A chromosome-level genome assembly of wild silkmoth, Bombyx mandarina

Abstract

The wild silk moth, Bombyx mandarina, is the closest relative of the domesticated silk moth, Bombyx mori. National BioResource Project of Japan (NBRP) maintains a B. mandarina strain derived from individuals captured at Sakado (Saitama, Japan) in 1982. Now, NBRP has developed chromosome replacement strains of B. mori. In each strain, one autosome of B. mori was replaced with the corresponding chromosome of B. mandarina. To facilitate the use of chromosome replacement strains and B. mandarina itself, we constructed a chromosome-level genome assembly of B. mandarina. Furthermore, we performed functional annotations of the genome assembly, i.e., transcriptome-based gene prediction, Assay for Transposase-Accessible Chromatin (ATAC)-seq, and PIWI-interacting RNA (piRNA)-targeted small RNA-seq. The assembly harbors 14,859 protein-coding genes and 628 piRNA clusters across three tissues: ovaries, testis, and embryos. ATAC-seq data comprehensively detected open chromosome regions, which will benefit when CRISPR/Cas9-mediated genome editing is conducted.

Background & Summary

B. mandarina is the closest relative of the domesticated silk moth, B. mori¹. After the ancestor species of B. mandarina and B. mori was initially domesticated in ancient China, it was brought to other Asian regions for raw silk production. We can find B. mandarina populations outside China, such as in the Korean peninsula and Japan^2,3. The phylogenetic relationships of B. mandarina populations in Asia are shrouded in mystery. Chinese B. mandarina has 28 homologous chromosomes, while Japanese B. mandarina has 27 pairs⁴. In Korea, located between China and Japan, we can find n = 28 and n = 27 individuals^3,5. Therefore, phylogenetic studies using ribosomal DNA sequences supported the close relationship between Japanese and Korean populations². However, studies using mitochondrial genome sequences have proposed a different hypothesis³: the Japanese population is more closely related to the B. mandarina population in southern China than the Korean population. This discrepancy might have resulted from the limited data exploited from rDNA or mitochondrial sequence analyses. Even though the chromosome numbers of Japanese and Korean B. mandarina are the same, it is still being determined whether the fused chromosomes in both populations have the exact evolutionary origin because any chromosomal genome assemblies of B. mandarina have not been constructed.

The National BioResource Project of Japan (NBRP) is an initiative that is obliged to maintain, manage, and attribute genetic resources in Japan. The silkworm section of the NBRP maintains not only mutant strains of B. mori but also wild silkmoths, including B. mandarina; the B. mandarina strain maintained by the NBRP originated from an individual captured in Sakado, Saitama in 1982, and has been repeatedly inbred by sib-cross for 42 years in The University of Tokyo and Kyushu University. This strain is probably the most highly purified B. mandarina strain (hereafter referred to as the “Sakado” strain). Moreover, NBRP has developed a series of consomic lines in which the B. mori chromosomes are replaced by B. mandarina chromosomes⁶. B. mandarina and B. mori differ in many physiological traits, such as larval motility and adult flight, which provide good material for genetic studies.

Therefore, we decided to construct the chromosome-scale genome assembly and annotate it for further phylogenetic and genetic studies. After we determined the genome sequences of the Sakado strain with long reads, the draft genome assembly was scaffolded with two scaffolding technologies: optical genome mapping and HiC-seq. As a result, we successfully constructed a chromosome-scale genome assembly of B. mandarina. As Banno et al. predicted⁴, single-copy orthologue anchored genome-wide alignment of B. mori genome (accession No. GCF_030269925.1) and the B. mandarina genome (accession No. GCA_030267445.2) revealed that homologous chromosomes of chromosome 14 and 27 of B. mori fused into a single chromosome in B. mandarina (Fig. 1; hereafter we refer to this chromosome as “A1” chromosome)⁴. Transcriptome-based gene prediction and functional annotation were also performed, identifying 14,859 protein-coding genes in this assembly. For the future application of the CRISPR/Cas9, we performed embryonic ATAC-seq to associate the positions of protein-coding genes and open chromatin regions since the enzymatic activity of Cas9 is less efficient in heterochromatin regions⁷. In addition to ATAC-seq, we performed piRNA-targeted small RNA-seq to identify piRNA clusters (piCs) since piRNA is involved in the sexual differentiation of Lepidoptera⁸.

