Genomic Variation, Population History, and Long-Term Genetic Adaptation to High Altitudes in Tibetan Partridge (Perdix hodgsoniae)

Abstract

Species residing across elevational gradients display adaptations in response to environmental changes such as oxygen availability, ultraviolet radiation, and temperature. Here, we study genomic variation, gene expression, and long-term adaptation in Tibetan Partridge (Perdix hodgsoniae) populations residing across the elevational gradient of the Tibetan Plateau. We generated a high-quality draft genome and used it to carry out downstream population genomic and transcriptomic analysis. The P. hodgsoniae populations residing across various elevations were genetically distinct, and their phylogenetic clustering was consistent with their geographic distribution. We identified possible evidence of gene flow between populations residing in <3,000 and >4,200 m elevation that is consistent with known habitat expansion of high-altitude populations of P. hodgsoniae to a lower elevation. We identified a 60 kb haplotype encompassing the Estrogen Receptor 1 (ESR1) gene, showing strong genetic divergence between populations of P. hodgsoniae. We identified six single nucleotide polymorphisms within the ESR1 gene fixed for derived alleles in high-altitude populations that are strongly conserved across vertebrates. We also compared blood transcriptome profiles and identified differentially expressed genes (such as GAPDH, LDHA, and ALDOC) that correlated with differences in altitude among populations of P. hodgsoniae. These candidate genes from population genomics and transcriptomics analysis were enriched for neutrophil degranulation and glycolysis pathways, which are known to respond to hypoxia and hence may contribute to long-term adaptation to high altitudes in P. hodgsoniae. Our results highlight Tibetan Partridges as a useful model to study molecular mechanisms underlying long-term adaptation to high altitudes.

Introduction

An important question in evolutionary biology is how organisms respond to a novel environment. Given the accelerating pace of environmental changes in recent times, it has become increasingly critical to gain a better understanding of organismal persistence (Fitzpatrick and Edelsparre 2018). Two major evolutionary processes mediate organismal persistence: (i) phenotypic plasticity (acclimation), which allows a particular genotype to express various phenotypes upon exposure to different environments, and (ii) genetic adaptation, arising from natural selection for a certain phenotype. These two processes are generally considered to operate on different time scales: acclimation is known to be the major mechanism of short-term rapid phenotypic changes, whereas genetic adaptation requires advantageous alleles, either from preexisting standing genetic variation (Barrett and Schluter 2008) or novel mutations (Barton 1998) and, hence, reflects a long-term species response. Yet, acclimation and genetic adaptation may interact to produce net changes in phenotypes that dictate organismal persistence (Donelson et al. 2019). For example, physiological acclimation may provide a fitness advantage, counteract the negative effects of rapid environmental changes (Laitinen and Nikoloski 2019), and facilitate long-term genetic adaptation (Suzuki and Nijhout 2006). However, acclimation may also be maladaptive and slow the effect of adaptive genetic changes (Rago et al. 2019).

Species residing across an elevational gradient demonstrate a suite of physiological traits showing both short-term plastic and long-term genetic adaptation in response to environmental changes such as temperature, oxygen availability, or ultraviolet (UV) radiation. One such trait is the response to the decreased partial pressure of oxygen (PO₂), i.e. hypoxia. Hypoxia imposes key physiological challenges due to changes in oxygen transport to tissues and related compensatory changes in cardiovascular functions (Naeije 2010). Rapid plastic responses to hypoxia include changes in blood composition, ventilation, and circulation (Monge and Leon-Velarde 1991; Hochachka 1998; Borras et al. 2010; Tufts et al. 2012; Gilbert-Kawai et al. 2014; McClelland and Scott 2019; Meir et al. 2019; Storz and Scott 2019). Furthermore, prolonged exposure to low PO₂ is associated with long-term genetic changes involving, e.g. elevated hemoglobin concentration (Beall 2000; Linck et al. 2021), increased hemoglobin-oxygen affinity (Storz et al. 2007; Projecto-Garcia et al. 2013; Natarajan et al. 2015), changes in hematology, relative heart mass (DuBay and Witt 2014), increased capillarization of muscle fibers (Monge and Leon-Velarde 1991; Scott et al. 2009; Butler 2010), changes in breathing patterns (Scott 2011; Lague et al. 2017), or additional changes that optimize metabolic functions (Storz et al. 2010; Dawson et al. 2020).

Genomics studies in different systems have identified several genes as targets of positive selection at high altitudes (Beall et al. 2010; Bigham et al. 2010; Simonson et al. 2010; Yi et al. 2010; Peng et al. 2011; Xu et al. 2011; Alkorta-Aranburu et al. 2012; Scheinfeldt et al. 2012; Huerta-Sánchez et al. 2013; Xiang et al. 2013; Xing et al. 2013; Crawford et al. 2017; Eichstaedt et al. 2017; Hu et al. 2017; Yang et al. 2017; Arciero et al. 2018; Jeong et al. 2018). For example, the hypoxia-inducible factor (HIF) pathway is known to play a key role in cellular oxygen response to hypoxia (Kaelin and Ratcliffe 2008; Lendahl et al. 2009; Greer et al. 2012; Semenza 2012, 2014; Samanta et al. 2017), and HIF pathway genes, such as EGLN1 (Egl-9 family hypoxia inducible factor 1) and EPAS1 (Endothelial PAS domain protein 1), are associated with high-altitude adaptation in humans (Beall et al. 2010; Bigham et al. 2010; Simonson et al. 2010; Bigham and Lee 2014; Foll et al. 2014; Petousi and Robbins 2014; Simonson 2015; Azad et al. 2017; Witt and Huerta-Sánchez 2019; O’Brien et al. 2020) and other mammals (Zhang et al. 2014; Liu et al. 2019; Schweizer et al. 2019; Witt and Huerta-Sánchez 2019; Pamenter et al. 2020).

Life at high elevations is physiologically challenging, yet a wide variety of species thrive in these regions, with birds being the most prolific. Birds at high altitudes have evolved specialized features in their lungs and heart for the maintenance of oxygen supply (Storz et al. 2010; Scott 2011; Lague 2017) and exhibit remarkable hypoxia tolerance (Faraci 1991; Scott et al. 2015; Meir et al. 2019). For example, the Bar-headed Goose (Anser indicus) has been used as a popular model for studying the physiological basis of elevational migration and exercise performance under hypoxia (Scott and Milsom 2006; Scott et al. 2009, 2015; Scott 2011; Lague et al. 2017). Other species such as hummingbirds (Projecto-Garcia et al. 2013; Rappole 2013), house wrens (Galen et al. 2015), tit-tyrant flycatchers (DuBay and Witt 2014), ducks (McCracken et al. 2009; Natarajan et al. 2015; Dawson et al. 2016; York et al. 2017; Lozano-Jaramillo et al. 2018; Graham and McCracken 2019; Ivy et al. 2019; Laguë et al. 2020), sparrows (Cheviron et al. 2014; Sun et al. 2016), and tits (Zhu et al. 2018a, b) have also been used to study high-altitude adaptation in birds. However, most of these studies have only leveraged pairwise comparisons between high and low altitude resident species, and the clinal variations in populations locally adapted across the elevational gradient are relatively less explored.

