Sample collection

Blood samples were collected from 196 African chickens: Ugandan (n = 72), Rwandan (n = 100), and Kuroilers (n = 24). Kuroilers, originally imported from India, were sampled from farms in Uganda. Five physically distinct farms were sampled in each sampling area to make up a single geographically defined ecotype, in an attempt to reduce stratification. There were six ecotypes (30 farms) for Rwanda from the areas of Huye, Kicukiro, Kirehe, Musanze, Nyagatare, and Rubavu. For Uganda there were three ecotypes (15 farms) from the areas of Kamuli, Luweero, and Masaka. Blood was collected using FTA cards (Additional file 1: Table S1).

Genotyping and quality control

Genotyping was conducted at GeneSeek (Lincoln, NE) using the Affymetrix Axiom ® 600 k Chicken Genotyping Array. SNPs were put through a quality control step in Plink [32, 33] based on the parameters of > 97 % call rate (-geno 0.03) and minor allele frequency (MAF) > 0.02. After filtering, 506,965 total SNPs remained. A total of 476,106 autosomal SNPs were available for downstream analyses.

Population stratification analysis

Hierarchical clustering analysis of the genotype data in JMP ® (www.jmp.com) was used to determine the relationships between the samples. This cluster analysis was run using the ward method [34]. PLINK [32, 33] was used to construct a multi-dimension scaling plot (MDS-plot) to examine population structure for stratification. MDS plots for this analysis were based on 196 × 196 matrix of genome-wide Identity-By-State (IBS) scores calculated based on pairwise comparisons of the genetic distances for all individuals. The APE [35] and PhanGorn [36] R packages were used to examine the relatedness of the birds by phylogenetic analysis using neighbor-joining (NJ) distance based tree construction methods (data not shown). Visualization of the corresponding trees was done using FigTree (http://tree.bio.ed.ac.uk/software/figtree) and Ninja (http://mesquiteproject.wikispaces.com/Additional+Mesquite+Packages). IBS values and inbreeding coefficients were calculated within Plink. Principal component analysis (PCA) was done using SNP and Variant Suite (SVS) [37] and JMP ® (www.jmp.com). Shared ancestry was also explored using the Admixture software [38] set at varying values of k, ranging from 1 to 9 to represent the number of ecotypes sampled, with a final optimal k = 3.

Runs of homozygosity analysis

Runs of homozygosity (ROH) analyses were carried out in Plink [32, 33] to examine genomic regions that harbor alleles driven to fixation using a SNP based sliding window approach. Runs of homozygosity were calculated for each individual and any ROH that overlapped between individuals within and between populations. For the individual and overlapping ROH analysis, a run was defined in Plink as  ≥ 250 SNPs, density of 50 kb/SNP, allowed gap of 1000 kb, and 3 heterozygous calls allowed within a run. The analysis focused on the overlapping ROH that were also analyzed using the additional parameters of allelic match threshold of 0.95 identity and 20 or more informative SNPs. The overlapping ROH are regions that overlapped across all populations and contained 10 or more individuals. Consensus regions between populations were defined as the region that was common to every bird, irrespective of length of the ROH, that Plink assigned to a “pool” because of shared overlapping ROH. The analysis of the overlapping/consensus regions was based on all samples as a whole and a “pool” of individuals sharing overlapping/consensus ROH was made up of birds from all three populations. A gene ontology enrichment analysis was conducted on a gene list from the ROH consensus regions [39, 40]. The gene list is based only on genes mapped to within the consensus region of the ROH. Gene ontology enrichment analysis results were considered statistically significant at a FDR cut-off < 0.15.

