Genome of root celery and population genomic analysis reveal the complex breeding history of celery

Summary

Celery (Apium graveolens L.) is an important vegetable crop in the Apiaceae family. It comprises three botanical varieties: common celery with solid and succulent petioles, celeriac or root celery with enlarged and fleshy hypocotyls and smallage or leaf celery with slender, leafy and usually hollow petioles. Here we present a chromosome‐level genome assembly of a celeriac cultivar and a comprehensive genome variation map constructed through resequencing of 177 representative celery accessions. Phylogenetic analysis revealed that smallage from the Mediterranean region represented the most ancient type of cultivated celery. Following initial domestication in this region, artificial selection has primarily aimed at enlarging the hypocotyl, resulting in celeriac, and at solidifying the petiole, leading to common celery. Selective sweep analysis and genome‐wide association study identified several genes associated with hypocotyl expansion and revealed that the hollow/solid petiole trait directly correlated with the presence/absence of a NAC gene. Our study elucidates the complex breeding history of celery and provides valuable genomic resources and molecular insights for future celery improvement and conservation efforts.

Introduction

Celery (Apium graveolens L., 2n = 2x = 22), an aromatic herb that grows annually or biennially (Li et al., 2018; Marongiu et al., 2013), is one of the most important vegetables in the Apiaceae family. As a low‐calorie and nutrient‐rich vegetable with medicinal properties for reducing blood glucose and serum lipids, it is abundant in vitamins, carotenoids, cellulose, flavonoids, volatile oils and antioxidants, which could help protecting cardiovascular health, and strengthening the heart. Nowadays, celery has been cultivated worldwide as a vegetable, and also widely utilised in the pharmaceutical and fragrance industries (Fazal et al., 2012; Kooti et al., 2014; Nagella et al., 2012; Sowbhagya, 2014).

The precise origin of celery remains uncertain due to the widespread distribution of wild celeries. However, numerous studies indicate that it likely originated in the Mediterranean region (Malhotra, 2006; Megaloudi, 2005; Quiros, 1993). Wild celeries have been discovered in archaeological sites dating back to the 7th–9th centuries B.C., including the Heraion of Samos and Kastanas in Greece (Megaloudi, 2005). Celery was first cultivated as early as 400 B.C. in Egypt and the Roman Empire for its medicinal properties. However, it wasn't domesticated as a true vegetable in Europe until the 16th century (Quiros, 1993). The introduction of celery to China can be traced back to around 2000 years ago in the Han dynasty. It was mentioned as a vegetable named 胡芹 (foreign celery) in the Chinese work “Qimin Yaoshu” by Jia Sixie in the sixth century A.D. Thus, although celery is believed to be originally from the Mediterranean basin, its cultivation for vegetable purpose began in China.

Depending on the specific part of the plant consumed, celery is typically categorised into three botanical varieties, A. graveolens var. dulce, A. graveolens var. rapaceum and A. graveolens var. secalinum. A. graveolens var. dulce, commonly known as celery or cultivated celery (hereafter referred to as ‘common celery’), is well known for its solid and succulent petioles, which is sweet in flavour and tender in texture. As the most economically important celery variety, common celery is cultivated worldwide and can be either cooked in stir‐fries or consumed raw in salad and juice. A. graveolens var. rapaceum, commonly known as celeriac or root celery, is characterised by its enlarged, fleshy hypocotyl, and commonly utilised in stews and soups. This variety was originally cultivated in the Mediterranean region and northern Europe. A. graveolens var. secalinum, commonly known as smallage or leaf celery, has slender, leafy, and usually hollow petioles with strong flavour and aroma. It can be stir‐fried or used as a flavorful addition to soups and is primarily popular in Asian and Mediterranean countries (Bruznican et al., 2020; Kokotkiewicz and Luczkiewicz, 2016; Quiros, 1993).

Unlike common celery, celeriac is characterised by its expanded hypocotyls, hollow petioles and strong flavour. To date, three genome assemblies of common celery have been reported (Cheng et al., 2022; Li et al., 2020; Song et al., 2020); however, none is available for celeriac, restricting the understanding of genetic differences among different varieties. In this study, we de novo assembled a chromosome‐scale reference genome of celeriac. Comparative analysis between celeriac and common celery genomes identified approximately 186 000 high‐confidence structure variants (SVs). We then conducted genome resequencing of 177 accessions of the three celery varieties collected from 26 countries and identified about 70 million single nucleotide polymorphisms (SNPs) and 13 million small insertions/deletions (indels). Phylogenetic, population structure and genetic diversity analyses revealed the breeding history of celery varieties. Furthermore, selective sweep scanning and genome‐wide association studies (GWAS) pinpointed several candidate loci associated with hypocotyl expansion and petiole hollowness. This study enhances our understanding of the origin and diversity of celery varieties and sheds light on the formation of enlarged hypocotyl and hollow/solid petioles. Moreover, this study provides valuable genome resources for further advancements in biological research and breeding strategies aimed at improving celery quality.

Results

Sequencing and assembly of a celeriac genome

The genome of the celeriac cultivar Alabaster (PI 662464) was de novo assembled using highly accurate long PacBio HiFi reads and then anchored into pseudochromosomes with Hi‐C data. A total of 79.62 Gb of HiFi reads were generated with an N50 length of 15.85 kb, covering approximately 23.32× of the celeriac genome with an estimated size of 3.41 Gb (Table S1 and Figure S1). The assembly had a total size of 3.26 Gb, containing 836 contigs with an N50 size of 66.05 Mb. Using the Hi‐C data, a total of 763 assembled contigs with a total length of 3.25 Gb were ordered and oriented into 11 pseudochromosomes ranging from 216.4 to 348.8 Mb in length, accounting for 99.76% of the assembly (Figure 1a and Table S2). BUSCO (Manni et al., 2021) assessment revealed that about 99.07% of the core conserved plant genes were fully represented in the Alabaster genome assembly. The annotation of the long terminal repeat (LTR) indicated an LTR assembly index (LAI) (Ou et al., 2018) of 15.79, suggesting a level of assembly comparable to reference quality. The assessment using Merqury (Rhie et al., 2020) revealed a consensus quality value (QV) of 53.19 and a completeness score of 97.76% (Table S2). Furthermore, Illumina short reads were mapped back to the assembly, resulting in a mapping rate of 99.95% and a coverage of 99.98% (Table S3). Collectively, these comprehensive assessments affirmed that the quality of the Alabaster genome assembly was high, better than the genome assemblies of two common celery cultivars, Ventura and Baili (Tables S2 and S3).

Figure 1.

