Convergent molecular evolution of thermogenesis and circadian rhythm in Arctic ruminants

Abstract

The muskox and reindeer are the only ruminants that have evolved to survive in harsh Arctic environments. However, the genetic basis of this Arctic adaptation remains largely unclear. Here, we compared a de novo assembled muskox genome with reindeer and other ruminant genomes to identify convergent amino acid substitutions, rapidly evolving genes and positively selected genes among the two Arctic ruminants. We found these candidate genes were mainly involved in brown adipose tissue (BAT) thermogenesis and circadian rhythm. Furthermore, by integrating transcriptomic data from goat adipose tissues (white and brown), we demonstrated that muskox and reindeer may have evolved modulating mitochondrion, lipid metabolism and angiogenesis pathways to enhance BAT thermogenesis. In addition, results from co-immunoprecipitation experiments prove that convergent amino acid substitution of the angiogenesis-related gene hypoxia-inducible factor 2alpha (HIF2A), resulting in weakening of its interaction with prolyl hydroxylase domain-containing protein 2 (PHD2), may increase angiogenesis of BAT. Altogether, our work provides new insights into the molecular mechanisms involved in Arctic adaptation.

1. Introduction

Convergent evolution, referring to the same or similar phenotypic traits achieved by two or more distinct lineages evolving under comparable ecological conditions, has been extensively studied, not least since the increased accessibility of genome data from a wide range of species [1,2]. Convergent phenotypic evolution provides a great opportunity to elucidate the genetic basis of adaptation to a particular environment because convergence most likely occurs as a response to similar selection pressures. Extreme environments, such as the Arctic, are useful settings for studying convergent evolution because they often impose exceptional selection pressures on organisms inhabiting them [3]. The muskox (Ovibos moschatus) and reindeer (Rangifer tarandus) [4] are the only two ruminants distributed mainly in the High Arctic, which is characterized by constant daylight in summer, and almost constant darkness, shortage of food and temperatures as low as −60°C in winter [5]. The muskox belongs to the family Bovidae and the subfamily Caprinae and is believed to have originated on the tundra of north-central Asia around one million years ago, and reached the American continent over the Bering Land Bridge about ninety thousand years ago (figure 1a) [6]. Today, they are naturally distributed across northern Canada and northeast Greenland (figure 1b), with several additional populations introduced by humans across the Arctic [7]. The reindeer is a member of the family Cervidae and the subfamily Capreolinae (figure 1a), and it is widely distributed throughout Arctic and subarctic regions (figure 1b). These two ruminants show a number of independent cold adaptations to the Arctic, including behavioural, physical and physiological adaptations [5,8,9]. The two species diverged approximately 18.75–20.88 Ma from an ancestor that occurred in temperate regions [10], so any shared Arctic adaptations must be the result of convergent evolution. One clear example of Arctic phenotypic convergence is the ability to deposit large amounts of brown adipose tissue (BAT) in newborn Arctic ruminants [5]. BAT is important for generating heat under cold conditions in animals. It does so by virtue of its multiple cellular lipid droplets for lipid oxidation, a very rich vascularization to distribute the heat, and the presence of a unique uncoupling protein (UCP-1) in the membrane of the numerous mitochondria specialized in heat dissipation [11]. Previous research has shown that BAT serves as the major source of non-shivering thermogenesis (NST) in newborn muskox and reindeer [12], contributing as much as 60–70% to the total metabolic capacity [13]. By contrast, the contribution of BAT to the total metabolic capacity of neonatal lambs is 40% [14]. Moreover, the unique extremes in seasonal light–dark cycles in the Arctic lead to a seasonal absence of circadian rhythmicity [15–18], which affects many other aspects of Arctic animal physiology and behaviour such as feeding, digestion, and metabolic requirements [16,19]. Hence, modifications to circadian rhythmicity are another important Arctic adaptation shared by muskox and reindeer.

Figure 1.

