Genomic insights into local adaptation of indigenous chickens

Abstract

Indigenous chickens are an essential part of biodiversity and a vital protein resource to humans, yet global warming and environmental changes pose serious threats to their survival and productivity. Therefore, assessing population adaptive capacity under shifting environments is crucial for breeding resilient animals, and guiding conservation strategies. Here, we integrated ecological and whole-genome resequencing data from 1022 chickens from 44 Chinese indigenous populations to reveal genomic signatures of local adaptation. From 87 agroclimatic variables, we identified eight dominant environmental factors including solar radiation, precipitation, diurnal temperature range, and five landcover variables (cropland areas, water areas, trees coverage, bare ground and shrubs coverage) that shape ecological niches of indigenous chickens. Landscape and comparative genomics analyses revealed both known and novel candidate genes, such as UNC80, PTPRO, NCOR2, CSF2RB, NXT2 and PALLD for the solar radiation, precipitation, diurnal temperature range, cropland areas, trees coverage and bare ground, respectively. Particularly, adaptive non-coding variants harbored in these genes exhibited spatial allelic changes across populations and acted as regulatory elements via chromatin accessibility and DNA methylation, influencing adaptation in a tissue-specific manner. Our findings underscore the rich genetic diversity of Chinese indigenous chickens and provide new insights into genomic mechanisms of local adaptation, offering valuable references for domestic animal breeding, conservation, and climate resilience.

INTRODUCTION

Chicken (Gallus gallus domesticus) is a ubiquitous livestock species, providing high-quality and affordable protein. In addition, it represents entertainment, cultural heritage, and scientific research value (Eda, 2021). Indigenous chicken populations inhabit diverse ecosystems and serve as valuable biomedical models. Their gene reservoirs are very critical for conservation biology and for long-term genetic improvement in response to changing breeding goals, such as adaptation to extreme climate conditions, disease resistance, and stress tolerance. However, the rapid expansion of commercial breeds has severely reduced the number of indigenous chickens, increasing their risk of extinction and the eventual loss of unique genotypes and adaptive traits (Padhi, 2016). This concern is amplified by the challenges of environment changing, underscoring the need to harness the genetic diversity of indigenous populations for climate-resilient agriculture (Meek et al., 2023).

Adaptive responses that enable survival and reproduction in extreme environments are often beyond the range of phenotypic plasticity. Locally adapted populations are therefore expected to experience strong selection and genomic modifications, resulting in changes in metabolism, physiology, and behavior (Rocha et al., 2021). Earlier comparative studies focused on single-gene or few-gene polymorphisms underlying adaptation, such as the heat shock proteins (HSPs) and heat shock factors (HSFs), which are related to the heat stress in animals (Balakrishnan et al., 2023; Gourdine et al., 2021; Hariyono & Prihandini, 2022; McManus et al., 2022). However, differential expression of candidate genes may evolve neutrally and not necessarily reflect adaptation. Meanwhile, causal variants may reside in distant regulatory elements or trans-acting factors, making it difficult to directly link genetic variation with expression data. Moreover, measuring and collecting fitness-related phenotypes is challenging due to long generation intervals, experimental constraints, and population-specific limitations (Savolainen & Lascoux, 2013). Recent advances in genome sequencing now allow affordable population-scale analyses, providing a powerful complementary strategy to study local adaptation across a species’ geographical distribution (Ellegren, 2014). Furthermore, new approaches including genotype-environment association analyses, landscape genomics, and multi-omics integration are increasingly applied for conserving species in wildlife species (Meek et al., 2023; Wang et al., 2025) and for breeding in crops (Campbell et al., 2025) and animals (Lai et al., 2025; Leng et al., 2024).

China’s vast geography spans wide gradients in latitude, altitude and topography, creating remarkable climatic and environmental diversity. This mosaic of environments provides a natural laboratory for studying ecological adaptation (Mi et al., 2021). Archaeological and genomic evidences indicate that domestic chickens originated from the red junglefowl subspecies (Gallus gallus spadiceus) in southwestern China ~10000 years ago (Wang et al., 2020a). Since then, more than 100 indigenous chicken breeds distributed across nearly all ecosystems have been developed across China (Zhang et al., 2023a). These populations represent an exceptional resource for investigating the molecular mechanisms of environmental adaptation. Previous studies examined chicken adaptation to high-altitude (Wang et al., 2015; Yuan et al., 2022), tropical and frigid (Shi et al., 2023; Tian et al., 2020), and desert or arid environment (Gu et al., 2020). However, the specific ecological environment is driven by many contributing factors, which were rarely investigated in the region-limited or single-variable studies. How geography and environmental complexity shape genetic variations in chickens is still fascinating.

To explore this, we sequenced and retrieved genomes of over 1000 chickens from 44 indigenous populations distributed across China, for which we characterized the functional relationship between genomic variants and agroclimatic variables. By integrating ecological and multi-layer omics analysis, we aimed to: 1) characterize the spatial patterns of genetic diversity and structure, 2) identify key environmental drivers of ecological niche partitioning, 3) dissect genomic signatures of local environment adaptation and allele frequency shifts across geographical ranges, and 4) elucidate potential regulatory mechanism linking adaptive variants to candidate genes.

MATERIALS AND METHODS

Chicken samples

A total of 1029 individuals from 44 domesticated chicken populations were used in our study. These chicken populations were distributed within three-quarters of China’s administrative divisions (26/34), representing most agro-ecological zones in China (Figure 1A; Supplementary Table S1). These areas encompassed diverse environmental conditions, including variations in temperature, solar radiation, precipitation, land cover and soil properties. Among these populations, a total of 735 blood samples from 20~30 female chickens per population were collected in 28 populations, while 294 individual sequencing data from 16 chicken populations were retrieved from previous studies (Gu et al., 2020; Huang et al., 2020; Li et al., 2023; Shi et al., 2023; Wang et al., 2020b; Xu et al., 2023; Zhi et al., 2023). Genomic DNA was extracted from the blood samples using DNeasy Blood & Tissue kit (Qiagen, Germany).

Figure 1.

Geographic location and genetic structure of Chinese indigenous chicken populations

A: Geographic origin of 44 chicken populations. Populations are grouped into six agroclimatic zones: North, Northwest, Middle, Southeast, Southwest, and Xizang. B: Principal component analysis (PCA) of the 44 chicken populations based on 1415317 pruned SNPs. C: Ancestry composition of 44 chicken populations inferred by ADMIXTURE analysis (K=2–45), with the optimal K as 27. D: Genetic divergence among populations measured by pairwise F_ST. The chicken populations are ordered by agroclimatic zone.

Genome sequencing, variant calling and quality control

DNA integrity was verified prior to the library construction. Paired-end libraries with insert sizes of 350 bp were constructed and sequenced on the BGISEQ-500 platform (BGI, China) to generate 150 bp paired-end reads. Low quality reads were filtered using fastp v.0.20.1 (Chen, 2023), and the clean reads were aligned to the chicken reference genome (GRCg7b) using BWA v.0.7.17 (Li & Durbin, 2010) with the default parameters. The Genome Analysis Toolkit GATK v.4.6.0.0 ‘MarkDuplicatesSpark’ module was subsequently used to filter out potential PCR duplicate reads (McKenna et al., 2010). The resulting alignments were reordered using Picard and then indexed using SAMtools v.1.13 (Li et al., 2009). SNP calling was processed according to the best practices for GATK by invoking ‘HaplotypeCaller’, ‘CombineGVCFs’ and ‘GenotypeGVCFs’ modules. We set a minimum quality score of 30 for both bases and mapped reads to call variants, and then the common SNP dataset was subjected to filter with the following rigorous criteria using the GATK ‘VariantFiltration’ module: (a) quality by depth<5.0, (b) mapping quality score<40.0, (c) strand odds ratio>4.0, (d) Fisher strand>60.0, (e) MQRankSum<–12.5, (f) ReadPosRankSum<–8.0, and (g) any three SNPs clustered in a 10 bp window. The high-quality SNP sets from 1 029 birds was filtered using PLINK v.1.90 software (Chang et al., 2015) with the following parameters: sample call rate>90%, SNP call rate>90%, minor allele frequency (MAF)>0.05 and Hardy–Weinberg equilibrium P-value<1e–5. After these steps, a total of 7276708 SNPs distributed across the 39 autosomes in 1022 chickens (seven chickens were filtered out due to low sample call rate) were obtained for downstream analyses. Genotypes were subsequently phased and imputed using Beagle v.4.1 (Browning et al., 2021) with default parameters to obtain genotyping rate and accuracy of 100%.

