A haplotype-resolved chromosome-level genome assembly of autotetraploid Chinese yam (Dioscorea polystachya) elucidates dioscin biosynthesis and regulation

Abstract

Chinese yam (Dioscorea polystachya) is extensively cultivated for nutritional and medicinal applications. However, the lack of a high-quality reference genome has hindered molecular genetic analysis and breeding advancements. Here, we present a haplotype-resolved chromosome-level assembly for this autotetraploid species, featuring a 1.56-Gb genome anchored to 80 chromosomes across four haplotypes and comprising 95 668 protein-coding genes. Following divergence from Dioscorea alata about 4.64 million years ago (Mya), D. polystachya underwent a specific whole-genome duplication ~1.42 Mya, resulting in an autotetraploid species without subgenomic dominance. Notably, the biosynthetic pathway genes of dioscin, an important steroidal saponin primarily accumulating in tubers, were generally over-retained in D. polystachya compared to the diploid species D. alata. Of these genes, 7-dehydrocholesterol reductase (Dp7-DR) promoted the accumulation of dioscin, exhibiting tuber-specific expression and strong inducibility by abscisic acid, based on transcriptome and gene function analyses. We determined that the transcription factor DpbZIP12 activates Dp7-DR transcription, as supported by yeast one-hybrid, dual-luciferase reporter, and electrophoretic mobility shift assays. Notably, overexpressing Dp7-DR or DpbZIP12 resulted in lower cholesterol levels and elevated dioscin levels, while silencing either gene produced opposite metabolic profiles. These findings delineate promising targets for manipulating dioscin content and expand genetic resources for enhancing yam nutritional quality.

Introduction

China is a major and ancient origin and domestication center of yam (Dioscorea spp.), with rich germplasm resources representing 55 species [1]. Among these, Dioscorea polystachya is the most widely distributed and cultivated species, valued for centuries as both a healthy vegetable and a source of traditional Chinese medicine because it is rich in high-quality starch and multiple functional compounds. Dioscin, a saponin from D. polystachya, serves as an important industrial precursor for synthesizing steroid hormones such as cortisone, pregnenolone, and progesterone [2], and it exhibits beneficial effects, including anti-inflammatory, hypoglycemic and anti-tumor activities [3, 4]. In plants, the biosynthesis of dioscin starts with isopentenyl diphosphate and dimethylallyl diphosphate produced by the mevalonate pathway and methylerythritol 4-phosphate pathways, respectively [5, 6]. These intermediates are subsequently converted into geranyl diphosphate and farnesyl pyrophosphate, leading to the production of squalene, 2,3-oxidosqualene, and cholesterol. Cholesterol serves as a precursor for steroidal saponins and steroidal alkaloids [7], which is modified by cytochrome P450 monooxygenases (CYP90B, CYP94, CYP72A et al.) and UDP-glycosyltransferases [8–11]. Although heterologous biosynthesis from cholesterol to diosgenin has been achieved in Nicotiana benthamiana and yeast [12, 13], the absence of an efficient transgenic system in Dioscorea species has hindered the functional validation of related genes. Furthermore, due to the dioecious reproductive system and clonal propagation characteristics of D. polystachya [14], a high-quality reference genome for this species remains unavailable. These limitations pose significant challenges for elucidating the regulatory mechanisms of the genes encoding key enzymes involved in dioscin biosynthesis in D. polystachya.

In recent studies, genome assemblies published for diploid accessions of Dioscorea rotundata [15], Dioscorea alata [16, 17], Dioscorea zingiberensis [6], and the Chinese herb Dioscorea nipponica [18] have provided valuable resources for understanding the mechanism of dioscin biosynthesis. However, intraspecific variation in ploidy levels is widespread across Dioscorea species. For instance, documented cytotypes for white yam (D. rotundata) and D. alata encompass diploid (2n = 2x = 40), triploid (2n = 3x = 60), and tetraploid (2n = 4x = 80) forms [15, 16, 19, 20], whereas the Chinese medicinal species D. zingiberensis has been reported to exhibit diploid (2n = 2x = 20), triploid (2n = 3x = 30), and tetraploid (2n = 4x = 40) cytotypes [5, 6, 14]. Nevertheless, the mechanisms underlying polyploidization and the evolutionary relationships among Dioscorea species with different ploidy levels remain to be fully characterized.

The D. polystachya cultivar ‘Ruichang’, originating in the Jiangxi Province more than 500 years ago, is distinguished among Chinese cultivars for its superior flavor. The tubers of this cultivar accumulate substantial carbohydrate reserves and abundant steroidal saponins [1]. Because of the value of these secondary metabolites, there has been great interest in understanding their biosynthetic pathways and the factors that influence their accumulation. In this study, we present a haplotype-resolved chromosome-level genome assembly of the autotetraploid D. polystachya cultivar Ruichang. Through integrated comparative genomic, transcriptomic, and gene functional analyses, we explored and elucidated the regulatory mechanisms behind dioscin biosynthesis, substantially advancing our mechanistic understanding of this pathway.

Results

Genome assembly and annotation

The stem of the Chinese yam (D. polystachya ‘Ruichang’) is slender, green, or purplish, and it exhibits a dextrorse (clockwise) twining growth habit. The leaves are typically opposite, heart shaped to triangular ovate. Bulbils (aerial tubers) commonly form in the leaf axils (Fig. S1A). The tubers are cylindrical, measuring ~60–80 cm in length and 6–8 cm in diameter, with yellowish brown skin and white flesh (Fig. S1B and C). The D. polystachya genome size was estimated to be ~1.6 Gb by flow cytometry; in addition, a k-mer distribution analysis indicated that the D. polystachya genome is tetraploid with a high heterozygosity level of 3.9% (Fig. S2A and B; Tables S1 and S2). We detected 80 chromosomes using the root tip squash method (Fig. S2C), suggesting the presence of 20 chromosomes for each haplotype (2n = 4x = 80).

To assemble the genome sequence of D. polystachya, we generated 41.20 Gb of HiFi long reads and 76.47 Gb of short reads (Table S3) and obtained a draft assembly comprising four haplotypes using hifiasm. This assembly was 1.56 Gb in length, with a contig N50 value of 4.83 Mb after removing redundant contigs. Using 268.12 Gb (~175×) of chromosome conformation capture (Hi-C) data, we independently scaffolded those contigs belonging to each haplotype into four sets of 20 pseudochromosomes (Table 1; Table S4). Briefly, we aligned Hi-C reads to the draft assembly and partitioned them according to their best matching haplotype. Guided by the resulting haplotype-partitioned Hi-C interactions, we anchored and oriented all contigs into 20 clusters per haplotype. We then used a merged Hi-C contact map to refine scaffolding among the four haplotypes, with manual adjustments to correct misassigned genomic regions. Finally, we generated a haplotype-phased genome assembly for D. polystachya (Fig. 1). Each of these four haplotypes (hereafter referred to as Hap1, Hap2, Hap3, and Hap4) consists of 20 pseudochromosomes with total sizes of 393.53, 380.31, 325.91, and 323.70 Mb, respectively (Table 1). Comparisons to the diploid genome of D. alata showed that they are highly colinear (Fig. S3). Moreover, each chromosome in D. alata had four corresponding syntenic pseudochromosomes in the D. polystachya genome, each representing one of the four haplotypes (Fig. S3).