Phylogenetic tree and geographical distribution of muskox and reindeer. (a) Maximum-likelihood tree for the 50 species obtained using fourfold degenerate sites. The number of convergent amino acid substitutive genes (CASG, black) of the muskox and reindeer are shown. (b) The geographical distribution of the muskox (red) and reindeer (blue) [6].

Most previous studies have focused on the morphological, physiological or behavioural aspects of adaptation in Arctic animals [5,6,20]. In this study, we de novo assembled the genome of the muskox, which provides an opportunity to investigate the genetic basis and molecular convergence of Arctic adaptation. Subsequently, we performed analyses of convergent genome evolution between reindeer and muskox, and integrated four BAT and three white adipose tissue (WAT) transcriptomes from goats (Capra hircus) to explore the functional genetic basis of Arctic adaptation. Based on these data, we identify genetic changes involved in the convergent Arctic adaptation of muskox and reindeer, including those associated with BAT thermogenesis and circadian rhythm.

2. Material and methods

(a) . Genome sequencing and assembly

Protocols used in animal collection were approved by the guidelines of the Northwest A&F University Animal Care Committee. We obtained ethylenediamine tetraacetic acid (EDTA) blood from a male muskox from Helsinki Zoo (studbook ID: SB#1939) and performed DNA extraction using the QIAGEN DNeasy Blood & Tissue Kit (QIAGEN, Valencia, CA, USA). The library was sequenced on an Illumina Hi-Seq X10 platform 150 bp paired-end (150PE) sequencer at BGI, generating a total of 180 Gb of Chromium linked-reads. Then, we applied Supernova (v. 2.0.1) [21] for genome assembly using about 60× coverage of reads. After that, the assembly was gap-closed using SOAPdenovo GapCloser (v. 1.12) [22] based on the linked-reads with barcodes being trimmed off. To obtain standard genome assembly metrics, we used the assemblathon_stats.pl script (https://github.com/ucdavis-bioinformatics/assemblathon2-analysis/blob/master/assemblathon_stats.pl). Benchmarking universal single-copy orthologues (BUSCO) [23] was used to assess the content of the genome assemblies using the mammalia_odb9 databases.

(b) . Transcriptome sequencing and analysis

BAT and WAT samples were collected from four neonatal goats (2 or 4 days old) and three goat kids (30 days old) in a local farm in Yangling, Shanxi, China. Total RNA was extracted from adipose tissues using TRIzol (Invitrogen, CA, USA). The sequencing libraries were constructed according to the manufacturer's instructions (NEBNext Ultra RNA Library Prep Kit from Illumina) and sequenced by the Illumina HiSeq 4000 platform and 150 bp paired-end reads were generated.

The raw reads were filtered by Trimmomatic (v. 0.33) [24]. High-quality reads were aligned to the goat reference genome (ARS1) with default parameters using HISAT2 (v. 2.0.3) [25]. The alignment results were sorted and merged by SAMtools (v. 0.1.19) [26]. Then the gene expression levels were quantified (in fragments per kilobase of transcript per million mapped reads, FPKM) by StringTie (v. 1.2.2) [27] based on the transcript annotation. Finally, we identified the differentially expressed genes (DEG) in BAT (relative to WAT) by negative binomial generalized linear models in DESeq2 (v. 1.20.0) [28].

(c) . Multiple sequence alignment

We used conserved genome synteny methodology to establish three high-confidence orthologous gene sets that included sequences from 21, 49 and 69 publicly available genomes and a de novo assembled muskox genome (electronic supplementary material, table S1). Briefly, LAST (v. last867) [29] (lastal -m 100 -E 0.05 index_name ruminant.species.fa | last -split -m 1 > result.maf) was used to align the genomes to the goat reference genome [30]. Then we used Multiz (v. 11.2) [31] to merge all the multiple alignment files (MAFs) into the multiple sequence alignment results. To reduce any effects of sequencing and assembly errors, annotation errors, pseudogenes, non-orthologous alignments and non-conserved gene structures in subsequent analyses, a series of rigorous filters were applied. After filtering, the three orthologous gene sets were generated from 22 species for the selection tests, 50 species for the estimation of convergent substitutions and 70 species for further verifying whether the identified convergent substitutions are shared uniquely by muskox and reindeer in 70 species.

