Extraordinary diversity of antiviral immune-related gene families across 38 bat species

Abstract

Background

Despite harboring diverse pathogenic viruses, bats rarely exhibit clinical symptoms of the diseases. Previous research has reported evolutionary characteristics in key antiviral gene families in bats, such as natural killer cell receptors, MHC class I genes, and type I interferons, suggesting that bats may possess an immune tolerance that allows them to host viruses asymptomatically. However, this hypothesis is based on limited datasets and requires more comprehensive examinations.

Results

We assembled a chromosome-level reference genome of the Chinese horseshoe bat (Rhinolophus sinicus), a recognized reservoir for SARS-like coronaviruses. By combining this genome with data from 37 other bat species spanning major lineages, we have discovered that the evolutionary signatures of these antiviral gene families exhibit lineage-specific characteristics in gene repertoire, genomic distribution, signaling modes, and expression patterns. Furthermore, we found that the evolutionary diversification of these antiviral gene families is largely influenced by the richness and diversity of viruses, particularly in bats that host more viruses.

Conclusions

Our findings offer insights into the immune adaptations of bats in response to viral infections and reveal a greater interspecies evolutionary heterogeneity in their antiviral immune systems than previously recognized.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12915-026-02571-1.

Background

Bats (order Chiroptera) are renowned among mammals for their ability to harbor a wide variety of zoonotic viruses, such as Lyssaviruses, Nipah, Hendra, Ebola, Marburg, and SARS-CoV-2, without showing symptoms of disease [1, 2], despite these pathogens causing severe illness in humans and other mammals [3]. Experimental infections further demonstrate that bats often exhibit limited or no clinical signs, even when exposed to viruses that are highly pathogenic in other hosts [4–6]. This unusual tolerance to viral infection has sparked growing interest in elucidating the underlying immunological mechanisms.

Previous genomic comparisons between bats and other mammals have revealed several adaptations that may contribute to bats’ unique antiviral immunity. These include changes in multiple immune gene families involved in viral detection, antigen presentation, interferon signaling, and inflammation regulation. For example, expansions and modifications of interaction motifs in natural killer (NK) cell receptors such as NKG2A and CD94, unique insertions within the antigen-binding groove of major histocompatibility complex class I (MHC-I) molecules, and alterations in type I interferon (IFN-I) gene copy number and expression have been reported in different bat species [7–10]. Further, dampened NLRP3 inflammasome activation, attenuated STING-dependent interferon signaling, alterations in NF-κB regulators, and the emergence of resistant variants of RIG-I-like receptors (RLRs) suggest a shift toward immune tolerance and controlled inflammatory responses [11–13].

Among these immune adaptations, the NKG2A/CD94 receptor complex represents a particularly intriguing case of evolutionary modification in bats. This heterodimer plays a pivotal role in NK-cell regulation, with its signaling outcome determined by the structural features of its subunits. Specifically, inhibitory signaling is mediated through immunoreceptor tyrosine-based inhibition motifs (V/IxYxxL/V) in the cytoplasmic domain of NKG2A, whereas activating signaling depends on a positively charged residue (Arg or Lys) in the transmembrane domain [7, 14]. Evidence from certain bat species suggests that expansions of NKG2A-like genes, coupled with a bias toward inhibitory motifs and elevated basal expression levels, might drive bats toward an inhibitory immune state [7]. By contrast, its partner CD94 is generally considered a conserved single-copy gene across mammals, with two cysteines near position 60 that are essential for disulfide-mediated heterodimerization and signal transduction [15, 16]. Complementing these NK cell receptor adaptations, bat MHC-I molecules also show unique structural features that may enhance viral antigen presentation. Specifically, some bat species possess 3- or 5-amino acid insertions within the antigen-binding groove of MHC-I. These insertions are thought to facilitate a distinctive salt bridge between residues at positions 80 and 86 (Asp/Glu and Arg), potentially enhancing peptide-binding affinity and antigen presentation efficiency [8, 9, 17]. In parallel, the IFN-I system shows substantial interspecies variation in gene copy number and functional properties, supporting effective viral control while minimizing prolonged inflammatory signaling [7, 10, 18]. NLRP proteins are key components of inflammasome-driven inflammation, and dampened NLRP3 activation in bats has been linked to reduced transcriptional priming and structural changes in the leucine-rich repeat domain [11, 19]. Collectively, these immunogenomic findings have led to the “virus tolerance hypothesis,” which postulates that bats favor tolerance-based mechanisms to limit immune-mediated pathology during viral infection [11, 20–22]. However, most data supporting this hypothesis derive from a handful of bat species, leaving the extent and variability of these adaptations across the bat phylogeny unresolved.

To address this gap, we assembled a chromosome-level genome for the Chinese horseshoe bat (Rhinolophus sinicus), a key reservoir for SARS-like coronaviruses [3], and performed comparative analyses with 37 additional bat species spanning nine families and 23 genera. We focused on the evolutionary patterns of major antiviral immune gene families, including NKG2A, CD94, MHC-I, IFN-I, RLR, and NLRP. Our results reveal substantial diversity in the evolution of these gene families among bats, challenging previous generalizations and highlighting complex evolutionary mechanisms underlying bats’ unique ability to coexist asymptomatically with pathogenic viruses.

Results

De novo assembly of the chromosome-level reference genome of the Chinese horseshoe bat (Rhinolophus sinicus) and an improved bat phylogeny

To investigate the prevalence of evolutionary features in antiviral immune-related gene families across different bat lineages, we collected a wide range of bat genomes from public databases. To safeguard the reliability of our analyses and reduce potential biases arising from incomplete assemblies, we assessed the completeness of each bat genome using the BUSCO tool suite [23]. We selected 37 bat genomic datasets with BUSCO completeness scores exceeding 85% for further investigation (Fig. 1A; Additional file 2: Table S1). These datasets included bat species from 9 families and 23 genera, covering major bat lineages.

Fig. 1.

Reconstruction of the phylogenomic tree across 38 bat species. A The quality of the genomic data of the 38 bat species. The yellow dashed line indicates 85% BUSCO completeness scores. B ASTRAL tree inferred from 1701 single-copy one-to-one orthologous gene trees. Each gene tree was estimated with IQ-TREE using the best-fit substitution model. The estimated divergence times among bat species are indicated. The black dots on the tree represent constraints used to calibrate the estimated divergence times. The red star indicates Rh. sinicus, which has chromosome-level genomic data generated in this study. The horse is used as the outgroup in this phylogenomic tree

Furthermore, we generated a chromosome-level reference genome for the Chinese horseshoe bat (Rhinolophus sinicus) using 175.9 Gb of Nanopore long reads (Additional file 2: Table S2) and 220.7 Gb of chromatin conformation capture (Hi-C) interaction data (Additional file 2: Table S3). The draft assembly from Nanopore reads exhibited a contig N50 of 42.9 Mb (Additional file 2: Table S4). Based on the Hi-C data, approximately 41.1% of the assembled contigs, representing 99.6% of the total contig length, were anchored into 18 chromosomes (2n = 36; Additional file 1: Fig. S1; Additional file 2: Table S5), consistent with the known karyotype of Rh. sinicus [24]. The final genome assembly spanned ~ 2.04 Gb, with a scaffold N50 of 166.6 Mb (Additional file 2: Table S4). Genome quality assessment using Merqury [25] and BUSCO [23] yielded a consensus quality value (QV) of 41.4 and a completeness score of ~ 94%, respectively (Additional file 2: Table S1). These metrics indicate a high level of contiguity compared to available high-quality bat genome assemblies [22]. By integrating various types of evidence, including 262.6 Gb transcriptomic data from multiple tissues (Additional file 2: Table S6), we identified 19,587 protein-coding genes in the Rh. sinicus genome. The resulting annotation achieved a BUSCO completeness score of 95.6% (Additional file 2: Table S1), further validating the accuracy and quality of the de novo genome assembly for the Chinese horseshoe bat and providing a solid foundation for subsequent genomic analyses and investigations.