Population genomic analysis

The imputed genome data were first pruned with the “indep-pairwise 25 5 0.2” using PLINK v.1.90, yielding 1415317 independent SNPs for population structure analysis. Principal components (PC) analysis was carried out on the pruned SNPs using the “pca” in PLINK v.1.90, and the first two PCs were extracted and visualized using R software. We used ADMIXTURE v.1.3.0 (Alexander et al., 2009) with default parameters to investigate population structure across all samples, with the number of clusters (K) being set from 2 to 45. By using pruned SNPs, we also constructed the Neighbor-Joining (NJ) phylogenetic tree based on the distance matrix using FastME 2.0 software (Lefort et al., 2015), and the visualization of tree was performed using Interaction Tree Of Life (iTOL, https://itol.embl.de). Linkage disequilibrium (LD) decay was assessed using PopLDdecay (Zhang et al., 2019). For the isolation-by-distance (IBD) analysis, VCFtools v.0.1.15 (Danecek et al., 2011) was applied to calculate the population differentiation coefficient (F_ST) using 50 kb sliding windows with 10 kb increments at each step. The matrix of F_ST/(1-F_ST) and the matrix of geographic distance (km) among populations were used to perform the Mantel tests, implemented in the vegan R package (Oksanen et al., 2018) with the significance determined based on 9999 permutations.

Environmental data acquisition and analysis

For each of the 44 chicken populations, the central coordinates of their original habitats (conservation farms or marketplaces) were recorded, based on. To better capture the local environmental conditions, nine additional geographic coordinates were selected within 2 km grids surrounding each population site using Google Earth Pro v.7.3. This resulted in 440 coordinate points from all populations, which were considered as “occurrence points” to predict the potential geographic distribution for chicken populations. Based on the occurrence points, we used the raster R package to obtain 87 agroclimatic variables, including 67 bioclimatic, eight landcover, one topographical and 11 pedologic variables (Supplementary Table S2) from the WorldClim (Fick & Hijmans, 2017), WorldCover (Zanaga et al., 2022) and ISRIC (https://www.isric.org/) databases at a spatial resolution of 30 arc-seconds. These variables reflect ecological factors relevant to chicken survival, such as climate and elevation (physiological tolerance), soil characteristics (food resources), and land cover (food availability and predation risk). After removing two variables lacking records within the occurrence ranges, the remaining 85 environmental variables were fed to R package ENMeval for model evaluation (Kass et al., 2021). Models were tuned with the maxnet algorithm and randomkfold partitions, testing regularization multipliers (RM) from 0.5 to 6.0 (increments of 0.5) and different feature class (FC) combinations: L=linear, Q=quadratic, H=hinge, P=product, T=threshold. The best FC and RM selected based on ΔAICc were applied in the Maxent software (Fourcade et al., 2014) to estimate the contribution of each variable. To improve the model's robustness and biological interpretability, variables with contributions >1% were retained for further analysis. To reduce collinearity, pairwise correlations among variables were calculated, and the variable with lower contribution was removed for correlated pairs (r>0.6) (Gheyas et al., 2021). Variables with variance inflation factor (VIF) <5 calculated using the usdm R package, were retained. The relative contribution and permutation importance of the retained variables were re-evaluated by Maxent with best model parameters output from ENMeval.

In the Maxent, 75% of occurrence points were used for the model training and the rest 25% were used as model testing data according to Guisan and Thuiller (Guisan & Thuiller, 2005). The remained environmental variables were used for predicting the environmental suitability of chicken populations. The model was replicated 10 times according to the subsample method to verify the prediction accuracy, and the area under the receiver operating characteristic curve (AUC) was used to evaluate the model’s performance, where an AUC value of 0.9~1.0 indicates a perfect prediction and value <0.5 represent a random prediction (Elith et al., 2011). The Jackknife test of each environmental variable was used to evaluate its importance. Finally, the environmental variables with contribution rate larger than 4% were considered as important factors for exploring the distribution of indigenous chickens (Gheyas et al., 2021). The continuous probability of distribution was divided into four groups using the reclassify tool in ArcGIS v.10.2 according to the previous studies: 0–0.05 (unsuitable due to severe environmental limitations), 0.05–0.33 (moderate environmental limitations), 0.33–0.66 (suitable without environmental limitations), 0.66–1.00 (optimal environmental conditions for local chicken populations) (Yu et al., 2022).

The environmental values for each population were calculated by the raster R package. By ranking each environmental variable, populations in the top three or bottom three for each variable were designated as “extremely high” (H group) and “extremely low” (L group) groups, respectively. For some variables, all populations with zero values of the variable were divided into “extremely low” group. Then, the niche differentiation analyses were performed between the two groups using the ENMTools v.1.0.4 R package (Warren et al., 2021). Niche overlap was quantified by Schoener’s D and Hellinger’s I, where a value of 0 suggests no overlap and 1 indicates complete overlap.

Genotype-environment association analysis

In this study, we applied two different methods to identify genomic variants associated with environmental variables. Firstly, we conducted redundancy analysis (RDA) to identify genomic variants showing significant relationships with multiple environmental axes. RDA has been shown to be one of the best genotype-environment association analysis (GEA) approaches for large SNP sets and exhibits low false positive rates (Capblancq & Forester, 2021). Herein, the partial RDA analyses was implemented using the R package vegan, with environmental variables specified as explanatory factors and genotypes as response variables, while controlling for latitude, longitude, and the first three principal components (PCs) derived from principal component analysis (PCA) of the genomic data. The significance of model constraints was assessed by using ANOVA like permutation test (anova.cca). Then, we selected SNPs that distributed in the two-tails of normal distribution of SNP loadings using a standard deviation cutoff of 3 along each significant RDA axis as significant environment-associated variants, which were further screened based on largest third-quantile value among all variables. Second, we employed a univariate latent-factor linear mixed model (LFMM) to detect associations between whole genomic variants and the eight environmental variables. The method is good at controlling type I errors by estimating residual structure. LFMM analysis was implemented in the R package LEA v.3.3.2 (Caye et al., 2019). By setting the number of ancestry clusters inferred from ADMIXTURE as latent factors, we computed regularized least squares estimates for LFMM to account for population structure using a ridge penalty. Then, significance thresholds for the associations were calibrated by using the genomic control method. The calibrated P-values of each variant were adjusted for multiple tests using a false discovery rate (FDR) correction, with FDR<0.01 considered significant.

The associated SNPs detected using both RDA and LFMM for each variable were united as the adaptive variants, which were used to investigate the role of geography and environment in shaping spatial genetic variations, compared to neutral (the 1415317 LD-pruned SNPs) variants (Sang et al., 2022). Mantel tests were used to test for correlations among F_ST(F_ST/1−F_ST), geographic distance (isolation-by-distance) and environmental distance (isolation-by-environment), in which environmental distance was represented by Euclidean distance of all scaled environmental variables. The significance of correlations was determined using 9999 permutations performed in the R package vegan.

Selection signature analysis for chicken populations from extremely high and low environmental conditions

Selection signature analysis (SSA) was performed using two complementary approaches, in which fixation index (F_ST) analysis compared the frequency of differential alleles between population groups in extremely high versus extremely low environmental conditions (Holsinger & Weir, 2009), and cross population extended haplotype homozygosity (XP-EHH) was designed to identify genomic regions under selection by comparing haplotype homozygosity between two groups (Sabeti et al., 2007). F_ST value was calculated in VCFtools v.0.1.15 using the Weir and Cockerham method (Weir & Cockerham, 1984), with a 50 kb sliding window and a 10 kb step size. The weighted F_ST were then standardized to zF_ST, and genomic regions within top 1% zF_ST values were retained for further analysis. Prior to the XP-EHH analysis, genotypes of high- and low-environmental value groups were phased using beagle v.4.1 software, respectively. XP-EHH analyses were carried out using selscan v.2.0.3 software (Szpiech & Hernandez, 2014) to obtain the XP-EHH score for each SNP, and then the standardized XP-EHH (XPEHH_std) and its absolute value of XPEHH_std () were calculated. In order to integrate the results of the two methods, were averaged across the same windows used in the F_ST analysis, and genomic regions within top 1% values were selected for further analysis. Based on the zF_ST and , the correlations between F_ST and XP-EHH were calculated using Pearson correlation method. To mitigate potential confounding effects of population structure and demographic history on SSA outcomes, we combined three chicken populations each in high and low-environment value group (Gheyas et al., 2021). Furthermore, the SHRUBS and WATER were excluded from SSA due to that the environmental values between two groups were very similar, and the number of populations with extreme values exceeded three per group.

