Identification of genomic regions that exhibit sexual dimorphism for size and muscularity in cattle

Abstract Sexual dimorphism, the phenomenon whereby males and females of the same species are distinctive in some aspect of appearance or size, has previously been documented in cattle for traits such as growth rate and carcass merit using a quantitative genetics approach. No previous study in cattle has attempted to document sexual dimorphism at a genome level; therefore, the objective of the present study was to determine whether genomic regions associated with size and muscularity in cattle exhibited signs of sexual dimorphism. Analyses were undertaken on 10 linear-type traits that describe the muscular and skeletal characteristics of both males and females of five beef cattle breeds: 1,444 Angus (AA), 6,433 Charolais (CH), 1,129 Hereford, 8,745 Limousin (LM), and 1,698 Simmental. Genome-wide association analyses were undertaken using imputed whole-genome sequence data for each sex separately by breed. For each single-nucleotide polymorphism (SNP) that was segregating in both sexes, the difference between the allele substitution effect sizes for each sex, in each breed separately, was calculated. Suggestively (P ≤ 1 × 10−5) sexually dimorphic SNPs that were segregating in both males and females were detected for all traits in all breeds, although the location of these SNPs differed by both trait and breed. Significantly (P ≤ 1 × 10−8) dimorphic SNPs were detected in just three traits in the AA, seven traits in the CH, and three traits in the LM. The vast majority of all segregating autosomal SNPs (86% in AA to 94% in LM) had the same minor allele in both males and females. Differences (P ≤ 0.05) in allele frequencies between the sexes were observed for between 36% (LM) and 66% (AA) of the total autosomal SNPs that were segregating in both sexes. Dimorphic SNPs were located within a number of genes related to muscularity and/or size including the NAB1, COL5A2, and IWS1 genes on BTA2 that are located close to, and thought to be co-inherited with, the MSTN gene. Overall, sexual dimorphism exists in cattle at the genome level, but it is not consistent by either trait or breed.


Introduction
Sexual dimorphism is the phenomenon whereby males and females of the same species are distinctive in behavior, size, or appearance (Berns, 2013).This is attributable to the combination of sex-specific genes on sex chromosomes, sexspecific expression of genes, and other regulatory mechanisms that are not yet widely understood (Pointer et al., 2013).Sexdependent differences have been documented for a whole range of traits in different species ranging from color, ornamentation, mating behavior, and size (McPherson and Chenoweth, 2012;Berns, 2013;van der Heide et al., 2016).An individual's sex is also known to have an influence on the growth of body tissues and could, therefore, affect carcass composition and weight distribution within the body tissue (Berg and Butterfield, 1976).Sexual size dimorphism is likely to have originated in mammals during evolution due to competition among males for access to females; males would fight one another to gain access to females and the winner, generally the bigger, stronger animal would mate with more females (Kirkpatrick, 1987;Katz, 2008).In selective breeding systems, breeding males are selected on numerous desirable traits and consequently competition for mates has been diminished in domesticated animals.Nonetheless, evidence of sexual dimorphism based on quantitative genetics approaches have been reported for several economically important traits in cattle, including growth rate (Koch and Clark, 1955;Marlowe and Gaines, 1958;van der Heide et al., 2016) and carcass traits (Crews and Kemp, 2001;Bittante et al., 2018).
Linear-type traits describing the muscular and skeletal characteristics of an animal are scored globally in both dairy (Veerkamp and Brotherstone, 1997;Berry et al., 2004) and beef (Mc Hugh et al., 2012;Mazza et al., 2014) cattle.These traits are typically considered as being genetically the same in both males and females; estimated genetic correlations of near unity between the same linear-type trait in different sexes of cattle substantiate this assumption (Doyle et al., 2018).Genetic correlations, however, are a manifestation of the cumulative effect of both linkage and pleiotropy across the entire genome and it is possible that the control of such traits by sex may differ in specific genomic locations.The objective, therefore, of the present study was to determine whether genomic regions associated with size and muscularity in cattle exhibited signs of sexual dimorphism.This knowledge will be useful in informing breeding programs of the potential improvement in accuracy achievable by evaluating males and females separately.