Establishing a clear phylogenetic framework is crucial for conducting comparative analyses of immunogenomic features in bats. Using 1701 high-quality single-copy one-to-one orthologs identified by OrthoFinder [26], we inferred a phylogeny for the 39 species, including 38 bats and a horse as the outgroup, using a coalescent-based approach [27]. The phylogenetic tree showed strong concordance with the currently recognized bat species tree at the family level (Fig. 1B) [28], while clarifying relationships among species within the same genus. Although the phylogenies inferred with coalescent and concatenation-based [29] methods were largely congruent, several key topological conflicts were observed within Pteropodidae. The coalescent-based tree placed Eidolon dupreanum as sister to Rousettus and Macroglossus sobrinus as sister to Pteropus, while the concatenation-based tree grouped Ei. dupreanum and Ma. sobrinus together, a clade with low bootstrap support (53%) (Additional file 1: Fig. S2). Notably, our analysis clarified phylogenetic positions among species in Pteropus, the most species-rich genus within the Pteropodidae family. Our coalescent-based phylogenomic tree places Pt. rufus as more closely related to Pt. giganteus, with Pt. vampyrus sister to this clade and Pt. pselaphon positioned as their outgroup (Fig. 1B). This contrasts with our concatenation-based tree, which supports a closer relationship between Pt. rufus and Pt. pselaphon, as well as between Pt. giganteus and Pt. vampyrus (Additional file 1: Fig. S2).

Accordingly, we estimated the evolutionary timelines among the 39 species. The resulting time-calibrated species tree (Fig. 1B) is highly consistent with previous findings based on fossil records [30, 31]. Our phylogenetic tree represents a densely sampled chiropteran phylogeny at the family level based on genomic data. It provides a reliable foundation for exploring the evolution of antiviral immune-related gene families in bat species.

NKG2A-like/CD94 heterodimers display substantial combinatorial diversity among bat species

To explore the generalizability of these evolutionary signatures across a broader range of bat species, we examined the functional gene contents of the NKG2A family in 38 bat species. By focusing on protein-coding genes with intact open reading frames, we aimed to better capture their biological relevance and significance. Our results showed significant variation in the number of intact NKG2A-like genes across different bat species (Fig. 2A; Additional file 2: Table S7). Notably, we found that Phyllostomus discolor and Ph. hastatus had evolved 17 and 16 intact NKG2A-like genes, respectively, suggesting that the phyllostomus bats may possess the highest number of intact NKG2A-like genes among bats. This observation is consistent with a significant expansion detected in their ancestral branch by CAFE analysis (Fig. 2A). By contrast, species within the Vespertilionidae family exhibited the fewest intact NKG2A-like genes, ranging from 1 to 3 (Fig. 2A; Additional file 2: Table S7). Furthermore, significant expansions were detected at multiple ancestral nodes and in several extant species of pteropodid bats, which aligns with the distribution of intact NKG2A-like genes on our bat phylogeny (Fig. 2B). Our phylogenetic analysis of these NKG2A-like protein sequences confirmed these observations, showing considerable diversification among bat lineages (Fig. 2B).

Fig. 2.

Evolutionary diversity and expression pattern of intact NKG2A-like genes in bats. A The numbers of NKG2A-like genes with only an activating motif, with only an inhibitory motif, and with both inhibitory and activating motifs are shown for different bat species. Nodes or lineages with significant expansion or contraction are marked with red or blue circles, respectively. Multiple sequence alignments display variations among bats and outgroups. The functional motifs of NKG2A-like genes are indicated by the dashed box. B The evolutionary relationship of NKG2A-like genes in bats and outgroups is inferred using the maximum likelihood analysis. Dashed lines represent the NKG2A-like genes identified in outgroups. The red, blue, and green terminal lineages denote NKG2A-like genes with only an activating motif, only an inhibitory motif, and both inhibitory and activating motifs, respectively. NKG2A-like genes from different bat families are color-coded and labeled. The numbers on the major branches correspond to bootstrap support values. Red and blue ancestral lineages represent putative species-specific and ancestral expansions, respectively. C The expression pattern of NKG2A-like genes is analyzed using transcriptomic data from 8 tissues across 8 bat species. The rows are organized based on the average expression of each NKG2A-like gene across all tissues for a specific bat species. Normalization of the NKG2A-like gene TPM (transcripts per million) values is performed by dividing them by GAPDH TPM values in the corresponding tissues

Through sequence comparison, we found that NKG2A-like genes in non-bat mammals mostly harbor inhibitory motifs. However, ~ 30% of the bat species examined lacked NKG2A-like genes that feature both activating residues and inhibitory motifs (Fig. 2A). These findings differ from earlier observations in a limited number of bat species [7], indicating substantial evolutionary diversity in NKG2A-like genes across different bat species.

A previous study suggested that higher expression of NKG2A-like genes with solely inhibitory motifs indicates an antiviral immune-inhibitory state in bats [7]. To assess the generality of this finding, we examined the expression patterns of NKG2A-like genes across various bat species. This analysis included our newly generated transcriptomic data from multiple tissues of the Chinese horseshoe bat (Additional file 2: Table S6). Combining this data with published transcriptomic data from 7 other bat species (Additional file 2: Table S8), we found that the dominance of expression among NKG2A-like genes with only inhibitory signaling motifs was not consistently observed across different bat species (Fig. 2C). For example, in Artibeus jamaicensis, a potential reservoir of multiple viral families [32], the two NKG2A-like genes (NKG2A-2 and NKG2A-5) featuring only inhibitory signaling motifs showed the lowest expression levels among all NKG2A-like genes (Fig. 2C). Additionally, in Ph. hastatus, a known natural reservoir of the hantavirus [33], two NKG2A-like genes with activating signaling motifs (NKG2A-1 and NKG2A-10) exhibited higher expression levels than most NKG2A-like genes with inhibitory signaling motifs (Fig. 2C). These findings significantly diverge from previous reports based on a limited number of bat species [7] and highlight the variability in expression patterns of NKG2A-like genes with different signaling motifs among bat species.

By analyzing intact CD94 protein-coding sequences across a wider range of bat species, we observed a high level of diversity in CD94 gene content among bats (Fig. 3A; Additional file 2: Table S7). Consistent with prior findings in the Egyptian fruit bat (Ro. aegyptiacus) and the flying fox (Pt. vampyrus), certain bat species such as Hipposideros pendleburyi (family Hipposideridae) and Ma. sobrinus (family Pteropodidae) had evolved multiple copies of CD94 (Fig. 3A; Additional file 2: Table S7). However, phylogenetic analysis revealed a diverse duplication of CD94 among bat species, indicating species-specific and ancestral duplications that occurred before speciation (Fig. 3B). More importantly, we identified only a single copy of CD94 in several bat species, consistent with observations in non-bat mammals such as Pt. alecto and Ei. dupreanum (family Pteropodidae), as well as most yinpterochiropteran bats (Fig. 3A). Many bat species exclusively possessed CD94 with conserved cysteines (Fig. 3A), aligning with findings in non-bat mammals [7]. Interestingly, CD94 proteins lacking conserved cysteines were exclusively found in pteropodid bats. The extensive expansions of CD94 proteins lacking conserved cysteines significantly contribute to the gene contents of pteropodid bats as compared to the conserved cysteine-containing CD94s (Fig. 3B). These findings suggest that pteropodid bats have evolved specific molecular adaptations in NK cell signaling.

Fig. 3.

Evolutionary diversity of intact CD94 genes in bats. A The numbers of intact CD94 genes with and without conserved cysteine residues are shown for each species. The dashed box in the multiple sequence alignment indicates the conserved cysteine residues in CD94 proteins of bats and outgroups. B The evolutionary relationship of CD94 proteins among bats and outgroups inferred using the maximum likelihood analysis. Dashed lines denote CD94 proteins identified in outgroups. CD94 proteins lacking conserved cysteine residues are indicated by red circles, while CD94 proteins possessing conserved cysteine residues are indicated by blue circles. The red and blue ancestral lineages represent putative species-specific and ancestral expansions, respectively. The numbers on the major branches correspond to bootstrap support values

The number of MHC-I genes and their genomic organization vary substantially among bat species

Our phylogenetic analysis revealed that most MHC-I sequences can be grouped at the family level (Additional file 1: Fig. S3), suggesting that these MHC-I genes evolved after the divergence of bat families. We examined the diversity of MHC-I gene repertoires across 38 bat species and observed significant interspecies variation in MHC-I genes. Consistent with expansion in the Egyptian fruit bat [7], we identified 19 intact MHC-I genes in the Leschenault’s rousette (Ro. leschenaultii) and 11 in the Madagascan rousette (Ro. madagascariensis) (Fig. 4A; Additional file 2: Table S7), suggesting that MHC-I gene expansion occurred on the ancestral branch of the rousette bats. Phylogenetic and CAFE analysis further confirmed this suggestion (Fig. 4A, B). In yangochiropteran bats, we observed another expansion of MHC-I genes on the ancestral branch of the phyllostomid bats (Fig. 4A, C). Particularly, Ph. discolor was found to have the largest number of intact MHC-I genes reported to date (n = 46; Fig. 4A; Additional file 2: Table S7). Notably, while expanded MHC-I gene repertoires were observed in several bat lineages compared to non-bat mammals, more diverse MHC-I repertoires were found among closely related bat species. For example, we identified 27 intact MHC-I genes in Myotis lucifugus, whereas only 2 were found in My. brandtii (Fig. 4A; Additional file 2: Table S7).

