Reconstructing the genome sequence of D. tokoro using genome assembly and linkage analysis

We reconstructed a chromosome-scale, haplotype-resolved genome assembly from female and male individuals of D. tokoro using whole-genome sequencing reads obtained using the Oxford Nanopore Technologies (ONT) and Illumina sequencing platforms (Fig 1E). The genome size of D. tokoro was estimated to be 388 Mb by flow cytometry analysis (S2 Fig). A previous linkage study suggested that this species follows an XY/XX male heterogametic sex-determination system [34]. Therefore, we generated separate assemblies from the filtered ONT reads for haplotypes 1 and 2 of a female individual (Waka1) collected from Tahara, Wakayama Pref., Japan, presumably with XX genotype. We also obtained separate assemblies using filtered ONT reads for haplotypes 1 and 2 of a male individual (Kita1) collected from Kitakami, Iwate Pref., Japan, presumably with XY genotype. For Waka1 individual (female), we obtained a haplotype 1 assembly with 212 contigs (425.1 Mb in total) and a haplotype 2 assembly with 346 contigs (342.2 Mb in total) after polishing the draft assemblies using Illumina reads (S1 Table). Likewise, for Kita1 individual (male), we obtained a haplotype 1 assembly with 172 contigs (414.4 Mb in total) and a haplotype 2 assembly with 440 contigs (300.4 Mb in total). Notably, the Waka1 haplotype 1 contained two telomere-to-telomere contig assemblies, and the Kita1 haplotype 1 contained six telomere-to-telomere contig assemblies. BUSCO analysis [56] showed that the percentages of complete embryophyte BUSCOs were 98.2% in the Waka1 haplotype 1 assembly, 81.9% in the Waka1 haplotype 2 assembly, 98.0% in the Kita1 haplotype 1assembly, and 76.0% in the Kita1 haplotype 2 assembly (S1 Table). The relatively low sizes of the haplotype 2 assemblies of Waka1 and Kita1 were probably caused by using the PECAT program, which prioritizes the assembly of haplotype 1; similar biases were previously reported [57].

To anchor the assembled sequences to chromosomes by linkage analysis, we used a pseudo-testcross approach [58] (Fig 1F), a method to address the linkage of two DNA markers, both of which are heterozygous in one parent and homozygous in another parent, by testing their co-segregation or independent segregation in the F₁ progeny population (S3 Fig). We obtained 186 F₁ progeny from a cross between Waka1 (female) and Kita1 (male) (S4 Fig) and genotyped these progeny and the two parent plants by restriction site–associated DNA sequencing (RAD-seq) [59]. We selected SNP and presence/absence polymorphism (PA) markers that were heterozygous in the female parent and homozygous in the male parent as female-parent-heterozygous markers (S5 Fig). We also obtained SNP and PA markers that were homozygous in the female parent and heterozygous in the male parent as male-parent-heterozygous markers. Using the pseudo-testcross scheme, we constructed two separate linkage maps using the female-parent-heterozygous markers and the male-parent-heterozygous markers (Figs 1G, 1H; S6, S7, S8, and S9). Finally, we combined the two linkage groups using the shared assemblies between the two linkage maps and generated pseudo-chromosomes 1–10 (S10, S11, S12, and S13 Figs); for simplicity, we refer to the pseudo-chromosomes as chromosomes hereafter. The chromosome-assigned sequences using the Waka1 (female) haplotype 1 and Kita1 (male) haplotype 1 references appears to be close to complete, with 10 chromosomes containing only 24 and 18 contigs, respectively (S1 Table).

To annotate the gene models, we obtained RNA-seq data from 18 different organs of D. tokoro and used protein homology information for D. alata TDa95/00328 and D. rotundata TDr96_F1. Based on the combination of transcriptome-based gene identification and ab initio gene prediction, we identified 21,308 genes for the Waka1 (female) haplotype 1 assembly, 17,710 genes for the Waka1 (female) haplotype 2 assembly, 23,200 genes for the Kita1 (male) haplotype 1 assembly, and 18,145 genes for the Kita1 (male) haplotype 2 assembly (S2 Table). Chromosomal synteny analysis based on collinear blocks with predicted genes revealed high collinearity between the Kita1 (male) haplotype 1–based chromosomes with the largest number of telomere-to-telomere assemblies and the chromosomes inferred using the three other reference genomes (Fig 1I). We determined that the chromosome-scale genome assemblies obtained from female and male individuals are useful to infer the structures of X- and Y-specific genomic regions.

Linkage study and mapping coverage analysis of Illumina short reads identify the X- and Y-specific genomic regions on chromosome 3

To identify the genomic region linked to the sex phenotype, we performed association analysis using the 186 F₁ progeny, comprising 38 females, 89 males, and 59 non-flowering individuals. Based on RAD-seq data from the 127 flowering F₁ progeny, we examined the association between the genotypes and sex phenotype of the F₁ individuals using Fisher’s exact tests (Figs 2A and S14). The log-transformed q-values (–log₁₀(q)) revealed significant associations between sex phenotype and the male-parent-heterozygous markers in the middle of chromosome 3 but did not detect any association using the female-parent-heterozygous markers (Figs 2B, 2C, and S15). These associations were also confirmed by a Fisher’s exact test using the same number of females and males randomly selected from the F₁ progeny (S16 Fig). Linkage analysis using simple interval mapping performed with R/qtl [60] also detected linkage between sex phenotype and the male-parent-heterozygous markers on chromosome 3, whereas no linkage was detected with the female-parent-heterozygous markers (S17 Fig). These results suggest that the sex of D. tokoro is determined by components on chromosome 3 and that the species has a male heterogametic sex-determination system of XY/XX.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. The dioecious wild yam D. tokoro contains X- and Y-specific regions on chromosome 3.