Materials and methods
Animal Care and Use Committee approval was not obtained for the present study as data were obtained from the existing Irish Cattle Breeding Federation (ICBF) national database (http:// www.icbf.com).

Phenotypic data
Linear-type traits are routinely scored in both registered and commercial beef herds by trained classifiers from the ICBF as part of the Irish national beef breeding program (Mc Hugh et al., 2012;Berry and Evans, 2014).The type traits used in the present study describe the muscular and skeletal development of the animal and include development of the hind quarter (DHQ), inner thigh (DIT), and loin (DL), thigh width (TW), wither width (WOW), wither height (WH), back length (BL), hip width (HW), and chest width (CW) and depth (CD).The five muscular traits were scored (Supplementary Table S1) on a scale of 1 (narrow) to 15 (wide), whereas the five skeletal traits (Supplementary Table S1) were scored on a scale of 1 (short or narrow) to 10 (long/ tall or wide).Data on these 10 traits were available on 147,704 purebred Angus (AA), Charolais (CH), Hereford (HE), Limousin (LM), and Simmental (SI) beef cattle scored between 6 and 16 mo of age between the years 2000 and 2016.
Data editing procedures and the justification for such edits are outlined in detail by Doyle et al. (2019Doyle et al. ( , 2020a, b), b).Animals were discarded from the data set if the sire, dam, herd, or classifier was unknown, or the parity of the dam was not recorded.Parity of the dam was subsequently recoded into 1, 2, 3, 4, and ≥5.Contemporary group was defined as herd-by-scoring date generated separately within each breed; each contemporary group had to have at least five records.Each of the 10 traits were separately standardized to a common variance within classifierby-year as described in detail by Brotherstone (1994).Following edits, data were available on 81,200 animals (Supplementary Table S2) consisting of 3,356 AA,31,049 CH,3,004 HE,35,159 LM,and 8,632 SI.

Generation of adjusted phenotypes
Prior to inclusion in the genome-wide association analysis, all phenotypes were adjusted within breed in ASREML (Gilmour et al., 2009) using the model: where y ijkl is the linear-type trait, μ is the overall mean, HSD i is the fixed effect of herd-by-scoring date (i = 11,130 levels), AM j is the fixed effect of the age in months of the animal (j = 11 classes from 6 to 16 mo), DP k is the fixed effect of the parity of the dam (k = 1, 2, 3, 4, and ≥5), Animal l is the random additive genetic effect of the animal where N(0, Aσ 2 a ), and e is the random residual effect where N(0, Aσ 2 e ); σ 2 a is the additive genetic variance, σ 2 e is the residual variance, A is the numerator relationship matrix, and I is an identity matrix.The adjusted phenotype used in the subsequent analysis was the raw phenotype less the fixed-effect solutions of HSD, AM, and DP.This approach of pre-adjustment of phenotypes was undertaken (as opposed to fitting directly in the association analysis) so as to generate a better estimate of especially the contemporary group effect that would have contained nongenotyped animals.

Genotype data
Of the phenotypic data set of 81 200 animals, 19 449 animals from the five beef breeds (Supplementary Table S2) were imputed to whole-genome sequence (WGS) as part of a larger data set of 638 662 multi-breed genotyped animals (Purfield et al., 2019).These 638 662 animals were genotyped using one of seven different genotype panels as described previously by Doyle et al. (2020a, b).The reference population used for imputation contained

BTA
Bos Taurus autosome GRM genomic relationship matrix HD high density MAF minor allele frequency PAR pseudoautosomal region QTL quantitative trait loci SNP single-nucleotide polymorphism WGS whole-genome sequence 90% male animals and 8% female animals; 2% of the reference population were of unknown sex.Each animal had to have a call rate ≥ 90% and only single-nucleotide polymorphisms (SNPs) with a known chromosome and position on UMD 3.1, and SNPs with a call rate ≥ 90% within the panel were retained for imputation.
All autosomes of genotyped animals were imputed to WGS following the steps outlined in Doyle et al. (2020a, b).Imputation of the pseudoautosomal region (PAR) and non-PAR regions of the X chromosome was undertaken separately.The non-PAR region was imputed for males and females separately.The PAR region of the X chromosome was defined from 143,861,798 to 148,823,899 bp (Mao et al., 2016).Regions of poor WGS imputation accuracy were discarded as described by Purfield et al. (2019).Furthermore, within each breed and each sex, all SNPs with a minor allele frequency (MAF) ≤ 0.002 were not considered further (Supplementary Table S2).The number of SNPs remaining for each sex in each breed is outlined in Supplementary Table S2.

