Chromosome-level genome assembly of the highly-polymorphic peacock blenny (Salaria pavo)

Abstract

The peacock blenny Salaria pavo is notorious for its extreme male sexual polymorphism, with large males defending nests and younger reproductive males mimicking the appearance and behavior of females to parasitically fertilize eggs. The lack of a reference genome has, to date, limited the understanding of the genetic basis of the species phenotypic plasticity. Here, we present the first reference genome assembly of the peacock blenny using PacBio HiFi long-reads and Hi-C sequencing data. The final assembly of the S. pavo genome spanned 735.90 Mbp, with a contig N50 of 3.69 Mbp and a scaffold N50 of 31.87 Mbp. A total of 98.77% of the assembly was anchored to 24 chromosomes. In total, 24,008 protein-coding genes were annotated, and 99.0% of BUSCO genes were fully represented. Comparative analyses with closely related species showed that 86.2% of these genes were assigned to orthogroups. This high-quality genome of S. pavo will be a valuable resource for future research on this species’ reproductive plasticity and evolutionary history.

Background & Summary

Phenotypic plasticity, which allows organisms to respond to their environment without altering their underlying genotype, is widespread in nature and is considered a potential adaptation for thriving in changing environments^1,2. Hence, it is hypothesized that phenotypic plasticity plays a key role in the evolution of phenotypic diversity among species². Remarkable examples of morphological, physiological, habitat, and behavioral diversity have been described in teleost fish, representing the richest evolutionary radiation of vertebrates, with more than 30,000 species described³. The African cichlids are a classical example of this phenomenon, where adaptive radiation allowed species into a variety of ecological niches over a short evolutionary time span^4,5. However, less focus has been given to the genomic mechanisms underlying plastic responses within species expressed either within (e.g. sex change in the bluehead wrasse Thalassoma bifasciatum⁶) or across (e.g. temperature effects on sex ratios in the zebrafish Danio rerio⁷) generations.

The peacock blenny Salaria pavo (Teleostei: Blenniidae; Fig. 1), a small intertidal fish found in the rocky shores of the Mediterranean and adjacent Atlantic regions⁸, is a remarkable example of reproductive plasticity. In this species, nest-holder males are larger than females and exhibit well-developed secondary sexual characters (SSCs), such as a head crest and a sex-pheromone-producing anal gland, which they use to attract females to their nests for spawning^9,10, providing sole parental care to eggs until hatching^11,12. The intensity of mating competition varies among populations and is primarily influenced by the abundance of nest sites, which impacts mate availability during the breeding season, leading to the appearance of polymorphic reproductive phenotypes. In populations inhabiting coastal lagoons, nest sites are scarce (e.g. Ria Formosa, southern Portugal) and a strong intrasexual competition for mates is present, leading to a sex role reversal in courtship behavior, with females becoming the courting sex^13–15. The scarceness of nests leads smaller and younger males to adopt an alternative male reproductive tactic (ART), behaving as female-mimics to approach nests defended by nest-holder males and sneak fertilizations¹⁶. During the breeding season, males’ territory is restricted to the nest, with females competing for access by displaying elaborate courtship behaviors, consisting of stereotypic movements and a transient nuptial coloration, more often than nest-holder males^13,14,17. Sneaker males switch to nest-holders after their first breeding season¹⁸, encompassing a transitional male morph which is reproductively inactive. Hence, in this species, the same male can express both male and female reproductive behavior during its lifetime. In contrast, rocky shore populations have abundant nest sites (e.g. Adriatic Sea, Gulf of Trieste), with nest-holder males establishing nests in rock crevices and aggressively defending a territory around the nest, where they display courtship behavior towards females, who usually assume a passive role in courtship responding with changes in coloration and few displays before entering the nest to spawn (i.e. ‘conventional’ sex roles)¹⁵. Interestingly, experiments under similar laboratory conditions indicate that the expression of courtship behavior in this species is plastic¹⁹, suggesting a genomic regulation of this behavior. This species thus exhibits high polymorphism in mating strategies, with plastic behavioral phenotypes in both sexes, and developmental plasticity in males (i.e. life-history traits).

Fig. 1.

Peacock blenny Salaria pavo nest-holder male, inside the nest, being courted by two females on the right and one sneaker male on the left.