Genome of Apium graveolens var. rapaceum Alabaster. (a) Hi‐C interaction heatmap of the Alabaster genome. (b) Features of the Alabaster genome. (i) Ideogram of the 11 chromosomes in Mb scale; (ii) repeat content (% nucleotides per Mb); (iii) gene density (number of genes per Mb); (iv) densities of SVs (outer) and SNPs (inner) in comparison to the Ventura genome (number of SVs and SNPs per Mb). (c) Synteny between the Alabaster genome and the genomes of common celery Ventura and Baili. Syntenic regions are indicated by grey colour, and inversions longer than 5 Mb are indicated by orange colour.

About 2.90 Gb of repetitive sequences were identified in the Alabaster assembly, accounting for 88.9% of the entire genome, a proportion similar to that observed in the Ventura and Baili genomes (Figure 1b and Table S4). The majority of transposable elements (TEs) belonged to the LTR category, with a total length of 2.66 Gb (81.7% of the genome). The two predominant types of LTR, Copia and Gypsy, constituted 39.2% and 30.6% of the genome, respectively (Table S4). Using a combination of ab initio, homology‐based and transcript‐guided methods, we predicted a total of 40 313 protein‐coding genes in the Alabaster genome. These genes had an average length of 3027 bp, an average coding sequence (CDS) length of 1093 bp and an average of 4.55 exons (Table S5). Notably, functional annotations could be assigned to 38 582 genes (95.7%) through comparison with public databases (Table S6).

Comparison of the celeriac and common celery reference genomes

Celeriac and common celery have undergone distinct selection processes, leading to significant disparities in crucial traits, which also determine their respective primary edible parts. To shed light on the genetic basis underlying the divergent phenotypes between these two varieties, we conducted a direct comparison of the celeriac genome (Alabaster) with the genomes of two common celery cultivars (Ventura and Baili). While the genomes exhibited good collinearity (Figure S2), a number of large SVs were evident (Figure 1c), including 19 large inversions (>5.0 Mb) between the Alabaster and Ventura genomes and 19 between the Alabaster and Baili genomes, with 12 of these inversions shared between the Ventura and Baili genomes compared to the Alabaster genome. These inversions were subsequently validated through both HiFi read mapping and Hi‐C maps (Figure S3).

Given the high similarity observed between the two common celery genomes and the superior quality of the Ventura genome, only the Ventura assembly was used in the subsequent analyses. A total of 185 954 SVs were identified between Alabaster and Ventura genomes. Among them, 158 097 (85.0%) were located in intergenic regions, while 27 857 (15.0%) were detected within gene bodies or promoter regions (defined as 2 kb upstream of the gene body) in either the Alabaster or Ventura genome (Table S7). Genes with CDS or promoters overlapping SVs showed functional enrichment in the biological processes of defence responses, cellular communication, signal transduction, responses to stimuli as well as responses to abscisic acid and alcohol (Figure S4). As celeriac and common celery are cultivated globally and exhibit different disease resistance for same pathogens and pests (Bruznican et al., 2020), these SVs may contribute to their differences in environmental adaptation and disease resistance. Furthermore, 673 genes in the Alabaster genome and 233 genes in the Ventura genome exhibited presence/absence variations (Tables S8 and S9). Notably, the unique genes identified in the Alabaster genome were enriched with those involved in glutathione catabolic processes, programmed cell death and response to reactive oxygen species, while those unique to the Ventura genome were enriched with genes involved in lipid transport and responses to toxic substances (Figure S5). Thus, these SVs and unique genes in celeriac and celery genomes might contribute to their phenotypic divergence and adaptation to distinct environments.

Population genome resequencing and analyses

To explore the genetic diversity and breeding history of celery, a total of 177 accessions, including 58 common celery, 60 celeriac and 59 smallage or leafy celery accessions, from 26 countries were selected for genome resequencing (Figure 2a, Table S10 and Figure S6). In total, we generated 10.4 Tb of cleaned sequence data, yielding an average depth of 18.0× for each accession (Table S11). These sequences were mapped to the Alabaster genome, leading to the detection of 70 274 110 SNPs and 12 928 426 indels. Among these variants, 3 882 537 SNPs (5.52%) and 409 557 indels (3.17%) located in gene coding regions, of which 1 753 532 SNPs lead to nonsynonymous mutation and 35 950 SNPs caused stop codon alterations. Additionally, 280 992 indels (2.17%) could cause frameshift mutations (Table S12).

Figure 2.

Genetic diversity of A. graveolens. (a) Geographic distributions of the 177 A. graveolens accessions. The size of the circle is proportional to the number of accessions. (b) Maximum‐likelihood phylogenetic tree and model‐based clustering of A. graveolens accessions. Different numbers of ancestral kinships (K from 2 to 5) are shown. Branch colors in the tree denote different varieties. Scale bars: 10 cm. (c) PCA plot of A. graveolens accessions. (d) Group‐specific LD decay plots. (e) Nucleotide diversity (π) and population divergence (F _ST) across the five groups. The value in each circle represents the nucleotide diversity for the corresponding group, while the value beside each line indicates population divergence between the two groups.

A phylogenetic tree was constructed using 72 677 SNPs at fourfold degenerate sites, with Petroselinum crispum as the outgroup, revealing the classification of celery accessions into three main clades (Figure 2b). Clade I was the basal clade, exclusively comprising smallage accessions primarily from the Mediterranean region. Clade II mainly consisted of celeriac with enlarged hypocotyls, while Clade III encompassed common celery and Chinese leafy celery, which are consumed for their petioles (Figures 2b and S7). We identified highly differentiated genomic regions (top 5% fixation index, F _ST) between Clade II and Clade III (Figure S8), which contained a total of 3440 genes (Table S13). Functional enrichment analysis revealed that these genes were primarily associated with biological processes such as cellular response to alcohol, cellular response to abscisic acid stimulus and cell division (Figure S9).

To better understand the breeding history of celery and explore the genetic basis of notable traits, such as enlarged hypocotyls, we further divided the population into five groups based on the genetic structure, phenotypic traits and geographic origins: G1 corresponding to Clade I, G2 and G3 within Clade II and G4 and G5 within Clade III (Figure 2b and Table S10). Principal component analysis (PCA) also supported this grouping (Figure 2c). A few accessions that could not be reliably assigned to specific groups based on the aforementioned information were excluded from subsequent comparative analyses. G1 occupied the most basal position on the phylogenetic tree, suggesting it may be genetically closest to wild celeries, consistent with the assumption that celery originated in the Mediterranean region (Kokotkiewicz and Luczkiewicz, 2016; Megaloudi, 2005). Notably, G1 also exhibited the shortest linkage disequilibrium (LD) decay distance (~2.25 kb; r ²  = 0.2604) (Figure 2d).