Genome-wide association
Whole-genome association analyses were performed within each sex in each breed separately using a mixed linear model association analysis in GCTA (Yang et al., 2011).Autosomal SNPs from the original high-density (HD) panel (i.e., 734,159 SNPs) were used to construct the genomic relationship matrix (GRM) for each sex within each breed as per Doyle et al. (2020a, b) who used the data from the present study but in a combined analysis of both sexes.In the association analyses of the X chromosome, all males were coded as homozygous for the genotyped allele for SNPs in the non-PAR region, while heterozygous SNPs were accepted in the PAR region.The model used for the within-sex and within-breed analysis was as follows: where y is a vector of preadjusted phenotypes, µ is the overall mean, x is the vector of imputed genotypes, b is the vector of additive fixed effects of the candidate SNP to be tested for association, u ∼ N(0, Gσ 2 u ) is the vector of additive genetic effects, where G is the genomic relationship matrix calculated from the HD SNP genotypes, and σ 2 u is the additive genetic variance, and e ∼ N(0, Iσ 2 e ) is the vector of random residual effects, where I is the identity matrix and σ 2 e is the residual variance.

Dimorphism
For each SNP that was analyzed in both sexes (i.e., segregating in both sexes), the difference between the allele substitution effect sizes for each sex, in each breed separately, was calculated using a t-test: where b x is the allele substitution effect in males (m) and females (f), SE is the estimated standard error of the allele effect, and n is the respective sample size.The presence of dimorphism was determined at each SNP based on the calculated P-value from the t-test statistic.An SNP with a P-value ≤ 1 × 10 −5 was assumed to have a suggestively different allele effect in the two sexes, whereas an SNP with a P-value ≤ 1 × 10 −8 was assumed to have a significantly different allele substitution effect in the two sexes.These levels of significance are generally consistent with recommended levels (Pe'er et al., 2008).Nonetheless, an additional test was undertaken in the CH and LM breeds where two populations were generated: the first population included half the males and half the females combined with the second population consisting of the remainder.Association analyses were undertaken in each population and the allele effects compared statistically to mimic the approach used in the present study to detect dimorphism.The significance of the difference in allele substation effect was not <1 × 10 −8 , thus providing confidence in this threshold used in this study.

Quantitative trait loci detection
To identify quantitative trait loci (QTL) regions that were dimorphic in more than one trait or more than one breed, each chromosome was split into 1 kb genomic windows and windows containing at least one suggestive (P ≤ 1 × 10 −5 ) or significant (P ≤ 1 × 10 −8 ) SNP were compared across the traits and breeds.

Results
The scale of measurement, mean, and standard deviation of the linear-type traits in each sex in each breed is in Supplementary Table S1.Single-nucleotide polymorphisms with evidence of significant (P ≤ 1 × 10 −8 ) dimorphism were detected for some traits, while suggestively (P ≤ 1 × 10 −5 ) dimorphic SNP were detected for all of the traits in all five breeds; however, these SNPs differed both by trait and by breed.