Table 1.

Assembly and annotation statistics.

Statistics	Value
Genome assembly
Contig sequence total/count	1562.96 Mb/653
Contig N50 length/count	4.83 Mb/105
Contig N90 length/count	1.32 Mb/351
Scaffolds length total/count	1437.24 Mb/80
Scaffold N50 length/count	19.04 Mb/32
Scaffold N90 length/count	13.53 Mb/68
Hap1 total/count	393.53 Mb/116
Hap2 total/count	380.31 Mb/117
Hap3 total/count	325.91 Mb/109
Hap4 total/count	323.70 Mb/120
Genome annotation
Total predicted gene/annotation gene	98 699/95 668
Repetitive sequences	887.18 Mb/61.88%
Number of genes with GO annotation	68 615
Number of genes with KOG annotation	82 632
Number of genes with NR annotation	95 647
Number of genes with Swiss annotation	70 412
Number of genes with TrEMBL annotation	93 331

Figure 1.

Distribution of genomic features within the D. polystachya genome. (A) Genomic landscape of D. polystachya. Chromosomes of yam (i), gene density (ii), TE density (iii), expression level (iv), noncoding RNA content of the genome (v), GC content (vi), paralogous genes on different chromosomes (vii) were displayed; 500-kb sliding windows. (B) Genome-wide all-by-all Hi-C interaction heatmap. Four haplotypes (Hap1, Hap2, Hap3, and Hap4) from each chromosome are shown.

We assessed the completeness of our genome assembly by conducting a Benchmarking Universal Single-Copy Orthologs (BUSCO) analysis. We identified 96.4% of the 1614 BUSCO genes in the D. polystachya genome as complete single-copy or duplicated genes (Table S5). For predicting protein-coding genes, we integrated homology-based, de novo, and transcriptome-based analyses, yielding 98 699 predicted gene models with an average gene length of 3440 bp, a mean exon length of 252 bp, and 5.25 exons per gene (Table 1). We assigned functional annotations to 95 668 (or 96.9% of total) genes using public databases (Table 1; Fig. S4). Repetitive sequences spanned 887.18 Mb across the four haplotypes, constituting 61.9% of the assembled genome (Table 1).

Disentangling the differential expression of D. polystachya haplotypes

To investigate the evolutionary relationships among the four haplotypes for each chromosome of the D. polystachya genome, we compared the D. polystachya genome to the D. rotundata genome and identified 4 937 003 single nucleotide polymorphisms (SNPs) in total. We then used these SNPs to reconstruct phylogenetic trees for each set of pseudochromosomes. The Hap1 and Hap2 versions of most (16/20) pseudochromosomes were phylogenetically close to each other, as were those of Hap3 and Hap4 pseudochromosomes (Fig. S5), suggesting two major haplotype lineages, represented by Hap1/Hap2 and Hap3/Hap4 (Fig. 2A). We identified sequence variations (including SNPs and insertions/deletions [InDels]) between all haplotype pairs to validate the genetic divergence of these four haplotypes. The numbers of genomic variants between Hap1 and Hap2 and between Hap3 and Hap4 were significantly smaller than those obtained for all other combinations (Fig. 2B; Tables S6 and S7). These consistent phylogenetic and genetic patterns among the four haplotypes support the reliability of the haplotype phasing.

Figure 2.

Chromosome phasing and subgenome dominance of D. polystachya. (A) Clustering of 4 937 003 chromosome-specific SNP enabled consistent partitioning of the genome into two clusters. Dioscorea rotundata was used as outgroup, and Hap1-Hap4 donate the four sets of chromosomes after phasing. Chr01 is displayed as a representative example, with the trends observed across most chromosomes. (B) Heatmaps of variation between subgenomes. Each cell represents the SNP density and InDel relationship between two subgenomes, with the color depth reflecting the degree of variation. (C) Variation in gene density between representative chromosomes in the two subgenomes of D. polystachya. Gene density variation along representative chromosomes without substantial segmental fragmentation in the two subgenomes of D. polystachya is shown. Gene density was calculated using 200-kb sliding windows. (D) Number of dominantly expressed genes between paralogous gene pairs from the two subgenomes of D. polystachya. Blue indicates Hap1 genes with higher expression than their Hap3 paralogs (Hap1/Hap3 fold change >1, 2, 3, or 4); green indicates Hap3 genes with higher expression than their Hap1 paralogs (Hap3/Hap1 fold change >1, 2, 3, or 4). (E) TE distribution in flanking regions of genes from different gene sets. TE density was calculated in 100-bp sliding windows over the 5-kb region upstream and downstream of the transcription start site. The yellow ellipse indicates the gene body.

Given the phylogenetic clustering of haplotypes (Hap1/Hap2 and Hap3/Hap4 as homologous pairs) (Fig. 2A and B), we designated Hap1 and Hap3 as representatives of two ‘diploid’ pseudogenomes in the tetraploid genome of D. polystachya and evaluated their divergence levels. Hap1 and Hap3 contain 25 679 and 24 056 genes, respectively. To mitigate effects from structural variation, we analyzed syntenic regions, identifying 17 258 genes in Hap1 and 17 356 in Hap3, indicating that Hap1 did not have a higher gene content (i.e. did not retain more genes) than Hap3 (Fig. 2C). Furthermore, we evaluated whether transcript levels are balanced. Using transcriptome deep sequencing (RNA-seq) data for D. polystachya tubers, we identified 20 681 genes expressed in Hap1 and 20 056 genes expressed in Hap3. Importantly, there was no significant bias in gene expression (transcripts per million [TPM] ≥ 2) levels between the syntenic paralogous pairs in Hap1 and Hap3 (Fig. 2D; Table S8). We also determined the density of transposable elements (TEs) in the syntenic regions of Hap1 and Hap3. The TE density in gene-flanking regions was also similar between these two pseudogenomes (Fig. 2E).

A comparative analysis of genes from Hap1 and Hap3 revealed very high sequence conservation, with over 95% of all gene pairs across the two haplotypes exhibiting very high (>95%) sequence identity (Fig. 3A). Moreover, when we checked for potential expression bias between the two haplotypes in various tissues (stems, roots, tuber peel, and tuber flesh), we determined that 2.2%, 2.4%, 1.7%, and 1.5% of expressed genes show haplotype-specific expression in the stems, roots, tuber peels, or tuber flesh, respectively (Fig. 3B; Table S9). In addition, 86 expressed genes displayed significant differences in their expression levels in a haplotype-specific manner among the four tissues analyzed (Fig. 3C). Genes with biallelic expression had notably lower Ka/Ks values than all other genes (Fig. 3D), suggesting that most of these genes with balanced expression between the two haplotypes Hap1 and Hap3 are more evolutionarily conserved than the other types of genes with imbalanced expression between the two haplotypes. Collectively, these findings suggest that the D. polystachya genome shows no clear expression dominance among haplotypes.

Figure 3.

