Background

The environmental variance (VE) of a trait is the within-individual variation of the phenotypic values of that trait due to environmental factors [1, 2]. VE is partially under genetic control, which means that individuals sharing the same environment can have different VE because they have different genotypes [1]. Indeed, there have been successful divergent selection experiments for VE in mice [3] and rabbits [4]. VE was recently proposed as a measure of animal resilience [5], which has been defined as an animal’s ability to cope with environmental disturbances and the rapid recovery of its productive performance [6, 7]. Differences in resilience have been reported in rabbit lines that were divergently selected for high and low VE of litter size (LS), with the line with a low VE of LS being more resilient [8]. According to Colditz and Hine [7], the immune system, nervous system and cell receptors are essential for modulating animal resilience and allowing detection of and response to environmental perturbations, such as pathogen infections.

Genome-wide associations studies (GWAS) for VE in livestock have identified candidate genes that are involved in the immune response, which boosts the inflammatory response [9,10,11]. In rabbit lines that were divergently selected for VE of LS, Casto-Rebollo et al. [11] identified functional mutations in candidate genes that are involved in the immune system, the nervous system, and the development of sensory structures. However, VE is a complex trait with a low heritability [12] and low phenotype accuracy, which makes the identification of all the loci that affect VE by GWAS only, difficult [13]. Analyses of signatures of selection, which do not require phenotype data, could help to identify more loci that affect VE of LS. Several methods for the detection of signatures of selection have been proposed and are based on different assumptions according to the pattern of positive selection to be detected [14]: (1) reduction of the local genomic variability, (2) modification of the spectrums of allele frequency, or (3) variation of the linkage disequilibrium (LD). However, because of these different assumptions, the three methods are not strongly correlated [15, 16] and, thus, have to be used in conjunction to identify the largest possible number of signatures of selection.

The aim of this study was to identify signatures of selection in rabbits that were divergently selected for high and low VE of LS during 13 generations [4], in order to determine the molecular mechanisms and pathways that control the VE of LS and animal resilience. We used three methods to identify signatures of selection in combination with whole-genome sequencing (WGS) analysis to highlight the genes with functional mutations. This study complements a previous GWAS for VE of LS using the same rabbit lines [11].

Methods

Animals and genotyping data

The rabbits used in this study were from generations 11 and 13 and from the base population of a divergent selection experiment for high and low VE of LS that was carried out at the University Miguel Hernández in Elche, Spain [4]. In total, 473 genotypes were used from 96 does from the base population, 282 from generation 11 (147 from the line with high VE of LS and 135 from the line with low VE of LS), and 95 from generation 13 (46 from the line with high VE of LS and 49 from the line with low VE of LS). Genomic DNA was isolated from blood and tissue samples using standard protocols. Genotyping was performed with the 200 K Affymetrix Axiom Orcun Single Nucleotide Polymorphism (SNP) array (Thermo Fisher Scientific) and quality control was done with the Axiom Analysis Suite 3.1 platform from Thermo Fisher Scientific and PLINK v.1.9 software [17]. Quality control removed animals with a call rate lower than 97% and SNPs that had a minor allele frequency (MAF) lower than 0.05, a missing genotype higher than 0.05, or an unknown position in the rabbit reference genome (OryCun v2.0.103). The missing genotypes were imputed with the Beagle v4.1 software [18]. After quality control, 452 genotypes and 97,155 SNPs remained in the dataset. A principal component analysis (PCA) was performed to study population structure and to identify outliers using the R package SNPRelate available from Bioconductor [19].

Identification of signatures of selection

Statistical analyses were performed to search for signatures of selection using the 274 genotypes from generation 11 (139 from the line with high VE of LS and 135 from the line with low VE of LS) and 90 genotypes from the base population. The 93 genotypes from generation 13 were kept for the validation analysis. Three methods were used to identify the patterns of signatures of selection (Fig. 1): (a) detection of runs of homozygosity (ROH), (b) quantification of the variation in LD patterns (VarLD), and (c) estimation of the fixation index (FST).

