The significance threshold for differential expression (DE) was set at False Discovery Rate (FDR) adjusted p-values (q-values) = 0.015. In Table 1 numbers and details of DE probes identified in eight separate contrasts are summarized. When all RJ-individuals were compared to all WL-individuals, DE was observed for 604 probes. DE between males of the two strains was observed for 410 probes and the corresponding female comparison revealed DE for 270 probes. A total number of 837 probes, corresponding to 779 unique transcripts were identified as DE in any of the three latter contrasts (Additional File 1) and overlap of DE between contrasts is presented as a Venn-diagram in Figure 1. Volcano plots in Figure 2 illustrate global distributions of log2(fold change in expression) (M-values) and q-values for all three contrasts performed between RJ and WL. Details regarding the 85 probes for which DE was observed in all three contrasts are presented in Table 2 and hierarchical clustering of these is presented as a heat map in Figure 3. Hierarchical clustering was also performed for all 837 probes showing DE in any contrast between the two chicken strains (Additional File 2). Close to two thirds (63%) of the probes showing DE-between females of the two strains had lower expression in WL females, which is a highly significant bias (χ₂, d.f. = 1, P < 0.001). There was no apparent difference in the direction of differential expression between males where 48% of probes showed lower expression in WL.

Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) was used as endogenous control probe for the microarray, and had been spotted in different concentrations in many separate spots. Upon examination of microarray fluorescence data one WL male was found to express GAPDH in levels around 50% lower than the other 19 individuals (Additional File 3). Since GAPDH had been chosen as reference gene, this individual was excluded from qPCR analyses.

Nine genes were chosen for verification of DE by quantitative PCR (qPCR). These nine genes were identified as DE between WL and RJ in the microarray analysis, some of them with high statistical significance for DE while others had q-values only slightly below 0.015. Data from qPCR revealed statistically significant DE (P < 0.05 in Student's t-test) between WL and RJ for all nine transcripts. Results from qPCR-analysis are presented in Figure 4 and in Figure 5. In Figure 6, M-values observed in qPCR-analysis are presented together with corresponding values from the microarray analysis.

In total 916 probes on the microarray were present in the four QTL-regions (ecirc1 = 147 probes, nc-bmd1 = 354 probes, bmd1 = 148 probes and tors1 = 267 probes). Of 837 probes showing DE between WL and RJ, 60 were localized to QTL-regions (Table 3) and these correspond to 57 unique transcripts. In Figure 7, relative expression of the 60 DE probes is graphically presented along each QTL-interval. Of 13,907 probes spotted on the microarray, 6.6% were present within QTL regions. Out of 837 probes showing DE, 7.2% were localized to QTL-regions. Thus, there was no apparent overrepresentation of DE in QTL-regions.

GO-analyses of DE-transcripts revealed statistically significant overrepresentation of functional classes belonging to all three top-level ontologies (cellular component, molecular function and biological process). Transcripts encoding ribosomal proteins were highly enriched in the ontology "Cellular Component" and in the ontology "Molecular Function", protein biosynthesis was similarly overrepresented. Table 4 lists GO-terms overrepresented among transcripts identified as DE between RJ and WL. GO-categories overrepresented among transcripts identified as DE between males and females is presented in Additional File 7.

Mean noncortical BMD (BMD of trabecular and medullary bone) was 282 ± 49 mg/cm³ for RJ females and 348 ± 27 mg/cm³ for WL females (Additional File 4). Male pQCT images are presented in Additional File 5.

Comparisons between males and females rendered the highest numbers of differentially expressed transcripts (Table 1). As expected, all females and no males expressed the female specific W-chromosome gene WPKCI, a gene believed to be involved in female sex determination in birds 32. In male birds, the femoral midshaft consists of an inner blood filled marrow cavity surrounded by an outer lining of cortical bone. The marrow cavity of the sexually mature female bird is, in addition to blood and air, also occupied by varying amounts of medullary bone lining the endosteal surfaces of cortical bone. Consequently, the female femur contains a tissue which is absent from the male femur and it can therefore be expected that a large proportion of DE observed between sexes is dependent on relative differences in cell type abundance. Another fraction of DE transcripts is likely to be directly or indirectly caused by factors not attributed to phenotypic differences, but rather to karyotype as it was recently shown that dosage compensation of the avian Z is not as effective as mammalian dosage compensation 33 (in birds males have two Z-chromosomes, whereas females have one Z- and one W-chromosome). GO-categories identified as overrepresented among transcripts which exhibit DE between sexes (Additional File 7) could consequently harbor functional categories associated to cell types present in varying ratios in female and male bones. Males and females had statistically significant differences in expression of certain known bone cell type markers. Alkaline Phosphatase (ALP) was expressed in higher levels in the female (M-value = 0.63) (indicating higher numbers of osteoblasts in the female bone). Phosphate-regulating Endopeptidase Homolog, X-linked (PHEX) also had higher expression in female bone (M-value = 0.64), which similarly would indicate higher osteoblastic numbers in female bone. The osteoclast marker Creatine Kinase B (CKB) was expressed in more than four times higher levels in female femurs than in male femurs. The higher expression of markers for both osteoblasts and osteoclasts suggests that bone remodeling is accelerated in female bone, and is likely attributed to females having higher turnover rates of bone tissue.