(d) . Phylogenetic analysis

The orthologous genes generated from 22 species and 50 species were separately aligned for phylogenetic analyses (figure 1a and electronic supplementary material, figure S1). The four-fold degenerate sites of 14 357 and 11 800 one-to-one orthologous genes were separately extracted and then concatenated separately by using in-house Perl scripts (https://github.com/1221li/comparative-genome.git). Subsequently, the sites were used for constructing maximum likelihood (ML) phylogenetic trees by IQ-TREE (v. 1.4.2) [32] using the default parameters.

(e) . Analysis of positively selected genes and rapidly evolving genes

To identify positively selected genes (PSG) and rapidly evolving genes (REG) of muskox and reindeer, lesser mouse deer (Tragulus kanchil) was designated as an outgroup. A species tree based on four-dimensional sites was used for CODEML implemented in PAML [33] and branch lengths were optimized on a gene-by-gene basis. The muskox and reindeer branches of the phylogenetic tree were designated as foreground branches for two separate positive selection analyses, one for each species. The PSG were identified by the branch-site model, in which an alternative model allows sites to be under positive selection on the foreground branch. The REG for each species were tested by the branch model, in which the alternative model allows different rates for different branches. Likelihood ratio tests (LRTs) were used to compare a model allowing sites to be under positive selection (dN/dS > 1), with a null model allowing sites to evolve neutrally and be under purifying selection (dN/dS < 1). The false discovery rate (FDR) corrected with the Benjamini & Hochberg method was applied to correct for multiple tests.

(f) . Protein three-dimensional structure simulation

The three-dimensional structure of the HIF1 domain in hypoxia-inducible factor 2 alpha (HIF2A) protein was predicted using AlphaFold [34], and then visualized using UCSF Chimera [35].

(g) . Molecular convergence analysis

To identify the convergent amino acid substitutive genes (CASG) between muskox and reindeer, we expanded our data set to include 50 species (figure 1a; electronic supplementary material, table S1). Molecular convergence analyses were performed by two methods: (1) The PCOC method (changes in amino acid profiles (PC model) and at least one change per site (OC model)), which considers convergent shifts in amino acid preference instead of convergent substitutions [36]. We used the PCOC detection pipeline with parameter ‘-f 0.9’. (2) Convergent amino acid substitutions were identified based on Zhang's method [37]. The site was assumed as convergent substitution if amino acids of all clades with the convergent phenotype at that site are the same now but not in their most recent ancestry. The convergence event counting and probability calculation were performed by conv_cal software [37]. The ancestral sequences of 50 species were inferred by the Codeml in PAML [33]. To filter out false positive sites resulting from chance amino acid substitutions, a Poisson test was used to verify whether the observed number of convergent sites was significantly higher than the expected number by chance under the JTT-f_gene and JTT-f_site amino acid substitution models, respectively [37]. To further confirm the accuracy of convergent sites, the convergence at conservative sites (CCS) method was performed [37,38], which greatly reduces random convergence at rapidly evolving sites and falsely inferred convergence caused by misinterpreting the sharing of the ancestral character [38].