Here, we study genomic variation, population structure, and long-term genetic adaptation in Tibetan Partridge (Perdix hodgsoniae) and its two closely related species, Daurian Partridge (Perdix dauurica) and Gray Partridge (Perdix perdix), that reside across the elevational gradient in the Tibetan Plateau (Fig. 1a). All three species are small built and have limited sexual dimorphism in plumage with little plumage iridescence (Johnson 1996; Zheng 2015). The Tibetan Partridge is distributed between 2,800 and 4,600 m only in the Tibetan plateau, the Daurian Partridge between 0 and 3,000 m across the east of Asia including China and Mongolia, and the Gray Partridge between 0 and 2,000 m, having a wide distribution across Eurasia, ranging from China to Scandinavia (Johnson 1996; Zheng 2015). Previous studies have investigated the breeding ecology (Zheng-wang et al. 1994; Cucco et al. 2010; Dorge et al. 2014) and phylogenetic relationships (Bro et al. 2000; Cao et al. 2012; Homberger et al. 2021; Kimball et al. 2021) of these three species. In comparison with the other two closely related species, Tibetan Partridge lives exclusively at high altitudes and hence is expected to demonstrate the signature of long-term genetic adaptation. The International Union for Conservation of Nature (IUCN) status of the Tibetan Partridge is the least concern (IUCN 2022), possibly due to its large habitat range and stable population trend. The Tibetan Partridge inhabits high-altitude meadows covered by grass, scrubs, and some bushes, where it forages on seeds and insects and relies more upon walking but is capable of short flights (Lu and Ciren 2002). Tibetan Partridge is a non-migratory bird that lives in a large group outside the breeding season and pairs for breeding during spring and summer (Lu et al. 2003). Local human communities in Tibet may occasionally capture Tibetan Partridge as food; however, there have been no known efforts to domesticate or breed Tibetan Partridges in captivity, nor have there been any known reintroduction programs aimed at establishing populations in areas where they have been extirpated. Tibetan Partridges are adapted to living in the wild, capable of surviving across the altitudinal gradient in Tibetan Plateau and hence are a suitable study system to explore processes of long-term genetic adaptation. Recently, draft genomes of the Tibetan Partridge have been published (Li et al. 2023; Zhou et al. 2023), but population-scale studies to examine genomic signatures associated with high-altitude adaptation have not yet been done in this species.

Fig. 1.

Sample locations and phylogeny. a) Distribution of Perdix species. The Gray Partridge (P. perdix) is widely distributed across Europe and West Asia from 0 to 2,000 m, the Daurian Partridge (Perdix dauurica) is distributed across north Asia from 0 to 3,000 m, and the Tibetan Partridge (P. hodgsoniae) is restricted to the Tibetan Plateau at high elevations from 2,800 to 4,600 m. We sampled individuals of P. perdix from Scotland (blue star), P. dauurica from Ningxia, China (pink star), and four populations of P. hodgsoniae (black dots) from different elevations in the Tibetan Plateau. b) Maximum likelihood phylogenetic tree based on autosomal SNPs. All branches have full local support based on the Shimodaira–Hasegawa test (Price et al. 2010). Bird images © Nan Wang, Peng Ding, and Yong Xia.

In this study, we have used an integrative approach of population genomics and transcriptomics to explore genetic diversity, population structure, and genomic signatures associated with long-term genetic adaptation to high altitudes in the Tibetan Partridge. We have identified genomic regions showing strong genetic differentiation and genes differentially expressed that are related to traits possibly associated with high-altitude adaptation (such as traits related to hypoxia, UV radiation, and temperature).

Results

Reference Genome Sequencing, Assembly, and Annotation

A DNA library was prepared from an adult female Tibetan Partridge, native to 4,000 m elevation using the 10× chromium linked-read approach (Zheng et al. 2016) and sequenced on the Illumina platform to generate paired-end 150-bp reads, resulting in ∼675 million read pairs (∼102 Gb of raw sequence data). We first generated a Kmer distribution (supplementary table S1 and fig. S1, Supplementary Material online) based on these short sequencing reads to estimate a genome size of 1.17 Gb. Based on this genome size estimate, we had ∼87× sequence coverage for generating the draft de novo genome assembly. We used a chromatin conformation approach, Hi-C (Belton et al. 2012), to further improve the contiguity of the draft genome. The estimated length of the final genome assembly was 1.02 Gb. The assembly contained 11,359 total scaffolds (minimum 1,000 bp to maximum 56.88 Mb) with scaffold N50 of 8.86 Mb (Table 1).

Table 1.

Assembly statistics and annotation metrics for P. hodgsoniae genome

Estimated genome size based on Kmer distribution	1.17 Gb
Estimated sequence coverage	87.18×
Total assembled genome size (Gb)	1.02
Final assembly sequence coverage	98.77×
Total number of scaffolds	11,359
GC content	41.26%
N50 scaffold length (bp)	8,859,506
Longest scaffold (bp)	56,880,207
Repeats	12.65%
BUSCO founda	8,096
Total genes annotated	21,239

^aOut of 8,338 highly conserved avian genes (Waterhouse et al. 2018).

Gene space completeness was assessed using BUSCO (Waterhouse et al. 2018). Among 8,338 genes highly conserved across birds, 8,008 (96.04%) complete BUSCO genes were identified (7,914 single copies and 94 duplicated) in the P. hodgsoniae genome assembly. Fragmented sequences of further 88 genes were identified (1.06%), and only 242 (2.90%) were missing (Table 1; supplementary fig. S2, Supplementary Material online). This result indicates a high degree of gene space completeness in the P. hodgsoniae genome.

We further characterized the transposable elements in the P. hodgsoniae genome using homology-based (Smit et al. 2013) and de novo approaches (Flynn et al. 2019). A total of 12.65% of the genome was estimated to be consisting of repeats (supplementary table S2, Supplementary Material online). Among these sequences, 8.05% are retroelements (6.83% Long Interspersed Nuclear Elements [LINE], 0.07% Short Interspersed Nuclear Elements [SINE], and 1.14% Long Terminal Repeats [LTRs]), and 0.92% are DNA transposons. These estimates are similar to transposable elements detected in other bird genomes (Kapusta and Suh 2017).

We also performed RNA-sequencing from multiple tissues (skeletal muscle, heart, liver, lungs) from the same individual that was used for genome sequencing. The RNA-seq data were further used to aid in genome annotation (supplementary table S3, Supplementary Material online) by combining with protein homology-based evidence and ab initio gene prediction models in two rounds of iteration using the MAKER2 genome annotation pipeline (Holt and Yandell 2011). This genome annotation procedure identified 21,239 protein-coding genes in the P. hodgsoniae genome. The Annotation Edit Distance, which provides a measurement for how well an annotation agrees with overlapping RNA-seq and protein homology data, was >0.5 in >98% of the gene models indicating the high quality of P. hodgsoniae genome annotation (supplementary fig. S3, Supplementary Material online).

Population-Scale Genomic Variation Among Partridges

We further carried out whole-genome sequencing of 20 individuals from four populations of P. hodgsoniae across its elevational range (2,917, 3,792, 4,000, and 4,296 m) and 5 individuals from each of the two sister species, P. dauurica (residing at 1,200 m) and P. perdix (residing at 200 m) (supplementary table S4, Supplementary Material online; Fig. 1a). Each bird was sequenced to ∼30× coverage, and reads were aligned to P. hodgsoniae genome to identify ∼24 million Single Nucleotide Polymorphisms (SNPs) that were variable within or between populations. We found a considerable amount of genetic diversity within each population, in the range of 1.06 × 10⁻³ to 2.31 × 10⁻³ (supplementary table S5, Supplementary Material online), similar to that reported in other bird populations (Ellegren 2013). The genetic diversity was highest in P. dauurica, which is consistent with previous findings of high genetic diversity in this species based on microsatellite variation (Cao et al. 2010).

