Cis-regulatory evolution reveals sensory trade-offs as a genetic basis for temporal niche evolution in tapirs

Abstract

Evolutionary shifts in diel activity patterns shape sensory remodeling across mammals, yet the genetic basis remains poorly understood. Tapirs represent a unique natural experiment, having reverted from a cathemeral ancestor to a nocturnal niche characterized by reduced vision but enhanced hearing and olfaction. Here, we investigate the genetic basis of this phenomenon by generating high-quality chromosome-level genomes for Tapirus terrestris and Tapirus indicus. Comparative analyses revealed extensive lineage-specific remodeling of genes and cis-regulatory elements linked to sensory pathways. Notably, functional validation via CRISPR-Cas9 editing of a tapir-specific conserved noncoding element (CNE74) upstream of the FLT1 gene in mice revealed coordinated sensory effects, including retinal degeneration and reduced visual acuity, yet enhanced auditory sensitivity. These findings suggest that regulatory element evolution may induce pleiotropic effects on competing sensory modalities, offering genetic insights into sensory evolution during temporal niche adaptation and potential relevance to human retinal vascular diseases.

Cis-regulatory evolution drives sensory trade-offs during tapir adaptation to nocturnality.

INTRODUCTION

Diel activity refers to the characteristic activity patterns of species within a 24-hour cycle (1). Although animals exhibit some flexibility in their diel activity patterns, many species exhibit a preferred diel temporal niche. Diel activity patterns profoundly influence internal rhythms, the ability of sensory systems to acquire environmental information, resource utilization efficiency, and energy metabolic balance in animals (2). These ecological pressures further drive adaptive evolution in species at morphological, physiological, and genetic levels (3). Converging evidence from anatomy, morphology, and molecular phylogenetics consistently supports the nocturnal ancestry of crown mammals (4–7). Mammals diversified rapidly once they transitioned from the nocturnal niche to other ecological niches, and such evolutionary shifts have occurred multiple times in mammals (2). Thus, identifying the naturally occurring molecular variations associated with diel niche traits could offer crucial insights into the genomic foundations of mammalian diversification.

Recent comparative genomic studies have begun to illuminate the genetic basis of sensory adaptations associated with diel activity shifts. The transition from diurnal to nocturnal activity in night monkeys (Aotus trivirgatus) is accompanied by pseudogenization of the short wavelength-sensitive opsin gene SWS1, likely reflecting relaxed selection for color vision in low-light environments (8). Similarly, the evolutionary shift from nocturnal to diurnal activity in the striped mouse (Rhabdomys pumilio) is associated with relaxed purifying selection on rod phototransduction genes and altered circadian regulation of peripheral tissues (9). These studies have primarily focused on protein-coding sequence evolution and circadian clock genes. In contrast, the role of regulatory element evolution remains largely unexplored in the context of temporal niche adaptation.

Tapirs (Tapiridae, Perissodactyla) represent an exceptional model for investigating the genetic mechanisms underlying diel niche shifts and sensory system evolution. Uniquely among ungulates, tapirs reverted from a cathemeral (day-and-night active) ancestor to a strictly nocturnal and crepuscular niche (active at dawn and dusk) (10–13). This twice-daily activity reversal drove a marked sensory reallocation: Tapirs exhibit regressive visual features, such as corneal opacity and reduced acuity (14), while enhancing auditory and chemosensory systems (15). Specifically, tapirs have unique auditory adaptations including an enlarged guttural pouch with sac-like branches that may amplify mid-low-frequency sound perception (16), a well-developed olfactory bulb (14), and a broad sense of taste (17). This coordinated sensory evolution, visual regression coupled with auditory and chemosensory enhancement, provides a natural experiment for understanding how activity pattern shifts drive adaptive changes across multiple sensory systems, making tapirs an ideal comparative model within mammals.

Furthermore, the four extant tapir species are distributed across two continents: three in Central and South America (Tapirus terrestris, Tapirus bairdii, and Tapirus pinchaque) and one in Asia (Tapirus indicus). All species are listed as endangered or vulnerable under the International Union for Conservation of Nature (IUCN) Red List and protected under the Convention on International Trade in Endangered Species of Wild Fauna and Flora (CITES) Appendices I and II, with T. indicus facing the highest extinction risk. Understanding the genomic basis of their sensory adaptations is critical for their conservation management in fragmenting habitats.

Here, we report the chromosomal-level genomes for two species representing the disjunct continental lineages: the South American T. terrestris (lowland tapir) and the Southeast Asian T. indicus (Asian tapir). Through comparative genomic analyses with other ungulates and broader mammalian out-groups, we identified lineage-specific genes variants and conserved noncoding elements (CNEs) in tapirs, involved in sensory system development and function. We investigate the regulatory potential of one Tapiridae-specific CNE (CNE74) upstream of FLT1 gene related to the retina and cochlea, using CRISPR-Cas9–edited mice. The Tapir_CNE74–edited mice exhibited reduced retinal thickness, diminished electroretinogram responses, and impaired visual acuity yet showed enhanced auditory sensitivity at specific frequencies compared to wild-type (WT) littermates. These findings suggest that regulatory element evolution can exert pleiotropic effects on multiple sensory systems and provide a plausible molecular framework for understanding temporal niche adaptation in mammals, with potential implications for human retinal vascular diseases.

RESULTS

Genome assembly, phylogenetic analysis, and chromosome evolution

By integrating PacBio sequencing in high-fidelity (HiFi) mode with Hi-C data (tables S1 to S3), we generated chromosome-level genome assemblies for T. terrestris and T. indicus (Fig. 1A, fig. S1, and table S4). The assemblies are ~2.57 Gb and 2.47 Gb in size, with contig N50 lengths of 70.38 and 131.30 Mb, respectively (Table 1). These assemblies represent a substantial leap in accuracy and contiguity over previous genomes (table S5 and figs. S2 to S4) (18, 19), evidenced by high quality values (QV) (78.47 and 63.84) and high Benchmarking Universal Single-Copy Orthologs (BUSCO) values (99.56 and 99.69%) (Table 1). Notably, we achieved gap-free, telomere-to-telomere assembly for 24 of 41 T. terrestris chromosomes and 6 of 26 T. indicus chromosomes (Fig. 1A and fig. S5). Deciphering the rearrangements in mammalian chromosomes is crucial for understanding speciation, adaptation, and disease susceptibility (20). T. indicus has the lowest chromosome number among extant perissodactyls (excluding the plains zebra) (21). On the basis of the four publicly available chromosome-level genomes (Diceros bicornis, Equus caballus, Bos taurus, and Canis lupus familiaris), we reconstructed the ancestral chromosomes of Tapiridae (using T. indicus as the reference genome). We identified 1086 conserved genomic fragments at 300-kb resolution and reconstructed 40 predicted ancestral chromosome fragments (PACFs) in Tapiridae, with a total length of ~2.32 Gb (fig. S6). We traced back the source of these PACFs for Tapiridae and found that there are more chromosome rearrangement events in T. indicus than in T. terrestris. Additionally, we identified 13 chromosome fusion events in T. indicus (Fig. 1B), indicating these fusions as the primary driver of its reduced karyotype. We annotated 40.40% of repetitive sequences in T. terrestris and 41.76% in T. indicus, with 20,772 and 21,044 protein-coding genes annotated, respectively (tables S6 to S8).

Fig. 1. Genome assemblies and evolutionary history of tapirs.

(A) Chromosome-scale landscape of T. terrestris and T. indicus. Circos plots display genomic features including chromosome length (a), guanine/cytosine (GC) content (b), repetitive elements (c), and gene density (d); interior links indicate syntenic blocks between the two species (e). (B) Synteny analysis revealing chromosomal homology and fusion events in T. indicus. (C) Phylogeny based on single-copy orthologs. Black numbers indicate divergence times [95% highest posterior density (HPD) intervals in blue], estimated using fourfold degenerate synonymous sites (4DTv) and four fossil constraints (table S9); red and green numbers denote expanded and contracted gene families, respectively. Pie charts represent reconstructed ancestral activity states (orange, nocturnal/crepuscular; gray, cathemeral; blue, diurnal) inferred from 196 mammals. The map (right) shows species distributions relative to tropical rainforest ecoregions [curated from (120)]. (D) Genome-wide heterozygosity levels compared with other endangered mammals. The IUCN categories: EN (endangered), NT (near threatened), EW (extinct in the wild), VU (vulnerable), and LC (least concern). (E) Demographic history inferred by pairwise sequentially Markovian coalescent (PSMC) analysis with the following parameters: “N25 -t15 -r5 -p 4+25*2+4+6” (default). The generation times for T. indicus and T. terrestris are 12 years and 11 years, respectively, with mutation rates of μ = 1.4473 × 10⁻⁹ and μ = 1.59203 × 10⁻⁸. The light gray and dark gray bars represent the Middle Pleistocene and Marine Isotope Stage 11 (MIS 11), respectively.

