An F2 population of females and barrows was produced from an intercross between two production sire lines: FH016 (Pietrain type, France Hybrides, St-Jean-de-Braye, France) and FH019 (Synthetic line from Duroc, Hampshire and Large White founders, France Hybrides). Eighteen F1 males and 72 F1 females were produced from 16 FH016 F0 males and 25 FH019 F0 females. F2 animals (1,370) were produced by mating each of the 18 F1 boars to the same group of one to four full sib F1 sows for the production of one to three litters. All animals were raised on the same farm and were slaughtered in the same abattoir in 32 successive batches. The number of records collected on F2 animals ranged from 700 to 1,350 according to the trait (Table 1 and Additional File 3).
Genomic DNA was extracted from white blood cell samples from F0 and F1 animals using a salting-out procedure  and from piglet tails docked at birth for F2 animals using a DNeasy tissue extraction kit (QIAGEN, Courtabeuf, France). One hundred and seventy microsatellite loci spanning all autosomes with an average spacing of 17 cM were selected based on predicted information content of available markers in this population. Information content was assessed from the number of alleles observed and average heterozygosity when genotyping all F1 males (data not shown). Microsatellites were genotyped by PCR and electrophoretic sizing of the PCR products. The sequence of primers used in PCR followed the reference pig genetic maps (, NCBI Map viewer Sscrofa5) or are summarized in Additional File 4 for complementary markers. Amplification was carried out with fluorescently labelled primers (M13 tag-tailed primer and labelled M13 primer in the same PCR reaction, Dyes IRD700/IRD800, Sigma-Genosys, Lyon, France). PCR was followed by pooling markers across different size ranges and a denaturing polyacrylamide gel electrophoresis revealed on a LICOR 4200 automated sequencer (Li-COR Inc., NE). Microsatellite alleles were called by PCR fragment sizing and binning using Li-COR SAGA genotyping software and a custom set of co-migrating size standards (fluorescently labelled PCR products from a set of plasmid inserts). We separated F1 parents and F2 progeny segregating alleles on common gels to allow a direct match between parents and progeny alleles. All gels were manually reviewed and edited for the assessment of genotyping quality and genotype validation. In addition to microsatellite genotypes, the MC4R and PRKAG3 V199I RFLP polymorphisms were genotyped as described [44, 45] and used as genetic markers. All F2, F1 and F0 animals were genotyped for all markers. We performed iterative Mendelian inheritance checks for each marker allele throughout the whole pedigree using the peeling algorithm implemented in the LOKI software package .
Multipoint linkage maps were constructed ab initio using CRIMAP 2.4 software . Genetic maps used for QTL detection were built from all the genotypes available from this population including carriers of the RYR1 and PRKAG3 mutations using the CRIMAP build option. However, the order of the closest markers was fixed as inferred from physical maps [48, 49] when a limited number of recombination events in this population did not allow proper assessment of marker order. The sex-averaged genetic maps used subsequently in IBD coefficient estimation and QTL detection analyses are reported in Additional File 4.
RYR1 and PRKAG3 genotypes
All F2 animals were individually genotyped for the RYR1 R615C and PRKAG3 R200Q mutations using previously described RFLP tests [3, 50]. These genotypes were used to define a population excluding carriers of either of the mutations we used for de novo QTL detection, while a control population included all the animals and was used for control detection of known mutations (mutant allele frequencies p RYR1 Cys615 = 0.08; p PRKAG3 Gln200 = 0.03). Additional File 3 lists the number of records per trait and per genotype analyzed in this control population including all animals or selectively removing mutation carriers. Mutation genotypes were subsequently included in the genetic marker genotypes used in QTL detection procedures, although they were only informative in the control population.
Meat quality traits
Early pH fall in loin muscle was recorded as pH at 45 minutes in the loin (pH-45) and was measured from a solution of 1 g of longissimus lumborum muscle dispersed 45 minutes post mortem in 9 ml of a 5 mM sodium iodoacetate - 150 mM potassium chloride buffer. Ultimate meat pH was recorded in loin (longissimus lumborum muscle, pH-LL) and ham (semimembranosus muscle, pH-SM) 24 h post mortem on half-carcasses. Intramuscular fat content (IMF) was measured as percent lipid (lipid weight:meat weight) as determined by chloroform/methanol extraction from freeze-dried longissimus lumborum sampled 30 h post mortem. Glycolytic potential (Glyc-P) was measured on the same longissimus lumborum samples and calculated from the enzymatic quantification of glycogen, glucose-6P and lactate concentrations .