(h) . Co-immunoprecipitation and immunoblotting of HIF2A and PHD2

Genes, including the HIF domain of muskox and reindeer HIF2A (419D, D-type), human prolyl hydroxylase domain-containing protein 2 (PHD2) and reverted HIF2A (E419, E-type), were synthesized by Sangon Biotech, Inc. (Shanghai, China). The three genes were ligated into pUC-57 vectors and then cloned into the sites of pCMV-HA or pCMV-Flag-N vectors to express proteins. Co-immunoprecipitation (Co-IP) and immunoblotting (IB) were performed as described previously [39]. Human embryonic kidney (HEK) 293 T cells were purchased from the American Type Culture Collection (Manassas, VA) and were cultured in 6 cm Petri dishes until reaching 90% confluency. Cells were transfected with 4 µg of PHD2-HA with 4 µg of HIF2A-Flag or HIF2A-D419E-Flag. Two days after transfection, cells were lysed in lysis buffer (Sigma) and subjected to Co-IP according to the manufacturer's instructions (Sigma). Cell lysates were incubated with protein G-sepharose beads conjugated with anti-Flag antibodies at 4°C overnight. After washing five times, the precipitates were resuspended in SDS-PAGE sample buffer, boiled for 5 min and run on a 10% SDS-PAGE gel. IB was conducted with rabbit monoclonal anti-HA (Cell Signaling Technology) antibodies.

(i) . Gene enrichment analysis

To assess which biological functions were enriched among convergent and selected genes in muskox and reindeer, we performed functional enrichment analyses on the CASG, REG and PSG in muskox and reindeer, respectively. To identify convergence at the functional level we looked at overlapping Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO) pathways between the species. KEGG pathways and GO terms enrichment analysis were conducted by the KEGG Orthology-Based Annotation System (KOBAS) (v. 3.0) [40], clusterProfiler [41] and Metascape [42]. The FDR calculated with the Benjamini & Hochberg method was applied to correct for multiple testing.

3. Results

(a) . Muskox genome assembly and comparative genomics

The genome of a male muskox was sequenced using Illumina Hi-Seq X10 platform 150 bp paired-end (150PE) sequencing at BGI, generating a total of 180 Gb of Chromium linked-reads. First, we used about 60× coverage of reads for genome assembly. After that, the assembly was gap-closed based on the linked-reads with barcodes being trimmed off. We obtained a 2.6 Gb assembly with contig N50 of 843 kb, scaffold N50 of 46.82 Mb and GC content of 41.7% (table 1 and electronic supplementary material, table S2). To test assembly completeness, we ran BUSCO [23]. We found that 93.7% of these genes are completely present in the muskox genome assembly (table 1).

Table 1.

Assembly statistics of muskox genome.

assembly	number/length
total sequence length (bp)	2 615 835 778
total ungapped length (bp)	2 601 841 192
number of contigs	14 255
N50 Contig length (bp)	843 055
N50 scaffold length (bp)	46 819 416
number of scaffolds	6709
GC content (%)	41.7
complete BUSCOs (C) (%)	93.7

In this study, combining published genome data with our de novo assembled muskox genome, orthologous genes were used to search for the CASG, REG and PSG, and finally overlapping functional pathways under selection in the two Arctic ruminants. We identified a total of 413 and 552 REG in muskox and reindeer, respectively (electronic supplementary material, tables S3 and S4). Among these genes, 94 and 150 genes furthermore had a significant signal of PSG identified by a PAML branch-site model in muskox and reindeer, respectively (electronic supplementary material, tables S5 and S6). We found 37 REG and 6 PSG shared between muskox and reindeer (electronic supplementary material, tables S7 and S8). For CASG, we used the PCOC method [36] and identified 443 genes that are under convergent evolution in muskox and reindeer (electronic supplementary material, table S9). Based on the alternative method of Zou & Zhang [37] implemented in the conv_cal software, a total of 175 genes were identified (electronic supplementary material, table S10), all of which were also included in the 443 genes identified by PCOC. To filter randomly occurring convergent amino acid substitutions, we performed the CCS method for the 175 gene set and detected 23 robust CASG between muskox and reindeer (electronic supplementary material, table S11). To further verify whether the identified convergent substitutions are shared uniquely by muskox and reindeer in 70 species, we checked the convergent substitutions and found 8 CASG (ERICH5, SRP68, CCDC187, EPAS1, B4GALNT1, MYOM2, PKN3 and ACADVL), showing convergent amino acid substitutions shared uniquely by muskox and reindeer in 70 species (electronic supplementary material, table S11).