(A) Schematic diagram of association analysis to identify genomic regions associated with sex phenotype. Single-nucleotide polymorphism (SNP) and presence/absence (PA) markers were detected in 38 females and 89 males in the F₁ progeny. (B) Manhattan plots of –log₁₀(q) values between SNP-type markers and sex phenotype based on the Kita1 (male) haplotype 1 assembly. The –log₁₀(q) values were obtained by Fisher’s exact test. The upper and lower plots show male-parent-heterozygous and female-parent-heterozygous markers, respectively. The yellow points in the upper plot indicate SNP-type markers significantly associated with sex phenotype. A significance threshold of 5% false discovery rate was adjusted by Benjamini-Hochberg correction: adjusted threshold = 1.31 for plots of male-parent-heterozygous markers and adjusted threshold = 2.00 for plots of female-parent-heterozygous markers. (C) Manhattan plots of –log₁₀(q) values between PA markers and sex phenotype based on the Kita1 (male) haplotype 1 assembly. The red and blue plots show PA markers significantly associated with sex phenotype. The male-parent-heterozygous markers in the middle of chromosome 3 show significant associations. (D) Alignment-based dot plot analysis of chromosome 3 between the Waka1 (female) haplotype 1 and the Kita1 (male) haplotype 1 showed low sequence matches in the middle of the chromosome. Blue arrows in the left and red arrows in the bottom indicate positions of the contigs for Kita1 (male) haplotype 1 and Waka1 (female) haplotype 1, respectively. Waka1 (female) haplotype 1 and Kita1 (male) haplotype 1 assemblies corresponding to chromosome 3 contained only three and two contigs, respectively, with the mismatch region located in the middle of contigs in both assemblies (Figs C and D in S1 Text). The inset panel shows a zoomed-in alignment of 20-30 Mb region of the chromosomes, including the mismatch region. The alignment-based dot plots were generated using D-Genies, a software performing large-scale genome alignment visualized as dot plots [131]. Chromosome 3 of Kita1 (male) haplotype 1 was used as the target sequence, and chromosome 3 of Waka1 (female) haplotype 1 was used as the query sequence. The full plot was generated with noise filtering, and the inset panel was generated without noise filtering, using the D-Genies algorithm that removes small matches based on the alignment size and frequency. (E) Mapping coverage analysis to differentiate between putative X- and Y-specific regions. The bar graphs indicate female and male mapping depth, obtained by mapping Waka1 (female) and Kita1 (male) Illumina sequences onto the haplotype 1 assembly from Waka1 (female). Coverage analysis revealed putative X-specific regions of chromosome 3, shown as red regions below the bar graphs. Waka1 haplotype 1 chromosome 3 was defined as the X chromosome. (F) Bar graphs showing female and male mapping depth, obtained by aligning Waka1 (female) and Kita1 (male) Illumina sequences onto the haplotype 1 assembly from Kita1 (male). Coverage analysis revealed putative Y-specific regions on chromosome 3, shown as blue regions below the bar graphs. Kita1 haplotype 1 chromosome 3 was defined as the Y chromosome. (G) Positions of XY gametologs on the putative X- and Y- specific regions. The illustrations of D. tokoro were obtained with TogoTV (https://doi.org/10.7875/togopic.2022.552 and https://doi.org/10.7875/togopic.2022.453; copyright 2016 DBCLS TogoTV/ CC-BY-4.0).

https://doi.org/10.1371/journal.pgen.1012123.g002

In the genomic region associated with sex phenotype, we detected a higher concentration of presence/absence polymorphism (PA) markers (358 markers in the 10-Mbp region around the marker showing the peak association value) than SNP markers (62 markers in the 10-Mbp region) (Fig 2B and 2C), indicating structural differences in the middle of chromosome 3. Alignment-based dot plot analysis of chromosome 3 between the female and male haplotype 1 assemblies detected a ~ 5 Mb mismatch in the middle of the chromosome (Fig 2D), confirming the differentiation of chromosome 3. The chromosome 3 assemblies corresponding to the female haplotype 1 and the male haplotype 1 contained only three and two contigs, respectively, with the mismatch region located in the middle of contigs in both assemblies (Figs 2D, 2C, and 2D in S1 Text), indicating that the observed structural differentiation is reliable. Therefore, we further examined the structural differences in this region by performing coverage analysis of Illumina short-read sequences of the male and female genomes mapped to the reference genome sequences. In diploid plants with male heterogametic sex determination (XY) and substantial structural differentiation between X and Y chromosomes, the Y-specific region is predicted to be absent from females. Furthermore, the coverage of the male X-specific region is thought to be one-half that of females (hemizygous) [61,62]. To assess whether this is the case for D. tokoro, we aligned Illumina short reads obtained from female and male individuals to each haplotype of chromosome 3 and studied the mapping depth. As the reference genome sequences for mapping, we used four different chromosome 3 assemblies: the Waka1 (female) haplotype 1 and 2 assemblies and the Kita1 (male) haplotype 1 and 2 assemblies. We identified a region in the middle of chromosome 3 that was enriched for sites covered by male reads at 0.5 × depth and female reads at 1 × depth in the Waka1 (female) haplotype 1 assembly (Fig 2E). Such consistent depth differences between male and female reads were not observed in other chromosomes (S18 Fig). When we mapped the Illumina short reads to the Kita1 (male) haplotype 1 assembly, we identified a region in the middle of chromosome 3 that was enriched for sites covered by male reads at 0.5 × depth and female reads at 0 × depth (Fig 2F). Such depth differences between male and female reads were not observed in other chromosomes (S19 Fig). We also observed such differences in depth between male and female reads when male and female plants collected from three different locations were used for mapping (S20 and S21 Figs). In contrast, we did not detect regions covered by male reads at 0.5 × depth and female reads at 1 × depth in the center of chromosome 3 of the Waka1 (female) haplotype 2 (S22A Fig), possibly due to incomplete assembly of haplotype 2. We detected a region covered by male reads at 0.5 × depth and female reads at 1 × depth in the Kita1 (male) haplotype 2 assembly (S22B Fig).

To confirm the putative SDRs identified by the association study and mapping coverage analysis, we applied two previously reported methods: SDpop, which detects SDRs by modelling the allele and genotyping frequencies for different segregation types [63], and RADSex, which detects SDRs based on the association of RAD markers with male and female individuals [64]. For SDpop analysis, we used Illumina shot read sequencies from five females and five males of D. tokoro collected from a natural population in Wakayama prefecture, central Japan. Based on the alignments of these sequencies to each haplotype reference, empirical Bayes posterior probabilities in the SDpop model inferred three segregation types at each polymorphic site: autosomal segregation type, X-hemizygosity segregation type and XY gametology segregation type. High posterior probabilities of XY gametology segregation type and low posterior probabilities of autosomal segregation type were detected in the middle of chromosome 3 of both the Waka1 (female) haplotype 1 assembly and the Kita1 (male) haplotype 1 assembly. These patterns suggest the presence of SDRs at positions consistent with the X- specific and Y-specific regions identified by the mapping coverage analysis (S23A and S23B Fig). For RADSex analysis, we used RAD markers of 127 F₁ flowering progeny from a cross between Waka1 (female) and Kita1 (male). The probability of association with sex, -log₁₀(p), was calculated for each marker by a Pearson’s chi-squared test between marker and sex phenotype in RADSex. Sex-associated markers with high -log₁₀(p) values were detected in chromosome 3 of both the Kita1 (male) haplotype 1 assembly (S24A Fig) and the Waka1 (female) haplotype 1 assembly (S24B Fig). These regions corresponded to the SDRs identified by the mapping coverage analysis (S24C and S24D Fig).

Based on the results of our association analysis (Fig 2B and 2C), alignment-based dot plot analysis (Fig 2D), mapping coverage analysis (Fig 2E and 2F) and the SDR detection analyses (S23 and S24 Figs), we define chromosome 3 of the Waka1 (female) haplotype 1 assembly as the X chromosome and chromosome 3 of the Kita1 (male) haplotype 1 assembly as the Y chromosome. These chromosomes are differentiated into putative X- and Y-specific regions spanning ~4.8 Mb with 54 candidate genes (S3 Table) and ~4.5 Mb with 56 candidate genes (Table 1), respectively. These candidate genes include X- or Y-specific genes and putative orthologous genes (gametologs; [45,46]). A total of 22 genes in the Y- region had corresponding genes in the X-region so that they were defined as XY gametologs (Fig 2G).