Fig. 1.

A chromosome-scale genome assembly of B. mandarina. (a) Hi-C contact map of the genome assembly. Each block represents a Hi-C contact between two genomic loci. (b) Alignment of B. mandarina A1 chromosome and B. mori 14 and 27 chromosomes. (c) Single-copy orthologue (SCO) anchored synteny plot of B. mandarina and B. mori genomes. Each SCO was linked in a thin line. The lines in Inverted syntenies are shown in red while direction-conserved syntenies are in blue.

Methods

Insects

B. mandarina (Sakado strain) was provided from NBRP silkworm (https://shigen.nig.ac.jp/silkwormbase/). The B. mandarina larvae were fed on fresh leaves of Morus alba under a long-day condition (16 h light/8 h dark) at 25 °C.

Genome assembly

Genomic DNA was extracted from silk glands of a male final instar larvae of B. mandarina using Genomic-tip 100/G (QIAGEN). Extracted HMW DNA was submitted for SMRTbell library preparation and the Illumina paired-end library preparation. The prepared libraries were sequenced on the PacBio RS II or illumina HiSeq2500 platform (Supplementary Table 1). The draft assembly was constructed with FALCON (v 0.7) and FALCON-Unzip (v 0.4.0)⁹. FALCON was performed with the following parameters:

genome_size = 500000000

seed_coverage = 30

length_cutoff_pr = 10000.

Paired-end reads were used to polish the draft assembly using Pilon (v 1.22) with default settings¹⁰. The polished assembly was initially scaffolded with optical genome mapping as previously described¹¹: Genomic DNA was isolated from the pupa immediately after the pupation for the optical genome mapping. DNA isolation was conducted using the Bionano Prep Animal Tissue DNA Isolation Fibrous Tissue Protocol (Bionano Genomics). DLE-1 was used for the direct label stain. The DNA labeling was conducted using the Bionano Prep Direct Label and Stain Protocol. The labeled samples were scanned on the Bionano Saphyr system using Saphyr Chip G2.3. The obtained data were analyzed using Bionano Access (v 1.8.2)¹² and Bionano Solve (v 3.8.2)¹³. In the single enzyme pipeline, to create a.cmap file (describing the labeled genomic regions from the obtained data), we used a wrapper script “pipelineCL.py” bundled with the Bionano Solve with default settings. The resulting.cmap file was submitted to “hybridScaffold.pl”, also bundled with the Bionano Solve, which bridges the contigs based on the location information for the labelled regions in the.cmap file. “hybridScaffold.pl” was executed with the option “-B 2 -N 2”. The obtained optical genome mapping data was deposited at DDBJ¹⁴. The second scaffolding was conducted using Hi-C sequencing. The Hi-C library was prepared with a Dovetail Omni-C kit on pupal genomic DNA. Fastp (v 0.20.1)¹⁵ was used for the quality trimming of Hi-C reads with ‘-q 25 -l 50.’ The Hi-C scaffolding was conducted through juicer (v 2.0)¹⁶ – 3D-DNA (v 180922)¹⁷ pipeline with default settings. Hi-C contact map shows 27 distinct domains (Fig. 1a), which equals the number of chromosomes of Sakado strain⁴. The basic metrics of the assembly are summarised in Table 1.

Table 1.

The basic metrics of the genome assembly of B. mandarina.

Assembly	Species	Sex	Karyotype	Scaffolds	Scaffold N50 (bp)	Assembly Size (bp)	Accession No.
Bma-NBRP_v1.0	Bombyx mandarina	Male (ref. 4)	2n = 54	27	16,433,000	419,602,541	GCA_030267445.2

Comparison of B. mandarina genome structure with the B. mori genome

In comparison to the B. mori genome (accession No. GCF_030269925.1)¹⁸ using Mummer (v 4.0.0)¹⁹, the so-called “M chromosome” turned out to be homologous to B. mori chromosomes 14 and 27 (Fig. 1b). Following the nomenclature of the chromosomes of T. varians¹¹, a bombycid species (i.e. the prefix ‘A’ is used if a fusion has occurred in comparison to the B. mori chromosome), this chromosome was termed “chromosome A1.” In addition, we performed single-copy orthologue anchored genome-wide alignment of the and the B. mandarina genome (accession No. GCA_030267445.2)²⁰. The single-copy orthologues shared with the two assemblies were extracted by BUSCO (v 5.4.6)²¹. MCScanX²² was used to create a riparian plot (Fig. 1c).