Table 1. Assembly statistics of the reference genomes of two tapir species assembled using PacBio-HiFi reads in this study.

	T. terrestris	T. indicus
Contig N50 (Mb)	65.14	87.67
Contig number	114	160
Total length of contigs (Mb)	2,567.76	2,468.89
Scaffold N50 (Mb)	70.38	131.30
Number of chromosomes	41 (39 + XY)	26 (25 + X)
Total length of assembled chromosomes (Mb)	2,542.18	2,440.07
Complete ratio of BUSCO (%)	99.56	99.69
QV (consensus quality value)	78.47	63.84

Using 8956 high-confidence single-copy orthologs, we resolved the phylogenetic position and divergence timing of tapirs (Fig. 1C). Divergence times were estimated on the basis of fourfold degenerate sites (4DTv), calibrated with four well-established fossil constraints (table S9) using MCMCTREE (see Materials and Methods). The result indicated that Tapiridae diverged from the Rhinocerotidae at ~51.52 million years ago (Ma) [95% highest posterior density (HPD) intervals, 45.70 to 60.52 Ma], while the split between T. terrestris and T. indicus occurred ~17.33 Ma (95% HPD intervals, 11.03 to 24.36 Ma) (Fig. 1C). These estimates remained robust across additional analyses using MCMCTREE and r8s on three datasets: (i) complete coding sequences (CDSs); (ii) 10 replicates of 10-Mb randomly sampled nucleotides from whole-genome alignments (WGAs); and (iii) complete WGAs (fig. S7). Furthermore, ancestral state reconstruction across 196 mammalian species reveals an evolutionary shift in activity patterns (table S10). The results indicated cathemeral activity at the last common ancestor (LCA) of Perissodactyla (support probability, 0.80) and Ungulata (support probability, 0.64). In contrast, the LCA of Tapiridae was reconstructed as nocturnal/crepuscular, with a support probability value of 0.87 (Fig. 1C and fig. S8). These results suggest a possible evolutionary transition from cathemeral to nocturnal/crepuscular activity in the tapir lineage.

Individual heterozygosity and population genetic diversity are well-established correlates of fitness-relevant functional variation (22). T. indicus and T. terrestris are classified as endangered and vulnerable species by the IUCN (10), respectively. Genome-wide heterozygosity analysis against other threatened species (23) revealed low genetic diversity in both tapirs (T. terrestris, 0.31%), with T. indicus showing extremely low heterozygosity (0.086%) (Fig. 1D). Additionally, we reconstructed the demographic histories of T. terrestris and T. indicus using multiple parameters of pairwise sequentially Markovian coalescent (PSMC) according to Hilgers et al. (24) (Fig. 1E and fig. S9). The results showed high consistency across all settings for both species, confirming the robustness of our findings. Specifically, the effective population size (N_e) of T. terrestris experienced a marked decline during 0.7 to 0.4 Ma (part of the broader Middle Pleistocene epoch from 0.778 to 0.126 Ma), coinciding with South American aridification and rainforest fragmentation (25). This was followed by a rise during Marine Isotope Stage 11 (~0.42 to 0.37 Ma) due to humid climate–driven forest recovery (26). In contrast, the N_e of T. indicus peaked during the Early Pleistocene (2.58 to 0.778 Ma), linked to the migration of its ancestors to the Sunda region (27), followed by a decline from the Middle Pleistocene to the present. This decline may have been influenced by Middle Pleistocene forest-to-grassland transitions in Sunda (28) and Late Holocene human activities (27). In summary, these results highlight the persistent conservation needs of Tapiridae species.

Gene evolution and tapir-specific CNEs

To explore the genetic basis of sensory adaptations associated with the diel evolutionary transitions in tapirs, we identified positively selected genes (PSGs), rapidly evolving genes (REGs), genes under relaxed selection, genes with specific amino acid substitutions, and expanded gene families in the ancestral lineage of tapirs. We identified 68 PSGs, 204 REGs, 421 genes under relaxed selection, 631 genes with tapir-specific mutations, and 10 expanded gene families (tables S11 to S14 and Fig. 1C). These genes are enriched in pathways related to sodium ion transport, sensory development, and circadian rhythms.

To assess functional enrichment, we performed Gene Ontology (GO) and pathway analyses using Metascape with a threshold of q value < 0.05. Results revealed that 53 REGs were significantly enriched in organic cation transport (GO:0015695, q value = 3.93 × 10⁻³), monoatomic cation transport (GO:0006812, q value = 0.035), sensory organ development (GO:0007423, q value = 0.014), and the circadian-related regulation of vitamin D receptor signaling pathway (GO:0070562, q value = 0.020) (table S15). Complementary analyses using Metascape and Enrichr revealed nominal enrichment of 15 relaxed selection genes in eye development (GO:0001654, P value < 0.01, q value = 0.056) (table S16) and statistically significant enrichment of 10 relaxed selection genes in the “retinal ganglion cell” category (adjusted P value = 0.037) (table S17), respectively. Moreover, several genetic variations are associated with circadian rhythms. The core circadian clock gene PER1 showed evidence of positive selection (29), while four REGs (TOP1, PTEN, CRTC1, and ATF4) could regulate key rhythm gene expression (fig. S10) (30–33).

We identified 472 tapir-specific CNEs using the WGAs of 13 mammal genomes (table S1), of which 320 CNEs showed evolutionary constraint (phyloP score > 1) when validated against 447-mammal multiple sequence alignments (see Materials and Methods and fig. S11, A and B) (34). Manual examination of three CNEs confirmed their high sequence conservation across mammals (fig. S11, C to E). All CNEs overlapped with candidate cis-regulatory elements (cCREs) from human ENCODE data, with 178 receiving additional validation from mouse ENCODE or cattle assay for transposase-accessible chromatin using sequencing (ATAC-seq) datasets (fig. S12). Furthermore, functional enrichment analysis revealed that these CNEs are significantly enriched for “embryonic morphogenesis” (GO:0048598, binomial false discovery rate q value < 0.01) and related developmental pathways (table S18). Notably, 55 CNEs are linked to genes involved in the development of vision, hearing, and olfaction (table S19).

Evolution of vision in tapirs

The transition to nocturnal and crepuscular lifestyles may have influenced visual evolution in tapirs, characterized by reduced visual reliance and corneal opacity (35). We identified 61 vision-related genes showing evolutionary changes in tapirs: 9 genes under relaxed selection, 12 genes with tapir-specific mutations, and 14 genes with tapir-specific CNEs. These genes involved in the development of corneal, lens, trabecular meshwork, and retinal structures (Fig. 2A and table S19). Specifically, the relaxed selection gene COL8A2 regulates the fate of corneal endothelial cells and is associated with corneal transparency (36). The two tapir-specific mutated genes (CRYBA2 and CRYBB2) encode crystallin proteins (fig. S13, A and B), essential structural components of the vertebrate lens that ensure precise light focusing on the retina for clear vision (37, 38). The trabecular meshwork plays a crucial role in regulating intraocular fluid and maintaining intraocular pressure (39). Three genes involved in trabecular meshwork development showed evolutionary changes: SLC9A1 (associated with tapir-specific CNE97), the relaxed selection gene GSN, and the tapir-specific mutated gene CXCR4 (figs. S13C and S14) (40–42).

Fig. 2. Evolution of vision in tapirs.

(A) Genetic alterations in eye development pathways. Genes are colored according to evolutionary signatures; solid and dashed lines indicate direct and indirect protein interactions, respectively. CLMN, calmin. (B) Uniform Manifold Approximation and Projection (UMAP) visualization of integrated adult mouse and human retinal single-cell transcriptomes. (C) Expression profiles of vision-related genes with tapir-specific variants across retinal cell types. (D) Locations of six tapir-specific amino acid substitutions (blue) within NRP1 functional domains (fig. S16). (E) Structural modeling of the NRP1 (green)–VEGFA (orange) interaction interface. Red dashed lines indicate distances between backbone α-carbon (Cα) atoms of key residues. (F) Coimmunoprecipitation (Co-IP) assays in SH-SY5Y cells showing reduced binding affinity of tapir-type NRP1 to VEGFA compared with the human-type protein. Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) serves as the loading control.

