Telomere-to-telomere haplotype-resolved reference genome reveals subgenome divergence and disease resistance in triploid Cavendish banana

Abstract

Banana is one of the most important crops of the world. Cavendish-type bananas, which have a monospecific Musa acuminata origin (AAA), account for around half of the global banana production, thereby are of great significance for human societies. However, until now, the high-quality haplotype-resolved reference genome was still undecoded for banana cultivars. Here, we reported the telomere-to-telomere (T2T) and haplotype-resolved reference genome of ‘Baxijiao’ (Cavendish) consisting of three haploid assemblies. The sizes of the three haploid assemblies were estimated to be 477.16 Mb, 477.18 Mb, and 469.57 Mb, respectively. Although with monospecific origins, the three haploid assemblies showed great differences with low levels of sequence collinearity. Several large reciprocal translocations were identified among chromosomes 1, 4, and 7. An expansion of gene families that might affect fruit quality and aroma was detected, such as those belonging to sucrose/disaccharide/oligosaccharide catabolic processes, sucrose metabolic process, starch metabolic process, and aromatic compound biosynthetic process. Besides, an expansion of gene families related to anther and pollen development was observed, which could be associated with parthenocarpy and sterility of the Cavendish cultivar. Finally, much fewer resistance genes were identified in ‘Baxijiao’ than in M. acuminata, particularly in the gene clusters in chromosomes 3 and 10, providing potential targets to explore for molecular analysis of disease resistance in banana. This T2T haplotype-resolved reference genome will thus be a valuable genetic resource for biological studies, molecular breeding, and genetic improvement of banana.

Introduction

Banana originated in Southeast Asia and the west Oceania region, where it was domesticated at least 7000 years ago [1]. Most banana cultivars are unseeded clones, mainly diploids or triploids, derived from wild fertile diploid Musa species. The genetic diversity of banana cultivars mostly resulted from hybridization events within Musa acuminata (A genome, 2n = 22), and between this species and Musa balbisiana (B genome, 2n = 22); meanwhile, to a minor extent, Musa schizocarpa (S genome, 2n = 22) and the species of Australimusa section (T genome, 2n = 22) also contributed to the genetic contents of edible banana cultivars [2]. Banana cultivars are among the most consumed fruits worldwide: over 100 million tonnes of banana are produced yearly [3]. Besides, banana cultivars are also the fifth most produced food crop in the least-developed countries; as a result, they play an important role in maintaining food security in these regions. The demand for cultivated bananas continues to grow; however, their production is greatly affected by a complex of biotic and abiotic stresses. For instance, banana yields are severely threatened by the disease fusarium wilt, which is caused by the fungus Fusarium oxysporum f. sp. cubense (Foc) [4]. Therefore, breeding new banana cultivars with improved traits, particularly genetic resistance to a wide and diverse group of pathogens, is crucial for future banana production; and a high-quality banana genome would no doubt facilitate such genetic manipulation.

Efforts to sequence banana genomes began over 10 years ago. D’Hont et al. [5] reported the first draft genome of a doubled-haploid Pahang (DH-Pahang) M. acuminata genotype of the subspecies malaccensis, providing a key stepping-stone for genetic breeding for banana. Since then, the DH-Pahang genome has been improved twice [6, 7]. Moreover, other Musa genomes, for instance M. balbisiana, Musa itinerans, M. schizocarpa, Musa troglodytarum, and Musa beccarii, have already been reported during the past decade [2, 8–12]. These high-quality banana genomes have provided necessary and valuable resources for further research on evolution of Musa genus, as well as domestication and breeding of banana. Technological advances have boosted sequencing capacities and made the sequencing of complex genomes from multiple species and cultivars available. Thanks to long-read sequencing, now telomere-to-telomere (T2T) gap-free assembly of chromosome sequences is possible, providing unprecedented opportunities to untangle genomic regions missed in previous assemblies [13]. However, there has long been a lack of high-quality genomes for banana cultivars, likely due to the complexity of the genomes [14–16].

In banana, the most efficient ploidy level for agronomic performance is triploid, which has given rise to more vigorous plants, higher sterility, and larger fruits without seeds. Cavendish-type bananas, which have a monospecific M. acuminata origin (AAA), account for around half of the global banana production [1, 17]. Nearly all of the global banana export trade relies on the clones derived from a unique triploid cultivar of Cavendish banana [17], leading to a narrow genetic basis in this type of banana. In this study, we developed a T2T reference genome of a banana cultivar ‘Baxijiao’ (AAA, Cavendish) consisting of three haploid assemblies, by integrating PacBio high-fidelity (HiFi) reads, Oxford Nanopore Technology (ONT) ultra-long reads, and Hi-C reads. This haplotype-resolved reference genome revealed great differences among the three haploid assemblies with low levels of sequence collinearity. In addition, some enriched biological processes of ‘Baxijiao’ specific gene families were found to be related to fruit quality and aroma, as well as parthenocarpy and sterility of the cultivar. Finally, the presence of much fewer resistance genes in the ‘Baxijiao’ genome than in M. acuminata genome shed light on the possible mechanisms of fusarium wilt susceptibility of this cultivar. We hope this high-quality genome will help to obtain a better understanding of banana domestication, and open a new range of opportunities for biological studies and the breeding of banana varieties.

Results

Assembly of highly contiguous and phased triploid reference genome

We generated the ‘Baxijiao’ genome by incorporating PacBio HiFi sequences, Oxford nanopore sequences, and Hi-C sequences. In total, we obtained 102.06 Gb (~72× coverages of the total genome size) of PacBio HiFi reads with N50 longer than 15 kb, 48.00 Gb (~34× coverages) of ONT reads with N50 of 80 kb, and 156.14 Gb (~110× coverages) of Hi-C data, respectively (Table S1, see online supplementary material).