Genome-Based Phylogeny and Population Demographic History

We further used the SNP data to build a phylogenetic tree using a maximum-likelihood approach (Fig. 1b; Price et al. 2010). As expected, each species was monophyletic, and P. dauurica and P. perdix were the sister clades to P. hodgsoniae. We dated phylogenetic splits using a Bayesian method (Lemey et al. 2009) and using the TimeTree database (Kumar et al. 2017) estimates of 4.37 million years ago as the split between P. hodgsoniae and the ancestor of the other two species. We inferred that the split between P. perdix and P. dauurica occurred about 2.82 million years ago (supplementary fig. S4, Supplementary Material online).

In addition to using methods to generate a single “species” tree by concatenating all possible “gene” trees, we also used a method that implements a full coalescent model (Bryant et al. 2012) to examine all possible gene trees across the genome. We observed that most posterior gene trees shared the same topology (supplementary fig. S5, Supplementary Material online), indicating greater certainty in the phylogenetic clustering in these populations. We also built separate phylogenetic trees for the Z chromosome and the mitochondrial genome (supplementary figs. S6 and S7, Supplementary Material online), and both trees had similar topology to the phylogeny based on autosomes. These results indicate a concordant pattern of evolution among different components of the genome (autosomes, sex chromosomes, and mitochondria) in partridges.

Within the Tibetan Partridge clade, the four populations formed separate clusters (Fig. 1b), indicating that they are genetically distinct. However, they were clustered in two main clades, one comprising populations from 2,917 and 4,296 m, and the second comprising populations from 3,792 and 4,000 m, indicating that elevation differences are not driving the major phylogenetic relationships among Tibetan Partridge populations. The phylogenetic clustering of these four populations was instead consistent with their geographic distribution and topographic isolation due to the river basins and the mountains (supplementary fig. S8, Supplementary Material online).

To support the phylogenetic relationships and possible gene flow patterns among the four Tibetan Partridge populations, we employed fastsimcoal2 (Excoffier et al. 2021) to build different demographical models based on the site frequency spectrum. We tested the phylogenetic model for two scenarios: (i) two separate genetic clusters for four populations of P. hodgsoniae (2,917/4,296 m and 3,792/4,000 m) as indicated by genome-wide phylogeny and (ii) a 3,792 m population phylogenetically closer to 2,917/4,296 m considering their close geographical distance in comparison with 4,000 m population (supplementary fig. S9, Supplementary Material online). The lowest Akaike Information Criterion (AIC) value supported the best-fit model with two separate genetic clusters. Similar models of expected gene flow between four populations supported possible gene flow events between populations from 2,917 and 4,296 m (supplementary fig. S10, Supplementary material online).

Genome-Wide Scans for Signals of Positive Selection

We used population branch statistics (PBS) to identify the signatures of positive selection in the genome among the Tibetan Partridge populations. PBS calculates allele frequency changes in a population relative to its population divergence from the other two populations being compared. This method has been proven to be a powerful tool for detecting genetic loci associated with high-altitude populations (Yi et al. 2010). We calculated the average PBS score in sliding windows of 15-kb across the genome using ANGSD (Korneliussen et al. 2014), comparing low-elevation population (2,917 m) against two high-elevation populations (4,000 and 4,296 m). We identified 56 genomic windows showing significant genetic divergence (PBS > 1) between these populations (Fig. 2a). The strongest signal was a ∼60 kb region including two 15-kb windows with the highest PBS score that overlapped the ESR1 gene. The pairwise F_ST scores between the 2,917 m population and each of the 4,000 and 4,296 m populations were also high in these two windows (supplementary fig. S11, Supplementary Material online). Most individual SNPs in this region had high PBS scores as well (Fig. 2b). A maximum-likelihood tree based on all SNPs from this ∼60 kb region revealed a clustering pattern different to the genome-wide phylogeny. Population from 2,917 m is the first branch to diverge within the Tibetan Partridge, and the other three populations (4,296, 3,792, and 4,000 m) cluster together (Fig. 2c). We also examined the genotypes of all SNPs in this region in all three species P. hodgsoniae, P. perdix, and P. dauurica (Fig. 2d), which revealed that the population from 2,917 m carries ancestral alleles and the other three Tibetan Partridge populations possessed derived alleles in this region, except for one individual from 3,792 m which carried ancestral alleles (and cluster with 2,917 m population in the tree) and one individual from 4,296 m that was heterozygous for this genetic locus. A fragment of ∼7 kb in this region appeared to be the core haplotype of differentiation as the pairwise linkage disequilibrium (LD) analysis for the variants within this 7 kb region showed a very high LD pattern (supplementary fig. S12, Supplementary Material online).

Fig. 2.

A major locus associated with high-altitude adaptation. a) Genome-wide PBS screen comparing low (2,917 m) and high-altitude (4,000 and 4,296 m) populations of the Tibetan Partridge. The strongest signal is marked with a red arrow that includes two 15-kb windows located in Scaffold 187. b) PBS scores for individual SNPs around these two 15-kb windows in Scaffold 187 indicate that this highly divergent region is extended across 60 kb. c) Maximum likelihood phylogenetic tree based on this 60 kb region shows a deep divergence between the 2,917 m population and all high-altitude 3,792, 4,000, and 4,296 m populations. d) Genotypes of individual SNPs within the 60 kb region shows that 2,917 m individuals are homozygous for ancestral alleles found in P. perdix and P. dauurica, while most high-elevation individuals are homozygous for derived alleles. A fragment of ∼7 kb appeared to be the core haplotype differentiating low-elevation from high-elevation species/populations. e) Phastcons scores (Siepel et al. 2005) for 135 SNPs strongly differentiated between 2,917 m and all high-altitude populations (F_ST = 1) overlap ESR1 gene.

This ∼60 kb genomic region consists of 5,622 SNPs. Among these, we identified 135 SNPs (2.4%) that were fixed for different alleles at 2,917 m and all other three populations (i.e. F_ST = 1 in 2,917 m vs. 3,792/4,000/4,296 m comparisons). We further aligned the Tibetan Partridge genome to the chicken genome using Cactus (Armstrong et al. 2020) and conducted a liftover of the coordinates of these SNPs into the chicken genome. Six of the 135 fixed differences occurred at highly conserved sites among vertebrates (Phastcons vert77 score > 0.8) (Siepel et al. 2005) and overlap the ESR1 gene (Fig. 2e). Five of these six highly conserved sites are noncoding (one upstream, four intronic), and one is a coding mutation located in the third exon of ESR1.

We also examined additional 54 genomic windows showing high genetic divergence (PBS > 1) between the low-elevation population (2,917 m) and the two high-elevation populations (4,000 and 4,296 m). There were 53 genes annotated in these regions (supplementary database S1, Supplementary Material online), which included genes such as PSMC2, PRKCD, GRIK2, DPCD, and CCNI, that were previously reported to be associated with hypoxia (Hao et al. 2019; Lim et al. 2021). We further examined metabolic pathways associated with these genes (see Materials and Methods for details) and identified 9 metabolic pathways enriched among these 53 genes (P-value ≤0.01; supplementary table S6, Supplementary Material online). Most of these pathways were associated with the regulation of cellular growth and apoptosis.