Angus
A total of 16,541,913 SNPs were segregating in the 1,044 males and 15,402,160 SNPs were segregating in the 400 females.Of these totals, 15,008,408 SNPs were segregating in both the male and female populations (Supplementary Table S2).Significant dimorphism (P ≤ 1 × 10 −8 ) was evident for a total of seven SNPs across just three traits (HW, TW, and DIT; Table 1), whereas suggestive dimorphism (P ≤ 1 × 10 −5 ) was evident for between 31 (DHQ) and 1,254 (HW) SNPs depending on the trait (Table 1).
In general, the allele substitution effects of the dimorphic SNPs tended to be in opposite directions in each sex (i.e., if the allele effect in the male population was negative, then the allele effect of the same allele in the female population was positive or vice versa; Tables 2and 4).The allele effects in the male population also tended to be closer to zero than those in the female population (Table 2 and 4) and the most significantly dimorphic traits tended to have a very low MAF in the female population.
Of the muscular traits investigated, the most significantly dimorphic SNP was an intronic SNP (P = 3.45 × 10 −9 ) located within the ADGRA3 gene on Bos Taurus autosome 6 (BTA6) and was associated with DIT (Table 2); this SNP had an allele effect of +0.12 (SE = 0.13) and an MAF of 0.026 in males but an allele effect of −3.68 (SE = 0.63) and an MAF of 0.003 in females.Of the skeletal traits, the most significantly dimorphic SNP was an intergenic SNP, rs208222963 (P = 9.86 × 10 -10 ), located on BTA8 that was associated with HW (Table 3); this SNP had an allele substitution effect of −0.23 (SE = 0.10) in males but +2.29 (SE = 0.40) in females with an MAF of 0.025 in the males and 0.005 in the females.
Of the 1-kb windows containing at least one suggestively associated dimorphic SNP, there was little overlap between the muscular and skeletal groups of traits.The only overlap in windows between the muscular and skeletal traits was between HW, DL and WOW (two windows on each of BTA23 located at 14.555 and 51.713 Mb), between HW and TW (one window on BTA8 located at 66.264 Mb), and between HW and DL (two windows on BTA27 at 18.141 and 18.403 Mb).Only one 1-kb genomic window was suggestively associated with three skeletal traits (WH, BL, and HW; Supplementary Figure S1a), and this was located between 66.386 and 66.387 Mb on BTA8, within the ENSBTAG00000006446 gene.The greatest overlap across traits in AA was between WH and HW where a total of twelve 1 kb windows across five chromosomes suggestively exhibited sexual dimorphism (Supplementary Figure S1a).Similar to the skeletal traits, only one 1 kb window was common to more than two muscular traits (DL, DIT, and TW) and this was located between 59.524 and 59.525 Mb on BTA24 (Supplementary Figure S2a).The largest overlap across all skeletal traits was between DL and TW where seven 1-kb windows exhibited suggestive dimorphism.Minimal overlap was detected among the remaining skeletal traits

Charolais
A total of 18,054,274 SNPs were segregating in the 4,641 CH males and 17,448,948 SNPs were segregating in the 1,792 CH females.Of these, 17,227,625 SNPs were segregating in both the male and female animals.Evidence of suggestive dimorphism (P ≤ 1 × 10 −5 ) existed for between 51 (DIT) and 3,051 (CW) SNPs depending on the trait (Table 1), whereas evidence of significant dimorphism (P ≤ 1 × 10 −8 ) was evident in all but three traits (i.e., DIT, HW, and WH).Of the muscular traits, the most significantly dimorphic SNP was rs110487743 (P = 1.36 × 10 −10 ), an intronic SNP located within the NAB1 gene on BTA2 which exhibited dimorphic associations for DL (Table 4).Of the skeletal traits, the most significantly dimorphic SNP was for CW and was an intergenic SNP, rs446294174 (P = 4.44 × 10 −16 ; Table 5) that had an allele effect of −0.03 (SE = 0.18) in males but −3.34 (SE = 0.36) in females with an MAF of 0.015 in males and 0.008 in females.
Of the 1-kb windows containing at least one dimorphic SNP, no windows were shared between the skeletal and muscular trait groups.Despite the lack of overlap between the skeletal and muscular traits in the CH, considerable dimorphism was detected across the muscular traits.Across trait dimorphism was detected in four of the five muscular traits (i.e., DHQ, DL, TW, and WOW; Supplementary Figure S2b) where eight 1-kb windows in common between 5.54 and 5.60 Mb on BTA2 contained a suggestively associated dimorphic SNP; only one gene, NAB1, was located within this region.An additional 22 1-kb windows on BTA2 were also deemed to exhibit across trait dimorphism for the muscular traits (Supplementary Figure S2b).Compared with the muscular traits, fewer windows containing a suggestive SNP were common among the skeletal traits.Seven 1-kb windows were common to CW and CD (Supplementary Figure S1b), one window on each of BTA4, BTA5, BTA12, and BTAX, and three windows on BTA13 that contained the BTBD3 gene.A single window on BTA15 at 72.31 Mb was common to both WH and BL.

