A set of 15 sires (Table 1) was used for re-sequencing to scan for loss of function mutations within the 1128 bp of translated sequence encoding the three GDF8 exons. No indels, synonymous or non-synonymous substitutions were detected, consistent with a single previous report for ovine GDF8 11 . A further 1860 bp of the gene was investigated to search for putative regulatory mutations and SNP suitable for association analysis. Re-sequencing regions of the promoter (283 bp), intron 1 (480 bp), intron 2 (695 bp) and the 3' untranslated region (402 bp; UTR) revealed a total of 6 SNP. None appear to be located within splice sites which would interfere with transcriptional activity. Four have been identified previously and are located in the promoter (g-41C>A), intron 1 (g+391G>T), intron 2 (g+4036A>C) and the 3'UTR (g+6723G>A) 10 . The remaining two appear novel and are located within intron 1 (g+2154C>T) and the 3'UTR (g+6871G>A). The genotypic status, using IUB convention, at each polymorphism (ordered by numerical position) within the 15 sires (S1 – S15) was as follows: S1 CGCAGG, S2 CGCARG, S3 MGCAGG, S4 CGCAGG, S5 MGCAGG, S6 MGCAGG, S7 CGCAGG, S8 CGCAGG, S9 MGCAGG, S10 MKCMGG, S11 CGCARG, S12 MKCMGG, S13 MKCAGR, S14 CGCARG, S15 MKYMGG.

Three SNP were selected for analysis within a larger set of animals (g-41C>A, g+4036A>C and g+6723G>A). The remaining SNP (g+391G>T, g+2154C>T and g+6871G>A) were not tested either due to low informativeness within sires 1 – 15 or unsuitability of assay design. Genotyping 326 animals from 12 breeds revealed that two SNP were polymorphic within every breed tested (g-41C>A and g+4036A>C, Table 2). The third SNP has been identified to cause muscle hypertrophy within Belgian Texel (g+6723G>A) 10 . Consistent with this finding, the loss of function allele was present in the homozygous form in 109 of the 116 Australian Texels tested (g+6723A p = 0.946, Table 2). This included sires sampled from 12 different producers and indicates the mutant allele is near fixation within commercial Australian Texel flocks. The allele was also present in breeds other than Texel. Testing of 77 White Suffolks, including 65 sires from 6 producers, revealed an allele frequency approaching 10% (p = 0.093, Table 2) while three Poll Dorset and one Lincoln were identified carrying g+6723A in the heterozygous form.

Two microsatellites expected to flank GDF8 (BM81124 and BULGE20) 8 and three SNP (g-41C>A, g+4036A>C and g+6723G>A) were used to assess haplotype diversity. Genotyping was performed using 15 Poll Dorset, White Suffolk or mixed breed sires (S1 – S15, Table 1) and their 1191 half-sib progeny. Both microsatellites were highly polymorphic with seven (BULGE20) and ten (BM81124) alleles observed within the 30 sire chromosomes tested. Marker segregation within family was used to assign allelic phase and construct five-marker haplotypes across the GDF8 locus (Table 1). A total of 20 different haplotypes were observed and none of the sires was homozygous indicating high levels of diversity. Table 1 shows that three sires were detected which carry the loss of function allele (g+6723A; sires S2, S11, S14). Inspection of the surrounding phase-known alleles at the other four markers revealed that in each case, g+6723A was observed on a single marker haplotype (BM81124 allele 218, g-41C, g+4036A, BULGE20 allele 141, Hap ID 4, Table 1).

The breed history of each sire (S1 – S15) was determined from five generation pedigrees. Poll Dorset, Texel, Suffolk and White Suffolk ancestors were identified and the number of each is recorded in Table 1. Three of the 15 sires have a Texel ancestor (S2, S11 and S14). In each case this corresponded with the presence of g+6723A, suggesting Texel as the likely origin of the mutant allele. This is consistent with both the high frequency of g+6723A observed within Australian Texels (Table 2) and its identification in Belgian Texel 10 . Assuming Texel ancestry is a prerequisite for presence of the allele, Sheep Genetic Australia's phenotypic and pedigree database 12 was interrogated to determine the prevalence of Texel ancestors across nine major breeds. Five generation pedigrees were constructed for 5116 progeny tested sires and a total of 590 had at least one Texel ancestor (Table 3). Coopworth had the largest proportion of sires with Texel ancestors (69/130; 0.531) followed by White Suffolk (204/1853; 0.11) while three breeds (Border Leicester, Dorper and Merino) contained no recorded Texel relatives (Table 3).