The haploid genome of ‘Baxijiao’ was estimated to be ~457.31 Mb with a heterozygous rate of 2.855% by GenomeScope 2.0 [18] based on a k-mer of 21 (Fig. S1, see online supplementary material). To achieve the T2T and haplotype-resolved genome, we adopted an assembly scheme that allows combining HiFi and Hi-C reads to assemble the phased genome coupled with ONT reads to fill the gaps (Fig. S2, see online supplementary material). Based on HiFi and Hi-C reads, hifiasm [19] was applied to build a phased diploid assembly, generating 581.77 Mb and 959.85 Mb primary contigs with N50 of 14.95 Mb and 16.15 Mb, respectively. With the aid of the T2T genome of M. acuminata ssp. malaccensis (X. Liu et al. unpublished results), the contigs were further ordered, orientated, and grouped. Then Hi-C reads were employed to anchor and remove some short contigs, in addition to splitting the larger contigs into two sets. NextDenovo (https://github.com/Nextomics/NextDenovo) was used to construct the ONT assembly, which was subsequently used to fill the gaps of the HiFi-assembled reference. After filling gaps, a nearly complete and phased reference genome named BXJ was generated, containing 33 pseudo-chromosomes (11 for each haploid assembly) with a total length of 1.42 Gb, and sizes of 469.57–477.18 Mb for haploid assemblies (Fig. 1A andTable 1; Table S2, see online supplementary material). The three haploid assemblies were named BXJ1, BXJ2, and BXJ3. The longest chromosome was chromosome 8 for each haploid assembly, with a size of 53.00 Mb, 54.63 Mb, and 52.65 Mb for BXJ1, BXJ2, and BXJ3, respectively. The shortest chromosome showed slightly different between BXJ1 and the other two haploid assemblies; specifically, chromosome 2 was the shortest chromosome in BXJ1 (33.52 Mb), while chromosome 11 was the shortest one in both BXJ2 (33.79 Mb) and BXJ3 (32.90 Mb). Using 100-kb intervals, the gene density, TE density, Copia density, Gypsy density, GC contents and potential centromere tandem repeats density were counted in the BXJ genome, respectively (Fig. 1A). This is the first T2T and haplotype-resolved reference genome for triploid bananas.

Figure 1.

Overview of the Cavendish reference genome. A The circos diagram of the ‘Baxijiao’ genome. The tracks represent the following elements (from outer to inner): (a) schematic representation of the 33 chromosome sequences, (b) density of centromeric repeats, (c) GC contents, (d) density of transposable elements, (e) density of the Gypsy elements, (f) density of the Copia elements, (g) density of genes. The innermost is syntenic relationships. B Hi-C interaction heatmap for the ‘Baxijiao’ genome. C HiFi coverage of the 33 pseudo-chromosomes. D Mapping rates of HiFi, Illumina and RNAseq reads in BXJ1, BXJ2, and BXJ3. E BUSCO assessments in BXJ1, BXJ2, and BXJ3. F Sequence collinearity among BXJ1, BXJ2, and BXJ3. Gray lines represent the collinearity blocks with length of 15 000 bp, while red lines represent the potential inversions.

Table 1.

Genome assembly statistics for ‘Baxijiao’ and MAv4

	Baxijiao			MAv4
	BXJ1	BXJ1	BXJ1
Cumulative size (bp)	477 162 536	477 181 898	469 568 689	468 821 802
HiC scaffolds	11	11	11	11
Complete BUSCO (N = 1614)	97.40%	97.80%	93.80%	98.80%
Gaps	33	40	71	15
After gap-closing	0	2	0	15
Number of Telomere	19	17	17	22
HiFi mapping rate	99.84%	99.80%	99.50%
Illumina mapping rate	97.95%	97.66%	96.36%
Repeats	53.76%	54.14%	55.13%	52.62%
LAI index	19.84	20.65	20.22
Number of predicted genes	37 185	37 241	37 178	36 979

Genome annotation and quality assessment

Within the three haploid assemblies, 256.48 Mb (53.76%), 258.29 Mb (54.14%), and 258.84 Mb (55.13%) of repetitive regions were identified (Table S3, see online supplementary material), similar to those in other banana genomes (e.g. M. acuminata: 52.62% [7]; M. beccarii: 55.99% [12]). Among them, the most abundant repetitive sequences were long terminal repeats (LTRs), which accounted for 36.36%, 35.44%, and 33.53% of the BXJ1, BXJ2, and BXJ3 assemblies, respectively (Table S3, see online supplementary material). A total of 37 185, 37 241, and 37 178 high-confidence protein-coding genes were predicted from BXJ1, BXJ2, and BXJ3 (Table 1), which were slightly more than the genes (36 979) predicted in the latest v4 version of the M. acuminata double-haploid genome (hereafter MAv4) [7]. The total lengths of the predicted genes in BXJ1, BXJ2, and BXJ3 were 142.78 Mb, 138.99 Mb, and 133.28 Mb, respectively. The average lengths of these protein-coding genes were 3 839 bp, 3 732 bp, and 3 584 bp.

Using the seven-base repeats (CCCATTT at the 5′ end and TTTAGGG at the 3′ end) as queries, we totally estimated 53 telomeres and constructed 8, 6, and 6 telomere-to-telomere chromosomes for BXJ1, BXJ2, and BXJ3, respectively (Table S4, see online supplementary material). The length of the identified telomeres ranged from 6 006 to 33 838 bp in BXJ1, 2 345 to 44 905 bp in BXJ2, and 4 494 to 26 964 bp in BXJ3. The longest telomere sequence was located at the right end of Chr10 in BXJ2, whereas the shortest one was located at the right end of Chr08 in BXJ2.

Centromere is a specialized nucleotide sequence in a chromosome that holds together the two daughter chromatids and is involved with the movement in mitosis and meiosis. Using Tandem Repeats Finder [20], the approximate locations of the centromeres in the ‘Baxijiao’ genome were identified, with lengths ranging from 924 207 to 3 415 439 in BXJ1, 930 592 to 3 844 332 in BXJ2, and 481 633 to 4 080 464 bp in BXJ3, respectively (Table S5 and Fig. S3, see online supplementary material).