Hereford
A total of 17,241,152 SNPs were segregating in the 727 HE males and 16,494,904 SNPs were segregating in the 402 HE females.Of these, 15,991,751 SNPs were segregating in both the male and female animals.In comparison to the AA and CH, evidence of suggestive dimorphism (P ≤ 1 × 10 −5 ) was evident for fewer SNPs in all of the traits, ranging from 39 (DIT) to 256 (CW) SNPs depending on the trait; there was no evidence of significant dimorphism (P ≤ 1 × 10 −8 ; Table 1).Similar to the AA, the allele substitution effect of the dimorphic SNPs in the males tended to be in the opposite direction to the allele effect of the same SNP in the females (Table 6and 8).The most significantly dimorphic SNP in the muscular traits was rs798960299 (P = 6.30× 10 −8 ), an intronic variant located within the NR5A2 gene on BTA16 with an allele effect of +0.42 (SE = 0.13) in males but −0.61 (SE = 0.14) in females (Table 6).Of the skeletal traits, the most significantly dimorphic SNP was rs381085044, an intergenic SNP on BTA1 that had a dimorphic association with CW represented by an allele effect of −0.22 (SE = 0.06) in males but +0.32 (SE = 0.08) in females (Table 7).
No 1-kb window that contained at least one dimorphic SNP was common to more than two traits.Limited dimorphism was found between CW and HW, where three windows between 40.75 and 40.78 Mb on BTA24 containing the PTPRM gene were suggestively associated with both traits (Supplementary Figure S1c).CW also had two separate windows exhibiting dimorphism on BTA13 between 27.46 and 27.47 Mb in common with CD, and one window on BTA14 at 45.485 Mb in common with WH.Two adjacent 1-kb windows on BTA8 at 25.965 Mb were common to both WH and BL.For the muscular traits, WOW had one window in common with each of DL (BTA10 at 48.327 Mb), TW (BTA10 at 50.420 Mb), and DHQ (BTA9 at 83.332 Mb).One 1-kb window was also common to both TW and DHQ (BTA16 at 68.580 Mb; Supplementary Figure S2c).

Limousin
A total of 18,056,913 SNPs were segregating in the LM males and 17,767,237 SNPs were segregating in the LM females.Of these, 17,482,131 SNPs were segregating in both the male and female animals.Between 46 (TW) and 1105 (CW) SNPs were suggestively dimorphic (P ≤ 1 × 10 −5 ), whereas three traits (i.e., CW, CD, and DHQ) had evidence of significant dimorphism (P ≤ 1 × 10 −8 ; Table 1).Of the muscular traits, the most significantly dimorphic SNP was rs42425148 (P = 1.85 × 10 −9 ), an intergenic SNP located on BTA1 that had a dimorphic association with DHQ (Table 8) represented by an allele effect of −0.01 (SE = 0.15) in males but −1.64 (SE = 0.23) in females.The most significantly dimorphic SNP for the skeletal traits was also an intergenic SNP, rs478688690 (P = 3.63 × 10 −11 ), located on BTA11, that had a dimorphic association with CW (Table 9), with an allele effect of +0.03 (SE = 0.29) in males and −3.86 (SE = 0.51) in females; this SNP had an MAF of 0.005 in males and 0.003 in females.Similar to the AA and CH, SNPs with a higher MAF tended to have an allele effect that was closer to zero.An intergenic SNP with dimorphic associations in HW (P = 5.71 × 10 −7 ) had an MAF of 0.493 in males and 0.486 in females, but the allele substitution effects were just −0.07 (SE = 0.01) in males and +0.06 (SE = 0.02) in females.Genomic regions that exhibited sexual dimorphism across traits in the LM were limited; eight 1-kb windows were found to be suggestively associated with three of the skeletal traits (WH, BL, HW; Supplementary Figure S1d), whereas only three windows were common between TW and WOW of the muscular traits (Supplementary Figure S2d).All eight windows associated with WH, BL, and HW were on BTA6, but no obvious candidate gene was located in the vicinity.The 40 windows suggestively associated with both CW and CD (Supplementary Figure S1d) were located on 15 different autosomes: BTA5, BTA6, BTA8, BTA9, BTA10, BTA11, BTA13, BTA16, BTA17, BTA18, BTA20, BTA22, BTA24, BTA26, BTA29, and BTAX.