Function annotation for candidate genes

For the genotype-environment association analysis (GEA), the adaptive SNPs were annotated using SnpEff v.4.3 (Cingolani, 2022). Genomic regions located in the 100 kb upstream and downstream of adaptive SNPs were annotated using Bedtools v.2.30.0 with the GRCg7b genome annotation. For SSA, genomic regions identified by F_ST and XP-EHH analysis were annotated using Bedtools v.2.30.0 with GRCg7b genome annotation. Finally, the common genes identified by both GEA and SSA were considered as candidate genes associated with environmental adaptation. Functional annotation and enrichment analysis of genes were performed with Gene Ontology (GO) knowledge base and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway database using clusterProfiler package (Wu et al., 2021) implemented in R. False discovery rate (FDR) method was used to adjust the P-values from the hypergeometric test. GO terms and KEGG pathways with FDR<0.05 were considered significantly enriched.

Identification of functional SNPs

To further characterize adaptive genomic regions, we used VCFtools v.0.1.15 to calculate nucleotide differences (θπ), genetic diversity (Tajima’s D) and fixation index (F_ST) statistics. The basic parameters were set to a 15 kb sliding window with a 15 kb step for calculating θπ. Tajima’s D was calculated in a 15 kb window and F_ST was calculated for every single SNP harbored in candidate genomic region. Linkage disequilibrium (LD) and haplotype block structure were evaluated with r² statistic and visualized with LDBlockShow (Dong et al., 2021) for SNPs and R package locuszoomr for genomic regions (Lewis & Wang, 2025), respectively. The haplotype network analysis was performed using R package geneHapR by removing individuals with heterozygous sites in the region (Zhang et al., 2023b).

To investigate the tissue-specific regulatory pattern of selected variants, we integrated 15 chromatin states including promoters (TssA, TssAHet, and TssBiv), TSS-proximal transcribed regions (TxFlnk, TxFlnkWk, and TxFlnkHet), enhancers (EnhA, EnhAMe, EnhAWk, EnhAHet, and EnhPois), repressed regions (Repr and ReprWk), quiescent regions (Qui), and accessible but did not coincide with any other measured epigenetic marks (ATAC islands) which covered 23 tissues including adipose, bursa of Fabricius, bone marrow, cecum, cerebellum, colon, cortex, duodenum, gizzard, hypothalamus, ileum, jejunum, liver, lung, muscle, proventriculus, spleen, eggshell gland, testis, trachea and thymus from publicly available database (Pan et al., 2023). The selected SNPs were also annotated with signal peaks including four histone modifications (H3K4me3, H3K4me1, H3K27ac and H3K27me3), CTCF, ATAC-seq and DNase-seq. Meanwhile, gene expression profiles across tissues were retrieved from ChickenGTEx database (Hou et al., 2026). To ensure consistency with our genomic dataset, the genomic coordinates of regulatory elements and functional features were converted from GRCg6a reference genome to GRCg7b reference genome using LiftOver with custom chain files, which were generated through whole-genome alignment using the transanno toolkit (https://github.com/informationsea/transanno) with minimap2chain module.

RESULTS

Population structure and genetic diversity of local chicken populations

In the present study, a total of 1029 chickens from 44 populations were investigated across China, with sampling sites categorized into six agroclimatic areas including the North (N), Northwest (NW), Middle (M), Southeast (SE), Southwest (SW) and Xizang (T) (Figure 1A). Whole-genome resequencing data of these chickens were generated using next generation sequencing (NGS) techniques on BGI and Illumina platforms. By aligned onto the chicken reference genome (GRCg7b), average>99% of the clean reads were mapped with an average depth of 14.40× and coverage of 96.23% (Supplementary Table S3). Among 35739816 high-quality SNPs across the 1029 samples, the majority of substitutions were A/G (37.20%) and C/T (37.19%) transitions (Supplementary Figure S1A). Rare SNPs with MAF<5% accounted for 60.41% of total variants, with most ranging between 0.01 and 0.05 (Supplementary Figure S1B), suggesting population-specific pattern of SNPs distribution. we performed principal component analysis (PCA) for the 1022 qualified samples based on 1415317 pruned autosomal SNPs. Along the first PC, the clusters transition from negative to positive values representing geographic order as Southwest (SW), Southeast (SE), Middle (M), North (N), Northwest (NW) and Xizang (T). The PC2 largely reflected the difference between chicken populations in North and South China (Figure 1B). Through deep inspection, we found chicken populations in SE separately clustered, whereas several chicken populations overlapped with others in other regions, such as the LDC, DGC and BSC in N; HDC and NYC in NW; LSC, GSC and MCL in M; NDC, HTC, GXS, HXC and WCC in SE; XZJ and LZC in Xizang (Supplementary Figure S2). The neighbor-joining (NJ) tree revealed similar results (Supplementary Figure S3).

To investigate population structure, we inferred ancestry proportions for all samples using ADMIXTURE. The number of ancestors that ADMIXTURE converged with minimum cross-validation (CV) error was 27. We found that BJC, BYC and JYC (N), LHC and XHC (M), PDC, LYC, XSC, JHW and HTC (SE), LSF, DWS and GZC (SW), and RKC (T) showed high genetic purity, whereas all NW chicken populations exhibited high admixture. Interestingly, populations in central China (LSC, GSC, MCL, HJC, DAC, and HLC) also showed similar admixture patterns, suggesting that extensive gene exchange existed throughout the history of these breeds (Figure 1C). We also assumed six agroclimatic areas as the six ancestries for all chickens and found that LHC (M), WBD (M), XSC (SE), DWS (SW) and RKC (T) were genetically independent. The remaining ancestry showed significantly different patter between chicken populations in Southwest and other part of China (Supplementary Figure S4). Moreover, the substantial admixture increased the genetic diversity of chicken populations, among which the physical distance that linkage disequilibrium (LD) decreased to half value (half-life) for pure populations (254 nt) were higher than that of populations with high admixture (194 nt) (Supplementary Figure S5).

The genetic differentiation among the 44 populations was assessed using weighted F_ST. The F_ST value ranging from 0.01 to 0.25 (mean=0.09) indicated a moderate genetic differentiation among populations. As shown in Figure 1D, the highest F_ST (0.25) was found between LHC and WBD. We observed that populations with high purity were significantly divergent from other populations (average F_ST=0.07–0.16), and the divergence among admixed populations showed area-specific differentiation. For example, the LDC and DGC (0.04), the NYC and XWC (0.04), the LSC, GSC, MCL, HJC, DAC and HLC (0.01–0.06), the HXC, GXS and WCC (0.01–0.03), the PDG and CSL (0.04) and the XZJ and LZC (0.03) were slightly divergent (F_ST=0.01–0.06) within each area compared to that among areas (F_ST=0.01–0.19). This was consistent with the clusters identified in the PCA. Although CLZ exhibited an admixed genetic structure, it was highly divergent from all other populations with F_ST ranging from 0.09 to 0.22, exceeding divergence seen in some purebreds like JYC, LYC, JHW, HTC, LSF and RKC from other populations. This suggested that CLZ represented an independent genetic lineage shaped by human migration and unique environmental conditions in the edge of the Qinghai–Xizang Plateau. Furthermore, the genome-wide estimates of nucleotide diversity (π) revealed that the π of CLZ was lower than almost all populations except for WBD and XSC (Supplementary Figure S6). Combined with high half-life LD decay of CLZ, this suggested that CLZ is a potentially new genetic resource which undergone a long-term natural selection and domestication.