The shared GO terms and KEGG pathways in both species were related to processes involved in BAT thermogenesis, including fatty acid catabolic process, mitochondrial gene expression and regulation of lipolysis in adipocytes (electronic supplementary material, figure S2a,b and tables S12–S15). These enriched pathways suggest that cold adaptations involving BAT thermogenesis could have evolved convergently in both species. We focus on these putatively convergently evolved pathways in the following.

(b) . Convergent metabolic-related genes involved in brown adipose tissue thermogenesis

To further investigate whether genes under selection in muskox and reindeer are truly related to BAT thermogenesis, we identified DEG between BAT and WAT by using transcriptomes from goats (electronic supplementary material, table S16). We found that a high proportion of the genes identified as REG, PSG or CASG were also DEG between these two adipose tissues. 169 REG and 34 PSG out of those identified in the two species, along with 6 CASG identified across the two species, had differential expression between BAT and WAT (electronic supplementary material, tables S17–S19). We next conducted GO and KEGG analyses using all these genes as input, and with all significant DEG as background. The significant enrichment of GO terms was consistent with the crucial components of BAT thermogenesis (electronic supplementary material, figure S3 and table S20), and the significant pathway was the cAMP signalling pathway (electronic supplementary material, table S21), which both confirmed the link to cold adaptations in muskox and reindeer. Specifically, several genes under selection and differentially expressed were related to these four components of BAT thermogenesis: (i) sympathetic nerve stimulation (PRKAA2, PRKAG2, ADCY7, GNAS, ADCY5 and ADCY6), (ii) lipolysis (NP-R, HSL and MGL), (iii) UCP1 transcription (FGF21, RPS6KA2, TSC2, AKT1S1, CREB3, MAPK12, CREB3L1, ARID1A, ZNF516 and UCP1), and (iv) oxidative phosphorylation (ACSL1, ACSL3, CPT1B, CPT1A, SLC25A29, SLC25A20, CPT2, UCP1, respiratory chain enzyme complexes and ATPase) (figure 2).

Figure 2.

Genetic changes correlated with BAT thermogenesis in muskox and reindeer. Thermogenesis pathway involving BAT in muskox and reindeer. Genes marked green are rapidly evolving genes of muskox; genes marked red are rapidly evolving genes of reindeer; genes highlighted in purple are rapidly evolving genes of muskox and reindeer; genes marked in orange are positively selected genes of reindeer; genes marked in pink are positively selected genes of muskox;genes marked in blue are differentially expressed genes.

(c) . Convergent angiogenesis involved in brown adipose tissue thermogenesis

Our comparative genomic analysis identified five genes related to angiogenesis across those identified as evolving fast in either muskox or reindeer: ANGPT4 (REG in reindeer), PTGS1 (REG in muskox, PSG in muskox and reindeer), THBS1 (PSG in muskox) [43], SEMA3F (encoding semaphorin 3F, REG in reindeer) [44] and HIF2A (also named epithelial PAS domain protein 1 EPAS1, CASG in muskox and reindeer) (figure 3a and electronic supplementary material, figure S4a,b). Of these, PTGS1 and HIF2A were identified both in muskox and reindeer (electronic supplementary material, table S8). Furthermore, HIF2A is one of the 8 CASG (ERICH5, SRP68, CCDC187, EPAS1, B4GALNT1, MYOM2, PKN3 and ACADVL, electronic supplementary material, table S11), showing a single convergent amino acid substitution E419D shared uniquely by muskox and reindeer in 70 species (figure 3a). The substitution resulted in modest change in the three-dimensional structure of HIF1 domain in HIF2A (figure 3b). To explore the functional consequences of the convergent amino acid substitution, we constructed a human HIF1 domain vector (wild type) and a muskox- and reindeer-convergent substitution vector (mutant type) by replacing the human alleles (E) with the convergently substituted alleles (D), and then performed a Co-IP experiment. Compared with the wild type, the mutant type bound less strongly to PHD2 (figure 3c). The mutations in the oxygen-dependent degradation domain of HIF2A decrease hydroxylation by PHD2, resulting in stabilization of HIF2A. One of the functional effects of HIF2A is a positive correlation of increased HIF2A with the expression of VEGF, an important angiogenic factor [45,46].