In total, 402 species belonging to 59 genera have been identified within the Blenniidae family (Fishbase³). The peacock blenny is one of the two species currently recognized within the Salaria genus, alongside Salaria basilisca²⁰. Previously, the genus included several freshwater species until a recent taxonomic revision by Vecchioni et al.²¹, which reclassified these species under the newly established genus Salariopsis. The phylogeographic history of the genus Salaria is complex, with the marine ancestor of S. pavo thought to have diverged from the ancestor of the freshwater lineage in the Middle Miocene²². Notably, within the Salariinae subfamily, male ARTs have been described in at least three additional species besides S. pavo: Scartella cristata²³, Salaria fluviatilis²⁴, and Parablennius sanguinolentus parvicornis²⁵, though the expression of ARTs varies among species.

Although the ecology, behavior, and physiology of S. pavo are well characterized, with brain transcriptomic profiles linked to each male morphotype and females²⁶, a complete genome assembly has been lacking, limiting the understanding of the genomic mechanisms associated with phenotypic plasticity in this species. Furthermore, epigenetic mechanisms have been proposed as key drivers of plasticity, allowing for rapid genome modulation in response to environmental cues through changes in epigenetic marks in gene regulatory networks that drive whole-organism performance²⁷. Therefore, we constructed a high-quality chromosome-level genome for S. pavo using a combination of DNBSEQ short reads, HiFi long reads, and Hi-C data. We also annotated the genome for repetitive elements, protein-coding genes and non-coding RNAs (ncRNAs). This genome offers the possibility for further studies on the genomic mechanisms underlying the behavioral and developmental plasticity observed in this species, and contributes to the genome database of the combtooth blennies, allowing further research on male ARTs, and evolutionary and biogeographic analyses within this family.

Methods

Fish sampling

One nest-holder male peacock blenny, S. pavo, collected at Culatra Island (Ria Formosa Natural Park, 36°59′N, 7°51′W, Portugal; for a description of the sampling area, see¹³) was euthanized using an overdose of buffered MS-222 (tricaine methanesulfonate anesthetic; Sigma-Aldrich), the blood and longitudinal muscle sampled immediately and snap-frozen in liquid nitrogen following BGI tissue sampling guidelines. Additionally, seven tissues from the same individual were dissected out – gill together with branchial arch, liver, testis, muscle, kidney, heart, and whole brain together with pituitary – and snap-frozen for RNA analysis. All samples were stored at −80°C until extraction. Fish procedures were performed in accordance with accepted veterinary practice under a “Group-1” license issued by the Portuguese National Authority for Animal Health (DGAV), permit number 0421/000/000/2013.

Library construction and sequencing

Sample extraction, library preparation and sequencing were performed by BGI Tech Solutions (Hong Kong, China). For short-read sequencing, gDNA extracted from muscle tissue was randomly fragmented with an insert size of 350 bp and sequenced using the DNBSEQ-G400 platform to generate 150 bp paired-end reads. The raw short reads were filtered by SOAPnuke v2.1.7²⁸ to remove adapters and low-quality reads with the parameters: -n 0.001, -l 10, -q 0.5, --adaMR 0.25,--apolyX 50, and --aminReadLen 150. A total of 79.1 Gb of clean data resulting in 107.49-fold coverage of the S. pavo genome was obtained (Table 1).

Table 1.

Statistics of sequencing data generated by DNBSEQ and PacBio.

Libraries	Total length (bp)	Read count	Read length (bp)	Sequence coverage (X)
DNBSEQ reads	79,104,612,300	263,682,041	150	107.49
PacBio reads	49,319,533,384	2,494,162	19,789	67.02
Total	128,424,145,684	266,176,203	—	174.51

For HiFi (high-fidelity) sequencing, high-molecular-weight (HMW) gDNA obtained from the blood sample was sheared to approximately 15 Kb selected using Sage ELF before preparing the PacBio HiFi library. The genomic library was sequenced in CCS mode on the Pacific Biosciences Sequel II System using two cells. A total of 49.3 Gb of long clean reads were generated, with mean lengths of 19.5 Kb and 20.0 Kb, respectively, resulting in 67.02-fold coverage of the S. pavo genome (Table 1).

