Chromosome-level genome assembly and annotation of female Acrossocheilus kreyenbergii

Acrossocheilus kreyenbergii (Cypriniformes: Cyprinidae), commonly known as the thin-jawed barbel, is a small-stream headwater specialist endemic to upland regions of China (Chen, 2025). This species is diagnosed by several distinctive morphological traits, most notably a discontinuous lower jaw divided into two separate lateral lobes and a narrow post-labial groove that measures less than one-third of the oral width (Jian et al., 2025) (Figure 1A-1B). A. kreyenbergii is widely distributed across the Huaihe, Yangtze and Pearl River basins in China and it is the most widely distributed species of Acrossocheilus (Yuan, 2009, Zhou et al., 2016, Chen, 2025, Zhao et al., 2018).It exhibits a strong preference for lotic habitats with sandy, pebbly, gravelly, or rocky substrates. Trophically, it is predominantly a benthic algivore, with aquatic insects serving as a supplementary dietary component (Chen and Fu, 2024). Sexual maturity is typically attained at approximately two years of age and the reproductive season extends from late spring to early autumn, during which adults deposit adhesive eggs onto hard substrates such as exposed rock surfaces (Chen and Fu, 2024).Regarding genetic resources, the mitochondrial genome of A. kreyenbergii has been characterized as a 16,849 bp circular molecule containing 13 protein-coding genes, 22 transfer RNA (tRNA) genes, 2 ribosomal RNA (rRNA) genes, and a single non-coding control region; its gene arrangement conforms to the vertebrate consensus (Zhou et al., 2016). However, a high-quality chromosome-level nuclear genome assembly and its associated functional annotations remain unavailable for this species.Fortunately, significant progress has been made in recent years in the genomics of the genus Acrossocheilus, with chromosome-level genome assemblies successfully generated for A. wenchowensis, A. longipinnis, and A. fasciatus (E et al., 2026, Xue et al., 2026, Yuan et al., 2024). These studies indicate that genome size within the genus is relatively conserved (870-936 Mb), with all assemblies exhibiting high contiguity and successful anchoring to 25 chromosomes (Yuan et al., 2024, Xue et al., 2026, E et al., 2026). These findings underscore a highly conserved karyotypic architecture within Acrossocheilus, providing a robust methodological framework for the genome assembly of A. kreyenbergii.The absence of a reference genome for A. kreyenbergii represents a conspicuous gap, hindering comprehensive comparative genomic analyses within the genus and across East Asian cyprinids. This limitation is particularly evident in evolutionary systematics, where phylogenies based solely on mitochondrial markers have often failed to resolve the contentious relationships between Acrossocheilus and allied genera such as Onychostoma (Yang et al., 2018). In contrast, whole-genome analyses have demonstrated the ability to estimate divergence times with greater precision; for instance, the split between A. wenchowensis and A. fasciatus has been robustly calibrated to approximately 5.25 million years ago (Xue et al., 2026). Therefore, a highquality nuclear genome for A. kreyenbergii is essential for elucidating the evolutionary history, facilitating functional gene discovery, and characterizing the adaptive genetic architecture of this endemic lineage.Here, we report a chromosome-level genome assembly of a female A. kreyenbergii.This high-quality genome assembly serves as a fundamental resource for advancing studies on A. kreyenbergii in molecular ecology, evolutionary biology, and aquaculture genomics.Acrossocheilus kreyenbergii; PacBio sequencing; de novo assembly; genome annotation.The genome of a female A. kreyenbergii specimen collected from Wuyuan County, Shangrao City, Jiangxi Province, China, was sequenced and assembled de novo. A total of 66.43 Gb of Illumina short-read data were generated for genome survey analysis, providing approximately 85× coverage (Supplementary Table S1). We performed k-mer analyses across multiple k values using these reads. The k = 17 