The retina mediates phototransduction and visual processing through a well-defined cellular hierarchy: Photoreceptors transmit signals to intermediate neurons, which relay processed information to retinal ganglion cells (RGCs) whose axons project to the brain via the optic nerve (43). Our analysis revealed extensive retinal-related genetic variation in tapirs. Specifically, three relaxed selection genes (PRDM1, RPGRIP1L, and NEUROD1) and seven tapir-specific mutated genes (CDH23, IFT172, MFRP, MKS1, WDR19, NR2E3, and PDE6B) are involved in the pathway of photoreceptor development (fig. S15) (44–53). Mutations of PRDM1 and NEUROD1 are associated with retinal pigment degeneration, a degenerative retinal disease that often leads to vision loss under low-light conditions (54, 55). Three additional relaxed selection genes (ATF4, NTN1, and EPHB2) are implicated in the development of RGCs (56–58). Relaxed selection on these genes is consistent with reduced selective constraint on photoreceptor maintenance in crepuscular-nocturnal environments. The vascular endothelial growth factor (VEGF) signaling pathway drives retinal vascular development by promoting endothelial cell proliferation, migration, and tube formation (59). We identified multiple tapir genetic variants affecting this pathway (60–64), including two tapir-specific mutated genes (NRP1 and ROCK2), two relaxed selection genes (NRP2 and PIK3R1), and a tapir-specific CNE (CNE74) located in the proximal promoter of FLT1 (Fig. 2A and fig. S16). Specifically, NRP1 and NRP2 encode co-receptors for VEGFA (60).

To characterize the expression patterns and functional roles of these vision-related genes, we examined their expression in mouse and human retinal cells using published single-cell data of the adult mouse and human retina (Fig. 2, B and C) (65–67). We integrated and reannotated eight cell clusters: rod/cone photoreceptors, Müller cells, ON/OFF-cone bipolar cells, RGCs, microglia, and amacrine cells (Fig. 2B, fig. S17, and table S20). Expression analysis revealed that NRP1 is expressed in ON bipolar cells, microglia, and amacrine, while NRP2 is specifically expressed in ON bipolar cells (Fig. 2C). Previous research has indicated that NRP1 binds with high affinity to VEGFA (60). We found that NRP1 harbors six tapir-specific mutations (Fig. 2D and fig. S16A), all within protein functional domains potentially affecting binding affinity with VEGFA (68): P63T, H223Y, and Y248H are situated within the CUB domain, while M437I and P475A are located in the F5/8 type C domain. AlphaFold3 (69) modeling revealed that H223Y and Y248H mutations alter NRP1-VEGFA binding distances. H223Y mutation reduces the distance from 4.9 to 4.3 Å, while Y248H mutation decreases it from 17.0 to 16.4 Å, comparing to original human-type residues (Fig. 2E). These results suggest that these two specific mutations of NRP1 in tapirs may affect the affinity for VEGFA. To further evaluate the combined impact of all six tapir-specific substitutions of NRP1 (P63T, H223Y, Y248H, M437I, P475A, and K744R) on VEGFA interaction, we performed molecular docking analysis using HDOCK (70). The optimal binding conformation of tapir-type NRP1 with VEGFA suggests reduced predicted binding affinity compared to human-type NRP1 (tapir type: −181.6 versus human type: −185.4; a higher docking score represents lower affinity) (fig. S18, A to F), suggesting reduced predicted binding affinity. Crucially, to experimentally validate these computational predictions, we introduced all six tapir-specific substitutions (P63T, H223Y, Y248H, M437I, P475A, and K744R) into human NRP1 (tapir-type NRP1) and performed coimmunoprecipitation (Co-IP) assays in SH-SY5Y neuroblastoma cells coexpressing NRP1 and VEGFA. While human WT NRP1 robustly coimmunoprecipitated with VEGFA, tapir-type NRP1 exhibited substantially reduced binding (Fig. 2F). These results suggest that these mutations of NRP1 in tapirs may impair the affinity to VEGFA. Given that precise NRP1-VEGF signaling is critical for retinal vascularization and neuronal survival, this reduced ligand affinity likely compromises retinal homeostasis. We propose that this functional impairment represents a molecular mechanism of the visual regression observed in tapirs.

FLT1 encodes a tyrosine protein kinase and also serves as a receptor for VEGFA (71). Previous studies indicate that mutations in FLT1 can lead to corneal neovascularization–related diseases, ultimately affecting vision (71). We identified a tapir-specific CNE74 located within the proximal promoter region of FLT1 [704 base pair (bp) upstream of the transcriptional start sites (TSS)], coinciding with an ATAC-seq peak from previous mouse embryonic day 14.5 (E14.5) embryos retinal data (Fig. 3A) (72). CNE74 harbors a 26-bp tapir-specific deletion that disrupts a predicted EBF1/EBF3 binding motif (Fig. 3A and table S21). Given that EBF1 controls Müller cell specification during retinogenesis (73) and FLT1 exhibits Müller cell-specific expression (Fig. 2C), this deletion might alter transcriptional regulation of FLT1 in tapir retina.

Fig. 3. Impaired vision caused by the tapir-specific CNE74 deletion in mice.

(A) Sequence alignment of CNE74 upstream of Flt1 highlighting the tapir-specific 26–base pair (bp) deletion (blue). (B) The corresponding 26-bp sequence was deleted using CRISPR-Cas9 in mice. (C) Sanger sequencing confirmation of the 26-bp deletion in gene-edited mice. (D) Representative fundus fluorescein angiography (FFA) and optical coherence tomography (OCT) images of mice. (E and F) Quantification of whole retinal (E) and outer nuclear layer (ONL) (F) thickness. (G to J) Retinal function assessment via electroretinogram (ERG). Representative scotopic traces (G) and amplitudes (H) at 0.01 cd·s/m²; photopic traces (I) and amplitudes (J) at 3.0 cd·s/m². (K and L) Visual acuity assessment. Schematic of the test chamber (K) and optokinetic spatial frequency thresholds (L) indicating reduced visual acuity in Tapir_CNE74 mice. (M) Retinal expression levels of VEGF signaling genes, shown in transcripts per million (TPM). (N) Gene set enrichment analysis (GSEA) ridge plots showing down-regulation of vision-related pathways. Data are presented as means ± SEM [n = 14 WT and 12 Tapir_CNE74 for (E) to (J); n = 13 WT and 10 Tapir_CNE74 for (L)]. n.s., not significant.

To assess the CNE74 function in vivo, we generated CRISPR-Cas9 gene-edited mice (Tapir_CNE74) by deleting the corresponding 26-bp sequence (Fig. 3, B and C). To exclude potential off-target effects, we used CHOPOFF (74) and Cas-OFFinder (75) to predict putative off-target sites in mouse genome (fig. S19). Whole-genome sequencing and PacBio HiFi sequencing of WT and Tapir_CNE74 mice revealed no divergent homozygous single-nucleotide variations or large insertions/deletions (InDels) at predicted off-target sites or within coding regions (see Materials and Methods and tables S22 and S23), confirming the specificity of the editing.