Fig. 4.

Evolutionary diversity of intact MHC-I genes in bats. A The numbers of intact MHC-I genes without insertion, with a 3-amino acid (aa) insertion, and with a 5-aa insertion, as well as those containing Asp/Glu^80/86 and Arg^80/86 residuals, are shown for each species. Nodes or lineages with significant expansion or contraction are marked with red or blue circles, respectively. A sequence logo plot shows a partial alignment of the MHC-I α1 domain in bats and outgroups. The 3- and 5-aa insertions are colored in red and blue, respectively. The dashed boxes indicate specific salt bridges formed between Asp/Glu^80/86 and Arg^80/86. The evolutionary relationships of MHC-I proteins among yinpterochiropteran bats (B) and yangochiropteran bats (C) are derived from Fig. S2 to show the putative MHC-I gene expansions. Red and blue lineages represent species-specific and ancestral expansions, respectively. The numbers on the major branches indicate the bootstrap support values

MHC-I genes in humans are located on the short arm of chromosome 6, specifically in the p21.3 band, and are organized into three genomic regions called the α, κ, and β duplication blocks [34]. Similarly, MHC-I genes were clustered into distinct blocks on a single chromosome homologous to human chromosome 6 in several phylogenetically distant mammalian species, including mouse, horse, and cow (Fig. 5). To gain a clearer understanding of the genomic organization and evolutionary patterns of MHC-I genes in bats, we mapped the identified MHC-I genes in 7 bat genomes that have been assembled at the chromosome level. Consistent with observations in other mammals, we found that two yinpterochiropteran bats, Rh. ferrumequinum and Rh. sinicus, had MHC-I genes located on a single chromosome homologous to human chromosome 6 (Fig. 5). However, this distribution pattern was not shared by all yinpterochiropteran bats; in Ro. aegyptiacus, MHC-I genes were found not only on the canonical MHC-I chromosome but also in two other genomic regions. These regions showed no collinearity with the genomic sequences surrounding the MHC-I genes in Rh. ferrumequinum and Rh. sinicus (Fig. 5). Furthermore, in four yangochiropteran bats, namely Ph. discolor, Pipistrellus kuhlii, My. myotis, and Mol. molossus, which spans three different families, MHC-I genes were located not only on the canonical MHC-I chromosome homologous to human chromosome 6 but also dispersed on another chromosome homologous to human chromosome 19 (Fig. 5). These findings indicate substantial lineage- and species-specific patterns in the chromosomal distribution of MHC-I genes among bat species.

Fig. 5.

Genomic map of the MHC-I region in mammals. The α, κ, and β duplication blocks are highlighted in red, blue, and green, respectively. Black boxes represent putatively functional MHC-I genes and gray boxes represent MHC-I pseudogenes. Non-MHC-I pseudogenes and non-coding genes were excluded from the illustration. The region enclosed by the red dashed line in Rh. sinicus and Rh. ferrumequinum indicates a putative inversion region

Our sequence comparisons indicated that while most bat species examined possessed these 3- and/or 5-aa insertions, they had varied evolutionary origins (Fig. 4A). Specifically, the 3-aa insertion was found exclusively in pteropodid bats, whereas the 5-aa insertion was observed in other bat species (Fig. 4A). Phylogenetic analysis revealed that most MHC-I sequences clustered based on the taxonomic classification of Yangochiroptera and Yinpterochiroptera (Additional file 1: Fig. S3). Interestingly, five canonical MHC-I protein sequences lacking the 3- or 5-aa insertions from Rh. ferrumequinum, Rh. sinicus, and Hi. pendleburyi of Yangochiroptera, as well as Miniopterus schreibersii and Molossus molossus of Yinpterochiroptera, clustered with MHC-I protein sequences from non-bat mammals (Additional file 1: Fig. S3).

Diversified type I interferon gene reservoirs among bat species

To explore the evolutionary patterns of IFN-I across a wide range of bat species, we conducted a comprehensive search for members of different IFN-I subfamilies, including IFN-α, IFN-β, IFN-ε, IFN-κ, IFN-δ, and IFN-ω, across various bat lineages. Our analysis reveals intriguing patterns in the evolution of type I interferon (IFN-I) genes across bat species. While Ro. aegyptiacus exhibited several IFN-α genes comparable to non-bat mammals, most bat species had undergone a reduction in the number of IFN-α genes (Fig. 6A, B; Additional file 2: Table S7). Surprisingly, genomic data from 12 vespertilionid bats revealed the complete absence of intact IFN-α genes (Additional file 2: Table S7). This finding aligns with an earlier report that identified only one IFN-α pseudogene in the My. lucifugus genome [35], suggesting that the loss of IFN-α likely occurred in the common ancestor of vespertilionid bats. Given the crucial role of IFN-α in mounting an inflammatory response to viral infections [36], the reduction or absence of IFN-α in many bat species, particularly vespertilionids, might reflect an evolutionary adaptation toward a diminished inflammatory response. Conversely, our results showed a remarkable expansion of IFN-ω in several bat species, with at least 10 copies found in Pt. giganteus, Ro. leschenaultia, Ro. aegyptiacus, My. lucifugus, and Pi. pipistrellus (Fig. 6A; Additional file 2: Table S7). Although the expansion in other bat species was less pronounced, most still had more IFN-ω genes, ranging from 2 to 7, than humans, having only one functional IFN-ω [37].

Fig. 6.

Evolutionary diversity of intact type I interferons (IFN-I) in bats. A The numbers of intact genes for six IFN-I subfamilies are shown. Nodes or lineages with significant expansion or contraction are marked with red or blue symbols, respectively, where circles, squares, and triangles denote the IFN-α, IFN-ω, and IFN-δ subfamilies. B The evolutionary relationship of intact genes for the six IFN-I subfamilies among bats and outgroups is inferred using the maximum likelihood analysis. Dashed lines represent IFN-Is from outgroups

Variations in the size of immune-related gene groups among bat species

Systematic evolutionary analyses of genomic data have revealed substantial interspecies variation in key antiviral immune-related gene families among bat species. To confirm interspecies variation in the numbers of immune-related genes among bats, we leveraged an independent dataset from the KEGG database to examine evolutionary variations in immune-related genes across bats and non-bat mammals. We totally retrieved 116,335 immune-related genes (Additional file 2: Table S9), encompassing 101 mammalian species from orders such as Chiroptera, Rodentia, Primates, Cetacea, Carnivora, and Artiodactyla. These genes were clustered into 1201 orthogroups (Additional file 2: Table S10) using OrthoFinder [26]. Our analyses revealed that Chiroptera displayed significantly smaller ROGUE values than other mammalian orders, except for Rodentia (P < 0.001, Mann–Whitney U tests; Fig. 7A). This suggests a generally higher degree of interspecies variation in the size of immune-related orthogroups within Chiroptera. To further validate this result, we applied the same correction method to estimate interspecies Euclidean distances for immune-related orthogroup sizes in the principal component dimensions for each mammalian order. Larger Euclidean distances indicate greater variation. Similarly, Chiroptera exhibited significantly larger Euclidean distances than the other four mammalian orders (P < 0.001, Mann–Whitney U tests; Fig. 7B). This reinforces the evidence for greater interspecies variation in the size of immune-related orthogroups within Chiroptera.

Fig. 7.

