetics Committee on Animal Research (H21-1).

Perirenal fat tissues of more than 27,500 Japanese Black steers were collected at two slaughterhouses from 2000 to 2008. The cattle were reared in different herds throughout Japan. Carcass traits such as cold carcass weight and marbling were systematically measured by certified graders and recorded at slaughterhouses. Sire DNA used to analyze the CW-1 region was obtained as semen.

First, animals were selected from the 15% extremes of the cold carcass weight distribution that did not include a half-sib in each extreme, resulting in 199 steers for heavy carcass weight and 299 steers for light carcass weight. Then, 658 steers that had been selected from both extremes for marbling and genotyped using the BovineSNP50 BeadChip (Illumina) were added to increase the sample numbers. The 658 steers included half-sibs and distributed from heavy to light carcass weight. The resulting population comprised 1156 steers with a normally distributed carcass weight (Additional file 1). The 1156 steers were the offspring of 347 sires and included at most 20 half-sibs.

The 1156 DNA samples were genotyped using the BovineSNP50 BeadChip (Illumina). The Btau4.0 assembly 25 was used to map the position of the SNP. Genotyping call rates were more than 95% in all samples. Among 54,001 SNP on the BovineSNP50 BeadChip, 39,011 SNP on autosomes fulfilled our quality control criteria: (1) call rate greater than 95%, (2) minor allele frequency greater than 0.01, and (3) p-value of a chi-square test for Hardy-Weinberg equilibrium greater than 0.001.

An association analysis was performed using EMMAX software 13 based on a linear mixed model with genomic related matrix. This software performs three-step analysis. First, the n × n genetic relatedness matrix of IBS between individuals is calculated using high-density SNP genotype data. Second, it uses a variance component model to estimate the restricted maximum likelihood parameters.

V a r Y = σ a 2 S ^ N + σ e 2 I

where Var(Y) is the variance of phenotype, σ_a² is the additive genetic variance, S ^ N is the n × n normalized genetic relatedness matrix, σ_e² is random environmental variance, and I is an identity matrix.

Third, a generalized least squares F-test for each marker was performed to detect associations accounting for the sample structure using the genetic relatedness matrix

y i = β 0 + β k X ik + η i V a r η = V ∝ σ a ^ 2 S ^ N + σ e ^ 2 I

where y_i is the phenotypic value of an individual iβ₀ is the fixed effect composed of age (month), slaughter year, and slaughterhouse to adjust the environmental effect, β_k is the effect size of marker kX_ik is the minor allele counts of marker k of individual i, and η_i is the error term that includes the genetic relatedness matrix. In a conditioned analysis, the genotypes of SNP associated with carcass weight were included as covariates. Haploview 26 was used to analyze LD among the SNP.

The genome screen was conducted using the microsatellite markers on the Shirakawa-USDA linkage map 27. QTL analyses were performed with the interval mapping method using a linear regression model for half-sib families 2829, as described previously 15. Linear regression analysis was performed using the following model:

y = X b + e ,

where y is the vector of phenotypic value, X is the design matrix of fixed effects composed of slaughter year, age (day), and probability of having the Q phase at a given location (Prob(Q)), b is the vector of fixed effects, and e is the residual error. b was estimated by the least squares method. An F-statistic value at each position was calculated from the residual sum square regressed with Prob(Q), and the total residual sum square without Prob(Q). The analysis was performed at 2-cM intervals along each chromosome. A threshold for significance of the F-statistic value was obtained by 10,000 random permutations of the phenotypic data 30. The 95% CI of the QTL locations was calculated using the bootstrapping method 31. Briefly, a set of offspring was chosen so as to be the same number as the original half-sibs by resampling from the original half-sibs. Resampling was repeated 10,000 times. The position of the F-statistic peak in each bootstrapping was collected. The CI was determined based on the distribution of the peaks. Therefore, the CI may be fragmented into separated regions, and not a single contiguous region.

The sires and a Q-homozygous steer were genotyped using a BovineHD BeadChip (Illumina) that includes BTA-52694-no-rs. The SNP Hapmap40466-BTA-82123 was genotyped by direct sequencing of the polymerase chain reaction (PCR) products using BigDye Terminator v.3.1 Cycle Sequencing Kit (Applied Biosystems) followed by electrophoresis using an ABI 3730 sequencer (Applied Biosystems).