Environmental variables contribute to ecological niche of chicken populations

Ecological niche model (ENM) was applied to predict the distribution of each chicken population, following the conceptual model of environmental factors governing species distributions (Soberón & Osorio-Olvera, 2023). Firstly, we built the models with different tuning parameters and evaluated their performance and detected that the ΔAICc reached zero, when the feature class (FC) was set as linear, quadratic and hinge (LQH) with RM=1 (Supplementary Figure S7). Then, we selected a total of 22 environmental variables with contribution rate larger than 1% for the further analysis (Figure 2A). After removing the highly correlated variables (r>0.6) and VIF>5, nine variables including BARE (areas with minimal vegetation), BIO_2 (mean diurnal range), CLAY (proportion of clay particles), CROPLAND (land used for crops), SRAD_11 (solar radiation in November), PREC_08 (precipitation in August), TREES (land covered by tree foliage), SHRUBS (woody plants) and WATER (areas covered by lakes and rivers) were fed to the ENMeval for automated tuning and evaluations of ENM. The optimal settings for ENM included five FCs (linear, quadratic, hinge, product and threshold) and a regularization multiplier value of 0.5 (Supplementary Figure S8), suggesting that the optimized model had a better fit and transferability to reduce overfitting of the chicken distribution model compared to the default parameters (FC=LQHPT, RM=1). After running 10 replicates under the optimized parameter settings, the average AUC values for Chinese local chicken reached 0.95, and average omission rate was closed to the predicted omission (Supplementary Figure S9), indicating the high accuracy and effectiveness of the optimized MaxEnt model in predicting the environmental suitability of the local chickens. The Jackknife test revealed that BARE, BIO_2, CROPLAND, PREC_08 and SRAD_11 with test gain >0.4 were the primary variables influencing the distribution of local chickens, whereas the CLAY contributed little in both regularized gain and AUC (Supplementary Figure S10). The relative contributions of each variable to the Maxent model showed that SRAD_11 contributed largest to the ecological niche of chicken populations (28%), followed by the CROPLAND (27%), WATER (12.5%), PREC_08 (9.1%), BIO_2 (6.5%), TREES (5.2%), BARE (4.7%), SHRUBS (4.2%), and the CLAY had the smallest contribution (2.9%) and permutation importance (1.7%) (Figure 2B).

Figure 2.

Ecological niche modeling (ENM) for Chinese indigenous chicken populations

A: Twenty-two environmental variables that contribute more than 1% to the distribution of indigenous chickens predicted by ENM using all 87 environmental variables. B: Relative contribution and permutation importance of nine environmental variables selected based on correlation (r≤0.6) and variance inflation factor (VIF≤5). C: The suitability maps of indigenous chicken populations in China generated by ENM using the nine selected environmental variables. D–I: Suitability maps for representative chicken population: BJC (D) in the North, CLZ (E) in the Northwest, LHC (F) in the Middle, XSC (G) in Southeast, DWS (H) in the Southeast and RKC (I) in Xizang, predicted from eight key variables (SRAD_11, CROPLAND, WATER, PREC_08, BIO_2, TREES, BARE and SHRUBS). The details of the variables, refer to Supplemental Table S2. Green star denotes sampling location. J: Niche identity tests between groups of populations from extremes (high vs. low) for each environmental variable. In each group, three or more chicken populations are included based on the value of environmental variable. The arrows indicate the observed niche equivalency, and the histograms represent the simulated (expected) equivalency. All differences between the observed index and the expected index rejected the hypothesis that environmental niches between regions were identical (P<0.01).

Based on the nine environmental variables, we found the highly suitable regions for chicken populations were located in the Southwest and Southeast China (Figure 2C), where rich biodiversity and a long history of traditional poultry farming were previously recognized and inferred for survival and formation of diverse local breeds (Zhuang et al., 2023). Moreover, we explored the suitability map for each chicken population by excluding the CLAY due to its negligible effect. The AUC values of all distribution models exceeded 0.9 in the suitability analysis, and we found that the eight remaining environmental variables exerted different impacts on the distributions of chickens (Supplementary Figure S11). For example, the ecological niche of BJC (N) and XSC (SE) was mainly affected by BIO_2, which make them adapted to the north China (high diurnal range) and the east coast of China (low diurnal range), respectively (Figure 2D, G). CLZ (NW), LHC (M), DWS (SW) and RKC (T group) adapted to mountain areas, northern China plain, humid-hot areas of southern China and Xizang environment, respectively (Figure 2E, F, H, I).

By focusing on the eight remaining variables, we selected three populations each with extremely high and low value, respectively (Supplementary Figure S12), between which niche differentiation were tested. For the BIO_2, the extremely high (H) group included NYC, XWC and SNC with average value of 15.59, and the GXS, PDC and WCC with average value of 7.07 were selected as extremely low (L) group. Similarly, we extracted H and L group for the BARE (0.68 vs. 0.01), CROPLAND (0.62 vs. 0.003), PREC_08 (292.87 vs. 3.33), SRAD_11 (14389.97 vs. 8199.07), TREES (0.92 vs. 0), SHRUBS (0.04 vs. 0) and WATER (0.23 vs. 0), respectively. For all variables, the empirical values of Hellinger’s I and Schoener’s D were significantly lower than those expected from 100 pseudo-replicated simulations (Figure 2J), suggested a clear niche differentiation affected by selected environmental variables. However, observed I and D of SHRUBS and WATER were larger than that of other variables, indicating smaller differentiation. Overall, ENM revealed the environmental diversity of Chinese indigenous chicken habitats and identified eight important environmental drivers of local adaptation. The significant niche difference was consistent to the genomic divergence between chicken populations with extremely high and low environmental variables.

Genotype-environment associations analysis

By integrating eight environmental variables with genomic data from 1022 individuals, we performed genotype-environment association (GEA) analyses to detect the environment-associated genetic variants using RDA and LFMM methods. In the RDA, the overall model was significant (P<0.01) under 500 permutations, explaining 3.04% of the total genetic variance (Supplementary note 1). This result was consistent with previous findings in Ethiopia indigenous chickens (Gheyas et al., 2021), supporting the polygenic inheritance of adaptive traits since only a small proportion of genomic variants are expected to be associated with several environmental predictors (Forester et al., 2018). The eight RDA axes in the model were significant with permutation P-value<0.05, explaining all the variance captured by the RDA model (Supplementary Figure S13, Supplementary note 1). The PCoA showed that first two RDAs explained 36.75% (RDA1, 19.12%; RDA2, 17.63%) of the variation of environmental variables. The RDA1 was positively correlated with most variables, including BARE, CROPLAND, PREC_08, SHRUBS, SRAD_11 and WATER, whereas RDA2 was correlated with and TREES. BIO_2 and CROPLAND pointed to opposite direction, in which BJC was positively correlated with high BIO_2, whereas PDC was positively correlated with CROPLAND. BARE and TREES also pointed to opposite direction, in which TGC was positively correlated with BARE, whereas CLZ was positively correlated with TREES. XHC was found in the swampy wetland (Figure 3A). We identified 35 688 outlier SNPs (1862~6549 SNPs per environmental variable), with correlations ranging from 0.03 to 0.41 (Supplementary Figure S13; Supplementary Table S4). Using r>0.2 as threshold, we obtained 307, 812, 510, 613, 113, 883, 1693 and 1436 SNPs for BARE, BIO_2, CROPLAND, PREC_08, SHRUBS, SRAD_11, TREES and WATER, respectively. LFMM was a complementary univariate landscape genomic method to test associations between genomic variants and environment variables, while accounting for background population structure (Frichot et al., 2013). We identified a total of 205~19126 SNPs were significantly (FDR<0.01) associated with one or more environmental factors. The largest number of associated SNPs was found in BIO_2, whereas the BARE and SHRUBS were associated with only ~200 each. Moreover, a total of 1190 SNPs affected two or more environmental factors (Supplementary Figure S14; Supplementary Table S5).

Figure 3.

Genotype-environment associations analysis for eight important environmental variables

A: Principal coordinate analysis (PCoA) for eight important environmental variables based on the RDA1 and RDA2. Arrow length denotes the relative contribution of each environmental factor to the principal components. B: Scatter plot for SNP statistics of RDA correlation coefficients (y-axis) and LFMM P-value (x-axis) for each environmental variable. Dashed lines denote the threshold (r≥0.2 and FDR≤0.01). Green points indicate candidate SNPs identified by both RDA and LFMM. C: The number of environment-associated SNPs and the number of annotated genes for each environmental variable. D: Significantly enriched GO biological processes and KEGG pathways for genes associated with each environmental variable. E: Isolation-by-distance analyses for local chicken populations based on neutral and adaptive variants, respectively. The shadow of linear regression denotes the 95% confidence interval. F: Isolation-by-environment analyses for local chicken populations based on neutral and adaptive variants, respectively. The shadow of linear regression denotes the 95% confidence interval.