For Hi-C sequencing, cells obtained from muscle tissue were treated with the crosslinking agent formaldehyde to fix the DNA and its binding proteins. After cell lysis, the DNA was digested with the restriction enzyme DpnII and marked with a biotinylated residue²⁹. The blunt ends of the crosslinked DNA fragments were then ligated using DNA ligase to form circular molecules, which were subsequently purified and sheared. The target DNA was captured through a biotin-streptavidin-mediated pull-down. Following the construction of the Hi-C library, sequencing was performed on the DNBSEQ-G400 platform as 150 bp paired-end reads. Quality control and filtering of the sequencing data were conducted using SOAPnuke v2.1.7²⁸ with the parameters: -n 0.001, -l 20, -q 0.3,–adaMR 0.25,–polyX 50, and–minReadLen 150. After filtering, 96.08 Gb of clean data were obtained, with Q20 and Q30 scores of 98.25% and 93.83%, respectively (Table 2).

Table 2.

Statistics of Hi-C clean data.

Clean Reads	Clean Bases	Q20 (%)	Q30 (%)	GC (%)
320,268,665	96,080,599,500	98.25	93.83	41.20

Strand-specific mRNA libraries were prepared for each tissue and sequenced on the DNBSEQ-G400 platform as 150 bp paired-end reads. Raw data with adapter sequences or low-quality sequences were filtered with SOAPnuke v2.1.7²⁸ using the following parameters: -n 0.001, -l 20, -q 0.4, --adaMR 0.25, --polyX 50, and --minReadLen 150. A total of 50.0 Gb of clean data was generated – gill together with branchial arch (7.2 Gb), liver (7.2 Gb), testis (7.2 Gb), muscle (6.8 Gb), kidney (7.2 Gb), heart (7.2 Gb), and whole brain together with pituitary (7.2 Gb). These RNA-seq datasets will be assembled into a reference transcriptome and the transcripts used for structural and gene annotation of the assembled genome.

Genome survey and de novo genome assembly

K-mer analysis was performed on 38 Gb of DNBSEQ sequencing data using Jellyfish v2.2.7³⁰ with K = 17. The genome size calculated from the K-mer number/depth was approximately 738.04 Mbp, with a corrected genome size of 717.65 Mbp, a genome heterozygosity rate of 0.55%, and a repeat sequence proportion of 32.55%. Subsequently, a preliminary assembly was conducted using SOAPdenovo2 v2.42³¹ with K = 41, resulting in a contig N50 of 148 bp, a total length of 1,306,005,430 bp, and a GC content of 40.72%. The 49.0 Gb of clean HiFi reads generated by Pacbio were then used in Hifiasm v0.19.5³² with default parameters to obtain the de novo assembly of the S. pavo genome. The assembled genome had a total length of 735.90 Mbp, with a contig N50 of 3.69 Mbp and a scaffold N50 of 31.87 Mbp (Table 3). BUSCO v5.4.5^33,34 was used to evaluate the completeness of the genome assembly with the Metazoa_odb10 dataset (954 markers). The evaluation results showed that 99.0% of the sequences in the reference dataset had a complete ortholog in S. pavo genome, including 97.5% complete and single-copy genes and 1.5% complete and duplicate genes, 0.7% of the genes were reported as fragmented and 0.3% of the genes were completely missing, which indicates a high level of genome completeness (Fig. 2).

Table 3.

Genome assembly statistics.

Type	Contig (bp)	Scaffold (bp)
Total Number	709	61
Total Length	735,896,725	735,961,525
Average Length	1,037,936	12,064,943
Max Length	12,769,392	40,169,789
Min Length	13,387	18,985
N50 Length	3,694,337	31,867,360
N50 Number	56	11
N90 Length	524,579	22,483,029
N90 Number	234	22

Fig. 2.

Genome assembly assessment. (a) BUSCO evaluation results. (b) Sequence depth distribution plot. (c) GC content and depth distribution plot.