For nine selected transcripts, DE observed in the microarray analysis was verified by qPCR-analysis (Figure 6), indicating true differential expression for most probes having been identified as DE. As could be expected, the magnitude of DE varied between the two methods, of which qPCR is likely to be more accurate due to having greater specificity than hybridization techniques.

Higher numbers of microarray probes showing DE were observed in contrast between males of the two strains (n = 410) than in the corresponding female contrast (n = 270). In reproductive female chicken the cyclical process of egg-laying is accompanied by a phase of intense bone remodeling. It is possible that the females were in different phases of bone remodeling at the time of tissue sampling and/or that female femurs had differing relative amounts of cortical or medullary bone, which could also be calcified to different degrees. Differences in bone remodeling due to egg-laying and differences caused by heterogeneous bone phenotypes could both result in female biological replicates being more heterogeneous than male ones, and could explain why less DE was seen in the female contrast. In Figure 1, it can be seen that the fraction of DE transcripts which overlap between the male specific- and the sex independent contrasts is higher than the corresponding overlapping fraction observed between the female- and sex-independent contrasts. We suggest that this can be attributed to a larger phenotypic heterogeneity among female than male biological replicates.

Phenotyping of the femurs revealed that WL females generally had slightly higher density of noncortical BMD (in the female metaphysis medullary- and bone trabecular together constitute what here is referred to as noncortical bone) in the metaphysis with mean values of 348 ± 27 mg/cm³ for WL females and 282 ± 49 mg/cm³ for RJ females (Additional File 4). From phenotyping performed by pQCT in an independent sample of female chicken we have observed that noncortical BMD of the femoral metaphysis is strongly positively correlated with medullary BMD of the femoral midshaft (r = 0.73) (Additional File 6). The strain differences in noncortical BMD of the metaphysis are not large enough to conclude that they can be attributed also to the femoral midshaft. Interestingly, the metaphysis of RJ female individual 19 (RJF19) had very little, if any medullary bone (Additional File 4) and consequently a low noncortical BMD. The absence of medullary bone from RJF19 is likely to confer absence of this bone type also in the midshaft. Exclusion of RJF19 from the female RJ vs. WL contrast did not significantly alter the set of probes for which DE was observed, nor did the exclusion result in any dramatic changes in the statistical significance of observed DE (data not presented). The femoral phenotyping was performed in the metaphysis whereas the RNA was prepared from the midshaft and the data should thus be regarded as a qualitative rather than quantitative addition to the study. pQCT-data obtained for the female femurs indicate large differences in bone size between the RJ and WL populations, but also indicate that female biological replicates should be relatively homogenous with the exception of RJF19. From the phenotyping of male femurs (Additional File 5), we draw no conclusion other than that there is an apparent size difference between the two populations.