Figure 3.

Muskox and reindeer convergent amino acid substitution in HIF2A. (a) One specific mutation (E419D) in the HIF1 domain of HIF2A. The HIF2A protein sequences of multiple orders (indicated with different colours) were aligned. The alignments of the E419D adjacent region are shown here. Species marked in different colours belong to different orders. (b) The three-dimensional structure of D-type (mutant type) and E-type (wild-type). (c) Co-immunopreciptation (Co-IP) assays showing that the muskox- and reindeer-mutated HIF2α (D-type) bound more weakly compared with WT HIF2α (E-type) proteins. PHD2-haemagglutinin (HA) was transfected with HIF2α (D-type) or HIF2α (E-type), and Co-IP was conducted. IB, immunoblot.

(d) . Circadian rhythm regulation in muskox and reindeer

In mammals, light regulates the circadian clock in the suprachiasmatic nucleus (SCN) of the hypothalamus via the phototransduction pathway (figure 4). Our comparative genomic analyses identified five REG (RYR1, CALM1, CALML4, CREBBP and ADCY6) in reindeer, four REG (ADCY5, MAPK3, CREBBP and CREB3) in muskox, and one PSG (CREB3) in reindeer that are involved in circadian clock pathways. RYR1 affects the signal transduction in the circadian rhythm pathway by regulating Ca²⁺ and K⁺ concentrations, which are crucial for the expression of clock genes in the SCN [47]. ADCY5 and ADCY6 are involved in adenylate cyclase-mediated pathways and affect the phosphorylation of CREB.

Figure 4.

Circadian rhythm regulation in muskox and reindeer. Light regulates the molecular clockwork in reindeer suprachiasmatic nucleus (SCN) neurons. Green genes are rapidly evolving genes of muskox; red genes are rapidly evolving genes of reindeer; orange genes are positively selected genes of reindeer; RGC, retinal ganglion cells; PACAP, pituitary adenylate cyclase activating polypeptide; SCN, suprachiasmatic nucleus; P, phosphorus.

4. Discussion

(a) . Convergence enhances BAT thermogenesis in Arctic ruminants

Previous phenotypic observations of Arctic animals have shown that typically precocious young (the muskox calves and the reindeer calves) have deposits of BAT, which compose about 1.1% of the body weight in newborn reindeer calves, and BAT in newborn muskox is at a higher proportion than in newborn reindeer [12,48]. BAT heat production is triggered by noradrenaline when the ambient temperature is low, which stimulates lipolysis of triglycerides (also called triacylglycerol, TG) in BAT. Lipolysis leads to the generation of free fatty acids (FFAs). FFAs activate mitochondrial UCP1 to produce heat, which is distributed to other tissues via a rich vascularization in BAT. Therefore, lipid metabolism, mitochondria and vascularity are all crucial components in BAT thermogenesis [49].

In this study, we found convergent genetic changes associated with BAT thermogenesis in muskox and reindeer. The two genes (LRP1 (PSG) and FASN (REG)) were identified as DEG between WAT and BAT in our comparative genome analysis (electronic supplementary material, figure S4c,d and table S22). LRP1 belongs to the LRP protein family, which encode low-density lipoprotein receptor proteins. Previous studies have demonstrated that the deficiency or expression disruption of LRP leads to mitochondrial dysfunction [50]. Furthermore, LRP5 (which was not among our candidate genes) was previously identified as a cold adaptation gene with the strongest positive selection signal found in humans from Siberia [51]. Another REG related to lipid metabolism, FASN, has undergone positive selection in Adélie penguin (Pygoscelis adeliae), a species adapted to extreme cold in the Antarctic region [52].