Differences in Gene Expression Profiles Among Native Populations Across the Elevational Gradient

We carried out blood transcriptome sequencing from individuals of Tibetan Partridge populations across the elevational gradient (2,917 to 4,609 m, n = 37, Fig. 3a). All 19 samples of Tibetan partridge used for genome sequencing were also used for blood transcriptome sequencing. However, we added additional samples (n = 18) to have a more detailed representation of populations across the elevation gradient and to use a larger sample size (Fig. 3a; supplementary table S7, Supplementary Material online). We chose blood for this study as it is a critical tissue for oxygen transportation that likely contributes to hypoxia adaptation. In addition, this approach allowed us to utilize a non-invasive method of collecting the tissue samples without the need of sacrificing the bird. We generated ∼8 to 24 million RNA-seq reads for each bird (supplementary table S7, Supplementary Material online). Using the annotation of the P. hodgsoniae genome, we quantified the transcript abundance for each individual using RSEM (Li and Dewey 2011).

Fig. 3.

Co-expression modules and DEGs among Tibetan Partridge populations across the elevational gradient. a) Sampling scheme for conducting blood transcriptome sequencing. Dots represent the number of individuals sampled from each location. Samples were divided into three altitudinal groups for differential expression analysis. b) Upper panel bar plot shows the total number of genes assigned to each co-expression module based on WGCNA analysis. Modules 6 and 19 were correlated with elevation. Genes not assigned to any module are shown as module 0. Lower panel bar plot shows the relative number of DEGs in each module; the total number of DEGs is shown on top of the bars. c) Heatmap showing normalized read counts across all samples for the 74 DEGs found in modules 6 and 19. d) Normalized read counts plots for strongest differentially expressed genes, upregulated (IGF2R) and downregulated (SLC1A4) in the high-altitude (>4,296 m) group.

Conducting differential gene expression analysis in a gradient setting where there are no clear definitions of “high” and “low” elevation populations is challenging. To mitigate the bias associated with categorizing birds into specific groups in such a gradient design, we employed a stepwise approach. We believe this is a robust approach to analyze data along the elevational gradient rather than only using a standard pairwise comparison by grouping populations into hypothetical “low” and “high” elevations. We first examined genes that are possibly co-expressed together in populations across the elevational gradient using weighted gene co-expression network analysis (WGCNA) (Langfelder and Horvath 2008). This analysis identified 19 modules of co-expressed genes (supplementary table S8, Supplementary Material online), and 2 of these modules, 6 (332 genes) and 19 (48 genes), were significantly correlated with elevation (P-value < 0.01; Fig. 3b, upper panel).

We then carried out differential gene expression analysis between two groups showing the strongest elevation differences, (populations from <3,623 m and from >4,296 m) using DESeq2 (Love et al. 2014). The choice of categorization for <3,623 and >4,296 m was made to use a similar sample size in each group while maximizing altitudinal differences between the groups. We identified 291 differentially expressed genes (DEGs) among these 2 groups (P-value <0.01; supplementary fig. S13, Supplementary Material online). Moreover, we found a higher proportion of the DEGs in the two modules correlated to altitude in the WGCNA analysis (Fig. 3b, lower panel) indicating consistency in the results of two different approaches. Module 6 had 71 DEGs, most of which were downregulated in higher elevation populations (Fig. 3c). Module 19 had 3 DEGs (DDR2, HSD17B7, and CORO1C), and they showed upregulation specifically in individuals from the 2,917 m population. Insulin-like growth factor 2 receptor (IGF2R) gene and solute carrier family 1 member 4 genes (SLC1A4) showed the strongest differences in gene expression among populations across the elevational gradient (P = 3.3 × 10⁻⁴ and 9.23 × 10⁻³¹, respectively). IGF2R was upregulated in populations >4,296 m, whereas SLC1A4 was downregulated in these populations (Fig. 3d).

We further performed metabolic pathways enrichment analyses for these 74 genes (71 + 3 = 74) that showed differences in expression associated with elevation in 2 independent analytical approaches (WGCNA and differential gene expression analysis). We found 6 metabolic pathways overrepresented in these 74 candidate genes, including the fatty acid metabolism pathway (P-value <0.01; supplementary table S9, Supplementary Material online). We also carried out a separate enrichment analysis in all the 291 DEGs that identified 9 enriched pathways, 3 of them related to glycolysis and sugar metabolism (supplementary tables S10 to S12 and fig. S14, Supplementary Material online).

Comparative Analysis of Candidate Genes Identified From Genomic and Transcriptomic Analyses

We identified 56 genomic windows showing significant genetic divergence (PBS > 1) by comparing low-elevation populations (2,917 m) against two high-elevation populations (4,000 and 4,296 m) (Fig. 2a; supplementary database S1, Supplementary Material online) from genomic analysis. We investigated whether any of the candidate genes identified in these regions of significant genetic divergence (PBS > 1) overlapped with DEGs discovered through transcriptome analysis. Unexpectedly, ESR1, the candidate gene that exhibited the most pronounced genetic divergence between low- and high-elevation populations was not expressed in the blood of Tibetan Partridges. So, we were unable to explore the variations in the expression profile of this gene among populations residing along the elevational gradient. None of the other genes from the regions of high genetic divergence showed differential gene expression among populations.

Additionally, we assessed the levels of genetic divergence (PBS score) in the genomic regions encompassing all 291 genes that exhibited differential expression among populations. However, none of these genes coincided with regions of high genetic divergence. Only one gene, MOSPD2, displayed a moderately high PBS score (0.93) (supplementary fig. S15, Supplementary Material online).

Although genes showing high genetic divergence did not demonstrate changes in their expression among populations, they may still interact with other genes within the same metabolic pathway and regulate their gene expression. Using the String protein interaction database (Szklarczyk et al. 2021), we identified that a majority of genes in regions of high genetic divergence do interact with genes that showed differences in expression associated with elevation (>0.9 confidence score, see Materials and Methods for details) (Fig. 4; supplementary fig. S16, Supplementary material online).

Fig. 4.

Candidate gene interaction. Network showing high confidence (score > 0.9) interactions among candidate genes identified from genomics and transcriptomics analysis (e.g. ESR1 interacts with seven DEGs). Only the largest interaction network is shown here (see the complete network in supplementary fig. S16, Supplementary material online).

For example, ESR1 (the gene showing the strongest genetic divergence) shares biological pathways and interacts with seven other genes that were identified to be differentially expressed among populations across the elevational gradient (Fig. 4). To investigate potential interactions between genes associated with the same metabolic pathways, we conducted an enrichment analysis of metabolic pathways among the genes displaying genetic divergence (identified through genomics analysis) and DEGs (identified through transcriptomics analysis). Our analysis revealed an overrepresentation of the neutrophil degranulation metabolic pathway among these gene lists (P < 0.01).

Discussion

In this study, we examined genomic and transcriptomic variation in natural populations of Tibetan Partridges residing across the elevation gradient in the Tibetan Plateau, including genomic comparisons with their two closely related species, Daurian and Gray Partridge. We generated a draft genome from a female Tibetan Partridge native to 4,000 m elevation using a linked-read sequencing approach. General statistics of genome contiguity and gene space completeness (genome size 1.02 Gb, N₅₀ 8.86 Mb, completeness 96.04%) are similar to the recently published genomes of Tibetan Partridge (Li et al. 2023; Zhou et al. 2023).

