Resource population
Two divergent breeds of chicken, WL and SLK were used to set up the F2 cross used in this study. WL chickens were from a flock that is maintained at the Roslin Institute and showed no incubation behaviour (0 % incidence) when tested by the same method as that used to record phenotypes for the F2 population (Fig. 1). SLK chickens were obtained from the Wernlas Collection (Shropshire, SY7 9BL), a certified rare breeds farm (now closed), and maintained at the Roslin Institute. In this breed, incidence of incubation behaviour reached 90.5 % when tested (Fig. 1). All matings were performed by artificial insemination. Three WL sires were crossed with eight SLK dams and two SLK sires were crossed with ten WL dams in the F0 generation to produce the F1 cross. Four males and 20 females from the F1 generation were used to establish the F2 population. Phenotypic data for incubation behaviour were successfully recorded on 280 F2 animals from 19 families. This population was already used in a study that determined the causative mutation for preaxial polydactyly [17]. All animal experiments were performed according to United Kingdom Home Office legislation and were approved by the ethics review group of the Roslin Institute.
Incubation behaviour phenotype
After hatching, chickens of the F1 and F2 generations were reared in floor pens on a short-day lighting schedule (8 h light and 16 h dark) for 16 weeks. After 16 weeks, the birds were transferred to new floor pens (4 m × 1 m) in groups of six or seven individuals on a long-day lighting schedule (16 h light and 8 h dark). Temperature was maintained between 18 and 23 °C and the birds had access to food and water ad libitum. Each pen contained nest boxes, with wood shavings and hard-boiled eggs to encourage incubation behavior. Daily behavioral observations recorded birds that displayed persistent nesting (i.e. not leaving the nest if challenged), raising feathers and clucking when approached. Birds for which this behavior persisted for two consecutive days were referred to as incubating. Freshly laid eggs were removed from the pens but not the hard-boiled eggs. The birds were maintained in these conditions up to the age of 1 year. The number of days between entering the pen and the onset of incubation behaviour was also recorded.
Treatment of the data for analysis
Data for analysis were prepared in two ways. First, for the trait termed ‘incubation status’, the data were categorised on a 3-point scale i.e. there were three categories of birds: category 1 included birds that clearly showed incubation behaviour including vocalisation, raising feathers when approached and sitting on the eggs; category 2 included birds that showed some signs of incubation behaviour, for example, vocalisation and/or raising feathers when approached, but no persistent sitting; and category 3 included birds that showed no signs of incubation behaviour. In the second approach, the data were split into cumulative time periods because we believed that the time needed for hens to start incubation behaviour might reveal information on the strength of the motivation for incubation. The experiment was split into five cumulative time periods, by categorising and analysing birds that showed full signs of incubation behaviour from 25 to 30, 25 to 36, 25 to 42, 25 to 48 and 25 to 53 weeks of age. However, since significant or suggestive results were only obtained for the first period, in this paper, we consider only the first period between 25 and 30 weeks of age, which is referred to as ‘early incubation behaviour’. The phenotyping data were considered as a bivariate trait with two classes i.e.: class 1, which included birds that did not display any incubation behaviour, and class 2, which included birds that did have incubation behaviour (See Additional file ‘Phenotype’ accessible at Dryad Digital Repository).
Genotyping
Blood samples were collected from all F0, F1 and F2 individuals and DNA was extracted as previously described [18].
Microsatellite markers
Based on an initial screen, 90 microsatellite markers that were known to be informative for these populations and to be spread across 23 autosomal linkage groups and the sex chromosomes were used to genotype F0, F1 and F2 animals (See Additional file ‘Genetic Map Distances’ and Additional file ‘Genotypes’ accessible at Dryad Digital Repository). Fragment sizes were determined using GENESCAN 3.1 DNA fragment analysis and GENOTYPER 2.1 (PE Biosystems, Foster City, USA).
