A complete telomere-to-telomere assembly of the Spinacia oleracea genome reveals Y chromosome evolution and centromere landscape

Dear Editor,

Spinach (Spinacia oleracea L.) is a diploid plant with n = 6 chromosomes (2n = 12) and is one of the world’s most important green leafy vegetables. The genus Spinacia is well suited for studying sex chromosome evolution because it is dioecious, has an XY sex determination system, and includes two wild species (She et al., 2023). The first chromosome-level spinach genome, assembled using next-generation sequencing reads in 2017, had only 47% of the sequence anchored to the chromosomes (Xu et al., 2017). In 2023, we reported a spinach genome (termed Sp_YY_v1), sequenced from a homozygous YY male of the inbred line 10S15 using Oxford Nanopore Technology (ONT) long reads. However, the assembly still contained 74 unmapped contigs (38.2 Mb) and 109 gaps (She et al., 2023). Similar limitations are also present in several recently assembled spinach genomes (Ma et al., 2022; She et al., 2023, 2024a, 2024b), and these incomplete genome assemblies have hindered the exploration of complex genomic regions such as centromeres and telomeres.

Here, we present a telomere-to-telomere (T2T) genome assembly of a YY male of inbred line 10S15. We integrated 90 Gb of PacBio High Fidelity (HiFi) data (∼90×), 65 Gb of ultra-long ONT reads (65× coverage; N50 = 100 kb), 30 RNA sequencing (RNA-seq) datasets from 10 different tissues, and 17 Gb of Isoform Sequencing (Iso-Seq) long-read data (Supplemental Tables 1 and 2; Supplemental Figures 1 and 2). HiFi+ONT and ONT reads were assembled independently using hifiasm and NextDenovo, respectively (Supplemental Figure 3). After removing contaminant contigs (Supplemental Methods; Supplemental Figure 4), we obtained eight contigs from the HiFi+ONT assembly (925.7 Mb; contig N50 = 139 Mb) and 41 contigs (925.7 Mb; contig N50 = 111 Mb) from the ONT assembly. Notably, all contigs from both assemblies could be anchored to six pseudo-chromosomes using a reference-guided approach based on the published Sp_YY_v1 assembly (She et al., 2023), with high alignment identity (> 90%; Supplemental Figure 5). The two gaps in the HiFi+ONT assembly were then filled using the ONT-based assembly and corrected ultra-long ONT reads (Supplemental Figure 6), yielding a gap-free T2T genome sequence of 926 Mb with 12 telomeres (AAACCCT telomeric repeats), designated Sp_YY_v2 (Figure 1A; Supplemental Tables 3 and 4).

Figure 1.

Overview of the gap-free telomere-to-telomere (T2T) spinach reference genome Sp_YY_v2 and characterization of the complete Y chromosome and centromeric regions.

(A) Collinearity analysis between Sp_YY_v2 and Sp_YY_v1 assemblies. Gray lines indicate collinear regions and orange indicates inversion regions. Three large-scale inversions (>1 Mb) are indicated by purple hexagons on chr1, chr3, and chr5. The sex-linked region (SLR) on chr4 includes a Y duplication region (YDR, yellow) and its flanking inversions (IV, green). The distributions of centromeric retrotransposons of maize (CRM) and the CRM probe, confirmed to be centromeric by fluorescence in situ hybridization (Li et al., 2019), are highlighted in red and blue, respectively.

(B) Transcript abundance (transcripts per million) of Sp4g026740.1 at different developmental stages of the stamen, based on RNA-seq data from Ma et al. (2022). Statistical significance was assessed using the Student–Newman–Keuls test (ɑ = 0.01).

(C) Neighbor-joining phylogenetic tree based on 14 248 Y-specific SNPs across 13 S. tetrandra (8 males and 5 females), 9 males of S. turkestanica, and 70 males of S. oleracea. The two groups of S. oleracea, SOLG1, which was mainly from Asia, and SOLG2, which was from Europe, were inferred from our previous study (She et al., 2024a). STU, S. turkestanica; STE, S. tetrandra; YG1–YG3, Y groups 1–3.

(D) Genotype distribution of 1000 randomly selected Y-specific SNPs across 92 Spinacia accessions. SNPs identical to Sp_YY_v2 are labeled as “Ref”, alternative alleles as “Alt”, and heterozygous calls as “Heter”. The bottom figure shows fiaxtion index values between different Y groups in recombining pseudo-autosomal regions, pericentromeric pseudo-autosomal regions (pPARs), and SLRs.