Allele-specific expression characteristics of D. polystachya. (A) Comparison of allele coding sequences between two haplotypes (Hap1 and Hap3). (B) Allele-specific expression (ASE) patterns in stem, roots, tuber peels, and tuber flesh. Genes with TPM >2 for both Hap1 and Hap3 alleles and less than 2-fold difference were classified as having ‘biallelic expression’. Genes with TPM >2 for both alleles but a >2-fold difference were defined as showing ‘increased expression of one allele’. When only one allele (either Hap1 or Hap3) had TPM >2 and the other ≤2, the gene was defined as showing ‘Hap1 dominant expression’ or ‘Hap3 dominant expression’, respectively. (C) Venn diagram showing the overlap of ASE genes in the four tissues. (D) Calculation of Ka/Ks values for various classes of ASE genes in tuber flesh, with outliers represented by dots and the interquartile range shown by boxes. ^*P ≤ 0.05; ^**P ≤ 0.001 (t-test).

Analysis of whole-genome duplication events and phylogenetic relationships of D. polystachya

To investigate D. polystachya evolution, we assessed the distribution of synonymous substitution per gene (Ks) between colinear gene pairs to identify signatures of whole-genome duplication (WGD) events. Based on the peaks of Ks values, we detected evidence for three WGD events in D. polystachya. A recent D. polystachya-specific tetraploidization event (Ks = 0.020) reflected the divergence time between Hap1/2 and Hap3/4. In addition, two shared WGD events corresponded to the delta (Ks = 0.82) and tau (Ks = 1.18) events, which are conserved in Dioscoreaceae and monocotyledonous species, respectively (Fig. 4A). We also identified a divergence between D. polystachya and D. alata (Ks = 0.063) (Fig. 4A). Collectively, the D. polystachya-specific WGD event resulted in autotetraploidization, confirming the Chinese yam variety ‘Ruichang yam’ as an autotetraploid.

Figure 4.

Auto-tetraploidization event in D. polystachya. (A) Ks values revealed a recent D. polystachya-specific WGD event during the evolution. (B) Phylogenetic tree for D. polystachya and eleven other plants. Gene family expansions and gene family contractions among the total changes are shown in the pie chart, the estimated divergence time (million years ago, Mya) is indicated at each node, pentagrams represent three WGD events.

We reconstructed a time-calibrated phylogeny of 12 plant species, including D. alata, D. rotundata, and D. zingiberensis, using 923 shared single-copy genes. Phylogenomic analysis dated the divergence of Dioscoreaceae from other monocot lineages to ~86.45 million years ago (Mya). Dioscorea zingiberensis separated from the Dioscorea group at ~32.55 Mya, whereas D. polystachya and its closest relative D. alata diverged ~4.64 Mya (Fig. 4B). Molecular dating placed the tau, delta, and D. polystachya-specific tetraploidization WGD events at 110, 32, and 1.42 Mya, respectively (Fig. 4B).

Using syntenic gene pairs between the D. alata and D. polystachya genomes, we estimated a background gene retention rate of 91.3%, calculated as the ratio of D. polystachya syntenic genes to (D. alata syntenic genes × 4). This high gene retention rate is consistent with the relatively recent divergence time between D. polystachya and D. alata. Notably, genes involved in the dioscin biosynthetic pathway showed an even higher retention rate of 98.4%, suggesting that nearly all these genes are preserved as four copies (one per haplotype) in the D. polystachya genome. Moreover, the copy number of six gene families involved in this pathway was further expanded through tandem duplication events, namely the GGPP, 24-Sterol methyltransferase type 1 (SMT1), C-4 Sterol methyl oxidase 1 (SMO1), CYP72A, CYP90B, and CYP94 gene families (Table S11). In addition, based on RNA-seq data of tuber tissues, we observed frequent divergence in haplotype expression among these genes across the four haplotypes (Table S11). These findings indicate that the tetraploidization that led to the formation of D. polystachya might have contributed to enhanced dioscin biosynthesis compared to that in D. alata through both higher gene copy number and diversified gene expression patterns.

Role of Dp7-DR in dioscin accumulation in D. polystachya

We conducted a temporal RNA-seq analysis of four tuber developmental stages, at 150 (T1), 170 (T2), 190 (T3), and 210 (T4) days after planting, using Hap1 of the D. polystachya genome as a reference (Fig. S6). A total of 8.65 Gb of clean reads was produced on average, reaching an average mapping ratio to the reference genome of 96.0% (Table S12). We identified differentially expressed genes (DEGs) through pairwise comparisons of the four developmental stages (Fig. S6B) and divided them into nine co-expression clusters based on K-means clustering analysis (Fig. S6C). Our previous study revealed rapid tuber development from T2 to T4 stages [21]. Notably, the DEGs in Clusters 4 and 6 were highly expressed at the T4 stage and wereenriched in Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways related to ‘steroid biosynthesis’ and ‘plant hormone signal transduction’ (Fig. S6D). Importantly, the levels of dioscin, the primary bioactive steroid saponin in yam tubers, peaked at the T4 stage (Fig. 5A), with tuber peel and flesh accumulating significantly higher concentrations than leaves or stems, and tuber peels having higher levels than tuber flesh (Fig. 5B).

Figure 5.

Identification of key genes related to dioscin biosynthesis. (A, B) Dioscin content in D. polystachya tubers at different developmental stages (A) and in different tissues (B). T1–T4 indicate four rapid growth stages (150, 170, 190, and 210 days after planting). Values are shown as means ± standard deviation from three biological replicates (n = 3). Different lowercase letters indicate significant differences, as determined by one-way analysis of variance (ANOVA), with P < 0.05. (C) Number of DEGs for the indicated comparisons. The x-axis of the UpSet plot represents the total number of DEGs identified in each group, and the y-axis shows the number of DEGs unique to each set across various tissues. (D, E) Relative Dp7-DR expression levels in yam tubers at different developmental stages (D) and in diverse tissues (E), as determined by RT-qPCR. The specific primers are listed in Table S2; ACTIN was used as a reference. Three biological replicates were used for each sample, and the 2^−ΔΔCt method was used to measure transcript levels. (F) Subcellular localization of Dp7-DR in N. benthamiana leaves. Confocal microscopy images of leaves expressing Dp7-DR-GFP (top) or GFP alone (bottom) were acquired using an LSM 880 confocal microscope (Zeiss, Germany) with the following settings: GFP (excited at 488 nm), ER marker SPER-mCherry (excited at 561 nm), and chlorophyll autofluorescence (excited at 640 nm). GFP, green fluorescent protein; SPER-mCherry, an ER marker; Chlorophyll, chlorophyll autofluorescence. Merged, image combining GFP fluorescence, ER fluorescence, chlorophyll autofluorescence, and bright field. Scale bars, 20 μm. (G–L) Relative expression levels of Dp7-DR (G, J), cholesterol content (H, K), and dioscin content (I, L) in D. polystachya bulbils with Dp7-DR silencing (G–I) or overexpression (J–L). Values are shown as means ± standard deviation from three replicates (n = 3). TRV, co-infiltration of the pTRV1 vector with the empty vector pTRV2. TRV-Dp7-DR, co-infiltration of pTRV1 with pTRV2-Dp7-DR. ^*P < 0.05; ^**P < 0.01; ^***P < 0.001 (t-test).

