Molecular investigation of the progenitors, origin and domestication patterns of diploid Chinese old garden roses

Abstract

Background and Aims

Chinese old garden roses are major contributors to the genetic development of modern roses. The RoKSN gene is associated with continuous flowering in roses and is proposed to have originated from Chinese wild roses. However, the wild roses that are implicated in the breeding of Chinese old garden roses and the origin of the RoKSN locus remain unidentified. We collected 25 of the most renowned and classic diploid Chinese old garden roses along with all related wild roses from East Asia. These roses were analysed with the aim of identifying the wild species that contributed to the genetic composition of Chinese old garden roses. In addition, we aimed to infer the geographical origin of the RoKSN gene and to develop a schematic overview of hybrid domestication of Chinese old garden roses.

Methods

We compared the haplotypes of internal transcribed spacers (nrITS), six nuclear single-copy genes and three chloroplast genes between Chinese old garden roses and wild roses. Additionally, we assessed genetic organization using 21 expressed sequence tag–simple sequence repeats to identify potential donor species that contributed to the emergence of these cultivars. Primers were designed for RoKSN to allow comparison of the gene across the entire distribution range of Rosa sect. Chinenses.

Key Results

Our findings confirmed that the majority of rose cultivars are descendants of early hybridization events. Rosa chinensis var. spontanea, R. odorata var. gigantea and R. multiflora var. cathayensis were the primary donors for the 25 cultivar roses. Chinese old garden roses were categorized into four groups. Ten cultivars were hybrids between R. chinensis var. spontanea and R. multiflora var. cathayensis, thereby forming the ‘Old Blush’ group. Five cultivars were hybrids between ‘Old Blush’ and the R. kwangtungensis species complex, thereby forming the ‘Slater’s crimson’ group. Six cultivars were hybrids between ‘Old Blush’ and R. odorata var. gigantea, thereby forming the ‘Tea Rose’ group, and three cultivars were hybrids that evolved from more than three donors. Moreover, we observed relatively close genetic proximity among Chinese old garden roses with an identical RoKSN-copia gene that is responsible for continuous flowering, which indicates a single origin for this retrotransposon-containing allele. Additionally, we determined that the haplotypes of the RoKSN-copia gene predominantly occurred in the Sichuan Basin region. In contrast, R. chinensis cultivated in the Ya’an region showed no markers of hybridization and displayed a genetic composition that was close to that of the wild species R. chinensis var. spontanea. This cultivar may represent the earliest mutated individual that bears the RoKSN-copia gene and may have served as a bridge from wild species to continuous-flowering old rose cultivars.

Conclusions

The study provides crucial evidence that elucidates the origin of cultivated roses and lays the groundwork for further analysis of the breeding history of Chinese old garden roses using genomic data.

INTRODUCTION

Roses are cherished as beloved ornamental flowering plants for gardening, potted flowers and cut flowers, and are one of the most important commercial plant crops worldwide. More than 35 000 varieties of modern roses exist with a wide assortment of novel traits. Rosa L. is a genus in the family Rosaceae that exhibits rich phenotypic diversity and considerable value in scientific, ornamental, and medicinal fields (Esselink et al., 2003; Roberts et al., 2003; Zhang et al., 2024c). Chinese roses were cultivated widely during the Han Dynasty (141–87 BC). So-called modern roses appeared in 1759 after Chinese roses (Rosa chinensis) were introduced to western countries (Cairns, 2003; Wang, 2007).

Rosa cultivars are classified horticulturally using three primary systems, i.e. the British Rose Society, American Rose Society and International Rose Federation classification systems. Although these schemes differ significantly from one another, the systems share clear demarcation for classification of old garden roses. This boundary is set uniformly in the year 1867, which was marked by the introduction of ‘La France’, the world’s first hybrid tea rose. Roses that existed prior to 1867 encompass both species and cultivated varieties and collectively are termed Chinese old garden roses or European old roses (Cairns, 2003).

Chinese old garden roses have introduced genes for superior horticultural traits, including continuous flowering, double petals, tea scent and yellow petal colour, to modern rose cultivars (Wylie, 1954; Martin et al., 2001). Moreover, numerous Chinese old garden roses possess many desirable characteristics. For example, R. chinensis f. viridiflora lacks petals but instead possesses an enlarged number of green sepals, whereas R. chinensis ‘Mutabilis’ has colour-changing petals (Wang, 2015). In addition, Chinese old garden roses in many cases introduced improved frost resistance to modern varieties (Ji et al., 2013; Fan et al., 2021; Li et al., 2022).

Although it is widely accepted that Chinese old garden roses contributed invaluable genetic resources to the development of modern roses (Martin et al., 2001; Rusanov et al., 2005), it is unclear which wild roses were involved in breeding of old roses. Although more than 200 wild roses exist, only 8–15 of these species may have been used to breed modern roses (Cairns, 2007). Exploration of the ancestral origins of Chinese old garden roses only identified lineages at the sectional level, which makes it challenging to pinpoint specific wild parental species. Moreover, the existing literature does not categorize Chinese old garden roses into groups, but instead describes these roses individually. For example, old roses in Chinese folklore are divided primarily into tea-scented varieties based on fragrance and flower colour and include R. odorata var. gigantea ‘Yueyuefen’ (pink petals) and ‘Yueyuehong’ (red petals) which are related to R. chinensis var. spontanea (Wang, 2015). Rosa sects Synstylae, Chinenses and Gallicanae generally are clustered together with Chinese old garden roses based on nuclear ribosomal internal transcribed spacer (nrITS) and maturase K (matK) evidence (Qiu et al., 2013). Rose cultivars may have evolved from the ‘Old Blush’ group to the R. odorata and ancient hybrid China groups based on simple sequence repeat (SSR) analysis (Tan et al., 2017). In addition, Chinese old garden roses and R. chinensis var. spontanea have a close genetic relationship (Wu, 2019), although the genetic components that are involved may be particularly complex (Jian et al., 2018; Yang et al., 2018). Phylogenetic analysis of certain Chinese old garden roses using single-copy nuclear genes revealed that these cultivars clustered together within sect. Chinenses, which renders it difficult to identify the exact species involved in hybridization (Cheng et al., 2025). Analysis of modern roses and certain Chinese old garden roses using whole genome single nucleotide polymorphisms (SNPs) indicated that Chinese old garden roses either cluster with modern roses (Cheng et al., 2025) or are assigned only to section-level components (Zhang et al., 2024c), which complicates the determination of the species involved. Although whole genome SNPs provide more informative loci, several issues remain to be resolved. Artificial selection and repeated selfing/backcrossing in the breeding of Chinese old garden roses may have diluted the genetic contributions of early wild species, making them undetectable in standard SNP or SSR analyses. As a result, the origins and domestication history of Chinese old garden roses remain unclear.