(E) Genome-wide distribution of gene density and TEs across all chromosomes in the Sp_YY_v2 assembly.

(F) Comparison of Gypsy retrotransposon clades in centromeric and non-centromeric regions of Sp_YY_v2. The pink rectangles indicate the significant enrichment of CRM in centromeric regions. Significance was determined using the Wilcoxon test. ns, not significant; ∗∗P < 0.01; ∗P < 0.05.

(G) Correlation analysis between CRM copy number and the percentage of CRM-related genes within each centromere on six chromosomes. The blue line represents the regression line; the gray shading indicates the 95% confidence interval. R, Pearson correlation coefficient.

(H) DNA methylation levels in the centromeric regions and the 2 Mb flanking regions across 6 chromosomes.

We employed multiple strategies to evaluate the accuracy and completeness of the Sp_YY_v2 assembly. Mapping rates exceeded 99.99% for both HiFi and ONT long reads (Supplemental Table 5) and 98.78% for RNA-seq data (Supplemental Table 6), with uniform coverage across the genome (Supplemental Figure 7). The Sp_YY_v2 assembly exhibited a quality value of 51.9 and a Benchmarking Universal Single-Copy Orthologs score of 99.2%, indicating a highly accurate and complete T2T assembly (Supplemental Table 7). The long terminal repeat (LTR) assembly index of 24.18 also suggested that Sp_YY_v2 can be regarded as a “gold-standard” genome (LTR assembly index > 20).

Compared to Sp_YY_v1 (She et al., 2023), the Sp_YY_v2 assembly significantly improves completeness and accuracy (Supplemental Table 8). The T2T assembly successfully anchored 72 previously unmapped contigs (38.2 Mb) and filled all 109 gaps (Supplemental Table 9), primarily from telomeric and centromeric regions (Supplemental Figure 8). Additionally, we corrected 14 structural inversions across chromosomes 1–5, ranging from 33 kb to 1.6 Mb (Figure 1A; Supplemental Table 10), which were validated using corrected ultra-long ONT reads (Supplemental Table 10; Supplemental Figure 9). The Sp_YY_v2 genome contains 701 Mb (75.71%) repetitive sequences, which is similar to Sp_YY_v1, and most of them are transposable elements (TEs), especially LTR retrotransposons (58.63%) (Supplemental Table 11). We predicted 28 247 protein-coding genes, 84% of which were supported by Iso-Seq data or showed expression (transcripts per million > 0.5) in at least one sample (Supplemental Table 2). BUSCO analysis identified 99.3% complete genes, an improvement over the 95.9% found in Sp_YY_v1 (Supplemental Tables 7 and 8), and more than 96.84% of genes could be annotated in the Kyoto Encyclopedia of Genes and Genomes, Non-redundant Protein Sequence, Swiss-Prot, Gene Ontology, Pfam, and Translated European Molecular Biology Laboratory databases (Supplemental Table 12).

Even though the YY male plant (with 0.12% heterozygosity) was derived from an inbred line purified for 10 generations, we identified 20 055 heterozygous loci based on resequencing data (Supplemental Methods), which corresponded mainly to 43 continuous heterozygous regions (top 1% heterozygosity windows) (Supplemental Figure 10A; Supplemental Table 13). Gene Ontology enrichment analysis indicated that these were enriched for genes involved in plant resistance, such as response to oxidative stress, response to stress, and response to stimulus (Supplemental Figure 10B; Supplemental Table 14). We also identified 802 heterozygous presence/absence variations based on long reads (Supplemental Methods), 55% of which were TEs, especially LTR and DNA elements (Supplemental Figure 11), suggesting that TEs probably contribute to variation between haplotypes.