Microsatellites were searched for in the genomic sequences and the primers were designed using Primer 3 32. Seventeen microsatellite markers from DIK7100 to DIK7116 were developed in this study. Marker information, such as primer sequences and genomic positions, is shown in Additional file 5. Genotyping was performed using PCR with a fluorescent-labeled reverse primer, followed by electrophoresis using ABI 3730 DNA analyzer (Applied Biosystems) and analysis using GeneMapper software (Applied Biosystems). The sires and their offspring were genotyped to determine the phase of the sires’ chromosomes. The haplotypes of the animals used for targeted resequencing are also shown in Additional file 5.

Linkage disequilibrium mapping was performed using 142 heavy carcass weight cattle and 145 light carcass weight cattle that were selected from the GWAS samples and did not include half-sibs in each group. A haplotype frequency comprising two contiguous microsatellites was estimated using an expectation-maximization algorithm in each group and the difference in haplotype frequencies between the groups was tested by Fisher’s exact test using a 2xn contingency table. The association of each haplotype of the three sires segregating the CW-1 QTL was also tested by Fisher’s exact test using a 2x2 contingency table. The frequency of q3 haplotypeAdditional file ( 5) was very low (< 6 alleles in both groups) in some regions, where an association of the haplotype was not tested.

Three sires segregating the QTL, three Q-homozygous sires, and a q-homozygous steer (Additional file 5) were subjected to sequence capture (NimbleGen custom array) followed by resequencing (Illumina GAIIx, 40-bp paired-end run). The Q-homozygous sires were determined as those that harbored at least one Q haplotype and homozygous Q alleles at Hapmap46986-BTA-34282 and did not segregate a carcass weight QTL on BTA 14 in the half-sib analyses using more than 255 offspring per family. The revised CW-1 region including 22.8-23.7 Mb on Btau4.0 and 24.5-25.5 Mb on UMD3.0 was targeted. Alignment and variant detection were performed using Illumina pipeline (CASAVA 1.7). As a reference genome assembly, UMD3.1 was used to identify variants in coding regions, while Btau4.0 was used to search variants in conserved segments across mammalian species. PhastCons scores were downloaded from the University of California, Santa Cruz web site 17. Obtained putative sequence variations were filtered by: (1) heterozygous in the three sires segregating the CW-1 QTL, (2) homozygous in the three Q- and the q-homozygous animals, and (3) opposite alleles between the Q- and the q-homozygous animals.

Coverage of the coding regions (UMD3.1) and highly conserved regions (PhastCons > 0.9, Btau4.0) was checked in each animal. The region between PLAG1 and CHCHD7 was not covered in every animal. Although the region was difficult to amplify by PCR, we detected the variations in heterozygous sires reported by Karim et al. 11. Ten nucleotides in the 3’UTR of TMEM68 were not covered only in the q-homozygous animal and were sequenced using an ABI 3730 DNA analyzer (Applied Biosystems). Other coding regions and highly conserved regions were covered in every animal. Regions with fewer than 4 reads in either of the animals were confirmed to have no sequence variations between the q-homozygous and a Q-homozygous animal.

Sequence variations were genotyped using the same population used for the LD mapping. The SNP were genotyped by direct sequencing of the PCR products, while the Indel and variable number of tandem repeat were amplified using a fluorescent-labeled reverse primer and genotyped in the same way as a microsatellite marker. The information, such as genomic positions, primer sequences, and alleles, is shown in Additional file 6. The genotypes were combined with the genotypes determined using a Bovine SNP50 BeadChip. Mixed-model association analysis was performed using EMMAX 13, adjusting for age, slaughterhouse, and year as covariate and fixed effects.

Genotypes of FJX_PLAPROTRI and NCAPG c.1326T > G, and a Q haplotype comprising Hapmap40466-BTA-82123 and BTA-52694-no-rs were used to estimate the effect size of CW-1CW2, and CW3, respectively. The 1156 GWAS samples were genotyped with FJX_PLAPROTRI (Additional file 6) and NCAPG c.1326T > G 3. A Q haplotype of CW-3 was determined using the sires segregating the QTL (Additional file 5) and diplotypes of the GWAS samples were estimated using fastPhase 33. The calculation was performed using EMMAX including a number of the Q alleles or haplotypes (0, 1, and 2) as X_ik in the model.