Evolutionary variations in the sizes of immune-related gene orthogroups among different mammalian orders. The ROGUE value (A) and the Euclidean distance (B) are used to quantify the size variation of immune-related gene orthogroups across different mammalian orders. Each point on the graph represents a bootstrap sampling. The center line indicates the median ROGUE value or median Euclidean distance. The lower and upper hinges represent the 25th and 75th percentiles, respectively. Statistical significance between the compared groups was assessed using the Mann–Whitney U test. ***, P < 0.001; n.s., not significant

To rule out the possibility that differences in genome quality account for variation in immune system gene numbers among species (Fig. 7), we conducted the same analysis across four non-immune systems: the circulatory, nervous, endocrine, and excretory systems. These systems share the same KEGG hierarchy as the immune system. If genome-quality differences were causing variation in immune system gene numbers among species, similar variation would be expected in non-immune systems. However, Chiroptera did not generally show lower ROGUE values or higher Euclidean distances in variation of gene numbers among species in these non-immune systems compared to other mammalian orders (Additional file 1: Fig. S4). These findings suggest that the non-immune systems did not exhibit the significant gene-number variation observed in bats’ immune system. Therefore, it is unlikely that the observed significant variation in immune system gene numbers among species is due to differences in genome quality across bats or other mammalian groups.

Viral diversity drives the evolution of bat antiviral gene families

The substantial variations observed in antiviral immune-related gene families and orthogroups among different bat species are likely influenced by the virome diversity of these species [3, 38]. In general, virus diversity appears to be higher in Yinpterochiroptera species than in Yangochiroptera species [38, 39]. To confirm this hypothesis, we compiled 17,847 bat-virus associations from the ZOVER database [40], covering 287 yangochiropteran species and 135 yinpterochiropteran species (Additional file 2: Table S11). Our analysis showed significantly higher viral richness in Yinpterochiroptera than in Yangochiroptera (P = 0.0092, Mann–Whitney U test; Fig. 8A). To further explore the relationship between viral diversity and the evolutionary signatures of NKG2A, CD94, MHC-I, and IFN-I, we estimated the Shannon biodiversity index based on viral richness for the bat species examined in our study (Additional file 2: Table S12). After controlling for phylogenetic independent contrasts, we found a general positive correlation between viral biodiversity and the presence of NKG2A, CD94, MHC-I, and IFN-I families in yinpterochiropteran bats compared to yangochiropteran bats (Fig. 8B).

Fig. 8.

Correlation between viral richness and the evolutionary signatures of NKG2A-like, CD94, MHC-I, and IFN-I gene families in bats. A A box plot shows the difference in viral richness between yangochiropteran and yinpterochiropteran bats. In the box plot, the lower edge and upper edge of a box represent the 25% quartile (q1) and 75% quartile (q3), respectively. The horizontal line inside a box indicates the median (md). The whiskers extend to the most extreme values inside inner fences, md ± 1.5(q3-q1). The Mann–Whitney U test was applied to assess differences in viral richness between the two bat suborders. B Scatter plots depict intact gene numbers for NKG2A-like, CD94, MHC-I, and IFN-I gene families plotted against viral diversity for yangochiropteran and yinpterochiropteran bats using phylogenetically independent contrasts (PIC). Each point and triangle in the scatter plots represents a yangochiropteran (red) and yinpterochiropteran (blue) bat, respectively. The lines represent linear regression fits for yangochiropteran (red) and yinpterochiropteran (blue) bats

Discussion

In this study, we created a chromosome-level reference genome for the Chinese horseshoe bat (Rh. sinicus), a known natural reservoir host for SARS-related coronaviruses [3]. We then conducted a comprehensive analysis of the evolutionary characteristics of key antiviral immune gene families, including NKG2A, CD94, MHC-I, and IFN-I, across 38 bat species. Our findings revealed substantial variation in these key antiviral immune gene families across bat species. Consistent with these genomic analyses, the significantly lower ROGUE values and higher Euclidean distances for immune-related orthogroups suggest that bats have evolved greater immune system diversity than other mammals. These findings suggest that the mechanisms underlying viral disease tolerance observed in some bat species may not apply uniformly across all bats. The diversity observed in bat immune gene contents suggests a more complex and varied picture of immune response than previously thought.

Several key phylogenetic incongruences were detected within Pteropodidae between our coalescent-based and concatenation-based analyses. The ASTRAL topology, consistent with recent phylogenomic studies [41, 42], resolved Ei. dupreanum as sister to Rousettus and Ma. sobrinus as sister to Pteropus. In contrast, the alternative grouping of Ei. dupreanum and Ma. sobrinus in the concatenation tree was weakly supported, suggesting that this relationship is likely unstable and may be driven by underlying genealogical discordance. Regarding the Pteropus genus, our coalescent-based tree places Pt. rufus as more closely related to Pt. giganteus, with Pt. vampyrus as the sister to this clade and Pt. pselaphon as the outgroup—a finding consistent with earlier phylogenetic analyses based on mitochondrial and nuclear sequence data [43, 44]. However, this conflicts with our concatenation-based tree, which supports an alternative topology grouping Pt. rufus with Pt. pselaphon and Pt. giganteus with Pt. vampyrus. Such phylogenetic discordance often stems from ILS and/or historical introgression, processes recently highlighted in the genomic evolution of Myotis bats, where they have been linked to the adaptive evolution of immune genes [45]. Our findings indicated that similar evolutionary forces may have shaped the evolution of the immune response during the early diversification of Pteropodidae. It is worth noting that our set of single-copy orthologs is smaller than those from studies with chromosome-level assemblies. This primarily stems from variability in initial genome quality and, critically, the lack of transcriptome data for over half of the included bat species, which severely limits the completeness of de novo annotations. Nevertheless, our ortholog set is larger than that reported in a comparable study of a similar taxonomic scale [42] and proved fully sufficient to reconstruct a highly supported species phylogeny. While the scale of our dataset limits genome-wide scans for ILS/introgression or the identification of underlying genes, it nevertheless motivates future investigation into the genomic signatures of ILS/introgression and their potential contribution to the evolution of viral tolerance.

To ensure that our findings were not influenced by variations in genome quality, we re-examined the evolutionary variation of the four key antiviral immune-related gene families using 17 chromosome-level, reference-quality bat genomes (Additional file 2: Table S1). The results continued to show significant lineage-specific variation in gene contents within these families (Additional file 1: Fig. S5), reinforcing the robustness of our initial analysis involving 38 bat species. To rule out the possibility that the observed variation in gene numbers within immune-related gene families across bat species was a technical artifact, we conducted a literature review of gene family studies. We selected five multigene families known for their stable sizes across mammalian species: the Feminization-1 (FEM1) [46], Heterotrimeric G proteins β (GNB) [47], CUG-BP, Elav-like (CELF) [48], Claudin (CLDN) [49], and Homeobox containing (HOX) [50] gene families. Using the same methodology applied to immune-related gene families, we identified members of each of these five gene families across the 38 bat genomes analyzed. We found that the gene numbers of these families were evolutionarily conserved among the 38 bat species (Additional file 1: Fig. S6). This result further suggests that the considerable variations observed in the sizes of immune-related gene families are genuine and not an artifact of genomic quality issues.