Analysis of putative selection signatures

Sample haplotypes were phased using FastPhase [41] for downstream analysis of selection signatures. The R package REHH [42] was used to calculate integrated haplotype score (iHS) and standardized log-ratio of the integrated EHHS (iES) between pairs of populations (Rsb) values to examine the populations for SNPs that displayed signals of selection. Both iHS and Rsb values were log transformed to normalize the data and calculated as per the method established by Voight et al., 2006 [43]. Statistical significance of iHS values were determined by use of the-log p-values generated by the REHH software package. The -log p-values were not adjusted for multiple test correction because all multiple test correction procedures proved to be too conservative due to the number of tests exceeding 400,000. To address the issue of statistical significance a very stringent nominal p-value (α = 0.001, −log α = 4) was set and iHS data was then ranked from lowest to highest. After ranking, the p-value (α = 0.001, −log α = 4) was used as the cut-off for all samples and populations. iHS values were considered extreme at iHS > |3.29| because this was the lowest iHS value at the p-value cut-off. Pairwise comparisons between populations were examined using the Rsb statistic. P-values for the Rsb statistic were low enough to be subjected to multiple test correction using a FDR cut-off of < 0.05. The use of a lower FDR was also necessary to reduce the number of significant results for a more focused downstream analysis. To carry out gene ontology term enrichment, SNPs were annotated to genes of interest using the Affymetrix NetAffx™ Analysis Center which, when supplied with a list of markers from one genotyping platform, will give a list of genes within, upstream, or downstream of a gene in that region.

Gene ontology enrichment analysis

Gene ontology (GO) term enrichment was performed using (GO)TermFinder [39]. Visualization and reduction of redundant terms of the GO enrichment results were carried out using ReVigo [40]. The significant GO terms were filtered for redundant terms to produce a focused list of the functions and processes under selection. The FDR for the enrichment tests was set at < 0.15. This cut-off was set higher than the threshold for the iHS or Rsb tests since it was a separate independent test. Another point that prompted the use of a higher FDR was the lack of annotations for the chicken genome, which effectively reduced the overall number of genes that could be analyzed.

Population structure analysis

The MDS plot (Fig. 1) showed overlap among Ugandan, Rwandan, and Kuroiler populations. The Ugandan and Rwandan ecotypes showed the highest degree of overlap between populations, with the Kuroilers showing the most discrete clustering of individuals. The amount of admixture, based upon identity by state, within each country (3 Ugandan ecotypes, 6 Rwandan ecotypes) showed crossover between sampling areas (Fig. 2). The clustering analysis based on the SNP genotyping calls indicated that the ecotypes (sampling location) assigned to the birds showed shared ancestry of genotypes between individuals across populations and ecotypes. These results were visualized as an admixture plot that also showed evidence of shared ancestry (Fig. 2).

Examination of runs of homozygosity (ROH)

The number and extent of individual ROH differed widely among the populations (unpublished data). Kuroilers had the fewest chromosomes that contained ROH, while the combined Ugandan ecotypes were the only population to have ROH on every chromosome except chromosome 16. Chromosome 16 showed no evidence of ROH based on the parameters in any of the populations. The median length of the ROH in a population was longest in Ugandan and shortest in Kuroiler. The amount of the genome covered by ROH per individual ranged from ~2 % to 40 % (unpublished data).

Examination of ROH overlapping consensus regions

Genes under putative selection within populations

The strongest iHS signal (p< 0.001; iHS > |4|) across all three populations was on chromosome 18 in variants annotated to intronic regions of the protein kinase C alpha (PRKCA) gene (Figs. 3 and 4) (Additional file 2: Table S2). It was the strongest overall signal in the Ugandan birds, and second strongest signal in the Rwandan and Kuroilers (p< 0.001; iHS > |4|). There were also strong selection signals that were unique to each of the three populations. In the Uganda ecotypes the gene cyclin-dependent kinase inhibitor 3 (CDKN3) on chromosome 5 had the next highest iHS value (iHS = |4.82|) with 41 statistically significant variants (p < 0.001). The gene deleted in liver cancer 1 (DLC1) on chromosome 4 had the highest iHS value in the Rwandan birds (iHS = |4.29|) supported by 44 statistically significant variants (p < 0.001). In Kuroilers, variants annotated to adrenomedullin (ADM) (iHS = |5.79|) on chromosome 5 and glutamate-cysteine ligase, modifier subunit (GCLM) on chromosome 8 (iHS = |4.40|) had the highest iHS values (Additional file 3: Table S3).