The quality and completeness of the BXJ assembly were evaluated in multiple ways. First, as estimated, the genome sizes of each haploid assembly (469.57–477.18 Mb) were slightly longer than that of MAv4 (~468.82 Mb) [7]; in addition, Hi-C data visualized by JuiceBox [21] presented a high consistency across all chromosomes, providing their accuracy of the ordering and orientation (Fig. 1B). Second, both HiFi and Illumina reads were aligned to the three haploid assemblies; the average HiFi read coverages of each chromosome were ~200 × (Fig. 1C); besides, 97.95%, 97.66%, and 96.36% Illumina reads, as well as 99.84%, 99.80%, and 99.50% HiFi reads were successfully aligned to BXJ1, BXJ2, and BXJ3, respectively (Fig. 1D), indicating sufficient and high coverage of the ‘Baxijiao’ genome by this assembly. Third, high LTR assembly index (LAI) scores (19.84, 20.65, and 20.22) were observed in the three haploid assemblies (Table 1). Finally, evaluation of completeness using BUSCO [22] revealed that complete sequences of BXJ1, BXJ2, and BXJ3 accounted for 97.40%, 97.80%, and 93.80% of the conserved core eukaryotic gene set, respectively (Table 1 and Fig. 1E). Overall, these results presented the high quality, accuracy, and reliability of the BXJ reference genome.

Whole genome comparison and syntenic analysis

BXJ1, BXJ2, and BXJ3 harbor a set of similar genomic features, such as similar genome sizes and gene numbers, as well as parallel repeat contents (Table 1). However, the syntenic analysis revealed low levels of nucleotide sequence collinearity between the haploid assemblies (Fig. 1F), which could be attributed to the different origins of M. acuminata subspecies of the subgenomes. Meanwhile, due to evolutionary conserved coding regions, high syntenic relationships of genes among the three haploid assemblies were observed. Genomic collinearity analysis revealed that 29 131, 27 571, and 28 838 transcripts matched 31, 37, and 37 syntenic blocks in the three pairwise comparisons (BXJ1 vs BXJ2, BXJ1 vs BXJ3, BXJ2 vs BXJ3), respectively. Substantial structural variations within the BXJ genome were identified using MCScan [23]. Several large reciprocal translocations were identified among Chr01, Chr04, and Chr07 (Fig. 2A). Specifically, syntenic relationships of Chr01 and Chr04 were observed, indicating the occurrence of a reciprocal translocation involving the distal end of the long arm of Chr01 in BXJ2 and a 0.47–0.50 Mb region in the long arm of Chr04 in BXJ1 and BXJ3 (Fig. S4A, see online supplementary material). In addition, another reciprocal translocation involving a 1.88–3.12 Mb region in the distal end of the long arm of Chr01 in BXJ1 and BXJ3 and the long arm of Chr04 in BXJ2 was also identified (Fig. S4B, see online supplementary material). Moreover, syntenic relationship of Chr01 and Chr07 was also observed, showing the existence of a reciprocal translocation involving the distal end of the long arm of Chr01 in BXJ1 and a 1.04 Mb region in Chr07 in BXJ2 (Fig. S4C, see online supplementary material).

Figure 2.

Whole genome comparisons within ‘Baxijiao’ and between ‘Baxijiao’ and other species in monocots. A Syntenic analysis at the gene level. Highlighted lines represent reciprocal translocations. The other lines represent syntenic blocks. B Inferred phylogenetic tree and the expanded (+) or contracted (−) gene families in ‘Baxijiao’ and other species in monocots. Number of expanded and contracted gene families at branch (share between species/BXJ haploid assemblies) and leaf nodes (share in specific species/BXJ haploid assemblies) has been denoted. C The top enriched biological processes for expanded and contracted genes in ‘Baxijiao’.

According to sequence identity, we obtained 18 263 groups of alleles between the haplotype-resolved BXJ genome and MAv4 (Table S6, see online supplementary material). These alleles are evenly distributed across the 11 chromosomes of the genome. The coding sequences of most allelic genes in the three haploid assemblies were highly similar to those in MAv4 (mean = 96.50%, 96.59%, and 96.62%). In addition, through calculating Ka/Ks values between the allelic gene pairs, we found that most allelic genes had experienced purifying selection (Ka/Ks < 1, Fig. S5, see online supplementary material).

Gene family evolution

Phylogenomic analysis was performed using the homologous genes from ‘Baxijiao’ and 15 other species in monocots, including the main ancestral contributors of banana cultivars: M. acuminata (A genome), M. balbisiana (B genome), M. schizocarpa (S genome), and M. troglodytarum (T genome). The phylogenetic tree suggested that the three haploid assemblies of ‘Baxijiao’ were clustered together, and sister to M. acuminata (Fig. 2B).

In total, 38 387 orthogroups were characterized from a set of 612 509 genes from the chosen monocot plant species. Based on the phylogenetic tree, potential expanded and contracted gene families were identified. In total, 447 significant expansion and contraction gene families consisting of 1 551 genes were specific to ‘Baxijiao’. A total of 117 gene families composed of 694 genes have significantly expanded in the ‘Baxijiao’ genome. GO enrichment analysis revealed that these expanded genes were significantly enriched in 199 biological processes (Table S7, see online supplementary material). The most enriched biological processes were related to response to stress and sucrose induced translational repression (Fig. 2C). Interestingly, some other significantly enriched biological processes were associated with sources of carbohydrates and pollen development, including response to disaccharide, response to sucrose, response to carbohydrate, acceptance of pollen, recognition of pollen, pollen-pistil interaction, and pollen germination. Meanwhile, 330 gene families composed of 857 genes have significantly contracted, which were enriched in 231 biological processes (Table S8, see online supplementary material). A large proportion of the enriched biological processes (30%) were associated with plant growth, morphogenesis, and hormones, such as cotyledon development, fruit development, anther development, floral organ morphogenesis, developmental process involved in reproduction, auxin efflux, hormone-mediated signaling pathway, and so on. In addition, some of the enriched biological processes (14%) were associated with response to external biotic and abiotic factors, including response to symbiotic fungus, response to bacterium, immune response-activating signal transduction, red or far-red light signaling pathway, and cellular response to blue light. Potential expanded and contracted gene families were also characterized in the haploid assemblies (Table S9, see online supplementary material). Interestingly, some significantly expanded gene families were enriched in aromatic compound biosynthetic process, sucrose catabolic process, disaccharide catabolic process, oligosaccharide catabolic process, sucrose metabolic process, and starch metabolic process, which might be important for fruit quality (Fig. S6 and Tables S10–S12, see online supplementary material). In addition, some expanded gene families were involved in anther, particularly anther wall tapetum development, such as anther morphogenesis and anther wall tapetum formation.