Variations in the size of gene families, especially those related to immunity, can have important biological implications. Changes in the number of genes within these families can affect an organism’s capacity to respond to pathogens, with larger families potentially offering a wider range of molecules for immune recognition and response. By studying interspecies variations in the size of key antiviral gene families among bats, we can assess the evolutionary diversity and functional divergence of the immune system across bat species. For example, the presence of highly diverse gene contents in both NKG2A-like and CD94 components among bat species indicates substantial variations in the combinatorial diversity of heterodimeric NKG2A-like/CD94 receptors. Certain bat species, such as Pt. giganteus and Ro. aegyptiacus, have evolved expansions of NKG2A-like and CD94 genes with diverse functional motifs (Figs. 2A and 3A; Additional file 2: Table S7). This diversification suggests that these species may have developed alternative antiviral strategies by varying the combinations of NKG2A-like and CD94, thereby enhancing their ability to bind a wide range of viral ligands. Conversely, other bat species, like My. myotis, exhibit more limited combinatorial diversity in their NKG2A/CD94 heterodimeric receptors. This reduced diversity may indicate a potentially narrower but more specialized suite of antiviral strategies, possibly focusing on a specific set of viral threats or compensating through other immune mechanisms. Overall, these findings suggest that the antiviral mechanisms mediated by NKG2A/CD94 receptors in NK cells among bat species are more complex than previously understood. In addition, Pavlovich et al. reported a notable expansion of IFN-ω genes in the Egyptian rousette, a host of Marburg virus, with 22 members compared to just one in humans. These IFN-ω genes have evolved to exhibit antiviral activity against viral infections, even at lower expression levels than those of their counterparts in other mammals [7]. Our study also identified similar species-specific expansions of the IFN-ω genes in several bat species, including Rhinolophus sinicus and Rhinolophus ferrumequinum (Fig. 6). A recent study on the IFN-I antiviral system in Rhinolophus bats highlights distinct IFN-ω responses in these bats compared to other mammals [51]. Notably, unlike the Egyptian rousette, Rhinolophus bats usually express their IFN-I genes constitutively, demonstrating exceptional antiviral activity [51]. Furthermore, our results revealed expansions of IFN-ω genes in other bat species as well (Fig. 6). IFN-ω usually induces weaker inflammatory responses compared to other IFN-I genes, as observed in cells from phylogenetically distant bat species Eptesicus fuscus and Ro. aegyptiacus [7, 52]. This suggests that the overall expansion of IFN-ω in bats might be an evolutionary strategy to mitigate excessive antiviral inflammation while still conferring effective immune defense. Yet, the significant variation in the IFNI-ω repertoire indicates lineage-specific differences in IFN-ω-mediated inflammatory responses across bat species. While these findings underscore the potential neo-functionalization of IFN-ω genes in bats, further functional studies are needed to explore the specific innovations of these expanded IFN-ω genes. Apart from IFN-α, IFN-ω, and IFN-δ, other subfamilies, including IFN-β, IFN-ε, and IFN-κ, appeared evolutionarily conserved across bats and other mammals, suggesting their fundamental role in mammalian inflammatory signaling (Additional file 2: Table S7). Notably, we found that the ancestral lineage of rousette bats underwent concurrent expansions in the IFN-α, IFN-ω, and IFN-δ gene families, suggesting a major innovation in the regulation of their inflammatory response.

Our investigation revealed that the bat ancestor experienced a significant expansion of MHC-I genes compared to other mammals (Fig. 4A). We propose that the expansion of this key gene family constituted a crucial genetic innovation, underpinning the bat’s remarkable capacity to adapt to diverse pathogens and to occupy a vast range of ecological niches. We also revealed considerable heterogeneity in MHC-I copy number among the 38 bat species studied—a variation evident even between closely related species such as My. lucifugus and My. brandtii (Additional file 2: Table S7). It should be noted, however, that several genomes in this study, including those of the above two Myotis bats, were assembled from short-read sequencing data. Such assemblies are prone to inaccuracies in structurally complex regions, such as the MHC-I loci, leading to redundant contigs or assembly gaps. These technical artifacts can lead to over- or underestimation of gene counts—a challenge previously documented not only for MHC-I [53] but also for other multi-copy gene families such as olfactory receptors [54]. Therefore, the reliability of the observed copy number variation may be compromised by the heterogeneity in genome assembly quality. To mitigate this concern, we repeated the analysis using only chromosome-level bat genome assemblies. Consistently, the results from these high-quality genomes confirmed our initial finding that MHC-I copy number is highly heterogeneous across bats. Notably, Ph. discolor was found to harbor the highest number of MHC-I copies among all mammalian species examined.

We further examined the distribution of MHC-I genes across various mammals. MHC-I genes are organized into three chromosomal regions in humans: the α, β, and κ blocks [34]. Consistent with previous research [55], we confirmed the presence of MHC-I genes in the α block for humans and mice (Fig. 5). However, we did not find MHC-I genes in the α block for other mammals examined in this study, such as dog, cow, pangolin, and all bats (Fig. 5). This absence suggests that MHC-I genes in the α block either originated in the common ancestor of Euarchontoglires or were lost in the common ancestor of the laurasiatherian mammals. While the distribution of MHC-I genes in the β block is relatively conserved evolutionarily, previous studies have reported the absence of MHC-I genes in the κ block in the genomic data of Ro. aegyptiacus [7], Desmodus rotundus [17], and Pt. alecto [8]. Our research on additional bat species largely aligns with these results, except for Rh. sinicus and Rh. ferrumequinum (Fig. 5). In these two species, we detected MHC-I genes in the κ block, marking the first instance of identifying such genes in this block within bats. This discovery raises interesting questions regarding the origin of MHC-I genes in the κ block of bats. The highly conserved collinearity of genes surrounding MHC-I genes within the κ block in Rhinolophus and other mammals suggests that MHC-I genes in the κ block likely originated from ancestral mammals. If this is the case, the loss of MHC-I genes in the κ block of bat genomes likely occurred independently twice: once in the common ancestor of yangochiropteran bats and once in the common ancestor of the Old World fruit bats. Although Rh. sinicus and Rh. ferrumequinum harbored a comparable number of MHC-I genes, they exhibited highly divergent genomic distributions. In Rh. ferrumequinum, MHC-I genes were predominantly located within canonical MHC-I genomic regions, particularly in the β block, whereas in Rh. sinicus, the majority of MHC-I genes were distributed outside the canonical MHC-I regions (Fig. 5). However, the immunological implications of this distinct distribution pattern remain unclear.

We identified a ~ 3.5 Mb genomic inversion between Rh. sinicus and Rh. ferrumequinum, located outside the canonical MHC-I region but in close proximity to an expanded cluster of MHC-I genes in Rh. sinicus (Fig. 5). Although the ancestral state of this inversion region remains unresolved, its presence in two closely related bat species raises intriguing questions regarding its evolutionary origins and functional implications. Inversions are known to suppress recombination between divergent haplotypes and have been implicated in fundamental evolutionary processes, including the expansion and contraction of gene families, alterations in gene expression through changes in genomic architecture, and the modulation of disease risk [56–58]. We therefore hypothesize that the inversion described here may have facilitated the divergent evolution of flanking genomic regions in these two species, potentially influencing MHC-I gene content and regulation, and thereby contributing to the emergence of distinct immune phenotypes. Future efforts to sequence and assemble this region from multiple individuals will be crucial to determine whether these inversion haplotypes represent fixed differences between the species. Interestingly, we identified five canonical MHC-I protein sequences from species of Yangochiroptera and Yinpterochiroptera, which clustered with MHC-I sequences from non-bat mammals (Additional file 1: Fig. S3). This suggests the existence of an ancient orthologous relationship among these sequences. Moreover, the broader presence of the 5-aa insertion in MHC-I proteins across bat species suggests that this insertion might contribute to forming a more robust salt-bridge chain between Asp/Glu and Arg, potentially allowing for tighter binding of MHC-I proteins to viral peptidomes compared to the 3-aa insertions. Further investigation is required to elucidate the structural and functional disparities between MHC-I proteins with these two distinct insertion types.

While the RLR gene family has experienced adaptive evolution in different mammalian lineages [59], indicative of its role in the ongoing evolutionary “arms race” against pathogens, our findings demonstrate a strict conservation in copy number—each of the three RLR members (DDX58, DHX58, and IFIH1) is present as a single copy in the vast majority of the bat and outgroup species examined (Additional file 1: Fig. S7; Additional file 2: Table S7). This finding of genomic conservation is mirrored at the functional level; specifically, the critical role of IFIH1 in antiviral defense is maintained in bats despite high sequence divergence [60]. Taken together, the evidence indicates that the core function of the RLR family is evolutionarily conserved, suggesting that the ability of bats to harbor zoonotic viruses is not primarily due to changes in their RLRs but rather a consequence of other features.