Putative genes under selection between populations

The populations were also compared pairwise by means of the Rsb statistic to identify additional genomic regions under selection. Statistical significance (FDR < 0.05) and the highest Rsb value (>|7|) were used to determine genes under possible selection. The Ugandan ecotypes vs. the Rwandan ecotypes showed selective pressure on chromosome 13 for Gamma-aminobutyric acid receptor, gamma-2 (GABRG2). Also annotated to the highest Rsb value on chromosome 13 was the gene teneurin-2 isoform 1 (TENM2). On chromosome 11, Beta, beta-carotene 15,15'-monooxygenase (BCMO1) reached statistical significance along with having a Rsb above |7|. The Ugandan and the Rwandan ecotypes both yielded similar results when compared to the Kuroilers. The strongest selection occurred on chromosome 1 with the gene olfactomedin 4 (OLFM4).

Gene set enrichment analysis

Population structure and admixture of populations indicate mixed genetic backgrounds

The observed population structure could partially be a result of using IBS values, which are based on the likelihood that identical alleles inherited by two individuals came from the same parent. This can lead to some uncertainty due to not having pedigree information, but it still allowed us to capture information on the population structure [44]. The amount of overlap seen across ecotype and populations may stem from flock management that allows unrestricted inter-mating of chickens from different genetic backgrounds. The admixture may be due to individuals having ancestors in multiple source groups that contribute to the shared ancestry. Another contributor to this admixture may be related to the movement of trade and selection parameters used by farmers to purchase new birds at market.

Possible factors contributing to signatures of selection

There are multiple environmental challenges that could lead to the occurrence of the discovered selection signals in the three populations. The geographic location of Uganda and Rwanda is near the equator and in tropical climates (Additional file 5: Figure S5). This, coupled with smallholder farm practices, places the studied birds in situations in which they would likely contact multiple environmental stressors that may affect their fitness. This would include challenges to their immune system from pathogens in the absence of vaccination. Also the equatorial locations of the samplings were exposed to high ultraviolet rays. The birds may also have experienced nutritive stress brought on by sub-optimal food in an environment in which birds must forage for food. The birds used in this study would have adapted to the environment over the years, likely through changes in allele frequency of a beneficial or detrimental allele.

Evidence of common signatures of selection across ecotypes for stress response

In livestock species, as well as humans and mice, environmental stressors such as pathogens or non-optimal temperatures can create cellular oxidative stress by generating reactive oxygen and nitrogen species effecting calcium signaling, apoptosis, vascular plasticity, growth and immune functions [16, 22, 45–50]. The chickens in our study displayed evidence of population similarities in how selection may have affected their response to environmental stress. Signatures are related to genes and signaling pathways involved in the reduction of ROS through utilization of Ca²⁺, lipids, and phosphorylated kinases. If the mobilization of Ca²⁺ is part of a prolonged response to chronic stressors it is possible that birds that scavenge for food have to make trade-offs in Ca²⁺ usage. An eggshell can contain a concentration 1.99 M Ca²⁺ [51]. Signal transduction of Ca²⁺ stored in the endoplasmic reticulum can be identified by increased cytoplasmic Ca²⁺ concentrations of 300–500nM [52]. The Uganda population showed GO terms related to endoplasmic reticulum (ER) stress response where Ca²⁺ is stored. Despite being such a small amount (300–500 nM) this diversion of resources may hinder other functions dependent on calcium. This mobilization of calcium for stress tolerance could also reduce the ability of birds to produce structurally competent shells and limit egg production. Pathways affected by calcium were also upregulated in heat stressed broilers [53]. We also uncovered selective pressure on negative regulation of multiple kinases involved in stress responses that may function to reduce pro-inflammatory cytokines. This reduction of pro-inflammatory cytokines had a lowering effect on metabolism, reducing stress during an immune response [10, 46, 54–56]. The most marked effect of the selection was the enrichment of terms related to response to stress, especially those indicating selection for or against cellular, endoplasmic reticulum (ER), and oxidative stress [46, 54, 57]. Other processes that reached significance include cell-cell signaling (all populations), VEGF signaling, cytokine binding/production (Uganda and Rwanda), and inflammatory response (Rwanda) (Tables 3, 4 and 5). Selection on the genes and processes reported may be related to cyto-protective effects being used by chickens to tolerate heat as a general stress response, because it is a constant in their environment [58, 59].