Fig. 1
figure 1

Methods of identifying patterns of signatures of selection. a Runs of homozygosity (ROH). From left to right: the number of consecutive homozygous SNPs increases, generating a genomic region where the individual is homozygous at all sites, i.e. a ROH. b Fixation index (FST). From left to right: allele frequencies of individuals in the population change until it differentiates into two different sub-populations (FST = 1). c Variation in linkage disequilibrium (VarLD), which searches for differences in linkage disequilibrium (LD) patterns between populations. From left to right: an advantageous allele (red star) can modify the LD in a population because of the selective sweep containing the SNPs surrounding it, i.e. the haplotype of this advantageous allele (Haplotype 2)

Full size image

Detection of runs of homozygosity

A ROH is a region of the genome that displays a local reduction of genetic variation, i.e. a genomic region for which the individual is homozygous at all sites, which indicates the presence of a locus that is affected by selection (Fig. 1a) [20]. Using an algorithm implemented in PLINK v1.9 [17], we identified the ROH in all the individuals from the base population and from the lines with high and low VE of LS of generation 11. The parameters were set according to Ceballos et al. [21]. This algorithm searches for stretches of consecutive homozygous SNPs on each chromosome using sliding windows of 500 kb that contain around 50 SNPs. SNPs with missing calls and more than one heterozygous SNP were not allowed in a window. The proportion of the overlapping windows that must be called homozygous to define any given SNP as in a homozygous segment was set to 0.05%. Two SNPs separated by more than 1 Mb belonged to two different homozygous segments. A homozygous segment was considered as a ROH if the number of consecutive SNPs exceeded 50 and the SNP density was higher than one SNP per 30 kb. A ROH must be a consensus genomic region in the selected animals to be a candidate signature of selection, i.e. it had to be identified in 50% of the animals in the line with low VE of LS (65), and in 50% of the animals in the line with high VE of LS (70).

Estimation of the fixation index

The fixation index (FST) was used to estimate the differences in allele frequencies between the lines with high and low VE of LS (Fig. 1b). The FST was calculated using Weir and Cockerham’s pairwise estimator method [22], implemented in the VCFtools v1.16 software [23]. The FST values were estimated in 500-kb sliding windows with a step size of 250 kb. Windows with less than ten SNPs were excluded from the analysis. FST values were weighted to take differences in sample sizes between populations into account (for further details see Weir and Cockerham [22]). Relevant FST windows were those with a weighted FST value equal or above the weighted FST value in the 99.9th percentile of the distribution for all the genomes. MAF was calculated in the base population and the lines with high and low VE of LS for the relevant FST windows. Those that showed divergent changes in MAF between the rabbit lines relative to the base population were considered to be putative signatures of selection. These windows were considered as resulting from an effect of genetic drift if the MAF between the lines with high and low VE of LS at generation 11 displayed the same change relative to the base population (increase or decrease) or if one of the lines did not show any change (i.e. had a MAF equal to that in the base population).

Quantification of VarLD scores

We used the VarLD software [24] to evaluate the magnitude of the differences in LD patterns (Fig. 1c) between two populations. We analysed the pairwise comparison of the three populations: base population with the line with high VE of LS (Base-High), base population with the line with low VE of LS (Base-Low), and between the lines with high and low VE of LS (High–Low). Sliding windows of 50 SNPs with a step size of one SNP were used to calculate the correlation matrix of each population per chromosome. The program computed the r2 metric for each pair of SNPs to determine the strength of LD in each window. The difference between the eigenvalues of the correlation matrices of both populations determined the VarLD score, which was standardized by the mean and the standard deviation of all the scores along each chromosome. A genomic window was relevant when its standardized VarLD scores were equal or above the standardized VarLD score in the 99.9th percentile distribution for all the genomes. The relevant windows identified in both the Base-High and Base-Low comparisons were considered as putative signatures of selection and the relevant windows identified only in the High–Low comparison were considered as resulting from the effect of gene drift.