To pinpoint key genes responsible for dioscin accumulation, we compared DEGs in tubers at Stages T3 and T4, as well as in the peel and flesh tissues, identifying 659 and 1298 DEGs, respectively, with 100 shared between the comparisons (Fig. 5C). Among these, Dp7-DR (DpolyH1c03G000046), encoding 7-dehydrocholesterol reductase (7-DR), is annotated to be the ‘steroid biosynthesis’ KEGG pathway (ko00100) (Fig. S7A). Expression analysis of dioscin biosynthetic pathway genes (Table S10) across the four haplotypes of D. polystachya revealed that all four copies of Dp7-DR (DpolyH1c03G000046, DpolyH2c03G000045, DpolyH3c03G000049, and DpolyH4c03G000003) exhibited high transcript levels, which correlated strongly with dioscin content (Fig. S7B and C; Table S13). These four haplotype sequences were identical (Fig. S8A). We isolated a 1305-bp Dp7-DR open reading frame (Fig. S8B) encoding a 434-amino acid protein that phylogenetically clusters with its orthologs Da7-DR from D. alata and Dr7-DR from D. rotundata (Fig. S8C; Table S14). Dp7-DR expression peaked at the T4 developmental stage, particularly in tuber peels and flesh (Fig. 5D and E), consistent with the elevated dioscin accumulation in these tissues. Furthermore, intracellular localization via Dp7-DR-GFP fusion in N. benthamiana leaves showed GFP signal colocalization with an endoplasmic reticulum (ER) marker (Fig. 5F), indicating predominant ER localization. This localization is consistent with its function as a key enzyme in the sterol biosynthesis pathway, thereby facilitating its role in dioscin production.

To investigate the function of Dp7-DR in dioscin biosynthesis, we employed virus-induced gene silencing (VIGS) and transient overexpression in D. polystachya bulbils (Fig. 5G–L and Fig. S9). Compared to the control, Dp7-DR-silenced bulbils exhibited significant reduction in Dp7-DR transcripts (Fig. 5G), increased cholesterol content, and decreased dioscin content (Fig. 5H and I). Conversely, overexpressing Dp7-DR in bulbils was characterized by reduced cholesterol and enhanced dioscin accumulation (Fig. 5J–L). Collectively, these results indicate that Dp7-DR plays a role in dioscin accumulation.

DpbZIP12 binds to the Dp7-DR promoter to promote its transcription

The promoter region of Dp7-DR contains 11 abscisic acid (ABA)-responsive elements (ABREs), as predicted by the PlantCARE database (Fig. 6A; Table S15). ABREs are involved in the response to ABA and can be recognized by basic leucine zipper (bZIP) A subfamily transcription factors. Treatment with ABA for 12 h resulted in a rise in dioscin content and higher expression levels for Dp7-DR (Fig. 6B and C). DpbZIP12, encoding a bZIP A subfamily member, was highly expressed in D. polystachya tubers based on RNA-seq data; moreover, its expression levels were strongly and positively correlated with dioscin content (Fig. 6D and Fig. S11A). To better understand the putative regulatory role of DpbZIP12 in dioscin accumulation, we investigated whether DpbZIP12 directly bound to the Dp7-DR promoter. Yeast one-hybrid (Y1H) assays confirmed that DpbZIP12 binds to the Dp7-DR promoter in yeast cells (Fig. 6E). We then assessed the effect of DpbZIP12 on Dp7-DR promoter activity using luciferase reporter assays (Fig. 6F). Nicotiana benthamiana leaves co-infiltrated with a DpbZIP12 effector construct and a reporter construct consisting of the Dp7-DR promoter driving the firefly luciferase (LUC) reporter gene exhibited notably elevated luminescence over those measured in the controls (Fig. 6G). We also performed an electrophoretic mobility shift assay (EMSA) with a probe containing the bZIP-binding site in the Dp7-DR promoter. Recombinant purified maltose binding protein (MBP)-DpbZIP12 bound to the biotin-labeled probe, and the intensity of the shifted band diminished when an unlabeled competitor probe was added (Fig. 6H). These findings suggest that DpbZIP12 binds to the Dp7-DR promoter to induce its transcription.

Figure 6.

DpbZIP12 binds directly to the Dp7-DR promoter to activate its expression. (A) Prediction of cis-acting elements in the Dp7-DR gene promoter, as determined by the PlantCARE database. (B, C) Dioscin content (B) and relative Dp7-DR expression levels (C) in response to treatment with 100 μM ABA. Values are shown as means ± standard deviation from three independent biological replicates. Different lowercase letters indicate significant differences, as determined by one-way ANOVA, with P < 0.05. (D) Heatmap representation of expression levels for the genes encoding subgroup A bZIP transcription factors for each of the four haplotypes at four tuber developmental stages (T1–T4). (E) Yeast one-hybrid assay showing that DpbZIP12 binds to the Dp7-DR promoter. The plasmids pHIS2-proDp7-DR and AD-DpbZIP12 were co-introduced into the yeast strain Y1H Gold. AD, GAL4 activation domain. SD/−LW, synthetic defined (SD) medium lacking leucine and tryptophan. SD/−LWH, SD medium lacking leucine, tryptophan, and histidine. 3-AT, 3-amino-1,2,4-triazole. (F) Diagrams of the plasmids used for the dual-luciferase reporter assays. The Dp7-DR promoter was cloned upstream of the firefly luciferase (LUC) reporter gene in pGreenII-0800-LUC, and DpbZIP12 was cloned in pGreenII-62-SK as effector plasmid. (G) Dual-luciferase reporter assay system showing that DpbZIP12 activates transcription of the Dp7-DR promoter. The empty vector pGreenII-62-SK was used as negative control. Agrobacterium cell suspensions containing either pGreenII-62-SK-DpbZIP12 or pGreenII-0800-LUC-proDp7-DR were mixed in 10:1 and 1:1 ratios (v/v). Values are shown as means ± standard deviation (n = 6). Statistical significance was determined by t-test for the comparison between the control and the 1:1 mixture of pGreenII-62-SK-DpbZIP12 and pGreenII-0800-LUC-proDp7-DR (^**P < 0.01). (H) EMSA showing the binding of recombinant purified MBP-DpbZIP12 to the Dp7-DR promoter. Biotin-labeled probes were mixed with MBP-DpbZIP12. Unlabeled probes served as competitor probes. MBP, maltose binding protein; proDp7-DR, Dp7-DR promoter. (I) Relative DpbZIP12 expression levels in response to treatment with 100 μM ABA. Values are shown as means ± standard deviation from three independent replicates. ^*P < 0.05; ^**P < 0.01; ^***P < 0.001 (t-test).