G2 and G3 denoted celeriac accessions characterised by enlarged hypocotyls. However, these two groups exhibited divergence in both geographical distribution and phenotypic traits. Almost all G2 accessions were from the Mediterranean coastal region, whereas G3 was predominantly composed of European accessions (Table S10). Notably, most elite celeriac cultivars belonged to G3, distinguished by significantly enlarged globular hypocotyls observed at the ground level. In contrast, G2 accessions displayed limited hypocotyl expansion (Figure S10), suggesting their primitive status in celeriac domestication. The LD decay distance in G3 (~675.70 kb; r ² = 0.3832) was notably greater than that observed in G2 (~2.96 kb; r ² = 0.2839), while the nucleotide diversity in G3 (π = 1.47 × 10⁻³) was substantially lower compared to that in G2 (π = 2.55 × 10⁻³). Consequently, we tended to regard G2 as semi‐domesticated celeriac, while G3 was classified as improved cultivar.

The primary consumed part of accessions in G4 (Chinese leafy celery) and G5 (common celery) is the petiole. Almost all G5 accessions featured solid and succulent petioles, whereas both solid and hollow petioles were common in G4. Interestingly, the only two G5 accessions with hollow petioles were collected from the United States and exhibited distinct genetic structure compared to other common celery cultivars. Traditional Chinese celery typically produces hollow stalks, and the solid‐stalked form observed in G4 cultivars is believed to result from introgressions from common celery (Xiao et al., 2021; Zhang, 2008). The G4 group of Chinese leafy celery showed comparable LD decay distance (~5.92 kb; r ² = 0.2884) and nucleotide diversity (π = 2.36 × 10⁻³) to the Mediterranean smallage group (G1). Both G4 and G1 are classified as A. graveolens var. secalinum for their leafy petioles and unexpanded hypocotyls, but G4 displayed a distinct genetic structure pattern from G1, suggesting their different genetic components. Considering the records that celery was introduced into China from the Caucasus region during the Han Dynasty and gradually cultivated into a type with slender petioles (Xiao et al., 2021; Zhang, 2008), we propose that Chinese smallage celery emerged through local independent breeding after the spread of ancient Mediterranean smallage celery to China. G5, the common celery group, exhibited the greatest LD decay distance (~984.61 kb; r ² = 0.4008) and the lowest nucleotide diversity (π = 1.39 × 10⁻³), suggesting that accessions in this group represent elite cultivars, consistent with the recent selection of milder, less ungrateful and solid‐stalked celery varieties over the past few hundred years (Malhotra, 2006; Quiros, 1993).

We then calculated the pairwise population divergence (F _ST) across the five groups. The lowest F _ST was observed between G1 and G2 (F _ST = 0.074), likely due to their close geographic distribution. The F _ST was also relatively low between G1 and G4 (F _ST = 0.107), which aligns with the historical records suggesting that Chinese celery was introduced from G1 during the Han Dynasty. In contrast, both G3 and G5 displayed strong genetic divergence from other groups (Figure 2e), consistent with their status as improved elite cultivars. As expected, the highest F _ST was observed between G3 and G5 (F _ST = 0.387), which contained cultivars with entirely different breeding directions: selections for succulent petioles in common celery and enlarged hypocotyls in celeriac.

A phylogenetic tree of celeriac, common celery and five other species suggested that celeriac diverged from common celery around 1.69 million years ago (Mya) (Figure S11). To further understand the evolutionary history of celery, we estimated the divergence times for each group. The results indicated that G1 diverged first at 1.72 Mya, while G2 and G3 diverged from their common ancestor at 1.55 Mya and G4 and G5 diverged from their common ancestor at 1.33 Mya (Figure S12). Subsequent pairwise sequentially Markovian coalescent (PSMC) analysis revealed that the five groups shared a similar demographic history: they diverged around 1.0 Mya, followed by two bottleneck events, the first occurring soon after their divergence and the second around 200 000 years ago (Figure S13).

Genetic basis of hypocotyl expansion in celeriac

Celeriac is characterised by its enlarged edible hypocotyl. However, little is known about the genetic basis of this important trait. We identified selection signals during the history of celeriac breeding through performing two comparisons: between G2 and G1 for the “formation” stage and between G3 and G2 for the “improvement” stage. A total of 1833 putative selective sweeps were identified in the comparison of G2 vs. G1, spanning 99.93 Mb and encompassing 1977 genes (Figure S14 and Tables S14 and S15), whereas 2719 sweeps were detected in G3 vs. G2, covering 164.75 Mb and involving 3443 genes (Figure 3a and Tables S16 and S17).

Figure 3.

Identification of candidate genes for hypocotyl expansion in celeriac. (a) Genome‐wide distribution of selective sweeps identified through comparison between G2 and G3 using XP‐CLR (cross‐population composite likelihood‐ratio) test (sliding window = 50 kb, step = 25 kb). Solid and dashed black lines define the top 1% and 5% XP‐CLR scores, respectively. Red vertical bars indicate two selective sweep regions that overlap with GWAS signals. (b) Manhattan plot of GWAS for the hypocotyl expansion trait. Solid grey and dashed black horizontal lines indicate the Bonferroni‐corrected significance threshold of GWAS at α = 0.01 and α = 0.05, respectively. Candidate genes identified by both GWAS and selective sweep analysis during celeriac improvement are displayed. (c) and (f) Local Manhattan plots showing the 15.5–17.5 Mb (c) and 14.0–16.0 Mb (f) regions on chromosomes 2 and 4, respectively. Red dots correspond to the most significant SNPs on chromosomes 2 and 4, respectively, and red lines indicate the positions of the two SNPs in the promoter regions of Agrc02g006400 and Agrc04g001310, respectively. Promoter regions are represented by grey boxes, and CDS regions are indicated by green boxes. (d) and (g) Allele frequencies of the two SNP loci in different celery groups. (e) and (h) Expression of Agrc02g006400 and Agrc04g001310 in hypocotyl tissues of common celery (C003) and celeriac (C190) cultivars. Values are means ± SD (n = 3).

Only 274 genes were selected during both the formation and improvement stages (Figure S15 and Table S18), indicating that only a small fraction of genes are under continuous selection during celeriac breeding. These genes were enriched with those associated with xylem and phloem development. Several potential genes associated with the growth and expansion of plant organs were identified, including Agrc07g000890 (encoding a cyclin‐dependent kinase inhibitor), Agrc07g032470 (encoding a cyclin‐dependent kinase) and Agrc10g029800 (encoding a cyclin), whose homologues have been reported as cell cycle regulators influencing cell division, thereby impacting plant development and organogenesis (Inagaki and Umeda, 2011; Inzé and De Veylder, 2006; Meijer and Murray, 2001). Additionally, Agrc09g028600 encodes a CONSTANS‐LIKE (COL) protein homologous to the Arabidopsis B‐box protein 11 (BBX11), which has been reported to negatively regulate hypocotyl elongation under various light conditions (Zhao et al., 2020). The continued selection of these genes may suggest their key roles in celeriac breeding.