Hi-C scaffolding and chromosome anchoring

The Hi-C sequencing data was assembled using the ALLHiC v0.9.8 pipeline³⁵ with the following parameters: --minREs 50 --maxlinkdensity 3 --NonInformativeRatio 2. Interaction signals between contigs were calculated based on the Hi-C data, and contigs were grouped according to the strength of these signals. Genetic algorithms were then used for random optimization to sort and orient the contigs within each group, thereby achieving genome anchoring and clustering. Finally, the heatmap was manually corrected using Juicebox v2.20.00³⁶ to obtain a chromosome-level genome. The assembly consisted of 25 chromosome-level sequences and 37 unanchored sequences (Fig. 3). The genome-wide Hi-C analysis shows that the last two chromosomes are similar in length and relatively shorter than the other chromosomes and are independently distinguishable while exhibiting a strong interaction with each other. Since the sequenced individual is male, and based on the observed patterns of interaction, the last two chromosomes are inferred to be the X and Y chromosomes. Consequently, the karyotype is 2n = 48, indicating a haploid genome of 24 chromosomes in accordance with the published karyotype for this species^37,38. The sequences anchored to chromosomes had a total length of 726,921,114 bp, out of a total genome length of 735,961,425 bp, resulting in 98.77% of the assembled sequences being assigned to chromosomes (Table 4).

Fig. 3.

Genome-wide Hi-C heatmap of S. pavo.

Table 4.

Assembly statistics for Hi-C.

Class	Scaffold Number	Total Length (bp)
place	25	726,921,114
unplace	37	9,040,311
total	62	735,961,425
Coverage rate	98.77%

Repeat and ncRNA annotation

First, the de novo genome-specific repeat library was generated using RepeatModeler v2.0.3³⁹. Afterwards, RepeatMasker v4.1.2-p1⁴⁰ was employed to identify and annotate repetitive sequences in the genome by comparing them to both the de novo repeat database and the RepBase database⁴¹. The results showed that 26.87% of the genome consisted of repeat sequences, with long interspersed nuclear elements (LINEs), short interspersed nuclear elements (SINEs), long terminal repeats (LTRs), and DNA transposons accounting for 0.09%, 4.18%, 6.85%, and 14.06% of the genome, respectively (Table 5). Based on the structural characteristics of tRNA, tRNAscan-SE v1.4⁴² was used to identify tRNA sequences in the genome. Due to the high conservation of rRNA genes, the rRNA sequences of closely related species Blennius ocellaris⁴³, Lipophrys pholis⁴⁴, and Salarias fasciatus⁴⁵ were used as reference sequences to identify rRNA in the genome through NCBI-BLAST + v2.2.26⁴⁶ alignment. The miRNA and snRNA information in the genome was predicted and annotated using the covariance model of Rfam v14.1⁴⁷. Finally, a total of 5,148 tRNAs, 5,849 rRNAs, 2,678 miRNAs, and 536 snRNAs were annotated in the S. pavo genome (Table 6).

Table 5.

Statistics of repetitive sequences in the S. pavo genome.

Type	Number	Length (bp)	Percentage (%)
SINE	6,840	668,490	0.09
LINE	186,876	30,745,474	4.18
LTR	291,227	50,398,843	6.85
DNA	861,528	103,442,202	14.06
Unknown	39,800	4,073,104	0.55
Total	1,341,271	197,738,351	26.87

Table 6.

Statistics of non-coding RNA annotation in the S. pavo genome.

Type		Copy (w)	Average length (bp)	Total length (bp)	% of genome
miRNA	miRNA	2,678	139.814	374,423	0.051
tRNA	tRNA	5,148	74.666	384,379	0.052
rRNA	rRNA	5,849	179.937	1,052,449	0.143
	18S	211	808.152	170,520	0.023
	28S	1,734	266.874	462,760	0.063
	5.8S	96	156.000	14,976	0.002
	5S	3,808	106.143	404,193	0.055
snRNA	snRNA	536	149.159	79,949	0.011
	CD-box	172	122.023	20,988	0.003
	HACA-box	81	160.173	12,974	0.002
	splicing	226	154.186	34,846	0.005
	scaRNA	56	197.964	11,086	0.002
	Unknown	1	55.000	55	0.000