Genetic Diversity and Population Structure

Genetic diversity was considerably high in all three partridge species, with the highest in P. dauurica, which is consistent with the wider distribution range and the effects of the Tibetan Plateau uplift and environmental changes on the phylogeographic structure of this species (Cao et al. 2012). The whole-genome phylogenetic tree indicated that P. perdix is a sister clade to P. dauurica, and P. hodgsoniae diverged earlier, which agrees with a previously published phylogeny based on the nuclear (C-MOS) and mitochondrial genes (CYTB and ND2) (Bao et al. 2010). The phylogenetic clustering of the four Tibetan Partridge populations was not consistent with their altitudinal distribution but correlated with their geographic distribution along the river basins. Populations of Tibetan Partridge from 2,917 and 4,296 m were sampled from the Yaluzangbu river drainage area, which is separated from the 3,792 m population (sampled from the Nujiang river drainage area), by a series of interconnected mountains. These mountains also serve as barriers for other species, such as the Tibetan eared pheasant and the white eared pheasant distributed at similar elevations (Wang et al. 2018b). Our result showed a similar tendency of topographic isolation.

We inferred the divergence time of P. perdix and P. dauurica to be 2.82 MYA, which is congruent with the previously estimated time based on fossil records (Bao et al. 2010). The first divergence within the P. hodgsoniae clade was estimated as 350,000 yr ago, which corresponds with the maximum extent of glacier development (Lei et al. 2014). The further divergence into the contemporary elevational populations possibly occurred during the Gonghe Movement (∼150,000 yr ago) when the plateau reached its present elevation (Lei et al. 2014) and likely the current river basins were established. Although our estimated times of divergence are well supported and confidence intervals are small, these times should be taken with caution because they are constrained by the estimated divergence time of the genus (4.37 MYA) based on the TimeTree database (Kumar et al. 2017).

Among four populations of Tibetan Partridge, we identified possible evidence of gene flow between populations residing at 2,917 and 4,296 m elevation. The ancestral population of Tibetan Partridge resided in high altitudes habitats in shrub areas, above the tree line (4,000 to 4,500 m) as lower elevations were not suitable habitats due to warmer temperatures and higher humidity (Wu 1980; Zheng 1997). However, the anthropogenic disturbance has changed these habitats along the Yaluzangbu River, allowing shrubs to grow at lower elevations and thereby expanding the habitats for Tibetan Partridge to lower elevations (Zheng 1997; Li et al. 2022). The population at 2,917 m was likely established from an ancestral population at higher elevations, which is consistent with our results of gene flow between the 2,917 and 4,276 m populations. Moreover, the 2,917 m population has lower diversity and a smaller effective population size compared with other sampled populations of the Tibetan Partridge, also suggesting that this population might have recently migrated to lower elevations.

Genome-Wide Scans for Signals of Positive Selection

We identified multiple genomic regions showing strong genetic divergence among populations of Tibetan Partridge residing across the elevational gradient. The genomic region showing the strongest divergence overlapped the ESR1 gene. This gene encodes an estrogen receptor, a ligand-activated transcription factor that regulates the transcription of many estrogen-inducible genes that play a role in growth, metabolism, sexual development, gestation, and other reproductive functions (Sayers et al. 2022). ESR1 is also known to mediate activation of the eNOS-NO-cGMP pathway that facilitates high-altitude acclimatization (Pooja et al. 2020) and demonstrates a signature of positive selection in highland human populations (Gnecchi-Ruscone et al. 2018; Pooja et al. 2021; Sharma et al. 2022). Hence, ESR1 is a likely candidate gene for high-altitude adaptation in Tibetan Partridge. Populations from 2,917 m shared similar alleles to those of Gray and Daurian partridge at the ESR1 locus, while populations from 3,792, 4,000, and 4,296 m carried different alleles. This suggests high-altitude populations of Tibetan Partridge carry non-ancestral, derived alleles at the ESR1 locus, indicating possible adaptive loci at high altitudes. ESR1 showed six SNPs, strongly conserved across vertebrates, fixed for different alleles between the ancestral (low altitudes) and the derived allele (high altitudes). These are possible candidate mutations to further explore their possible role in the regulation of ESR1.

We also identified additional genes, e.g. PSMC2, PRKCD, and GRIK2 showing significant genetic divergence (PBS > 1) between the 2,917 m and 3,792/4,000/4,296 m populations of Tibetan Partridges. PSMC2 (proteasome 26S subunit, ATPase 2) is involved in the ATP-dependent degradation of ubiquitinated HIF-α in the cytosol (Gillespie et al. 2022). As HIF-α degradation is retarded under decreasing concentrations of oxygen (Jiang et al. 1996), PSMC2 is a key gene for cellular response to hypoxia. PRKCD (protein kinase C, delta) governs the cellular homeostatic response against hypoxic stress by activating autophagy and apoptosis (Chen et al. 2009). PRKCD is also involved in the UV-induced death effector pathway of keratinocytes undergoing apoptosis and plays an important role in epidermal homeostasis (Denning et al. 1998). GRIK2 (glutamate ionotropic receptor kainate type subunit 2) is a neurotransmitter receptor activated in a variety of normal neurophysiologic processes (Sayers et al. 2022) and acts as a thermoreceptor conferring sensitivity to cold temperatures (The UniProt Consortium 2021). GRIK2 has been previously identified as a candidate gene associated with high-altitude adaptation in humans and other animals (Ai et al. 2014; Foll et al. 2014; Gaur et al. 2020). Together with ESR1, these additional genes are possible candidate genes associated with high-altitude adaptation in Tibetan Partridges.

Biological pathways enriched among the set of all 53 genes showing high genetic divergence (PBS > 1) between low- and high-elevation populations of Tibetan Partridge are mostly involved in the regulation of cellular growth and apoptosis. Reduced oxygen supply activates mitochondria to produce reactive oxygen species, reactive nitrogen species, and other free radicals, causing oxidative stress, cell damage, and apoptosis (Li and Wang 2022). Strong divergence in genes associated with the regulation of apoptosis pathways is a possible indication of genetic adaptation to long-term chronic hypoxia in Tibetan Partridges.

While the genomic analysis has identified interesting candidate genes, our findings do not support a “simplified” narrative where only a few genes and a small number of genetic variants show significant genetic divergence between populations at high and low elevations. Instead, our data reveal a more complex picture, revealing at least 56 genomic regions that exhibit signals of positive selection. This result is not surprising, as elevation changes impose multiple selective pressures on organisms, such as hypoxia, UV exposure, and cold, leading to various physiological differences across populations residing at different elevations. We expect these physiological differences to be regulated by multiple genomic regions. Given the scope of the current work, it is challenging to conduct a comprehensive functional analysis of a large number of genetic variants within these 56 genomic regions and predict their functional role. We also did not measure any physiological traits that are known to be associated with high-altitude adaptation (DuBay and Witt 2014; DuBay 2018; DuBay et al. 2020; Qu et al. 2020; Swanson et al. 2020). So, our study is also not able to characterize associations between these candidate genetic variants and adaptive traits to identify their possible causative mutations. Future studies that involve the measurement of such physiological traits (e.g. hematological traits associated with blood oxygen carrying capacity) and large-scale genotyping of these candidate genetic variants among additional populations of Tibetan Partridge across the elevational gradient will allow a better functional interpretation of the candidate genes we have identified in this study.

Gene Expression Differences in Tibetan Partridge Populations Across the Elevational Gradient

To conduct pairwise differential gene expression analysis in a gradient setting is not straightforward, as we do not have strictly defined “high” and “low” elevation groups. So, to minimize the bias of categorizing birds into specific groups in such a gradient design, we utilized a stepwise approach. First, we identified modules of co-expressed genes in populations across the elevational gradient and identified which of these modules are significantly correlated with elevation using weighted gene coexpression network analysis. This analysis does not require “priori” information on elevation groups. Second, we overlapped the results of co-expression network analysis with an independent differential gene expression analysis between two groups showing the strongest elevation differences (populations from <3,623 m and from >4,296 m). The choice of categorization for <3,623 and >4,296 m was made to use a similar sample size in each group for doing differential gene expression analysis while maximizing altitudinal differences between the groups. This combined approach allowed better utilization of transcriptome data from multiple populations across the elevational gradient.