ABA treatment elevated DpbZIP12 expression levels within 12 h (Fig. 6I). Silencing DpbZIP12 in bulbils resulted in significantly lower Dp7-DR transcript levels (Fig. 7A and B), leading to greater accumulation of cholesterol concomitant with lower dioscin levels (Fig. 7C and D). By contrast, overexpressing DpbZIP12 led to lower cholesterol content, along with elevated dioscin content and higher Dp7-DR expression (Fig. 7E–H). These results demonstrate that DpbZIP12 activates transcription from the Dp7-DR promoter, thereby enhancing its expression and promoting dioscin biosynthesis.

Figure 7.

DpbZIP12 promotes the accumulation of dioscin in bulbils. (A–H) Relative expression levels of DpbZIP12 (A, E) and Dp7-DR (B, F), cholesterol (C, G), and dioscin contents (D, H) of silenced (A–D) or overexpressing DpbZIP12 (E–H). RT-qPCR was performed using the primers listed in Table S2; ACTIN was used as a reference. Three biological replicates were used for each sample, and the 2^−ΔΔCt method was used to measure transcript levels. Values are shown as means ± standard deviation from three replicates (n = 3). ^*P < 0.05; ^**P < 0.01; ^***P < 0.001 (t-test). TRV, co-infiltration of pTRV1 with pTRV2 as negative control; TRV-DpbZIP12, co-infiltration of pTRV1 with pTRV2-DpbZIP12. (I) Proposed model of ABA-mediated regulation of dioscin biosynthesis, with DpbZIP12 activating Dp7-DR transcription.

Discussion

A recent WGD resulted in the formation of autotetraploid D. polystachya and contributed to dioscin biosynthesis

Polyploidization is closely linked to the diversification of plant species and plays an important role in plant genome evolution [22]. In addition to the tau and delta WGD events shared by monocotyledonous and Dioscorea species, respectively, D. polystachya experienced its own WGD event about 1.42 Mya after its divergence from D. alata (Fig. 4), the closest species in evolutionary terms among the yams whose genomes have been sequenced [23]. This recent WGD event resulted in the autotetraploidy of D. polystachya. Polyploidy confers several agronomically significant advantages in plant breeding, most notably larger organs, buffering against deleterious mutations, and greater heterozygosity. Harnessing these attributes is important for the development of cultivars with greater yields, enhanced product quality, and heightened tolerance to biotic and abiotic stressors [24–26]. For instance, tetraploid potato (Solanum tuberosum) and alfalfa (Medicago sativa) are more productive than their diploid counterparts [27, 28]. Polyploidy substantially augments carotenoid accumulation in maize (Zea mays), with the seeds of artificial tetraploid maize lines demonstrating concentrations up to 220% higher than diploid seeds [29]. A naturally occurring tetraploid Arabidopsis (Arabidopsis thaliana) accession exhibits enhanced salinity tolerance through polyploidy-mediated regulation of leaf potassium homeostasis [30]. Significant divergence in dioscin content exists between D. polystachya and its diploid ancestor D. alata [18], denoting polyploidy contributes to dioscin accumulation in the autotetraploidy of D. polystachya in this study.

One possible explanation to dioscin accumulation after WGD in D. polystachya is higher gene copy number and gene-dosage effects. Genes associated with the dioscin biosynthetic pathway were more prone to retention and are present in more copies in D. polystachya than in D. alata (Table S11). In hexaploid camelina (Camelina sativa), combinatorial sequence variation across its three FATTY ACID DESATURASE 2 (FAD2) homoeologs induced by gene editing generated substantial phenotypic diversity, yielding oil profiles with oleic acid concentrations ranging from 10% to 62% among different edited lines, illustrating how oil content is correlated with allelic dosage in an additive manner [31]. Similarly, genome-wide association studies in sweet potato (Ipomoea batatas) revealed genomic regions associated with 23 agronomic characteristics and exhibiting dosage effects. Specifically, expression of the expansin gene IbEXPA4 was negatively correlated with the dosage of its favorable allele, which suppressed IbEXPA4 transcription and enhanced storage root weight. In the same study, tuber flesh coloration exhibited a positive dosage effect with Orange (IbOr) haplotypes derived from the ‘Nancy Hall’ lineage with orange flesh [32]. Besides, in D. nipponica and D. zingiberensis, an expansion of the CYP90, CYP94, and UGT73 gene families promoted dioscin biosynthesis and enhanced their dioscin content [5, 18].

Dioscorea alata is domesticated in Southeast Asia [14], and it is distributed in the tropical and subtropical regions, whereas D. polystachya thrives in temperate regions and covers the largest cultivation area for yam in China [33] with a broader ecological niche than D. alata. Previous studies have indicated that ≤33% of Dioscorea species display varying levels of ploidy, even within the same species [14]. One possible explanation is that asexual reproduction promotes polyploidy as a by-product of adaptive selection during prolonged periods of asexual reproduction, a strategy used by plants to protect themselves from extreme environmental damage [34]. Polyploidization results in expansion of gene families, and subsequent gene fractionation promotes gene sub-/neo-functionalization, ultimately enhancing the adaptation and genome plasticity of plant species.

Dioscin levels in D. polystachya are regulated by ABA and DpbZIP12

The biosynthesis of plant saponins is modulated by phytohormones through the regulation of key enzymes or gene expression. ABA is a key regulator of secondary metabolism [35], enhancing flavonoid biosynthesis in tomato [36], promoting carotenoid accumulation in citrus (Citrus sinensis) fruits [37, 38], stimulating terpenoid biosynthesis in hemp (Cannabis sativa) [39], and regulating artemisinin biosynthesis in sweet wormwood (Artemisia annua) [40]. At present, few in-depth studies have explored the regulation of dioscin biosynthesis in plants. We showed that treating yams with ABA resulted in a significant rise in dioscin levels at 12 h post-treatment, followed by a drop at 24 h post-treatment (Fig. 6). ABA initially promotes the accumulation of secondary metabolites by upregulating the expression of genes encoding the respective biosynthetic enzymes, but prolonged ABA treatment may trigger feedback mechanisms. Long-term ABA treatment may induce the expression of negative regulators of ABA signaling (e.g. Protein phosphatase 2C [PP2C], Sucrose nonfermenting 1-related protein kinase 2 [SnRK2]), accelerating the degradation of secondary metabolites to mitigate phytotoxicity from their excessive accumulation [41]. Concurrently, plants experiencing metabolite-induced stress may redirect energy resources toward essential maintenance such as membrane repair or osmotic homeostasis, potentially diverting substrates from secondary metabolism [42]. We thus suspect that a single ABA treatment elicits significant dioscin accumulation, whereas prolonged application may trigger a feedback mechanism, resulting in lower levels of secondary metabolites.