Eleven of the 15 resource sires were heterozygous for at least one of the three GDF8 SNP (g-41C>A, g+4036A>C and g+6723G>A, Table 1). This allowed linkage analysis to be performed using data collected on their half-sib progeny. Each SNP was genotyped across all 15 families (1191 progeny) before regression analysis was performed using 50 phenotypic traits (Table 4). Table 5 shows the majority of significant marker by trait associations were detected for g+6723G>A. The mutation had significant effects on muscling (P < 0.05 for loinwt, emac, emdc, emwc and forequarter) highly significant effects on fatness (P < 0.01 for cfatc and imf) and significant effects on subjective measures of eating quality (P < 0.05 for flavour, tender and satisfaction). The mutant allele (g+6723A) was associated with increased muscling, decreased fatness and a decrease in perceived eating quality (Table 5) but had no detected effect on shear force, ultrasonically derived muscling and fat depth, live weight or yield. Significant effects were detected for g-41C>A on forequarter weight (p < 0.05), meat tenderness (P < 0.05 for two shear measurements avkgf0 and diff072) and for +4036A>C on muscling (P < 0.05 for emwc).

To search for evidence of function alleles other than g+6723A, microsatellites BM81124 and BULGE20 were genotyped across all 15 half-sib families. Haplotype effects were estimated for a set of five widely measured live-animal and carcass traits using both across-family and within-family analysis (Table 6). The microsatellite haplotype carrying g+6723A (Hap4, Table 1) showed significant across-family effects (Table 6) for the same set of traits identified by single marker analysis (Table 5). Comparison of Hap4 effects between sires (within-family analysis) revealed inconsistent effects for every trait. Six haplotypes which do not contain g+6723A showed significant effects. Two of these (Hap3 and 18, Table 6) were present in sires which carry Hap4 and their associated within-family effects arise from the indirect impact of g+6723A. The remaining four haplotypes, however, were identified in wildtype sires and suggest additional and currently unidentified functional alleles may be present at the ovine GDF8 locus. Two haplotypes showed effects on ultrasound eye muscle depth (emd1; Hap2 and Hap6), one on slaughter measurements of fat and muscle (grknf; Hap1) and one on eye muscle area (Hap13, Table 6). In nearly every case, within-family analysis revealed inconsistent associations between sires which carried the same haplotype (Table 6).

Six primer sets were designed from available ovine sequences [GenBank: AY918121, AF393618, AF266758] to span the three exons of the sheep GDF8 gene (primers MSTN1 - 6, Table 7). A seventh primer pair was designed from available cattle sequence (AF320998) to amplify a portion of the 3' untranslated region (MSTN7F/R, Table 7). Primer sets were used to amplify and sequence fragments from 15 sires (S1 – S15, Table 1). PCR was conducted in 20 μl reactions containing 50 ng of template DNA, 0.5 μM of both forward and reverse primer, 1.5 mM MgCl₂, 200 μM dNTP, 0.5 U HotStarTaq™ DNA Polymerase and reaction buffer (Qiagen, Victoria, Australia). Amplified products were directly sequenced in both forward and reverse orientation using ABI Prism™ Big Dye Terminator 3.2 chemistry and the ABI 3130xl Genetic Analyser (Applied Biosystems, Victoria, Australia). Sequence was trimmed and aligned using Sequencher version 4.2 (Gene Codes Corporation, Ann Arbor, MI, USA). SNP were named in relation to the first base in the ATG translation start site, as given in GenBank accession DQ530260 10 .

SNP detection was performed using the fluorogenic 5' nuclease assay formatted for analysis using Applied Biosystems (ABI) 7900HT 'TaqMan' sequence detection system (primer and probe sequences are recorded in Table 7 using prefix 'g'). Reactions were conducted in 10 μl containing 2 ng of genomic DNA, 2.25 μM of each PCR primer, 0.8 pM of each TaqMan probe and 5 μl of TaqMan Universal Master Mix (ABI). Cycling was performed as follows: 2 min at 50°C and 10 min at 95°C; 45 cycles of 95°C for 30 sec and 60°C for 60 sec. End point allele discrimination analysis was performed using the Sequence Detection System (SDS) software version 2.2.2 (ABI). Primers for amplification of microsatellite markers BM81124 and BULGE20 (Table 7) were synthesized with fluorescent tags NED and VIC respectively. PCR was performed as described for exon re-sequencing before fragment sizes were determined using the ABI 3130XL and associated GeneMapper software version 3.7 (Applied Biosytems).

Allele frequencies at three GDF8 SNP were determined within each of 12 breeds (Table 2). The DNA samples used have been described previously 17 . Additional population samples from the Australian Texel (n = 111) and White Suffolk breeds (n = 65) were collected from commercial sheep farmers as part of this work. Blood was collected on Guthrie cards before DNA extraction and whole genome amplification was performed using the 'Genomiphi' DNA kit according to the manufacturers instructions (Amersham Biosciences). Genotyping of all animals was performed using 'TaqMan' SNP assays described above.