We then conducted GWAS analysis for the hypocotyl expansion trait using the 177 accessions, identifying a total of 16 candidate genes associated with hypocotyl expansion. Among them, six were also detected in the selective sweeps during celeriac improvement (Figure 3b and Tables S19 and S20), and five of these six genes were expressed in hypocotyl tissues (Figure S16). The most significant SNP on chromosome 2 (Chr02:16331290) located in the promoter region of Agrc02g006400 (Figure 3c). Notably, 92% of G3 accessions shared the TT genotype at this SNP locus, while the remaining accessions in this group exhibited the TC heterozygous genotype. In contrast, only 8% of the G2 accessions harboured the TT genotype, 25% displayed the TC heterozygous genotype and the remaining (67%) had the CC genotype. Remarkably, all accessions from other groups lacking hypocotyl expansion exclusively harboured the CC genotype (Figure 3d). Agrc02g006400 encodes a bHLH transcription factor (TF). The bHLH family TFs are universally involved in plant development, metabolism and stress response (Hao et al., 2021; Sun et al., 2018). Transcriptome analysis revealed a lower expression of Agrc02g006400 in the expanded hypocotyl, which was further confirmed by quantitative reverse transcription PCR (RT‐qPCR) (Table S20 and Figure 3e).

The next significant SNP on chromosome 4 (Chr04:14246328) located in the promoter region of Agrc04g001310 (Figure 3f). The TT genotype of this SNP locus was dominant in G3 (92%), less prevalent in G2 (28.6%), but completely absent in all other groups. Conversely, the AA genotype predominated in groups with nonexpanded hypocotyls (87.5% in G1, 100% in G4 and 98.1% in G5) (Figure 3g). An 84‐bp deletion in the promoter region of this gene, which contains a Myb‐binding site, exhibited a similar distribution pattern to that of the SNP (Figure S17). Agrc04g001310 encodes a jasmonic acid ZIM‐domain protein and its homologue in potato could regulate tuber initiation and formation in stolon tips (Begum et al., 2022). The expression of Agrc04g001310 was significantly repressed in expanded hypocotyl tissues of celeriac comparing to common celery (Table S20 and Figure 3h), indicating its potential function in celeriac hypocotyl expansion.

Genetic basis of petiole hollowness

The distinguishing feature of common celery is its thick, solid and succulent leaf petioles, whereas the petioles of celeriac and smallage are often slender, shorter and hollow. The hollow/solid trait largely determines the texture of celery petioles. However, the genetic basis of this trait remains largely unknown. We observed that the petioles of hollow varieties initially exhibited a solid structure and gradually became hollow during growth (Figures 4a and S18). Conversely, solid celery varieties maintained their solid petiole structure throughout the entire growth period. We selected the celeriac cultivar Hongcheng (C190) with hollow petioles and the common celery cultivar Ventura (C003) with solid petioles for comparative transcriptome analysis. Petioles were collected at 55 days after planting (DAP), a stage at which the petioles of the hollow line from inner to outside corresponded to the solid (P1), cavity emergence (P2) and cavity expansion (P3) periods (Figure 4a). Differential expression analyses of the three stages between the two lines revealed a consistent upregulation of 809 genes and a consistent downregulation of 816 genes in the hollow line (Figure 4b; Tables S21 and S22). The consistently upregulated genes were significantly enriched with multiple pathways associated with programmed cell death triggered by reactive oxygen species and cell wall catabolism (Figure S19), suggesting that programmed cell death may play a crucial role in the formation of hollow petioles in celery, consistent with the lysogenic mechanism of cavity formation (Evans, 2003).

Figure 4.

Identification of candidate genes for petiole hollowness in celery. (a) Petioles at different developmental periods. Scale bars: 1 mm. (b) Numbers of up‐ and downregulated genes during petiole growth in the hollow line compared with the solid line. (c) GWAS for hollowness based on SNPs (top) and SVs (bottom). Solid grey and dashed black horizontal lines indicate the Bonferroni‐corrected significance threshold of GWAS at α = 0.01 and α = 0.05, respectively. (d) Local alignment of chromosome 4 between Ventura and Alabaster. Syntenic regions are indicated by grey colour, and the insertion in Alabaster is indicated by red line. The promoter region is represented by the grey box, and the CDS regions are indicated by the green boxes. (e) Allele frequency of the 4144‐bp insertion encompassing the Agrc04g007090 gene in hollow and solid petiole accessions. (f) Cross‐section images of the inflorescence stem stained with Evans blue from 75‐day‐old Arabidopsis plants. Scale bars: 200 μm.

GWAS of the petiole hollow/solid trait identified a particularly strong association signal on chromosome 4 (Figure 4c and Table S23) that harboured a total of 55 genes, of which 35 were expressed (FPKM ≥ 1) in at least one period of the hollow or solid line (Table S24 and Figure S20). Among these genes, Agrc04g007090, which encodes a NAC TF, was upregulated across all three periods of petiole development in the hollow line compared with the solid line (Figures S20 and S21). We further performed GWAS of the hollow‐stalk trait using SVs. A 4144‐bp indel (IND_W_40871), which spanned the entire Agrc04g007090 gene, was identified as the most significantly associated SV (Figure 4c and Tables S25 and S26). This gene exhibited a presence/absence variation between the Alabaster and Ventura genomes, i.e., present in Alabaster and absent in Ventura, due to the indel (Figure 4d). Among the 177 sequenced lines, this 4144‐bp insertion was present in all hollow‐stalked accessions and absent in all solid lines (Figure 4e). To further validate the correlation of this insertion and the hollow‐stalk trait, we genotyped the insertion based on resequencing data of 51 celery or leafy celery accessions (depth >5×) from a recent study (Cheng et al., 2022). We found that all the 18 hollow‐stalked accessions contained this insertion, while all the 33 solid‐stalked ones didn't (Table S27 and Figure S22).

Phylogenetic analysis of Agrc04g007090, Arabidopsis NACs and a sorghum NAC TF D revealed that Agrc04g007090, ANAC074 and D belonged to the same large clade (Figure S23). Both ANAC074 and D have been reported to induce programmed cell death in stem parenchyma cells. To better understand the functional roles of Agrc04g007090, we conducted a complementation assay by heterologously expressing the Agrc04g007090 gene in Arabidopsis anac074 mutant plants (Figure S24). The inflorescence stems of wild‐type, anac074 mutant and Agrc04g007090‐overexpressing plants were stained with Evans blue at 75 days after planting, revealing that most pith parenchyma cells in the wild type were stained, indicating cell death (Figures 4f and S25), whereas the pith parenchyma cells in the anac074 mutant were not stained, consistent with previous findings (Fujimoto et al., 2018). Notably, the Agrc04g007090‐overexpressing plants restored cell death in the pith parenchyma of anac074 inflorescence stems (Figures 4f and S25). This demonstrates that the Agrc04g007090 gene is involved in cell death in the stem pith parenchyma. Therefore, we propose that Agrc04g007090 likely controls the formation of hollow petioles in celery by regulating the programmed death of petiole pith cells.

