Rosa rugosa is a perennial shrub tree that grows to 1–1.5 m tall and is native to Eastern Asia. It blooms and produces edible hips (the seed pods) in summer and early autumn. R. rugosa has been utilized in many ways. Because of its attractive pink flowers, R. rugosa is often used to create windbreaks and hedges. It has also been cultivated in North America and Europe as an introduced ornamental plant. The fruits of R. rugosa possess antioxidant activity and antibacterial activity due to their high contents of phenolic and flavonoid compounds and ascorbic acid1,2. It is able to control soil erosion and is planted along highways in Germany and Denmark3. Because of the high level of biosynthesis of pleasant volatile compounds in its flowers, R. rugosa has been used as an important source for the production of essential oil4. In breeding, R. rugosa has been widely used for breeding salt-resistant Rosa varieties. Although R. rugosa has many advantages, research on its molecular breeding and domestication has not even begun, partly due to the lack of high-quality genome sequences.

Also known as rugged rose, R. rugosa can adapt to many environmental conditions, such as salinity and alkaline soils, shade, frigid temperatures, drought, and high humidity. These excellent abilities make R. rugosa ideal for gene mining and molecular breeding to produce novel Rosa varieties. In some places, R. rugosa has become invasive5, attesting to its ability to adapt to new environments. However, the molecular mechanisms underlying this adaptability are largely unknown.

Following the rapid development of genome sequencing technologies and bioinformatic technologies, hundreds of angiosperm genomes have been reported6,7,8. The Rosa genus includes ~200 species with quite different morphological traits9. Within the Rosa genus, the first draft genome sequence of wild Rosa multiflora was released in 201810. Since then, two chromosome-level genomes of Rosa chinensis, also known as Chinese rose, have been released11,12. For R. rugosa, only the chloroplast genome13 and mitochondrial genome14 have been reported. A high-quality genome sequence for R. rugosa would not only enable comparative genomic studies of Rosa species but also reveal the mechanisms underlying its ornamental traits, such as floral biology and its unique ecology.

Here, we report the first chromosome-level genome assembly of R. rugosa, relying on HiFi sequencing and Hi-C scaffolding technology. Based on this high-quality genome assembly, we studied the genomic structural differences between R. rugosa and R. chinensis. We also revealed the genetics responsible for floral biology. The mechanisms that account for its evolution and adaptation to harsh environments were explored here as well.

Results and discussion

Genome sequencing and assembly

We used a combination of sequencing technologies, including PacBio-CCS (HiFi), 10X genomics, and Hi-C, to construct the reference genome for R. rugosa. We obtained a total of 59.24 Gb HiFi clean data and 80.91 Gb 10X genomics clean data, respectively. We employed K-mer-based statistics to predict genome size, and it was estimated to be 454.78 Mb. The assembled genome is 382.64 Mb with a contig N50 of 15.36 Mb (Table 1), significantly longer than that in R. chinensis (contig N50 = 3.4 Mb)12 or woodland strawberry Fragaria vesca (contig N50 = 7.9 Mb)15. The GC content of the R. rugosa genome was 39.30% (Table 1), which was very similar to that of F. vesca (38.98%) and R. chinensis (38.84%). To assemble the contigs into chromosomes, we applied Hi-C sequencing technology and anchored 98.21% of the sequences onto 7 chromosomes (Fig. 1, Supplementary Table 1). Based on this high-quality genome assembly, we evaluated the genome completeness of R. rugosa using BUCSO with the embryophyte_odb10 database. The genome assembly completeness reached 93.2%, and the gene prediction completeness reached 94.4%. We further compared the completeness of R. rugosa with the released Rosaceae genomes of R. chinensis, strawberry (F. vesca), peach (P. persica), apple (M. domestica) and pear (P. bretschneideri). Their proportions were similar to those of R. rugosa (Supplementary Table 2), indicating the high quality of our genome assembly.

Table 1 Statistics of the R. rugosa genome assembly and annotation
Fig. 1: The Hi-C interaction heatmap of R. rugosa showing that the contigs were assembled into 7 chromosomes.
figure 1

A The fruit, flower, and leaf of R. rugosa. The tender leaf was sampled for genome sequencing. B The heatmap showed the 7 assembled chromosomes

Genome components

The R. rugosa genome was composed of 50.27% repetitive sequences (Table 2). Most of these repetitive sequences are long terminal repeats (LTRs), including Gypsy and Copia, accounting for 26.75% of the total genome. The proportion of LTRs in R. rugosa was much greater than that in Fragaria spp. such as F. vesca (~16%)16 and F. nilgerrensis (16.5%)17 but slightly less than that in R. chinensis (28.3%)12, suggesting the rapid evolution of LTRs in Rosaceae plants. The R. rugosa genome encodes 39,704 protein-coding genes, close to the number in R. chinensis12. Moreover, we annotated 37.32%, 87.58%, and 23.03% of genes using the Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG) and Clusters of Orthologous Groups (COG) databases (Supplementary Figs. 1, 2, 3). We mapped the genes and repetitive elements to the 7 chromosomes (Fig. 2).