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Differential expression analysis of genes located in Y-specific regions.

https://doi.org/10.1371/journal.pgen.1012123.t001

X- and Y-specific regions are likely located in the pericentromeric regions

Recombination suppression is a key step in SDR evolution. This suppression is caused by various mechanisms, such as chromosome rearrangements including inversions and transpositions [6,65,66]. Recombination suppression has also been reported even in the absence of structural rearrangements, accompanied by gene loss and accumulation of repetitive DNA sequences [67,68]. Accumulation of repetitive DNA in the non-recombining regions (NRY) surrounding the SDR are reported as a conspicuous feature of sex chromosomes [69–73]. In D. tokoro, the alignment-based dot plot analysis showed no major inversion between the X and Y chromosomes (Fig 2D). The X- and Y-specific regions revealed to be located in a genomic region with lower gene density and higher density of repetitive DNA sequences, represented by Ty3 long terminal repeats (LTRs) (Fig 3A and 3B). Co-localization of X- and Y-specific regions to the gene-paucity and repetitive DNA-rich regions suggests that D. tokoro X- and Y-specific regions likely correspond to the pericentromeric regions (near the centromere) of chromosome 3.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. X- and Y-specific regions locate in putative peri-centromeric regions.

(A) Repetitive DNA sequences accumulate in X-specific regions on chromosome X (chromosome 3 of Waka1 haplotype 1 assembly), suggesting that these regions are located in the peri-centromeric regions. The bar graphs indicate gene and repeat density on chromosome 3. The heatmap indicates gene and Ty3 retrotransposon density. Distributions of genes and repeat densities along the chromosomes were calculated by a sliding window analysis (window size = 500 kb and step size = 100 kb). For each window, the percentage bp coverage of the gene or repetitive element sequence was calculated with the number of non-N basepairs as denominators. Ty3 element belongs to “Repeats” category. (B) Repetitive DNA sequences accumulate in Y-specific regions of chromosome Y (chromosome 3 of Kita1 haplotype 1 assembly), suggesting that these regions are located in pericentromeric regions.

https://doi.org/10.1371/journal.pgen.1012123.g003

We also evaluated the recombination rate around the X- and Y-specific regions by comparing the relationship between the linkage distances (in centimorgans, cM) and physical distances (in base pairs, bp) (S10 and S12 Figs). Linkage distance, recombination frequency between a pair of DNA markers on a chromosome, was obtained based on the segregation of RAD markers among the F₁ progeny. Although we observed a small reduction in linkage distances as compared to physical distances in the center of chromosome 3, the pattern of relationship between linkage and physical distances was not greatly different from those of other chromosomes (S10 and S12 Figs). This suggests that the NRYs, even if exist, may be restricted to the small X- and Y-specific regions near the centromere with suppressed recombination, while most part of chromosome 3 constitutes pseudoautosomal regions (PARs) undergoing recombination. Based on these observations, we suggest that centromere-mediated recombination suppression has contributed to the evolution of D. tokoro X- and Y-specific regions possibly located in the vicinity of the centromere.

Most Dioscorea species are dioecious. However, it remains unclear whether the identified sex determination system, involving X- and Y-specific regions, is ancient and predates species divergence. Therefore, we calculated the distribution of synonymous site divergence (d_S) between the putative orthologous genes located in X- and Y-specific regions to test whether the X–Y differentiation is relatively ancient or recent compared with species divergence within the genus. We used all predicted genes (23,200 genes) with complete coding sequences from D. tokoro Kita1 haplotype 1 assembly as the reference, and obtained (1) 76,090 orthologous or paralogous gene pairs by comparison with the Waka1 haplotype 1 assembly of D. tokoro, (2) 61,220 gene pairs by comparison with D. alata genome, (3) 75,576 gene pairs by comparison with D. rotundata genome. The d_S values between possible orthologous gene pairs located on X- and Y-specific regions (XY gametologs) were compared to the three estimates of d_S values: (1) D. tokoro intraspecific d_S values; d_S values of all orthologous and paralogous gene pairs between the Kita1 haplotype 1 reference genome and the Waka1 haplotype 1 reference genome, (2) interspecific d_S of D. tokoro–D. rotundata, (3) interspecific d_S of D. tokoro–D. alata (S25A Fig). The median d_S value of 22 XY gametologs was 0.019 (S25B and S25C Fig), which is comparable to the median of D. tokoro intraspecific d_S values, 0.015. In contrast, it is > 20-fold lower than the median divergence between D. tokoro and D. rotunada (median of d_S value: 0.49) or D. tokoro and D. alata (median of d_S value: 0.48), (S25A Fig), suggesting that the divergence time between Y- and X- located genes are not as old as the divergence time between the D. tokoro and two other Dioscorea species.

In X–Y systems, divergence between homologous X- and Y-linked genes typically increases with distance from PARs, forming a stepwise pattern referred to as evolutionary strata. In D. tokoro, we also examined the presence of evolutionary strata, which are observed as groups of X–Y gene pairs showing distinct levels of synonymous divergence. We applied a threshold of d_S < 0.5 to filter out potential paralogous X–Y gene pairs with excessively large d_S values, based on the maximum d_S values observed in SDRs in previous studies [22,51]. We also evaluated it with a less stringent threshold of d_S < 1.0. The distribution of d_S values of X–Y gene pairs in D. tokoro showed no clear evolutionary strata throughout the X and Y chromosomes (S26 Fig); we could not detect any signature of successive extension of recombination suppression following the X–Y differentiation in D. tokoro.

Finally, we conducted an interspecific comparison of SDRs among the Dioscorea species (Fig 4). In the genus Dioscorea, multiple sex-determination systems, including XY/XX and ZZ/ZW systems, have been reported and the different systems are scattered across its phylogeny (Fig 4A). We performed synteny analysis between D. tokoro and two other Dioscorea species: D. alata, which has an XY/XX system, and D. rotundata, which has a ZZ/ZW system. The 20–30 Mb region of sex chromosomes (chromosome 3) of D. tokoro, including the SDRs, were syntenic with the sex chromosome (chromosome 6; [35,44]) and chromosome 13 of D. alata (Fig 4B). In contrast, the SDR of D. rotundata is located in the terminal region of its sex chromosome (chromosome 11; [27]), and the sex chromosome shows no collinearity with the sex chromosomes of either D. tokoro or D. alata (Fig 4B). These results suggest transitions of the sex-determination system and SDRs within the genus Dioscorea (Fig 4C).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Possible transitions of sex chromosomes in the genus Dioscorea.