By integrating results from RDA and LFMM, we identified 13, 276, 66, 69, 39, 19, 307 and 500 overlapping SNPs, corresponding to 90, 1860, 546, 578, 315, 110, 2167 and 2865 genes for BARE, BIO_2, CROPLAND, PREC_08, SHRUBS, SRAD_11, TREES and WATER, respectively (Figure 3B, C; Supplementary Table S6). Gene enrichment analysis showed that BARE-associated genes were overrepresented in cytokine-cytokine receptor interaction and metabolic pathways including amino acid metabolism (xenobiotics by cytochrome P450, tyrosine, retinol and pyruvate), glycolysis and fatty acid degradation. CROPLAND-associated genes were overrepresented in biological processes, such as cellular response to chemical stimulus, response to bacterium and chemotaxis. Interestingly, WATER-associated genes were overrepresented in four of the biological processes identified in the CROPLAND, involving same genes including AvBD (avian β-defensin) gene family (1, 6, 8, 9, 10, 11, 13), DEFB4A and SBDS. (Figure 3D; Supplementary Table S7). The AvBD represented a key family of antimicrobial host defense peptides in birds and might also influence nutrient metabolism under varying nutrient availability, as indicated by CROPLAND (Fu et al., 2023).

We used the 1283 variable-associated SNPs that were regarded as adaptive variants to explore patterns of isolation-by-distance (IBD) and isolation-by-environment (IBE) compared to neutral variants (excluding adaptive SNPs). We found that genetic divergence calculated using both adaptive and neutral SNPs were not correlated with geographic distance (Figure 3E). In contrast, adaptive SNPs showed a significant IBE indicated by partial Mantel tests (Figure 3F), whereas the neutral SNPs was not significant, indicating that genetic variation of the adaptive variants was mainly influenced by the environment. Our results suggested that the adaptive SNPs were relatively robust to the confounding effects of population structure and highlight the central role of ecological factors in shaping genomic variation across landscapes.

Selection signatures associated with environmental extremes in indigenous chickens

The niche differential analysis for BARE, BIO_2, CROPLAND, PREC_08, SRAD_11 and TREES suggested that the significant environmental divergence prompted the genomic changes between chicken populations with extremely high and low environmental variables. Given their small niche differentiation in more than three populations in the extreme group, the SHRUBS and WATER were discarded before the selection signature analysis and candidate gene analysis. For the six retained drivers, we applied two methods to detected the potential genomic regions and variants associated with environmental changes. First, we performed F_ST analysis and extracted top 1% selected regions (N=935) with zFst threshold larger than 3.34, 3.44, 3.29, 3.34, 3.37 and 3.42 for BARE, BIO_2, CROPLAND, PREC_08, SRAD_11 and TREES, respectively. These regions were annotated with 548–679 Ensembl genes per variable, the gene sets of which were independent with few overlaps (Supplementary Figure S15A; Supplementary Table S8). Function annotation revealed that about half (41.97%~49.30%) of genes lacked gene symbols. GO and KEGG enrichment showed that MAPK signaling pathway and calcium signaling pathway were significantly (FDR<0.05) enriched by BARE-associated genes (Supplementary Figure S15B). In the XP-EHH analysis, the signatures of individual SNP were merged into 935 windows, in which top 1% selected regions were obtained with threshold exceeding 2.82, 2.89, 2.86, 2.79, 2.80 and 2.83 for the six variables, respectively. These genomic regions were annotated with 425–510 Ensembl genes for the six variables (Supplementary Figure S15C; Supplementary Table S9), in which the BARE-associated genes and the CROPLAND-associated genes were significantly (FDR<0.05) overrepresented in eight and four KEGG pathways, respectively (Supplementary Figure S15D). Among these pathways, alpha-linoleic acid metabolism, linoleic acid metabolism and GnRH signaling pathway were shared by two variables. Similarly, the gene sets of the six variables from XP-EHH were independent with few overlaps, and 41.39%~53.18% genes lacked gene symbol annotation.

Notably, we observed a modest but significant positive correlation ranging from 0.23 to 0.39 between the F_ST and the XP-EHH results (P<2.2e-16) across the six environmental variables (Supplementary Figure S16), consistent with previous studies and confirming the robustness of our analysis (Ma et al., 2015). However, the overlap between the top 1% signatures detected by both methods was low, with significant correlations observed only for BIO_2 and PREC_08 (Figure 4A). This highlighted the complementary roles of two methods in detecting significant selection signatures. We further screened top genes defined as those annotated to the top 0.1% selection signatures in one method or top 1% selected genes shared by two methods (Supplementary Table S10). This yielded 168, 194, 107, 214, 126 and 184 top genes for BARE, BIO_2, CROPLAND, PREC_08, SRAD_11 and TREES, respectively (Figure 4B; Supplementary Table S11). Unlike to the distribution pattern in F_ST and XP-EHH, we found that top genes were common to two or more variables, with overlap proportion ranging from 46.26% to 71.73% (Figure 4B, C). Spearman correlation analysis indicated that this overlap was not owing to the number of shared chicken populations used in the SSA (Figure 4D). Functional enrichment revealed that calcium signaling pathway was significantly enriched by BARE-associated genes including CHRM2, CAMK2G, CAMK2D, CYSLTR2, CACNA1B and FLT1. (Figure 4E). Moreover, we mapped these genes to the chicken QTL database and revealed associations with 66 traits, covering growth, health, feeding, fatness, exterior, digestion, egg and meat quality traits, physiology and reproduction traits (Figure 4F; Supplementary Table S12). Notably, only a small proportion of genes (9–39) were overlapped with known QTLs, suggested that the candidate genes were more potential to affect traits, such as adaptation, since the QTLs mapping was generally conducted for economic traits and those associated with environment changes were rarely reported in chickens (Hu et al., 2022). Few environment-associated genes were overlapped with QTLs, together with few enriched pathways by environment-associated genes confirmed polygenic nature of adaptation in local chicken populations, which were a complex combination of multiple minor gene effects (Sang et al., 2022).

Figure 4.

Selection signature analysis between high and low group for six key environmental variables

A: Correlations between top 1% standardize values of F_ST (x-axis) and XPEHH (y-axis). The number of top 1% regions is listed in the parentheses. B: Shared and unique top genes, defined as the top 0.1% selection signatures in either method and top 1% selected genes shared by F_ST and XPEHH for each environmental variable. C: Upset plot for top genes across environmental variables. Red numbers mark the shared genes and chicken populations between two environmental variables used in the D: correlation analysis using Spearman method. E: Significantly enriched GO biological processes and KEGG pathways for top genes associated with each environmental variable. F: QTL annotation for top genes based on the chicken QTLdb.

Geographic distribution of genomic variants underlying adaptive responses to climate changes

According to previous analysis results, SRAD_11, PREC_08 and BIO_2 are three most important climatic factors contributing to the ecological niche of local chicken population (Figure 2B). SRAD_11, representing total solar radiation in November, was strongly correlated with annual solar radiation (r=0.59, P=2.6e-5; Supplementary Figure S17A), making it a robust indicator of adaptation to ultraviolet and heat stress. We identified PRMT3, EPC2 and TPK1 that were associated with SRAD_11 to be involved in the regulation of DNA damage and repair, which are crucial in how organisms respond to the solar radiation (Angrand, 2022; Brobbey et al., 2022; Jessulat et al., 2021) (Figure 5A). The common genomic regions identified by SSA and GEA included MAML2, C20H20ORF112, POU6F2 and one gene cluster (Supplementary Table S13). The gene MAML2 encoded the Mastermind-like proteins that acted as a vital transcriptional coactivator in the Notch signaling, which was responded to high-dose radiation (Banerjee et al., 2020). Within MALM2, we identified 11 SNPs with F_ST>0.5 in high LD (Supplementary Figure S18A, B). We selected an index SNP rs15536665 (F_ST=0.65) to show the alleles frequency pattern and found that chicken populations originated from high SRAD_11 in Xizang and Southwest areas possessed more G allele (Supplementary Figure S18C). In the gene cluster, we firstly identified index SNP rs316099750 that was divergently selected between high and low group (grey shadow). Given that the negative Tajima's D value suggested an excess of rare genetic variants in a population compared to that under neutral evolution, we further focused on two genomic regions with –0.57 and –0.36 of Tajima’D in the low SRAD_11 group in the gene cluster, among which the genomic region ranging from 2745001 bp to 2760000 bp (red shadow in Figure 5B) contained significant divergent values across the three statistics. The region was located in UNC80 gene and harbored an index intronic SNP rs316174149 (F_ST=0.56). The allele frequency patterns showed that chicken populations that originated from areas with high SRAD_11 possessed more A allele, whereas the T allele was dominant in chicken populations that originated from areas with low SRAD_11 (Figure 5C; Supplementary Figure S19). Notably, the Sichuan basin was exposed to less solar radiation, as shown in Figure 5C. We then investigated the gene frequency of rs316174149 for chicken breeds in Sichuan basin including Miyi chicken (MYC), Pengxian Yellow chicken (PXC), Emei Black chicken (EMB), Jiuyuan Black chicken (JYB) and Jinyang silky chicken (JYS) (Li et al., 2019), frequency of which was consistent with pattern that T allele was dominant in these chicken populations in contrast to that A allele was dominant in the Xizang fowl (TF) and Red jungle fowl (RJF) in tropical areas (Figure 5D). Given that most non-coding variants were located in regulatory elements, we further annotated the SNP based on the atlas of chicken regulatory elements (Pan et al., 2023). Five regulatory elements including EnhAWk, ATAC, Repr, ReprWk and Qui were annotated for rs316174149 across 23 tissues (Figure 5E). Correspondingly, we also found significant H3K27me3 signals for hypothalamus, duodenum, and eggshell gland, as well as ATAC-seq signals for kidney (Figure 5F). Moreover, the expression data from Chicken GTEx (Guan et al., 2025) showed that the UNC80 was highly expressed in the hypothalamus (Figure 5G). UNC80 encodes a key component of the NALCN sodium channel complex that regulates neuronal excitability and is actively expressed in the nervous systems (Chua et al., 2020; Wie et al., 2020). These suggested that the rs316174149 might act as a repressor to alter gene expression within the hypothalamus, mediating the long-term adaptation to solar radiation (Vrettos et al., 2025).