Table 2 Repeat sequences in the R. rugosa genome
Fig. 2: The high-quality genome assembly of R. rugosa allows the visualization of genomic details.
figure 2

A The 7 chromosomes. B Gene density (window size = 50 kb), C LTR density (window size = 50 kb). D GC content distribution (windows size = 50 kb). E Synteny blocks in R. rugosa

Evolution of the R. rugosa genome

To study the evolution of the R. rugosa genome, we constructed a species tree of R. rugosa and representative Rosaceae species using phylogenomics. We obtained 321 high-confidence single-copy nuclear genes across 8 eudicot species. R. rugosa is closely related to R. chinensis, diverging ~5.26 million years ago (Fig. 3). Although they are close relatives in the Rosa genus, the gene orthogroups differ greatly in these two species, gaining 5418 and 1764 in R. chinensis and R. rugosa, respectively, and losing 2404 and 4676 in R. chinensis and R. rugosa, respectively. The genus Rosa could be divided into two groups: group I: Pimpinellifoliae+Rosa+Carolinae and group II: Gallicanae+Synstylae+Chinenses+Laevigatae+Caninae+Banksianae+Microphyllae+Bracteatae. This significant orthogroup difference may be because R. rugosa belongs to Group I and R. chinensis belongs to Group II18.

Fig. 3: A phylogenomic species tree of R. rugosa and eight other representative species.
figure 3

This tree was constructed using 321 stringent single-copy nuclear protein-coding genes, showing gene family contraction and expansion. The numbers above the branches represent the number of gene families with either expansion (red) or contraction (blue). The numbers at the node indicate divergence time and 95% confidence interval. The species used in the tree are Arabidopsis thaliana, Fragaria vesca, Rosa chinensis, Rosa rugosa, Prunus persica, Malus domestica, and Pyrus bretschneideri. The red star indicates the whole-genome duplication in the ancestor of apple (M. domestica) and pear (P. bretschneideri)

We then explored the orthogroup variations between R. rugosa and R. chinensis. We studied both the contracted and expanded orthogroups in these two species (Table 3). We showed that R. rugosa lost several orthogroups, including OG0000650 (aldolase superfamily), OG0000325 (aminotransferase-like), OG0001051 (IBR domain-containing), OG0000709 (NB-ARC domain-containing disease resistance), and OG0000761 (NB-ARC domain-containing disease resistance), but had more NB-ARC domain-containing disease resistance protein genes than OG0000869.

Table 3 The expansion and contraction of orthogroups between R. rugosa and R. chinensis

The publications of hundreds of angiosperm genomes6 has revealed that polyploidization events have occurred frequently, with at least four waves19, contributing to the genomic materials for innovation20. We calculated the gene Ks values in R. rugosa, R. chinensis, and Vitis vinifera. We found that their shared feature is a single peak at 1.3-1.5 (Fig. 4A–C). We then compared the whole-genome syntenic patterns and still did not find any recent WGD. These results show that the Rosa species experienced only the eudicot-specific WGT event, similar to grapes21. This result is consistent with previous reports for other Rosa species22.

Fig. 4: Rosa rugosa experienced only the core eudicot-specific gamma whole-genome triplication, with no recent polyploidization.
figure 4

A Ks distribution of R. rugosa paralogs. B Ks distribution of R. chinensis paralogs. C Ks distribution of P. persica paralogs. D Cross-species comparison using several eudicots, including grape (V. vinifera), R. rugosa, R. chinensis, strawberry (F. vesca), and peach (P. persica). The red lines indicate genomic shuffling across these Rosaceae species using grapes as an outgroup

We compared the syntenic patterns of R. rugosa with those of other representative species (Fig. 4D). We showed that R. rugosa has very conserved syntenic relationships with grape. For example, VvChr1 and VvCh5 correspond to RrChr1, VvChr9, and VvChr11, and half of VvChr14 matches RrChr6 (Fig. 4D, Supplementary Fig. 4). In the genomes of R. rugosa and R. chinensis, every chromosome matched each other well. However, when compared with two other Rosaceae species, namely, peach (P. persica) and wild strawberries (F. vesca), we found that a large segment composed of 10.44 Mb of chromosome was reversed in R. chinensis but in the exact same order in other species (Fig. 4D, Supplementary Fig. 5). In addition, we found that a segment 1.56 Mb in length was translocated in F. vesca. These results suggest that genomes within the Rosa genus are very conserved in terms of synteny and that small genetic changes could contribute to morphological variations.