We evaluated retinal morphology in 6-month-old WT and Tapir_CNE74 mice using fundus photography, fundus fluorescein angiography (FFA), and high-resolution spectral-domain optical coherence tomography (SD-OCT). Fundus photography and FFA revealed minimal overt retinal degeneration in Tapir_CNE74 mice (Fig. 3D), whereas SD-OCT detected significant reductions in overall retinal thickness and outer nuclear layer (ONL) thickness in Tapir_CNE74 mice (Fig. 3, E and F), indicating photoreceptor layer thinning. ERG recordings demonstrated impaired retinal function. Under dark-adapted conditions, both a-wave and b-wave amplitudes decreased at stimulus intensities of 0.01, 3.0, and 10.0 cd·s/m², indicating an attenuated response to the rod-driven pathway (Fig. 3, G and H, and fig. S20). The b-wave of the photopic (cone photoreceptor) ERG was also decreased at a stimulus intensity of 3.0 cd·s/m² (Fig. 3, I and J). We further conducted behavioral tests to assess the visual acuity of the mice. Compared with WT mice, Tapir_CNE74 mice exhibited diminished spatial frequency response to perceived patterns with identical contrast and rotating velocity (Fig. 3, K and L). Transcriptomic analysis of Tapir_CNE74 mice retinal tissue identified 91 differentially expressed genes (DEGs) compared to WT controls, with significant down-regulation of VEGF signaling components including Flt1 and downstream effector Pdpk1 (Fig. 3M, fig. S21, and table S24). Functional enrichment analysis using Metascape revealed that DEGs were significantly enriched in vascular-related pathways, including positive regulation of cell migration (GO:0030335, q value = 0.010) and regulation of vasculature development (GO:1901342, q value = 0.010) (table S25). Subsequent gene set enrichment analysis (GSEA) revealed coordinated suppression of visual development pathways, including “regulation of vasculature development,” “neural retina development,” and “eye photoreceptor cell differentiation” (Fig. 3N). In contrast, pathways associated with mitochondrial and related functions were significantly up-regulated (fig. S22). Together, these data indicate that CNE74 deletion compromises photoreceptor layer integrity and retinal function, reduces visual acuity, and perturbs angiogenic and developmental signaling, consistent with sensory adaptation associated with the shift from cathemeral to nocturnal/crepuscular niches in Tapiridae.

Evolution of auditory sense in tapirs

To explore the evolutionary genetic basis of acute hearing in tapirs, we identified 51 hearing-related genes through comparative genomics, including 2 PSGs, 9 genes with tapir-specific mutations, and 40 genes with tapir-specific CNEs (Fig. 4A and table S19). Expression patterns of these genes were evaluated using published single-cell data from mouse cochlear tissues (76). Eleven clusters of cells were annotated on the basis of known cell-type markers from published literature (fig. S23). Clusters identified consisted of mesenchyme (MES), fibrocyte (FIB), proliferating cells (PRO), vascular cells (VE), hair cells (HC), intermediate cells/melanocytes (INT), sensory epithelial cells (SE), pericytes (PER), glial cells (GLI), neurons (NEU), and macrophages (MAC) (Fig. 4B, fig. S23, and table S26). Briefly, two genes with tapir-specific mutations (PAX3 and MPDZ), expressed in the stria vascularis (SV), generate the endocochlear potential in the scala media and maintain ion homeostasis (fig. S24, A and B) (76). CEMIP with tapir-specific mutations and DDR2 (associated with CNE67) are expressed in the cochlear mesenchymal cells (Fig. 4C and fig. S24C). Two genes with tapir-specific mutations (EPYC and IKZF2) and COL27A1 (associated with CNE298) are expressed in sensory epithelial cells of the cochlea (Fig. 4C and fig. S24, D and E). Two PSGs (CABP2 and MARVELD2) and CHRNA10 with tapir-specific mutations are involved in the development of hair cells (Fig. 4C and fig. S25A) (77–79). Notably, CABP2 also shows convergent positive selection in echolocating birds and mammals with high auditory sensitivity (80). ATP2B1, ATP2B2, and SLC12A6 harbor tapir-specific mutations and contribute to ion homeostasis in the cochlea (fig. S25, B to D) (81). ATP2B1/ATP2B2 mediate Ca²⁺ homeostasis, while SLC12A6 supports K⁺ buffering near hair cells (81). The K cycle, together with the homeostasis of K⁺, Na⁺, Ca²⁺, and pH in the endolymph, is essential for normal inner ear function (82). The DCHS1 gene carries four tapir-specific mutations (fig. S25F) and is required for middle ear development; pathogenic variants in DCHS1 have been linked to structural abnormalities of the middle ear (83).

Fig. 4. Evolution of auditory sense in tapirs.

(A) Genetic variants associated with inner ear development colored by different evolutionary signals. (B) UMAP visualization of cell types in the adult mouse cochlea, based on reanalysis of raw data from Qin et al. (76). (C) Cochlear expression profiles of genes with tapir-specific variants. (D) Auditory brainstem response (ABR) thresholds indicating enhanced sensitivity at 16 kHz in Tapir_CNE74 mice. dBSPL, decibels sound pressure level. (E) Cochlear Flt1 expression levels (TPM) of WT and Tapir_CNE74 mice. (F) Volcano plot of cochlea DEGs of Tapir_CNE74 versus WT mice (log₂ fold change > 1 and adjusted P value < 0.05). (G and H) Sequence alignments of tapir-specific deletions in CNE320 upstream of DDR2 (G) and CNE298 upstream of COL27A1 (H) (blue, tapir sequences). (I and J) Luciferase reporter assays in SH-SY5Y cells assessing regulatory activity of CNE320 (near DDR2) (I) and CNE298 (near COL27A1) (J). Data are presented as means ± SEM [n = 10 WT and 6 Tapir_CNE74 for (D)]. n.s., not significant.

Consistent with its known pleiotropic role in the retina and cochlea (84), we found that FLT1 is specifically highly expressed in cochlear vascular cells (Fig. 4C). Vascular cells, as the principal targets of Norrin signaling, are critical for maintaining the cochlear microenvironment crucial to hair cell survival (85). We found that CNE74, located upstream of FLT1, also overlaps with ATAC-seq peaks obtained from published mouse embryonic E11.5 cochlear data (fig. S26) (77). Notably, existing evidence indicates that the transcription factor EBF1, which binds to this CNE, plays a functional role in cochlear development (86). To characterize the hearing effects of the CNE74 upstream of FLT1, we examined pure tone frequency measurement (8 to 32 kHz) of 6-month-old Tapir_CNE74 mice and WT mice. Compared to WT mice, Tapir_CNE74 mice exhibited a robust improvement in hearing sensitivity at 16 kHz, a frequency falling within the optimal hearing range of mice (Fig. 4D) (87). The auditory brainstem response (ABR) thresholds at 16 kHz shifted from ~60 dB in WT mice to ~40 dB in Tapir_CNE74 mice, representing a substantial 20-dB enhancement in hearing sensitivity. To elucidate the molecular mechanisms driving this phenotype, we performed bulk RNA sequencing on cochlear tissues. The analysis revealed that Flt1 expression levels were significantly reduced in Tapir_CNE74 mice relative to those in WT mice (Fig. 4E), accompanied by broader transcriptomic changes with 44 DEGs, 43 of which were up-regulated (Fig. 4F and table S27). Several DEGs are critical for hearing, including Mmp8, Kng1, Ptafr, and Hspa1b, whose dysregulation or loss of function has been associated with altered auditory performance (88–91). Functional enrichment analysis suggested an association with immune-related pathways, such as lymphocyte homeostasis (GO:0002260, P value = 2.83 × 10⁻⁶, q value = 0.06) (table S28). This observation is further supported by GSEA analysis, which indicated that Tapir_CNE74 mice exhibited activation of immune-response programs (e.g., immunoglobulin complex) and suppression of cholinergic signaling terms (fig. S27). These transcriptomic signatures suggest potential alterations in cochlear homeostasis and the local microenvironment. However, the precise contribution of these specific molecular changes to the auditory phenotype in gene-edited mice warrants further verification in future studies.

Despite CNE74, we have identified 40 other tapir divergent CNEs associated with hearing related genes (table S19). For instance, CNE320 and CNE298 located in the proximal promoter of DDR2 (−47 bp from the TSS) and COL27A1 (−124 bp from the TSS), respectively. COL27A1 encodes collagen fibers and is expressed in the basement membrane of the cochlea. Mutations in this gene may cause Steel syndrome, and result in hearing impairment (92). Both of CNE320 and CNE298 are overlapped with ATAC-seq peaks retrieved from published mouse embryonic E11.5 cochlear data (Fig. 4, G and H) (77). CNE320 contains a predicted binding site for ZNF85, a transcriptional repressor with broad tissue expression according to the Human Protein Atlas. CNE298 contains a predicted binding motif for the transcription factor ZIC1, which is involved in the development of the neuroepithelium during mouse inner ear development (93).