The strongest DEG across populations was IGF2R. Hypoxia is known to induce the expression of IGF2, which binds to its receptor, IGF2R, which acts as a G protein-coupled receptor, causing activation of the mitochondria-mediated apoptosis pathway (Chen et al. 2015). The upregulation of the IGF2R gene in high-altitude populations of Tibetan Partridge is consistent with this hypoxia response.

During hypoxia, cells undergo metabolic changes to meet their energy requirements. One such adaptation involves altering the way glutamine is metabolized. Instead of being primarily oxidized in the tricarboxylic acid cycle, glutamine is redirected to a process called reductive carboxylation. This shift allows cells to generate essential metabolites even under low-oxygen conditions. The transcription factor HIF-1, a key gene that regulates the cellular response to hypoxia, has been shown to decrease the expression of enzymes associated with glutamine metabolism (Sun and Denko 2014). The downregulation of the glutamine transporter SLC1A4 in high-altitude populations of Tibetan Partridge aligns with its potential response to hypoxia.

Metabolic pathway enrichment analysis identified glycolysis and sugar metabolism pathways enriched in DEGs among Tibetan Partridge populations residing across the elevational gradient. One of the major intracellular adaptations to severe hypoxia is the transition from oxidative phosphorylation to glycolysis (Robin et al. 1984). Hence, prolonged hypoxia exposure may impact the activity of genes associated with glycolysis and sugar metabolisms pathways. One of the genes in the pathway, LDHA (lactate dehydrogenase A), catalyzes the conversion of lactate to pyruvate and is a key step of anaerobic glycolysis (Sayers et al. 2022) and shows significant changes in gene expression among populations of Tibetan Partridge. Lactate dehydrogenase has been associated with high altitude adaptation in human (Horscroft et al. 2017), yak (Kuang et al. 2010), and lizards (Wang et al. 2018a) and hence is a candidate gene possibly associated with high-altitude adaptation in Tibetan Partridge.

We have identified a list of genes that are differentially expressed among populations residing across the elevational gradient. However, it is important to note that not all these genes may be considered “adaptive.” The observed differences in their expression levels could be attributed to various physiological changes that may not necessarily be adaptive. Some of these gene expressions may also show “maladaptive” responses, a known mechanism in hypoxia-related traits (Dempsey and Morgan 2015). In addition, we have only utilized blood tissue for the identification of DEGs and were not able to sample additional tissues for transcriptome analysis due to logistics and permit issues. Many key genes we have identified as candidates for long-term adaptation to chronic hypoxia may be differentially expressed and physiologically relevant in other tissues. For example, certain HIF pathway genes are known to play a critical role in particular cell types (e.g. vascular endothelial cells and bone marrow macrophages) (Rodriguez et al. 2021). Studies of tissue-specific gene expression from multiple tissues and cell types are needed in the future to have a better understanding of gene functions associated with high-altitude adaptation in Tibetan Partridges.

Comparative Analysis of Genomic and Transcriptomic Data

The whole-genome sequencing data from populations of Tibetan Partridge collected from four different elevations (2,917, 3,792, 4,000, and 4,296 m) allowed us to characterize candidate genomic regions showing high genetic divergence among these populations. Transcriptomics analysis allowed us to characterize variations in gene expression profiles in blood tissue among these same populations of Tibetan Partridge. We further compared whether any genes associated with high genetic divergence also show changes in gene expression among populations. However, we did not find any genes showing differences in gene expression associated with regions of high genetic divergence.

We do not necessarily expect all candidate genes identified using the population genomics approach to show differential gene expression in blood, as not all genes are expressed in this tissue. For instance, we did not find ESR1 (our strongest candidate from genomic comparisons) expressed in blood samples in the Tibetan Partridge. For one bird, we had generated transcriptome data of additional tissues for genome annotation, and we found ESR1 expressed in the liver, muscle, heart, and lungs. But, we found no expression in blood samples in all 37 birds used for genome and transcriptome sequencing. ESR1 is known to express in blood, specifically in lymphocytes in humans (Pierdominici et al. 2010), so it was interesting to identify no ESR1 expression in Tibetan Partridge blood. However, our findings are consistent with other studies in birds, which have reported the absence of ESR1 expression in blood, e.g. tree swallow (Tachycineta bicolor) (Bentz et al. 2019), chicken, and turkey (Bastian et al. 2021).

However, we found that most of the candidate genes identified from genomic and transcriptomic comparisons are reported to interact and are involved in the same or related metabolic pathways (based on the human protein interaction databases). ESR1 interacts with GAPDH, ALDOC, TALDO1, and LDHA, which were differentially expressed among populations of Tibetan Partridges. These results allow us to predict that the genetic variants showing strong divergence in the ESR1 gene may regulate other genes within the same metabolic pathways that are expressed in the blood. In an ideal scenario, it would have been preferable to investigate the tissue-specific gene expression of ESR1 using various tissues. However, due to logistical constraints, it was not feasible to employ invasive methods for collecting additional tissue samples from these birds.

We found the neutrophil degranulation pathway enriched among the list of candidate genes showing high genomic divergence and differential expression. Neutrophils are key immune cells responding to tissue damage and inflammation and are known to increase the transcriptional activity of hypoxia-inducible genes (Campbell et al. 2014) and hence may play an important role in the adaptation of Tibetan Partridge to high elevations. The findings from the comparative analysis of genomic and transcriptomic data suggest that genetic changes in genes operating upstream in a metabolic pathway could potentially influence the expression of downstream genes that were identified as differentially expressed among Tibetan Partridge populations. However, comprehensive investigations into the functional aspects of these candidate genes and their associated pathways will be necessary to delve into the potential involvement of non-coding regulatory genetic changes in the long-term adaptation to high altitudes in Tibetan Partridges.

Candidate Genes Identified in Previous Studies of High-altitude Adaptation in Other Systems

The HIF signaling pathway is known to play a key role in the regulation of cellular utilization of oxygen and glucose and mediates the ability of the cell to cope with decreased oxygen tension (Kumar and Choi 2015). Several HIF pathway genes, such as EGLN1 and EPAS1, are known to be associated with high-altitude adaptation in humans (Beall et al. 2010; Bigham et al. 2010; Simonson et al. 2010; Bigham and Lee 2014; Foll et al. 2014; Petousi and Robbins 2014; Simonson 2015; Azad et al. 2017; Witt and Huerta-Sánchez 2019; O’Brien et al. 2020), other mammals (Zhang et al. 2014; Liu et al. 2019; Schweizer et al. 2019; Witt and Huerta-Sánchez 2019; Pamenter et al. 2020), reptiles (Li et al. 2018), and birds (Graham and McCracken 2019). We did not find evidence of genetic divergence or differential expression in EPAS1 and EGLN1 genes among Tibetan Partridge populations. However, we have identified differential gene expression in other genes (such as LDHA, ALDOC, and MA2PK1) that regulate HIF-1α, which is a key gene from the HIF signaling pathway.