Since we did not find significant expansion or loss of genes related to the salt stress response or water stress in R. rugosa compared to R. chinensis (Table 3), we then investigated the contribution of WGT to R. rugosa.

We studied the WGT and its contribution to floral evolution in R. rugosa and R. chinensis. R. rugosa has large, pink, and fragrant flowers. We analyzed the genes retained after WGT to determine whether floral genes could have expanded after this ancient polyploidization event. Gene Ontology annotation of all R. rugosa protein-coding genes showed that 146 genes, compared to 67 genes in R. chinensis, were involved in floral organ development (Fig. 5A, Supplementary Figs. 6, 7), suggesting that R. rugosa retained many more genes for floral-related traits. Floral organ development was divided into four categories, including floral organ development, floral whorl development, floral organ morphogenesis and floral organ formation, according to the agriGO analyses. Among them, 34 genes, including kinase proteins (LRR kinases) and transcription factors (KNOX/ELK, MYB, zinc finger and MADS-box), were involved in all four aspects in R. rugosa (Fig. 5B). However, only 13 genes were involved in all four aspects in R. chinensis (Supplementary Table 3). Then, we compared the floral organ determination genes and the MADS-box genes in R. rugosa, R. chinensis, and A. thaliana. We found a total of 92 MADS-box genes in R. rugosa, slightly more than that in R. chinensis (84 MADS-box genes) (Supplementary Fig. 8). The S-locus of R. rugosa was investigated for the first time and compared with other Rosaceae species (Supplementary Fig. 9). The results showed that there were 19 F-box genes and one S-RNase gene in R. rugosa. Unlike Prunus spp., R. rugosa’s S-locus size was similar to that in Maleae spp., suggesting that the self-incompatibility recognition mechanism was closer to or belonged to the multifactor recognition model.

Fig. 5: The floral developmental genes were retained after gamma WGT in R. rugosa.
figure 5

A The Venn diagram shows the distribution of genes involved in floral organ development, floral whorl development, floral organ morphogenesis, and floral organ formation from R. chinensis and R. rugosa. B Annotation of the 34 genes involved in four aspects of floral development in R. rugosa identified a series of kinase and transcription factor genes

R. rugosa plants are economically important partly due to the high essential oil production of their flowers. Monoterpenes are the main constituents of essential oils, accounting for 50–70% of the total content23,24. Due to the lack of genome sequences, only a fraction of genes could be identified using transcriptomes or comparative genomic studies24. Here, a total of 53 terpene synthases (TPSs), which are key genes responsible for terpene biosynthesis, were identified from the genome of R. rugosa (Supplementary Fig. 10). The RrTPSs were distributed into five subfamilies (TPS-a, b, c, g and e/f) based on clustering with TPS identified from model angiosperm species25. Eighteen and four RrTPS genes were found to belong to the TPS-g and TPS-b subfamilies, respectively. Because TPS-g and TPS-b are mainly involved in monoterpene biosynthesis, these 22 RrTPS genes are the main candidates responsible for the high-level production of monoterpenes in essential oil. Twenty-six RrTPS genes were identified to be members of the TPS-a subfamily with putative sesquiterpene synthase functions. In addition, the TPS family in R. rugosa contains two members in the TPS-c subfamily and 3 members in the TPS-e/f subfamily. Further phylogenetic analysis indicated that each RrTPS gene, a member of TPS-g, has corresponding orthologs in the genome of R. chinensis (Supplementary Fig. 10), suggesting a close evolutionary relationship between the two TPS families from R. rugosa and R. chinensis.