Discussion

Celery has undergone extensive selective breeding over the past few centuries, resulting in significant phenotypic differences between varieties. In this study, we present, for the first time, a high‐quality chromosome‐scale genome assembly of celeriac. Additionally, genome resequencing of a global collection comprising 177 celery accessions has elucidated the relationships and breeding history of the three celery varieties. By identifying candidate genes associated with hypocotyl expansion in celeriac and the solid petiole trait in common celery, our study enhances the understanding of the genetic basis underlying these agronomically important traits and contributes to the development of biological research and breeding strategies for this important vegetable crop.

Celeriac and common celery represent two distinct directions of selection, leading to completely different plant phenotypes, with celeriac featuring enlarged hypocotyls, hollow petioles and a strong flavour. Through direct genome comparison, we identified abundant SVs between celeriac and common celery. GO enrichment analysis of genes with SVs in CDS or promoters revealed that many of these genes were involved in defence response and response to stimuli (Figure S4). Previous studies have indicated that celeriac and common celery exhibit both similar and different disease resistances to same pathogens or pests (Bruznican et al., 2020). Moreover, the heat tolerance of celery is positively correlated with petiole hollowness, with solid celery showing greater heat tolerance (Li et al., 2023). Thus, SVs between celeriac and common celery may contribute to their different disease resistance and environmental adaptations. Furthermore, we found that unique genes identified in the Alabaster genome were associated with glutathione catabolic processes, programmed cell death and responses to reactive oxygen species (Figure S5). Among them, Agrc04g007090 was highlighted by SV‐GWAS as the key candidate gene responsible for the formation of hollow and solid petioles in celery. Therefore, these unique genes may play critical roles in the phenotypic differences between varieties. Future in‐depth studies on these genes will enhance our understanding of the underlying causes of intervarietal differences and contribute to the selection and improvement of desirable traits.

Celery is believed to have originated in the Mediterranean region, with early forms characterised by slender and hollow petioles. The basal position of Mediterranean smallage on the phylogenetic tree supports this assumption. Following the initial domestication, diverse celery varieties have emerged due to intensive selection efforts aiming at enhancing different edible parts. Celeriac and common celery exhibit significant morphological differences from Mediterranean smallage, with each having a clear direction of selection, i.e., selection for enlarged hypocotyls to form the celeriac type and for solid succulent petioles to form the celery type. Chinese leafy celery closely resembles Mediterranean smallage in its botanical form, featuring slender and leafy petioles, while they displayed distinctive genetic structure patterns. China was the first place to cultivate celery as a vegetable, where it was primarily used as a side dish and a spicy flavouring. Thus, while the original plant form of celery has been preserved, improvement has altered its strong flavour, making it more palatable. Moreover, China has domesticated the local leafy celery in various ways to enrich its colour and texture. In the 1970s, large quantities of common celery from the West were introduced into China and crossed with local leafy celery to produce solid‐petiole varieties. All of these have allowed Chinese celery to maintain a high level of genetic diversity, distinct from celeriac and common celery.

The domestication of celeriac is believed to have occurred in two distinct phases, with the first phase primarily taking place in the Mediterranean region. During this initial phase, semi‐domesticated celeriac with a well‐defined hypocotyl between the petiole and the root system emerged; however, the expansion of hypocotyl in this phase was limited. In the second phase, celeriac spread to Europe where it underwent intensive selection processes, leading to the development of modern improved celeriac cultivars with significantly expanded hypocotyls. These varieties are now widely cultivated as elite celeriac. The emergence of common celery occurred relatively recently, first in Italy during the 16th century, where solid‐stalked varieties were bred and later spread to France and England (Malhotra, 2006). During the 17th–19th centuries, European countries such as France made significant improvements to solid‐stalked celery by enhancing its petiole plumpness and crispness while reducing its medicinal flavour. This resulted in a crunchy, sweet and juicy solid‐stalked celery, aligning with European and American dietary preferences for consuming raw celery as salad. Consequently, the development of solid petiole became the primary objective of common celery selection, leading to the rapid dissemination of common celeries with solid petioles throughout Europe and the United States. As a result of this historical trajectory, cultivated common celery exhibits limited diversity.

Hypocotyl expansion is the key distinguishing characteristic of celeriac from other varieties. Through selective sweep and GWAS analyses, we have pinpointed six genes within the selective sweep regions, each hosting significantly associated SNPs in its gene body or promoter region. Notably, two genes, Agrc02g006400 and Agrc04g001310, exhibit significant expression difference between expanded and non‐expanded hypocotyls, suggesting that regulation of their expression might directly influence hypocotyl expansion in celeriac.

The structure of the petiole plays a crucial role in determining the texture and taste of celery. The petioles of common celery are solid and succulent, whereas those of celeriac and smallage are often slender and hollow. Our investigation revealed that cavities within hollow petioles developed gradually as development progressed. Previous studies have elucidated two mechanisms for the formation of cavities (known as aerenchyma) in plant tissues: schizogeny, involving cell separation and lysigeny, resulting from programmed cell death (Evans, 2003). Lysigenous aerenchyma formation has been observed in many important crops, such as maize, rice and wheat (Gunawardena et al., 2001; Nilsen et al., 2020; Steffens et al., 2011). Key signalling molecules such as reactive oxygen species, ethylene and calcium are recognised for initiating PCD during lysigenous aerenchyma formation in plants (De Pinto et al., 2012; Ren et al., 2021; Trobacher, 2009). As the upregulated genes along hollowness development in hollow‐stalk celeries are enriched in pathways associated with programmed cell death, we infer that the formation of hollow petioles in celery aligns with the lysogenic type of cavity formation. Through GWAS analysis, we have delineated the genetic basis underlying the hollow/solid trait. Our findings indicate that this trait directly correlates with a structural variant that impacts the presence or absence of the NAC gene, Agrc04g007090. Notably, this gene is present in all hollow accessions while absent in all solid accessions. Additionally, the same variant was detected in Chinese solid leafy celery accessions, implying that they may have derived from the introgression from common celery. Functional characterisation of Agrc04g007090 in Arabidopsis demonstrated it participated in PCD, suggesting its role in hollow petioles in celery.

Preferences for specific traits in cultivated celeriac and common celery have led to intensive artificial selection, resulting in a significant loss of diversity. In contrast, smallage from the Mediterranean region still retains a high diversity and can serve as a valuable genetic resource for future celery breeding and improvement endeavours. Hence, conserving smallage from the region of celery origin becomes imperative. Chinese local celery also exhibits high diversity, with a wide range of colours and textures. However, with the introduction and widespread cultivation of common celery in China, coupled with the market's preference for solid‐petiole traits and the development of more hybrid solid‐stalked varieties, traditional Chinese leafy celery is facing a decline. Consequently, the conservation of traditional Chinese local celery is becoming increasingly urgent. In summary, our study not only sheds light on the origin and breeding history of celery but also provides valuable resources for the genetic improvement of this important crop.