bZIP transcription factors are essential components of ABA signaling [43, 44]. In citrus, CsbZIP44 acts as a central regulator of ABA-induced carotenoid biosynthesis by binding to the promoters of carotenoid metabolism-related genes and activating their expression [37, 38]. Similarly, overexpression of AabZIP1 in A. annua elevated artemisinin production via transcriptional activation of Amorpha-4,11-diene synthase (ADS) and CYP71AV1 [45]. SmbZIP7 and SmbZIP20 in the perennial herb Danshen (Salvia miltiorrhiza) regulate tanshinone and artemisinin production [46]. OsbZIP79 in rice (Oryza sativa) regulates production of the diterpenoid antitoxin [47]. Two Asparagus racemosus bZIP transcription factors bind to the promoter region of the steroidal saponin biosynthesis gene Squalene epoxidase (ArSQE) [48]. Although bZIPs influence the accumulation of diverse secondary metabolites, their role in dioscin biosynthesis is uninvestigated, particularly within Dioscorea species. In this study, we identified an ABA-responsive bZIP transcription factor, DpbZIP12, that can bind to the Dp7-DR promoter to modulate dioscin biosynthesis (Figs 6 and 7, and Fig. S10). Among the DEGs in bulbils silenced for or overexpressing DpbZIP12, 18 DEGs assigned to the KEGG pathway ‘plant hormone signal transduction’ were specifically involved in ABA signal transduction, including genes encoding ABA receptors (PYL), protein phosphatase 2C (PP2C), kinases (SnRKs), ABA-responsive element binding factors (ABFs), and abscisic acid-insensitive 5 (ABI5) (Fig. S11C). Additional research is required to elucidate the regulatory mechanisms and interactions between ABA signal-related proteins and related transcription factors in the control of dioscin biosynthesis.

Of the ~600 Dioscorea species globally, only a limited number of cultivars are commercially utilized, representing untapped genetic potential and necessitating the exploration of accelerated biotechnological resources. Based on the released genomic data combined with the identification of dioscin-related genes, the key genes closely related to dioscin biosynthesis, transport, and accumulation can be further explored. For example, SNPs in the promoter of CYP94D144 (identified in D. zingiberensis) are associated with variation in diosgenin content, suggesting its utility as a molecular breeding target [49]. Upregulating CYP94D144 expression, which could be achieved via clustered regularly interspaced short palindromic repeats (CRISPR)/CRISPR-associated nuclease 9 (Cas9)-mediated gene editing, should enhance dioscin production and expedite breeding progress. In this study, we demonstrated that the DpbZIP12–Dp7-DR module regulates dioscin content, providing an excellent target for the study of dioscin biosynthesis and the genetic engineering of varieties with high dioscin content.

In conclusion, we assembled a haplotype-resolved chromosome-level genome from autotetraploid D. polystachya. In addition to the tau WGD event shared by monocotyledonous species and the delta event WGD associated with Dioscorea, D. polystachya experienced its own WGD event about 1.42 Mya, giving rise to its autotetraploid genome. Notably, dioscin biosynthetic pathway genes were generally over-retained compared to the diploid species D. alata (98.4% vs. 91.3%). We also demonstrated that Dp7-DR is involved in dioscin biosynthesis and identified a previously unknown regulator of dioscin biosynthesis, DpbZIP12, which stimulates dioscin production in yam tubers (Fig. 7I).

Materials and methods

Plant materials and treatment

Chinese yam (D. polystachya var. ‘Ruichang yam’) plants were grown at the experimental base of Jiangxi Agricultural University in Nanchang (28°45′53′′N, 115°49′41′′E), Jiangxi Province, China. Young leaves were collected for whole-genome sequencing, and roots, young stems, and tubers were sampled for RNA-seq analysis. The leaves of D. polystachya plants were sprayed with 100 μM abscisic acid (ABA), and samples were randomly collected at 0, 12, and 24 h after treatment, immediately flash frozen in liquid nitrogen and stored at −80°C until use.

Genome size estimation

Illumina short reads were preprocessed using SOAPnuke (v1.6.5) for quality control, followed by genome size estimation via GenomeScope across multiple k-mer values. Separately, fresh leaves of D. polystachya and soybean (Glycine max, internal control) plants were collected and processed, and their purified nuclei were stained with propidium iodide. Nuclear DNA content was quantified using a FACSCalibur flow cytometer [50].

Genome assembly, scaffolding, and phasing separation of subgenomes

The tetraploid genome of D. polystachya was assembled using hifiasm (v0.19.3-r572) [12] with PacBio HiFi long reads and Hi-C data, where the ‘--n-hap 4’ parameter was specified to enable phasing into four haplotypes. This step yielded preliminary haplotype-resolved assemblies, each representing one of the four homologous haplotypes.

To refine these assemblies, Purge_Dups (v1.2.5) [51] was employed to remove heterozygous and redundant sequences in a haplotype-specific manner. Briefly, HiFi and short reads were mapped to the entire preliminary assembly using minimap2 (v.2.28). Subsequently, the ‘split_paf.py’ script was used to partition the alignment results into four haplotype-specific subsets based on the phasing outcome from hifiasm. The coverage information of the mapped reads was then utilized to identify and purge redundant regions corresponding to each haplotype. This refinement step produced four haplotypes with sizes consistent with the estimates from flow cytometry, and the retained sequences were further verified by alignment with the genome of a closely related species (D. alata) to ensure no valid coding regions were erroneously removed.

For scaffolding, Hi-C reads were aligned to the refined assembly using BWA-mem (v.0.7.17). The aligned reads were then partitioned into four haplotype-specific subsets based on their best alignment scores, with multimapped reads retained and assigned to the corresponding haplotypes using a probability model implemented in HiC-Pro (v3.1.0). The Allhic (v0.9.8) pipeline was then run to scaffold each of the four subgenomes separately. Specifically, each haplotype-specific Hi-C subset was used to construct 20 pseudochromosomes per haplotype via the combined workflow of Juicer (v1.5) [52], 3D-DNA (180922), and Juicebox (v1.11.08) [53].

The resulting Juicebox ‘assembly’ files for each haplotype were merged and grouped according to homologous relationships identified via Nucmer (v3.23) [54]. Manual curation was subsequently performed based on the integrated Hi-C contact map of all four haplotypes. Contigs erroneously assigned between haplotypes were corrected with Juicebox based on the principle that contigs within the same haplotype should exhibit stronger Hi-C interactions than those between haplotypes. In addition, interchromosomal contacts were examined to reorder homologous pseudochromosomes across haplotypes, ensuring that chromosomes within the same haplotype displayed stronger global interaction signals compared to those across haplotypes. Through this strategy, a final assembly was obtained consisting of 80 pseudochromosomes, with 20 chromosomes per haplotype.

Genome annotation

Protein-coding genes were predicted through an integrative approach combining ab initio predictions from Genemark (v4.61_lic) [55] and AUGUSTUS (v3.3.3) [56] with homology-based evidence, including plant protein sequences from the nonredundant (Nr) [57], EuKaryotic Orthologous Groups (KOG) [58], Gene Ontology (GO) [59], Swiss-Prot [60], and TrEMBL [61] databases, as well as a de novo transcriptome assembly constructed from D. polystachya RNA-seq data using Trinity (v2.11.0) [62].

Repeat sequences in the D. polystachya genome were identified de novo using RepeatModeler [63] and annotated with RepeatMasker [64] utilizing the repeat library.

Genome-wide synteny analysis among yam species

Adjacent syntenic gene pairs were employed to delineate large-scale colinear genomic blocks between subgenomes. Considering local structural variations or potential assembly errors, syntenic gene pairs occasionally lacked immediate adjacency to neighboring syntenic pairs in one or both subgenomes. Syntenic fragments were consolidated into a single colinear block if separated by <50 genes or <300 kb in either genome [65].

