Animal design and microarray setup
Animal design, genotyping and microarray setup are previously described by Le Mignon et al. . Briefly, the animal design corresponds to 45 male offspring produced by a sire known to be heterozygous for a QTL affecting abdominal fatness (AF) on chromosome 5 with a location confidence interval extended from 156 cM to 187 cM and a significant effect of 1.03 phenotypic standard deviation. This sire are not heterozygous for other AF QTL on GGA1, GGA3 and GGA7 previously detected in a three-generation F0-F1-F2 design performed by intercrossing two experimental chicken lines divergently selected for abdominal fatness from which the sire has been produced. Genotyping for GGA5 chromosome was performed for 10 markers (ADL0292, ADL0023, MCW0238, ADL0233, MCW0026, SEQF0079, SEQF0080, SEQF0082, SEQF0085, ROS330 at 83, 100, 125, 151, 162, 166, 175, 187, 190, 192 cM respectively). Markers were chosen from available markers  or developed for this program . The six additional SNP markers were developed from the chicken genome sequence assembly and correspond to rs15678496, rs15683152, rs15685956, rs16689818, rs15691594, rs14531246 at 67, 77, 80, 86, 89 and 95 cM respectively. Gene expression measurements were obtained from the livers of these animals using a 20 K chicken oligo array (Ark-genomics). 11213 genes (55 % of the 20461 genes) were selected as expressed in the liver. The raw and normalized microarray data were deposited in the Gene Expression Omnibus (GEO) public repository . The accession number for the series is GSE12319 and the sample series can be retrieved with accession numbers GSM309564 to GSM309609.
The animal labels were defined as follows: F1 to F20 for the 20 fattest animals, L1 to L20 for the 20 leanest animals and I for the 5 intermediates.
All experiments were conducted under Licence N°; 37-123 from the Veterinary Services, Indre et Loire, France and in accordance with guidelines for care and use of animals in Agricultural Research and Teaching (French Agricultural Agency and Scientific Research Agency).
Factor analysis method
The method takes into account the gene dependence structure and consequently, the impact of dependence on the multiple testing procedures for high-throughput data. Indeed, genes can have similar expression profiles because they are involved in common pathways but independently of the variable of interest (AF in our case). The common information shared by all the variables (i.e. gene expressions) and independent of the variable of interest is modeled by a factor analysis structure. An EM algorithm is used to estimate the model. Once the factor model is estimated, factor-adjusted test statistics are obtained by correction of the classical tests from the effect of the common factors. David Causeur's team showed that the resulting tests statistics are asymptotically uncorrelated, which improves the overall power of the multiple testing procedure (, ). The algorithm is implemented in the "FAMT" R package available from CRAN. As in Blum et al. , the raw expression data set is adjusted for the estimated independent factors, which results in the so-called factor-adjusted expression data.
QTL and eQTL mapping
QTL (eQTL) mapping consists in mapping on the genome, regions that control the variation of a complex trait (expression trait). Before QTL analyses, the AF trait values of the sire family (71 birds) were adjusted for hatch and dam effects by two-way variance analysis, including body weight at slaughter as a covariate (SAS GLM procedure). For the eQTL analyses, no adjustment of the gene variables was performed for hatch and dam effects because of the small size of the population studied (45 birds). QTLMap software based on an interval mapping method described by Elsen et al. , was used to detect QTL (or eQTL) affecting the AF trait (or a gene expression phenotype). The statistical variable for testing the presence of no QTL (or no eQTL) versus one QTL (or one eQTL) at one location and also of one QTL versus two, was an approximate likelihood ratio test (LRT) . Significance thresholds were empirically determined for AF QTL and transcript level eQTL from 2000 simulations performances assuming a polygenic model with a given heritability (h2 = 0.5). The widely used "one LOD drop-off method" was applied to obtain 95% confidence intervals of the QTL location . QTLMap software was also used to test an interaction between the proximal and distal QTL using the "interaction model" testing the hypothesis « No QTL versus 1 QTL in interaction with another one fixed in our study at 168 cM » 
We considered that a gene has an eQTL colocalizing with an AF QTL if the CI of the eQTL region was overlapping the CI of the QTL region.