(A) Distributions of sex determination systems on a phylogenetic tree of 183 species of Dioscorea. Each of the sections Stenophora and Enantiophyllum contains species with XY/XX and ZZ/ZW systems. The phylogenetic tree was adopted from [41]. D. rotundata was not included in the phylogenetic analysis of Dioscorea in [41], but it belongs to the section Enantiophyllum [147]. (B) Gene order–based synteny analysis of chromosomes of D. tokoro (male haplotype 1), D. alata [30] and D. rotundata [29]. Blue-colored bands between D. tokoro and D. alata indicate synteny blocks spanning 20-30 Mb of chromosome 3 (male haplotype 1) in D. tokoro, including the SDRs, and their corresponding regions in D. alata. Blue-colored bands between D. alata and D. rotundata indicate synteny blocks of chromosomes 6 and 13 in D. alata and their corresponding regions in D. rotundata. Red-colored bands indicate synteny blocks of chromosome 11 in D. rotundata and the corresponding regions in two other species. (C) Comparative schematic representation of sex chromosomes in three Dioscorea species.

https://doi.org/10.1371/journal.pgen.1012123.g004

The Y-specific genes BLH9 and HSP90 are specifically expressed during early male flower development

The male heterogametic sex determination (XY) in D. tokoro suggests that Y-specific genes or miRNAs function in sex determination. To identify the candidate genes or miRNAs involved in sex determination, we conducted transcriptome analysis via RNA-seq and small RNA-seq, given that small RNAs have also been reported to play roles in sex determination [12,74]. In D. tokoro, male and female flowers show sex-specific phenotypes after the bud stage (Fig 5A and 5B), indicating that the determination of sex phenotypes takes place before this stage. Therefore, we obtained RNA-seq and small RNA-seq data from male and female flowers at five stages of development as well as non-reproductive organs (Fig 5C and 5D). We mapped the RNA-seq reads to the Kita1 (male) haplotype 1 assembly, which includes the Y chromosome.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. The candidate genes BLH9 and HSP90 are located on Y-specific regions and are expressed at high levels during early male flower development.

(A) Five stages of male flower development used for RNA-seq and small RNA-seq analyses. (B) Five stages of female flower development. (C) Non-reproductive parts of the plant, including the shoot apex, leaf, stem, and rhizome. The shoot apex is indicated with ellipses. (D) Parts of the rhizome, including the rhizome stem, apex, root, bud, and storage region. Each rhizome region is indicated with ellipses. (E) Heatmap of the expression levels of genes in the Y-specific region. The boxes under the heatmap show the DESeq2 results for each gene. The gray boxes indicate p-value < 0.05 based on differential expression testing by negative binomial generalized linear model or log₂FC > 2 based on the log₂ ratio of the mean of normalized counts in each group in two comparisons: male (S0, S1, S2) vs. female (S0, S1, S2) and male (S0, S1, S2) vs. non-reproductive organs. BLH9 and HSP90, highlighted by rectangles in the heatmap, are highly expressed during early stages of male flower development. (F) Coverage of short reads from male and female individuals on BLH9. BLH9 was covered only by male short reads. (G) Relative expression levels of BLH9 in male and female flowers at five stages of development and in non-reproductive organs. (H) Coverage of short reads on HSP90 from male and female individuals. HSP90 was covered only by male short reads. (I) Relative expression levels of HSP90. All expression levels are shown as normalized transcripts per million (TPM) values.

https://doi.org/10.1371/journal.pgen.1012123.g005

The Y-specific regions contained 56 potential protein-coding genes (Fig 5E and Table 1) and one primary miRNA (No. 20 in S4 Table). Of these, 14 genes (genes 1–14 in Fig 5E) were Y-specific and more highly expressed during the three stages of early male vs. female flower development based on the hypothesis test for differential expression of the RNA-seq data using negative binomial generalized linear models (p-value < 0.05), whereas no primary miRNA showed differential expression between male and female flowers. A sequence similarity search with BLASTX revealed that the 14 genes included ten genes encoding hypothetical proteins without functional information and four genes encoding proteins with functional information (Table 1).

We also compared the expression levels of genes between male and female flowers based on the log₂ ratio of the mean of normalized counts (log₂[fold change], log₂FC). Within the 14 genes identified by differential expression analysis described above, we identified ten genes (genes 1, 4–10, 12, and 13 in Fig 5E) that showed more than 4-fold higher expression (log₂FC > 2) during the early stages of male vs. female flower development. The ten genes included two genes encoding proteins with functional annotations: A. thaliana BEL1-LIKE HOMEODOMAIN PROTEIN 9 (AtBLH9) and rice (Oryza sativa) HEAT SHOCK PROTEIN 81–1 (HSP81–1). The eight remaining genes encode hypothetical proteins without functional information.

When we compared the expression levels of the ten genes on the Y-specific region between male flowers and non-reproductive organs, only two genes (genes 1 and 5 in Fig 5E) had higher expression in male flowers than in non-reproductive organs (p-value < 0.05), and both these genes showed significantly higher expression in male flowers based on normalized counts (log₂FC > 2). Thus, we identified two Y chromosome–specific genes showing significantly higher expression in male flowers during the three stages of early development (p-value < 0.05 and log₂FC > 2) compared to female flowers and non-reproductive organs. The first gene, named BLH9, is a male-specific gene covered only by male-specific short-read sequences (Fig 5F) that is upregulated in male flowers during the three stages of early development (Fig 5G). A sequence similarity search by BLASTX showed that BLH9 is closely related to the A. thaliana homeobox gene AtBLH9, which belongs to the BELL family within the TALE (three-amino-loop-extension) superfamily (S5 Table). In the phylogenetic tree including all putative homologues of the TALE superfamily of D. tokoro, BLH9 was placed in the same clade as AtBLH9 together with a D. tokoro gene, DtKita1h1 03399.t1.p1, located on chromosome 1 (S27 Fig). The second gene, HSP90, is also male specific (Fig 5H) and was upregulated in male flowers during the three stages of early development (Fig 5I). HSP90 shares similarity with the A. thaliana chaperone gene HEAT SHOCK PROTEIN 90–4 (AtHSP90.4) (S5 Table). In the phylogenetic tree including all putative homologues of the HSP90 protein family of D. tokoro and A. thaliana, HSP90 showed a similarity to AtHSP90.1–AtHSP90.4, but its DNA sequence appeared to have diverged from other HSP genes of D. tokoro (S28 Fig). The male specificity of both genes was confirmed by PCR amplification of the DNA fragment using gene-specific primers and DNA templates from individuals of two natural D. tokoro populations: the HNMK population (northern Japan) and the KMMT population (southern Japan) (S29 Fig). In summary, The Y-specific genes BLH9 and HSP90, specifically expressed during early male flower development, were identified as the primary candidate sex-determining genes.

BLH9 shares similar functions with AtBLH9 and suppresses fruit development in A. thaliana

A genetic transformation method for D. tokoro has not yet been established. Among plants in the genus Dioscorea, successful transformation has been reported only in D. rotundata [75], which belongs to a section (Sect. Enantiophyllum) distant from D. tokoro (Sect. Stenophora) and possesses a ZZ/ZW sex-determination system [27]. Therefore, it is difficult to validate the functions of D. tokoro Y-specific genes by transformation of Dioscorea species.