No commonly used bone cell markers were identified as differentially expressed between RJ and WL in the sex-independent contrast, nor in the two sex-specific contrasts. However, some genes for which differential expression was observed have previously been attributed roles in bone metabolism. For example WD-repeat containing protein 5 (WDR5), which accelerates osteoblast and chondrocyte differentiation 3435 and whose expression is induced by bone morphogenic protein 2 (BMP2). Figure 4C and Figure 6 show that WL-individuals expressed WDR5 in lower amounts than did RJ-individuals. Over the last few years, the wnt-signaling system has been attributed roles important to bone metabolism (reviewed in 36), and has become one of the most extensively studied signaling systems in the bone field. Wnt inhibitory factor 1 (WIF1) was identified as DE between females of the two strains, with WL having higher levels than RJ. WIF1 is a secreted protein that inhibits some wnt-proteins from binding to their receptors in a manner distinct from that of the secreted frizzled related protein receptors (SFRPs) 37. Interestingly, WIF1 is located in a RJ/WL QTL-region originally identified for body weight and growth 38, which also has pleiotropic effects on many bone traits 11 as well as several other phenotypes 3940. WL females were found to have stable mRNA levels of WIF1 whereas expression was heterogeneous in RJ-females (Figure 5B). Strong expressional regulation of WIF1 has been observed during BMP2 induced osteoblastic differentiation of murine C2C12 and MC3T3 cells 4142. The latter study reported that treatment of C2C12-cells with LiCl had a stimulatory effect on WIF1 expression, indicating that WIF1 expression is regulated by wnt-signaling components. Furthermore, co-expression of WIF1 and the transcription factor RUNX2 in osteoblastic regions of all bones in the head and the axial skeleton was observed in mice at embryonic day 16.5 41. RUNX2 is a master regulator of osteoblastic differentiation, and co-expression between WIF1 and RUNX2 therefore indicates a role of WIF1 in this process. Furthermore, WIF1 has been reported to be expressed in trabecular but not cortical bone of mature mice 42. This is intriguing because trabecular bone has a much faster turnover as compared to cortical bone, a relationship analogous to the contrasting lability between medullary- and cortical bone in chicken. Syndecan 3 (SDC3) has also previously been implicated in bone signaling 43. It had higher expression in WL and its expression pattern was highly correlated with WIF1 in hierarchical clustering of all probes representing the 779 transcripts identified as differentially expressed between WL and RJ (Additional File 2). We speculate that in reproductive female chicken, expression of WIF1 and SDC3 is induced in medullary bone in response to bone remodeling accompanying egg-laying.

Two probes targeting immunoglobulin-like receptor CHIR-A ([GenBank:CN233569] and [GenBank:CN233552]) indicated lower levels of expression in WL than in RJ (M-value = -1). The CHIR-A protein shows homology to osteoclast associated receptor (OSCAR), which has been shown to inhibit the formation of osteoclasts from bone-marrow precursor cells stimulated by osteoblasts 44. Although CHIR-A has yet not been anchored to a chicken chromosome, synteny between human 19q13.4 and chicken micro-chromosome E64 45 suggests location on this micro-chromosome.

LIM and senescent cell antigen-like domains 1 (LIMS1) and Ran binding protein 2 (RANBP2) are closely located on chicken chromosome 1 and were both expressed in higher amounts in RJ (Table 2). LIMS1 is a focal adhesion protein which together with integrin linked kinase 1 (ILK1) and alpha-parvin (PARVA) forms a complex which regulates cell shape, motility, and survival 46. Also associated to focal adhesions, cytoplasmic protein tyrosine kinase (PTK2) and vinculin (VCL) were expressed in higher amounts in RJ (Table 2). PTK2 is activated by binding of ligands such as integrins and growth factors to their receptors and it was recently proposed that interactions between vinculin, talin, and actin filaments constitute a slippage interface between the cytoskeleton and integrins 47. Large inter-individual differences in expression levels were observed for EGF-like repeats- and discoidin I-like domains-containing protein 3 (EDIL3). qPCR revealed that all female WL and one female RJ had at least 95-fold higher expression levels of EDIL3 than the other individuals most of which had no detectable expression (Fig. 3G). The EDIL3 protein adheres to endothelial cells through integrin A5/integrin B3 receptor binding 4849 and promotes angiogenesis by inducing expression of pro-angiogenic molecules 50. Possibly, a stable expression of EDIL3, as seen in WL-females is a result of them having a high turnover rate of the vascular medullary bone due to intense selection for egg production. The individual heterogeneity in expression among RJ-females could similarly be explained by their lower rate of egg production, with medullary bone not being subject to constant remodeling.

The observation that several DE transcripts have functional convergence at integrin signaling/focal adhesion is interesting and makes it tempting to speculate that these signaling pathways have become perturbed during domestication or due to selection for egg production in WL, possibly resulting in altered bone metabolism.