Repetitive elements annotation in the genomes

Repetitive annotation of the B. mandarina genome was briefly summarised here: repetitive elements in the genome assembly were identified using RepeatModeler (v 2.0.4)²³ with the “-LTRstruct” option for performing an LTR structural search. In the following repeat annotation process using RepeatMasker (v 4.1.2)²⁴, the resulting “consensi.fa.classified” file was specified by “-lib” option²⁴. Among the repetitive elements, LTR, non-LTR (LINE or SINE), DNA transposons, and rolling circles were extracted and the density information of those repetitive groups is visualized by circlize (v 0.4.16)²⁵ (Fig. 2).

Fig. 2.

General genome annotation information. Summary of B. mandarina genome characteristics. The outermost to the innermost circle are I. chromosome ideograms; II. GC content; III. GC skew; IV. LTR element density; V. non-LTR retrotransposon density; VI. DNA transposon density; VII. rolling circle density; and VIII. gene model density.

RNA sample preparation for RNA-seq, sRNA-seq, and Isoform-seq (Iso-seq)

RNA samples derived from 12 different tissues (Supplementary Table 2) were prepared precisely as previously described¹¹. Total RNA was extracted using TRIzol reagent (Invitrogen) according to the manufacturer’s protocol. Embryos were sampled 24 hours after oviposition. The two aliquots of testis and ovary-derived RNA samples were subjected to RNA-seq and sRNA-seq, respectively. The three aliquots embryo-derived RNA samples were subjected to sRNA-seq, RNA-seq, and Iso-seq, respectively.

Library preparation for RNA-seq, sRNA-seq and Iso-seq

The sRNA-seq library was prepared using TruSeq small RNA kit (illumina) according to the manufacturer’s protocol with a slight modification. To target piRNA, a region of 147–158 nucleotides was extracted in the purification step of the cDNA construct using BluePippin (Sage Science). The constructed library was sequenced on the illumina HiSeq2500 platform (illumina). Except for an RNA sample from embryos, the RNA-seq libraries were prepared using TruSeq stranded mRNA kit (illumina) according to the manufacturer’s protocol. The embryonic RNA-seq library was prepared using NEBNext Poly(A) mRNA Magnetic Isolation Module (New England BioLabs) and the NEBNext® Ultra™ ll Directional RNA Library Prep Kit (New England BioLabs) according to the manufacturer’s protocol. The constructed RNA-seq libraries were sequenced on the illumina NovaSeq6000 platform (illumina). For Iso-seq, the library was constructed using Sequel Iso-seq Express Template Prep (Pacific Bioscience) according to the manufacturer’s protocol. The constructed library was sequenced on the PacBio Sequel platform (PacBio).