To explore the biological role of the candidate gene BLH9, we overexpressed this gene and AtBLH9 in A. thaliana. Phylogenetic analysis of D. tokoro BLH9 and putative homologues of BLH9 from monocotyledonous species (Asparagus officinalis, Dioscorea alata, D. rotundata, Oryza sativa and Phoenix dactylifera) as well as A. thaliana TALE (three-amino-loop-extension) superfamily proteins showed that AtBLH9 shared the greatest amino acid sequence similarity with BLH9 and its putative homologues (Fig 6A). AtBLH9 is involved in inflorescence architecture and fruit development, as overexpression or knockout of AtBLH9 led to abnormal inflorescence development, including reduced inflorescence height and shorter fruit length, as well as irregular internode elongation, with extremely short and long internodes, respectively [76–81]. To compare the biological roles of AtBLH9 and BLH9, we overexpressed AtBLH9 and BLH9 in A. thaliana plants driven by the CaMV35S promoter. For phenotypic evaluation, we used T₂ plants transformed with a construct expressing BLH9 or AtBLH9. Wild-type A. thaliana Col-0 plants were used as a control.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. The D. tokoro Y-specific gene BLH9 plays similar roles to AtBLH9 in inflorescence development.

(A) Phylogenetic tree of BLH9 from D. tokoro, putative homologues of BLH9 from monocotyledonous species (Asparagus officinalis, Dioscorea alata, Dioscorea rotundata, Oryza sativa and Phoenix dactylifera), and TALE superfamily proteins from A. thaliana. The circles in the tree nodes indicate bootstrap values. Tree scale indicates amino acid substitutions per site. (B) Whole-plant phenotypes of Col-0, control (Cont.) and overexpression (OX) lines of BLH9, and control (Cont.) and overexpression (OX) lines of AtBLH9. Relative expression levels were calculated using the 2^{− ΔΔCT} method, with gene expression in the overexpression lines (OX) defined as 2^{− ΔΔCT} > 2 and low expressing lines defined as 2^{− ΔΔCT} ≤ 2 (S31 Fig). The low expressing lines were used as control (Cont.). (C) Main inflorescence heights of the five lines. Red plus symbol, average; center line, median; box limits, upper and lower quartiles; whiskers, 1.5x interquartile range; points, each measurement. (D) Inflorescence of Col-0. White arrow indicates an internode. (E) Inflorescence of the BLH9 overexpression line. (F) Mean fruit lengths of the five lines. (G) Variation in internode length of the five lines.

https://doi.org/10.1371/journal.pgen.1012123.g006

We evaluated the relative expression levels of BLH9 and AtBLH9 in all T₂ plants by RT-qPCR using the 2^−ΔΔCT method (S30 Fig). The T₂ plants were categorized into low-expressing (2^{− ΔΔCT} ≤ 2; indicated as Cont.) and overexpressing (2^{− ΔΔCT} > 2; indicated as OX) groups. Compared to Col-0 plants and the low-expressing T₂ plants (BLH9 Cont. and AtBLH9 Cont.), the T₂ overexpressing plants (BLH9 OX and AtBLH9 OX) showed shorter average main inflorescence height, but the differences between BLH9 OX and Col-0 plants and that between BLH9 OX and BLH9 Cont. were not statistically significant (Fig 6B and 6C). BLH9 OX had shorter fruits than the Col-0 plants and the BLH9 Cont. (Fig 6D, 6E, and 6F). We also evaluated irregular internode elongation by examining variation in internode length in the main inflorescence. A large variation in internode length was observed in AtBLH9 OX (Fig 6G) but not in BLH9 OX. These results suggest that BLH9 from D. tokoro shares some but not all functions with AtBLH9 from A. thaliana and may be involved in suppression of fruit development.

Plant materials

To construct the reference genome, a female D. tokoro plant (Waka1; original code: DT49) collected from Tahara, Wakayama Pref., Japan (33°32’16.8“N, 135°51’36.0”E) was crossed with a male D. tokoro plant (Kita1; original code: 110628-5) collected from Waga-Sennin, Kitakami, Iwate Pref., Japan (39°17’42.0”N, 140°53’45.6”E). The 186 F₁ progeny comprised 38 female plants, 89 male plants, and 59 non-flowering plants. To identify sex-linked regions, female and male individuals were collected from three wild populations: a population in northern Japan (KTKM; Kitakami, Iwate Pref., Japan; 39°18’25.0”N, 140°54’07.0”E); a population in central Japan (SHG; Koka, Shiga Pref., Japan; 34°56’24.1”N, 136°13’05.3”E); and a population in southern Japan (FKOK; Kasuya, Fukuoka Pref., Japan; 33°38’08.7”N, 130°30’38.2”E). For SDRs confirmation, five females and five males were collected from a wild population (KZGW; Kozagawa, Wakayama Pref., Japan; 33°32’24.1”N, 135°46’29.3”E). For transcriptome analysis, RNA-seq data were collected from 18 tissue samples. Male and female flowers were collected from wild populations in Takizawa and Kitakami, Iwate Pref., Japan, and non-reproductive organs were collected from Kita1. For small RNA-seq, 15 samples were collected from the wild population in Koka, Shiga Pref., Japan. To check the male specificity of the candidate genes, five females and five males were collected from two wild populations: one in northern Japan (HNMK; Hanamaki, Iwate Pref., Japan; 39°22’10.0”N, 141°09’16.0”E) and one in southern Japan (KMMT; Kumamoto, Kumamoto Pref., Japan; 32°53’34.8”N, 130°39’22.7”E).

DNA/RNA extraction, library preparation, and sequencing

For Oxford Nanopore Technologies (ONT) sequencing, genomic DNA was extracted from fresh leaves of Waka1 (female) and Kita1 (male) using NucleoBond HMW DNA (Macherey-Nagel, Düren, Germany). The DNA was subjected to size selection and purification with Short Read Eliminator XL (Circulomics, Baltimore, MD, USA). Long-read libraries were constructed using a Ligation Sequencing Kit V14 (SQK-LSK114). The libraries were sequenced on R10.4.1 flow cells (FLO-PRO114M) using a PromethION 2 Solo device. The raw sequencing data were subjected to base calling using dorado v0.8.1.