Loins were sliced 30 hours post mortem, and two 3-4 cm slices (326 g +/- 65 g; including 11th-12th ribs) were collected. A first chop was used for loin colour measurement, raw meat shear force measurements and chemical analysis. The other chop was kept at 4°C for 48 h in a closed polyethylene plastic bag and then grill-cooked in a 240°C dry oven for 30 min. Drip loss was calculated as 100(1 − weight after storage/weight before storage). The same chop was weighed again after cooking and dripping and loin cooking yield was calculated as 100(weight after cooking/weight after storage). Raw meat and cooked loin instrumental tenderness were assessed by the measurement of shear force (Warner-Bratzler 60° cell, maximum strength over shearing of 1 cm diameter meat cylinders, average of ten repetitions). Loin lightness (CIE L*), redness (CIE a*), and yellowness (CIE b*) colorimetric parameters were recorded using a Minolta CR300 colorimeter (Konica Minolta Sensing Europe BV, Roissy, France). CIE L*, a* and b* colour measurements were acquired with the same equipment from deboned semimembranosus muscle 96 h post mortem. An in-depth account of procedures used for recording meat quality traits was reported previously [51, 52].
Carcass composition and growth traits
Backfat thickness and loin muscle cross-section area were recorded on live pigs at 105 kg bodyweight from echographic pictures of loin sections at the last rib level acquired using a 3.5 MHz linear transducer probe and an Aloka 500-V echograph (Aloka, Tokyo, Japan). Fat thickness and muscle depth were recorded from carcasses at slaughtering using the carcass grading system in use at the abattoir, Fat-O-Meter (SFK, Herlev, Denmark). Fat depth and muscle depth from this equipment are calculated from readings of an optical probe inserted through the skin, the back fat layers and the loin muscle and recording penetration depths at transitions between fat (white) and muscle (red) layers. Records were acquired at a dorsal (last rib) and a lumbar position, on a lateral line located 10 cm from the mid-line. A hand-operated optical probe (Introvison Ltd, Hitchin, UK) used on carcasses 24 h post mortem allowed a direct visual score of fat depths at similar locations. Fat depth was also recorded with a ruler on ham cuts directly above the humerus and on half-carcass mid-line cuts at a dorsal and lumbar position. The primal cut weights of ham without feet or loin but with bones were recorded at carcass cutting. Loin rib eye area was measured from digitized pictures of sliced chop. Individual birth weight was recorded on the first day of life and growth average daily gain was calculated as total weight gain during the ad libitum fed growth period (35 kg to 105 kg) divided by the number of days for this period. Elementary statistics for all traits analyzed along with abbreviations used and short definitions are presented in Table 1 for the F2 population excluding carriers of RYR1 and PRKAG3 mutations.
The genetic variability of the population used in this study, where phenotypes were recorded in a single F2 generation, can be described as a segregation of F1 alleles and can be analyzed by interval mapping for QTL segregation within each of half-sib or full-sib F2 families . However, parental haplotypes from these outbred populations cannot be treated as fixed in each parent population, and actual sizes of half-sib or full-sib families in our F2 population limits the detection power of within-family analyses or restrict analysis to the largest families. However, additional relationships exist in our pedigree beyond segregation within full-sib and half-sib families as F1 males and females used in different half-sib families were selected from the same F0 litters. Additionally, as all F1 females mated to the same F1 male were full sibs, relationships between F2 half sibs include sharing of identical F0 alleles transmitted by the different F1 females in addition to the segregation of paternal alleles. Use of sampling-based approaches associated with pedigree peeling makes it possible to infer sharing of identical-by-descent (IBD) founder alleles among all phenotyped (F2) animals .
We present results from a QTL detection analysis based on population-wide analyses of covariance between relatives associated with their IBD coefficients for each particular genomic position considered, as could be inferred from pedigree and inheritance of linked marker alleles. QTL detection was performed using a two-step variance component estimation method as proposed by George  and subsequently demonstrated in livestock [9, 55]. Briefly, univariate mixed models of variance were fitted to trait observations under either an additive polygenic model (H0) or a QTL model (H1: additive polygenic effect and QTL effect) for each of the tested positions. Both variance analysis models included the same fixed effects describing slaughter or fattening batches (all traits) and sex (carcass composition traits, cooked loin shear force, glycolytic potential and intramuscular fat percentage). Carcass weight or live weight was used as a covariate in all carcass composition trait models. A covariate for chop weight was used in the loin cooking yield and cooked loin shear force models. An independent random effect was used to fit a common environment defined as birth in a common litter in the analysis of individual birth weight and average daily gain. Additive genetic effect was set up using a three-generation pedigree structure in addition to the phenotyped animals, and QTL effects were fitted using the IBD relationship matrix estimated for the given genome position. IBD relationship matrices were estimated using a reversible jump Monte Carlo Markov chain method every 4 cM along linkage groups using the software package LOKI 2.4.6  and sampling over 20,000 iterations. No assumptions were made regarding IBD status of founder alleles originating from the same parental line, i.e. all IBD probabilities of founder alleles were set to 0, irrespectively of founder parental line origin. This was set to describe expected allele heterogeneity within these outbred parental lines, with known admixture history (one of the two parental lines being a synthetic line). Variance components were estimated using a residual maximum likelihood (REML) method with ASREML 2.0 software .