Transcriptome-based gene prediction

BRAKER3 (v 3.0.8) was used for gene prediction^26,27. The RNA-seq and Iso-seq data were submitted to BRAKER3 separately²⁸, and Tsebra finally merged their respective prediction²⁹. The detailed information on transcriptome data is summarised in Supplementary Table 2. Quality trimming for short read data was conducted using fastp (v 0.20.1)¹⁵ with following options: ‘-q 28 -l 80. ‘ Trimmed short read data were submitted to BRAKER3 using the ‘--rnaseq_sets_ids’ option. The short reads were aligned to the genome assembly by hisat2 (v 2.2.1)³⁰. Iso-seq data were generated consensus for each read cluster according to the following procedure³¹: Iso-seq subreads were converted to circular consensus sequences (ccs) using ccs v 6.4.0 with options ‘--minLength 10 --maxLength 100000 --minPasses 0 --minSnr 2.5 --minPredictedAccuracy 0.0.’ lima (v 2.7.1) was used to remove primer sequences from the CCSs with options ‘--isoseq --peek-guess --ignore-biosamples.’ After the trimming of adaptors, PolyA tail trimming and concatemer removal were performed by isoseq3 (v 3.8.2) in ‘refine’ mode with option ‘--require-polya.’ Finally, isoform-level clustering was conducted by isoseq3 in ‘cluster’ mode with option ‘--use-qvs.’ The resulting clustered.bam file was submitted to BRAKER3. Before gene prediction with Iso-seq data, BUSCO analysis on the genome assembly was conducted to obtain complete and single-copy BUSCO sequences³⁰. Complete and single-copy BUSCO sequences were submitted to BRAKER3 with an Iso-seq-derived bam file. Since we had two Iso-seq datasets (Supplementary Table 2), we ran BRAKER3 for them separately. The resulting gene models were also submitted to BUSCO analysis³⁰, scoring 94.5% completeness (Fig. 3a). The basic metrics of gene models were summarised in Table 2.

Fig. 3.

BUSCO assessment and the top 10 GO assignments of the predicted gene models. (a) BUSCO scores of the gene models (top) and the genome assembly (bottom). (b) Overall top 10 GO assignments to gene models.

Table 2.

Statistical summary of the constructed gene models.

No. of protein coding gene	14,859
Average CDS length [bp]	1526.6
Average exon length [bp]	230.50
Average intron length [bp]	1791.4

Functional annotation of gene models

The deduced amino acid sequences of gene models were submitted to EnTAP⁶ for functional annotation. A protein similarity search was conducted against the latest complete UniProtKB/TrEMBL protein data set and complete UniProtKB/Swiss-Prot data set using diamond (v 0.9.14)³². A protein orthology search was also conducted against the Evolutionary genealogy of genes: Non-supervised Orthologous Groups (EggNOG) databases30 to assign Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG) terms, and protein domains from pfam³³ and smart³⁴. Additional family and domain search was performed against tigrfam³⁵, sfld³⁶, hamap³⁷, cdd³⁸, superfamily³⁹, prints⁴⁰, panther⁴¹, and gene3d⁴² using InterProScan (v 5.68-100)⁴³. The results of functional annotation are summarized in Table 3. The top 10 GOs assigned to the gene models are shown in Fig. 3b without distinguishing between molecular function, biological process, and cellular component. The top 10 GOs for each category are shown in Supplementary Fig. 1.

Table 3.

Brief summary of functional annotation.

		Similarity search			Ontology search		total
		EggNOG	TrEMBL	Swiss-prot	EggNOG**	InterPro
aligned	informative	11,717	11,054	7,135	9,839	12,153	annotated***	13,111
	uninformative*	0	2,039	176	2,895	—
unaligned		3,142	1,766	7,548	2,125	2,706	unannotated***	1,748

*When the query sequences were aligned to sequences whose description contains any of conserved/predicted/unnamed/hypothetical/putative/unidentified/uncharacterized/unknown/uncultured/uninformative, such alignment was categorized as “uninformative,” and the query sequence was treated as an unannotated sequence.

**In this column, queries with at least one GO term were treated as “Informative,” while queries without GO terms were treated as “Uninformative” “Unaligned” in this column means queries without protein family assignment.

*** “Annotated” means at least one match yielded from any of the databases. “Unannotated” means no match yielded from all databases.

ATAC library preparation and data processing

Another batch of early embryo samples was subjected to Isoform-seq (Iso-seq), and small RNA-seq was subjected to ATAC-seq. Fragmentation and amplification of the ATAC-seq libraries were conducted according to Buenrostro et al.⁴⁴. The constructed libraries were sequenced on the Illumina HiSeq. ATAC-seq reads were pretreated with fastp and mapped to the genome with bwa-mem2 (v 2.2.1)⁴⁵. Alignments containing mismatches were then removed using bamutils (v 0.5.9)⁴⁶. Next, we removed duplicated reads using GATK MarkDuplicates (v 4.1.7)⁴⁷. The resulting bam files were converted to bigwig files using deepTools bamCoverage (v 3.5.1)⁴⁸. Heatmap was created using deepTools computeMatrix, and the starting point of the gene model was set to the reference point (Fig. 4).