Selection pressure identified by ROH analysis indicate similar biological processes between populations

The results from the ROH analysis indicate that the populations have experienced selective pressure on their genomes from the environment. Runs of homozygosity (ROH) are defined as long stretches of homozygous genotypes within a genome, thought to be the result of consanguinity and identity by descent (IBD) inheritance [44, 60–62]. It is possible that ROH carry genes or variants under positive selection or that smaller ROH are related to the hitchhiking effects of selective sweeps [60, 63]. Analysis of ROH has been used to uncover alleles detrimental to health in humans and livestock [61, 63–65]. However, it may be possible that the ROH shared between the three populations point to conservation of alleles necessary for survival. These may be alleles that provide protection from oxidative and other stressors due to the environment. The existence of statistically significant hits for calcium ion activity and apoptotic regulation may, respectively, be related to kinase activation and amino acid recycling in the birds as a means of maintaining energy to put toward redox functions. It is possible that the birds are diverting resources to autophagy to conserve resources to put into increasing calcium activity to lower metabolic temperatures. This may be possible since it has been shown that blood Ca²⁺ levels are inversely related to body temperature [66].

Genes putatively under selection from environmental challenges among and within ecotypes

Protein kinase C alpha (PRKCA) found on chromosome 18 appears as a strong signal in all three populations. The gene PRKCA is a calcium activated, phospholipid dependent kinase that is involved in gene expression, inflammation, prolactin secretion, as well as, the regulation of multiple cell processes including inflammation and wounding [67, 68] (Fig. 4). It is also upregulated in response to endothelial injury. In chicken, PRKCA carries out functions involved in gene expression and kinase activity, and is involved in prolactin secretion [67, 69]. The selective pressure on PRKCA may be related to the negative effect that a challenging environment has on overall body weight and feed efficiency [22, 70], which could indicate nutritive stress. In addition to its functions in livestock, PRKCA acts as a messenger to stimulate prolactin secretion [71], which in humans has been linked to psychosocial stress responses [72].

Protein kinase C alpha is also involved in the biological process of prolactin secretion (GO:0070459) possibly through protein-protein interactions with the prolactin receptor (PRLR). This is interesting due to recent studies showing evidence of PRLR’s connection to heat tolerance in livestock [73, 74]. The interaction between PRKCA and PRLR indicates that both genes may contribute to the birds’ response to environmental stressors. In humans, PRKCA regulates induction of NF-kappa-B inhibitor alpha (NFKBIA/IKBA), which mediates host defense and inflammation responses, as well as plays a role in the activation of MAPK signaling and cyclin-dependent kinase (CDK) complex formation. Previous studies have indicated a role of MAPK and NF-κB pathways in the regulation of heat [70]. Regulation of NF-κB may be part of the response to inflammation triggered by ROS. The possible reduction in metabolic heat by PRKCA may be related the protein-protein interactions it has with suppressor of cytokine signaling 1 (SOCS1) and suppressor of cytokine signaling 3 (SOCS3) (Fig. 5). Both SOCS1 and SOCS3 help to battle ROS and inflammation and may play a role in the cascade needed to restore lost homeostasis [53, 75, 76]. Other SOCS genes have also been shown to respond under heat stress, which is a possible indication of a link between heat stress tolerance and cytokines [53].