Methods

Plant materials, classification and phenotyping

A total of 177 celery accessions were obtained from the U.S. National Plant Germplasm System or purchased from local markets in China and the United States. The accessions were grown in the greenhouse at Wuhan Botanical Garden, Chinese Academy of Sciences (Wuhan, Hubei Province), at 16–20°C with 10 h of light and 14 h of darkness. The celeriac cultivar Alabaster (PI 662464) was selected for de novo assembly, and the other 176 accessions were subjected to genome resequencing. These accessions were categorised into three botanical varieties based on their morphological characteristics observed in two consecutive years of 2020 and 2021, including hypocotyl enlargement, petiole hollowness, petiole texture, flavour and plant architecture. Hypocotyl enlargement was assessed by investigating the hypocotyl phenotype of each accession at 90–120 DAP. For petiole hollowness, the outermost petiole of each accession was cut transversely at 40 DAP to determine the presence or absence of hollowness.

Library construction and sequencing

For PacBio sequencing, high‐molecular‐weight DNA was isolated from fresh young leaves of Alabaster. A SMRT library was constructed following the standard SMRTbell library preparation protocol and sequenced on a PacBio Sequel II platform in the CCS (circular consensus sequencing) mode to produce HiFi reads. Illumina paired‐end libraries with insert sizes of ~350 bp were constructed using the Illumina Genomic DNA Sample Preparation kit following the manufacturer's instructions. Hi‐C libraries were prepared following the proximo Hi‐C plant protocol. Five different tissues (leaves, roots, petioles, hypocotyls and flowers) of Alabaster were collected for RNA‐seq. RNA‐seq libraries were constructed for each tissue using the NEBNext Ultra RNA Library Prep Kit for Illumina according to the manufacturer's recommendations. The Illumina paired‐end, Hi‐C and RNA‐seq libraries were sequenced on the Illumina X Ten platform.

For genome resequencing, a paired‐end sequencing library with the insert size of ~350 bp was constructed for each of the 176 celery accessions from genomic DNA extracted from fresh leaves and then sequenced on the DNBSEQ‐T7 platform of BGI.

De novo genome assembly

The genome size of Alabaster was estimated using Jellyfish v2.2.10 (Marcais and Kingsford, 2011) and GCE v1.0.2 (Liu et al., 2013) with Illumina sequencing reads. HiFi reads were de novo assembled into contigs using hifiasm v0.13‐r308 (Cheng et al., 2021) with default parameters. Raw Hi‐C data were filtered using HiCUP v0.8.0 (Wingett et al., 2015) to obtain valid read pairs, which were aligned to the assembled contigs using Juicer (Durand et al., 2016). The resulting alignments were used to construct pseudochromosomes using 3D‐DNA v180114 (Dudchenko et al., 2017) and further manually corrected and sorted using Juicebox v1.11.08 (Robinson et al., 2018).

The quality of the genome assembly was assessed using the LTR assembly index (LAI) (Ou et al., 2018), BUSCO v5.1.2 with the ‘embryophyta_odb10’ database (Manni et al., 2021) and Merqury (Rhie et al., 2020). The accuracy of the genome assembly was further assessed by aligning the Illumina reads to the assembly using BWA v0.7.17‐r1188 (Li and Durbin, 2009), and SAMtools v1.7 (Danecek et al., 2021) was then used to calculate the mapping rate and the genome coverage ratio.

Genome annotation

For repeat annotation, a de novo repeat library was constructed using RepeatModeler v2.0.1 (Flynn et al., 2020). LTR_finder v1.07 (Xu and Wang, 2007) and LTRharvest (Ellinghaus et al., 2008) were used to identify LTRs in the Alabaster genome, and LTR_retriver v2.9.0 (Ou and Jiang, 2018) was used to integrate and filter the identified LTRs to construct a nonredundant LTR library. The final repeat library was generated by combining the LTR library with the de novo repeat library. The combined library was then utilised to scan the Alabaster genome for repeat sequences using RepeatMasker v4.1.0 (Tarailo‐Graovac and Chen, 2009).

Protein‐coding genes in the Alabaster genome were identified using a combined method that included ab initio prediction, comparisons to homology proteins, and transcript evidence derived from RNA sequencing. Ab initio prediction was performed using AUGUSTUS v3.3.3 (Stanke et al., 2008). For transcript evidence, RNA‐seq reads from five tissues of Alabaster were aligned to the assembled genome using HISAT2 v2.2.1 (Kim et al., 2019), and transcripts were subsequently assembled from the read alignments using Stringtie v2.1.6 (Kovaka et al., 2019). TransDecoder v5.5.0 (http://transdecoder.github.io) was then used to predict coding regions from the assembled transcripts. Protein sequences from Arabidopsis thaliana, Oryza sativa, Daucus carota, Coriandrum Sativum and Apium graveolens Ventura were collected as protein homology evidence. Finally, the MAKER v2.31.10 pipeline (Cantarel et al., 2008) was used to predict gene models by integrating evidence from ab initio prediction, transcript assembly and protein homology. To improve the confidence of the predictions, predicted protein‐coding genes with AED values > 0.3 and FPKM < 1 were filtered out.

The predicted protein‐coding genes were functionally annotated by comparing their protein sequences against public databases including GenBank nr (https://www.ncbi.nlm.nih.gov/), KEGG (https://www.kegg.jp/), Swiss‐Prot and TrEMBL (https://www.uniprot.org/), using DIAMOND v2.0.13.151 (Buchfink et al., 2021). Protein domains were identified in the protein‐coding genes by searching against the InterPro database (http://www.ebi.ac.uk/interpro/) using InterProScan v5 (Jones et al., 2014). Gene Ontology (GO) terms were then assigned to each gene according to the corresponding InterPro entries.

Genome synteny and structure variation

Genome sequences of Ventura and Baili were aligned to the Alabaster genome using minimap2 v2.24 (Li, 2018) with the parameter ‘‐ax asm5’. SyRI v1.6.3 (Goel et al., 2019) was then used to generate the syntenic plot and identify inversions. The genome alignments between Alabaster and Ventura were filtered using Assemblytics v1.1 (Nattestad and Schatz, 2016). SV detection and genotyping followed a previously published pipeline (https://github.com/GaoLei‐bio/SV) (Wang et al., 2020). The identified SVs were annotated using Annovar v20180416 (Wang et al., 2010) based on the annotations of both Alabaster and Ventura genomes.