R. rugosa can adapt to drought, salinity, and alkaline soils and can even become invasive in some places3. However, R. chinensis does not have such abilities. By pathway enrichment of all R. rugosa genes (Supplementary Fig. 2), we showed that 850 genes in R. rugosa were involved in environmental adaptation. To trace the origin and evolution of these stress-related genes, we found that two pathways, salt stress and water stress (water deprivation or drought), were significantly retained and enriched after WGT (Fig. 6Afor R. rugosa, Supplementary Fig. 11for R. chinensis). In each module of R. rugosa, the number of genes was significantly higher than that in R. chinensis. Furthermore, we constructed a Venn diagram (Fig. 6B, C) to show the genes that might be involved in cross talk related to these abiotic stresses. Eventually, we found 11 and 7 genes in R. rugosa and R. chinensis that were predicted to be involved in these four abiotic stress responses, respectively (Supplementary Table 4, Supplementary Table 5). Notably, among these module genes, we found that two paralogs of DREB2A-INTERACTING PROTEIN 2 (DRIP2) in R. rugosa had been subjected to positive selection pressure (Fig. 6D, Supplementary Table 6). Furthermore, we found two drought/water stress-related DRIP2 genes in R. rugosa but only one in R. chinensis or Arabidopsis, with potential gene neofunctions in R. rugosa’s adaptation to stressful environments. Meanwhile, we found that the number of PTR3 genes, which encode dipeptide and tripeptide transporters involved in responses to high NaCl concentrations, expanded to 7 in R. rugosa but only four in R. chinensis, 5 in F. vesca, 3 in P. persica and 3 in A. thaliana. Two PTR3 (PEPTIDE TRANSPORTER 3) genes under positive selection pressure were detected (Fig. 6E, Supplementary Table 6). Therefore, these genes might provide R. rugosa with its unique ability to adapt to high salinity environments and water stresses.

Fig. 6: Abiotic stress-related genes were enriched in Rosa rugosa.
figure 6

A The agriGO modules of salt stress- and water stress-related genes predicted using Arabidopsis orthologs of R. rugosa duplicated genes after WGT. B, C Venn clustering of 193 and 103 abiotic stress-related genes from R. rugosa and R. chinensis, respectively. D, E The DRIP2 genes have two paralogs in R. rugosa but one in R. chinensis and Arabidopsis. The PTR3 genes have 7 paralogs in R. rugosa but 4 in R. chinensis. The DRIP2 paralogs and two PTR3 paralogs in R. rugosa have been subjected to strong positive selection pressure

Finally, as shown in Fig. 7, we constructed the salt stress response pathway of R. rugosa. Meanwhile, we compared the differences in the number of genes between A. thaliana, R. rugosa and R. chinensis (Supplementary Table 7). There was no difference in the number of genes among these sampled species, suggesting that R. rugosa did not cope with salt stress using gene dosage, but rather using transcription-, translation-, or metabolome-level regulation.

Fig. 7: The deduced salt stress response signaling pathway in R. rugosa.
figure 7

The related R. rugosa genes were listed in rectangular boxes


As a popular ornamental plant, R. rugosa is widely cultivated. The flowers of R. rugosa have been utilized for essential oil production and dried to produce flower tea. The economic value of this plant will certainly grow if molecular breeding accelerates the production of novel cultivars with optimized essential oil content and improved floral traits. A high-quality reference genome will provide a map for the identification of genes responsible for key agronomic traits and provide insights into how rugosa rose evolved during its long evolutionary history. This study provides for the first time the valuable resource of a R. rugosa genome for the rose research community. Through analysis of the genome sequence of R. rugosa and comparative genomic analyses, we provide novel insights into the biology, ecology and evolution of R. rugosa from three main perspectives. From the perspective of structural genomics, we show a large reversed segment in R. chinensis and a translocation in strawberry. From the perspective of floral biology, we found that more MADS-box genes were retained in R. rugosa than in R. chinensis, suggesting their potential roles in floral development in R. rugosa. From the perspective of stress biology, a number of stress-related genes were found to be specifically expanded and retained in R. rugosa, potentially contributing to its adaptation to stressful environments.

Materials and methods

Plant samples and DNA/RNA extraction

The R. rugosa plants were sampled from Nanjing Agricultural University. For genome sequencing, we collected mature healthy R. rugosa leaves. For transcriptome sequencing, the roots, stems, and leaves of R. rugosa were obtained. All samples were quickly frozen in liquid nitrogen and stored in a −80 °C freezer. We used a QIAGEN® Genomic DNA extraction kit (Cat#13323, QIAGEN) to extract genomic DNA according to the standard operating procedure provided by the manufacturer. We isolated total RNA for RNA sequencing by TRIzol reagent according to the manufacturer’s instructions.

Sequencing and library construction

We used a total of 15 µg genomic DNA to construct a SMRTbell target size library for PacBio-HiFi sequencing according to a standard protocol. We sheared genomic DNA to the expected size of fragments for sequencing on a PacBio Sequel II instrument with Sequencing Primer V2 and Sequel II Binding Kit 2.0 in Grandomics. To construct the Hi-C library, we digested cross-linked chromatin into units with Dpn II, marked by incubation with biotin-14-dCTP and ligated the units by biotinylation. Finally, the ligated genomic DNA was sheared to 100 bp by StLFT technology and sequenced using the DIPSEQ platform at BGI, China. One microgram of sample RNA was used to build an RNA library with a TruSeq RNA Library Preparation Kit (Illumina, USA) following the manufacturer’s recommendations.