Validation

The putative signatures of selection were validated by identifying those detected in the animals of the base population and of generation 13 (45 from the line with a high VE of LS and 48 from the line with a low VE of LS) applying the methods described above (Section on “Identification of signatures of selection”). Only those that were detected in both analyses (i.e. in generations 11 and 13) were considered as true signatures of selection.

Identification of candidate genes

Candidate genes were detected by searching for functional mutations in the genomic regions considered as true signatures of selection. Functional mutations were identified using whole-genome sequencing (WGS) data from two pools of DNA from breeding males in generation 10, i.e. all the fathers of animals from generation 11. Pools of DNA were prepared for each rabbit line (27 animals per line) and sequenced by Illumina Technology with an average depth of 27×.

WGS data were pre-processed following Elston et al. [25] with the following steps: (1) indexation to the reference genome (OryCun v2.0.103), (2) removal of adapters and low-quality read ends, (3) alignment to OryCun v2.0.103, and (4) identification of duplicates. Then, variant calling was performed using the GATK Best Practices pipeline [26] in three steps: (5) creation of raw VCF files for the high and low VE of LS lines, (6) variant filtering, and (7) variant annotation (for further information see Casto-Rebollo et al. [11]).

A variant was considered as a functional mutation if it affected the transcription unit of a gene, i.e. (a) if it was located in the UTR regions, (b) if it was a missense or frameshift mutation, or (c) if it affected a splicing site. The gene ontologies (GO) of each candidate gene were extracted using the biomaRt package available from Bioconductor to R [27]. The Ensembl 103 release database [28] was used to access the Oryctolagus cuniculus v2.0.103 information.

Results

Analysis of the population structure using principal component analysis showed a clear separation between the base population and the lines with high and low VE of LS (Fig. 2). The individuals from generation 13 displayed the same family structure than that of their ancestors from generation 11.

Fig. 2
figure 2

Principal component analysis of the genotyped data. Representation of the first (PC1) and second principal component (PC2) of the genotype data from the base population (orange) and the lines with high (right) and low (left) VE of LS in generations 11 (dot) and 13 (triangle)

Full size image

Signatures of selection for VE of LS

The ROH, FST and VarLD methods (Fig. 1) identified putative signatures of selection for VE of LS, using the animals from generation 11 and the base population. Analysis of the contiguous homozygous segments identified 6230 consensus ROH, which were detected in at least two animals of the base population and of the lines with high and low VE of LS. Of these 6230 consensus ROH, 720 were identified in at least 70 does of the line with high VE of LS and 65 does of the line with low VE of LS. These 720 consensus ROH were considered as putative signatures of selection because they were detected in at least 50% of the animals of each rabbit line. The FST analysis identified eight genomic regions with a weighted FST value equal or above 0.35 (99.9th percentile) on Oryctolagus cuniculus chromosome (OCU)2 (104.5–105 Mb), OCU9 (89–89.75 Mb), OCU12 (8.75–9.5 Mb), OCU14 (121–121.5 Mb), and OCUX (81–81.75 Mb) (see Additional file 1). However, only the regions on OCU2 (104.5–105 Mb), OCU12 (8.75–9.5 Mb) and OCUX (81–81.75 Mb) were considered as signatures of selection for VE of LS because they showed consistent divergent changes in MAF between the lines with high and low VE of LS relative to the base population (see Additional file 1). The VarLD analysis identified three genomic regions that showed differences in LD patterns and overlapped between the lines with high and low VE of LS on OCU13 (89.31–90.54 Mb), OCU14 (0.014–2.27 Mb), and OCU17 (28.78–29.92 Mb). The highest VarLD scores, 12.55 and 12.25, were obtained for OCU14 in the Base-High and Base-Low comparison, respectively. These three genomic regions were proposed as putative signatures of selection because their LD patterns differed relative to the base population.