Simmental
A total of 18,257,175 SNPs were segregating in the SI males, and 17,814,297 SNPs were segregating in the SI females, whereas 17,319,250 of these SNPs were segregating in both the males and females.Between 87 (BL) to 679 (CD) SNPs were suggestively dimorphic (P ≤ 1 × 10 −5 ), whereas no SNP was significantly dimorphic (P ≤ 1 × 10 −8 ; Table 1).Once again, the most significantly dimorphic SNPs tended to have a low MAF and a large allele effect size.The most significantly dimorphic SNP associated with any of the muscular traits was rs110995439 (P = 1.40 × 10 −8 ), an intron variant located within the GPC5 gene on BTA12 that had a dimorphic association with DIT; the allele substitution effect in the males was −0.85 (SE = 0.34), whereas the allele substitution effect in the females was +1.76 (SE = 0.30; Table 10).Of the skeletal traits, the most significantly dimorphic SNP was an intergenic SNP, rs437227524 (P = 1.98 × 10 −8 ), that had a dimorphic association with CD (Table 11) and had an allele substitution effect of −0.08 (SE = 0.24) in the male population but −2.64 (SE = 0.42) in the female population with an MAF of 0.004 in the males and 0.002 in the females.An intronic SNP, rs133629874 (P = 2.20 × 10 −7 ), located within the MMRN1 that had a dimorphic association with CW had an MAF of 0.439 in males and 0.472 in females with an effect size of +0.14 (SE = 0.03) in males and −0.13 (SE = 0.04) in females.
Few 1-kb windows containing a suggestive SNP overlapped among the muscular and skeletal traits.A single 1-kb window on BTA2, approximately 0.1 Mb from the IWS1 gene, contained suggestively dimorphic SNPs for all of DIT, DL, and CW.Of the skeletal traits, no genomic windows exhibited suggestive associated dimorphism (Supplementary Figure S1e), and no window was suggestively associated with three or more muscular traits (Supplementary Figure S2e).Three 1-kb windows, all located on BTA18 between 3.78 and 3.80 Mb, contained dimorphic SNPs for both TW and WOW (Supplementary Figure S2e); these windows were located approximately 0.3 Mb from the CNTNAP4 gene.One window located on BTA28 at 41.045 Mb, and six windows on the X chromosome between 114.572 and 114.578Mb were dimorphic for both TW and DL.A single 1-kb window was common to each of DIT and TW (13.541Mb on BTA8), DHQ and WOW (102.906Mb on BTA6), and WH and BL (145.410Mb on BTAX).

Across breed
The numerically smaller breeds of AA, HE, and SI had approximately 1.5 times more SNPs segregating in only one sex than the numerically larger breeds of CH and LM.Of the total autosomal SNPs that were segregating in both sexes, between 86% (AA) and 94% (LM) had the same minor alleles in both sexes.However, differences (P ≤ 0.05) in allele frequencies between the sexes were observed for between 36% (LM) and 66% (AA) of the total autosomal SNPs that were segregating in both sexes.
The vast majority of SNPs and 1-kb windows associated with any one trait were breed specific.The most windows displaying dimorphic characteristics in common for more than one breed was for CW in both the CH and LM with nine 1-kb windows common to these breeds occurring on BTA2 at 89.527 Mb (n = 1), on BTA12 between 87.128 and 87.303 Mb and containing the ENSBTAG00000032038 gene (n = 6), and on BTA24 between 31.661 and 31.671Mb (n = 2).Also for CW, a single 1-kb window on BTA6 approximately 0.1 Mb from the SLC34A2 gene displayed dimorphic associations in both the AA and the CH.A single 1-kb window containing the ENSBTAG00000046311 gene on BTA10 had dimorphic associations with HW in both the AA and SI.In the CH and the SI, two common windows, one at 91.126 Mb on BTA7 and one at 70.827 Mb on BTA9, exhibited dimorphism with CD in both breeds.