Additionally, several metabolic-related genes involved in BAT thermogenesis, including UCP1 transcription (CREB3 (PSG, REG) and FGF21 (REG)), fatty acid beta-oxidation (ACOT13 (REG) and ACADVL (CASG, REG)) and oxidation respiratory chain (COX11 (REG)) genes, were also identified [53,54] (electronic supplementary material, figure S5a–d). Notably, the REG FGF21 in muskox, the metabolic hormone fibroblast growth factor 21 gene, works in an autocrine/paracrine manner to increase UCP1 expression, hence stimulating sympathetic nerves to increase the abundance of brite/beige cells in WAT (‘browning’) and thereby induce thermogenesis [55,56]. Mice lacking in FGF21 had an impaired ability to adapt to chronic cold exposure [57].

In addition to the increased metabolic capacity provided by more mitochondria, BAT is also characterized by a rich vascularization, which allows more efficient exchange of nutrients, heat and oxygen via the bloodstream, required for adaptive thermogenesis in a cold climate. We therefore hypothesized that the enrichment of genes involved in angiogenesis among our CASG, REG and PSG could be related to the increased vascularization demands from a large BAT reserve. Our enrichment analysis showed that the candidate genes were enriched in angiogenesis and HIF-1 signalling pathways (electronic supplementary material, figure S2), suggesting that genes involved in angiogenesis may play a role in the BAT-associated cold adaptation of muskox and reindeer. The REG ANGPT4 in reindeer encodes angiopoietin, which facilitates endothelial cell activity and promotes angiogenesis. Similarly, the PSG PTGS1 in muskox and reindeer stimulates angiogenesis and upregulates VEGFA [58,59]. Notably, SEMA3F induces angiogenesis by inhibiting the Myc-regulated synthesis of THBS1 in epithelial cells [60], as well as affecting vascular patterning [44]. In mice, SEMA3F knockdown severely impairs vascularization [60]. Our results are consistent with research linking convergent genomic substitutions in SEMA3E and THBS1 to vascular development and thermoregulation in multiple marine mammal lineages [61].

Furthermore, a convergent amino acid substitution was observed in muskox and reindeer in the domain region of HIF2A, in a site highly conserved across vertebrates. Deficiency or expression disruption of this gene led to severe vascular defects and even death [62], and atherosclerosis [63] in mice. Our results showed that the E419D mutation unique to muskox and reindeer may reduce the degradation of HIF2A and thereby increase angiogenesis. We propose that the convergent HIF2A in muskox and reindeer is associated with transferring heat from BAT to other parts of the body. Although the abundance of blood vessels is thought to be associated with BAT, we cannot rule out the possibility that they are also used for restricting heat loss from legs, feet and the respiratory tract of Arctic mammals, as Arctic animals have multi-channelled arteriovenous blood vascular bundles in their limb extremities and a richly vascularized mucosal layer in nasal cavities for nasal heat exchange regulated by vasomotor adjustments in the nasal mucosa. Altogether, our analyses provide evidence that convergent substitutions of angiogenesis-related genes in muskox and reindeer might be associated with their phenotypic convergence in BAT-mediated adaptation to the cold Arctic environment.

(b) . The scarcity of molecular convergence events in muskox and reindeer

Convergent evolution may lead to the presence of molecularly redundant proteins [64]. Haemoglobin genes provide a simple example to detect the molecular convergence events of Arctic ruminants. Since haemoglobins from muskox and reindeer display unique thermodynamic properties characterized by a reduction in enthalpy required to unbind oxygen at low temperatures [9], we specifically searched these animals' haemoglobiniencoding genes for CASG. However, we did not find any convergent amino acid substitutions in either α- or β-chains of haemoglobins (HBA, HBB) (electronic supplementary material, figure S6c,d) [65], which could be due to the functional redundancy of amino acids and thereby coding regions. However, HBB did evolve rapidly in muskox (electronic supplementary material, figure S6a,b), which mirrors the accelerated evolution of haemoglobins in another iconic cold-adapted species, the extinct mammoth [65,66].