To assess the regulatory activity of these CNEs in a more physiologically relevant context, we performed dual-luciferase reporter assays in two cell types: SH-SY5Y neuroblastoma cells (a neuronal model) alongside human embryonic kidney (HEK) 293T cells (a nonneuronal model). Tapir-specific CNE320 and CNE298 exhibit significantly reduced transcriptional activity in SH-SY5Y neuroblastoma cells (Fig. 4, I and J), suggesting that these tapir-specific deletions may alter regulatory potential activity in neuronal contexts. Notably, the same CNEs exhibited increased activity in nonneuronal HEK293T cells (fig. S28, A and B), reflecting cell-type–specific transcription factor repertoires and chromatin contexts. This context-dependent regulatory pattern is consistent with previous studies showing that enhancer activities can vary markedly across cell types (94).

These results suggest that tapir-specific genetic variants, particularly the FLT1 CNE74, might have functional consequences for auditory system development and performance. While direct auditory measurements in tapirs are not currently available, we can infer the ecological significance of enhanced sensitivity at 16 kHz through phylogenetically informed comparisons and behavioral observations. Horses, closely related to tapirs, exhibit optimal hearing sensitivity in the range of 1 to 16 kHz (95), suggesting that 16 kHz likely falls within or near the functionally relevant hearing range for tapirs as well. Previous behavioral studies have documented that tapirs engage in rich vocal communication, including various whistles, squeaks, and alarm calls (96). We hypothesize that enhanced auditory sensitivity at 16 kHz would confer several adaptive advantages for intraspecific communication and detect approaching predators. However, we acknowledge that this interpretation remains hypothetical and requires further validation through direct physiological measurements of tapir auditory capabilities and additional genetic evidence.

Evolution of olfaction and taste in tapirs

Olfactory receptors (ORs) are G protein–coupled receptors with seven transmembrane domains, capable of detecting various odor molecules in the environment (97). We performed genome-wide identification of OR genes across tapirs and out-group mammals. Results revealed that tapirs exhibit a higher number of OR genes than other herbivorous ungulates (except omnivorous pig) while maintaining similar pseudogene proportions comparable to other ungulates (Fig. 5A). Phylogenetic reconstruction clearly segregated intact functional OR genes into class I and class II categories (fig. S29). Notably, tapirs have the largest repertoire of class I genes among the examined species (Fig. 5A). This expansion likely reflects sensory trade-offs and lineage-specific ecological adaptations to novel niches during diel niche shift. Moreover, PTCH1 is an important component of the SHH signaling pathway. Mutations in this gene cause anosmia, nasal plate morphology changes, and the generation of olfactory sensory neurons and olfactory ensheathing cells (98). We identified three mutations specific to tapirs in PTCH1 (N349S, V406A, and K418E) (Fig. 5B). Although these three sites are not located in specific domains, three-dimensional protein structure modeling suggests that N349S substantially shortens the nearby β sheet (Fig. 5C and fig. S30), which may affect the function of PTCH1.

Fig. 5. Evolution of olfaction and taste in tapirs.

(A) Comparative repertoire of olfactory receptor (OR) genes across nine mammalian species. Bars indicate counts of intact (I), truncated (T), and pseudogenized (P) genes; tapirs exhibit a notable expansion of class I ORs. (B) Sequence alignment highlighting three tapir-specific amino acid substitutions (N349S, V406A, and K418E) in PTCH1. (C) Structural superposition of tapir (colored) and human (gray) PTCH1, predicting a shortened β sheet induced by the N349S mutation (fig. S30). (D) Evolutionary alterations in taste receptor pathways. Genes are colored by evolutionary status: red, PSGs; gray, genes with tapir-specific mutations; black, putative pseudogenes (TAS2R38 and TAS2R3).

In terms of taste, vertebrates perceive five main taste categories: sweet, umami, bitter, sour, and salty, which are activated through physical interactions between taste receptors and external taste molecules (99). Taste receptor genes belong to the TAS1R and TAS2R subfamilies (99). Tapirs predominantly consume succulent twigs, leaves, and wild fruits, particularly aquatic plants, which distinguishes them significantly from obligate herbivores (17). We detected multiple genetic variations in tapir taste receptor genes (Fig. 5D). Notably, TAS2R38 and TAS2R3 are lost in tapirs, which belong to TAS2R gene family associated with bitterness (100) (fig. S31). TRPM5, a critical transducer for bitter, sweet, and umami taste (101), shows signatures of positive selection in tapirs. We also identified seven tapir-specific variants in TRPV1 (fig. S32), a receptor involved in salty taste (102); three substitutions (K139S, M163V, and K201R) are located in the ankyrin repeat domain, potentially affecting channel function.

Together, the enlarged class I OR repertoire, lineage-specific PTCH1 substitutions, and reshaped taste gene landscape suggest potential chemosensory evolution in tapirs. A broadened capacity for water-soluble/volatile detection via class I ORs aligns with frequent use of riparian and aquatic habitats. PTCH1 changes may modulate olfactory system development and sensitivity, although the functional effects remain to be verified. Concurrently, loss of bitter receptors TAS2R38/TAS2R3 reduced bitter-receptor diversity, coupled with enhanced downstream transduction (TRPM5) and modified ion channel properties (TRPV1), which is consistent with selective tolerance for certain plant secondary compounds and efficient detection of energy-rich fruits and aquatic vegetation. These patterns seem to be consistent with tapirs’ riparian habitat use, nocturnal activity, and diverse plant diet, although functional validation is required in future studies.

DISCUSSION

In this study, using PacBio HiFi and Hi-C technologies, we successfully assembled high-quality chromosome-level genomes for two tapirs (T. indicus and T. terrestris). Our results indicate that T. indicus exhibits a continuous decline in effective population size and lower heterozygosity, providing insights into the population genetics of species diversity and conservation for T. indicus. Through comparative genomic analysis, we identified a range of genetic variations related to sensory adaptations involving vision, hearing, olfaction, and taste in tapirs. While previous research has mainly focused on vision and olfaction in the evolutionary shifts of diel niches, for example, in rodents and night monkey (8, 9, 103, 104) and predominantly focused on protein-coding changes, our results extend previous findings by demonstrating that evolutionary transitions in diel activity patterns involve coordinated changes across multiple sensory modalities at both coding and regulatory levels. Specifically, we identified gene variants related to vision, audition, olfaction, and gustation, accompanied by extensive changes in cis-regulatory elements. This multifaceted genomic remodeling, encompassing both protein function and gene regulation, suggests that adaptation to crepuscular/nocturnal activity requires system-wide reorganization of sensory systems. These findings broaden our understanding of the genomic architecture underlying temporal niche evolution and suggest that diel activity transitions impose selection on integrated sensory systems at multiple regulatory levels rather than isolated changes in protein-coding sequences.

We functionally validated one Tapiridae-specific CNE (CNE74) upstream of FLT1 gene related to the retina and cochlea, using CRISPR-Cas9–edited mice. The Tapir_CNE74–edited mice exhibited pleiotropic effects, resulting decreased visual acuity but increased auditory sensitivity. By analyzing the bulk transcriptome of the retina and cochlea in gene-edited mice, we found that Tapir_CNE74 affects Flt1 expression in both retina and cochlea. While Flt1 is down-regulated in both tissues, the DEGs show limited overlap between cochlea and retina with only one gene (Diaph3). This pattern suggests that, while CNE74 regulates Flt1 expression in both tissues, the downstream consequences are tissue specific or cell specific, likely reflecting distinct regulatory networks. This could be one explanation for the pleiotropic functions of Tapir_CNE74 in different tissues, but we should also recognize that it may be influenced by other regulatory factors such as different cellular chromatin states. These findings suggest that regulatory element evolution may have pleiotropic effects on multiple sensory systems and provide a genomic framework for understanding temporal niche adaptation in mammals (Fig. 6).

Fig. 6. Sensory trade-offs associated with FLT1 regulatory evolution.

Schematic illustrating the CRISPR-Cas9 editing of the 26-bp CNE74 sequence upstream of Flt1 (top) and the resulting physiological consequences (bottom). The genetic modification leads to a sensory trade-off characterized by compromised visual function (eye) but enhanced auditory sensitivity (cochlea), providing a potential genomic mechanism for temporal niche adaptation in tapirs.