For Illumina sequencing, genomic DNA was extracted from Waka1 (female) and Kita1 (male) using a NucleoSpin Plant II Kit (Macherey-Nagel). Libraries for Waka1 (female) were constructed using a Collibri ES DNA Library Prep Kit for Illumina Systems (Invitrogen, Camarillo, CA, USA) and a TruSeq DNA PCR-Free LT Library Prep Kit (Illumina, San Diego, CA, USA). Libraries for Kita1 (male) were constructed using a TruSeq DNA PCR-Free LT Library Prep Kit. The libraries were sequenced using the MiSeq and HiSeq X systems. Genomic DNA was extracted from female and male individuals from northern (KTKM), central (SHG), and southern (FKOK) Japan using a DNeasy Plant Maxi Kit (Qiagen, Hilden, Germany). The DNA was purified by phenol/chloroform extraction and ethanol precipitation. Sequencing libraries were constructed using a Collibri ES DNA Library Prep Kit for Illumina Systems with a fragment length of ~350 bp and sequenced using the HiSeq X system (150 bp paired-end reads). Genomic DNA was extracted from five females and five males from KZGW population using Maxwell RSC Plant DNA Kit (Promega, Madison, WI, USA). Sequencing libraries were constructed using a Collibri PCR-free ES DNA Library Prep Kit for Illumina Systems with a fragment length of ~350 bp and sequenced using the Novaseq X Plus system (150 bp paired-end reads).

RAD-seq was performed as previously described [27]. Genomic DNA was extracted from fresh leaves of Waka1, Kita1, and 186 F₁ individuals using a NucleoSpin Plant II Kit (Macherey-Nagel). The DNA was digested with the restriction enzymes PacI and NlaIII and used to prepare libraries, and 75-bp paired-end reads were sequenced on the Illumina NextSeq 500 platform. Adapters and unpaired reads were removed using FaQCs and PRINSEQ lite. The filtered RAD-seq reads were used for linkage mapping and association analysis (S1 Data).

The RNA-seq data were obtained from 18 samples, including male and female flowers and non-reproductive organs of D. tokoro. The samples included male and female flowers at five stages of development: inflorescence stage 0, inflorescence stage 1, inflorescence stage 2, buds, and flowers. The samples also included eight non-reproductive organs: the vegetative shoot apex, leaf, stem, root apex, rhizome root, rhizome stem, rhizome bud, and rhizome storage tissue. Total RNA was extracted from the samples using an RNeasy Plant Mini Kit (Qiagen). The cDNA library was constructed using a TruSeq RNA Sample Prep Kit V2 (Illumina) and sequenced on the Illumina NextSeq 500 platform.

Small RNA-seq data were obtained from 15 samples, including male and female flowers and non-reproductive organs of D. tokoro. The samples included male and female flowers at five stages of development: inflorescence stage 0, inflorescence stage 1, inflorescence stage 2, buds, and flowers. The samples also included three non-reproductive organs: the vegetative shoot apex, male and female leaves, and male and female stems. Total RNA was extracted from the samples using Ambion Plant RNA Isolation Aid (Ambion, Austin, TX, USA), and the small RNA fraction was isolated using a mirVana miRNA Isolation Kit (Ambion). The small RNA libraries were constructed using a NEBNext Multiplex Small RNA Library Prep Set for Illumina (New England BioLabs, Ipswich, MA, USA) and DNA Clean & Concentrator-5 (Zymo Research, CA, USA). The libraries were sequenced on the NovaSeq 6000 platform.

Reference assembly

The genome size of D. tokoro Kita1 was estimated by flow cytometry using nuclei prepared from fresh leaf samples. Dioscorea rotundata accession TDr96-F1 (570 Mb) [27] was used as an internal reference standard. DNA from isolated nuclei was stained with propidium iodide and analyzed using a Cell Lab Quanta SC Flow Cytometer (Beckman Coulter, USA). The genome size of D. tokoro was estimated to be ∼388 Mb (570 Mb × 0.68) (S2 Fig).

To construct the reference genome based on the predicted genome size, the ONT sequencing data for Waka1 (female) and Kita1 (male) were filtered using chopper v0.8.0 [115]. The filtered reads from Waka1 (female) were assembled using PECAT [57], resulting in 212 contigs with N50 of 19,154,004 bp and a total size of 425.1 Mb for haplotype 1, and 346 contigs with N50 of 3,042,808 bp and a total size of 342.2 Mb for haplotype 2. The filtered reads from Kita1 (male) were also assembled using PECAT [57], resulting in 128 contigs with N50 of 33,851,599 bp and a total size of 415.0 Mb for haplotype 1, and 415 contigs with N50 of 1,379,312 bp and a total size of 300.8 Mb for haplotype 2. The assemblies were polished using Racon v1.5.0 [116] and Medaka v1.7.2 (Oxford Nanopore Technologies, 2018). The assemblies were also polished twice with Hypo v1.0.3 [117] using Illumina short reads from Waka1 (female) and Kita1 (male) filtered using FaQCs v2.08 [118].

TE and gene annotation

To annotate repetitive sequences, de novo repeat libraries were constructed for each assembly using EDTA (The Extensive de novo TE Annotator) v2.1.0 [119]. Using the resulting files, each genome FASTA file was soft-masked with the perl script make_masked.pl provided in EDTA.

For transcriptome-based gene identification using RNA-seq data from 18 samples of D. tokoro, poly(A) sequences and reads shorter than 50 bp were removed from the raw RNA-seq reads with FaQCs v2.10 [118]. Low-quality bases on the ends of reads with an average quality score of <20 were trimmed using PRINSEQ lite v0.20.4 [120]. Low-quality reads with an average read quality score of <20 were also removed using PRINSEQ lite v0.20.4. To obtain transcriptomes, the filtered RNA-seq reads were aligned to the assembled contigs with HISAT2 v2.2.1 [121]. The transcriptomes were assembled using StringTie v3.0.0 [122], and open reading frame regions were identified using TransDecoder v5.5.0 [123]. Gene models were excluded from further analysis if their coding sequences (CDSs) contained an internal stop codon, if the CDS lengths were not multiples of three, or if they encoded proteins shorter than 50 amino acids.

BRAKER2 v2.1.6 [124] was used for ab initio gene prediction using the filtered RNA-seq data from D. tokoro and protein homology information for D. alata and D. rotundata. Protein sequences of D. alata TDa95/00328 and D. rotundata TDr96_F1 were downloaded from NCBI (accession number GCA_020875875.1 and GCF_009730915.1, respectively). The predicted genes were categorized into three groups: gene models fully supported by protein hints, gene models at least partially supported by protein hints, and gene models without any support using the python script selectSupportedSubsets.py in BRAKER. Gene models fully supported by protein hints were selected, and gene models in which the CDS contained an internal stop codon, was comprised of a number of nucleotides that was not a multiple of three, or encoded a protein of <50 amino acids were removed.

Finally, all predicted genes were merged using GffCompare v0.12.6 [125]. Ab initio predicted genes were selected with no or less overlapping with genes predicted by StringTie. Gene predictions obtained from GeneMark in BRAKER2 were excluded. The final annotations include 21,308 genes in the Waka1 (female) haplotype 1 assembly, 17,710 genes in the Waka1 (female) haplotype 2 assembly, 23,200 genes in the Kita1 (male) haplotype 1 assembly, and 18,145 genes in the Kita1 (male) haplotype 2 assembly. To evaluate the completeness of the gene sets in the final scaffolds, BUSCO (Bench-Marking Universal Single Copy) v5.5.5 [56] was used with “transcriptome” as the assessment mode and Embryophyta odb10 as the database (S2 Table).