The determination of genome sequences of R. chinensis ‘Old Blush’ (Raymond et al., 2018), R. chinensis ‘Slater’s crimson China’ (Zhang et al., 2024b) and R. odorata var. gigantea (Zhou et al., 2024) make it feasible to study the origin of ancient roses at the genomic level. Thus, the components of Chinese old garden roses are derived primarily from sects Chinenses and Synstylae (Zhou, 2016; Zhang et al., 2024c; Cheng et al., 2025). While phylogenetic relationships between the Chinenses and Synstylae sections are still being debated (e.g. chloroplast paraphyly in Zhang et al., 2022), recent nuclear data (6048 single-copy genes; Cheng et al., 2025) support their monophyly, consistent with morphological distinctions (solitary flowers/free styles vs. corymbose inflorescences/connate styles). Given this evidence, we maintain the taxonomic status of both sections in our study. Analysis of modern roses and certain ancient roses using whole genome SNPs provided additional informative loci, but numerous issues need to be addressed (Zhang et al., 2024c). First, the close genetic distance and low nucleotide diversity among Chinese old garden roses (Scariot et al., 2006; Tan et al., 2017) resulted in these cultivar roses clustering together (Zhang et al., 2024c; Cheng et al., 2025), which complicates the dissection of parental components and assignment to specific wild species. Moreover, selective breeding for morphological traits, coupled with repeated selfing or backcrossing with a single wild species, may have diluted the genetic contributions of the species involved in early hybridization, which may not be detectable in cluster analysis (Wu, 2019). In addition, population genetics data of polyploid and diploid species were analysed together in admixture (Zhang et al., 2024c; Cheng et al., 2025) which necessitates further scrutiny due to allele dosage uncertainty (Meirmans and van Tienderen, 2013; Dufresne et al., 2014; Meirmans et al., 2018). Furthermore, certain putative parental species in sect. Synstylae, including R. taiwanensis and R. paniculigera, have not been analysed due to limitations in sampling wild roses. We address these issues further in the Discussion section.

Continuous flowering is a significant characteristic of ancient roses and inheritance of this trait is controlled by a pair of recessive homozygous loci (Li et al., 2015; Saint-Oyant et al., 2018). Reduced expression of the RoKSN gene is crucial for this trait which is attributed to a 9-kb copia retrotransposon insertion. This insertion results in haploinsufficiency of the RoKSN gene which may be linked to epigenetic modifications, as well as retrotransposon insertions along the chromosome (Iwata et al., 2012; Bai et al., 2021). The continuous-blooming trait is within the mapping region of the RoKSN gene which may originate from wild Chinese species (Soufflet-Freslon et al., 2021; Tan et al., 2024). Further understanding the source of this gene is essential for studying the evolution of ancient roses.

Certain Chinese old garden roses exhibit polyploidy, e.g. R. chinensis ‘Simianjing’ (2n = 3x; x = 7) and R. chinensis ‘Ruanxianghong’ (2n = 4x) (Luo et al., 2009). Polyploidy commonly occurs as a result of the merger of genomes of diploid species (Doyle and Coate, 2019). Certain polyploid wild roses in North America may stem from local diploids; for example, R. carolina L. is a tetraploid species derived from the allopolyploidization of the diploid species R. blanda Ait. and R. palustris Marshall (Joly et al., 2006; Joly and Bruneau, 2007 ). Almost all species within R. sects Synstylae and Chinenses are diploids (Jian et al., 2014; Li et al., 2017), with a sporadic report of polyploidy (Roberts et al., 2009). The parents of ancient roses are derived mainly from these two sections (Zhang et al., 2024c; Cheng et al., 2025). We speculate that the earliest ancient rose varieties may be diploid and derived from the hybrid offspring of the species from Rosa sects Synstylae and Chinenses. Investigation of the genetic composition of diploid cultivars is key to elucidating the origin and cultivation of ancient roses.

In this study, we subjected the most renowned and classic Chinese old garden rose cultivars, presumed to be diploid, using chromosome counting and flow cytometry to determine polyploidy status. Subsequently, we analysed seven nuclear and three chloroplast barcodes, and 21 pairs of SSR molecular markers that are characterized by rich polymorphism, good repeatability and high amplification efficiency. These markers are distributed across the chromosomes of Rosa. Genetic relationship and population structure analyses were conducted on 257 Rosa samples to explore the wild progenitors of Chinese old garden roses. The analysis reveals important genetic background information for understanding the origin and domestication mechanisms of both Chinese old garden roses and modern roses. Furthermore, the study provides a strong theoretical foundation for the genetic exploration and continued breeding of important traits in roses.

MATERIALS AND METHODS

Taxon sampling and collection of putative parent species distribution data

Twenty-five diploid Chinese old garden roses (Fig. 1; Supplementary Data Table S1) were collected from the main rose cultivation areas in China and were planted in the Suzhou Huaguan Yuanchuang Horticulture Technology Company, Jiangsu, China. Related wild roses were collected mainly the from Chengdu Institute of Biology (CIB). Rose cuttings were propagated and planted within the CIB in the Maoxian Mountain Ecosystem Research Station, Chinese Academy of Sciences (CAS) and Shifang Experimental Base of CIB, CAS. For haplotype analysis, the wild rose species sequences used in BLAST searches with Chinese old garden roses were sourced from Zhang et al., 2025, which includes a total of 228 individuals (43 species and nine varieties from sect. Synstylae and three species from Chinenses), covering all currently accepted species in these two sections. Based on the haplotype analysis results, for subsequent haplotype network construction (Figs S3 and S4) and SSR analysis involving wild species, we included only the putative wild parental species of Chinese old garden roses. The SSR analysis encompassed 232 wild rose individuals from eight species and one variety (Table S2), along with the 25 Chinese old garden roses. The six phased, genotyped nuclear sequences and the three concatenated chloroplast gene sequences used are provided in Figs S3 and S4, while the unphased sequences have been uploaded to NCBI. We examined specimens and identified the wild species using a plant flora checklist (Huang, 1994; Iwatsuki et al., 2001; Gu, 2003), by comparison with the morphology of the type specimen, and by BLAST sequence alignment with the data in our previous study (Zhang et al., 2025). All three species within Rosa sect. Chinenses across the entire distribution range were collected (Table S2 and samples described by Li et al., 2023) to detect different alleles and shared haplotypes with Chinese old garden roses for origin analysis of the RoKSN gene (Tan et al., 2024). Silica gel-dried leaf material was utilized for DNA extraction for each individual. Voucher specimens were deposited at the herbarium at CIB. Rosa odorata ‘Parks’ Yellow’ is believed to be lost in history (Wang, 2015) and was replaced by a progeny of R. odorata ‘Park’s Yellow Tea-scented’. The distribution data of the putative parent species was collected in the Chinese Virtual Herbarium (CVH, https://www.cvh.ac.cn) with clear erroneous records removed and our fieldwork samples (Zhang et al., 2025).

Fig. 1.

The 25 diploid Chinese old garden roses sampled in this study. Photos by Cheng Zhang.