In addition, a relaxed circadian rhythm activity in winter and summer is the only adaptation that may be truly unique to polar animals [47]. Previous studies have confirmed that muskox may follow ultradian cycles (less than 24 h) of alternate foraging and resting/digesting bouts [16], which is similar to the Svalbard reindeer [48], Adélie penguins in Antarctica [49] and Svalbard ptarmigan [50]. The genetic mechanism underlying the suspension of a circadian clock in muskox remains unknown. Only 23 genes with convergent amino acid substitutions between muskox and reindeer were identified and none of them was involved in convergent circadian rhythm pathways. These results implied that the molecular convergence events are rare in Arctic ruminants.

(c) . Convergent evolution at the level of molecular pathways

Convergent genetic changes in either the same nucleotide position in different lineages, the same genes but at different sites, or the same pathways but at different genes, can all result in phenotypic convergence. There are now several studies showing examples of animal species independently evolving the same pathways to adapt to similar selection pressures. For example, brain and neuronal development pathways are under selection in cat [67], rabbit [68] and goat [69] domestication, and pathways related to hypoxia and nutrition metabolism have convergently evolved in yak and Tibetan antelope [70] as a parallel adaptation to high altitude [71]. However, it is unknown whether Arctic adaptations share a common genetic architecture, i.e. whether convergent phenotypic Arctic adaptations are caused by molecular or genetic convergence [3]. Our findings showed that the positive selection has been acting on different genes. We therefore mainly find molecular convergence in Arctic adaptations at the level of pathways, which is consistent with a recent study in three Arctic Brassicaceae species, which showed a lack of convergent evolution at the codon and gene levels, but a large overlap in classic temperature stress pathways under selection [3]. Convergence in pathways rather than genes also seems to be the predominant mode of genetic convergence in evolutionary syndromes, such as the domestication syndrome [67,68,72] and the high-altitude syndrome [71,73,74].

Altogether, our results indicate molecular convergence at the level of pathways in Arctic adaptation of muskox and reindeer, some of which is tentatively shared even with much more divergent animal taxa such as polar penguins. These findings improve our understanding of the molecular mechanisms of Arctic adaptation and provide novel insights into the study of BAT and obesity in humans.

Ethics

Protocols used in animal collection were approved by the guidelines of the Northwest A&F University Animal Care Committee.

Data accessibility

The assembled genome for the muskox (Ovibos moschatus) has been deposited in the NCBI under project number PRJNA789692. The RNA-seq data obtained in this study have been deposited in the China National GeneBank (http://www.cngb.org) and assigned the sample accession numbers CNS0387345 to CNS0387347 and CNS0387354 to CNS0387357. The RNA-seq data also have been deposited in NCBI under accession numbers SRR15145151 to SRR15145154, SRR15145163, SRR15145165 and SRR15145167.

The data are provided in electronic supplementary material [75].

Authors' contributions

M.L.: formal analysis, investigation, resources, visualization, writing—original draft, writing—review and editing; X.L.: formal analysis, methodology, resources, visualization; Z.W.: formal analysis, methodology, resources, visualization; G.Z.: resources, validation; N.W.: resources, visualization; M.D.: resources; S.L.: methodology, resources; C.Y.: methodology, resources; G.M.: methodology, resources; H.S.: methodology, resources; C.H.: resources; G.X.: resources; Y.L.: resources; Z.h.L.: resources; W.W.: resources; Y.J.: resources, supervision; R.H.: resources, supervision, writing—review and editing; Y.W.: funding acquisition, investigation, resources, supervision, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed herein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

This work was supported by the National Key Research and Development Programme of China (grant number 2021YFF1001000), the National Natural Science Foundation of China (grant number 32170627) and the Postdoctoral Innovative Talents Support Program of China (grant number BX20200282).