The 726 putative signatures of selection identified in the ROH, FST and VarLD analyses were validated in 93 animals from generation 13. Among these 726 putative signatures of selection, 134 (see Additional file 2) were considered as true signatures of selection (Fig. 1), i.e. 129 ROH based on patterns of homozygous segments, two VarLD regions based on differences in LD patterns on OCU13 (89.31–90.54 Mb) and OCU14 (0.014–2.27 Mb), and three FST regions based on changes in allele frequencies on OCU2 (104.5–105 Mb), OCU12 (8.75–9.5 Mb) and OCUX (81–81.75 Mb). Finally, among these 134 true signatures of selection, the genomic regions did not overlap among the three methods.

Candidate genes for VE of LS

Nine hundred genes were identified in the genomic regions with positive selection patterns for VE of LS. Candidate genes were identified by searching for functional mutations in the 900 genes using WGS data. In total, 212,845 variants (single-nucleotide variants (SNVs) and insertion/deletion variants (INDEL) were identified in the true signatures of selection (Table 1). Among these 212,845 variants, 1196 were relevant (207 INDEL and 989 SNVs) based on their location in the transcription unit of 373 of the 900 identified genes. These 373 genes (proposed as candidate genes) are involved in many biological processes (see Additional file 3). Most of them (237) are found in basic common gene ontologies (GO) such as: protein binding (GO:0005515), cytoplasm (GO:0005737) and nucleus (GO:0005634), although, they also have a pleiotropic effect on other biological processes (see Additional file 3). One hundred and eleven genes (29.76%) are involved in biological processes related to immune response (e.g., GO:0030335 and GO:0071356), stress response (GO:0042594), reproduction and embryo development (GO:0001701), and/or carbohydrate and lipid metabolism (e.g., GO:0005739 and GO:0055114).

Table 1 Effects of the variants (SNVs and INDEL) identified in the genomic regions with true signatures of selection for VE of LS
Full size table

We highlighted the GATA3, FKBP10, KAT2A, CYP1B1, BRCA1, PGM3, and ACE genes that have a pleiotropic effect on the immune system, lipid and carbohydrate metabolism, and reproduction and embryo development.

Among the 1196 functional mutations, we found 10 INDEL that were fixed (1/1) in one of the rabbit lines and absent (0/0) in the other and that affect the TTC23L, FBXL20, GHDC, ENSOCUG00000031631, SLC18A1, CD300LG, MC2R, and ENSOCUG00000006264 genes (Table 2). These genes are involved in biological processes related to stress response (MC2R), energy, carbohydrate, and lipid metabolism (ENSOCUG00000006264 and MC2R), nervous system (MC2R, SLC18A1, and FBXL20), immune response (ENSOCUG0000000626), behaviour (FBXL20), cell maintenance (TTC23L and GHDC), and other processes (ENSOCUG00000031631 and CD300LG). For each of the MC2R and ENSOCUG00000006264 genes, one of the variants was a novel INDEL in the 3ʹUTR (Table 2) based on the two alleles in the reference rabbit genome OryCun v2.0.103. These novel alleles were fixed (2/2) in the line with high VE of LS and absent in the line with low VE of LS (Table 2).

Table 2 Functional mutations (INDEL or SNVs) fixed in one of the rabbit lines and absent in the other line
Full size table

Discussion

Divergent lines provide good biological material for genomic studies since they are selected for a unique trait and share the same environment. Previous studies on intramuscular fat (IMF) in rabbits and pigs [16, 29], and on antibody response and feather pecking behaviour in chickens [30, 31], using divergently selected lines, successfully detected signatures of selection, and identified associated genomic regions. The principal component analysis based on genotype data in our study showed a clear separation between the two divergent rabbit lines (Fig. 2), which agreed with the remarkable phenotypic differentiation in VE of LS (4.5% from the mean in the base population; Blasco et al. [4]). Thus, we searched for signatures of selection for VE of LS, by combining the ROH, VarLD and FST methods (Fig. 1) to identify genes and pathways that were modified during 13 generations of selection.