analysis based on these reads estimated a genome size of 777.3 Mb, with a heterozygosity rate of 0.92% and a repeat content of 46.08% (Supplementary Table S1; Supplementary Figure S1).Analyses at k = 21, 25, and 27 corrected genome size estimates of 763.79,767.22,and 775.81 Mb,respectively (Supplementary 2B1D; Supplementary Table S3S4-S4S5). The final assembly size (917.72 Mb) was larger than the k-mer-based estimates (763.79-777.3 Mb). This discrepancy is not attributable to k-mer size selection, as estimates obtained using four k-mer sizes (k = 17, 21, 25, 27) were consistently smaller than the assembly. Rather, it likely reflects the high repeat content (50.07%) of the genome, which leads short-read k-mer-based surveys to underestimate repetitive and heterochromatic regions. The assembled size falls within the reported range (870-936 Mb) for other chromosomelevel genomes of Acrossocheilus (Yuan et al., 2024, Xue et al., 2026, E et al., 2026).Repetitive element annotation, combining homology-based and de novo-based approaches, identified 459.53 Mb of repetitive sequences, representing 50.07% of the genome (Figure 3A; Supplementary Table S5S6). Tandem repeats comprised 42.37 Mb, representing 4.62% of the genome assembly (Supplementary Table S5S6). Gene prediction, integrating homology-based, ab initio, and transcriptome-assisted methods, identified a total of 26,056 protein-coding genes, of which 25,102 (96.34%) were functionally annotated against public databases (Swiss-Prot, NR, KEGG, and InterPro) (Figure 3B2A-B; Supplementary Figure S3S4,S6S7). A total of 42.94 Gb of clean RNA-seq data were generated from seven six tissues (muscle, brain, intestine, hepatopancreas, eye, kidney, and skin) to support genome annotation (Supplementary Table S1). Non-coding RNA annotations identified 31,19737,350 ncRNAs, comprising 15,84621,756 tRNAs, 10,1979,954 rRNAs, 2,574 miRNAs, and 2,823 snRNAs (Supplementary Table S7S8). These tRNA and rRNA counts correspond to nonredundant genomic loci matching zebrafish (Danio rerio) tRNA and rRNA references under stringent thresholds (identity ≥ 90%, query coverage ≥ 85%, alignment length ≥ 60 bp); overlapping hits were merged and only the best hit per locus was retained. We note that these counts cannot fully discriminate functional genes from pseudogenes, repetitive copies, or partial matches based solely on sequence similarity.Assembly completeness was assessed using BUSCO (v5.2.2) against the vertebrata database (vertebrata_odb10). Of the 3,640 orthologous groups, 3,615 (99.3%) were complete, 20 (0.5%) were fragmented, and 5 (0.2%) were missing (Supplementary Figure S2). Telomeric repeat analysis identified canonical motifs at 31 of 50 chromosome ends, with both ends supported for 12 chromosomesnearly all chromosomal ends (Figure 42C, Supplementary Table S9). Illumina short reads mapped at a rate of 99.47%, with 99.52% genome coverage. Base-level accuracy, assessed using Merqury, yielded a consensus quality value (QV) of 36.97. To further assess potential exogenous contamination, we performed a taxonomic assessment of the 190 unplaced scaffolds using BlobToolKit via BLASTN searches against the NCBI nt database. All aligned sequences were assigned to metazoan taxa, and no bacterial, fungal, or other non-metazoan sequences were detected. In parallel, GC content analysis in 10 kb windows revealed a uniform GC content of 37.58%, with no evidence of exogenous contamination (Supplementary Figure S4). These results indicate that the unplaced scaffolds are of host origin and that no additional decontamination steps were required.Synteny analysis with zebrafishzeabrfish (Danio rerio) confirmed the expected collinearity and chromosome-scale assembly accuracy (Figure 2C1E).A wild female A. kreyenbergii (total length 14.70 cm, body weight 71.40 g) was collected from Raohe River at Wuyuan County, Shangrao, Jiangxi Province, China (29°13′5.1″N, 117°40′50.3″E). Prior to tissue collection, the fish was euthanized using MS-222 (Sigma-Aldrich, #A5040). To maintain tissue integrity, all dissection work was performed on ice, with careful excision of multiple