Discussion
Although several studies in cattle have investigated the presence of sexual dimorphism using a quantitative genetics approach (Crews and Kemp, 2001;van der Heide et al., 2016;Bittante et al., 2018;Doyle et al., 2018), no previous study in cattle has attempted to detect evidence of sexual dimorphism at the genome level or to compare these effects across multiple breeds of cattle.From previous quantitative genetics studies, differences in genetic parameters by cattle sex have been observed for growth rate (Koch and Clark, 1955;Marlowe andGaines, 1958), postweaning gain (van der Heide et al., 2016), feed intake and efficiency, fleshiness scores, carcass weight and yield (Bittante et al., 2018), as well as longissimus muscle area and backfat (Crews and Kemp, 2001).In contrast, negligible differences in genetic parameters between the sexes were detected for early growth traits such as weaning weight (Koch and Clark, 1955;van der Heide et al., 2016).
A single trait measured in different environments (or different sexes) can be regarded as separate traits which are genetically correlated (Falconer, 1952).In many situations, the genetic correlations of the same trait taken in different environments are less than unity, indicating that selection   and females having different fitness optima for a phenotype (Rice, 1984;Lucotte et al., 2016).In cattle, differences in allele frequencies between the sexes may also have arisen due to these differential fitness optima for males and females where selection occurred at a given locus in one sex but not the other or due to different selection pressures at that locus in each sex (Lucotte et al., 2016); for example, selection for a female trait could have very different effects on genetically correlated traits in females compared with those traits in males (Bittante et al., 2018).Intersex differences in the frequency of a given allele may also be due to intersex differences in recombination rates; up to 75% of species that undergo recombination in their genome have different recombination rates per sex (Burt et al., 1991;Wyman and Wyman, 2013).In the majority of species, the male recombination rate is generally lower than in females (Poissant et al., 2010;Wyman and Wyman, 2013).It is thought that this lower recombination rate is advantageous to males as it maintains combinations of beneficial genes that have undergone sexual selection (Trivers, 1988); however, studies in both cattle (Ma et al., 2015) and sheep (Maddox and Cockett, 2007) reported that the male recombination rate in these species is actually higher than the females.

Across breed
Based on a series of within-breed genome-wide associations undertaken across both sexes combined using the data from the present study, Doyle et al. (2020a, b) detected little to no commonality in the genomic regions associated with each type trait across breeds.This indicates that the underlying genetic basis of the same trait in each breed is quite different; therefore, it was somewhat expected that the regions exhibiting dimorphism may also differ by breed.In general, the British breeds (AA and HE) had fewer suggestively or significantly dimorphic SNPs than the continental breeds but the British breeds had a greater percentage of SNPs exhibiting significant differences in allele frequencies between the sexes than the continental breeds.The location of the most significantly dimorphic SNPs also differed across the breeds.The differences observed between the breeds may be due to actual differences in the genetic basis of sexual dimorphism among the breeds, as previously observed between the Belgian Blue and Piemontese cattle breeds (Bittante et al., 2018), or may be simply due to differences in the statistical power to detect QTL due to the differences in breed-specific population sizes (Doyle et al., 2020a).Actual differences in the genetics underlying each trait may be attributable to different mutations affecting specific genes in each breed, such as the breed-specific MSTN mutations, or may, more likely be attributable to different QTL being affected by different selection pressures within each breed (Bittante et al., 2018).

Genes exhibiting dimorphism
SNPs exhibiting sexually dimorphism were located within or close to a number of different genes that have previously been associated with muscularity and/or size in cattle (Doyle et al., 2020a, b).Three genes on BTA2 (NAB1, COL5A2, and IWS1) containing dimorphic SNPs associated with multiple traits are thought to either be in strong linkage disequilibrium with MSTN (Grade et al., 2009) or have been previously identified as being located within a QTL also containing MSTN that was associated with muscularity in beef cattle (Doyle et al., 2020a).The MSTN gene has already been documented as being responsible for muscular hypertrophy in cattle (Grobet et al., 1997;McPherron and Lee, 1997) and is widely known as the causal variant for many muscularity and carcass traits in cattle (Casas et al., 2000;Allais et al., 2010;Purfield et al., 2019).Another candidate gene for muscularity that exhibited evidence of sexual dimorphism was the PDHX gene on BTA15 that contained dimorphic SNPs for 3 of the muscularity traits in CH and has previously been associated with carcass quality traits in beef cattle (Karisa et al., 2013).