The evolutionary forces driving the fixation of the pleiotropic Tapir_CNE74 deletion remain open to interpretation. We hypothesize a scenario where the nocturnal niche acted as a permissive environment that altered the cost-benefit dynamic of this sensory trade-off. Under low-light conditions, the fitness penalty of CNE74-mediated retinal degeneration may have been minimized because of relaxed visual constraints. Conversely, the concomitant enhancement of auditory sensitivity could have conferred a distinct survival advantage for individual communications or predator detection in the dark. Thus, we postulate that, despite the functional “cost” to the retina, the aggregate fitness gain potentially favored the variant’s retention. While our population genomic scans in Asian tapirs [11 published genomes (105) and our newly sequenced genome] did not yield classical selective sweep signals around the FLT1 locus and other vision-related genes (fig. S33, A to H, and tables S29 to S33), this absence might be confounded by severe demographic history. Asian tapirs display numerous short run-of-homozygosity (ROH) segments (<500 kb), indicative of historical bottlenecks. The extensive short ROHs and low effective population size observed in PSMC (Fig. 1D and fig. S34, A and B) point to a history of strong genetic drift, which can obscure haplotype-based signatures of past selection. It is conceivable that genetic drift facilitated the initial survival of the CNE74 deletion in a small ancestral population, potentially enabling subsequent positive selection on auditory function to maintain it. This plausible synergy between ecological opportunity and demographic stochasticity may have collectively shaped the unique sensory landscape of modern tapirs.

The identification of the FLT1-associated CNE74 offers a potential genomic target for understanding the mechanisms of retinal neovascularization, which underlies many fundus diseases (106). The VEGFA-FLT1/VEGFR2 axis is a primary therapeutic target for physiological angiogenesis (106). However, current anti-VEGF therapies face challenges such as nonresponse, resistance, and the burden of frequent injections (107). Receptor-targeted strategies, including FLT1, are therefore under investigation (108). In this context, the tapir-specific CNE provides an evolutionary perspective on how endogenous noncoding elements modulate FLT1 expression. While clinical translation remains challenging because of species differences, pleiotropy, and safety concerns, this element provides testable leads for dissecting mechanisms of angiogenesis-related retinal disease.

While our study provides primary evidence linking specific regulatory elements to sensory phenotypes, several limitations should be acknowledged. (i) Our functional validations were conducted in heterologous cell lines and mouse models, which provide valuable mechanistic insights but serve only as indirect proxies for tapir physiology. This constraint is unavoidable in research on endangered species where invasive experimentation is precluded. Future studies should aim to bridge this gap by establishing species-specific cellular models, such as induced pluripotent stem cells and retinal/cochlear organoids, to validate these candidate variants in a native genomic context. The primary contribution of our study is therefore to identify and prioritize candidate loci for future validation in tapirs. (ii) Despite functional validation of several CNEs and genes, our genome-wide screen identified additional vision- and hearing-related genes and candidate regulatory elements (table S19) that may collectively contribute to the sensory phenotype. Systematic experimental validation of these additional candidates, both individually and in combination, will be necessary to fully elucidate the polygenic architecture underlying sensory evolution in tapirs. (iii) We propose that both natural selection and demographic events (e.g., population bottlenecks) might have driven the evolution of the sensory system in tapirs. However, our population genomic analyses were based solely on data from Asian tapirs, which limits our ability to distinguish between adaptive and neutral evolutionary scenarios across the tapir phylogeny. Additional population genomic data from other extant tapir species and, if feasible, ancient DNA from extinct tapirid lineages would help strengthen the demographic analysis.

Despite these limitations, our study establishes a comprehensive framework for dissecting the molecular mechanisms underlying sensory evolution in the context of temporal niche transitions. The candidate loci identified here provide valuable resources for future studies, and our interdisciplinary methodology may be applicable to investigating adaptive evolution in other endangered or nonmodel species, particularly those undergoing similar diel niche shifts.

MATERIALS AND METHODS

Animals

This study collected animals legally in accordance with the policies of the Animal Research and Ethics Committee of Guangzhou Zoo and Guangzhou Wildlife Research Center. All animal experiments were approved by the Animal Research and Ethics Committee of Guangzhou Wildlife Research Center (approval number GZZ0020241201). Blood samples were collected from one female T. indicus and one male T. terrestris during routine health checks, provided by Guangzhou Zoo and Guangzhou Wildlife Research Center. The blood samples were used for genomic and Hi-C sequencing. The CNE74 CRISPR-Cas9 editing was performed in WT mice (Mus musculus, strain C57BL/6). Six-month-old mice were used for retinal and cochlear transcriptome sequencing.

Genome sequencing

For T. terrestris and T. indicus, we performed sequencing using PacBio in HiFi mode. SMRTbell library construction and sequencing were performed at Haorui Genomics (Xi’an, China) for T. indicus and T. terrestris, following PacBio’s official protocol to prepare ~20 kb SMRT Bell libraries. After obtaining the sequence data, we processed the raw reads using SMRT Link (v8.0, URL: https://github.com/PacificBiosciences/pbcommand) and applied the conservative convergent site (CCS) method with default parameters.

Hi-C libraries were constructed following a standard protocol with minor modifications. Briefly, blood cells from T. terrestris and T. indicus were isolated and cross-linked. Following lysis, chromatin was digested overnight with the restriction enzyme Mbo I. Proximity ligation was carried out to reconnect chromatin ends, followed by cross-link reversal with proteinase K at 65°C and phenol-chloroform purification. Biotin was removed from unligated ends, and DNA was sheared to 300 to 500 bp. Biotinylated fragments were then enriched with streptavidin C1 beads. Following A-tailing and adapter ligation, we amplified the libraries (12 to 14 cycles) for sequencing on the Illumina platform (HRORUI GENOMICS, Xi’an, China).

Genome assembly

PacBio HiFi reads and Hi-C reads were used to perform de novo genome assembly of T. terrestris and T. indicus. The final genome assemblies were generated following contig-level purging, Hi-C-based scaffolding, and a round of manual curation (see Supplementary Methods).

Assembly quality evaluation

We assessed genome completeness and consensus QV using Merqury (109). The QV score indicates the level of confidence in the accuracy of the shared sequences within the sequencing data. Furthermore, we assessed genome assembly completeness with BUSCO (v5.1.2) (110) using the mammalia_odb10 database.

Genome annotation

Repetitive elements in the genomes of T. terrestris and T. indicus were identified by their matches to Repbase (v.20181026) using RepeatMasker (v.4.1.6) (111). To identify protein-coding genes, we used both de novo and homology-based method to annotate genes (see Supplementary Methods).

Phylogenetic tree construction and divergence time estimation

The phylogenetic relationship of the two tapirs and out-group species was reconstructed (table S1). On the basis of 8956 single-copy BUSCO genes from 11 species, we reconstructed a maximum likelihood (ML) species tree and estimated divergence times using four fossil calibrations (table S9). The robustness of divergence time estimates at internal phylogenetic nodes was assessed using fourfold degenerate sites (4dTV sites), full coding sequences (CDS), and genome-wide random samples (fig. S7, A and B), with additional calibration provided by genome-wide alignments (fig. S7C) (see Supplementary Methods).

Collinearity analysis and chromosome evolution analysis

We performed a whole-genome syntenic analysis among the four Perissodactyla species (T. indicus, T. terrestris, E. caballus, and D. bicornis) to identify homologous blocks, followed by a systematic alignment of tapir genomes with multiple out-groups to reconstruct the Tapiridae ancestral chromosomes (see Supplementary Methods).

Ancestral state reconstruction

Activity pattern data of 196 mammalian species were obtained from published literature (table S10). On the basis of the phylogenetic tree obtained from Timetree (https://timetree.org/home), we reconstructed ancestral states using Mesquite v3.04 (112) based on the phylogenetic tree generated in this study. The ML method was applied under the Markov k-state 1 model, which assumes equal transition probabilities among character states.

Genome wide heterozygosity and historical effective population size analysis

Genetic diversity and historical dynamics are crucial for species survival. We estimated genome-wide heterozygosity and mutation rates for the two tapir species (see Supplementary Methods). The demographic trajectories were then reconstructed using PSMC (24) with the parameter sets “N25 -t15 -r5 -p 4+25*2+4+6” (default), “2+2+25*2+4+6,” and “1+1+1+1+25*2+4+6” (fig. S9).

Expansion and contraction of gene family

Gene families for T. indicus and T. terrestris (this study), D. bicornis, Equus asinus, E. caballus, B. taurus, Sus scrofa, Camelus dromedarius, C. familiaris, M. musculus, and Homo sapiens were constructed, and gene family expansion and contraction analyses were performed (see Supplementary Methods). Functional enrichment analysis of the expanded gene families at the Tapiridae ancestral node was conducted using Metascape (v3.5) (113), with P value cutoff set at 0.01.