tissues including muscle, brain, intestine, hepatopancreas, eye, kidney, and skin. Immediately after dissection, each tissue sample was flash-frozen in liquid nitrogen and stored at -80 °C until subsequent DNA and RNA isolation.For muscle tissue, high-molecular-weight (HMW) genomic DNA was extracted using the standard phenol-chloroform protocol (Green and Sambrook, 2017). Three methods were employed to assess the quality of the extracted DNA: 1.0% agarose gel electrophoresis, 0.8% pulsed field gel electrophoresis (PFGE), and a quantification using a Qubit 4.0 fluorometer (Thermo Fisher Scientific, USA). To perform chromatin conformation capture (Hi-C) scaffolding, muscle tissue was collected from the same specimen. Hi-C library preparation followed standard protocols (Belton et al., 2012, Rao et al., 2014), including formaldehyde cross-linking, chromatin digestion with MboI, biotin labeling, proximity ligation, and DNA purification. Library quality was assessed using an Agilent 2100 Bioanalyzer and qPCR. Libraries that passed quality inspection were sequenced on the Illumina NovaSeq 6000 platform in 150-bp PE mode.Total RNA was extracted from each of the seven six collected tissues using TRIzol TM reagent (Thermo Fisher Scientific, USA), and residual DNA was removed using DNase I (NEB, USA). RNA quality was evaluated using a Nanodrop spectrophotometer and an Agilent 2100 Bioanalyzer. Only RNA samples meeting the following quality criteria utilized, taking advantage of its high customization and publication-ready visualization capabilities.Genomic repetitive elements were categorized into two major groups: interspersed repeats, which are dominated by transposable elements (TEs), and tandem repeats. To comprehensively annotate TE-associated sequences, a dual approach integrating homology-based alignment and ab initio prediction was adopted. For the homologybased detection component, RepeatMasker and RepeatProteinMask (Tarailo Graovac and Chen, 2009) (https://www.repeatmasker.org/) were employed to conduct searches against the RepBase database (Jurka et al., 2005). Meanwhile, de novo repeat identification was carried out using three specialized tools: LTR_FINDER (Xu and Wang, 2007), RepeatScout (Price et al., 2005) and RepeatModeler (Flynn et al., 2020).Only sequences exceeding 100 bp in length and with a gap ("N") content of less than 5% were selected to establish a non-redundant de novo TE library. A combined library, integrating results from both homology-based and de novo predictions, was then used for the final TE screening with RepeatMasker. Tandem repeats were identified using Tandem Repeats Finder (TRF) (Benson, 1999).Gene structure annotation was conducted using a combination of three approaches: homology-based prediction, ab initio prediction, and transcriptome-based prediction.For homology-based prediction, protein sequences of seven fish species-including D. rerio (GCF_049306965.1), A. fasciatus (generated by our group), Onychostoma macrolepis (GCA_012432095.1), Ctenopharyngodon idellaus (GCF_019924925.1), Megalobrama amblycephala (GCF_018812025.1), Puntigrus tetrazona (GCF_018831695.1), and Oryzias latipes (GCF_002234675.1) -were obtained from the NCBI RefSeq database (https://ftp.ncbi.nlm.nih.gov/genomes/refseq). These protein sequences were first aligned to the target genome using TBLASTN (v2.2.26) with an E-value threshold of ≤ 1×10⁻⁵ (Altschul et al., 1997) splice-aware alignment was then performed using GeneWise (v2.4.1) (Birney et al., 2004) to refine the alignment results. For ab initio prediction, a pipeline integrated fivetwo tools was used: AUGUSTUS (v3.2.35) (Stanke and Waack, 2003) GeneID (v1.4) (Parra et al., 2000), GENSCAN (v1.0) (Burge and Karlin, 1997), GlimmerHMM (v3.04) (Majoros et al., 2004), and SNAP (2013-11-29) (Korf, 2004).In transcriptome-assisted prediction, clean RNA-Seq reads were processed through two parallel approachesd: de novo assembly with Trinity (v2.1.1) (Grabherr et al., 2011) and genome-guided transcript reconstruction. For the latter, reads were aligned to the genome using HISAT2 (v2.0.4) (Kim et al., 2015), followed by