We also examined signatures of genetic divergence and gene expression for additional 494 genes that have been identified as candidates for high-altitude adaptation in other studies (Hao et al. 2019; Lim et al. 2021; see Materials and Methods for details). Eleven genes from the list (PEX5, MED6, PSMD2, ACER2, GADD45G, RHOA, SRP54, SCP2, PSMD2, ACER2, and RHOA) overlapped in our list of candidate genes showing genetic divergence or differential expression among populations of Tibetan Partridge. Evidence of molecular convergence in high-altitude adaptation among divergent species is relatively less explored, and our results should be interpreted with care. Future comparative genomics and transcriptomics studies using a multispecies comparative framework will be needed to examine the evidence of such molecular convergence associated with high-altitude adaptation.

Materials and Methods

Study Samples

Blood samples were collected from 29 wild adult individuals (supplementary table S4, Supplementary Material online) from three species: Gray Partridge (P. perdix) from Scotland (n = 5 from 200 m), Daurian Partridge (P. dauurica) from Ningxia Province, China (n = 5 from 1,200 m), and Tibetan Partridge (P. hodgsoniae) from the Tibetan Plateau (n = 5 from 2,917 m, n = 5 from 3,792 m, n = 4 from 4,000 m, and n = 5 from 4,296 m). We also collected additional tissue samples (heart, lungs, liver, and muscle) from a female Tibetan Partridge from 4,000 m. These samples were used to generate a reference genome of the Tibetan Partridge, annotate the genomes, and carry out population-scale whole genome resequencing. We further carried out blood transcriptome sequencing from Tibetan Partridge populations (n = 37) across the elevational gradient. All 20 samples of Tibetan Partridge used for genome sequencing were used for blood transcriptome sequencing. However, we collected additional samples (n = 17) to have a more detailed representation of populations across the elevation gradient and to use a larger sample size for the transcriptomic study. All samples were stored in RNA later until downstream processing, and sampling was done under the supervision of a licensed veterinarian.

To reduce variation among individuals at the time of sampling (particularly for transcriptome analysis), we only collected blood samples from adult birds, all from the same field season with an approximate balance of males and females for each population. It is challenging to maintain a constant ambient conditions while sampling birds in a field setting; however, we made sure to use similar capture techniques and tried to expose every bird to a similar level of stress while sampling.

Reference Genome Sequencing, Assembly, and Annotation

DNA extraction and genome sequencing were performed by the commercial provider, Grandomics Biosciences in Wuhan, China. A 10× Chromium GEM sequencing library was prepared using the extracted DNA from one female Tibetan Partridge from 4,000 m according to the manufacturer's recommended genome assay linked-read protocol. The library was sequenced on the Illumina NovaSeq platform to generate paired-end 150 bp reads from 400 bp average insert size fragments. The amount of sequencing was targeted to ∼100× coverage, as recommended for generating a de novo reference genome using a linked-read sequencing approach.

We used a Kmer-based approach (Vurture et al. 2017) to estimate genome size, heterozygosity, and repeat content from short-sequencing reads. These linked reads were further used to generate a reference genome using Supernova.v.2.1.1 (Weisenfeld et al. 2017) with default parameters. A chromatin conformation approach, Hi-C (Belton et al. 2012), was used to further improve the contiguity of the draft genome. We generated 100 Gb of short-sequencing data based on the Hi-C library and used the standard workflow (Yamaguchi et al. 2021) for better scaffolding of the draft genome generated by Supernova. Genome contiguity statistics, such as scaffold N₅₀ and the total number of scaffolds, were calculated using custom perl scripts. We further compared the draft genome assembly against a set of conserved genes in Aves using BUSCO version 4.0.6 (Waterhouse et al. 2017) to assess gene space completeness. The annotation of the P. hodgsoniae genome was done using the MAKER2 annotation pipeline (Cantarel et al. 2008). We generated protein homology evidence using publicly available proteome datasets downloaded from the Ensembl database (Cunningham et al. 2019) as well as transcriptome sequencing of multiple tissues (heart, lungs, liver, and muscles) of a P. hodgsoniae individual. These RNA-seq data were combined with protein homology-based evidence and ab initio gene prediction models in two rounds of iteration using the MAKER pipeline to annotate the P. hodgsoniae genome.

We further screened for transposable elements in the P. hodgsoniae genome using combined homology-based and de novo approaches using Repeatmasker version 4.1.0 (Smit et al. 2013; Flynn et al. 2020). We used two different sets of repeat libraries: (i) a reference repeat library downloaded from the Repbase database (release 20190301) (Bao et al. 2010) and (ii) a de novo custom-built species-specific repeat library generated for P. hodgsoniae using RepeatModeler v.2.0.1 (Flynn et al. 2020). The usage of the species-specific repeat library increased the accuracy of the detection and annotation of transposable elements.

Assignment of Putative Chromosomes

We further carried out the whole-genome alignment of the P. hodgsoniae genome against the chicken genome (GalGal6) (Cunningham et al. 2019) using Cactus, a reference-free whole-genome aligner (Armstrong et al. 2020). The alignment was used to lift over genomic coordinates from the Tibetan Partridge genome to the chicken genome using halTools (Hickey et al. 2013). Because of the interspecies structural variation, we do not expect perfect synteny between genomic scaffolds of Tibetan Partridge and chicken chromosomes; hence, we assigned respective scaffolds of the Tibetan Partridge genome to a putative chromosome if >50% of its sequence lift over to a specific chromosome on the chicken genome. To recover mitochondrial genomes, we used genomic and transcriptomic short sequence reads in the Mitobim (Hahn et al. 2013) pipeline, using publicly available mitochondrial DNA from P. hodgsoniae (Genbank EU845766) as a bait.

Population-Scale Sequencing, Alignment, and Variant Calling

We used DNA from 29 partridge samples (i.e. Gray Partridge n = 5, Daurian Partridge n = 5, and Tibetan Partridge n = 19) for population-scale genome resequencing. Each DNA sample was uniquely tagged with a sequence index during the multiplexing library preparation protocol. The libraries (average fragment size about 400 bp) were sequenced using the Illumina NovaSeq platform to generate 150 bp paired-end reads. The amount of sequence per bird was targeted to ∼30× coverage. Similarly, we employed a Trizol-based procedure to extract RNA and then used Nanodrop 2000 to evaluate the quality. The RNA sample was further used to construct a fragment library with an insert size of 300 bp, and the libraries were sequenced on the Illumina HiSeq platform with a paired-end 150 bp sequencing strategy. The target sequencing data for each RNA-seq library were 2 Gb. The RNA-seq was performed at Shanghai OE Biotech Company in China.

All short sequence reads were quality checked using FASTQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/), and we used BWA (Li and Durbin 2009) with default parameters to map the genomic reads from each individual against the reference genome assembly of P. hodgsoniae. The alignments were further checked for PCR duplicates using PICARD (http://picard.sourceforge.net/). We used Genome Analysis Toolkit (GATK) (McKenna et al. 2010) for base quality recalibrations, insertion/deletion (INDEL) realignment, SNP, and INDEL discovery and genotyping across all 29 samples simultaneously according to GATK best practice recommendations (Van der Auwera et al. 2013). Quality filtering of the raw variant calls was done according to an in-house filtering pipeline that we have previously developed (Lamichhaney et al. 2015). The missing and low-quality genotypes from the call set were inferred separately for each population using BEAGLE (Browning and Browning 2007).