Generation of chromosomes using the pseudo-testcross approach

The pseudo-testcross approach [58] was used to construct chromosomes from the assembled contigs. SNP-type heterozygous markers and PA-type heterozygous markers were obtained from RAD-seq data from Waka1, Kita1, and 186 F₁ individuals. Parental line–specific heterozygous markers were identified as previously described [29] with several modifications (S1 Text). Linkage maps were constructed based on the heterozygous SNP-type and heterozygous PA-type markers. For each marker set (female-parent-heterozygous and male-parent-heterozygous marker sets), the markers were converted to genotype-formatted data to construct genetic linkage maps using MSTmap v1.0 [126]. After performing MSTmap, complemented-phased duplex linkage groups were generated by coupling- and repulsion-type markers under the pseudo-testcross approach. Finally, we pruned correlated flanking markers to remove redundant markers. After all filtering steps, two parental-specific linkage maps were constructed and visualized using R/qtl v.1.70 [60] and Asmap [127]. Based on the two parental-specific linkage maps, the contigs were anchored and linearly ordered as chromosomes using ALLMAPS [128]. Gene order–based synteny of the four constructed reference genomes was detected using MCscan [129]. Detailed methods are shown in S1 Text.

Identification of sex-linked regions by association analysis and coverage analysis

To identify sex-determination systems and sex-linked regions, association analysis was conducted using RAD-seq data from the 127 flowering F₁ progeny, which segregated into 38 females and 89 males. The RAD-seq data were aligned to the newly constructed female and male chromosomes using BWA v0.7.18-r1243-dirty [130]. SNP-type markers and PA-type markers were identified based on these alignments (S1 Text).

The associations between the genotypes and sex phenotypes of the 127 flowering F₁ individuals were calculated using Fisher’s exact test. The q-value of Fisher’s exact test was obtained for each marker by comparing the frequencies of particular alleles and sex phenotypes categorized as female or male. The false discovery rate (FDR) was set to 0.05 and was corrected by Benjamini-Hochberg correction. The log-transformed q-values (–log10(q)) for each position were visualized as Manhattan plots. In addition, the associations between the genotypes and sex phenotypes were calculated using Fisher’s exact test with equal numbers of randomly selected female and male F₁ individuals (38 females and 38 males). The associations between the genotypes and sex phenotypes were also checked by simple interval mapping using R/qtl v.1.70 [60]. The alignment-based dot plot between the chromosome 3 of Waka1 (female) haplotype 1 and Kita1 (male) haplotype 1 were generated using D-Genies [131]. Chromosome 3 of Kita1 (male) haplotype 1 was used as the target sequence, and chromosome 3 of Waka1 (female) haplotype 1 was used as the query sequence. The full plot was generated with noise filtering, and the inset panel was generated without noise filtering, using the D-Genies algorithm that removes small matches based on the alignment size and frequency.

The structural differences in chromosome 3 of Waka1 (female) haplotype 1 and Kita1 (male) haplotype 1 were detected by coverage analysis using Illumina short reads from four combinations of male and female individuals: male parent Kita1 and female parent Waka1 as well as male and female individuals collected from northern, central, and southern Japan. The Illumina short reads were filtered using FaQCs v2.10 [118] and aligned to each reference genome. The alignment to each haplotype reference was conducted using BWA v0.7.17-r1188 [130]. The alignment depths were calculated with Samtools v1.16.1 [132]. The alignment depths were normalized by dividing the values with the mean depth of all positions on the chromosomes. To reduce the effects of error on the alignment depths, a sliding window approach was employed (window size = 150 kb, step size = 10 kb). We identified regions in which the male depth was 0.5 (half of mean normalized depth) ± 0.15 and the female depth was 1 (mean normalized depth) ± 0.15, and regions in which the male depth was 0.5 (half the mean normalized depth) ± 0.15 and the female depth was < 0.15.

SDRs were also estimated using two previously reported methods, SDpop [63] and RADSex [64]. For SDpop analysis, Illumina short reads from five females and five males from KZGW population filtered using FaQCs v2.08 [118] and aligned to each reference genome: Waka1 (female) and Kita1 (male) haplotype 1. The alignment to each haplotype was conducted using BWA v0.7.17-r1188 [130]. After obtaining SNP-based genotypes using Bcftools v1.15 [132], posterior probability for segregation types were calculated by SDpop [63]. In this analysis, only the polymorphic sites located within coding regions of orthologous gene pairs of chromosome 3Y (from Kita1 haplotype 1) and chromosome 3X (from Waka1 haplotype 1) were used. The orthologous gene pairs (<1 × 10⁻⁵⁰ in BLASTP analyses, max_target_seqs = 2) were obtained following [13,22,133]. For RADSex analysis, we used RAD-seq data from 127 flowering F₁ individuals. Based on the filtered RAD-seq data and sex phenotypes, probability of association with sex, -log10(p), were calculated by RADSex v1.2.0 [64].

Based on the results of our association analysis, alignment-based dot plot analysis, mapping coverage analysis and SDRs detection analysis, regions including depth differences within the 23–29 Mb interval on X chromosome and within the 20–30 Mb interval on Y chromosomes were defined as X and Y-specific regions, respectively. The start position (23,395,001 bp) to end position (28,209,000 bp) of regions, in which the male depth was 0.5 (half of mean normalized depth) ± 0.15 and the female depth was 1 (mean normalized depth) ± 0.15, in X chromosome were defined as X-specific regions. The start position (23,759,001 bp) to end position (28,256,000 bp) of regions, in which the male depth was 0.5 (half of mean normalized depth) ± 0.15 and the female depth was < 0.15, in Y chromosome were defined as Y-specific regions. In this step, start and end positions of the depth differences were detected using higher-resolution sliding windows (window size = 20 kb, step size = 1 kb).

Genome structures of the X and Y chromosomes

Gene density and repetitive sequence accumulation values were obtained from gene annotation and the predicted repetitive sequences of the Waka1 (female) and the Kita1 (male) haplotype 1 assembly. Repetitive sequences included Copia LTR retrotransposons, Ty3 (gypsy) LTR retrotransposons, hAT TIR transposons, and Helitrons predicted by EDTA v2.1.0. Gene density and retrotransposon accumulation were visualized using the jcvi package [129].

Synonymous site divergence (d_S) between X- and Y-linked gametologs was obtained following [13,22,133] with slight modifications. Orthologous or paralogous gene pairs (<1 × 10⁻⁵⁰ in BLASTP analyses, max_target_seqs = 2) obtained from three comparison: female individual (Waka1) and male individual (Kita1) of D. tokoro, male (Kita1) D. tokoro and D. alata [30], male (Kita1) D. tokoro and D. rotundata [29]. The gene pairs were aligned in codon frames using MAFFT v7.475 [134] with the option “--localpair --maxiterate 1000” and Pal2Nal v14 [135]. Based on the in-codon-frame alignments, we calculate the Jukes and Cantor corrected values of d_S using KaKs_Calculator v2.0 [136]. Gene pairs located in the Y-specific regions of male D. tokoro and the X-specific regions of female D. tokoro were defined as XY gametologs. The gametologs with d_S values > 1 may not be real orthologs but may represent paralogs, so these were removed from XY gametologs. To evaluate the d_S values of all gene pairs across X and Y chromosomes, we applied a threshold of d_S < 0.5 based on the maximum d_S values in the SDRs reported in the previous studies [22,51] since these gene pairs may represent paralogs, not orthologs (S26A and S26C Fig). We also applied a less stringent threshold of d_S < 1.0 (S26B and S26D Fig).