Chromosomal ploidy analyses

Chromosomal ploidy analyses followed procedures described previously (Wang et al., 2020). Two distinct standard plants (Lycopersicon esculentum Miller and Zea mays L.) were used as outer calibrations and the diploid taxon R. chinensis ‘Sanguinea’ was used for inner calibration. Traditional chromosome compressions were also carried out in parallel to assess the ploidy levels of old roses. Nuclear genome sizes were estimated with fresh leaves using flow cytometry based on the improved LB 01 buffer [15 mmol L⁻¹ Tris, 2 mmol L⁻¹ Na₂EDTA, 0.5 mmol L⁻¹ spermine tetrahydrochloride, 80 mmol L⁻¹ KCl, 20 mmol L⁻¹ NaCl, 0.1 % (v/v) Triton X-100, adjusted to pH 7.5 with 1 mol L⁻¹ NaOH]. The buffer was passed through a 0.22-µm filter, β-mercaptoethanol was added to 15 mm, and the buffer was stored at −20 °C in 10-mL aliquots. Chromosomal ploidy of old roses was assessed using a BD FacsCalibur (Becton Dickinson, San Jose, CA, USA) flow cytometer, which was fitted with an argon-ion laser operating at 488 nm. Fluorescence from propidium iodide was detected via the 620-nm fluorescence-2 filter. Data acquisition parameters remained consistent across all samples. The sample flow rate was maintained at approximately 100 nuclei s^–1, with a minimum of 6000 nuclei collected for each analysis. Data were processed using Cell Quest software (Becton Dickinson). A densely clustered nuclei region in the dot plot was gated to exclude irrelevant signals. The average coefficient of variation (CV) values for G1 peaks served as a quality metric, and results with CV values below 5 % were deemed reliable. Histograms were further evaluated using ModiFit v.3.0 (https://www.vsh.com/products/mflt/index.asp). Flow cytometry results were validated by performing chromosome compression analysis on four individuals. Living plants were gathered from the field and transferred to the glasshouse at CIB, CAS. Root tips were treated with 2 mm 8-hydroxyquinoline for 3–4.5 h, fixed in Carnoy’s solution for 0.5–24 h, softened in 1 n HCl at 60 °C for 57 min, and then stained with Carbol fuchsin. Chromosome numbers were determined using a light microscope (Olympus, Tokyo, Japan).

DNA extraction and genotyping

Total genomic DNA was extracted from silica-dried samples with the TIANGEN Plant Genomic DNA Extraction Kit (Tiangen Biotech, Beijing, China) following the manufacturer’s protocols. Three chloroplast regions (ndhF-rpl32, ndhJ-trnV and psbJ-petA) (Zhu et al., 2015) and seven nuclear markers, including nrITS and six single-copy genes (GBSSI, SebI, RosCOS0536, RosCOS2599, RosCOS3576 and PosF21), were selected for amplification and sequencing. Gene locations were identified by aligning sequences to the Rosa chinensis ‘Old Blush’ reference genome (GenBank accession: SAMN07737764) using NCBI BLAST (RRID: SCR_004870), and were visualized with TBtools v.2.225 (Chen et al., 2023). The six single-copy genes were located on different chromosomes (Supplementary Data Fig. S1). The RoKSN gene was amplified with primers designed for the RoKSN fragment using Primer BLAST in NCBI. Primers were designed for the flanking fragment of the Chinese old garden rose RoKSN retrotransposon region due to the absence of the retrotransposon in wild roses (Bai et al., 2021).

PCR amplification was performed in 25-μL volumes using a GeneAmp PCR System 3730xl DNA Analyzer (Applied Biosystems, Foster City, CA, USA). The reaction set-up comprised 18.75 μL of double-distilled water, 2.5 μL of 10× PCR buffer, 0.25 μL of Taq DNA polymerase (TransGen Biotech, Beijing, China), 0.5 mmol L⁻¹ dNTPs, 1.0 mmol L⁻¹ of each primer and approximately 1.0 μL of unquantified genomic DNA extract. PCR amplification began with initial denaturation at 95 °C for 4 min, followed by 38 cycles of 94 °C for 40 s, primer-specific annealing temperatures (ranging from 53 to 59 °C) for 1 min, and extension at 72 °C for 1 min, with a final extension at 72 °C for 10 min. Most nuclear gene data were obtained through direct sequencing. Sanger sequencing was performed using an Applied Biosystems 3500 device at Tsingke Biotech ( Beijing, China); 3′ end sequencing was not performed if the 5′ end was sequenced fully. However, if one strand showed anomalous peaks caused by indels or poly structures, sequencing of both strands was conducted. When sequencing failed or heterozygous (double) peaks were present, sequences were phased using DnaSP v.6.12.03 (Rozas et al., 2017) or were cloned for further analysis using the pClone007 Versatile Simple Vector Kit (Tsingke Biotech). Colony PCR was performed according to the manufacturer’s instructions with this kit. Briefly, single bacterial colonies were resuspended in 15 µL of sterile water, mixed thoroughly by pipetting, and 2 µL of this bacterial suspension was used as the PCR template. Subsequently, single colonies that gave positive PCR results were inoculated into LB liquid medium with ampicillin selection and incubated overnight at 37 °C with shaking at 250 rpm. Plasmid DNA was extracted and appropriate restriction enzymes were selected for digestion. After confirming the absence of cleavage, which was indicative of the presence of a cloned insert, the remaining DNA was used for sequencing.

Sequence alignment and phylogenetic analysis

Nucleotide sequences were processed and assembled using DNAStar LaserGene SeqMan v.3.1 (DNASTAR, Madison, WI, USA). Alignments were performed with Clustal W (Thompson et al., 1994) and subsequently refined manually in BioEdit v.7.0.9.0 (Hall, 1999). Mononucleotide repeats and indels were omitted due to uncertainties in homology. Haplotypes were generated based on the obtained sequences (linked to specific geographical regions) using DnaSP v.6.12.03 (Rozas et al., 2017). TCS haplotype networks for each marker were then constructed using PopART v.1.7 (Clement et al., 2002). The three chloroplast fragments were concatenated for analysis as a single marker. The nrITS data were also used to construct maximum likelihood (ML) phylogenetic trees using RAxML v.8.1.5 (Stamatakis, 2014) with 1000 rapid bootstrap replicates and split networks in SplitsTree v.6.1.16 beta (Huson and Bryant, 2006) using the neighbor-net algorithm.

Genotyping EST-SSR polymorphisms

Expressed sequence tag (EST)-SSRs were identified from the transcriptome of R. multiflora var. cathayensis (GenBank accession SAMN15581433) using MISA v.1.0 software (Thiel et al., 2003) to allow the design of primers targeting Rosa species. Primers for each EST-SSR were generated subsequently with Primer 3 v.0.4.0 (Koressaar and Remm, 2007; Untergasser et al., 2012). In total, 128 new primers from the EST-SSR database were synthesized for PCR along with 20 previously published primers (Esselink et al., 2003; Rusanov et al., 2005; Kimura et al., 2006; Saint-Oyant et al., 2007; Zhang et al., 2024a). Length polymorphisms of the EST-SSRs were evaluated via polyacrylamide gel electrophoresis coupled with silver staining. Sequence alignments were performed to identify indels in flanking sequences. Primer regions associated with potential null alleles were excluded. All 257 samples underwent genotyping following this optimization. PCR involved initial DNA denaturation at 95 °C for 5 min, followed by 10 cycles of denaturation at 95 °C for 30 s, annealing at 55 °C (decreasing by 0.5 °C per cycle) for 30 s, and extension at 72 °C for 1 min, with a final extension at 72 °C for 7–10 min. Fragment sizes of PCR products for 19 EST-SSRs (Supplementary Data Fig. S1; Table S3) were analysed using an ABI 3830xl DNA analyser (Applied Biosystems) at Sangon Biotech (Shanghai, China).