The other kinase, cyclin-dependent kinase inhibitor 3 (CDKN3) found on chromosome 5 only appears as a strong signal in the Ugandan ecotypes. As a dual specific protein phosphatase, CDKN3 functions to inhibit cell cycle processes by blocking kinases targeting cyclin-dependent kinase 2 (CDK2), which causes cell cycle arrest and apoptosis in response to stress signals. The CDK2 gene is also part of the FoxO signaling pathway, which regulates transcription factors involved in gene activation of physiological cellular processes such as apoptosis, glucose, oxidative and nutrient stimuli stress resistance. Another function of CDKN3 is its ability to associate with cyclin C for activation and regulate different cell cycle events [67, 69, 77]. The selection pressure on CDKN3 in the Ugandan ecotypes may be related to its part in cell cycle regulation and ability to interact with phosphotyrosine or phosphoserine residues [67]. It functions to inactivate CDK2 which is part of a pathway engaged in DNA damage/telomere stress induced senescence, cell cycle arrest, and apoptosis in response to stress signals. To prevent senescence caused by oxidative stress, CDKN3 may function to inhibit CDK2. Studies have shown that oxidative stress can trigger activation of Protein Tyrosine Kinases (PTKs), which leads to the release of intracellular calcium in in vitro chicken cells and rat macrophages [49, 50]. The gene CDKN3 has also been shown to be considerably down regulated in mice undergoing oxidative stress from simulated environmental exposures [78].

In addition to the strong signal on chromosome 18, the Rwandan ecotypes also had a strong signal on chromosome 4 variants connected to the Rho GTPase activating protein, DLC1. This gene has functions involved in lipid binding and Rho GTPase activator activity and is also involved in negative regulation of stress fiber assembly. The gene DLC1 is also involved in the activation of cysteine-type endopeptidase activity involved in the apoptotic process, as well as positive regulation of the execution phase of apoptosis [67, 69, 77, 79]. For the Rwandan ecotypes, DLC1 may be involved in a population-specific mechanism of increasing cell migration and expanding cellular redox and lipid binding capabilities [67, 79, 80].

The gene with the strongest signal in the Kuroilers, ADM, is a hypotensive peptide that functions in hormone activity and in chicken is part of the biological processes of vasculogenesis (GO:0001570), positive regulation of angiogenesis (GO:0045766), and negative regulation of vasoconstriction (GO:0045906) [67, 69, 77, 81]. It also protects endothelial cells from cardiac stress [21]. In humans it is a vasodilator with functions that regulate fluid and electrolyte homeostasis and also has a role in the biological processes of response to cold (GO:0009409), hypoxia (GO:0001666), starvation (GO:0042594), LPS (GO:0032496), and wounding (GO:0009611) [45, 67, 82]. Other studies have shown ADM to also have anti-microbial, anti-apoptosis, and antioxidant functions [45, 67, 82, 83]. In the chicken, ADM has strong vascular modulation and functions as an antioxidant. Oxidative stress increases expression of inflammatory responses by endothelial cells [45, 68, 82] and ADM is shown to protect cardiovascular cells from oxidative damage [82, 83]. The selective pressure on this gene seen in the Kuroilers does not occur in the African ecotypes and may represent a result of artificial selection for stress tolerance in the Kuroilers during their development in India. The gene GCLM in chickens is part of the response to oxidative stress (GO:0006979), apoptotic mitochondrial changes (GO:0008637), and negative regulation of extrinsic apoptotic signaling pathway (GO:2001237). In humans, GCLM is also part of the response to nitrosative stress (GO:0051409) [67, 69, 77]. Nitrosative stress is caused by the formation of reactive nitrogen species (RNS) from cellular nitric oxide (NO) or it reacting with oxidative stress molecules to inflict cellular and vascular damage [84]. Other genes of interest that were in statistically significant regions of the genome based on the iHS values included DnaJ (Hsp40) homolog, subfamily C, member 5 (DNAJC5) and Collagen alpha-2(VI) chain (COL6A2). The heat shock protein DNAJC5 appeared in all 3 populations and functions as a molecular chaperone and a negative regulator of neuron apoptosis. It is also part of the GABA synthesis, release, re-uptake and degradation pathway and the protein processing in endoplasmic reticulum pathway. Collagen alpha-2(VI) chain (COL6A2) only appeared in the Kuroiler samples and was the only stop-gain (high impact) variant to reach statistical significance based on the iHS analysis. It is involved in focal adhesion and is part of the collagen biosynthesis and modifying enzymes, collagen degradation, and integrin cell surface interaction pathways [67, 69].