Recombination suppression during sex chromosome evolution leads to TE accumulation and gene degradation, complicating complete Y chromosome assembly. Here, we report a complete Y chromosome (chromosome 4 [chr4]) with a length of 205.40 Mb, an increase of 9.92 Mb relative to Sp_YY_v1 (195.48 Mb), primarily from approximately 5 Mb of newly anchored contigs on the short arm of the Y chromosome (Supplemental Figure 8). Based on this, we resolved the complete architecture of the Y chromosome, including the pericentromeric region (PR), which undergoes infrequent recombination, and the sex-linked region (SLR) (Supplemental Figure 12; Supplemental Table 15). PRs showed a higher TE content (83.50%) and lower gene density (22 genes/Mb) than non-PRs (31.59% TE; 45 genes/Mb) (Supplemental Figures 12 and 13). The Y chromosome contains a 24.89 Mb SLR, including 10.74 Mb of male-specific regions, which we termed the Y duplication region (YDR) and its flanking inverted repeats (Figure 1A). Sex-determining genes tend to be located in the YDR, where we identified a total of 57 protein-coding genes (8 more than in Sp_YY_v1), 93% of which were supported by Iso-Seq or Illumina RNA-seq data (Supplemental Figure 14). Based on annotation and expression patterns of these genes across 71 samples from different tissues (Supplemental Tables 2, 16, and 17), we identified a novel candidate male-determining gene, Sp4g026740.1 (SpRAD), which is a small-MYB (234-bp coding sequence) RADIALIS-like gene. SpRAD is predominantly expressed in stamens and shows significantly higher expression (P < 0.01, Kruskal-Wallis test) during early stamen development (Figure 1B; Supplemental Figure 15; Supplemental Table 18). Unlike its homolog in hexaploid persimmon, which promotes gynoecium development (Masuda et al., 2022), SpRAD expression is not detectable in female flowers, suggesting a divergent role (Supplemental Figure 15; Supplemental Table 18). SpRAD is shared between cultivated spinach (S. oleracea) and its closest wild relative (Spinacia turkestanica). While synonymous site divergence with Spinacia tetrandra suggests a divergence time of ∼4.6 million years ago (mya; 95% CI: 4.4–4.7; Supplemental Figure 16; Supplemental Table 19), our results indicate that SpRAD originated after the split from S. tetrandra (∼6.3 mya) but before the divergence between S. oleracea and S. turkestanica (∼0.8 mya) (She et al., 2024b), which is consistent with the occurrence of a turnover event (including a male-determining factor) in the recent common ancestor of S. oleracea and S. turkestanica (She et al., 2025).

To further investigate Y chromosome evolution in S. oleracea and S. turkestanica, we applied a SNP-calling pipeline (Supplemental Figure 17; Supplemental Methods) and identified 14 248 Y-specific SNPs using 133 S. oleracea accessions (43 females and 70 males) and 21 S. turkestanica accessions (12 females and 9 males) from our previous study (Supplemental Table 20) (She et al., 2024a). We constructed a neighbor-joining tree using 92 Spinacia accessions, including 70 male S. oleracea individuals—representing two major lineages, SOLG1 (which first spread to Asia) and SOLG2 (which later spread to Europe) (She et al., 2024a)—9 male S. turkestanica, and 13 S. tetrandra accessions as outgroups (Figure 1C). The majority of SOLG1 and SOLG2 accessions formed distinct clades, referred to as YG1 and YG3, respectively, with the SOLG1 lineage diverging earliest—potentially reflecting the domestication origin of spinach—whereas S. turkestanica and the remaining S. oleracea accessions clustered into a third group (termed YG2), a pattern also supported by principal-component analysis (Supplemental Figure 18A). Notably, the S. turkestanica Y chromosomes did not form the base branch of the phylogeny but instead clustered with those of some of the S. oleracea accessions, suggesting potential gene exchange between the two species. Three distinct Y haplotypes were identified, each corresponding to one of the three Y groups (Figure 1D; Supplemental Figure 18B), with divergence significantly concentrated in the low-recombination regions of the pericentromeric pseudo-autosomal regions (pPARs) and the SLR, where suppressed recombination facilitates the accumulation of genetic variants (van der Bijl and Mank, 2025). Fixation index analysis between the different Y groups confirmed significantly higher divergence in the pPARs and SLR compared to the recombining pseudo-autosomal regions (rPARs) (Figure 1D; Supplemental Figure 19).

Our gap-free T2T genome also allowed us to explore the architecture and context of centromeres in spinach. The enrichment of the centromeric retrotransposons of maize (CRM) probe, which was confirmed to be centromeric by fluorescence in situ hybridization (Li et al., 2019), revealed six putative centromeres in Sp_YY_v2 (Figure 1A and 1E). A total of 21 CRM elements were detected, 72% (15) of which were localized to centromeric regions (Figure 1A and 1E; Supplemental Figure 20; Supplemental Table 21). We delineated the locations and boundaries of the 6 centromeres by integrating the 2 approaches above (Supplemental Table 22) and confirmed completeness of the centromeres by mapping the corrected ultra-long ONT reads (> 30 kb) to the centromeres of Sp_YY_v2 (Supplemental Figure 21). The 6 centromeres ranged from 1.96 Mb (chr5) to 4.49 Mb (chr3) in length, with no correlation (R = 0.028) to chromosome size (Supplemental Figure 22).