Estimation of Intra- and Inter-species Genetic Diversity and Phylogeny Reconstruction

We used VCFtools (Danecek et al. 2011) to calculate various measures of intra- and inter-species genetic diversity (nucleotide diversity, SNP density, heterozygosity, and Watterson's theta). We used these estimates to further calculate effective population size using custom bash scripts. We used FastTree (Price et al. 2010) to infer approximate maximum-likelihood phylogenies with standard parameters for nucleotide alignments. Local support values for each node were estimated using Shimodaira–Hasegawa test implemented in FastTree. We also generated separate phylogenies for autosomes, Z and W chromosomes, and mtDNA. In addition to using methods to generate a single “species” tree by concatenating all possible “gene” trees, we also used SNAPP (Bryant et al. 2012), which implements a full coalescent model to examine all possible gene trees across the genome.

Dating the Nodes in the Phylogeny and Demographic History

We used SNAPP (Bryant et al. 2012) to infer a time-calibrated tree using a subset of SNP data (190,689 biallelic SNPs spaced 5 kb apart). Because the addition of multiple individuals per population only increases computational time, but not accuracy or precision (Stange et al. 2018), we reduced the number of individuals in each population to one. We used a true random number generator (https://www.random.org/) to select the individuals that would be included in the analysis. We constrained the age of the Perdix genus at 4.37 MYA based on TimeTree database estimates (http://www.timetree.org/) with 0 offsets and an standard deviation of 0.005 million years. TimeTree provides divergences times as a summary of times available in scientific literature; in this case, the 4.37 MYA time was based on two studies (Jetz et al. 2012; Stein et al. 2015). BEAST v.2.6.3 (Bouckaert et al. 2019) was ran for 5 million generations with sampling every 250 generations on the CIPRES server (http://www.phylo.org/); the first 10% was discarded as burn-in. Tracer 1.7.2 (Rambaut et al. 2018) was used to ensure that all parameters reached convergence (ESS > 200), and a maximum clade credibility tree was produced with TreeAnnotator 2.6.4 (Drummond and Rambaut 2007). The tree was visualized using the packages ggplot2 (Wickham 2016), ape (Paradis and Schliep 2019), and treeio (Wang et al. 2020) in R v.4.0.5 (R Core Team 2021). Estimated times of major geological events in the area were taken from Bao et al. (2010) and Lei et al. (2014).

To test different expected demographical models among the four Tibetan Partridge populations, we employed fastsimcoal2 (Excoffier et al. 2021). We generated site frequency spectrum data based on the sequence alignment files using ANGSD (Korneliussen et al. 2014). We then tested models for two phylogenetic scenarios: (i) two separate genetic clusters of P. hodgsoniae (2,917/4,296 and 3,792/4,000 m) as indicated by genome-wide phylogeny and (ii) a cluster with 3,792 m population phylogenetically closer to 2,917/4,296 m (following their geographical proximity). We also tested models for three gene flow scenarios among these populations on the optimal phylogenetic model. We carried out 1,000,000 simulations and 100 expectation conditional maximization cycles for each model. We repeated the simulation 50 times for each model, which helped us estimate the global maximum likelihood for each model. We used the simulation with the highest estimated likelihood to calculate the AIC value to identify the best-fit model.

Signatures of Selection Associated with Long-term Adaptation to High Altitudes

We scanned the genome in non-overlapping 15-kb windows to estimate PBS (Yi et al. 2010) that utilize pairwise F_ST values between three populations to quantify genetic divergence along each branch of their corresponding three-population tree. We used one low-elevation (2,917 m) and two high-elevation (4,000 and 4,296 m) populations of Tibetan Partridge to calculate PBS. The genomic windows with high PBS scores (>1) were analyzed for gene content. We further calculated pairwise LD between variants in the candidate genomic region using VCFtools (Danecek et al. 2011). We also examined the sequence conservation of SNPs that were fixed for different alleles at low-altitude and high-altitude populations across 77 vertebrate genomes using the PhastCons77 database (Siepel et al. 2005).

Comparative Transcriptomics Analysis

We generated RNA-seq data from 37 individuals of the Tibetan Partridge in their native habitat across the elevation gradient (2,917 to 4,609 m). During quality control of the sequence data, we identified five individuals with poor sequence quality and excluded them from downstream analysis. We used a “two-step” approach to analyze this transcriptome data, which we believe is a robust approach to utilize data along the elevational gradient rather than standard pairwise comparison by grouping populations into hypothetical “low” and “high” elevations.

Weighted Gene Coexpression Network Analysis

We first examined genes that are possibly co-expressed together in populations across the elevational gradient using WGCNA (Langfelder and Horvath 2008) for finding clusters (modules) of highly correlated genes. For this analysis, we excluded transcripts with low read counts (sum of counts across individuals ≤32). We used the DESeq2 (Love et al. 2014) function “variance stabilizing transformation” to normalize gene counts to correct for variation among different sequencing libraries and used the resulting matrix with 10,161 genes and 32 individuals as input for network analyses. We choose a value of 4 as the β power parameter to build the adjacency matrix (signed scale-free R² ≥ 0.8 and mean connectivity 100.9). We used the one-step function “net” for network construction and module detection.

Differential Gene Expression Analysis

We then carried out differential gene expression analysis between two groups showing the strongest elevation differences (populations from <3,623 m and from >4,296 m). The choice of categorization for <3,623 and >4,296 m was made to use a similar sample size in each group for doing differential gene expression analysis while maximizing altitudinal differences between the groups. We used RSEM (Li and Dewey 2011) to calculate counts and abundance values per transcript and tximport (Soneson et al. 2015) to import these data into DESeq2 (Love et al. 2014) in R (R Core Team 2021). We further excluded transcripts with low read counts (sum across all individuals = 32) and used the default analysis “DESeq” and the option “contrast” for comparing two groups to identify DEGs. DESeq2 by default also normalizes the raw read counts by library size and transcript lengths. All reported P-values were adjusted P-values, which account for multiple comparisons using the Benjamini–Hochberg method implemented in DESeq2.

Pathway Enrichment Analysis

To identify relevant pathways associated with candidate genes identified from our genomic and transcriptomic analyses, we used g:profiler (Raudvere et al. 2019), Reactome (Gillespie et al. 2022), and String (Szklarczyk et al. 2021) databases.

We further used String to build a network of interactions among candidate genes from both genomic and transcriptomic analyses. We used the human protein interaction database, only kept the genes with high confidence of interaction (score >0.9), and removed genes that were not connected in the network. We also used GSEA (Subramanian et al. 2005) to analyze candidate gene lists, ranking all genes resulting from DEseq2 analyses by their P-value and the sign of the change (the sign of the change × −log₁₀P-value).

Candidate Genes Comparison With Other Studies

We compare the candidate genes that we found in this study with 494 candidate genes associated with high-altitude that was previously reported in other systems. We got these list of genes by combining the list published by Lim et al. (2021; supplementary table S2, Supplementary Material online) and the genes identified by Hao et al. (2019).

Supplementary material

Supplementary material is available at Molecular Biology and Evolution online.

Author Contributions

S.L., P.W., and N.W. conceived the study. N.W. carried out fieldwork, collected the materials, performed experimental work, and explored the ecology and biogeography of the study species with contributions from Q.Z., J.J., and J.D. S.L. led the bioinformatics analysis of data. C.P. and P.W. performed the bioinformatics analysis with contributions from M.A.B., N.D., and L.C. N.W., C.P. S.L. wrote the paper with input from all other authors. All authors approved the manuscript before submission.

Data Availability

All raw data (Illumina sequencing reads) and reference genome of P. hodgsoniae have been submitted to NCBI under BioProject accession number PRJNA896916. Scripts and workflow for all analyses done in this paper is available on our lab GitHub page (https://github.com/sangeet2019/Tibetan_Partridge).

Conflict of interest statement. None declared.