Resistance genes

Plant genomes typically include a great number of disease resistance (R) genes. There are two main categories of plant R proteins: membrance-bound pattern recognition receptors (PRRs) and intracellular nucleotide-binding, leucine-rich repeat receptors (NLRs). PRRs are either receptor-like kinases (RLKs) or receptor-like proteins (RLPs) [24]. In this study, we first employed RGAugury [25] to predict resistance gene analogs (RGAs) in ‘Baxijiao’ and M. acuminata. A total of 68, 52, and 69 NBS-encoding genes were predicted in BXJ1, BXJ2, and BXJ3, respectively, compared to 124 in MAv4 (Table 2). The prediction for RLK- and RLP-type RGAs also suggested that ‘Baxijiao’ harbors considerably fewer R genes than M. acuminata (Table 2). The distribution of the above predicted RGAs in MAv4 and ‘Baxijiao’ genomes was presented in supplementary Fig. S7 (see online supplementary material). In addition, we applied the NLR-annotator [26] to characterize the genomic regions likely associated with disease resistance. An NLR locus predicted might be a gene with a complete or partial open reading frame but might also be the trace of a pseudogenized sequence in the investigated genome [26]. A total of 123, 126, and 140 NLR loci were detected in BXJ1, BXJ2, and BXJ3, respectively (Table 2; Tables S13–S15, see online supplementary material), compared to 128 loci in MAv4 [7]. NLR loci were unevenly distributed on all 11 chromosomes (Fig. 3). Some NLR loci were found to cluster together preferentially. Three major clusters of the predicted NLR loci were observed in the BXJ genome: two in Chr03 and one in Chr10 (Table 3). It is worth noting that a large proportion of NLR loci in the BXJ genome were found to be located in intergenic regions (BXJ1: 37.4%; BXJ2: 59.5%; BXJ3: 38.6%). Only 77, 51, and 79 resistance genes were predicted in the three haploid assemblies compared to 122 genes in MAv4, similar to the prediction for NBS-encoding RGAs. Especially in the gene clusters in the distal end of the long arm of Chr03 (BXJ1: 43.02–43.80 Mb; BXJ2: 41.36–42.09 Mb; BXJ3: 36.85–37.62 Mb), all NLR loci were predicted to be pseudogenized sequences (Fig. 3 and Table 3). In addition, in the gene cluster in Chr10, all NLR loci were predicted to be pseudogenized in BXJ2 (21.93–22.73 Mb), while most NLR loci in BXJ1 (19.29–20.01 Mb) and BXJ3 (20.93–21.73 Mb) remained to be functional according to gene annotation (Fig. 3 and Table 3). Overall, much fewer resistance genes were identified in ‘Baxijiao’ than in M. acuminata.

Table 2.

Summary of the predictions by RGAugury and NLR-annotator

Genome	RGAugury			NLR-annotator
	NBS encoding	RLK	RLP	NLR loci	NLR gene
MAv4	124	752	109	128	122
BXJ1	68	554	63	123	77
BXJ2	52	541	87	126	51
BXJ3	69	468	62	140	79

Figure 3.

The distribution of NLR loci in MAv4 and ‘Baxijiao’ genomes. Lines on chromosomes suggest that the loci are located in coding regions and intergenic regions, respectively. The syntenic relationships of the gene clusters in chromosome 3 and 10 are presented in the right panel.

Table 3.

Comparison of NLR clusters between the BXJ and MAv4 assemblies

Chr	Location (Mb)				No. NLR genes / No. NLR loci
	BXJ1	BXJ2	BXJ3	MAv4	BXJ1	BXJ2	BXJ3	MAv4
Chr03	39.20–39.73	37.36–37.99	32.82–33.48	36.32–36.95	3/8	0/12	19/20	15/16
Chr03	43.02–43.80	41.36–42.09	36.85–37.62	40.22–40.98	0/26	0/17	0/33	14/18
Chr10	19.29–20.01	21.93–22.73	20.93–21.73	24.69–25.42	8/13	0/18	10/17	12/13

No. NLR genes: number of NLR genes predicted.

No. NLR loci: number of NLR loci detected.

Homoeolog expression patterns in phased triploid reference genome

To determine patterns of homoeolog expression, we focused on 52 593 expressed genes (17 531 syntenic triads in total 18 263 triads, 95.99%) that had a 1:1:1 gene correspondence across the three haploid assemblies, and a homoeolog expression of >0.5 transcripts per million (TPM) in at least one tissue (Table S16, see online supplementary material). For each of the 52 593 expressed triads, we first standardized the relative expression of the BXJ1, BXJ2, and BXJ3 homoeologs for each individual tissue and combined analysis. We then carried out a global analysis integrating data across all four tissues (Fig. 4A). On average, ~40% of homoeolog triads showed non-balanced expression patterns, with higher (homoeolog-dominant) or lower expression (homoeolog-suppressed) from a single homoeolog respecting the other two. As for each tissue, balanced triads ranged from 49.45% in fruit to 55.00% in leaf (Fig. 4B). The homoeolog-dominant triads were found at moderate frequencies for each tissue (ranging from 11.95% to 12.83%), whereas the homoeolog-suppressed triads were more frequent (from 33.06% to 37.72%). Genes from syntenic triads in the dominant groups had higher absolute transcript abundance than genes in the balanced or suppressed groups (Fig. 4C). In addition, we found that the non-balanced triads showed subtly higher Ka/Ks values than the balanced triads (Fig. 4D). GO enrichments were carried out on the non-balanced triad; however, significant functional enrichment was not observed for these genes. The density of non-balanced expression genes was presented in Fig. 4E, which showed higher abundance of these genes along the regions near the ends of chromosomes.