The genetic composition of Chinese old garden roses was determined with a Bayesian clustering method using STRUCTURE v.2.3.4 (Pritchard et al., 2000) based on individual SSR data. We did not test that Hardy–Weinberg equilibrium occurred within random mating populations as Chinese old garden roses were cultivated through artificial selection (Wang et al., 2022). In artificial selection (e.g. selective breeding for trait improvement), selective pressure and limited genetic variation often lead to the frequent transmission of specific allele combinations to offspring which results in a significant increase in linkage disequilibrium between loci. Individuals possessing desirable traits are selectively bred during this process which further increases linkage disequilibrium (LD) at specific loci and their adjacent regions (Bergelson and Purrington, 1996; Tanksley and McCouch, 1997). Due to the involvement of ancient rose varieties, we did not perform an LD test on the data, but the LD test of our EST-SSR loci was analysed previously in wild roses (Zhang et al., 2024a). Genetic cluster numbers (K) were tested across a range from 1 to 18. Each analysis consisted of 1600 000 Markov chain Monte Carlo (MCMC) iterations following an initial burn-in period of 800 000 iterations. The use of long burn-in times and extended run lengths ensured the consistency of the STRUCTURE results, with 20 replicate analyses conducted. Posterior probabilities for K, including LnP(D), Var[LnP(D)] and the change rates of LnP(D) between successive K values, were calculated. The optimal number of clusters was determined by maximizing LnP(D) (Falush et al., 2003) and using the ΔK method (Evanno et al., 2005) through Structure Harvester (Earl and von Holdt, 2012). CLUMPP v.1.2.2 (Jakobsson and Rosenberg, 2007) employed the Greedy consensus algorithm with 100 000 iterations and a randomized input order to average the 20 replicate results. The H′ statistic was used to measure similarity (Jakobsson and Rosenberg, 2007) and a structured cluster diagram was created.

For cultivars with a component proportion of two wild rose species >0.9 in STRUCTURE, NewHybrids v.1.1 beta (Anderson and Thompson, 2002) was employed to determine the genetic background, including hybridization and backcrossing, of Chinese old garden roses. For parental combination inferred by STRUCTURE, two independent analyses were conducted using NewHybrids. The analysis involved running a total of 1600 000 MCMC chains, with the first 800 000 chains discarded during the burn-in phase. Jeffreys-type priors were utilized for allele frequencies and mixing proportions. Posterior probabilities (q) were calculated to assess the likelihood of an individual belonging to specific genotype frequency categories. The q value in STRUCTURE represents the proportion of an individual’s genotype derived from each K group. Conversely, q in NewHybrids indicates the probability of an individual being classified into categories such as parental purebreds, F1 hybrids, F2 hybrids or first backcross generations. A threshold value (Tq) was applied to classify individuals accurately: q ≥ 0.9 in STRUCTURE or ≥0.75 in NewHybrids was used to identify purebreds, whereas values below these thresholds indicated hybrids or introgressed individuals (Burgarella et al., 2009). Additionally, principal coordinate analysis (PCoA) was conducted using GenAlEx v.6.502 (Peakall and Smouse, 2012). Data standardization was performed prior to calculating the covariance matrix, and the R package ggplot2 (Wickham, 2016) was utilized to visualize the PCoA clustering outcomes. Nei’s genetic distance (DA) (Nei and Chesser, 1983) among the 257 individuals was calculated using PowerMarker v.3.25 (Liu and Muse, 2005).

Morphological trait measurement and analysis

Phenotypic data were collected from specimens in CIB. Sixteen morphological traits that are known to be important for cultivar identification of Chinese old garden roses and their putative parents were selected, including 24 qualitative traits and nine quantitative traits (Huang, 1994; Gu, 2003; Guan et al., 2024; Supplementary Data Table S4). Leaf traits (stipule margin shape, hairs on leaves abaxially, leathery texture, hairs on petioles, number of leaflets, size/shape among leaflets, top leaflet length/width and lateral second leaflet length/width) and flower traits (hairs on calyx tube/stigma, hairs on pedicel and flower diameter) were observed for data measurement and analysis.

Kaiser–Meyer–Olkin (KMO) and Bartlett’s sphericity tests during factor analysis using IBM SPSS Statistics v.24 (Released 2016; IBM, Armonk, NY, USA) were performed to evaluate the robustness of partial correlations within the dataset and to verify alignment with an identity correlation matrix. Missing data points were addressed through imputation utilizing the R package mice (van Buuren and Groothuis-Oudshoorn, 2011). The prcomp function in R was applied to perform principal component analysis (PCA) for the dataset, and the R package factoextra v. 1.0.7 (Kassambara, 2020) was then used to extract and visualize PCA results.

RESULTS

Estimation of nuclear DNA content indicates that Chinese old garden roses are diploid

Flow cytometry analysis revealed that all 25 Chinese old garden roses analysed were diploid (Supplementary Data Table S1). Chromosome sizes were similar, ranging from 0.49 to 0.60 pg, with the smallest chromosomes being those of R. odorata ‘Hume’s Blush Tea-scented’, R. chinensis ‘Sanguinea’ and R. hybrid ‘Mutability’ (0.49–0.52 pg) based on chromosome compression results for R. chinensis ‘Sanguinea’. The ploidy levels of the three old rose cultivars determined from flow cytometry were consistent with those obtained by the traditional chromosome tableting technique (Fig. S1).

Phylogenetic tree and splits network based on nrITS

In the ML phylogenetic tree (Fig. 2A), most nrITS haplotypes of Chinese old garden roses clustered together with Rosa chinensis var. spontanea North, forming a distinct clade. Only five Chinese old garden rose cultivars, including R. odorata ‘Park’s Yellow Tea-scented’, R. odorata ‘Light Yellow Tea-scented’, Tengchong, China, R. chinensis ‘Dark red’, R. odorata ‘Hume’s Blush Tea-scented’, and R. hybrida ‘Clotilde Soupert’, either entirely or partially, share nrITS haplotypes with R. odorata var. gigantea.

Fig. 2.

ML phylogenetic tree. (A) Maximum likelihood bootstrap support (BS) values are shown along the branches (the circle represents BS = 0–94 %). For the NeighborNet network based on nrITS (B) in Chinese old garden roses, the wild rose species are marked with alphanumeric labels.

Multiple box-like or reticulate structures were observed in the splits network analysis (Fig. 2B), especially among the Chinese old garden roses labelled 1–19, 24 and 25, which suggest conflicting phylogenetic signals and past hybridization events. Chinese old garden roses (indicated in dark red) displayed extensive connections to both R. chinensis var. spontanea North and R. odorata var. gigantea, suggesting frequent hybridization events and their key role in the domestication of Chinese old garden roses.

Haplotype phasing analysis of barcodes uncovers the origins of Chinese old garden roses

After excluding single-base repeats and ambiguously aligned indels, gaps were coded as single mutational events; the sequence matrix information is summarized in Table 1. All loci had haplotype diversity values exceeding 0.85, demonstrating the high resolution of our selected markers in differentiating among the studied species.

Table 1.

Overview of polymorphism data of each site.