In the analysis of signatures of selection, the identification of genomic regions under positive selection depends on the method applied [14]. Using the ROH, FST and VarLD methods, we identified 134 true signatures of selection for VE of LS with no overlapping between methods. Indeed, as each method is based on different assumptions (Fig. 1), correlations between results are low [15, 16], which makes it difficult to detect overlaps between the identified signatures of selection. The methods used to detect signatures of selection should be considered independently of the sources of QTL detection. By combining the three methods, ROH, FST and VarLD, we were able to identify most of the selection forces that affect the trait of interest, and we considered these 134 true signatures of selection as independent patterns of positive selection for VE of LS. However, only one FST window on OCU3 (51–51.75 Mb) agreed with a variance-controlling locus (vQTL), which was previously identified in a GWAS [11] that used the same animals from the base and the generation 11 populations. The three genes (SLIT3, FOXI1, and FGF18) with functional mutations located in this vQTL could be the most relevant genes that play a role in the control of VE of LS. They are involved in biological processes related to immune response, stress response, and/or development of sensory structures, which are relevant pathways to modulate resilience [7]. However, we identified this signature of selection in animals from generation 13 using a less conservative threshold of 99.5th percentile (weighted FST of 0.37). This FST window showed a weighted FST of 0.21 at generation 11 and 0.39 at generation 13. This stronger differentiation in allele frequencies in the population from generation 13 highlights the importance of this region for VE of LS. However, the difference between the weighted FST at generations 11 and 13 could also be an effect of the reduced sample size (95) at generation 13, which may hide the true changes in allele frequencies between the generations.

Previous studies in divergent populations showed that some overlapping occurred between the signatures of selection obtained by FST analysis and a few QTL identified by GWAS [16, 29,30,31]. In contrast to our study, those studies used populations from a long-term divergent selection (during 40 generations) [30, 31], or from a selection for a highly heritable trait (intramuscular fat; IMF) [16, 29]. The fact that VE of LS has a low heritability and accuracy [12] could hinder the identification of the genomic regions under selection in GWAS. Moreover, the small or moderate size of the effect of the variants could produce a sweep that is not large or strong enough to be detected as a signature of selection [32].

The identification of relevant loci for complex traits results in a large number of candidate genes due to their polygenic nature [33]. These genes are usually involved in multiple pathways or biological processes that may not be interrelated, such that searching for a relationship between these and the trait of interest is challenging. Along the same line, WGS analysis could be useful to identify the most relevant candidate genes that underlie the complex traits under study. In this study, we identified 900 genes that spanned genomic regions with patterns of signatures of selection for VE of LS, and among these, 373 presented functional mutations that affect their transcription unit (see Additional files 4 and 5). However, given these 373 genes that are implicated in a wide range of functional categories (see Additional file 6), it remains difficult to identify the most relevant molecular mechanism involved in VE of LS. Moreover, these genes may not have a clear relationship with VE of LS since they may be acting indirectly, by modulating the core genes underlying VE of LS [34]. For this reason, we could only make hypotheses based on the genes directly involved in previously identified biological pathways for VE of LS.

Previous studies, developed by Argente et al. [8, 35] and Beloumi et al. [8] on the same rabbit lines as those used in this study, showed line-differences in immune response biomarkers (plasma cortisol, leukocytes, and acute-phase protein levels), in plasma concentrations of cholesterol and triglycerides, and in mortality [8, 35]. Among the 373 genes, 59 were related to immune response, six to stress response, and 49 to energy metabolism, carbohydrate metabolism or lipid metabolism (see Additional file 3), which could explain the differences reported by Argente et al. [8] and Beloumi et al. [36]. In addition, we found 38 genes involved in reproduction and embryo development (see Additional file 3) that could clarify the correlated response of VE of LS with embryo implantation, embryo survival and litter size traits [36, 37]. In our study, we highlighted the genes, GATA3, FKBP10, KAT2A, CYP1B1, BRCA1, PGM3, and ACE, because they have a pleiotropic effect (see Additional file 3). The ontologies of these genes are related to the immune system, lipid and carbohydrate metabolism, and reproduction and embryo development, supporting all the previously reported evidence [8, 35,36,37].