Protein-coding gene prediction and annotation

Based on the genome annotation information of closely related species S. fasciatus⁴⁵ (Sfas), Gouania willdenowi⁴⁸ (Gwil), Amphiprion ocellaris⁴⁹ (Aoce), Mugil cephalus⁵⁰ (Mcep), and the model species Oryzias latipes⁵¹ (Olat), Augustus v3.5⁵² and SNAP v2013.11.29⁵³ were trained for de novo gene prediction of S. pavo genome. Additionally, gene homology prediction was conducted by aligning S. pavo genome with the protein-coding sequences of the previously mentioned species using BLAST + ⁴⁶ and Genewise v2.4.1⁵⁴. Furthermore, the RNA-seq data was assembled using Trinity v2.8.5⁵⁵, and the transcripts used to predict the gene structures by aligning the sequences with S. pavo genome using BLAT⁵⁶. Finally, the gene sets predicted by these three methods were deduplicated using EVidenceModeler v1.1.1⁵⁷ and refined with PASA v2.5.3⁵⁸, resulting in the prediction of 24,008 genes. RNA-Seq reads were aligned to the genome using HISAT2 v2.2.1⁵⁹ with default parameters to obtain baseline gene expression support for the predicted genes. The average transcript length obtained from the three methods was 14,092.38 bp, while the average CDS, exon, and intron lengths were 1,641.78 bp, 173.35 bp, and 1,469.85 bp, respectively (Table 7). In comparison with S. fasciatus, G. willdenowi, A. ocellaris, M. cephalus, and O. latipes, the results showed that the number of genes, average gene length, average CDS length, average exon length, and average intron length in S. pavo are similar to those of these species, indicating the high-quality and accuracy of the S. pavo genome assembly (Fig. 4 & Table 8). The gene sets were then compared with the Swiss-Prot, NR, Pfam, KEGG, and InterPro protein databases using DIAMOND v0.8.22⁶⁰/BLAST⁴⁶ (E-value < 10-5), annotating the functions of 20,442, 22,110, 18,806, 19,800, and 22,926 genes, respectively (Fig. 5). After integrating and deduplicating the results, a total of 23,323 genes (97.2% of all genes) were functionally predicted. Orthology inference was performed between predicted peptides of the S. pavo genome (N = 24,008) and annotated peptides (including all alternative splice versions) in S. fasciatus⁴⁵ (N = 39,222), G. willdenowi⁴⁸ (N = 43,478), A. ocellaris⁴⁹ (N = 47,242), M. cephalus⁵⁰ (N = 44,626), and O. latipes⁵¹ (N = 44,766), using OrthoFinder v2.5.5⁶¹ with the multiple sequence alignment-based tree inference (MSA)⁶² option (Fig. 6). The results showed that 86.2% of the predicted genes (20,700/24,008) in S. pavo could be assigned to orthogroups (Fig. 6b). Among these, only a low percentage – 2.3% (556/20,700) – were present in species-specific orthogroups (Fig. 6c), while the majority of the (15,750/20,700) had orthologs in all species (Fig. 6d). Additionally, the ratio of shared orthologs and orthogroups supports the expected phylogeny (Fig. 6e).

Table 7.

Statistics of gene structure annotation of the S. pavo genome.

Type	Gene Set	Number	Average transcript length (bp)	Average CDS length (bp)	Average exons per gene (bp)	Average exon length (bp)	Average intron length (bp)
De novo	Augustus	36,282	8,342.07	1,239.89	6.63	187.10	1,262.19
	SNAP	49,111	20,417.64	1,196.63	7.91	151.23	2,780.65
Homolog	Aoce	20,658	13,651.11	1,770.44	9.91	178.71	1,333.88
	Olat	19,859	13,193.08	1,727.30	9.62	179.59	1,330.43
	Mcep	20,223	13,803.83	1,771.45	9.94	178.27	1,346.36
	Sfas	21,074	11,679.72	1,496.00	8.53	175.48	1,353.30
	Gwil	19,439	13,385.95	1,725.71	9.72	177.49	1,336.76
RNAseq	transcripts	46,357	21,938.85	3,670.60	12.27	299.12	1,620.80
	PASA	42,219	9,513.77	1,232.76	7.47	165.04	1,280.03
EVM	EVM	32,236	11,024.55	1,367.41	7.64	179.04	1,454.91
PASA	PASA	31,902	11,508.10	1,392.88	7.76	179.59	1,497.27
Final set	Final	24,008	14,092.38	1,641.78	9.47	173.35	1,469.85

Fig. 4.

Comparisons of the genomic elements of closely related species.

Table 8.

Statistical information on gene structures in closely related species.