Of 779 unique DE transcripts, 57 were found to be localized within previously identified QTL-regions for bone traits (Table 3). Three of these QTLs had sex dependent effects in the RJ/WL-intercross population previously studied 11. QTL ecirc1 on chromosome 1 had sex independent effects and the WL-allele conferred larger circumference of the femoral endosteum. At QTL nc-bmd1 also on chromosome 1, female RJ-allele homozygotes were bestowed with higher noncortical BMD (medullary and/or trabecular BMD), whereas at QTL bmd1 on chromosome 2 the WL-allele was associated with higher noncortical BMD as well as higher total BMD in females. For tors1 on chromosome 20, female WL-allele bearers have more rigid femurs in rigidity tests by torsion. In Table 3 is indicated for which WL/RJ-contrast DE was observed, as this may aid in interpreting the significance of an overlap between DE and QTL.

Peroxisomal D3, D2-enoyl-CoA isomerase (PECI) which is present in QTL-region bmd1 was expressed in higher levels in RJ (tables 1 and 2). PECI is a peroxisomal enzyme catalyzing an isomerization step required for the beta-oxidation of unsaturated fatty acids, and has not previously been implicated in bone signaling pathways. The genomic location of PECI is very close to the peak of QTL bmd1 11 and the statistical significance as well as magnitude of differential expression (sex-independent M-value of -0.42) is similar regardless of contrast between WL and RJ. Due to these observations and despite no prior association to bone metabolism, we regard PECI as a QTL-candidate gene. Ribosomal protein L18A (RPL18A) was expressed in higher levels in RJ (M = -0.27) and the gene resides within QTL-region ecirc1. In GO-enrichment analysis, transcripts encoding ribosomal proteins were identified as overrepresented among DE transcripts, making RPL18A a QTL-candidate gene.

The probe targeting [GenBank:CN223165] was expressed in higher levels in WL (M = 0.40 and 0.32 in male- and sex independent contrast between WL and RJ, respectively). In protein BLAST search, [GenBank:CN223165] had highest similarity to a gag/pol polyprotein from avian myeloblastosis virus and was also highly similar to an avian endogenous retroviral insertion localized within the confidence interval of the pleiotropic QTL ecirc1. Interestingly, several isolates of avian myeloblastosis-associated virus (MAV) and avian leukosis virus (ALV) can induce osteopetrosis in chicken and may also affect growth (reviewed in 51). Osteopetrosis is a bone remodeling disorder characterized by an increase in bone density, and it is therefore tempting to speculate that the retroviral insertion could be causative of the pleiotropic QTL, for which the WL-allele confers higher phenotypic values both for bone and growth traits.

Theoretically, any genetic element which segregates between RJ and WL and which is localized to a QTL-region could be responsible for QTL-effects. Based on gene function as well as magnitude and statistical significance of DE, the best candidates among herein identified DE-genes include: WIF1, PECI, RPL18A and the retroviral insertion represented by a probe targeting [GenBank:CN223165].

GO-term enrichment-analysis of herein identified DE-probes could lead to identification of transcripts involved in signaling pathways having been perturbed between RJ and WL. It is however important to keep in mind that GO-enrichment analysis should be regarded primarily as a rough tool, which could aid in interpretation of data but also that obtained results require manual revision. The list of herein identified differentially expressed transcripts was enriched for certain GO-terms (Table 4). Prominent GO-categories identified in this study (Table 4), were those associated to the ribosome and to protein biosynthesis. Upon examination of probes contributing to overrepresentation of GO-terms; "cytosolic large and small ribosomal subunits", "protein biosynthesis", "ribosome", "structural constituent of ribosome", "RNA-binding" and "intracellular", we observed that 18 probes, representing transcripts encoding 15 ribosomal proteins were largely responsible for enrichment of these GO-terms. Inter-individual differences in expression for probes enriched in GO:0005842 and GO:0005843 are presented as a heat map in Figure 8. Probes for all 15 transcripts had lower expression levels in WL than in RJ, and had a mean M-value of -0.31. BLAST-searches of probe sequences for each ribosomal protein against all chicken cDNAs and against the chicken genome revealed localization to separate loci in the chicken genome and very low inter-sequence homology among probes, thus excluding the possibility of cross-hybridization as explanation for the observed overrepresentation. Of the eighteen probes enriched in GO:0005842 and GO:0005843, seven were derived from red junglefowl EST-libraries while the remaining eleven were derived from White Leghorn EST-libraries, suggesting that SNPs in the WL were not causative of differential expression. Interestingly, eukaryotic translation initiation factor 2, subunit 3 (EIF2S3) and Eukaryotic translation initiation factor 4, gamma 2 (EIF4G2) were also identified as DE in all contrasts between RJ and WL (Table 2) and were, analogously to the ribosomal proteins, expressed in lower levels in WL. When examining transcripts identified as DE in the sex independent contrast we noted that Eukaryotic translation elongation factor 1 alpha 1 (EEF1A1) and Eukaryotic translation elongation factor 2 (EEF2), were both expressed in lower levels in WL (M-values = -0.25 and -0.32, respectively). The observation that transcripts encoding 15 ribosomal proteins as well as several translation initiation and elongation factors had lower levels of expression in WL is intriguing. We speculate that one route, through which selection acted when desired traits were established in the WL, involved the alteration of pathways affecting protein biosynthesis.