Single nucleotide polymorphisms (SNPs)
After an initial analysis for potential QTL, the number of markers on chromosomes 2, 5, 7, 9, 13, Z and linkage group E22C19W28 which had putative QTL was increased and linkage groups which were not represented in the initial microsatellite screen were added (19, 20, 21, 22, 24, 25, LGE64) by genotyping the entire population for 384 SNPs. SNPs were selected from the Ensembl genome browser (http://www.ensembl.org/Gallus_gallus/Info/Index) and a list of published validated SNPs [19]. SNP genotyping was done using the GoldenGate genotyping assay with VeraCode technology (Illumina) and Bead Studio software was used to analyse SNP data and, as a first quality control, to remove poor quality or uninformative SNPs. Of the 384 SNPs, 218 were informative and of suitable quality. After these 218 SNPs were added to the map, a further 32 SNPs were added in the peak region of the best QTL on chromosome 5, some of these SNPs being located in coding regions. Genotyping was also carried out using the GoldenGate genotyping assay with VeraCode technology (Illumina). These SNPs were selected by comparing the whole-genome sequences of the SLK and WL breeds, which were obtained by Illumina®/Solexa next-generation sequencing. (See Additional file ‘Genetic Map Distances’ and Additional file ‘Genotypes’ accessible at Dryad Digital Repository).
Software used
All pedigree information, marker genotypes and phenotypic data were stored in the ResSpecies database [20]. Each marker was checked for individual and pedigree errors using the related ResSpeciesGenotypeChecker (http://resspecies.org) software prior to submission and poor quality data were removed [21].
Map construction
For map construction, both microsatellites and SNPs were exported from ResSpecies in the CRIMAP format. Marker order and map distances were estimated by using CRIMAP 2.4 software (http://saf.bio.caltech.edu/saf_manual/crimap-doc.html). For chromosomes 5 and 8 that contained many SNPs, the size of the maps was very large. Thus, we used the CRIMAP CHROMPIC option to identify unlikely double crossovers, which reduced the map of chromosome 5 to a more realistic length. For chromosome 8, it was necessary to construct a map by interpolation between the microsatellite genetic map and the SNP physical map that was obtained by using the 2006 genome build. The CHROMPIC option FLIPS was used with a 4-marker window to obtain the most likely order. A sex-average linkage map was built by using the BUILD option. All markers used in this study were referenced in the chicken genome database (http://www.genome.uc-sc.edu/) to obtain their positions on the genome. The linkage map used in the analysis is in Additional file ‘Genetic Map Distances’ accessible at Dryad Digital Repository). For the comparison between the genetic and physical maps, all positions were converted to the 2011 build of the chicken genome (galGAL4) using the liftOver tool (http://genome.ucsc.edu/util.html).
QTL mapping
Genotypic and phenotypic data for each trait were exported from the ResSpecies database. The interval mapping method [22] for QTL analysis was carried out using GridQTL [23] which is a grid-based portal version of the QTL Express program [24]. A linear model for the additive and dominance effects of a QTL at a given position was analysed by least squares for each trait with the additive effect defined as half the difference in the mean phenotypes between the two homozygotes and the dominance effect as the difference between the mean phenotype of the heterozygote and the average mean of the homozygotes. Family, pen and pen year were included as fixed effects in the model for ‘incubation status’ and for each of the cumulative periods. F-statistic profiles were generated at 1-cM intervals. We analysed the data by assuming that there were one or two QTL on each chromosome. The significant QTL that were detected were added as cofactors in the analysis of the remaining chromosomes in order to reduce background noise and increase the power of the analysis to identify additional QTL.
To give a realistic estimate of the proportion of the total phenotypic difference between breeds explained by a QTL, we estimated the mean phenotypic values for each genotype at the nearest marker to the QTL position that was fixed for the alternative QTL alleles between the SLK and WL breeds. The values for the trait were calculated for the WL–WL, SLK-WL and SLK–SLK allele combinations, and the size of the QTL effect was estimated as the proportion of the total phenotypic difference between breeds explained by the QTL expressed as a percentage. Thus, the QTL effect at the marker was twice the estimated additive effect divided by the phenotypic difference between the founder breeds for each trait. This is referred to as the ‘phenotypic difference’ to distinguish it from the ‘phenotypic variance’.
Determination of significance thresholds
In single- and two-QTL analyses, significance thresholds were determined by conducting 5000 permutations [25] and 1000 bootstraps were used to generate 95 % confidence intervals [26, 27]. A QTL was considered as significant if its F value was greater than the P ≤ 0.05 experiment-wide threshold value [28] and as suggestive if it exceeded the P ≤ 0.05 chromosome-wide threshold. With 23 independent autosomes analysed, this is approximately equivalent to the expectation of 23 × 0.05 = 1.15 false positive QTL per genome scan.