Figure 5.

Functional SNPs associated with adaptation to the solar radiation

A: Genome-wide selective signals of standardized F_ST and XP-EHH for chicken populations from high versus low solar radiation regions. Grey dashed lines highlight the top 1% threshold. Orange points represent candidate regions, and black points mark top annotated genes associated with solar radiation. B: Selection signatures for the gene cluster on Chromosome 7 that identified by both gene-environment association analysis and selection signature analysis. θπ and Tajima’s D are calculated in a 15 kb window with a 15 kb step. F_ST is calculated for each individual SNP. Grey shadow highlights genomic region harbored the index SNP, while the red shadow highlights candidate region harbored SNP rs316174149 used for the next analysis. C: Genotype frequencies of rs316174149 across 44 chicken populations. D: Allelic frequencies of rs316174149 across eight chicken populations distributed in Sichuan basin, Xizang and tropical area. A and T represent the alleles of this SNP. MYC, PXC, EMB, JYS, JYB, TF, RJF and I_RJF denote Miyi chicken, Pengxian yellow chicken, Emei black chicken, Jinyang silky chicken, Jiuyuan black chicken, Tibetan fowl, Red jungle fowl and Red jungle fowl in Indonesian. E: Regulatory activity across 23 different tissues annotated for rs316174149. F: Epigenomic annotation for rs316174149 based on the H3K4me3, H3K4me1, H3K27ac, H3K27me3, CTCF, ATAC-seq and DNase-seq. G: Expression profile of gene UNC80 across 32 chicken tissues.

PREC_08, an indicator of monthly precipitation, was strongly correlated with annual precipitation (r=0.75, P=5.8e-9; Supplementary Figure S17B), making it a robust marker of adaptation to humid versus arid environments. We identified SSA genes including ABCB10, ANK2, KCNA1, PRMT1, SLC22A15 and TRAPPC11 associated with PREC_08 were involved in molecule transportations to enhanced utilization of metabolic water in arid-dwelling animals (Rocha et al., 2021) (Figure 6A). The common genomic regions identified by SSA and GEA were annotated to PTPRO, SOX5, SPOCK1 and four gene clusters (Supplementary Table S14). In PTPRO, Tajima’D and F_ST analyses identified a region (62835001 bp~62840000 bp) with negative Tajima’D (–0.18) for the high PREC_08 group, contrasting with a positive Tajima’s D (1.64) in low PREC_08 group (Figure 6B), indicated that an excess of low frequency SNPs was under putative selective sweep in the chickens that originated from areas with high precipitation. In this region, 10 SNPs with F_ST>0.25 were detected. The most divergent SNP was an intronic SNP rs15292135, which was in LD block with six nearby SNPs (Figure 6C). The allele frequency of rs15292135 showed that C allele was dominant in the chicken populations adapted to the arid environment, whereas T allele was dominant in the chicken populations adapted to the humidity environment (Figure 6D). The chromatin state analyses showed that rs15292135 lied in short regions for most tissues annotated by ATAC, Repr, ReprWk and Qui (Figure 6E). Notably, the key roles of kidney and trachea were confirmed by the significant signals identified in the ATAC-seq data (Figure 6F), suggesting that rs15292135 affected the chromatin accessibility, altering gene regulation and contributed to the adaptation of humid or arid environment. Among four gene clusters, we scanned genomic regions with large difference among θπ, Tajima’s D and F_ST value on the chromosome 1. In the region, the index SNP rs3384544359 that possessed highest F_ST (0.46) was located in the region ranging from 64545001 bp to 64560000 bp. However, the MAF of rs3384544359 was low (0.13), resulting in almost fixed A allele in many chicken populations (Supplementary Figure S20). Therefore, we focused on another region ranging from 64590001 bp to 64605000 bp (red shadow in Supplementary Figure S21A), in which we identified a set of SNPs with high F_ST that were under putative selection sweeps. The index SNP rs312280416 with a MAF of 0.4 was in high LD with surrounding SNPs, locating in the AEBP2 (Supplementary Figure S21B). The allele frequency pattern showed that the frequency of G allele increased across chicken populations along with increase of the precipitation, suggesting that the SNP and candidate gene AEBP2 were involved in the adaptation of humidity (Supplementary Figure S21C).

Figure 6.

Functional SNPs associated with adaptation to the rainfall

A: Genome-wide selective signals of standardized F_ST and XP-EHH for chicken populations from high- and low-precipitation areas. Grey dashed lines highlight the top 1% quantiles. Orange points represent candidate regions, and black points mark top genes associated with precipitation. B: Selection signatures for gene PTPRO, identified by both gene-environment association analysis and selection signature analysis. θπ and Tajima’s D are calculated in a 15 kb window with a 15 kb step. F_ST is calculated for each individual SNP. C: Linkage disequilibrium (LD) block for SNPs with F_ST≥0.25 in the selected region, shown as mean r2 for paired SNPs. D: Allelic frequencies of the associated SNP rs15292135 across chicken populations from areas with extremely high and low precipitation. E: The regulatory activity across 23 different tissues annotated for rs15292135. F: Epigenomic annotation for rs15292135 based on the H3K4me3, H3K4me1, H3K27ac, H3K27me3, CTCF, ATAC-seq and DNase-seq.

For the BIO_2, we also identified a number of genes that were previously reported to be associated with temperature adaptation (Supplementary Figure S22A). For example, DIO3 encodes a deiodinase enzyme that converts the inactive thyroid hormone T4 to its active form T3. Cold exposure decreases hepatic DIO3 activity in chicks, leading to less T3 degradation. The increased T3 levels contribute to enhanced metabolic rate and heat production, which are essential for thermoregulation in cold environments (Al Amaz and Mishra, 2024). Using same pipeline, we obtained 2 intersected genes (EVL, NCOR2), 1 lncRNA (ENSGALG00010004931) and 6 gene clusters (Supplementary Table S15). We identified a 30 kb genomic region in the NCOR2 that was differentially selected between high and low groups across three indices, with Tajima’D=–0.71 in low BIO_2 group (Supplementary Figure S22B). Accompanied with F_ST value of single SNP, we detected that an intronic SNP rs316512683 with the highest F_ST (0.31) and in medium LD (0.46~0.66) with surrounding SNPs with F_ST>0.25 (Supplementary Figure S22C). The A allele of rs316512683 was mainly distributed in the north region that is characterized by large diurnal temperature difference, whereas the G allele was almost fixed in other chicken populations (Supplementary Figure S22D, E). Moreover, we detected SNP rs316512683 as a repressor in the heart via H3K27me3 modification, contributing to the thermal adaptation (Supplementary Figure S22F, G). Among gene clusters, we identified candidate SNP rs14836021 harbored in the intron of EPS8. As a signaling adaptor, it was involved in the EGFR pathway that promote cell survival and tissue repair under temperature-induced damage (Tito et al., 2025). The A allele of rs14836021 was almost fixed in the region that is characterized by small diurnal temperature difference, while the G allele increased in the north region (Supplementary Figure S23). Another SNP rs317631326 was an index SNP located in the DSEL and was in high LD with surrounding SNPs. The A allele of rs317631326 was dominant in NYC, XWC, SNC and BJC with extremely high BIO_2 (Supplementary Figure S24). DSEL encoded enzymes for dermatan sulfate biosynthesis, a key extracellular matrix component influencing cell behavior, metabolism, and homeostasis (Mizumoto & Yamada, 2022).