PSG and REG identification

We identified selection signatures for 8956 single-copy orthologs in our 11-species dataset using PAML codeml (v4.8) (114). To identify PSGs on Tapiridae branches, we compared branch-site models that permitted a codon site class with the ratio of nonsynonymous to synonymous substitutions (dN/dS) >1 on foreground branches against branch-site null models. Positive selection sites were identified using Bayes empirical Bayes in PAML.

We identified REGs in tapirs, defined as genes with elevated dN/dS ratios, using the PAML branch model. The two-ratio model (model = 2) differentiates between evolutionary rates for background and foreground (tapir) branches, while the one-ratio model (model = 0) assumes a uniform rate for all branches. Genes that exhibited a P value, calculated through the chi-square statistic, below 0.01, along with a higher ω value in the foreground lineage, were considered REGs. Last, we performed functional enrichment analysis using Metascape (v3.5) (113) on the PSGs and REGs identified at the ancestral node of tapirs, with P value cutoff set at 0.01.

Relaxed selection analysis

We used HyPhy RELAX model (v2.5.14) (115) to identify genes under relaxed selection with default parameters and the established tree topology. The Benjamini-Hochberg method adjusted the P values, identifying genes with adjusted P values < 0.05 as significantly relaxed. Furthermore, we performed functional enrichment analysis on the relaxed genes from the ancestral tapir node using Metascape (v3.5) (113) and Enrichr (116), with adjusted P value cutoff set at 0.05.

Analysis of CNEs

We constructed 13-way WGAs referenced to the human genome (GRCh38). On the basis of nonconserved models generated by PhyloFit, we identified highly conserved elements (HCEs) using PhastCons with a rho parameter of 0.3. These HCEs were strictly filtered to remove coding sequences and retain only those overlapping human ENCODE cCREs. We then identified tapir-specific CNEs on the basis of the following criteria: “(i) ≥90% sequence identity within the variant region among in-group species; (ii) ≥80% sequence identity in the 20-bp flanking regions across all species; and (iii) a variant region length of >6 bp.” Potential artifacts were removed by realigning human and bovine homologs to the tapir genome to confirm synteny. To ensure that these loci represent evolutionarily constrained regions, we filtered for elements with a mean PhyloP score of >1 based on 447-mammal alignments (GRCh38 reference). Last, CNEs were functionally annotated to target genes using GREAT, defining the regulatory domain as the region within 300 kb of a transcription start site. Detailed procedures are provided in Supplementary Methods.

Analysis of ATAC-seq data

To investigate the regulatory activity of the identified tapir-specific CNEs, we curated a collection of epigenomic resources, comprising mouse ENCODE data and ATAC-seq profiles from bovine tissues and mouse sensory organs (retina, whole eye, and cochlea; table S34). ATAC-seq data from different tissues were aligned to respective reference genomes, followed by identification of open chromatin regions and subsequent motif enrichment analysis within CNEs located in these regions (see Supplementary Methods).

Identification of amino acid changes

We applied the CCS method to identify tapir-specific amino acid substitutions across 8956 single-copy orthologous genes. To exclude potential assembly errors and individual variation, we mapped second-generation sequencing reads from four tapir individuals across the genomes and visually inspected the alignments. Detailed methods are provided in Supplementary Methods.

OR identification and classification

Mammal ORs were collected from the Horde database (https://genome.weizmann.ac.il/horde/) as reference sequences. The ORs were aligned with the genomes of nine species, including T. indicus (this study), T. terrestris (this study), D. bicornis, E. asinus, E. caballus, B. taurus, S. scrofa, C. dromedarius, and C. familiaris, using ORFAM (https://github.com/jianzuoyi/orfam). Complete OR genes and pseudogenes were identified using the func and pseudo programs in ORFAM, respectively. We generated multiple amino acid sequence alignments and reconstructed the phylogenetic trees based on ML (see Supplementary Methods).

Reanalysis of public single-cell transcriptomes from retina and cochlea

To comprehensively characterize retinal cell diversity and examine hearing-related gene expression, we collected expression matrices from multiple datasets: adult mouse retina (GSM4995561, GSM7184523, and GSM7184525) and adult human retina (GSM5567531, GSM5567532, and GSM5567534) and mouse cochlea (GSM6716222). We performed independent downstream analysis using the Seurat package.

Raw data were filtered using parameters: “min.cells = 3, min.features = 200, 50 < nFeature_RNA < 10,000, and percent.mt < 20.” For the retinal datasets, homologous gene sets between mouse and human were obtained following established method (117). To mitigate batch effects and species-specific variance, datasets were integrated using canonical correlation analysis. Specifically, gene counts were normalized, and the top 2000 highly variable genes from each dataset were identified. Integration anchors were determined (FindIntegrationAnchors), and data were consolidated (IntegrateData) with dims = 1:20. For the cochlear dataset, data were normalized and scaled directly without integration. Both integrated retinal data and scaled cochlear data underwent principal components analysis and Uniform Manifold Approximation and Projection (UMAP) for visualization. Cell clustering was performed at a resolution of 0.1 for retinal cells and 0.2 for cochlear cells. Marker genes were identified using the FindAllMarkers function with the Wilcoxon rank sum test (log₂ fold change > 1; min.pct > 0.25).

Co-IP and IB of NRP1 and VEGFA

Synthesized sequences of human NRP1 and tapir-specific NRP1 (harboring mutations P63T, H223Y, Y248H, M437I, P475A, and K744R) were cloned into pCDNA3.1-HA/Flag vectors (Tsingke Biotechnology). The SH-SY5Y (human neuroblastoma cell line) was obtained from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China) and authenticated via short tandem repeat profiling. Cells were maintained in Dulbecco’s modified Eagle’s medium (DMEM; ExCell) supplemented with 10% fetal bovine serum (FBS), penicillin (100 U/ml), and streptomycin (100 μg/ml) at 37°C in a 5% CO₂ humidified incubator.

Co-IP and immunoblotting (IB) were performed as described previously (118). Briefly, SH-SY5Y cells were transfected with NRP1-Flag or NRP1-HA plasmids. After 48 hours, cells were harvested and lysed on ice for 30 min in lysis buffer containing protease inhibitors. Lysates were centrifuged at 12,000 rpm for 15 min at 4°C, and the supernatants were collected as total cell lysates. For Co-IP assays, the lysate was divided into input, immunoprecipitation (IP), and immunoglobulin G (IgG) control groups (1:4:2 ratio). The IP and IgG groups were incubated with anti-VEGFA antibody (Proteintech, 19003-1-AP) or control IgG at 4°C for 12 to 18 hours, followed by 4-hour capture with Protein A/G beads. Immunocomplexes were washed three times with cold wash buffer, resuspended in protein loading buffer, and denatured by boiling for 10 min. Protein samples were separated by SDS–polyacrylamide gel electrophoresis, transferred to polyvinylidene difluoride membranes (Merck Millipore), and blocked with 5% skim milk. Membranes were immunoblotted overnight at 4°C with the following primary antibodies: mouse anti-HA (Proteintech, 2367; 1:1000), rabbit anti-Flag (Proteintech, 14793; 1:1000), anti-VEGFA (Proteintech, 19003-1-AP; 1:1000), and mouse anti–glyceraldehyde-3-phosphate dehydrogenase (Proteintech, 60004-1-Ig; 1:1000). The membranes were then incubated with a mixture of horseradish peroxidase–conjugated goat anti-mouse (Proteintech, SA00001-1) and goat anti-rabbit (Proteintech, SA00001-2) secondary antibodies (diluted 1:1000) for 1 hour at room temperature. After three 15-min washes with tris-buffered saline with Tween 20 (TBST), protein signals were visualized using SuperSignal West Pico Chemiluminescent Substrate (Thermo Fisher Scientific).

Dual-luciferase reporter assays

To assess the regulatory potential of identified CNEs, selected sequences were cloned into the pGL3-Basic vector (sequence verified by Sanger sequencing). SH-SY5Y human neuroblastoma cells were cultured in DMEM (ExCell) supplemented with 10% FBS, penicillin (100 U/ml), and streptomycin (100 μg/ml). HEK293 cells were cultured in DMEM (Gibco) containing 10% FBS, 2 mM glutamine, penicillin (100 U/ml), and streptomycin (100 μg/ml). Cells were maintained at 37°C in a humidified 5% CO₂ incubator. For reporter assays, cells were seeded in 24-well plates at a density of 4 × 10⁵ cells per well 24 hours before transfection.