Species	Number	Average gene length (bp)	Average CDS length (bp)	Average exons per gene	Average exon length (bp)	Average intron length (bp)	Genome size (Mb)
Salaria pavo	24,008	14,092.38	1,641.78	9.47	173.35	1,469.85	735.96
Salarias fasciatus	25,108	12,012.46	1,516.02	8.78	172.74	1,349.79	797.49
Gouania willdenowi	22,774	19,931.28	1,797.42	10.39	173.04	1,931.73	937.15
Amphiprion ocellaris	23,040	18,584.98	1,846.01	10.56	174.74	1,750.18	863.47
Mugil cephalus	23,268	13,897.58	1,826.60	10.49	174.10	1,271.73	634.85
Oryzias latipes	22,094	16,477.55	1,842.13	10.57	174.26	1,529.15	734.04

Fig. 5.

Gene prediction and functional annotation of the S. pavo genome. (a) Venn diagram of the gene sets predicted using a gene overlap greater than 50% between two or more of the following methods: de novo, genes supported by EVM-integrated de novo predictions; homolog, genes supported by EVM-integrated homologous predictions; RNA, genes supported by EVM-integrated RNA-seq data predictions. Gene predictions had a baseline gene expression, RPKM, greater than 1. (b) Venn diagram of functional annotation based on different protein databases: InterPro, Swiss-Prot, NR, and KEGG. The numbers represent the gene count.

Fig. 6.

Summary of orthology inference of predicted proteins between S. pavo and closely related species. (a) Evolutionary relationship among S. pavo, S. fasciatus, G. willdenowi, A. ocellaris, M. cephalus, and O. latipes. (b) Number of genes per species that could be placed in an orthogroup. (c) Number of orthogroups that are specific to each species. (d) Total number of peptides with orthologs in at least one other species. (e) Heatmap of the number of orthogroups containing each species pair (top right) and one-to-one orthologs between each species (bottom left).

Data Records

The assembly and all DNA and RNA sequencing data have been deposited at NCBI under BioProject accession number PRJNA1151695. This project contains the GenBank chromosome level assembly SPavo_v1.1 (GCA_043647535.1⁶³), the transcriptomic sequencing data (SRR30403554⁶⁴, SRR30403555⁶⁵, SRR30403556⁶⁶, SRR30403557⁶⁷, SRR30403558⁶⁸, SRR30403559⁶⁹, SRR30403560⁷⁰), DNBSEQ WGS sequencing data (SRR30415240⁷¹), the PacBio sequencing data (SRR30416695⁷²), and the Hi-C sequencing data (SRR30417930⁷³). The genome annotation files are available in Figshare⁷⁴.

Technical Validation

Genome assembly assessment

The completeness of the genome was evaluated using BUSCO, which revealed that 99.0% of the BUSCO genes were complete in the assembled genome (Fig. 2), indicating a high genome completeness. Short-read libraries were aligned to the assembled genome using BWA v0.7.8⁷⁵ to determine alignment rate and coverage depth. The results showed an alignment rate of 97.36% and coverage of 99.74%, demonstrating high accuracy of the genome assembly. Genome assembly quality was further assessed using Merqury v1.3⁷⁶, resulting in a QV (quality value) of 40.5061, indicating genome accuracy exceeding 99.99%. In summary, the evaluation results from these software tools collectively indicate a high level of consistency, completeness, and accuracy in the assembled genome.

Hi-C assembly validation

The genome anchoring rate was 98.77%, and the Hi-C heatmap displayed a well-ordered pattern of interaction contacts (Fig. 3). These findings collectively indicate that the Hi-C-assisted assembly process yielded high-quality outcomes.

Genome annotation

Gene prediction using EVidenceModeler, integrating de novo, homologous, and RNA-seq data, resulted in the prediction of 24,008 genes in the S. pavo genome (Fig. 5a), with an average transcript length of 14,092.38 bp (Table 7). Comparative analysis with closely related species showed that S. pavo has a similar number of genes and comparable average gene, CDS, exon, and intron lengths (Table 8). Additionally, 97.2% of the predicted genes could be functionally annotated (Fig. 5b). Orthology inference with closely related species showed that 86.2% of the predicted genes in S. pavo could be assigned to orthogroups (Fig. 6). Together, these results indicate a high-quality and accuracy of the S. pavo genome assembly.

Author contributions

D.G. conceived the research project. S.D.C. collected the samples and curated the data, C.J. and L.S. performed the analyses. S.D.C, C.J., L.S., L.Z., D.G. wrote and revised the manuscript.

Code availability

All software and pipelines used in this study were executed according to the manual and protocols of the published bioinformatic tools. The versions of the software have been given in the Methods.

Competing interests

The authors declare no competing interests.