Figure 4.

Homoeolog expression bias in syntenic homoeolog triads. A Ternary plot showing relative expression levels of 17 531 syntenic triads (52 593 genes) in ‘Baxijiao’. Each circle represents a gene triad with a BXJ1, BXJ2, and BXJ3 coordinate suggesting the relative contribution of each homoeolog to the overall triad expression (an example is shown on the top left). The gray circles represent the balanced triads, while the colored circles in vertices correspond to the triads from the homoeolog-dominant groups, and the colored circles close to edges and between vertices correspond to the triads from the homoeolog-suppressed groups. Box plots indicate the relative contribution of each homoeolog based on triad assignment to the seven groups, including BXJ1D, BXJ2D, BXJ3D representing BXJ1, BXJ2 and BXJ3 dominant respectively, and BXJ1S, BXJ2S and BXJ3S represent BXJ1, BXJ2, and BXJ3 suppressed. B Contribution of each group of homoeolog expression bias across the 4 tissues and in the combined analysis. C Box plot of absolute TPM expression abundance for each haploid genome from the seven groups. D Violin plots of the Ka/Ks values for the seven categories. The balance category shows slightly lower Ka/Ks values than the homoeolog-dominant and homoeolog-suppressed groups. E The distribution of homoeolog expression bias genes in all 33 chromosomes.

Discussion

Banana is among the most popular fruits and the fifth most produced food crop worldwide. The main traits selected in banana during domestication are parthenocarpy and sterility, which ensure the formation of fleshy and edible fruits without seeds. Genetic improvement and breeding for banana is challenging, because it is hard to generate high-quality parthenocarpic and sterile varieties through the recombination of non-parthenocarpic and fertile wild species. In this context, the genomic characterization of banana cultivars and a deep understanding of the wild-to-domesticate transition are crucial for successful banana breeding schemes. Here we assembled the T2T and haplotype-resolved reference genome of ‘Baxijiao’, a representative of the most commercialized Cavendish bananas. Our incorporated assembly scheme involving HiFi reads, ONT ultra-long reads, and Hi-C reads generated a high-quality BXJ genome, achieving the assembly goals of high contiguity and accuracy. Multiple evaluations including genome size and consistency, read mapping, LAI scores, and BUSCO supported the high quality, accuracy, and reliability of the BXJ assembly. Moreover, we identified the locations of the telomeres and centromeres in the chromosomes. To the best of our knowledge, this is the first T2T and phased reference genome for triploid bananas. This genome will thus provide valuable genetic resources for further research on banana domestication, and molecular breeding and genetic improvement of banana cultivars.

The haplotype-resolved reference genome of triploid Cavendish cultivar showed sequence and expression divergences among subgenomes

The domestication process has left signatures in the genomes of domesticated species. Cultivated bananas were derived from inter- or intra-specific Musa hybrids [2]. Most banana cultivars were derived from M. acuminata (A genome), a complex of subspecies that are geographically segregated in distinct continents and islands in Southeast Asia [27]. Four specific M. acuminata subspecies have been raised as the main parental contributors of banana cultivars, namely banksii, burmannica, malaccensis, and zebrina [1]. Accumulating evidence has put forward the subspecies banksii from New Guinea, zebrina from Java, and malaccensis from Malaysia and Sumatra as the main contributors to the genetic makeup of banana cultivars, and the minor involvement of burmannica [28, 29]. In addition, several recent studies on banana cultivars with pure M. acuminata origins proposed the presence of missing ancestral gene pools involving the setup of banana cultivars [15, 16, 30]. Cavendish-type bananas have a monospecific M. acuminata origin (AAA). Previous studies have established the contribution of the ‘Mlali’ subgroup, of banksii and zebrina/microcarpa ascendance, as 2 N gamete donor to Cavendish bananas; these studies also suggested that the N gamete was delivered by an AA diploid with a malaccensis affiliation [27, 31]. However, the hybridization events during banana domestication were far more frequent than previously expected, leading to over three contributors to Cavendish bananas. It has been reported that Cavendish bananas harbored mixed mosaic profiles with genomic regions of M. acuminata subspecies banksii, zebrina, and malaccensis, along with M. schizocarpa, and two unknown contributors [16]. As for ‘Baxijiao’, there existed low levels of sequence collinearity between any two haploid assemblies, consistent with the complex inter- or intra-specific hybrid origin of Cavendish bananas. Furthermore, the chromosomal rearrangements, the expanded/contracted gene families, as well as the homoeolog expression patterns also exhibited differences among the three haploid assemblies, further supporting the complex story of hybridization in triploid bananas.