Genome read mapping and variant calling

Raw resequencing reads were processed using fastp v0.20.1 (Chen et al., 2018). The cleaned reads were aligned to the Alabaster genome using BWA‐MEM (Li and Durbin, 2009), and the duplicate alignments were marked using Picard v2.26.11 (http://broadinstitute.github.io/picard/). Genome variants were identified using the Haplotyper module of the Sentieon^® Genomics software (Weber et al., 2016). Raw SNPs and indels were filtered using GATK v4.2.0 (McKenna et al., 2010), with parameters ‘QD < 2.0 || FS > 60.0 || MQ < 40.0 || MQRankSum < −12.5 || ReadPosRankSum < −8.0’ for SNPs and ‘QD < 2.0 || FS > 200.0 || ReadPosRankSum < −20.0’ for indels. Finally, SNPs and indels were annotated using Annovar v20180416 (Wang et al., 2010).

Phylogenetic and population structure analyses

SNPs with a minor allele frequency greater than 0.05 and a missing rate less than 10% were used for population genomic analyses. SNPs at fourfold degenerated sites (72 677 SNPs) were used to construct a maximum‐likelihood phylogeny tree employing IQ‐TREE v2.2.5 (Minh et al., 2020) with 1000 bootstraps. Petroselinum crispum was used as the outgroup. Illumina reads of Petroselinum crispum were downloaded from the NCBI Sequence Read Archive (accession no. ERR5639094) and aligned to the Alabaster genome for SNP calling, following the method described above. The resulting phylogenetic tree was visualised using the online tool iTOL (https://itol.embl.de/). Principal component analysis (PCA) was performed using Plink v1.90 (Purcell et al., 2007). Population structure analysis was performed using STRUCTURE v2.3.4 (Hubisz et al., 2009), with the number of subpopulations (K) set from 1 to 9, and each K was run 10 times. The optimal K (K = 2) was inferred using the online tool structureHarvester (https://taylor0.biology.ucla.edu/structureHarvester/).

Nucleotide diversity (π) and inter‐population fixation index (F _ST) were computed using VCFtools v0.1.17 (Danecek et al., 2011) with a window size of 500 kb and a step size of 50 kb. LD analysis was performed using PopLDdecay v3.41 (Zhang et al., 2018), where the squared correlation coefficient (r ²) between paired SNPs was calculated in each subpopulation with the parameters ‘‐‐ MaxDist 1000 ‐‐ MAF 0.05’. LD decay was measured by the r ² value of all paired SNPs within 1000 kb and the physical distance at which the pairwise correlation coefficient dropped to half of its maximum value was used as the LD decay distance.

Divergence time estimation and population demographic analysis

The species phylogenetic tree was constructed based on 438 single‐copy genes using RAxML v8.2.12 (Stamatakis, 2014) with the maximum likelihood method. Divergence times across different species were estimated with Mcmctree v4.10.7 (Yang, 2007). The time correction points of grape and lettuce (111.4–123.9 Mya), lettuce and ginseng (76.0–97.4 Mya) and ginseng and carrot (54.3–69.0 Mya) obtained from TimeTree website (http://www.timetree.org) were used to calibrate the time estimates. The divergence times of the five A. graveolens groups were inferred using r8s v1.81 (Sanderson, 2003), based on the divergence time between A. graveolens var. rapaceum and A. graveolens var. dulce. Demographic histories for the five A. graveolens groups were reconstructed using the pairwise sequentially Markovian coalescent model (Li and Durbin, 2011) with parameters “‐N25 ‐t15 ‐r5 ‐p ‘4 + 25*2 + 4 + 6’”. A mutation rate of 7.35 × 10⁻⁹ per site per year (Song et al., 2020) and a generation time of 2 years were used for A. graveolens.

Selective sweep identification

Selective sweeps were detected using the XP‐CLR package (Chen et al., 2010) The XP‐CLR scores were computed across the Alabaster genome with sliding windows of 50 kb and a step size of 25 kb. Genome regions with the top 5% highest XP‐CLR scores were identified as potential selective sweeps.

Genome‐wide association study

GWAS was conducted separately with SNPs and SVs, using a total of 28 169 082 SNPs and 54 397 SVs with MAF ≥ 0.05 and missing rate ≤ 0.1. GWAS was run with the mixed‐linear model using the EMMAX package (Kang et al., 2010). A kinship (K) matrix, generated with the emmax‐intel64 program in the EMMAX package, and a population structure (Q) matrix corresponding to the top three PCs were combined as a covariate to correct population stratification. Quantile–quantile and Manhattan plots were generated using the R package CMplot (Yin et al., 2021). Significance thresholds for GWAS were determined based on Bonferroni‐corrected P values of 0.05 and 0.01, corresponding to raw P values of 1.77 × 10⁻⁹ and 3.55 × 10⁻¹⁰, or −log10(P) values of 8.75 and 9.45, respectively, for SNPs, and 9.19 × 10⁻⁷ and 1.84 × 10⁻⁷, or −log10(P) values of 6.04 and 6.74, respectively, for SVs.

RNA sequencing and data analysis

Petioles and hypocotyls of two different cultivars, celeriac line Hongcheng and celery line Ventura, were collected at 55 DAP and 105 DAP, respectively. Each sample consisted of three biological replicates. Total RNA was isolated from these tissues using TIANGEN RNAsimple Total RNA Kit for subsequent RNA sequencing and RT‐qPCR analysis. RNA‐seq libraries were constructed and sequenced on the Illumina NovaSeq6000 platform. RNA‐seq reads were mapped to the Alabaster genome using HISAT2 v2.1.0 (Kim et al., 2019). Raw read counts for each gene were generated using the featureCounts function in the Rsubread package v2.12.3 (Liao et al., 2019) and normalised to FPKM values using edgeR v3.40.2 (Robinson et al., 2010). Differential expression analysis, based on raw read counts, was performed using the DESeq2 v1.34.0 (Love et al., 2014). Genes with at least twofold change and an adjusted P < 0.05 were considered as differentially expressed.

Arabidopsis transformation and microscopic observations

The Columbia‐0 (Col‐0) ecotype was used as the wild‐type Arabidopsis in this study. The mutant line anac074 was purchased from Arashare (http://www.arashare.cn). All plants were grown in the grown room at 22°C with a 16‐h light/8‐h dark cycle. To construct the binary vector for plant transformation, the full‐length coding sequence (CDS) of Agrc04g007090 was inserted into the pCambia1300‐221 vector to generate 35S::Agrc04g007090. Agrobacterium tumefaciens GV3101 harbouring the corresponding plasmid was then transformed into Arabidopsis using the floral dipping transformation method to obtain stable transformed plants. Transgenic T1 plants were selected on plates containing solid modified MS medium supplemented with 25 mg/L hygromycin B. Quantitative PCR was performed using Hieff® qPCR SYBR Green Master Mix (Yeasen Biotechnology), with the Arabidopsis Actin2 gene as the internal reference. The relative expression levels of the target genes were calculated using the ΔΔCT method. Cross sections of inflorescence stems from 75‐day‐old wild‐type, anac074 mutant and Agrc04g007090‐overexpressing Arabidopsis plants were stained with 0.1% (w/v) Evans blue and observed using a Leica DMi8 inverted fluorescence microscope.