Genome assembly and quality evaluation

Approximately 59.2 Gb of raw HiFi sequencing reads was obtained from the rosa DNA library. We first used HiCanu v2.2.126 for preliminary assembly of the rosa genome. Then, Redundans v 0.14a27 was performed to remove the redundant sequences. A total of 150.6 Gb of Hi-C data were obtained to anchor the contig onto the chromosome. We aligned Hi-C reads to assembly by BWA v 0.7.17-r118828. Next, the draft assembly genome was scaffolded with Hi-C reads by 3D-DNA v18011429. Then, Juicer was used to filter the sequence and cluster it, and the Juicerbox tool30 was applied to manually adjust chromosome construction. We finally anchored the scaffolds on seven chromosomes. In addition, the BUSCO v3.0.231 pipeline was used to assess the completeness and accuracy of the R. rugosa genome with the embryophyte_odb10 dataset, which contains 1614 BUSCO gene sets.

Genome annotation

To annotate the repeat sequence in R. rugosa, RepeatModeler v2.0.132 and RepeatMasker v4.1.033 were searched using Repbase TE libarary (v2018.10.26) from the Repbase server ( To predict the protein-coding gene R. rugosa, we combined de novo gene prediction, homology-based prediction and RNA-seq-based prediction. SNAP v2006.07.2835 and AUGUSTUS v3.3.336 were used for de novo prediction with the parameter file trained on F. vesca, M. domestica, P. persica, P. bretschneideri, R. chinensis and R. occidentalis. For homology-based and RNA-seq-based gene identification, F. vesca, M. domestica, P. persica, P. bretschneideri, R. chinensis and R. occidentalis genomes were searched. Then, we mapped RNA-seq data to the genome by Hisat2 v2.2.137 and obtained gene models with SAMtools v1.7.138. These transcripts and the genes from the six homologous species were analyzed with GeMoMa v1.6.4 software to identify protein-coding genes39. Finally, we merged the gene models with EVidenceModeler V1.1.140 from SNAP v2006.07.28, AUGUSTUS V3.3.3 and GeMoMa v1.6.4. We annotated the COG/KOG41, Gene Ontology42 and KEGG pathways43 of rosa protein sequences on the eggNOG-mapper online website ( and used HMMER v3.3.144 with the Pfam database45 to identify the functions of all proteins.

Construction of phylogenetic trees and estimation of divergence times

We used OrthoFinder v2.4.046 to generate clusters of gene families from rugged rose (R. rugosa), Arabidopsis (A. thaliana), strawberry (F. vesca), M. domestica, P. persica, P. bretschneideri, R. chinensis and V. vinifera. We aligned the single-copy proteins generated from OrthoFinder v2.4.046 by MUSCLE v3.8.155147. Based on the single-copy nuclear genes from the MUSCLE results, we used RAxML v8.2.1248 and ASTRAL-II v5.7.349 to construct the phylogenetic tree with the maximum-likelihood method. Then, we used the MCMCTree pipeline of the PAML v4.950 program to calculate the divergence times of the eight species. We marked the split times of Rosids and Rosaceae that were downloaded from the TimeTree website (

Gene family expansion and contraction

Based on the gene family and gene count statistics of OrthoFinder v2.4.0, we used CAFÉ v4.2.151 to determine the expansion and contraction gene families of R. rugosa, A. thaliana, F. vesca, M. domestica, P. persica, P. bretschneideri, R. chinensis and V. vinifera with a p value < 0.01.

Synteny and WGD

To find the synteny blocks between R. rugosa, R. chinensis and V. vinifera, the python version of MCScan (JCVI v1.1.7)52 was applied to compare proteins to proteins. We set 30 genes as the minimum in a syntenic region. Furthermore, we constructed a Circos map by Circos v0.5253. To analyze whole-genome duplications in rosa, we calculated and mapped the Ks values and distribution by wgd v1.1.054.

Genes under positive selection

To analyze positively selected genes, we chose F. vesca, M. domestica, P. persica, P. bretschneideri and R. chinensis to identify orthologs by WGD. ParaAT v2.055 and KaKs_Calculator v2.056 were used to detect the genes under positive selection. Next, we used BLASTP to search for homologous genes between R. rugosa and A. thaliana. AgriGO v2.057 was used to annotate the GO, and we drew a Venn diagram on an online website (