In this study, we detected an expansion of gene families related to various processes that might affect fruit quality in ‘Baxijiao’, such as sucrose/disaccharide/oligosaccharide catabolic processes, sucrose metabolic process, starch metabolic process, and aromatic compound biosynthetic process. Soluble sugars (e.g. sucrose, glucose, and fructose) have a significant impact on fruit flavor and taste. Soluble sugars also provide precursors for the synthesis of other compounds associated with fruit quality such as amino acids, organic acids, and antioxidant micronutrients. In general, the sugar content of many modern fruits changed to a great extent due to continuous selection and breeding efforts [32, 33]. The sugar contents are quite low at the early post-harvest stage of banana. During fruit development, banana accumulates a large reserve of starch (15–35% w/w of their fresh weight). However, due to a complex metabolism of starch shifting from synthesis to breakdown, the starch content at late ripening decreases drastically and the soluble sugars accumulate [34]. This transition from starch to sugar appears to be accountable both for pulp sweetening and for providing energy to metabolic processes that determine other quality characteristics of ripe banana; for instance, synthesis of volatile compounds, color change of fruit peel, and even pulp softening, thereby playing important roles in final fruit quality [35]. In addition, aromatic compounds are essential for fruit quality because the aroma is a major indicator of fruit flavor. Banana fruits have a pleasant flavor. Typical flavor compounds are synthesized during banana ripening, forming unique aroma signatures of banana fruits [36]. The aromatic profiles of many banana cultivars have been broadly uncovered, indicating that the characteristic banana aroma was the result of a complex mixture of volatile compounds [37–39]. In this sense, the expanded biological processes associated with carbohydrate metabolism and aromatic compound biosynthesis in the ‘Baxijiao’ genome might be a different feature from its wild ancestors formed in its domestication process and play important roles in its fruit quality. We also observed an expansion of gene families associated with anther and pollen development in the ‘Baxijiao’ genome. The cultivar ‘Baxijiao’ is triploid and parthenocarpic, thereby producing fruits that are free of seeds, while M. acuminata has fruits containing numerous seeds making them inedible [40]. Relatively, according to our observation, ‘Baxijiao’ has thin anthers without pollens, but M. acuminata harbors anthers with larger size and full of pollens. Therefore, these expanded gene families in the ‘Baxijiao’ genome could be responsible for the reproductive difference between cultivated ‘Baxijiao’ and its wild ancestors. The anther wall tapetum is a layer of cells that provides a source of nutrition for the pollen grains as they mature [41, 42]. Here the expanded gene families were enriched in several biological processes related to anther wall tapetum, which were expected to play crucial roles in the parthenocarpic characteristic of ‘Baxijiao’. Some contracted gene families in the ‘Baxijiao’ genome were involved in the biological processes associated with response to external biotic and abiotic factors. Accumulating evidence indicates that banana cultivars have generally reduced resistance to pathogens and environmental stresses than their wild ancestors [43, 44]. These contracted gene families will provide candidates for further investigations on the genetic basis of the reducing resistance in banana cultivars.

Here we also investigated the homoeolog expression patterns in ‘Baxijiao’. On average, ~40% of homoeolog triads presented non-balanced expression patterns with higher (homoeolog-dominant) or lower expression (homoeolog-suppressed) from a single homoeolog respecting the other two, similar to the findings in other crops like Brassica napus (36.5%) and bread wheat (~30%) [45, 46]. Although the genes with homoeolog expression bias were not significantly enriched in any biological process, their distributions were not stochastic, with a preference to locate in the regions near chromosome ends that showed higher recombination rates than more centric sequences in flowering plants [47]. In addition, we found that the non-balanced triads showed subtly higher Ka/Ks values than the balanced triads, suggesting that the genes with homoeolog expression bias might be under more relaxed selection pressure and have slightly higher evolutionary rates.

Disease resistance and banana breeding

Plant pathogens continue to be a significant challenge to banana yields. Over recent decades, banana production has been severely affected by the soil-borne fungus Foc, which is the causal agent of the fusarium wilt of banana. The pathogen can be transmitted by plant material, water, and soil [48]. Although human societies have made many efforts to manage fusarium wilt of banana, none of these controls has been sufficiently effective for sustainable production of susceptible varieties in Foc-infested soils [49]. During past decades, Cavendish plantations have been threatened by tropical race 4 (Foc-TR4), a new strain of Fusarium. Recognition of chromosome regions that confer Foc resistance will facilitate marker-assisted selection in banana breeding schemes. R genes encode proteins that detect pathogens, thereby playing a key role in crops’ remarkable immune responses. Here we found that considerably fewer resistance genes were identified in ‘Baxijiao’ than in wild banana M. acuminata. Recently, Chen et al. [44] have identified a major quantitative trait loci (QTL)controlling resistance to the subtropical race 4 (STR4, affecting Cavendish bananas in the subtropics) of Foc in M. acuminata ssp. malaccensis. This QTL is located on the distal end of the long arm of chromosome 3 (36.20–42.50 Mb), corresponding to the regions including two NLR gene clusters (36.32–36.95 Mb and 40.22–40.98 Mb) found in the MAv4 genome [7]. Interestingly, we found that in the relative gene clusters in the ‘Baxijiao’ genome, nearly all NLR loci were located in the intergenic regions, indicating that they might be pseudogenized. Given this QTL-controlling resistance to STR4, the presence of much fewer NLR genes in these gene clusters could be responsible for Foc susceptibility of ‘Baxijiao’. This finding further confirmed the distal end of the long arm of chromosome 3 as an important region in mediating Foc resistance in banana. The pseudogenized NLR loci were also identified in the gene clusters in chromosome 10, indicating a possible involvement of these genomic regions in pathogen resistance too. Interestingly, the patterns of expressed NLR loci differed among the three haploid assemblies. For instance, nearly all NLR loci in a gene cluster in chromosome 3 remained functional in BXJ3, while the NLR loci in other gene clusters in the same chromosome were almost pseudogenized in BXJ1 and BXJ2; in addition, the NLR loci in the gene cluster in chromosome 10 were totally pseudogenized in BXJ2, but most loci remained functional in the other two haploid assemblies. These findings lend further support that the ‘Baxijiao’ subgenomes were of different origins, and suggested that they might contribute differently to disease resistance. The different patterns of expressed NLR loci will therefore serve as reliable references for accurate banana breeding programs in future. To summarize, the R gene clusters in chromosomes 3 and 10 might be potential targets to perform the molecular analysis of resistance in banana, which will require further research.