Transfection was performed using Lipofectamine 2000. Briefly, 2 μg of pGL3-basic luciferase reporter plasmid and 0.02 μg of internal control plasmid pRL-TK were mixed with 2 μl of Lipofectamine 2000 transfection reagent in 50 μl of serum-free DMEM and incubated for 15 min at room temperature. The mixture was cotransfected into cells, and cellular lysates were collected using lysis buffer (RG027-1) after 36 hours. Luciferase activity was assessed using a dual-luciferase reporter assay kit (Vazyme Biotech Co. Ltd., China) and measured with a Synergy H1 multimode microplate reader (BioTek, USA). The relative luciferase activity was calculated by normalizing firefly luciferase signals to Renilla luciferase signals. Data were analyzed using GraphPad Prism software (v9.0.0).

Generation of Tapir_CNE74 gene-edited mice

Knockout mice with a 26-bp deletion in the CNE74 region, which is specific to tapirs relative to out-group species, were designed and generated by Model Organisms Center Inc. (Shanghai, China). Briefly, Cas9 mRNA was synthesized in vitro using the mMESSAGE mMACHINE T7 Ultra Kit (Ambion, TX, USA) following the manufacturer’s protocols, and the resulting mRNA was purified with the MEGAclear Kit (Thermo Fisher Scientific, USA). The sequences were chosen as the single guide RNA (sgRNA) targeting Cas9 (gRNA1: 5′-GAGGGAGTCTGTGTCTTTTT-3′, PAM: CGG; gRNA2: 5′-TCCCTCCCTTGGGACCCGCG-3′, PAM: CGG), which was also synthesized in vitro using the MEGAshortscript Kit (Thermo Fisher Scientific, USA) and subsequently purified with the MEGAclear Kit. Both the transcribed Cas9 mRNA and sgRNA, along with a 120-bp single-stranded oligodeoxynucleotide, were coinjected into the zygotes of C57BL/6J mice. F0 mice with the anticipated mutations were then crossed with WT C57BL/6J mice to generate F1 mice. The genotypes of the F1 mice were determined by polymerase chain reaction (PCR) and confirmed through sequencing (table S35).

Analysis of off-target effects

To exclude the possibility that the observed phenotypes resulted from off-target effects, we performed a systematic evaluation (see Supplementary Methods). Sequencing data confirmed successful disruption of the target sites, with no substantial off-target effects detected (fig. S19, A to C). Specifically, we identified 6169 potential off-target sites. Sequencing analysis of the Tapir_CNE74 mutant identified 114 sequence variants distinct from those in the WT and reference genomes (table S23). These variants included 108 single-nucleotide polymorphisms (SNPs) and six InDels. All variants were heterozygous and predominantly located in intergenic regions, with no variants detected within coding sequences (table S23 and fig. S19D). None of these variants overlapped with the predicted off-target sites. Additionally, structural variant analysis based on HiFi reads showed no detection of significant large-fragment deletions in Tapir_CNE74 mutants. Therefore, the observed phenotypes are most likely caused by the intended disruption of the target region rather than off-target effects.

Electroretinogram (ERG)

The retinal ERG was recorded using an Espion electroretinography system (Diagnosys, USA). Six-month-old mice were dark-adapted overnight and then anesthetized with a combination of pentobarbital sodium and xylazine hydrochloride injection. Pupils were dilated with 0.5% tropicamide for 10 min. Ofloxacin ophthalmic ointment was applied to keep the mouse corneas moist. Reference electrodes were carefully inserted into both side cheeks, gold ring electrodes were attached to the corneas of both eyes, and ground electrodes were inserted into the tail. The chart pattern was then checked to confirm the stability of the ERG waveform. A light-emitting diode light source provided the flash stimuli for electroretinography. Dark-adapted ERGs were recorded at 0.01, 3.0, and 10.0 cd·s/m². After stable background illumination at 30 cd·s/m² for 5 min, light-adapted ERGs with intensity of 3.0 cd·s/m² were elicited. The results are presented as the means ± SEM, and the statistical significance was assessed using two-tailed Student’s t test and one-way analysis of variance (ANOVA). All the experiments were conducted using a blinded design.

Fundus images and high-resolution SD-OCT images

Mice were maintained under anesthesia. Before fundus photography examination, ofloxacin ophthalmic ointment was applied to the eyes to improve connection with the machine (Micron III, Phoenix Research Labs). For FFA, 10% sodium fluorescein was injected intraperitoneally at 0.01 ml/g of body weight (119). All fundus images and optical coherence tomography (OCT) images were acquired by setting the optic nerve head as references to find the most suitable position. Final OCT images were obtained by averaging each 30 pictures (Envisu R2000, Leica). Fuji software was used to measure the thickness of whole retina and the thickness of ONL. The results are presented as the means ± SEM, and the statistical significance was assessed using one-way ANOVA. All the experiments were conducted using a blinded design.

Measurement of visual acuity in mice

Visual acuity in mice was measured using an OptoDrum (StriaTech, Tübingen, Germany). The animal was placed on a high platform, surrounded by rotating stripes containing spatial frequencies. The system captures the head reflection movements of mice at different spatial frequencies with 100% contrast and a rotational speed of 12°/s through cameras. Last, the highest spatial frequency that each mouse could see was identified and recorded as the response threshold of that mouse. The conductor of these tests was blinded to the background of mice. The results are presented as the means ± SEM, and the statistical significance was assessed using two-tailed Student’s t test.

Auditory brainstem response

ABR in mice was measured using a TDT RZ6 (Tucker-Davis Technologies) system. Animals were anesthetized and placed on a heating pad to maintain body temperature. We inserted the recording electrode behind the measuring ear, the reference electrode on the apex of the mouse brain, and the grounding electrode behind the opposite ear. We tested the minimum recognizable stimulus decibel levels that mice could hear at frequencies of 8, 16, 24, and 32 kHz. Specifically, at each stimulation frequency, the detection starts at 90 dB and decreases in increments of 5 dB until the test curve of the mouse no longer has an amplitude. We set this as the ABR threshold of the mouse in this frequency. The conductor of these tests was blinded to the background of mice. The results are presented as the means ± SEM, and the statistical significance was assessed using one-way ANOVA.

Transcriptome sequencing

To investigate whether gene expression in Tapir_CNE74 mice changes during visual and auditory development, we performed sequencing on the retinas and cochleae of WT and Tapir_CNE74 mice. After decapitation, both sides of eyes and otic capsules were removed and placed in 1× phosphate-buffered saline for further retina and cochlea dissection. Total retina RNA and cochlea RNA were extracted, respectively, using TRIzol reagent according to the instruction manual. The cDNA libraries were constructed using the Illumina TruSeq RNA Library Preparation Kit and sequenced on the Illumina NovaSeq 6000 platform, generating paired-end reads of 150 bp. After quality control of the raw RNA sequencing data, the reads were aligned to the mouse reference genome, followed by the acquisition of high-quality aligned reads and removal of PCR duplicates. Gene expression quantification and differential expression analysis were subsequently performed, and functional enrichment of the DEGs was conducted through GSEA (see Supplementary Methods for details).

Selective sweep identification and ROH calling

On the basis of whole-genome resequencing data from 12 Asian tapirs [11 from our published dataset (105) and one newly sequenced in this study], we performed comprehensive genomic analyses to identify selective sweeps and ROHs. After read mapping and stringent SNP filtering, genome-wide selection signals were scanned using composite likelihood ratio (CLR) and integrated haplotype score (iHS) methods across 50- and 20-kb window sizes. Genomic regions exhibiting empirical P values within the top 1% for both CLR and iHS were defined as candidate selective sweeps. Functional annotation and population genetic analyses, including Tajima’s D, were subsequently conducted on genes within these regions. Additionally, genome-wide ROH segments were systematically identified to assess homozygosity patterns (see Supplementary Methods).

Supplementary Materials

The PDF file includes:

Supplementary Methods

Figs. S1 to S34

Legends for tables S1 to S35

References

Other Supplementary Material for this manuscript includes the following:

Tables S1 to S35