Conclusion
Although many significantly and suggestively sexually dimorphic SNPs associated with the muscular and skeletal type traits were identified in the present study, the location and effect sizes of these tended to be both trait specific and breed specific.Both the allele substitution effect sizes and the allele frequencies of the dimorphic SNPs also differed by sex.This indicates that although sexual dimorphism exists in cattle at a genome level, it occurs at a low frequency but also differs both by trait and by breed.

Table 1 .
The number of suggestively dimorphic (P ≤ 1 × 10 −5 ) and significantly dimorphic (P ≤ 1 × 10 −8 ; in parenthesis) SNPs for each trait in each breed 1 CW, chest width; BL, back length; HW, hip width; WH, withe height; DHQ, development of hind quarter; DIT development of inner thigh; DL, development of loin; TW, thigh width; WOW, width of withers.Where there is no parenthesis, no significantly dimorphic SNP was detected.

Table 2 .
The location of the most significantly dimorphic SNPs, limited to the top 5, which were associated with the muscular traits in the AA 1 DHQ, development of hind quarter; DIT, development of inner thigh; DL, development of loin; TW, thigh width; WOW, width of withers.a Intergenic variant.b Intron variant.c Downstream gene variant.d Upstream gene variant.

Table 3 .
The location of the most significantly dimorphic SNPs, limited to the top 5, which were associated with the skeletal traits in the AA 1 WH, wither height; BL, back length; HW, hip width; CW, chest width; CD, chest depth.a Intergenic variant.b Intron variant.c Missense variant.d 3′ UTR variant.

Table 4 .
The location of the most significantly dimorphic SNPs, limited to the top 5, which were associated with the muscular traits in the CH 1 DHQ, development of hind quarter; DIT, development of inner thigh; DL, development of loin; TW, thigh width; WOW, width of withers.a Intergenic variant.b Intron variant.c Downstream gene variant.

Table 5 .
The location of the most significantly dimorphic SNPs, limited to the top 5, which were associated with the skeletal traits in the CH 1 WH, wither height; BL, back length; HW, hip width; CW, chest width; CD, chest depth.a Intergenic variant.b Intron variant.c Downstream gene variant.d Upstream gene variant.*SNP has not been mapped on the latest build

Table 6 .
The location of the most significantly dimorphic SNPs, limited to the top 5, which were associated with the muscular traits in the HE 1 DHQ, development of hind quarter; DIT, development of inner thigh; DL, development of loin; TW, thigh width; WOW, width of withers.a Intergenic variant.b Intron variant.c Downstream gene variant.d Upstream gene variant.*SNP has not been mapped on the latest build.

Table 7 .
The location of the most significantly dimorphic SNPs, limited to the top 5, which were associated with the skeletal traits in the HE a Intergenic variant.b Intron variant.c Downstream gene variant.d Upstream gene variant.*SNP has not been mapped on the latest build.

Table 8 .
The location of the most significantly dimorphic SNPs, limited to the top 5, which were associated with the muscular traits in the LM 1 DHQ, development of hind quarter; DIT, development of inner thigh; DL, development of loin; TW, thigh width; WOW, width of withers.a Intergenic variant.b Intron variant.c Upstream gene variant.*SNP has not been mapped on the latest build.

Table 9 .
The location of the most significantly dimorphic SNPs, limited to the top 5, which were associated with the skeletal traits in the LM a Intergenic variant.b Intron variant.c Downstream gene variant.*SNP has not been mapped on the latest build.

Table 10 .
The location of the most significantly dimorphic SNPs, limited to the top 5, which were associated with the muscular traits in the SI a Intergenic variant.b Intron variant.c Upstream gene variant.*SNP has not been mapped on the latest build.

Table 11 .
The location of the most significantly dimorphic SNPs, limited to the top 5, which were associated with the skeletal traits in the SI a Intergenic variant.b Intron variant.c Downstream gene variant.d Upstream gene variant.*SNP has not been mapped on the latest build.