Our genomic analyses uncovered several reorganization events within the NLRP gene family in bats, characterized by lineage-specific diversification and loss. Notably, NLRP1 has undergone expansion in Myotis bats, whereas NLRP1, NLRP4, NLRP8, and NLRP13 have been lost or truncated in Pteropodidae (Additional file 1: Fig. S8; Additional file 2: Table S7), suggesting substantial functional divergence in this lineage. We further confirm the previously reported dampened bat NLRP3 function across a broader range of bat species, linking it to the mechanism of alternative transcriptional initiation [11]. Moreover, our data reveal more complex regulatory patterns: while Yangochiroptera uniformly possess a single transcription start site, Yinpterochiroptera exhibit divergent initiation patterns. Specifically, Pteropodidae and Hipposideridae each possess two transcription start sites, whereas Rhinolophidae, the closest relatives of Hipposideridae, possess only one (Additional file 1: Fig. S9). The sequence similarity surrounding the transcription start sites also differs considerably between Pteropodidae and Hipposideridae (Additional file 1: Fig. S9). This diversity in initiation mechanisms suggests that the regulatory mechanisms in NLRP3 did not originate from a common ancestral modification but instead emerged independently in these lineages. Such convergent evolution underscores a persistent selective pressure to dampen inflammatory responses across bat lineages. It is important to note, however, that these predicted promoter architectures and initiation patterns, particularly those observed in Hipposideridae, await validation by direct transcriptomic evidence. Collectively, these findings highlight both the diversity of NLRP gene evolution and the diverse evolutionary strategies bats employ to dampen NLRP3-mediated inflammation.

Our results demonstrate that the copy number of gene members in key antiviral families (NKG2A, CD94, MHC-I, IFN-I) is more closely related to viral diversity in yinpterochiropteran bats than in yangochiropteran bats. This suborder-specific relationship suggests distinct evolutionary trajectories in immune adaptation between the two major bat lineages. This observation aligns with and extends previous findings on distinct evolutionary and functional adaptations in several key immune genes between Yinpterochiroptera and Yangochiroptera, including those in gene expansion/contraction [21, 61, 62], evolutionary rates [62, 63], and expression profiles [11, 64]. Taken together, evidence from both the present and previous studies implies that Yinpterochiroptera and Yangochiroptera may have faced divergent pathogenic landscapes early in their evolutionary divergence, driving them toward distinct immune adaptation strategies. Additionally, the stronger correlation in Yinpterochiroptera implies that their antiviral defense strategies may have been more directly shaped by gene family expansion and contraction in response to a richer virome, a pattern that could be rooted in their early ecological divergence. In contrast, the lack of a similarly strong correlation in Yangochiroptera suggests their antiviral defenses rely more heavily on other mechanisms. Collectively, our findings underscore that bats have evolved a remarkable diversity of antiviral strategies, which are more varied and species/lineage-specific than previously recognized, reflecting a sophisticated evolutionary arms race finely tuned to distinct ecological and virological landscapes.

Conclusions

In summary, our study employed large-scale comparative genomic analyses across major bat lineages, revealing substantial lineage-specific features in key antiviral gene families. Moreover, our findings provide compelling genetic evidence for a previously underestimated degree of interspecies heterogeneity within the bat immune system. While our findings were supported by analyses of chromosome- and reference-level bat genomes, we recognized that variations in genome quality could affect the precise member counts of antiviral gene families across bat species. We believe that further investigations involving additional bat species with high-quality genomic data will reinforce our conclusions regarding the lineage-specific characteristics of key antiviral gene families in bats.

Methods

Genomic resources from public databases

We obtained publicly available genomic data for 48 bat species from the NCBI (www.ncbi.nlm.nih.gov) and DNA ZOO (www.dnazoo.org) databases, which encompassed nine distinct bat families. To reduce potential biases arising from incomplete assemblies while maximizing taxonomic representation, we assessed the completeness of all downloaded bat genomes using BUSCO v5.2.2 [23]. Based on these assessment results, we set a completeness threshold at 85% to encompass all nine bat families in our study. This value was chosen after a per-family assessment revealed that the genome of Noctilio leporinus [65] (the only representative of the family Noctilionidae) scored 87.5%, the lowest among the highest-quality assemblies from each family. Selecting an 85% threshold successfully retained all 9 families while yielding 37 genome assemblies for downstream analysis (Additional file 2: Table S1).

Genome sequencing and assembly for the Chinese horseshoe bat (Rhinolophus sinicus)

We generated the first chromosome-level genome assembly of the Chinese horseshoe bat (Rh. sinicus). Genomic DNA was extracted from the muscle tissue of an adult individual collected in Kunming, China, using the Qiagen DNeasy Blood & Tissue Kit (Qiagen, Germany). The purity of the extracted DNA was assessed using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, USA), and concentration was measured with a Qubit 3.0 Fluorometer (Invitrogen, USA). Long DNA fragments were isolated using the BluePippin system (Sage Science, USA). Sequencing libraries were prepared with the Ligation Sequencing Kit (Oxford Nanopore, UK; SQK-LSK109). Specifically, DNA ends were repaired using the NEBNext Ultra II End Repair Kit (New England Biolabs, USA), followed by ligation of the sequencing adapters supplied with the ligation sequencing kit. The libraries were loaded onto R9.4.1 flow cells and sequenced on a Nanopore PromethION platform (Oxford Nanopore Technologies, UK). Base calling of the raw fast5 data was performed with Albacore v1.25 (https://github.com/dvera/albacore).

Additionally, Hi-C data were generated to improve the assembly’s continuity. Chromatin was fixed in situ with formaldehyde and digested with the restriction enzyme DpnII. The resulting 5′ overhangs were filled in with biotinylated nucleotides, and blunt-end ligation was performed. After DNA extraction, the Hi-C library was constructed using the Nextera Mate Pair Sample Preparation Kit (Illumina, USA). The extracted DNA was sheared, end-repaired, A-tailed, and ligated to Illumina adapters. Biotinylated fragments were captured using streptavidin beads and amplified by PCR. The final Hi-C libraries were quantified on a Qubit 3.0 (Thermo Fisher Scientific, USA) and sequenced on an Illumina HiSeq 4000 platform (Illumina, USA) with a paired-end 150 bp pattern.

The genome was assembled by integrating Nanopore long reads and Hi-C data. Raw Nanopore reads were self-corrected using NextDenovo v1.0 [66]. The corrected reads were then assembled into a draft genome assembly using wtdbg v2.4 [67]. The assembly was polished with Racon v1.3.3 [68] using 4 iterations and all error-corrected Nanopore reads. Hi-C data were processed with HiC-Pro v2.11.1 [69] to remove low-quality interactions, and then aligned to the polished assembly to support contig scaffolding. Finally, LACHESIS [70] was employed to cluster, order, and orient contigs into chromosome groups using a bottom-up hierarchical clustering approach. Genome accuracy and completeness were evaluated using Merqury v1.3 [25].

Annotation of the Rh. sinicus genome

To predict protein-coding genes in the Rh. sinicus genome, we employed the Broad Institute eukaryotic annotation pipeline [71], which mainly uses various approaches, including RNA-sequencing-based, genome-aligning-based, homology-based, and ab initio gene predictions. For RNA-sequencing-based gene prediction, we collected transcriptomic data from 7 tissues: heart, liver, spleen, lung, kidney, brain, and lymph node (Additional file 2: Table S6). These transcriptomic reads were assembled using both de novo and genome-guided approaches with Trinity v2.4.0 [72]. The resulting assembled transcripts from the two methods were then integrated using PASA v2.4.1 [73] to generate a consensus gene set based on overlapping transcript alignments. For genome-aligning-based gene prediction, we used TOGA v1.1.1 [74] to project annotations of protein-coding genes from multiple reference genomes onto the Rh. sinicus genome. TOGA requires pairwise genome alignment chains between the reference and query genomes, coding transcript annotations for the reference genome, and a file linking gene and transcript isoforms. We applied TOGA to genome alignments to project gene annotations from human (Homo sapiens; GRCh38.p13), mouse (Mus musculus; GRCm38.p6), and greater horseshoe bat (Rh. ferrumequinum; mRhiFer1_v1.p) [75] onto our bat genome. For the homology-based gene prediction, we selected 9 high-quality genomes: human, mouse, dog (Canis familiaris) [76], flying fox (Pteropus vampyrus) [77], greater horseshoe bat, little brown bat (Myotis lucifugus) [78], greater mouse-eared bat (Myotis myotis) [79], Natal long-fingered bat (Miniopterus natalensis) [80], and Kuhl’s pipistrelle (Pipistrellus kuhlii) [81] (Additional file 2: Table S13). To predict gene structures, we used TblastN (BLAST v2.10.0) [82] to align all protein sequences from each species to the assembled genome. Then, we applied GeMoMa v1.7.1 [83] for gene structure prediction. For ab initio gene prediction, we used the BRAKER v3.0.1 pipeline [84] in native mode, which incorporates both RNA-seq and protein for training. This step utilized a compilation of transcriptomic data (Additional file 2: Table S6) and extrinsic protein orthologs from 9 mammals (Additional file 2: Table S13). To generate a comprehensive, nonredundant gene set, we integrated the four gene prediction datasets using EVM v1.1.1 [71], partitioning the assembled genome into 1 Mb segments with 100 Kb overlap. Gene models were determined for each segment by assigning weights to the evidence sources as follows: PASA: 10; TOGA: 8; GeMoMa: 5; BRAKER3: 1. Finally, we applied PASA once again to update the EVM consensus predictions by adding annotations for untranslated regions and gene models for alternatively spliced isoforms. To ensure a high-quality set of protein-coding genes, we implemented three filtering criteria on the predicted transcripts: (i) amino acid sequences had to be ≥ 50 residues in length; (ii) protein sequences needed to map to the NCBI nonredundant protein database with a cutoff of E < 10⁻⁵; and (iii) the ratio of the alignment length to the query length had to be ≥ 0.4. By applying these criteria, we identified 19,587 protein-coding genes, which closely align with findings from the chromosome-level genome of Rh. ferrumequinum [22], a species closely related to Rh. sinicus.