Upon examination of transcripts contributing to other overrepresented GO-terms no apparently perturbed pathways were identified, possibly due to the limited coverage of the microarray.

Individual RNA-samples were subject to DNAse treatment using TURBO-DNAfree (Ambion) and equal amounts of RNA-samples were then reversely transcribed with the High Capacity cDNA reverse transcription kit (Applied Biosystems). Sample cDNA was diluted 20-fold with nuclease-free water and a reference chicken cDNA-sample was diluted 2-, 5-, 10-, 20-, 50-, 100- and 500-fold. Ten μl 2 × TaqMan^® Universal PCR Master Mix, No AmpErase^® UNG (Applied Biosystems) was mixed with 9 μl diluted cDNA and 1 μl of Taq-man gene specific assay mix (Applied Biosystems). This mix was subjected to 40 cycles of polymerase chain reaction amplification using the ABI Prism 7900 Taqman instrument (Applied Biosystems), where the amount of fluorescence is measured after each cycle. Diluted cDNA-samples from each individual were analyzed in triplicate and each dilution of the reference cDNA was analyzed in duplicate for the purpose of standard curve generation. For all TaqMan assays, correlation coefficients of standard curves exceeded R² = 0.98 and standard curve method was subsequently used for relative quantification of target abundance. All assays were designed with the Assays-by-Design software (Applied Biosystems) and featured gene specific primers and exon-spanning reporter probes (Table 5). Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) was used as reference gene and for each individual the relative expression level of each transcript was normalized relative to GAPDH-expression.

Chromosomal intervals were defined for our four previously identified QTLs for femoral bone traits 11, entitled ecirc1, nc-bmd1, bmd1 and tors1. Confidence intervals (CIs) of QTLs were defined as one unit LOD-drops at both sides of peak LOD-scores. For all QTLs, CIs were delimited by centiMorgan positions located in between microsatellite markers. As a conservative measure, QTL-regions were defined as chromosomal positions confined by microsatellite markers flanking CIs. Chromosomal positions were obtained from the chicken genome draft sequence build 2.1 62.

Clones on the microarray were attributed Gene Ontology (GO) terms by BLAST-analysis of all individual probe sequences against GO peptide sequences obtained from the Gene Ontology project website 63. BLAST search hits with a similarity cut-off of greater than E = e^-10 were retained for use in a modified version of the EASE-software 64. In EASE, GO-term enrichment analysis was performed for probes differentially expressed between all WL and all RJ. In the analysis, each GO-term was assigned an EASE score which is a conservative adjustment of Fisher's exact probability utilizing a jackknife approach. Enriched GO-terms (EASE-score < 0.05) comprising six or more DE probes on the microarray were retained for further analysis.

Clustering was performed in MultiExperiment Viewer (MeV) 65. Normalized fluorescence intensity values of each dye-swapped experiment were averaged separately for sample and reference channels. Thereafter, for each probe and individual, averaged sample and reference fluorescence values were log2-transformed. Average linkage hierarchical clustering was performed using Euclidian metric. In heat-maps the color of features (probes) were determined by log2(sample/reference).

The femurs, from which RNA had previously been prepared from the midshaft were subjected to phenotyping by peripheral Quantitative Computerized Tomography (pQCT). pQCT was performed with the Stratec pQCT XCT Research SA instrument (Stratec Medizintechnik) operating at a resolution of 70 μm. Noncortical BMD, which in the female bird reflects BMD of both trabecular and medullary bone was determined ex vivo, with one metaphyseal pQCT scan of the region situated at approximately 5% of bone length from the distal end of femur, and the noncortical bone was defined by setting an inner threshold to density mode (600 mg/cm3).