By searching for the most relevant functional mutations, we found seven promising genes that contained an INDEL with the alternative allele fixed in one rabbit line and absent in the other (Table 2), i.e., TTC23L, FBXL20, GHDC, ENSOCUG00000031631, SLC18A1, CD300LG, MC2R, and ENSOCUG00000006264, and which are the most relevant for VE of LS. However, the functional mutation detected in the MC2R and ENSOCUG00000006264 genes were even more interesting (Table 2). In our rabbit lines, both genes have lost the reference allele present in the rabbit reference genome OryCun v2.0.103 and display two different variants of the INDEL in their 3ʹUTR region. In both genes, one of the INDEL variant (the alternative variant for the reference genome) was fixed in the line with a high VE of LS. The other (a new variant) was fixed in the line with a low VE of LS (Table 2). The MC2R gene encodes the adrenocorticotropic hormone (ACTH) receptor, which controls ACTH and the level of cortisol [38]. Cortisol is an important molecule that regulates fat metabolism to mobilize glucose and mediates stress response and inflammatory response [39]. ENSOCUG00000006264 is an orthologue of the retinol binding protein 1 (RBP1) gene that is involved in the homeostasis of retinoid acid (RA) and the regulation of the vitamin A metabolism. Retinoid acid could be involved in many immunological functions, such as the control of inflammatory and tolerogenic immune response [40].

When comparing the results with previously identified VE loci (vQTL), we found evidence of the implication of the immune system in VE, in line with the results reported by Argente et al. [8] and Belloumi et al. [35]. In their GWAS, they identified genes that are related to the triggering of inflammatory response [9, 11] and belong to the HSP (heat shock protein) gene family [10, 11, 41], which can also modulate stress response, inflammatory response as well as the levels of glucose and fatty acids [42], and the fertilization and preimplantation of embryos [43]. Elevated levels of cortisol induce the synthesis of HSP to trigger stress response and cellular adaptation [44,45,46]. The INDEL variant on the MC2R gene could affect the stability of the transcribed mRNA, affecting the expression of the ACTH receptor that modulates the cortisol response. Thus, differential cortisol response could affect the stress and inflammatory response of animals, supporting the evidence of an effect on animal resilience [8, 35,36,37].

Animals can maintain their performance (be more resilient) when they can discriminate environmental stimuli from background. Berghof et al. [5] proposed VE as a measure of animal resilience, while Argente et al. [8] showed that the line with a low VE of LS was more resilient than the line with a high VE of LS. According to Colditz and Hine [7], the immune system, cell receptors, and nervous and sensory structures are essential for coping with environmental disturbances. In this study, we identified 59 genes that are involved in the immune system, 23 in sensory perception, six in animal behaviour and 35 in the nervous system (see Additional file 3), which support the correlated response of VE of LS in rabbit resilience [8]. Among these 59 genes, SLC18A1, FBXL20, and again MC2R were highlighted because they had also GO related to the modulation of the nervous system and/or behaviour (see Additional file 6). However, the role of the highlighted candidate genes in controlling VE of LS and animal resilience requires further studies to investigate their direct effect on the VE of LS. Although the 373 identified genes with functional mutations were considered as candidate genes for VE of LS, their direct or indirect implication in modulating the VE of LS needs to be assessed.

Conclusions

We identified 373 candidate genes with functional mutations for VE of LS in rabbits by combining independent methods of detection of signatures of selection and WGS data. These genes supported the biological pathways that were previously reported to be related to VE of LS and involved in immune response, lipid and carbohydrate metabolism and stress response. These candidate genes could also explain the correlated response of the VE of LS in embryo implantation, embryo survival and litter size. Two novel INDEL variants were fixed in the line with high VE of LS and absent in the line with low VE of LS, one in the MC2R gene and one in the ENSOCUG00000006264 gene. These promising functional mutations are located in genes that are involved in stress response and in the retinoid acid biosynthetic process, which could also control the immune response, respectively. This study expands on a previous GWAS for VE of LS in rabbits and identified additional molecular mechanisms and pathways for VE of LS and animal resilience. However, the real implications of these genes in VE of LS still need to be assessed through functional analyses.