Geographic distribution of genomic variants underlying local landcover adaptation

For the landcover, we identified three physical materials at the land surface including CROPLAND, TREES and BARE with non-negligibly effect on the animal’s local adaptations (Figure 2B). For the CROPLAND, a few top genes were annotated in the selection signatures analysis (Figure 7A; Supplementary Table S11), which were rarely reported to be associated with cropland coverage and availability. However, genes including TPK1, COA3, G6PC1and LEPROT were demonstrated to play important role in metabolism, digestion and energy homeostasis. For example, G6PC1 encodes glucose-6-phosphatase, crucial for maintaining blood glucose levels between meals (Claxton et al., 2022). After overlapping with associated genes identified in landscape genomic analysis, we detected a continuous genomic regions ranging from 51548866 bp to 51626768 bp on chromosome 1 (Supplementary Table S16). The θπ and Tajima’s D indicated that chicken populations located in the areas with high CROPLAND were under intensive selection (Figure 7B). We further screened a 10 kb region with most significant value across three indices. Within this region, an index SNP rs794437951 with highest F_ST (0.65) was in high LD with eight surrounding SNPs with F_ST>0.5. The allele frequency pattern showed that the index SNP was almost fixed to T allele for most chicken populations (Supplementary Figure S25), which probably due to the low minor allele frequency (MAF). Therefore, we profiled the MAF of nine SNPs, among which an intronic SNP rs732137070 harbored in ENSGALG00010014829 carried highest MAF (0.26), and the surrounding SNPs were in high LD with rs732137070 (r²=0.53~0.74) (Figure 7C, D). The chicken populations originated from areas with high CROPLAND possessed more G allele of rs732137070, while the A allele of rs732137070 was dominant in chicken populations originating from mountainous areas with few croplands (Figure 7E, F). The chromatin state annotation revealed that rs732137070 lied within the EnhAWk of hypothalamus and testis, suggested transcriptional activation at this site in the two organs (Supplementary Figure S26). However, no any regulatory peak was detected for the SNP.

Figure 7.

Functional SNPs associated with adaptation to the cropland coverage

A: Genome-wide selective signals of standardized F_ST and XP-EHH for chicken populations from areas with extremely high and low cropland use. Grey dashed lines highlight the top 1% threshold. Orange points represent candidate regions, and black points mark top genes associated with cropland coverage. B: Selection signatures for gene cluster that identified by both gene-environment association analysis and selection signature analysis. θπ and Tajima’s D are calculated in a 15 kb window with a 15 kb step. F_ST is calculated for each individual SNP. C: The minor allele frequency of SNPs surrounding the index SNP rs794437951. D: Linkage disequilibrium (LD) block for SNPs with F_ST≥0.5 in the selected region, shown as mean r2 for paired SNPs. The selected SNP rs732137070 is in high LD with surrounding SNPs. E: SNP genotype frequencies of rs732137070 across 44 chicken populations in China. F: Allelic frequencies of rs732137070 across 44 chicken populations.

For the TREES, we identified ITGB2 that belonged to integrin-beta gene family (Figure 8A; Supplementary Table S11), in which ITGB1 was previously reported to associated with land use pattern by RDA analysis in chickens (Gheyas et al., 2021), suggesting that the cell-to-cell and cell-to-extracellular matrix interactions regulated by integrin-beta were essential for individuals adapted to the specific environment. Using same pipeline, we identified six genes and four gene clusters distributing on the chromosome 1, 4, 10 and 23 (Supplementary Table S17). Some genes were involved in the neural development and signal transduction (FOXP2, TRPA1, PRKCE, and NRXN3), which contribute to the ability in sensor, communicating, mating and seeking in the dense tree areas. For example, TRPA1 was demonstrated to be participates in sensory processes, such as cold sensation (Zhang et al., 2022) and hearing (Vélez-Ortega et al., 2023), which were essential for living in the forest. In the TRPA1, we identified SNP rs317006675 in the intron to be under selection (Supplementary Figure S27A). The C allele of rs317006675 was significantly dominant in chicken populations from high TREES group and significantly decreased in chicken populations from areas with few trees (Supplementary Figure S27B). We then refined the four gene clusters, in which we identified most divergently selected regions on the chromosome 4 (Supplementary Figure S28). By focusing on genomic region with highest F_ST (red shadow in Figure 8B), we found a large divergence (Δθπ=0.0023 and ΔTajima’s D=2.81) between high and low TREES group. The region harbored 17 SNPs with F_ST>0.25. The index SNP rs314594662, a synonymous variant in the NXT2, was in high LD with eight surrounding SNPs. Chicken populations that originated from high trees coverage possessed more G allele, while the A allele was dominant in chicken populations originated from northwest areas with few trees (Figure 8C, D). Given that the nine SNPs located in the gene NXT2 were in high LD, we further analyzed the haplotypes between high and low TREES group. The nine SNPs formed five unique haplotypes across 62 samples. Among these haplotypes, CCGTGAGGA (H001) was dominant in terms of total amount and frequency in high TREES group, whereas the SNP sites in low TREES group were heterozygous and haplotypes were sporadic (Figure 8E), confirmed that NXT2 was under intensive selection in chickens to adapted to the environment with dense coverage of trees. Moreover, the chromatin state and regulatory annotation revealed that the candidate SNPs might be active enhancers within chromatin accessibility and DNA methylation region in bone marrow, trachea, kidney, liver, heart, gizzard, ileum, colon and cecum (Supplementary Figure S29).

Figure 8.

Functional SNPs associated with adaptation to tree coverage environment

A: Genome-wide selective signals of standardized F_ST and XP-EHH for chicken populations from areas with high and low tree coverage. Grey dashed lines highlight the top 1% quantiles. Orange points represent candidate regions, and black points mark top genes associated with tree coverage. B: The selection signatures for gene cluster that identified by both gene-environment association analysis and selection signature analysis. θπ and Tajima’s D are calculated in a 15 kb window with a 15 kb step. F_ST is calculated for each individual SNP. The red shadow highlight genomic region under extensive selection with an index SNP rs314594662 located in gene NXT2. C: Genotype frequencies of rs314594662 across 44 chicken populations. D: Allelic frequencies of rs314594662 across 44 chicken populations in China. E: Genomic position of candidate SNPs and a haplotype network in gene NXT2. The symbols on the line between haplotypes represent number of variants.

For the BARE, we identified a bunch of top genes like CACNA1B, COL11A1, MGAT4C, SLC4A10, SLC25A20 was associated with starvation and dehydration that were typical in BARE environment (Rocha et al., 2021, 2023) (Supplementary Figure S30A; Supplementary Table S11). Although no common genes were found across the four methods, a total of 31 genes were identified by any three of the four methods (Supplementary Figure S30B). Notably, the common genes identified in the GEA were rarely under intensive selection based on F_ST (n=6) and XPEHH (n=1), indicating polygenic inheritance of chicken adaption to the BARE environment. Nevertheless, the genomic regions under intensive selection were from natural and artificial interventions. We further focused on common SSA genes that overlapped with RDA or LFMM, from which we obtained three genes and one gene cluster in the RDA and one gene, one lncRNA and two gene clusters in the LFMM, respectively (Supplementary Table S18). By scanning individual SNP among these candidates, we found significant selection signatures in PALLD across three indices in RDA. An intronic SNP rs313027723 harbored in the gene was divergently selected between extremely high and low BARE group (Supplementary Figure S30C). The SNP was in a completed LD block with surrounding SNPs (F_ST>0.25), and the haplotype analysis showed that two haplotypes were existed in the populations. One of the haplotypes (CCAATAG) was abundant and dominant in low BARE group, whereas the TGGGCGA was less belonging to the high group (Supplementary Figure S30D, E). The G allele of rs313027723 showed increasing trend in the chickens found in the areas with extremely bare ground (Supplementary Figure S30F). For LFMM, the candidate region was mapped to the chromosome 3. An intronic SNP rs316823963 located in the GPR63 was screened out and in high LD (r²=0.61) with an index SNP rs15388384 located in the 3’UTR (Supplementary Figure S31A). Functionally important, we showed the allele frequency pattern of rs15388384, where the C allele was increased with increasing bareness of the ground (Supplementary Figure S31B). Furthermore, two SNP rs313027723 and rs15388384 were annotated as regulatory SNPs in the brain and heart, respectively (Supplementary Figure S32), suggesting their tissue-specific regulatory patterns.