DNA barcoding	Length (bp)	Variable sites	Parsimony-informative sites	Single sites	Number of haplotypes	Haplotype diversity (Hd)	Nucleotide Diversity (Pi)	CG content (%)
nrITS	607	27	24	3	18	0.864	0.01135	56.3
GBSSI	914	63	39	24	63	0.966	0.01181	36.7
SebI	739	26	16	10	33	0.920	0.00673	42.5
RosCOS0536	499	43	41	2	42	0.931	0.02054	36.6
RosCOS2599	917	24	18	6	50	0.924	0.00940	37.7
RosCOS3576	466	29	23	6	39	0.942	0.01341	35.8
PosF21	673	34	25	9	38	0.951	0.01147	33.9
psbJ-petA&ndhJ-tabE&ndhF-rpl32	3179	71	63	8	20	0.872	0.00269	30.0

Phase analysis of the haplotypes of 25 Chinese old garden roses was conducted and the sequences were compared with wild Rosa species in China. Most haplotypes were shared across both wild species and cultivars, although some cultivated individuals possessed unique haplotypes. Each network illustrated distinct patterns of haplotype distribution based on the clustering and topology of the network (Supplementary Data Figs S3 and S4). Due to the high morphological and haplotype similarity between Rosa multiflora and R. daishanensis, R. kwangtungensis and R. taiwanensis, as well as R. chinensis var. spontanea South and R. lucidissima, each of these pairs was treated as a single taxon in the haplotype analyses (Fig. 3). Most nuclear gene haplotypes of Chinese old garden roses were the same as those of wild species R. chinensis var. spontanea, R odorata var. gigantea and R multiflora var. cathayensis, which indicates that these three species are important parents of ancient roses. The haplotypes of R. chinensis cultivated in Ya’an and R. chinensis var. spontanea were similar, which suggests that the former may be a natural mutant of the latter. In addition, the haplotypes of 11 cultivars, i.e. R. chinensis var. semperflorens, R. chinensis (cultivated in Suining, Sichuan, China), R. chinensis (cultivated in Hainan, China), R. chinensis f. viridiflora, R. chinensis ‘Old Blush’, R. chinensis ‘Pompon de Paris’, R. chinensis (cultivated in Changshu, Jiangsu, China), R. chinensis ‘Minima’, R. hybrida ‘Fortune’s Five Color’, R. hybrid ‘White Fortune’s Five Color’ and R. chinensis ‘Emie Gray’, are derived from R. chinensis var. spontanea and R multiflora var. cathayensis which indicates that the cultivars are hybrid offspring of these two wild roses. Among these 11 cultivars, R. chinensis (Hainan), R. chinensis f. viridiflora, R. chinensis ‘Old Blush’, R. chinensis ‘Pompon de Paris’, R. chinensis (Changshu) and R. chinensis ‘Minima’ had identical haplotypes among the seven genes that were tested which suggests that these six cultivated species may originate from bud mutations of the same clone plant.

Fig. 3.

Haplotype phasing analysis of nrITS, six single-copy genes and chloroplast genes of diploid Chinese old garden roses compared with wild roses. Each of the following pairs was treated as a single taxon: Rosa multiflora and R. daishanensis, R. kwangtungensis and R. taiwanensis, and R. chinensis var. spontanea South and R. lucidissima. The last column presents maternally inherited plastid markers, while the other columns represent nuclear markers.

The cultivars R. chinensis ‘Dark red’, R. chinensis ‘Slater’s Crimson China’, R. chinensis ‘Slater’s Crimson China, Cl.’ and R. chinensis ‘Semi-double, Cl.’ possessed two genes that exhibited the same haplotype as the R. kwangtungensis species complex which therefore may be involved in hybrid breeding of these four cultivated species. The nrITS locus of R. chinensis ‘Dark red’ also had a shared haplotype with R. odorata var. gigantea which suggests that multiple wild roses were involved in breeding of the former species. The main haplotypes of six cultivars, i.e. R. hybrid ‘Bermuda Spice’, R. chinensis ‘Sanguinea’, R. odorata ‘Xiangyun Pink’, R. odorata ‘Park’s Yellow Tea-scented’, R. odorata ‘Light Yellow Tea-scented’ (cultivated in Tengchong, China) and R. odorata ‘Hume’s Blush Tea-scented’, were shared with R. chinensis var. spontanea, R. odorata var. gigantea and R. multiflora var. cathayensis. Rosa odorata var. gigantea has played a particularly important role in breeding of these cultivars.

Cultivars R. hybrid ‘Clotilde Soupert’, R. chinensis ‘Minima’ (cultivated in Taiwan, China) and R. hybrid ‘Mutability’ had haplotypes that were shared with wild species, among which R. multiflora var. multiflora or R. daisanensis is an important parent of R. hybrid ‘Clotilde Soupert’. In addition, R. kwangtungensis or R. taiwanensis is a key progenitor of R. chinensis ‘Minima’. The genetic composition of R. hybrid ‘Mutabilis’ is the most complex with at least five wild roses participating in the hybrid breeding of this species. Finally, the haplotypes of R. glomerata were identified in R. odorata ‘Park’s Yellow Tea-scented’ and R. hybrid ‘Mutabilis’ which suggests that R. glomerata may be involved in breeding of these cultivars.

Genetic structure analysis by microsatellite markers reveals the genetic history of wild Rosa species and Chinese old garden roses

In view of the preceding results of haplotype analysis, clustering analysis of putative wild Rosa species and 25 Chinese old garden roses was conducted. The haplotype of R. paniculigera, which is distributed in Japan, was observed only in RosCOS3576 of R. hybrida ‘Clotilde Soupert’ and was excluded from subsequent EST-SSR analysis. The number of alleles (k) per locus varied from 5 to 17, with an average of 8.29 alleles per locus for the 21 polymorphic EST-SSR loci (Supplementary Data Table S2). Values for the polymorphic information content varied from 0.171 to 0.831 with an average of 0.63 which suggests sufficient discriminatory power. The 21 EST-SSR loci were not in LD (Table S3). Bayesian clustering analyses suggested that the optimum k value was 6 (Fig. S2A, B). Twenty iterations in STRUCTURE showed very high similarity with the highest H′ of 0.949 (Jakobsson and Rosenberg, 2007).

Clustering analysis indicated that R. chinensis var. spontanea may be divided into northern and southern types, with the Sichuan Basin as the geographical boundary (Fig. 4A). Rosa lucidissima clustered together with southern types which is consistent with previous observations (Li et al., 2023). The morphologies of these two species also are very similar, although R. lucidissima possesses dense bristles with glands on branches, pedigrees and receptacles that distinguish this species from R. chinensis var. spontanea. The R. multiflora species complex, R. kwangtungensis species complex and R. glomerata all are within sect. Synstylae and have distinct genetic features. In contrast, R. multiflora var. cathayensis, R. multiflora var. multiflora and R. daishanensis share the same genetic components. All of these species belong to the R. multiflora species complex which has common morphological features, including pectinate margins on the stipules and hairless stigma. Rosa kwangtungensis and R. taiwanensis have similar genetic configurations but with differences in leaf hairs with the former having leaflets that are abaxially pubescent and the latter displaying abaxially glabrous leaflets.

Fig. 4.

Bayesian clustering plots of diploid Chinese old garden roses and the population of their putative parents using 21 EST-SSR loci in STRUCTURE and NewHybrids. (A) Individual cluster assignment at K = 6. (B) Genetics components of the Chinese old garden roses. (C) Genetic makeup and hybrid introgression identified using NewHybrids for a pairwise comparison.

Among the 25 Chinese old garden roses, R. chinensis cultivated in Ya’an was closest to that of the wild species R. chinensis var. spontanea, reaching 99.29 % similarity. In addition, the similarity with R. chinensis var. spontanea also reached 98.29–98.98 % in R. chinensis var. semperflorens, R. chinensis (Suining), R. chinensis (Hainan), R. chinensis f. viridiflora, R. chinensis ‘Old Blush’, R. chinensis ‘Pompon de Paris’, R. chinensis (Changshu) and R. chinensis ‘Minima’. The compositions of the last six cultivars were identical which is consistent with the results of barcoding. The genetic compositions of R. hybrida ‘Fortune’s Five Color’ and R. hybrida ‘White Fortune’s Five Color’ were also consistent, with only morphological differences in flower colour: the former is pink, whereas the latter is white. These two cultivars may be clones of different bud mutations.