TEs comprised ∼91% of centromeric sequences, predominantly LTRs (85%) (Supplemental Table 23; Supplemental Figure 23A). Comparison between TEs from centromeric and non-centromeric regions revealed that only Gypsy LTRs were significantly enriched (P < 0.001, Wilcoxon test) in centromeres (Figure 1E; Supplemental Figure 23B), comprising 44% of centromeric and 16% of non-centromeric sequences, suggesting that Gypsy had a major influence on the formation of centromeres. In contrast to most other plant centromeres, no centromere-specific tandem repeats were detected, and the low similarity between repetitive sequences further supported this result (Supplemental Figure 24), consistent with a previous report on cotton centromeres (Huang et al., 2024).

The centromere-located Gypsy elements could be classified into 9 clades, of which 2 were predominant in the centromeric regions: Retand (number 66) and CRM (number 15), (Figure 1F), accounting for 41% and 22% of the Gypsy sequences, respectively (Supplemental Table S24). Notably, compared with Retand, CRMs were centromere-specific sequences (Figure 1F; Supplemental Figures 20 and 25), indicating that CRM distribution is a useful marker for identifying spinach centromeres. It is generally accepted that centromeres are composed largely of repetitive sequences rather than coding genes (Huang et al., 2024). However, we found that chr1, chr3, and chr4 have higher gene density within their centromeres than the others (Supplemental Figure 26). Moreover, the proportion of genes overlapping with CRM (termed CRM-related genes) in centromeres was higher on chr1, chr3, and chr4 (mean 55.5%) compared to other chromosomes (mean 16.7%). This proportion was significantly positively correlated with CRM copy number (R = 0.9, P = 0.015) (Figure 1G; Supplemental Figure 27), suggesting that these genes may facilitate CRM expansion. Interestingly, the majority of CRM-related genes (84.1% on average) encode reverse transcriptases, which are essential for obtaining newly copied CRMs (Supplemental Figure 27; Supplemental Table 25). This likely explains the high CRM copy number in the centromeres of chr1, chr3, and chr4. Additionally, analysis of 30 RNA-seq datasets from inbred line 10S15 revealed that centromeric genes have significantly lower expression levels than non-centromeric genes (P < 2.22e−16, Wilcoxon test; Supplemental Figure 28), consistent with findings in cotton (Chang et al., 2024). DNA methylation analysis showed reduced CH (A/C/T) G and CHH methylation levels in centromeric regions relative to flanking regions of six chromosomes, whereas CpG methylation levels remained similar (Figure 1H). This pattern differs from that observed in pepper and cotton; in pepper, DNA methylation is lower in centromeric than in non-centromeric regions (Chen et al., 2024), whereas in cotton, no such reduction is observed (Chang et al., 2024).

In summary, we successfully completed the first gap-free T2T spinach genome assembly that captured all 6 centromeres and 12 telomeres. This complete genome enables detailed characterization of centromere and Y chromosome structure and evolution. The high-quality spinach reference genome provides a valuable resource for advancing spinach research and facilitating the genetic improvement of spinach.

Data and code availability

The raw data reported in this paper have been deposited in the BIG Data Center (https://ngdc.cncb.ac.cn) under accession number CRA020566. The genome assembly and presence/absence variations between the haplotypes are available at https://zenodo.org/records/15395809.

Funding

This work was supported by the Central Public-Interest Scientific Institution Basal Research Fund (Y2023QC07), the Chinese Academy of Agricultural Sciences Innovation Project (CAAS-ASTIP-IVFCAAS), and the China Agricultural Research System (CARS-23-A-17).

Acknowledgments

No conflict of interest is declared.

Author contributions

W.Q. designed the study. H.S., Z.L., and W.Q. analyzed the data. H.S. and Z.L. wrote the manuscript. W.Q., Z.L., H.Z., and Z.X. prepared the samples. W.Q., J.W., X.W., and F.C. revised the manuscript.

Published: June 7, 2025