Annotation of unannotated public bat genomes

Among the 37 publicly available bat genomes studied, 21 were found to have no genomic annotation information (Additional file 2: Table S1). These 21 genomes were masked with RepeatMasker v4.0.5 [85] using a custom repetitive sequence database constructed with RepeatModeler v1.0.4 [85] and the RepBase27.05 library [86]. Because most of these genomes lack transcriptomic data in public databases, we applied homology-based and ab initio annotation methods to predict protein-coding genes in each repeat-masked genome. For homology-based predictions, protein sequences from the 9 high-quality genomes that were used in Rh. sinicus genome annotation were aligned with each repeat-masked bat genome using TblastN [82]. Thereafter, the gene structures of the corresponding genomic region in each BLAST hit were deduced by using GeneWise v2.2.0 [87]. For the ab initio predictions, AUGUSTUS v3.3.2 [88], GlimmerHMM v3.0.4 [89], geneid v1.4.5 [90], and SNAP [91] were used to identify candidate protein-encoding genes in the masked genome with self-trained model parameters. All the homology-based and ab initio evidence were integrated into a consensus gene annotation using EVM v1.1.1 [71] by assigning weights to the evidence sources as follows: GeneWise: 5; AUGUSTUS: 1; GlimmerHMM: 1; geneid: 1; SNAP: 1. Lastly, we implemented the three filtering criteria that were used in Rh. sinicus genome annotation on the predicted protein-coding genes to produce the final total gene set. Our analyses encompassed genomic data from 38 bat species along with 7 well-characterized mammalian genomes (Homo sapiens, Mus musculus, Canis familiaris, Bos taurus [92], Equus caballus [93], Manis pentadactyla [94], and Sorex araneus [95]), which served as outgroups (Additional file 2: Table S1). BUSCO was used again to assess the completeness of the protein-coding annotation across all mammalian genomes, including 38 bats and 7 outgroup species. These datasets were used for subsequent comparative genomics and molecular evolutionary analysis.

Reconstruction of the phylogenomic tree of bats and divergence time estimation

To examine the evolutionary patterns of antiviral immune-related gene families in bats, we reconstructed the phylogenetic tree for the 38 bat species. We identified 1701 single-copy one-to-one orthologous protein-coding genes shared among these bats and the outgroup E. caballus using OrthoFinder v2.4.0 [26]. To account for potential confounding effects of ILS, we primarily inferred the species tree using ASTRAL v5.7.7 [27]. Specifically, the protein sequences of each orthologous gene were aligned with MUSCLE v3.8.31 [96], and these alignments were converted to protein-coding sequence alignments using PAL2NAL v14 [97]. The best-fit substitution model for each gene was determined with ModelFinder [98] in IQ-TREE2 [29], and a gene tree was estimated for each alignment. The resulting 1701 gene trees were then used as input for ASTRAL to compute the species tree. We also concatenated the individual gene alignments into a partitioned supermatrix and estimated a maximum-likelihood tree using IQ-TREE2, assigning model partitions for each gene. To assess the reliability of the tree topology, we used UFBoot2 [99] with 1000 bootstrap replicates.

To estimate species divergence times, we used the MCMCTree tool from the PAML v4.9 package [100]. The analysis used the phylogenetic tree inferred by ASTRAL [27] and the concatenated sequences of 1701 single-copy orthologous genes as input data. It was performed under the independent clock model with the HKY85 + G substitution model for 200,000 Markov Chain Monte Carlo (MCMC) generations, following a burn-in of 20,000 generations. Divergence times were calibrated using five fossil constraints within Chiroptera: a maximum of 55.0 million years ago (Ma) for the base of Rhinolophoidea [28]; a time range of 38.0–56.0 Ma for the Vespertilionidae-Molossidae split [101]; a 47.8–61.6 Ma range for the emergence of Yangochiroptera [31]; a 30.0–41.3 Ma range for the Mormoopidae-Phyllostomidae split [31]; and a 16.0–34.0 Ma range for the emergence of Phyllostomidae [30]. Additionally, we used the molecular divergence date of 76.0–85.0 Ma between Chiroptera and Perissodactyla, as estimated in TimeTree (http://timetree.org/).

Evolutionary analysis of key immune-related gene families

Automatic gene prediction and annotation were often inaccurate for multi-copy genes located in complex genomic regions, particularly MHC-I genes, which were frequently missing, fused, or misidentified [102]. Therefore, we manually re-annotated the key antiviral immune-related gene families—NKG2A, CD94, MHC-I, IFN-I, RLR, and NLRP—by using BLAST and GeneWise with genomic data from 38 bat species. We used the protein sequences of these gene families from six well-assembled and annotated bat genomes (Rh. ferrumequinum, Ro. aegyptiacus [103], Ph. discolor [104], My. myotis, Pi. kuhlii, and Mol. molossus [105]), as well as human and mouse, as query sequences to identify homologous genes in other bat genomes. To ensure the searches were both accurate and comprehensive, we applied the following criteria for selecting the functional members of the antiviral immune-related gene families: (i) a BLAST e-value cutoff of less than 10⁻⁵; (ii) surrounding genomic sequences of the identified hits were analyzed using the GeneWise to obtain complete open reading frames; and (iii) reverse BLAST searches were conducted to confirm the corresponding homologous genes. Upon manual annotation, we discovered potential mis-annotation issues even in well-annotated genomes, indicating that our approach can improve the completeness of genome-wide annotation. For example, a total of 6 NKG2A genes were missing from the original NCBI annotations of the six well-assembled and annotated bat species (data not shown). Additionally, we performed the same search protocol in the genomes of 7 non-bat mammalian species, including human, mouse, dog, cow, horse, shrew, and pangolin. This comprehensive approach allowed us to explore the evolutionary dynamics of these gene families across a broad range of mammals.

After identifying all gene members with intact open reading frames (intact genes) in antiviral immune-related gene families, we generated multiple sequence alignments of protein sequences for each family using MUSCLE v3.8.31 [96]. PartitionFinder2 v2.1.1 [106] was employed to determine the most suitable evolutionary model for each protein alignment (Additional file 2: Table S14). Based on the best-fit evolutionary model, we inferred the maximum likelihood phylogeny for each gene family using RAxML v8.2.11 [107]. The resulting phylogenies were visualized using iTOL [108]. Additionally, we used the WebLogo tool (https://weblogo.berkeley.edu/) to assess and visualize the functional domains of these proteins.

To minimize the potential impact of genome incompleteness on the inferred copy number and distribution of MHC-I genes, we further analyzed chromosome-level genomes to validate the variation patterns observed across all bat genomes. This analysis included seven chromosome-level bat genomes: Rh. sinicus, Rh. ferrumequinum, Ro. aegyptiacus, Ph. discolor, My. myotis, Pi. kuhlii, and Mol. molossus (Additional file 2: Table S1). The MHC-I regions of human, mouse, dog, cattle, horse, and pangolin were used for comparative analysis. MHC-I genes lacking part of the coding region—including the leader peptide, extracellular domains α1, α2, α3, transmembrane domain, or cytoplasmic tail—due to sequencing gaps or fragmentation were classified as partial. Those with a complete coding region but containing premature stop codons were considered pseudogenes. For all species except Rh. sinicus, the annotation of MHC-I genes and their surrounding genomic regions was retrieved from the annotation files associated with their respective genome assemblies (Additional file 2: Table S1). Finally, we updated the genome annotations of these species with the positional information of manually identified MHC-I genes. To avoid overestimating MHC-I gene counts due to redundant assembly sequences, we excluded sequences shorter than 5 Kb that contained MHC-I genes. For a clearer visualization of MHC-I genomic organization, partial MHC-I genes, tRNA, rRNA, and uncharacterized genes were excluded.