Some parental species were detected in the SSR results but were not observed in the barcode analysis. Rosa chinensis ‘Emie Gray’ contained 14.59 % of the components of the R. kwangtungensis species complex in clustering analysis. Rosa chinensis ‘Dark Red’ possessed 18.87 % of the genetic components of R. odorata var. gigantea, and R. chinensis ‘Slater’s crimson China’, R. chinensis ‘Slater’s crimson China, Cl.’ and R. chinensis ‘Semi-double, Cl.’ each contained 21.86–22.6 % of the genetic features of the R. kwangtungensis species complex. Rosa hybrid ‘Bermuda Spice’, R. chinensis ‘Sanguinea’, R. odorata ‘Xiangyun pink’, R. odorata ‘Park’s Yellow Tea-scented’, R. odorata ‘Light Yellow Tea-scented’ (Tengchong) and R. odorata ‘Hume's Blush Tea-scented’ mainly contained genetic features of both R. chinensis var. spontanea (71.35, 68.58, 65.76, 64.66, 61.72 and 56.56 %, respectively) and R. odorata var. gigantea (24.84, 30.42, 32.13, 32.73, 37.14 and 41.53 %, respectively). The genetic compositions of R. hybrid ‘Clotilde Soupert’, R. chinensis ‘Minima’ (Taiwan) and R. hybrid ‘Mutability’ were relatively complex. Rosa hybrid ‘Clotilde Soupert’ contained features of R chinensis var. spontanea (54.67 %), R. odorata var. gigantea (20.83 %) and R. multiflora var. cathayensis (23.56 %). Instead, R. chinensis ‘Minima’ (Taiwan) and R. hybrid ‘Mutability’ are derived mainly from R. odorata var. gigantea, the R. kwangtungensis species complex and R. multiflora var. cathayensis, with proportions of 53.12, 22.34, 21.86 %, respectively, for the Taiwan cultivar, and 47.27, 24.02 and 22.24 %, respectively, for R. hybrid ‘Mutability’ (Fig. 4A, B).

The NewHybrids data (Fig. 4C1) demonstrated that R. odorata ‘Hume’s Blush Tea-scented’ and R hybrid ‘Bermuda Spice’ were the F1 hybrid generation of R. chinensis var. spontanea and R. odorata var. gigantea. Figure 4C2 shows that R. chinensis ‘Sanguinea’, R. chinensis ‘Dark red’, R. odorata ‘Xiangyun Pink’, R. odorata ‘Light Yellow Tea-scented’ (Tengchong) and R. odorata ‘Park’s Yellow Tea-scented’ may be the F2 generation hybrids of these parental types. The data also suggest that R. chinensis ‘Emie Gray’, R. chinensis ‘Slater’s crimson China’, R. chinensis ‘Slater’s crimson China, Cl.’ and R. chinensis ‘Semi-double, Cl.’ may be hybrid offspring of R. chinensis var. spontanea and the R. kwangtungensis species complex, followed by multiple backcrosses with R. chinensis var. spontanea.

In PCoA (Fig. 5), PCoA1 (46.97 %) and PCoA2 (20.2 %) explained 67.17 % of the total genetic variation which is a relatively high value. Most of the Synstylae species were closer to each other than to R. glomerata (Fig. 5, red cluster). Rosa odorata var. gigantea from sect. Chinenses grouped together with the Synstylae species. Among the 25 Chinese old garden roses (Fig. 5, purple cluster), R. chinensis var. spontanea and R. lucidissima were relatively close, which indicates that these two varieties are the most important parents (Fig. 5). Rosa odorata ‘Xiangyun Pink’, R. hybrida ‘Mutabilis’, R. chinensis ‘Minima’ and R. chinensis ‘Emmie Gray’ are genetically intermediate which suggests a mixed genetic background. Rosa chinensis (Ya’an) genetically was close to R. chinensis var. spontanea. Although R. chinensis var. semperflorens and R. chinensis (Suining) also placed within R. chinensis var. spontanea groups in PCoA based on EST-SSR data (Fig. 4), the components of R. multiflora var. cathayensis in these two cultivars were only detected in the haplotype analysis. We hypothesize that these two cultivars are hybrids between R. chinensis var. spontanea and R multiflora var. cathayensis, followed by repeated backcrossing with the single-petal rose which led to dilution of the contribution from R. multiflora var. cathayensis.

Fig. 5.

Principal coordinate analysis (PCoA) based on data from 21 EST-SSR loci of Chinese old garden roses and putative wild parents. Each colour indicates a different Rosa group with ellipses showing 95 % confidence intervals.

Nei’s genetic distance analysis revealed that R. daishanensis, R. multiflora var. multifora, R. taiwanensis, R. kwangtungensis, R. chinensis var. spontanea and R. lucidissima exhibited close genetic relationships. Of the 25 Chinese old garden roses, only R. chinensis (Ya’an) had a short genetic distance to the wild species R. chinensis var. spontanea, whereas other old roses showed relatively close genetic distances to each other (Fig. 6). In summary, the results from both the haplotype and microsatellite analyses indicated that R. chinensis (Ya’an) exhibited the closest genetic structure to the wild-type single-petal rose, with no evidence of hybridization with other wild species.

Fig. 6.

Nei’s genetic distance (DA) among the diploid Chinese old garden roses and the population of their putative parents based on 21 EST-SSR loci.

Origin analysis of the RoKSN gene

The origin of the RoKSN gene was analysed by amplifying both the wild-type and RoKSN-copia genes for sequencing (Fig. 7A). RoKSN-copia fragments were present in all 25 Chinese old garden samples and shared a single haplotype. The fragments lacking retrotransposons were compared to those of R. chinensis var. spontanea using a Network graph. The RoKSN gene was identified at haplotype 16 (Fig. 7B) which clusters with haplotypes from R. chinensis var. spontanea populations in Tianquan, Wushan, Cangxi, Anyue and Chengdu (Fig. 7C). This observation was explored further by population sampling in these five regions and amplifying the RoKSN gene. The proportions of this haplotype corresponding to the RoKSN gene were 55, 33.3, 22.7, 20 and 15.6 % in Tianquan, Anyue, Wushan, Cangxi and Chengdu, respectively (Fig. 7D). Thus, the haplotypes of the RoKSN-copia gene predominantly occurred in the Sichuan Basin region.

Fig. 7.

Detection and distribution of different alleles at the RoKSN locus without recombination of the copia element. (A) Amplification primers RoKSN-F and RoKSN-R were designed for wild rose whereas RoKSN-F, LTR1-R and LTR2-F, and RoKSN-R were designed for Chinese old roses. (B) TCS haplotype networks based on RoKSN of Chinese old roses and putative parents. (C) ML phylogenetic tree based on RoKSN. (D) Right: distribution of samples from Rosa sect. Chinenses and the distribution of RoKSN. Left: proportion of haplotypes identical to the flanking fragments of the RoKSN gene. (E) R. chinensis var. spontanea from Ya‘an, Sichuan, China. Photo by Qi Yu.