Fig. 4.

Heatmap around gene bodies of ATAC-seq on early embryos. The ATAC-seq peaks of 3 kb upstream and downstream from the start codon of protein-coding genes are shown. The peaks of ATAC-seq are concentrated at the 5′ end of the gene body, which is a typical result of ATAC-seq.

Small RNA mapping

The small RNA reads were trimmed using Trim Galore (v 0.6.6)⁴⁹ in small RNA mode. The trimmed small RNA reads were mapped to the genome assembly, allowing up to 3 nucleotide mismatches using Hisat2 (v 2.1.0)³⁰ and ngsutils (v 0.5.9)⁴⁶. The information for each library is summarized in Supplementary Table 2.

piRNA cluster detection

The piC detection was performed as previously described^11,50. proTRAC (v 2.4.4)⁵⁰ was used with options ‘-clsize 5000 -pimin 23 -pimax 29 -1Tor10A 0.3 -1Tand10A 0.3 -clstrand 0.0 -clsplit 1.0 -distr 1.0-99.0 -spike 90-1000 -nomotif -pdens 0.05.’ As a result, we successfully identified 560 piRNA clusters in the three tissues (Fig. 5). The identity of piC is defined by the two nearest gene models. If multiple piCs were predicted between such two genes, such piCs were treated as a single piC. The genomic positions of piCs identified in testes, ovaries, and early embryos were visualized by RIdeogram (v 0.2.2)⁵¹ (Fig. 5a). The aggregation relationship of those piCs was visualized by ComplexUpset (v 1.3.3)⁵² (Fig. 5b).

Fig. 5.

piRNA clusters on B. mandarina genome. (a) piCs distribution detected in early embryos (box), pupal ovary (circle), and pupal testis (triangle). (b) UpSet plot visualising piCs that are each assigned to each tissue. The vertical bars correspond to the intersections. When the circles corresponding to tissues are connected by a line, the bar above circles represents the number of piCs commonly identified in concerning tissues. For example, the yellow bar indicates the number of piCs identified in all tissues. The identity of piCs was defined by the nearest two gene models: When comparing piCs identified in different tissues, if the nearest upstream and downstream gene models are the same, those piCs were treated as the same piC.

Data Records

The raw sequence data reported in this paper have been deposited in DDBJ. Genomic data for the draft assembly were deposited under the accession code PRJDB5778⁵³ while genomic data for scaffolding were deposited under the accession code PRJDB13954¹⁴. The genome assembly is available under the accession code GCA_030267445.2²⁰. Except for embryonic transcriptome data, all transcriptome data. i.e. RNA-seq and Iso-seq data were registered under the accession code PRJDB13954¹⁴. Embryonic RNA-seq, ATAC-seq and Iso-seq data are available under the accession code PRJDB13955⁵⁴. The annotated gene models have been deposited to the figshare repository⁵⁵.

Technical Validation

BUSCO (v 5.4.6)²¹ with lepidoptera_odb10 lineage dataset was used to assess the quality of gene models. For comparison, the results are summarised in Fig. 3, together with the results of BUSCO analysis for the genome assembly. 92.8% of the complete and single-copy BUSCO sequences were present in the gene models, while 93.4% of the complete and single-copy BUSCO sequences were in the genome assembly. BUSCO completeness scores were almost the same between the genome assembly and the gene model, suggesting that the gene prediction process is highly accurate across all genome regions.

Author contributions

J.L. designed the research plan, performed RNA extraction, analyzed the obtained data, and wrote the manuscript. T.S. and T.K. also designed this research plan. T.S. also performed the data analysis. T.K. reared B. mandarina larvae and collected their silk glands for genome sequencing. A.T. extracted the genomic DNA from the sample and was responsible for library preparation and sequencing operation. K.Y. and S.S. prepared the Hi-C seq library.

Code availability

Programs exploited in this study were executed with the default parameters except where otherwise specified in the Methods section.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41597-025-04395-0.