To investigate the expansion and contraction patterns of the NKG2A, CD94, MHC-I, IFN-I, RLR, and NLRP gene family across the bat phylogeny, we performed a CAFE5 [109] analysis using gene families clustered by OrthoFinder from the proteomes of 38 bat species and the horse. We manually corrected the gene copy number of the target gene family to reflect our manual annotation counts. The resulting gene families were further filtered to exclude those exhibiting extreme size variation (> 100) across the phylogeny. CAFE5 was then used to estimate the size of each family at each ancestral node and to compute family-wise P values, identifying families with significant non-random expansion or contraction, as well as branches across the species tree with significant changes in gene count.

Expression patterns of NKG2A-like genes

To examine the expression patterns of NKG2A-like genes in bats, we combined transcriptomic data from various tissues of Rh. sinicus (Additional file 2: Table S6) with data from 7 other bat species, available from the NCBI database (Additional file 2: Table S8). Initially, we filtered the transcriptomic data for quality and removed PCR duplicates using Fastp v0.23.2 [110]. The filtered transcriptomic data from different samples were then pseudoaligned to the protein-coding sequences annotated in the corresponding bat genome assemblies. To ensure the reliability of these annotated NKG2A-like genes, we also verified them using BLAST. Gene expression was quantified at the transcript level as transcripts per million (TPM) using kallisto v0.46.1 [111]. The TPM values for these genes were normalized by dividing the TPM of GAPDH in the corresponding tissue. To visualize gene expression patterns, we generated a heatmap of log-transformed, normalized TPM values using Seaborn v0.12.2 [112].

Identification of orthogroups related to the immune system in different mammalian orders

To further validate interspecies variations in the number of immune-related genes among bats, we utilized an independent dataset from the KEGG database (www.genome.jp/kegg/). KEGG provides detailed gene annotations and species information across various taxa, allowing us to compare size variations of immune-related genes between bats and other mammalian groups. We collected all gene IDs listed under the KEGG PATHWAY: 5.1 Immune system (ko: 04640, 04610, 04611, 04613, 04620, 04624, 04621, 04622, 04623, 04625, 04650, 04612, 04660, 04658, 04659, 04657, 04662, 04664, 04666, 04670, 04672, and 04062) for mammalian species across six orders. This included 18 species from Chiroptera, 18 from Rodentia, 22 from Primates, 6 from Cetacea, 24 from Carnivora, and 13 from Artiodactyla (Additional file 2: Table S9). After removing redundant gene IDs among these 101 mammalian species, we converted the gene IDs to refSeq protein IDs using the gene2refseq database (https://ftp.ncbi.nlm.nih.gov/gene/DATA/gene2refseq.gz). Subsequently, we retrieved protein sequences from the NCBI database using refSeq protein IDs, selecting the longest isoform for each gene for further analysis. We then utilized OrthoFinder to cluster these genes into orthogroups across these species (Additional file 2: Table S10).

To assess interspecies variation in immune-related orthogroups within each mammalian order, we employed the entropy-based model ROGUE [113]. The ROGUE value ranges from 0 to 1, with lower values indicating higher heterogeneity or complexity. To address potential bias from uneven sample sizes across mammalian orders, we randomly selected 10 species from each order and estimated the ROGUE value, repeating this process 1000 times for each order. To improve robustness, we also used average Euclidean distances along the principal component analysis (PCA) dimensions to assess interspecies variation in immune-related orthogroups within each mammalian order. We employed a bootstrap approach to generate 1000 replicates of clustering matrices for these orthogroups, with each replicate involving resampling 10 species from each order. PCA was then performed across the bootstrap replicates, and we calculated the average Euclidean distances in the first 10 PCA dimensions.

Correlation between virus richness and evolutionary signatures of antiviral immune-related gene families in bats

To explore why different bat species have evolved varying sizes of immune-related gene families, we examined the correlation between virus richness and the evolutionary signatures of antiviral immune-related gene families in bats using virus records from the ZOVER database [40]. This dataset included 17,847 bat-associated viruses across 287 species from the suborder Yangochiroptera and 135 species from Yinpterochiroptera (Additional file 2: Table S11). We estimated viral richness for each bat species, defining it as the number of unique viral species found in a host species [1]. We further assessed the correlation between evolutionary signatures of the key antiviral immune-related gene families (NKG2A, CD94, MHC-I, and IFN-I) and viral diversity in both yangochiropteran and yinpterochiropteran bats. In our study, 30 out of 38 examined bats had viral records, with 17 in Yangochiroptera and 13 in Yinpterochiroptera (Additional file 2: Table S12). To quantify virus diversity, we calculated the Shannon biodiversity index for each bat species using the vegan R package. To account for the potential influence of phylogenetic relationships on our correlation analyses, we conducted a phylogenetically independent contrasts (PIC) analysis using the ape package in R [114].

Abbreviations

BLAST

Basic Local Alignment Search Tool

BUSCO

Benchmarking Universal Single-Copy Orthologs

CD94

Killer cell lectin like receptor D1

CELF

CUG-BP, Elav-like

CLDN

Claudin

DDX58

DExD/H-box helicase 58

DHX58

DExH-box helicase 58

FEM1

Feminization-1

GAPDH

Glyceraldehyde-3-phosphate dehydrogenase

GeMoMa

Gene Model Mapper

GNB

Heterotrimeric G protein β

Hi-C

Chromatin conformation capture

HOX

Homeobox containing

IFIH1

Interferon induced with helicase C domain 1

IFN-I

Type I interferons

ILS

Incomplete lineage sorting

KEGG

Kyoto Encyclopedia of Genes and Genomes

LRR

Leucine-rich repeat

MAVS

Mitochondrial antiviral-signaling protein

MCMC

Markov Chain Monte Carlo

MHC-I

Major histocompatibility complex class I

NCBI

National Center for Biotechnology Information

NK

Natural killer

NKG2A

Natural killer group II member A

NLRP

NOD-like receptor pyrin domain containing protein

NLRP3

NLR family pyrin domain containing 3

PASA

Program to Assemble Spliced Alignments

PCA

Principal component analysis

PIC

Phylogenetically independent contrasts

QV

Consensus quality value

RLRs

RIG-I-like receptors

RNA-seq

RNA sequencing

ROGUE

Ratio of Global Unshifted Entropy

SRA

Sequence read archive

TOGA

Tool to infer Orthologs from Genome Alignments

TPM

Transcripts per million

ZOVER

Zoonotic and Vector-borne Viruses Database

Authors’ contributions

The project was supervised and designed by ZL and HHZ. GSL, YYR, SYZ, PC, and QYH conducted the research and analyzed the data. ZL and GSL wrote the paper with contributions from other authors. All authors read and approved the final manuscript.

Funding

This study was supported by grants from the National Key R&D Program of China (2023YFA1800500), the National Natural Science Foundation of China (32192422, 32192420, 32330014, 32525015, 32270444, 32470448, 32470436, and 32570499), the Yunnan Fundamental Research Projects (202102AA310055), and the Yunnan Revitalization Talent Support Program Top team (202505AT350003, 202405AS350022).

Data availability

All the data generated or analyzed in this study are available in the Article and its Supplementary Information. The genomic and transcriptomic data we generated have been deposited in the NCBI database under BioProject PRJNA1048078. These data can also be accessed in the Science Data Bank (10.57760/sciencedb.13888; 10.57760/sciencedb.14354). The annotation files for 21 previously unannotated bat genomes, generated by our annotation pipeline in this study, are deposited in the Science Data Bank (10.57760/sciencedb.27175).

Declarations

Ethics approval and consent to participate

All animal procedures were conducted in accordance with the ethical guidelines and approved by the Animal Research and Ethics Committee of the Kunming Institute of Zoology, Chinese Academy of Sciences (approval number: IACUC-PA-2023–03-059).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.