Comparative analysis of the morphology of Chinese old garden roses and their hypothesized progenitors

PCA was applied to 26 traits in the 25 Chinese old garden roses. The KMO test of sampling adequacy (0.763) and Barlett’s test of sphericity (χ² = 2208.342, P < 0.001) indicated the data were acceptable for factor analysis (Cerny and Kaiser, 1977). The first five principal components accounted for 67.6 % of the total variance, and the first two main principal components accounted for 33.2 % and 9 %. The factor loading threshold >0.2 was considered significant. The 15 characters with positive factor loading >0.2 on the first axis included flowering frequency, inflorescence type, flower colour, number of leaflets and apical leaflet width. Seven characters with positive factor loading >0.2 on the second axis included petal number and anther colour (Fig. 8B–D). The Chinese old garden roses form a distinct clade that showed closer similarity to Rosa species in R. sect. Chinenses in PC1, while exhibiting greater affinity to species in sect. Synstylae in PC2.

Fig. 8.

Principal component analysis of 16 morphological traits and significance test of five quantitative characters. (A) Scree plot of the factor analysis. (B) Variance contribution rate analysis for 36 factors. (C) Variable factor map for 16 factors. (D) PCA with 95 % confidence intervals shown in ellipses.

DISCUSSION

We collected 25 diploid traditional Chinese old garden roses, most of which were hybrids from the crossbreeding of species from Rosa sects Chinenses and Synstylae. One haplotype from RosCOS3576 in R. hybrida ‘Clotilde Soupert’ was identical to R. paniculigera from Japan and was not examined further (Fig. 3). Key molecular evidence revealed that Chinese old garden roses originated from reticulate evolution mainly from wild species within R. chinensis var. spontanea, R. lucidissima, R. odorata var. gigantea, R. multiflora var. cathayensis and the R. kwangtungensis species complex (R. kwangtungensis or R. taiwanensis).

Putative parental species of diploid Chinese old garden rose hybrids from Rosa sects Chinenses and Synstylae

Analysis of haplotypes, genetic components and EST-SSRs allowed identification of several species from R. sects Chinenses and Synstylae as the parental species of diploid Chinese old garden roses. However, R. chinensis cultivated in Ya’an exhibited no hybridization signals. The genetic composition of this variety closely resembled that of the wild species R. chinensis var. spontanea. Other rose cultivars were confirmed as descendants of past hybridizations with R. chinensis var. spontanea, R. odorata var. gigantea and R. multiflora var. cathayensis as main donors for the 25 varieties of cultivated roses. The ancient roses of hybrid origin were divisible into four categories including ten cultivars that were hybrids between R. chinensis var. spontanea and R. multiflora var. cathayensis, and five cultivars that were hybrids involving R. chinensis var. spontanea, R. multiflora var. cathayensis and the R. kwangtungensis species complex. Six cultivars were hybrids among R. chinensis var. spontanea, R. multiflora var. cathayensis and R. odorata var. gigantea, and three cultivars were hybrids with more than three donors. The genetic distances between ancient roses are relatively close with the same RoKSN-copia gene in all 25 diploid Chinese garden old roses that leads to continuous flowering. These observations suggest that the RoKSN-copia alleles have a common single origin.

Certain species with few features were detected in gene haplotype analysis. For example, the genetic components of R. chinensis ‘Pompon de Paris’, R. chinensis var. semperflorens, R. chinensis f. viridiflora, R. chinensis ‘Old Blush’ and R. chinensis ‘Emmie Gray’ (also known as ‘Ren Mian Tao Hua’ in Zhang et al., 2024c) were identical to those of R. chinensis var. spontanea based on SNPs in admixture analysis (Zhang et al., 2024c) and SSRs in this study. However, R. multiflora var. cathayensis haplotypes were found in two or more loci by barcodes in those cultivars (Fig. 3). Rosa multiflora var. cathayensis was not sampled previously for parental analysis (Zhang et al., 2024c). It was suggested recently that the genetic components of R. hybrida ‘Fortune’s Five Color’ (also known as ‘Feng Ye Lu Hua’ in Zhang et al., 2024c) were principally R. chinensis var. spontanea and an unidentified rose. Our results further indicated that R. hybrida ‘Fortune’s Five Color’ and R. chinensis ‘Dark Red’ were hybrid offspring of R. chinensis var. spontanea and R. multiflora var. cathayensis.

Both R. chinensis ‘Emmie Gray’ and R. chinensis ‘Slater's Crimson China’, which is one of the four ancient roses, possessed 14.6 % and 21.9 % genetic components, respectively, of R. wichuraiana (also known as R. luciae) that is indigenous to Hubei, Sichuan and Guizhou in China (Zhang et al., 2024c). Examination of numerous specimens here revealed that certain individuals of R. kwangtungensis with glabrous abaxial leaflets often are mistaken for R. luciae which frequently has 7–9 leaflets with rounded apices and is distributed only in Zhejiang in China, South Korea and Japan. In contrast, R. kwangtungensis possesses 5–7 small leaves with acute apices and is disseminated in Guangdong, China.

Both R. chinensis ‘Sanguinea’ (Dan Ban Xing Hong Yue Yue Hong) and R. odorata ‘Xiang Yun Pink’ hybridized with R. chinensis var. spontanea (or R. lucidissima) and R. odorata var. gigantea, which was consistent with our EST-SSR results. However, a contribution from R. multiflora var. cathayensis in the single-copy nuclear gene PosF21 was also detected which suggests that two species may have been involved in early stage breeding of R. chinensis ‘Sanguinea’. Chloroplast genotype analysis implicated R. lucidissima and R. odorata var. gigantea, respectively (Fig. 3).

The main genetic components of R. hybrida ‘Clotilde Soupert’ are derived from R. chinensis var. spontanea and R. chinensis ‘Old Blush’, although R. multiflora contributes 23.6 % to the composition of this variety. Rosa multiflora var. multiflora was detected in multiple nuclear gene and chloroplast haplotypes, which may be an initial parent of R. hybrida ‘Clotilde Soupert’.

The genetic anatomy of R. hybrida ‘Mutabilis’ is complex. This cultivar is a blend of R. chinensis ‘Old Blush’, R. chinensis var. spontanea, R. odorata var. gigantea as well as species in sect. Synstylae (Zhang et al., 2024c) which was verified by the analysis here. Interestingly, the GBSSI nuclear gene in R. odorata ‘Parks Yellow Tea-scented’ and R. hybrida ‘Mutabilis’ shares the same haplotype with R. glomerata which has not been reported previously in ancient rose parents. We further clarified that members of the hybrid stem group involved in this hybridization are derived mainly from the R. multiflora var. cathayensis and R. kwangtungensis species complexes. As previous analysis did not encompass chloroplast genes or genomes (Zhang et al., 2024c), the current data assisted in identifying the initial mother plants of these old roses.