Pairwise comparisons of populations reinforced selection toward genes and functions related to oxidative stress

The comparison between the Ugandan and Rwandan ecotypes pointed selective pressure on gamma-aminobutyric acid A receptor, gamma 2 (GABRG2), an inhibitory neurotransmitter in vertebrates that mediates neuronal inhibition by binding to GABA receptors and opening an integral chloride channel. This is consistent with the results we observed in DNAJC5. Gamma-aminobutyric acid A receptor, gamma 2 (GABRG2) is also involved in the response to hypoxia in chickens [67, 69, 77]. The gene teneurin transmembrane protein 2 (TENM2) also plays a part in neural development and regulation of proper nervous system connections and carries out calcium-mediated signaling using intracellular calcium sources, hemophilic and heterophilic cell-cell adhesion [67, 69, 77]. The calcium functions of TENM2 may help to stimulate the necessary calcium release to activate PRKCA into activation when stressors appear. On chromosome 11, beta, beta-carotene 15,15’-monooxygenase (BCMO1), functions as a mono-oxygenase activator involved in cellular redox reactions as scavengers of oxygen radicals for photo-protection [67, 69, 77]. BCMO1 might experience selection related to protection from uv-induced DNA damage related to the generation of oxidative stress. When Ugandan and Rwandan ecotypes were compared to Kuroilers, the genes with the strongest signals for selection were the same but had different markers for the same gene regions in each comparison. Olfactomedin 4 (OLFM4) functions as a negative apoptotic factor and an extracellular matrix glycoprotein involved in cell adhesion. In chickens, it negatively regulates I-kappaB kinase/NF-kappaB signaling and the immune response. The between-population results for GO enrichment also reinforce the GO term results seen within population with slightly more emphasis on signaling, activation, and transport functions.

Unique and shared features under selection in Kuroilers compared to native African ecotypes

Kuroilers showed a great deal of overlap with the African ecotypes in biological processes that reduce the effects of oxidative and metabolic stress. Unique to the Kuroilers were selection on prostaglandin-endoperoxide synthase activity, a target of NF-κB and negative regulation of interleukin-10 (IL-10) production. Both are a part of reduction of oxidative stress and have anti-inflammatory effects. The differences in the guided selection of the Kuroilers can be seen in the many growth and behavior related GO terms that are enriched within the Kuroiler population and not observed in the African ecotypes. In Kuroilers, the GO enrichment results from the iHS analysis have larger gene lists, which may signify that under artificial selective breeding for stress tolerance, larger genomic areas or QTL regions may have been selected upon. This may be the opposite of natural selection in the African ecotypes, which may be focused on smaller regions of the genome. This is reflected in the genes annotated to similar terms seen in the Ugandan and Rwandan populations related to stress, inflammatory responses, and apoptosis. The length of the natural selection experienced by the African populations may have affected this also. This would have led to more recombination events, leading to smaller genomic regions under selection than what was identified in the Kuroilers, which is a recently developed breed. It appears that all three populations have found ways through natural and artificial selection to tolerate environmental stressors. What this overlap in GO terms also shows is that both the artificial selection for stress tolerance (Kuroilers) and natural selection (Ugandan and Rwandan ecotypes) share a biological link. Previous studies showed evidence that exposure to one stressor, such as heat, can lead to protection from other stressors through cyto-protective memory present in the activation of protective signaling [58, 59, 85]. It is possible that the underlying purpose of the selection seen in the populations are involved in histone and transcriptional modifications leading to cross-tolerance brought about by numerous small adaptions to a challenging environment. The regulation of these types of signaling processes (i.e. MAPK, NF-κB) and the initiation of regulatory kinases and genes (PRKCA, CDKN3, ADM, OFLM4) that assist in activating processes that protect cells from oxidative damage. Selective pressure on PRKCA also indicates the presence of prolactin signaling, shown to be a heat stress related signaling system in cattle. There is also evidence of selective pressure for protection of cardiovascular health under metabolic stress. In the sampling populations of Uganda and Rwanda there was some indication of adaption to environmental challenges from heat present in the some of the observable phenotypes of the birds. Heat stress related phenotypes such as frizzled and Naked Neck were observed within the sampling populations.