A QTL detection test was computed at each of the scanned positions using a likelihood ratio test (LRT) of −2(logikelihood-H1 − loglikelihood-H0).
It was suggested that this statistic followed a chi-squared distribution from 1 DF (for one specific position) to 2 DF when testing over a genetic interval . The maximum LRT recorded for each scanned chromosome was compared with chromosome-wise thresholds obtained by simulation (see below) to assess QTL detection. The sum of estimated variance components as estimated from a polygenic additive model (H0) in a population excluding RYR1 and PRKAG3 mutation carriers was used to compute reference trait standard deviation in summary statistics of Table 1. The proportion of variance explained by each QTL effect was estimated as the ratio between the variance component associated with the IBD relationship matrix at most significant QTL position and the sum of variance components estimated from the same model (Additional File 5). In both cases, variance estimates refer to total phenotypic variance after correction for fixed effects and covariates used.
We checked total variance inflation or deflation between significant QTL models and polygenic (null hypothesis) models. Out of the significant and suggestive QTLs reported, we noted five cases where QTL model total variance deviated from additive model total variance by more than 2% (namely QTL for LL-a* (+5%), LL-b* (+3%) on SSC6, QTL for ADG on SSC1 (+3%) and QTL for F-FOM-B on SSC3 (−3%) and SSC5 (−4%)). In all other significant and suggestive QTL models, the sum of variance components was within 2% of the sum of variance components estimated under a polygenic model. We checked the sampling error of estimated QTL variance component, as estimated from the square root of the diagonal element of the average information matrix . For all significant QTLs, this sampling error reported as a proportion of estimated variance component (component/sampling error ratio), ranged from 1.73 to 2.82 over 28 significant QTLs, with an average value of 2.07.
Significance thresholds and confidence intervals
We inferred chromosome-wise significance thresholds from the distribution of QTL detection statistics observed on phenotypes simulated under the null hypothesis. We ran QTL detection using the same marker genotypes (IBD coefficients) and pedigree on sets of simulated phenotypes modelled from our pedigree structure as carrying polygenic additive variation only (h
2 = 0.3). However, to save computing, we ran QTL autosome scans on a large number (5,000) of simulated phenotype datasets but using a smaller subset of animals (305 animals; 4 half-sib families). We also scanned QTLs on four representative test chromosomes (SSC1, SSC6, SSC15, and SSC18) over fewer (1,000) full-sized simulated datasets modelled from the population structure used for a typical meat quality trait record (SF-cook, n = 760). A summary of selected quantile values from the distribution of maximum LRT detected on these datasets simulated under the null hypothesis is tabulated for each chromosome in Additional File 6. Chromosome-wise significance thresholds were found to vary to some extent according to linkage group size, but were not influenced by dataset size (Additional File 6), consistent with the expected distribution of nominal test statistic. Empirical simulation-based chromosome-wise thresholds were found to be close to corresponding quantiles from a χ2
2DF distribution, the upper bound theoretical estimate proposed for a genetic interval  (1% quantiles were found ranging from 7.7 to 9.5 over the 18 autosomes, compared with a 1% quantile from χ2
2DF of 9.21).
Simulations were performed only once and quantiles drawn from the null hypothesis LRT distribution of the largest set of simulations (5,000) using small datasets for each chromosome were used as chromosome-specific chromosome-wise significance thresholds throughout all traits analyzed in this work, irrespective of the actual number of records in each analysis. We considered only maximum QTL detection statistics across all positions along each linkage group against these chromosome-wise thresholds without considering any secondary peaks. Genome-wise significance levels were adjusted using a Bonferroni correction as suggested previously , and QTLs are reported as significant for a 5% genome-wise significance level (equivalent to a 0.285% chromosome-wise level for independent scanning of 18 autosomes per genome). QTLs detected with an LRT above 1% or 5% chromosome-wise threshold but below the 5% genome-wise threshold are reported as suggestive QTLs. A 1% genome-wise threshold is proposed to support interpretation of the most significant QTL detection tests in figures and tables as drawn from a χ2
2DF distribution and using the same Bonferroni correction for the testing of 18 chromosomes (1% genome-wise threshold corresponding to a 0.058% chromosome-wise threshold).
Confidence intervals are proposed on the basis of a -1 LOD drop-off and are reported as 95% confidence interval map segments where the loglikelihood of the QTL model is higher than loglikelihood (max) − ln(10) . Predicted QTL genotypic effects were obtained from solutions of mixed model equations for QTL models fitted at most significant chromosome position defined by maximum LRT.
Comparison of results with published QTLs
We identified parallels between significant QTLs detected in this study and QTLs reported in the literature for similar traits by searching the online directory of published QTLs, the pig section of the AnimalQTLdb release 12 . However, in most cases, the very large number of QTLs (5,986) described in pigs, and the very large confidence intervals associated with F2 and backcross studies, prevent formal identity confirmations with our detected QTLs.