In conclusion, we present a high-quality T2T and haplotype-resolved reference genome of the Cavendish banana ‘Baxijiao’. The low nucleotide sequence collinearity and the differences among the three haploid assemblies closely mirror the complicated domestication history of the cultivar. Some enriched biological processes of ‘Baxijiao’-specific gene families were found to be related to fruit quality, parthenocarpy, and sterility of the cultivar. In addition, the presence of much fewer R genes in the ‘Baxijiao’ genome than in M. acuminata genome shed light on the possible mechanisms of fusarium wilt susceptibility of this cultivar. This genome assembly will provide a valuable reference for understanding domestication, molecular breeding and genetic improvement for banana.

Materials and methods

Plant materials, DNA extraction, and sequencing

Plant materials of Cavendish banana (‘Baxijiao’) were collected in South China Botanical Garden, Chinese Academy of Sciences. Tissue samples including fresh roots, leaves, flowers, and fruits were harvested, then they were immediately frozen in liquid nitrogen and preserved at −80°C for further DNA/RNA extraction. Using the CTAB method, high-quality genomic DNA was extracted from leaf tissue samples. The concentration and purity of the extracted DNA were evaluated using a Qubit fluorometer (Thermo Fisher Scientific). DNA integrity was checked by gel electrophoresis.

For PacBio HiFi sequencing, a standard SMRTbell library was constructed with SMRTbell Express Template Prep Kit 2.0 according to the manufacturer’s recommendations (Pacific Biosciences, CA, USA). The library was then sequenced on a PacBio Sequel II system. The Oxford Nanopore SQK-LSK109 kit was used to prepare the ONT library following the manufacturers’ instructions, then a Nanopore PromethION platform was used for sequencing. For Hi-C sequencing, the genomic DNA was cross-linked with formaldehyde and extracted for Hi-C library preparation. The library was paired-end sequenced on an Illumina NovaSeq platform (Illumina, CA, USA). For Illumina sequencing, paired-end 150-bp reads were generated on the Illumina NovaSeq platform. For RNAseq, total RNA was isolated from all tissue samples using the NEBNext^® Ultra^™ II Directional RNA Library Prep Kit for Illumina^® (New England Biolabs, MA, USA). Paired-end 150-bp reads were also generated on the Illumina NovaSeq platform. All sequencing was carried out at Anhui Double Helix Gene Technology Co., Ltd (Anhui, China).

Haplotype-resolved reference genome assembly and gap filling

Consensus reads (HiFi reads) were produced through CCS software (https://github.com/PacificBiosciences/ccs) with default parameters. Hi-C reads were filtered by fastp version 0.23.2 [50]. The obtained highly accurate HiFi reads were first used to estimate genome size and heterozygous rate with GenomeScope version 2.0 [18], based on K-mer counts produced by jellyfish [51]. Subsequently, HiFi reads and Hi-C reads were de novo assembled to generate the draft contig genomes of ‘Baxijiao’ using hifiasm v0.16.1 [19], which included two primary haplotypes: a first primary haploid assembly with smaller genome size and a larger one containing unphased contigs from two haploid genomes. Hi-C data were then used to anchor and remove some short contigs for the primary genome. Briefly, ragtag v2.1.0 [52] was first used to sort, orientate, and cluster the first haploid contigs guided by a T2T version of M. acuminata ssp. malaccensis genome (X. Liu et al., unpublished results). Meanwhile, the contigs were anchored into 11 pseudo-chromosomes by using Juicer v1.6 [53] and 3D-DNA v180922 [54] in turn. Besides, JuiceBox v2.20.00 [21] was introduced for visualizing Hi-C data and manual modification in order to obtain the first haploid assembly containing 11 pseudo-chromosomes. Then the same workflow combining ragtag, Juicer, 3D-DNA and JuiceBox was applied to the other primary haplotype with unphased contigs, and the redundant Hi-C signals were used to separate the second and the third haploid assemblies. The tools agp2assembly.py and 01_2assemblyto2.py in JuiceBox were then used to group the contigs from the three haploid assemblies together and arrange them in a sequential manner. Meanwhile, through examination on the final turn, the file containing the combined contigs and assemblies from all three haploid assemblies was introduced to JuiceBox, and some excess contigs belonging to the third haploid assembly could be manually adjusted according to Hi-C signals. Finally, we generated 33 pseudo-chromosomes based on our high-covered Hi-C data and high-accurate HiFi reads. For gap filling, the ONT ultra-long reads were first used for de novo assembly by NextDenovo (https://github.com/Nextomics/NextDenovo), and polished by Nextpolish [55] based on the HiFi reads and the Illumina reads with default parameters. Then the ONT polished contigs were mapped to the primary genome using minimap2 v2.24-r1122 [56]. Because there are only 144 gaps in the primary genome, we examined the breakpoint locations with the Integrative Genomics Viewer (IGV) [57] tool and manually filled the gaps according to the alignments. After filling gaps, a nearly complete and phased reference genome named BXJ was generated. The scheme showing ‘Baxijiao’ genome assembly is presented in Fig. S2 (see online supplementary material). Telomere regions were identified using TIDK v0.2.1 (https://github.com/tolkit/telomeric-identifier). Centromeric regions were identified based on the results of Tandem Repeats Finder v4.09 [20] using default parameters and TE libraries.

Assembly validation and quality assessment

HiFi and Illumina reads were used to evaluate the accuracy of BXJ. HiFi reads were mapped to the assembly using minimap2, while Illumina reads were aligned with BWA v0.7.17-r1188 [58]. The mapping rate was calculated using BamTools v2.5.1 [59]. The LTR Assembly Index (LAI) calculated from LTR_retriever v2.9.0 [60] was introduced to assess the genome assembly quality. BUSCO v5.4.3 [22] was used for the evaluation of assembly completeness of the haplotype-resolved genome with the ‘embryophyta_odb 10’ database.