DISCUSSION

Coping with environmental changes is common to all species. Unlike studies that focus on intraspecific adaptive variation to mitigate extinction risks in wildlife (Sang et al., 2022), we leveraged the abundant genetic diversity in adaptive variations and environment-resilience in poultry for sustainable global food supply (Boettcher et al., 2015). Compared to commercial chicken strains used in the intensive farming, indigenous populations exhibit superior adaptive traits, such as environmental resilience, expression of natural behaviors including foraging ability, and disease resistance, thereby providing valuable resources for new breeding goals that fit changes in climate, husbandry practices and consumer demands (Gheyas et al., 2021). In this study, we collected and generated genomic data for over 1000 Chinese indigenous chickens, representing the largest population dataset spanning all agroclimatic zones of China to date. We systematically analyzed the genetic diversity and structure of 44 chicken populations, which confirmed the purity of some populations like BYC (Wang et al., 2023a), JYC (Yue et al., 2025), LHC (Wang et al., 2023b), as well as the admixture of populations like CSL (Xu et al., 2023), WCC (Qi et al., 2024) previously reported following the analysis of whole genome resequencing data. The large-scale population genomic analysis also corroborated the admixture of NDC (Huang et al., 2020), CHC (Zhong et al., 2022). Regionally, Xizang populations (SNC and RKC) formed a genetically distinct cluster, likely shaped by prolonged geographic isolation and independent domestication history (Li et al., 2023). In addition, chicken populations from central China (LSC and GSC in Henan province, MCL and HJC in Hubei province, DAC and HLC in Hunan province) displayed similar genomic structure, consistent with local pig populations (Du et al., 2024), indicating that the two species experienced similar demographic histories and human interventions, such as selective breeding, migration and ecologically diverse habitats (Wiener & Wilkinson, 2011).

Animal adaptation is a long-term systematic process to cope with local stressors like heat, drought, disease and nutritional challenges, through a combination of genetic selection and physiological changes over generations. Traditionally, it is difficult to dissect the genetic basis of environmental adaptation due to the complexity of environment posing as selection pressures. With aid of ecological methods, we disentangled the important environmental factors of adaptation from a large array of climatic and soil variables. We then integrated large-scale genomic data to identify the genomic regions associated with the divergent environmental variables using multiple complementary genomic approaches. These methods were effectively and robustly applied in plants and animals, offering more reliable results (Gheyas et al., 2021; Teng et al., 2025; Wu et al., 2024). Notably, our analysis identified solar radiation (SRAD_11), cropland use (CROPLAND), rainfall (PREC_08), temperature difference (BIO_2), trees coverage (TREES) and bare ground (BARE) as most important environmental drivers, providing a more comprehensive insight compared to studies based on only climate variables (Liu et al., 2025b; Niu et al., 2024; Zhao et al., 2025) or a smaller geographical region (Gheyas et al., 2021). These variables were further demonstrated to be key factors for comparative genomic analysis by niche difference test, which was a result of genetic selection over centuries by environmental conditions and human intervention on their livestock species (Velado-Alonso et al., 2022). Moreover, the genetic divergence calculated by adaptive SNPs showed a significant association with environmental distances based on these variables, reinforcing the importance and selection of environmental variables in shaping adaptive differentiation of chicken populations, similar to other livestock like sheep (Lv et al., 2014) and pig (Liu et al., 2025a). Functional annotation confirmed the roles of the identified genes in environmental adaptation, such as BARE-associated genes that were overrepresented in energy metabolism (amino acid, fatty acid and glycolysis) pathways that help animals manage energy stores to survive periods of food and water scarcity (Cheng et al., 2023; Rocha et al., 2021, 2023).

Environmental adaptation accompanied by micro-evolution within species is generally dictated by the genome (Bomblies & Peichel, 2022). By leveraging genomic data using multiple methods, we presented robust evidence for adaptive genes including PTPRO and NCOR2 which are well-known genes linked to precipitation and diurnal temperature difference, respectively. We also discovered novel genes including UNC80, CSF2RB, NXT2, and PALLD for solar radiation, cropland use, trees coverage and bare ground, respectively. NCOR2 is a nuclear receptor corepressor that played a crucial role in regulating thermogenesis in brown adipose tissue and for maintaining body temperature (Richter et al., 2022), hence its association with BIO_2. PTPRO modulates the glomerular pressure and filtration rate in the kidney, affecting conservation and loss of water, which enables adaptation to the humid and dry environments (Ozaltin et al., 2011; Wharram et al., 2000). CSF2RB encoded receptors that bound to ligands of colony stimulating factor (CSF), and is conserved in the vertebrates (Mun et al., 2020). Knockout of CSF receptors significantly caused fatty acid changes in the liver in the zebrafish (Meng et al., 2021), suggesting that its alteration can modify metabolism of chickens under different feed availability regimes. In addition, the pleiotropic genes associated with BIO2, SRAD_11 and BARE indicated that interaction of climatic and landcover factors with the genome may drive local adaptation, although more investigation is needed that consider human-mediated management. The variants that suffered from intensive selection in the candidate genes were mostly located in non-coding region, which implied that the regulatory elements played dominant role allowing more subtle and fine-tuned evolutionary changes with fewer negative side effects (Bomblies & Peichel, 2022; Haygood et al., 2010). By integrating Chicken GTEx data, we found these SNPs possessed different chromatin state across tissues, for example, acting as repressors in hypothalamus through H3K27me3 modification in response to solar radiation; adaptation to the humidity/aridity through altering chromatin accessibility (ATAC-seq) in kidney and trachea. These indicated that the non-coding SNPs generally regulated gene function in a tissue-specific manner (Albert & Kruglyak, 2015; Khurana et al., 2016). Furthermore, the rapid adaptation controlled by these variants were prevalent across local chicken populations (Agrawal & Hastings, 2023; Durkin et al., 2024; Henschen et al., 2023; Nair et al., 2024). Therefore, the corresponding interactions between key genes and genomic variants explained their potential regulatory mechanism for environmental adaptation.

Specifically, our findings provide genomic resources to further understand the genetic basis and regulatory mechanisms of adaptation in chickens and lays a foundation for future research with potential application in genetic resource conservation and animal breeding under diverse environments and climate change, and may accelerate comparative studies on adaptation in other species. Nevertheless, several limitations should be acknowledged. First, we found substantial adaptive genes lacked gene symbols, which limited the function interpretation. A more well-assembled genome and completed genome annotation is needed for the functional annotations. Second, the relatively low sequencing depth and the presence of modest breed admixture may reduce the statistical power, potentially leading to the omission of some causal variants. Third, our study CRISPR-based functional validation and population-specific regulatory omics data for investigating SNP-specific regulatory mechanisms. In addition, we should recognize that adaptive signals are generally not stable and shift rapidly due to the climate change and modern breeding practices. Future studies incorporating higher-depth sequencing, larger sample size and field experiment areas, complementary multi-omics technologies, and validation experiment could strengthen the applicability of the conclusions, and enhance the translational relevance for agricultural and biomedical research.

In conclusion, we dissected the agroclimatic factors and identified key environment factors driving the ecological niche partitioning of Chinese indigenous chickens. By incorporating landscape genomic analysis, comparative genomic analysis and atlas of chicken regulatory elements, we proposed candidate genes and SNPs that affected the environmental adaptation in a tissue-specific manner. Furthermore, we compiled a comprehensive list of potential variant-gene associations to serve as a valuable resource for future functional studies and breeding programs.

SUPPLEMENTARY DATA

Supplementary data to this article can be found online.

Funding Statement

This work was supported by the National Key Research and Development Program of China (2021YFD1200302), China Agriculture Research System of MOF and MARA (CARS-40), and Agricultural Science and Technology Innovation Program of the Chinese Academy of Agricultural Sciences (ASTIP-IAS16)

DATA AVAILABILITY

The genomic data of indigenous chickens has been deposited in National Genomics Data Center (NGDC, https://www.cncb.ac.cn/) at PRJCA046667 and the Science Data Bank under accession number https://doi.org/10.57760/sciencedb.010m2.