Syntenic orthologs were identified in the genomes of D. polystachya and other Dioscoreaceae species (D. zingiberensis and D. alata) using mSynOrths with default parameters [66]. Syntenic genes identified between the D. alata and D. polystachya genomes served as the foundation for determining the baseline gene retention rate, defined as the proportion of syntenic genes retained in both genomes relative to the total number of syntenic gene pairs, and the same analytical framework was applied to quantify gene retention within the dioscin biosynthetic pathway.

Comparison of expression levels between paralogs

Expression levels were compared between paralogs using RNA-seq data from leaves and tubers of D. polystachya plants. Paralogous gene pairs of the two haplotypes Hap1 and Hap3 were analyzed. The TPM values for genes from the two sets of chromosomes were counted using featureCounts [67]. Paralogous pairs with TPM values <2 were excluded to focus on expressed genes. Subsequent transcript quantification in D. polystachya was performed via Salmon (v0.13.1) with default parameters, enabling comprehensive statistical evaluation of gene expression patterns across all subgenomes.

Variation in TE distribution around genes

To assess TE density in the regions flanking genes in D. polystachya, a 100-bp sliding window (10-bp step size) was applied. Specifically, the nucleotides derived from TEs were counted across 5-kb regions upstream (5′) and downstream (3′) of each gene. The ratio of nucleotides from TEs to those not derived from a TE was calculated per 100-bp sliding window. Subsequently, the average TE ratio was determined for the D. polystachya genes and plotted to estimate the TE density in their flanking regions.

Ks analysis

The coding sequences of paralogous or orthologous gene pairs were aligned using Synorths [66]. The synonymous nucleotide substitution rate per synonymous site (Ks) was then calculated for each pair through the Nei and Gojobori method using KaKs_calculator [68]. Ks values were computed for intraspecific comparisons (D. polystachya, D. zingiberensis, D. alata) and interspecific comparisons (D. polystachya vs. D. zingiberensis, D. polystachya vs. D. alata, D. alata vs. D. zingiberensis), as well between Hap1 and Hap3 in D. polystachya, employing the equation Ks = 2μT, where μ denotes per year substitution rate and T displays divergence time [16].

Analysis of genome evolution

To dissect the phylogenetic relationship between D. polystachya and related species, genomic data from 12 species (D. alata, D. polystachya, D. rotundata, D. zingiberensis, Brachypodium distachyon, O. sativa, Ananas comosus, Musa acuminata, Phalaenopsis equestris, Asparagus officinalis, Colocasia esculenta, and A. thaliana) were subjected to gene family clustering analysis. Orthologous groups conserved across these genomes were identified through Orthofinder (v2.5.4) [69]. An ultrametric phylogenetic tree was reconstructed with r8s (v.1.81) [70]. Divergence times for each node were estimated via the MCMCtree program within the PAML package [71]. Gene family expansion and contraction dynamics were assessed using Cafe5 [72] to evaluate evolutionary trajectories across lineages.

Comparative analysis among homologous haplotypes and RNA-SEQ analysis

SNP and InDel variants were detected using Nucmer, with variant density quantified across genomic regions using a 300-kb sliding window. Heatmaps illustrating SNP and InDel distribution were generated to elucidate evolutionary relationships among six haplotype pairs (H1–H2, H1–H3, H1–H4, H2–H3, H2–H4, H3–H4) at chromosomal resolution. To validate the accuracy of phasing assignments, phylogenetic trees were reconstructed using SNPs between D. polystachya and the closely related species D. rotundata. IQ-TREE (v2.2.6) [73] was used to generate maximum-likelihood trees for each set of four homologous chromosomes.

RNA-seq analysis was performed on young stems, roots, tuber peels, and tuber flesh. Raw RNA-seq reads were trimmed and aligned with HISAT2 (v.2.2.1) [74] and analyzed for allele-specific expression using DESeq2, applying a false discovery rate (FDR) <0.05 and |log₂(fold-change)| > 1 [75].

Expression patterns from each haplotype were categorized as either biallelic (both alleles not significant) or imbalanced. The latter group was divided into three categories according to previous methods [76]: monoallelic expression of Hap1, monoallelic expression of Hap3, and higher expression of one allele. A criterion was established to determine the classification: haplotype-specific expression patterns in stems, roots, tuber peels, and tuber flesh. Specifically, genes with TPM > 2 for both Hap1 and Hap3 alleles and less than a 2-fold difference between the two haplotypes were classified as having ‘biallelic expression’. Genes with TPM > 2 for both alleles but at least a 2-fold difference in expression levels between the two haplotypes were defined as showing ‘higher expression of one allele’. When only one allele (either Hap1 or Hap3) had TPM > 2 and the other ≤2, the gene was defined as showing ‘Hap1 dominant expression’ or ‘Hap3 dominant expression’, respectively. Calculation of fragments per kilobase of exon model per TPM was performed using StringTie v2.1.4 [77].

Transcriptome datasets and clustering of gene expression profile analysis

Data filtering of transcriptome datasets of four tuber developmental stages at 150 (T1), 170 (T2), 190 (T3), and 210 (T4) days after planting was performed using SOAPnuke (v1.6.5), and clean reads were aligned using HISAT2 [74], assembled, and quantified using Bowtie2 (v2.4.5) [78]. Gene expression levels were normalized to TPM [79], and DEGs were identified using the DESeq2 package [75] with the criteria of a fold change >1.5 and an FDR-adjusted P value <.05. Gene ontology (GO) term and KEGG pathway enrichment of DEGs and heatmap generation were performed using the R package clusterProfiler [80].

Gene isolation, subcellular localization, real-time quantitative polymerase chain reaction, and end-point RT-PCR

The full-length coding sequence of Dp7-DR was cloned from D. polystachya tubers using a pClone007 versatile simple vector kit (Tsingke Biotechnology, Beijing, China). The sequences of proteins related to 7-DR from other species were identified by BLASTP search using NCBI and are listed in Table S14. A phylogenetic tree was reconstructed using MEGA7, employing a maximum likelihood model [81]. The promoter sequence and important cis-acting elements of Dp7-DR were analyzed using PlantCARE (https://bioinformatics.psb.ugent.be/webtools/plantcare/html/).

For subcellular localization analysis, the coding sequence of full-length Dp7-DR (excluding the stop codon) was amplified and cloned into the pBWA(V)HS-GLosgfp vector (Table S16). The resulting recombinant plasmid and the empty vector were separately introduced into Agrobacterium tumefaciens strain GV3101. Positive colonies were cultured in LB medium and then resuspended in infiltration buffer (10 mM MES, 120 μM AS, 10 mM MgCl₂, pH 5.6) to an OD₆₀₀ of 0.6. Bacterial suspensions carrying the recombinant or control vectors were mixed in a 1:1 (v/v) ratio with SPER-mCherry (an ER localization marker) [82]. GFP fluorescence signals were observed 3 days after infiltration using an LSM 880 confocal laser microscope (Carl Zeiss, Germany).