Whole genome alignment and syntenic analysis

The syntenic relationships between haploid assemblies were determined using nucmer v4.0.0rc1 [61] with default parameters. Delta-filter was then launched with parameters ‘-i 90 -l 15000’. LINKVIEW v1.1 (https://github.com/YangJianshun/LINKVIEW) was used for visualizing syntenic relationships at chromosome level. In order to further check large inversion regions, we used IGV and JuiceBox to examine the breakpoint regions of inversions according to the validation methods mentioned in Tang et al. [62] and Zhou et al. [63] (Figs S8–S9, see online supplementary material). MCscan [23] pipeline was used to find structure variations among the BXJ haploid assemblies. Briefly, the ‘jcvi.compara.catalog’ module with parameter ‘—cscore = 0.99’ was performed to find ortholog; then, ‘jcvi.compara.synteny’ module with ‘—minspan = 30’ was used to build the syntenic regions; finally, ‘jcvi.graphics.karyotype’ module was carried out to visualize the syntenic relationships.

Genome repeats and gene annotation

TEs libraries were identified by RepeatModeler v2.0.1 [64] with default parameters. RepeatMasker v4.1.1 (http://repeatmasker.org/) was used to identify repeats with the consensus TEs libraries. Then the repeat-masked genome was imported to MAKER3 v3.01.03 [65] for gene annotation. All high-confidence Swiss-prot protein sequences were imported for homology prediction by Protein2Genome with parameters protein2genome = 1. Transcripts from the four tissues were used by Est2Genome with parameters est2genome = 1, then the ab-initio prediction model was constructed by AUGUSTUSv3.3.2. AUGUSTUS models were tested based on two divided datasets, one for test and another for evaluation, and results showed our models were sufficiently reliable. Finally, the MAKER3 pipeline was run again to obtain high-quality gene annotations combining homology predictions, transcripts evidence predictions, and de novo predictions. Funannotate v1.8.7 (https://github.com/nextgenusfs/funannotate) was used to functionally annotate genes based on blast results from these databases: Pfam, UniProt, GO, KEGG, InterPro, dbCAN, EggNOG, and MEROPS.

Gene family expansion and contraction

Genes families in ‘Baxijiao’ and 15 other species in monocots were characterized using Orthofinder v2.5.4 [66] by comparing their protein-coding gene sequences. The single-copy ortholog sequences, which strictly existed in all above species, were then aligned using mafft v7.310 [67]. Subsequently, the concatenated sequences were subjected to phylogenetic analysis using RAxML v8.2.12 [68] with 2000 bootstraps. Once the phylogenetic tree was constructed, MCMCTree [69] was applied to estimate divergence times and a total of nine species pairs were selected as calibration points, whose estimated divergence times were acquired from the TimeTree website (Table S17, see online supplementary material). Two MCMCTree runs were performed to ensure the convergence of the divergence time estimation. CAFE v5.0 [70] was carried out to identify potential expanded and contracted gene families. For significantly expanded and contracted gene sets, GO enrichment analysis was conducted using TBtools v1.108 [71].

Identification of resistance genes

Heritable genetic variation for disease resistance in crop species is often affected by disease resistance (R) genes. RGAugury v2.2 [25] was used to predict resistance gene analogs (RGAs) in ‘Baxijiao’ and M. acuminata. The search for NLR loci was carried out by NLR-Annotator [26], a tool for de novo genome annotation of NLR loci independent of transcript support. A potential NLR locus might correspond to a complete or partial gene and might also be the trace of a pseudogenized sequence.

Homoeolog expression patterns in ‘Baxijiao’

After filtering by fastp, RNAseq reads were aligned to the haplotype-resolved BXJ genome using HISAT2 v2.2.1 [72]. featureCounts v2.0.4 [73] was then used to quantify RNAseq data as counts. To identify alleles in the BXJ genome, we applied DIAMOND v2.0.15 [74] to align the protein sequences among ‘Baxijiao’ haploid assemblies and MAv4, and obtain three types of alleles: three alleles paired with each other (1:1:1 correspondence), two alleles paired with each other, and the remaining, based on the identity scores. To explore patterns of homoeolog expression, we focused on the gene triads that had a 1:1:1 correspondence across the three haploid assemblies. A homoeolog with expression level >0.5 transcripts per million (TPM) in at least one tissue was considered expressed. Relative expression levels of the BXJ1, BXJ2, and BXJ3 homoeologs across triads were standardized following Ramírez-González et al. [46], based on the normalized absolute TPM and the relative expression level of each homoeolog was normalized as expression_BXJ1 = TPM(BXJ1) / (TPM(BXJ1) + TPM(BXJ2) + TPM(BXJ3)), expression_BXJ2 = TPM(BXJ2) / (TPM(BXJ1) + TPM(BXJ2) + TPM(BXJ3)), and expression_BXJ3 = TPM(BXJ3) / (TPM(BXJ1) + TPM(BXJ2) + TPM(BXJ3)). The relative expression level of each homoeolog per triad was introduced to plot the ternary diagrams using the R package ggtern [75]. Based on these plots, we described seven groups of homoeolog expression bias: a balanced group, with similar relative expression levels from the three homoeologs, and six homoeolog-dominant or homoeolog-suppressed groups, defined on the basis of the higher or lower relative expression levels from a single homoeolog compared to the other two (Table S18, see online supplementary material). KAKS Calculator v2.0 [76] was used to calculate Ka/Ks values of different expression bias categories. For homoeolog-dominant or homoeolog-suppressed genes, GO enrichment analyses were carried out using TBtools.

Author contributions

Y.Z. and X.J.G. designed and supervised the research; H.R.H., X.L., and R.A. wrote the manuscript; H.R.H., X.L., and X.W. analysed the data; H.R.H. and W.M.L. collected the experimental materials. All authors contributed to manuscript revision, read and approved the submitted version.

Data availability

The Cavendish (BXJ) raw genome sequencing reads are available from the National Center for Biotechnology Information (NCBI) under project ID PRJNA957115 and the National Genomics Data Center (NGDC) under project ID PRJCA016940. The Cavendish assembly and annotation files have been submitted to FigShare [77].

Conflict of interest statement

The authors report no conflict of interest.

Supplementary data

Supplementary data is available at Horticulture Research online.