To compare the sex chromosomes of D. tokoro (male), D. alata [30] and D. rotundata [29], Gene order–based synteny of chromosomes was detected using MCscan [129]. The orthologous regions were identified using the “jcvi.compara.catalog ortholog” and the “jcvi.compara.synteny screen” function with at least 30 collinear gene blocks. The detected collinearity was visualized using the “jcvi.graphics.karyotype” function.

Identification of candidate genes and miRNAs for sex determination

For transcriptome analysis of genes, the filtered RNA-seq data from 18 samples, including male flowers, female flowers, and non-reproductive organs, were aligned to the assembled contigs with HISAT2 v2.2.1 [121], and the aligned reads were counted using the featureCounts function in Subread v2.0.1 [137]. Differential expression analysis was performed with DESeq2 v3.15 [138] for two comparisons: male vs. female flowers at three stages of early development (inflorescence stages 0, 1, and 2) and male flowers at three stages of early development vs. non-reproductive organs. The FDR threshold was set to 0.05.

For transcriptome analysis of miRNAs, small RNA-seq data from 15 samples, including male flowers, female flowers, and non-reproductive organs, were filtered using FaQCs v2.10 [118] and Seqkit v2.3.0 [139]. Based on the small RNA-seq data, miRNAs were predicted using miRDeep-P2 v1.1.4 [140] as described previously [141]. The trimmed small RNA-seq reads were then aligned to the predicted miRNA references with Bowtie v1.3.1 [142]. The aligned reads were counted using the perl script bam2ref_counts.pl [141]. Differential expression analysis was performed with DESeq2 v3.15 [138] to compare male vs. female flowers at three stages of early development (inflorescence stages 0, 1, and 2). The FDR threshold was set to 0.05.

In the first step of candidate gene and miRNA identification, 56 genes and one primary miRNA located on Y-specific regions were identified. In the second step, within the genes and miRNA on the Y-specific regions, 14 genes (and no miRNA) were detected that were Y-specific and highly expressed during these three stages of early male vs. female flower development (q-value < 0.05). Within the 14 genes, ten genes were identified that were significantly upregulated in male vs. female flowers during the same three stages of development (log₂FC > 2). In the third step, within the ten genes, two genes were identified that were highly expressed in male flowers compared to non-reproductive organs during these three stages (q-value < 0.05). The two genes were significantly upregulated in male flowers compared to non-reproductive organs during the three stages (log₂FC > 2). Finally, two Y-specific genes were selected with significantly upregulated expression in male flowers during the early three stages of development (p-value < 0.05 and log₂FC > 2) in both transcriptome comparisons. The similarities of the candidate gene products were compared to known proteins using BLASTX with the Swiss-Prot function. The male-specific expression of the two candidate genes was confirmed by PCR amplification from five females and five males from two wild populations, one from northern Japan and one from southern Japan.

Overexpressing BLH9 and AtBLH9 in Arabidopsis thaliana

The homologous genes of BLH9 were subjected to phylogenetic analysis using protein sequences of BLH9, putative homologues of BLH9 from monocotyledonous species (Asparagus officinalis, Dioscorea alara, D. rotundata, Oryza sativa and Phoenix dactylifera) and TALE superfamily members in A. thaliana. The sequences were aligned using MAFFT v7.475 with the L-INS-i strategy [143]. Based on the alignment, a phylogenetic tree was constructed with the maximum likelihood method using iqtree v2.1.2 [144]. The tree was visualized using Interactive Tree Of Life (iTOL) v7.0 [145].

To overexpress BLH9 and AtBLH9 in A. thaliana, full-length cDNA clones were used for In-Fusion cloning. Total RNA was extracted from D. tokoro and A. thaliana using a Maxwell RSC Plant RNA Kit (Promega) and was reverse-transcribed into cDNA using a ReverTra Ace qPCR RT Kit FSQ-101 (TOYOBO, Osaka, Japan). BLH9 fragment was amplified from the D. tokoro cDNA using KOD FX Neo (TOYOBO). AtBLH9 was amplified from A. thaliana cDNA using PrimeSTAR GXL DNA Polymerase (Takara Bio). Each cDNA fragment was cloned into EcoRI and BamHI sites of the pBICP35 binary vector containing the CaMV35S promoter [146] using the In-Fusion HD Cloning Kit (Takara Bio). pBICP35::BLH9 and pBICP35::AtBLH9 were transformed into Agrobacterium tumefaciens strain GV3101::pMP90 by electroporation. A. thaliana transformation was performed by the floral dip method with A. tumefaciens strains carrying the binary expression plasmids, and kanamycin-resistant T₁ plants were selected. T₂ lines that showed a 3:1 segregation ratio for kanamycin resistance were selected, as the segregation ratio supports the idea that these lines contain a single T-DNA insertion.

To examine inflorescence phenotypes, 16 plants each were grown from eight lines: Col-0, three T₂ lines transformed with the construct for BLH9, and four T₂ lines transformed with the construct for AtBLH9. To eliminate the effects of kanamycin selection on plant growth, all seeds of T₂ plants and Col-0 were grown in Murashige and Skoog medium without kanamycin. The expression levels of BLH9 and AtBLH9 in T₂ plants were confirmed by RT-qPCR (S2 Data). The primers used in RT-qPCR are listed in S1 Text. The T₂ plants were separated into four groups based on the relative expression levels of the inserted genes calculated by the 2^−ΔΔCT method (S26 Fig): BLH9 Control (2^−ΔΔCT ≤ 2), BLH9 OX (2^−ΔΔCT > 2), AtBLH9 Control (2^−ΔΔCT ≤ 2), and AtBLH9 OX (2^−ΔΔCT > 2). At 30 days after the plants were transplanted to soil, three phenotypes were measured: inflorescence height, mean fruit length, and internode length. All phenotypic data are listed in S3, S4, and S5 Datas. The mean inflorescence heights and mean fruit lengths were compared by Wilcoxon rank sum test between the five groups: Col-0, BLH9 Control, BLH9 OX, AtBLH9 Control, and AtBLH9 OX. The variations in internode lengths between the five groups were compared by F-test. The p-values were adjusted by Bonferroni correction.

All detailed materials and methods are shown in S1 Text. All sequencing data are available under the Umbrella BioProject number PRJNA1223176. The genome assemblies and annotation files are available at zenodo: https://doi.org/10.5281/zenodo.14899024.