transcript assembly with Cufflinks (v2.2.1) (Trapnell et al., 2009). Ultimately, the gene predictions derived from these distinct approaches were integrated using EvidenceModeler (EVM, v1.1.1) and further refined with the Program to Assemble Spliced Alignments (PASA) (Haas et al., 2008) to generate a high-quality, non-redundant gene set. After PASA refinement, we applied the following filtering criteria to retrieve the final gene set: (i) gene models exhibiting ≥20% overlap with transposable elements (TEs) were discarded; (ii) genes models containing premature stop codons or ambiguous N bases were discarded; (iii) gene models supported solely by de novo prediction evidence with RPKM < 1 across all tissues were discarded; and (iv) single-exon gene models lacking both homology and transcriptome support with RPKM < 5 were discarded. Following filtering, each gene was represented by a single transcript isoform.Functional annotation of the predicted genes was performed using multiple complementary approaches. Protein sequences were aligned to the Swiss-Prot database (Bairoch, 2000); http://www.uniprot.org/) using BLASTP with an E-value cutoff of ≤ 1×10⁻⁵ for functional assignment. InterProScan (v5.31) (Mulder and Apweiler, 2007); https://www.ebi.ac.uk/interpro/) was used to identify protein domains and functional motifs, with corresponding Gene Ontology (GO) terms assigned accordingly.Additionally, DIAMOND (v0.8.22) was used to search homology against the NR database (ftp://ftp.ncbi.nih.gov/blast/db) with an E-value threshold of < 1×10⁻⁵.Moreover, KEGG pathway mapping was performed to annotate the biological pathways associated with the predicted genes (Ogata et al., 1999).Non-coding RNAs (ncRNAs) were annotated using dedicated tools for each ncRNA class. Transfer RNAs (tRNAs) were identified using tRNAscan-SE (Lowe and Eddy, 1997). , while rRibosomal RNAs (rRNAs) were detected via BLASTN (E-value ≤ 1×10⁻⁵) against rRNA reference sequences from Homo sapienszebrafish (Danio rerio).For both tRNA and rRNA predictions, sequences were aligned to the corresponding zebrafish references, and hits were retained only when satisfying the following thresholds: identity ≥ 90%, query coverage ≥ 85%, and alignment length ≥ 60 bp.Overlapping hits (≥ 80% reciprocal overlap) were merged into a single locus, and only the best-scoring hit was kept to eliminate redundant and repetitive matches. The reported tRNA and rRNA counts therefore corresponded to non-redundant genomic regions matching tRNA and rRNA references, rather than validated full-length functional genes; partial matches and potential pseudogenes cannot be fully excluded based on sequence similarity solely. For microRNAs (miRNAs) and small nuclear RNAs (snRNAs), INFERNAL was used to search the Rfam database (Ashburner et al., 2000) for their identification.In this study, a set of complementary approaches was used to evaluate the quality and completeness of the genome assembly. Assembly completeness was assessed using BUSCO (v5.2.2) against the Vertebrata lineage database (vertebrata_odb10).Chromosome structural integrity was further validated by identifying telomeric repeat sequences using tidk (v0.2.41) (https://zenodo.org/records/10091385) withCypriniformes-specific motifs. To verify assembly accuracy, Illumina short-read sequencing data were aligned to the assembled genome using BWA-MEM (v0.7.8) and variant calling was performed using SAMtools (Li and Durbin, 2009). Base-level assembly accuracy was additionally evaluated using Merqury (Rhie et al., 2020). GC content across the genome was calculated in 10-kb sliding windows using SOAP.coverage (v2.7.7) (Li et al., 2009). Potential exogenous contamination was evaluated using BlobToolKit (v4.5.0). The 190 unplaced scaffolds were queried against the NCBI nt database via BLASTN (E-value ≤ 1×10⁻⁵), and the resulting taxonomic assignments were visualized alongside GC content and sequencing depth. All aligned hits were assigned to metazoan taxa; no bacterial, fungal, viral, or other non-metazoan sequences were recovered. Accordingly, no further decontamination steps were carried out.