Conflicts of interest

The authors declare no conflicts of interest.

Author contributions

L.G., Z.F. and S.G. conceived the project. E.L., S.G., L.G., P.W., C.H., X.Y., M.Q., S.L., M.Z., X.L. and G.F. performed experiments. E.L., X.Y., X.W., M.Q., Y.Z., H.P., Y.Y., H.L. and L.G. analysed the data. E.L. wrote the draft manuscript. L.G., Z.F., X.W. and Y.Z. revised the manuscript.

Supporting information

Figure S1. 17‐mer depth distribution of the Alabaster Illumina genomic reads. Genome size of Alabaster was estimated based on the formula: Total number of 17‐mers/position of peak depth = 208,125,874,722/61 = 3,411,899,585 bp.

Figure S2 Synteny between the three celery genomes. (a) Synteny between celeriac Alabaster and common celery Ventura genomes. (b) Synteny between celeriac Alabaster and common celery Baili genomes. (c) Synteny between Ventura and Baili genomes.

Figure S3 Examples of inversions between Alabaster and Ventura supported by Hi‐C contact maps and PacBio HiFi read alignments. (a) Inversions on chromosome 2. (b) Inversions on chromosome 7.

Figure S4 GO terms enriched in genes with SVs in their coding sequences or promoter regions.

Figure S5 GO terms enriched in genes only present in Alabaster (a) or Ventura (b).

Figure S6 Plant morphology of the three celery varieties. (a) celeriac (C052, Alabaster). (b) Common celery (C003, Ventura). (c) Smallage (C143, Tiganching).

Figure S7 Maximum‐likelihood phylogenetic tree and geographic origins of A. graveolens accessions.

Figure S8 Genome‐wide distribution of the fixation index (F _ST) between Clade II and Clade III. Dashed black horizontal lines indicate the top 5% of the highest F _ST.

Figure S9 GO terms enriched in genes located in highly differentiated genomic regions between Clade II and Clade III.

Figure S10 Examples of the expanded hypocotyls of celeriac accessions from G2 and G3 groups. (a) C084 from the G2 group at 100 DAP. (b) C052 from the G3 group at 100 DAP.

Figure S11 Species phylogeny with estimated divergence times. Red dots indicate calibration points obtained from the TimeTree website.

Figure S12 Phylogeny and estimation of divergence times across the five A. graveolens subpopulations, with Petroselinum crispum as the outgroup.

Figure S13 Estimates of effective population sizes (Ne) of G1 (Mediterranean smallage group), G2 (semi‐domesticated celeriac), G3 (celeriac), G4 (Chinese leafy celery), and G5 (common celery). g, generation time (years); μ, mutation rate.

Figure S14 Genome‐wide distribution of selective sweeps identified through comparisons between G1 and G2. Solid and dashed black lines define the top 1% and 5% XP‐CLR scores, respectively.

Figure S15 Venn diagrams showing numbers of genes within G2/G1 and G3/G2 selective regions.

Figure S16 Expression patterns of six candidate genes associated with hypocotyl expansion in the hypocotyl tissues of common celery (C003) and celeriac (C190).

Figure S17 Gene structure and SV allele frequency distribution of Agrc04g001310.

Figure S18 Morphology of the hollow line ‘Hongcheng’ (C190) and the solid line ‘Ventura’ (C003).

Figure S19 Gene Ontology (GO) enrichment analysis of consistently up‐ (a) and down‐regulated (b) genes in the hollow line across the three developmental stages.

Figure S20 Expression patterns of 35 candidate genes associated with the hollow/solid petiole trait (FPKM ≥1) in petioles of the hollow line ‘Hongcheng’ (C190) and the solid line ‘Ventura’ (C003).

Figure S21 Expression of Agrc04g007090 in petioles of the hollow line ‘Hongcheng’ (C190) and the solid line ‘Ventura’ (C003). Values are means ± SD (n = 3).

Figure S22 Allele frequency of the 4144‐bp insertion encompassing the Agrc04g007090 gene in hollow and solid petiole populations.

Figure S23 Phylogenetic analysis of Agrc04g007090, Sobic.006G147400.1.p (sorghum D protein), and NAC transcription factors from Arabidopsis.

Figure S24 Expression of Agrc04g007090 in Arabidopsis wild type, anac074 mutant, and anac074 mutants overexpressing Agrc04g007090 (OE1‐9). Values are means ± SD (n = 3).

Figure S25 Cross‐section images of Evans blue‐stained inflorescence stems of 75‐day‐old Arabidopsis wild type, anac074 mutant, and anac074 mutant overexpressing Agrc04g007090. Scale bars: 200 μm.

Table S1 Summary statistics of genome sequencing data for Alabaster.

Table S2 Summary statistics of the celery genome assemblies.

Table S3 Summary statistics of Illumina reads mapped to different celery genomes.

Table S4 Summary statistics of the repeat sequence classification in celery genomes.

Table S5 Statistics of protein‐coding genes predicted from the celeriac genome.

Table S6 Statistic of gene functional annotations of the celeriac genome.

Table S7 SVs between genomes of Ventura and Alabaster.

Table S8 Genes present in the Alabaster genome but absent in the Ventura genome.

Table S9 Genes present in the Ventura genome but absent in the Alabaster genome.

Table S10 Information of the 177 accessions used in this study. The order of the accessions is according to that of Fig. 2b.

Table S11 Sequencing data statistics of the 177 celery accessions.

Table S12 Annotation of SNPs and Indels.

Table S13 Genes contained in the top 5% most differentiated region between Clade II and Clade III.

Table S14 Selective sweeps during celeriac formation.

Table S15 Genes in selective sweeps during the formation of celeriac.

Table S16 Selective sweeps during celeriac improvement.

Table S17 Genes in selective sweeps during the improvement of celeriac.

Table S18 Genes in selective sweeps during both the formation and improvement of celeriac.

Table S19 List of SNPs significantly associated with the hypocotyl expansion trait.

Table S20 List of candidate genes for the hypocotyl expansion trait.

Table S21 List of up‐regulated genes in hollow line (C190) across the three developmental periods compared to the solid line (C003).

Table S22 List of down‐regulated genes in hollow line (C190) across the three developmental periods compared to the solid line (C003).

Table S23 List of SNPs significantly associated with the hollow/solid petiole trait.

Table S24 List of candidate genes identified by SNP‐GWAS for the hollow/solid petiole trait.

Table S25 List of SVs significantly associated with the hollow/solid prtiole trait.

Table S26 List of candidate genes identified by SV‐GWAS for the hollow/solid petiole trait.

Table S27 Sequence statistics for the 51 celery accessions from Cheng et al., iScience 2022, 25:104565.

Data availability statement

Raw sequencing data and genome assemblies have been deposited in Genome Sequence Archive (GSA; https://ngdc.cncb.ac.cn/gsa/) under the BioProject accession number PRJCA022437.