Total RNA from yam roots, stems, leaves, and tubers was extracted for qPCR analysis using a TaKaRa MiniBEST Plant RNA Extraction Kit (Beijing, China), followed by reverse transcription into first-strand cDNA with a PrimeScript RT Kit. Three biological replicates were used for each sample, and the 2^−ΔΔCt method was employed for analysis using ACTIN (DpolyH1c19G001005) as the reference gene [4]. Specific primers used for each gene are listed in Table S16.

Virus-induced gene silencing and transient overexpression in yam bulbils

The Sol Genomics Network (SGN) VIGS Tool (https://vigs.solgenomics.net/) [83] was used to predict the specific sequences for gene silencing. Following amplification, sequences were inserted into a pTRV2 vector carrying tobacco rattle virus RNA 2. The resulting construct was introduced into Agrobacterium strain GV3101. VIGS was conducted in yam bulbils following a protocol similar to that used for gladiolus [82]. A mixture of two Agrobacterium cell suspensions each containing pTRV1 or pTRV2 (1:1, v/v) was infiltrated into yam bulbils. Negative controls included an Agrobacterium cell suspension carrying pTRV1 and the empty pTRV2 vector (TRV). Yam bulbils subjected to VIGS were collected for real-time quantitative polymerase chain reaction (RT-qPCR) analysis to evaluate the effectiveness of gene silencing.

The full-length coding sequence of candidate genes was amplified and cloned into the pSuper1300 vector to create overexpression constructs (Table S16). Bulbils of D. polystachya was infiltrated with an Agrobacterium suspension harboring the target plasmid or the empty vector as control in infiltration buffer, adjusted to a final OD₆₀₀ of 0.8. Tissue samples were collected from the infiltration sites for subsequent metabolite quantification and gene expression analysis.

Extraction of dioscin and cholesterol and content measurements

The contents of dioscin and cholesterol were measured using an HPLC system (Shimadzu Nexera LC-30A) coupled with a tandem mass spectrometer (MS/MS) (AB Sciex QTrap 4500). Briefly, 200 mg of samples were homogenized with 1.2 ml of isopropanol, followed by ultrasonic extraction for 10 min at 4°C and centrifugation for 10 min. The resulting supernatant was collected and filtered using a 0.22-μm microporous filter membrane for cholesterol measurement. For dioscin measurements, the supernatant was concentrated to dryness at 4°C, resuspended in 70% (v/v) methanol, and filtered. Samples were separated using a Waters Acquity UPLC BEH C₁₈ column (100 mm × 2.1 mm × 1.8 μm) at 45°C. Data were collected in dynamic-multiple reaction monitoring mode, with specific transitions controlled at specific times based on the retention time of the analytes. Concentrations were determined using an external calibration curve of the corresponding standard compounds (Fig. S10).

Yeast one-hybrid assay, dual-luciferase reporter assay, and electrophoretic mobility shift assay

In the Y1H assay, the full-length DpbZIP12 coding sequence, obtained from cDNA derived from total RNA isolated from D. polystachya tubers, was cloned into the pGADT7 vector (Table S16). A promoter fragment of Dp7-DR (475 bp) was amplified from genomic DNA extracted from young D. polystachya leaves and cloned into the pHIS2 vector (Table S16). After cotransforming the yeast strain Y187 with the pHIS2-pro-Dp7-DR reporter plasmid and the pGADT7-DpbZIP12 effector plasmid [84], positive yeast colonies were selected and plated onto synthetic defined (SD) medium lacking leucine and tryptophan (SD/−Leu/−Trp) for 3 days. The positive colonies were resuspended and spotted onto the SD/−Leu/−Trp medium and SD/−Leu/−Trp/−His medium containing 40 mM 3-amino-1,2,4-triazole.

For the dual-luciferase reporter assay, the full-length DpbZIP12 coding sequence was cloned into the pGreenII 62-SK vector to generate the effector plasmid (Table S16), which was then introduced into A. tumefaciens strain GV3101. A Dp7-DR promoter fragment (475 bp) was cloned into the pGreenII 0800-LUC vector to create a reporter plasmid (Table S16), which was introduced into Agrobacterium strain GV3101. The empty vector served as the negative control. Agrobacterium cell suspensions, each harboring the effector or reporter construct, were mixed and co-infiltrated into the leaves of N. benthamiana plants. A dual-luciferase reporter assay was performed as described previously [84]. The activities of Firefly luciferase (LUC) and Renilla luciferase (REN) were quantified using a commercial assay system (Promega) and detected with a Synergy™ H1 hybrid multimode microplate reader (BioTek). Experiments were performed independently at least three times (n = 3). Promoter activity was evaluated using relative LUC/REN ratios.

For the EMSA, the full-length DpbZIP12 coding sequence was cloned into the pDONR207 entry vector (Invitrogen) using the Gateway system, and then recombined into the pHMGWA expression vector. The MBP-DpbZIP12 plasmid was introduced into Escherichia coli BL21 cells, and protein production was induced with 0.5 mM IPTG. EMSA was then conducted using a LightShift Chemiluminescent EMSA Kit (Pierce Biotechnology, Rockford, IL, USA) with biotin-labeled DNA probes (5′-TTGTGCACACGTGTCGTGACAGGTCTCACACGTGGCAAGT-3′). To test the specificity of binding, unlabeled competitor probes (10×, 20× and 50× excess) were added to some samples. The reaction mixtures were separated on a polyacrylamide gel and visualized using a ChemiDoc XRS system (Bio-Rad Laboratories, Hercules, CA, USA).

Statistical analysis

Statistical analyses were carried out using a paired two-tailed Student’s t-test in SPSS version 19.0, with significance levels indicated as ^*P < 0.05, ^**P < 0.01, and ^***P < 0.001 for comparisons between two groups. Figures were generated using Origin 8.0. Data are presented as mean values ± standard deviation.

Author contributions

Y.H. and Q.Z. conceived and designed the experiments. N.S., Y.X., and T.L. performed most of the experiments. J.S. extracted the genomic DNA. N.S., Y.X., T.L., and S.W. conducted genome assemblies, genome annotations, and repeat content analysis. N.S., P.W., Q.Z., and T.C. performed the RNA-seq, mapping, and data processing. A.A., S.L., J.L., and Z.L. commented on the manuscript. N.S., Y.X., Y.H., and Q.Z. wrote the paper with input from all authors.

Data availability

The assembled genome sequences are available under accession code PRJCA034592 at the Genome Warehouse (GWH) of the China National Center for Bioinformation (CNCB-NGDC, https://ngdc.cncb.ac.cn/gwh/), and all transcriptome raw sequence read data were uploaded to the China National GeneBank DataBase (CNGBdb, https://db.cngb.org/) website with the accession number CNP0008466. For comparative analysis, the genome of D. alata was obtained from YamBase (https://yambase.org/), and the gene, protein, and annotation files for D. rotundata were sourced from http://genome-e.ibrc.or.jp/home/bioinformatics-team/yam.

Conflicts of interest statement

The authors declare that they have no conflicts of interest.

Supplementary material

Supplementary material is available at Horticulture Research online.