The Sichuan Basin is the most likely origin of the RoKSN gene

Cheng et al. (2025) further confirmed, through genome-wide association study analysis, that the TFL1 (terminal flower 1, known as KSN in Rosa) locus, located on chromosome 3 at 14.6–15.6 Mb, controls the regulation of flowering time. The RoKSN alleles in all 25 Chinese old garden roses share a unique genotype, carrying the same copia transposon insertion (Fig. 7B, C), which suggests that the RoKSN-copia variants have a single progenitor. Wild populations of R. sect. Chinenses distributed in China were collected to further investigate the geographical origin of the gene (Fig. 7D; Li et al., 2023) which demonstrated that the haplotypes of the RoKSN-copia gene fragments were identical in only five populations, i.e. Tianquan, Anyue, Wushan, Cangxi and Chengdu, all located in the Sichuan Basin (Fig. 7E). We infer that the RoKSN-copia gene predominantly occurred in the Sichuan Basin. Our molecular results suggest that the R. chinensis cultivar in Ya’an may represent the earliest mutated individual that bears the RoKSN-copia gene in a wild species. This cultivar potentially served as a genetic bridge from wild species to the continuously flowering old rose cultivars, and provides a foundation for studying the origins and mechanisms of mutations that are associated with continuous flowering. This hypothesis is supported by the presence of R. chinensis var. spontanea in Ya’an (Fig. 7E). This cultivar shares a similar colour-changing flowering process, transitioning from pink to a deep red colour in the later stages of flowering, with R. chinensis from the same region.

Domestication and cultivation history of diploid Chinese old garden roses

Morphological analysis indicated that many characteristics of Chinese old garden roses, including cup-shaped flowers, glandless and glabrous pedicels, and entire stipules, closely resemble those of R. chinensis var. spontanea. This finding corroborates previous predictions that the latter is the primary progenitor of Chinese old garden roses (Meng, 2012; Wang, 2015). Furthermore, the predominant inflorescence type among these roses is corymbose, which is a trait that may be derived from species within Rosa sect. Synstylae. The high degree of morphological similarity observed among Chinese old garden roses combined with genetic clustering results suggest that the ancestral progenitors of these varieties may have originated from a limited number of individuals during the domestication progress.

As outlined above, together with the natural distribution area of the putative parents (Supplementary Data Fig. S6), we infer that the RoKSN genes in old roses that bloom all year round originated from the Sichuan Basin. Rosa chinensis var. spontanea with the RoKSN-copia gene hybridized with R. multiflora var. cathayensis in the local Sichuan area, thereby producing the ‘Old Blush’ group of old roses. This group was introduced into the Yunnan region in China and then crossed with the local wild rose R. odorata var. gigantea to generate the ‘Tea Rose’ group. The ‘Old Blush’ rose was introduced subsequently to Taiwan and Guangdong regions and hybridized with the local R. kwangtungensis species complex to produce the ‘Slater's Crimson’ group (Fig. 9).

Fig. 9.

Schematic diagram of the hybrid domestication process of Chinese old garden roses. Letters represent retrotransposon insertion (a) and putative hybridization events (b–e).

CONCLUSIONS

We collected 25 of the most renowned and classic diploid Chinese garden old roses along with all the related wild roses from East Asia with the aim of identifying potential ancestral donor species responsible for the development of known cultivars. The genetic backgrounds of these varieties were investigated by assessing the haplotypes for nrITS and six single-copy genes, and evaluating genetic structure using 21 EST-SSRs. Barcodes may be more effective than admixture clustering methods in pinpointing parental species with diluted genetic components from early hybridization. Our analysis of the distribution of putative parental species and the origin of the continuous flowering gene RoKSN-copia enabled reconstruction of the history of hybridization and domestication of old roses. Notably, several morphological traits of Chinese old garden roses closely resemble those of wild single-petalled R. chinensis, whereas the corymbose inflorescence shape may have originated from species in sect. Synstylae. The cultivation and domestication of these ancient Chinese cultivars over the past century have resulted in the accumulation of numerous ideal agronomic traits. These traits have been preserved and further selected in modern roses which are one of the most important ornamental crops worldwide. Due to gene dilution from subsequent hybridizations, the components of certain parental species involved in early hybridization may not be visible in admixture clustering plots even using whole genome SNPs. However, these early hybridization parents may be identified accurately by combining barcoding with EST-SSRs. Identifying wild parent species provides a foundation for subsequent analyses of the hybridization history of Chinese old garden roses using genomic data.

Supplementary data

Supplementary data are available at Annals of Botany online and consist of the following. Supplementary file 1: Supplementary figures and tables. Figure S1. Microscopic examination of rose chromosomes. Figure S2. Distribution of six single-copy nuclear genes, RoKSN, and 21 microsatellite loci on chromosomes utilized in this study. Gene locations were identified by aligning sequences to the Rosa chinensis ‘Old Blush’ reference genome (GenBank accession: SAMN07737764) using NCBI BLAST. Figure S3. TCS haplotype networks based on nuclear genes of 228 samples of Rosa sections Chinenses and Synstylae. A, nrITS; B, SebI; C, GBSSI; D, RosCOS0536. The size of circles correlates with the number of accessions representing a haplotype (the large circle in the key represents ten haplotypes, while the small circle represents one haplotype). Black dots correspond to intermediate missing haplotypes absent in the sample. Resolved closed loops are shown with dotted lines. Haplotype colours and key designations (in the upper right corner) correspond to taxonomic complex of Rosa sects Chinenses and Synstylae (as in Fig. S4). Figure S4. TCS haplotype networks based on nuclear genes of 228 samples of Rosa sections Chinenses and Synstylae. A, RosCOS2599; B, RosCOS3576; C, PosF21; D. the three chloroplast genes. Figure S5. Estimation of the optimal number of clusters (K) based on STRUCTURE analysis. A, posterior probability of 20 replicates per K (mean ± s.d.); B, the distribution of ΔK. Figure S6. The natural distribution of the putative parents of Chinese old garden roses in China. Distribution data were collected from the Chinese Virtual Herbarium (CVH, https://www.cvh.ac.cn) and our fieldwork samples (Zhang et al., 2025). Table S1. Nuclear DNA content and genome size of 25 diploid Chinese old garden roses. Table S2. Detailed information of wild rose samples used for microsatellite analysis in this study. Table S3. Primer list for hybridization identification. Table S4. The 21 pairs of EST-SSR primer sequences. Table S5. List of qualitative and quantitative characters used in numerical analyses.

Supplementary file 2: Genotyping results of 21 microsatellite loci in 257 Rosa samples.

Supplementary file 3: Morphological trait measurements of Rosa samples for PCA (Fig. 8).

Supplementary file 4: Alignment of six phased nuclear sequences and three concatenated chloroplast gene sequences analysed in Supplementary Data Figs S3–S4.

Funding

This research was supported by the Natural Science Foundation of Sichuan Province (Grant No. 2022NSFSC0012) to X.-F.G.

Author contributions

Gao X.F. offered study supervision and acquired funds. Zhang C. and Jiang Z.Z. conceived and designed the experiments. Zhang C. performed the molecular experiment and wrote the manuscript. Yang S.Z. helped to collect the samples. Li S.Q. helped with the experiment. Liang Z.L. conducted chromosome compression experiments. All the authors read and approved the final manuscript for publication.

Data availability

The data that support the findings of this study are available from GenBank (NCBI). GenBank accessions: ndhF-rpl3-2R: PQ616323–PQ616374, trnL-trnF: PQ616375–PQ616423, psbJ-petA: PQ616424–PQ616473, nrITS: PQ466191–PQ466428, SebI: PQ486539–PQ486776, GBSSI: PQ486777–PQ487014, RosCOS0536: PQ487028–PQ487265, RosCOS2599: PQ487279–PQ487515, RosCOS3576: PQ487547–PQ487784, PosF21: PQ616137–PQ616174, KSN: PQ616175–PQ616322.