Pedigree records from 5116 progeny tested sires were obtained from Sheep Genetics Australia (SGA) which provides a national genetic information and evaluation service to the sheep industry 12 . Each pedigree was searched for the presence of Texel ancestors (Table 3).

Fifteen half sib families were generated using Poll Dorset, White Suffolk or mixed breed sires (the breed composition of each sire is presented in Table 1). The sires were drawn from Australian industry flocks and mated with Merino ewes. A total of 1566 progeny were born in 2000 and 2001 at the Struan Research Center, South Australia. The description, code, trait mean and units of measurement for each trait are given in Table 4. Live weight was recorded at birth (bwt) and at approximately 30 day intervals (prewwt1, prewwt2, pwwt1 - 8) until a final weight was recorded at day 300 (flnwt). Ultrasound measurements of fat and eye muscle depth were recorded at approximately day 210 (cfat1, emd1) and day 260 (cfat2, emd2). A total of 1224 progeny were slaughtered at approximately 300 days of age and measured for hot carcass weight, eye muscle width, eye muscle depth, eye muscle area, fat depth at the C site, loin weight, lean meat yield as measured by Viascan and fat and muscle depth at the GR site. A subset of carcasses (350) was dissected into their component parts in a boning room to generate the yield of lean meat, weight of bone, weight of trimmed fat, loin fat weight, weight of boned forequarter, weight of hind leg and weight of remaining meat including tenderloin and shank. Objective measurements of eating quality included mean Warner Bratzler shear force at zero, 72 hrs and 5 days post slaughter 18 , total glycogen at slaughter 19 , total lactate, pH decline, ultimate pH, lightness of the loin measured by chromameter and intramuscular fat content predicted by near infrared spectrophotometry following calibration for sheep 20 . Sensory measurements of subjective eating quality traits were collected on samples from 528 progeny using methods described elsewhere 21 . Ten consumers rated liking of smell, tenderness, juiciness, flavour and hedonic overall liking using a range of 1–100. Satisfaction was ranked using a range of 1 – 5. For each trait, higher scores indicate favourable sensory perception and the averaged score from all ten consumers was used. A composite score was calculated which comprised 0.2.Tender + 0.1.Juicy + 0.3.Flavour + 0.4.Ovlike.

Regression analysis of phenotypes on single marker allele variants was performed, fitting three SNP simultaneously (g-41C>A, g+4036A>C and g+6723G>A, Table 5). The model was:

y i = μ + s i + ∑ k = 1 n m a r ker ⁡ s b k g i j k + e i MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGacaGaaiaabeqaaeqabiWaaaGcbaGaemyEaK3aaSbaaSqaaiabdMgaPbqabaGccqGH9aqpiiGacqWF8oqBcqGHRaWkcqWGZbWCdaWgaaWcbaGaemyAaKgabeaakiabgUcaRmaaqahabaGaemOyai2aaSbaaSqaaiabdUgaRbqabaGccqWGNbWzdaWgaaWcbaGaemyAaKMaemOAaOMaem4AaSgabeaakiabgUcaRiabdwgaLnaaBaaaleaacqWGPbqAaeqaaaqaaiabdUgaRjabg2da9iabigdaXaqaaiabd6gaUjabd2gaTjabdggaHjabdkhaYjGbcUgaRjabcwgaLjabckhaYjabdohaZbqdcqGHris5aaaa@538F@

where y_ij is the phenotype on animal j from sire i, s_i is the polygenic effect of the sire and g_ijk is the paternal allele code (0 or 1) for the marker allele that animal j received from sire i for the k^th marker, and b is the regression coefficient (estimated marker effect) of marker k. Phenotypes were pre-adjusted for relevant fixed effects (birth year, management group, birth type, and sex) using an additive linear model. Slaughter traits were also corrected for effect of batch (9 slaughter batches over 2 years) and eating quality traits were corrected for test night (15 levels), birth-year and sex. Marker regression analysis was performed using the R-Program 22 . In subsequent analysis, two models were used to determine the association between phenotypes and haplotypes constructed using variation at microsatellites BM81124 and BULGE20. In both models, regression was used where each sire haplotype was fitted under the hypothesis it carried a QTN. Identity by Descent (IBD) probabilities were calculated for each putative QTL at the midpoint between BM81124 and BULGE20. In the first model, haplotype contrasts were calculated independently within each of the 15 half-sib families. The covariance between identical haplotypes in different families was set to zero, implying that the QTL effects were allowed to vary between sire families. The resulting solutions for haplotype effects are referred to as the 'within-family' analysis (Table 6). In the second model, haplotype effects across families were correlated, with the covariance determined by haplotype similarity using previously published methods 23 ('across-family analysis', Table 6). For both models, a likelihood ratio (LR) test was calculated, indicating the ratio of log likelihoods of a model with haplotype effect over a model without haplotype effects.