Whole-genome duplication (WGD) or polyploidization has been suggested to be prevalent throughout the evolutionary history of plants, acting as an important evolutionary force promoting diversification, speciation, and adaptation to new environments1,2,3,4. Neofunctionalization or subfunctionalization of duplicated genes post-WGD, as well as species-specific gene retention and loss, lead to expansion and contraction of gene families, resulting in the generation of morphological and adaptive diversity5,6,7. However, the process of species differentiation after ancient WGD is often accompanied by the occurrence of genome rearrangement events, resulting in the generation of new genes or changes in the expression and regulation mode of existing genes, which may result in the emergence of new traits and promote speciation8,9. Based on the origin and level of ploidy, polyploidy can be divided into two types, autopolyploidy and allopolyploidy. The former involves duplication of the diploid genome in a single species; whereas the latter is formed by a genomic combination between two or more distant species3,10. Although the relative proportion of the two types are comparable, allopolyploids are generally expected to have higher evolutionary advantages due to the pairing of chromosomes from each parent, which often results in more chromosomal structural variations, novel gene interactions, and morphological innovations3,11. In general, polyploid plants exhibit greater vigor, stronger resistance, wider environmental adaptability, and higher biomass and economic yield than their diploid relatives8,11,12,13. However, the underlying genomic basis of ecological adaptation and subsequent diversification remain poorly understood in most polyploid species.

Chloridoideae, containing approximately 124 genera and 1602 species, is the fourth largest subfamily of the grass family (Gramineae)14. Most species of this subfamily use the C4 photosynthetic pathway, and the first evolutionary transition from C3 to C4 photosynthesis was found in this subfamily about 32–25 million years ago (Ma)15. Species of this subfamily possess strong tolerance to arid environments14, such as Oropetium thomaeum and C. songorica16,17. With a base chromosome number of x = 10, it is suggested that more than 90% of species within this subfamily are polyploids, including the staple grain crop teff in Ethiopia and many turfgrass and forage species18. Ancient WGD events may have contributed to the stress tolerance, emergence of valuable traits, and diversification of these grasses17,18. Thus, this subfamily may serve as a valuable system for investigating potential impacts of WGD, subsequent subgenome divergence, functional innovations, genome rearrangement, and diversification in polyploid plants. Recent advances in sequencing technologies have facilitated genome assemblies for polyploid species. To date, 16 reference genomes of this subfamily have been reported, 13 of which are polyploid. Five genomes were used for comparative examination of desiccation tolerance and sensitivity19,20. Whereas the rest mainly focused on the genomes themselves and investigated the divergence between subgenomes or the genomic basis of adaptation to extreme environments16,17,21,22. Among these studies, however, few have focused specifically on evolutionary divergence after polyploidization. Thus, a comprehensive understanding of how subsequent speciation and diversification occurs in these polyploid genomes remains unknown9.

We focus on a newly named subtribe, Orininae, belonging to the tribe Cynodonteae, Chloridoideae14,23. This subtribe includes two genera, Orinus (3 species) and Cleistogenes (17 species), which are often difficult to separate morphologically. For example, Hao (1938)24 has treated one Orinus species, O. kokonorica (K.S. Hao) Tzvelev, as a Cleistogenes. However, there are two distinct morphological differences between the two genera. Cleistogenes has hidden cleistogamous spikelets concealed within the upper sheaths, which is not found in Orinus; and Orinus has elongated rhizomes covered with leathery and glossy scales, while Cleistogenes has a cespitose habit or very short rhizomes25. C. songorica is an allotetraploid species (2n = 4x = 40) and possesses a genomic basis for its dimorphic flower differentiation and drought adaptation17. Our karyotyping also revealed a chromosome number of 40 for O. kokonorica (Fig. 1a), indicating that this species is a tetraploid plant. However, there is no reference genome for Orinus, which largely limits our understanding of its evolutionary history and genomic divergence between these two genera. Here, we selected O. kokonorica as a case study and performed comparative genomics between O. kokonorica and C. songorica to investigate genomic basis of adaptive evolution and diversification after polyploidization.

Fig. 1: Characterization of the allotetraploid O. kokonorica genome.
figure 1

a Morphology of O. kokonorica. 1: Plants in native habitat. 2: Chromosome counts (2n = 40) in an O. kokonorica cell. 3: Plant with leaves, roots and rhizomes. 4: Underground root system. 5: Plant grown in greenhouse. b Collinearity within the O. kokonorica genome. Tracks from inside to the outside correspond to (1) gene synteny between chromosomes; (2) gene density; (3) LTR/Gypsy retrotransposons; (4) LTR/Copia retrotransposons; (5) ncRNA density; (6) GC content; (7) Pseudochromosomes of O. kokonorica. c Dotplots showing the Ks values of homologous genes between the Oko (O. kokonorica) and Cso (C. songorica) genomes. d Gene synteny between the Oko, Cso, and Otho (O. thomaeum) genomes.

Orinus kokonorica is mainly distributed in the northeastern Qinghai-Tibet Plateau (QTP) at elevations between 2400 and 4200 m26, whereas C. songorica widely occurs in semi-arid and desert areas in central Asia at relatively low altitude25. Both species are important forage resources with good palatability and high nutritious value for local livestock in arid regions. The two species display differences mainly in flower and rhizome morphology and in altitudinal distributions. Whereas dimorphic flower in C. songorica may have contributed to its survival and reproductive success under drought conditions17, the elongated rhizomes in O. kokonorica may store and allocate nutrients for the plant growth and protect dormant buds underground for overwintering in the QTP. In addition, with elongated rhizomes, the species has developed a strong underground root system (Fig. 1a), which may not only contribute to its high drought tolerance, but also has high ecological significance for sand fixation and water conservation in alpine arid regions25. In this study, we first performed genome assembly and RNA-seq-based transcriptomics under different stress treatments for O. kokonorica. Then, using publicly available genomic and transcriptomic data for C. songorica, we conducted comparative genomic and transcriptomic analyses to elucidate the evolutionary history of the two species and the genomic basis of their subsequent divergence.


Genome sequencing, size estimation, and assembly

The genome size of O. kokonorica was estimated to be ~520 Mb by K-mer analysis based on 35.29 Gb of cleaned Illumina data (Supplementary Fig. 1a, Supplementary Table 1). A combination of Illumina, Nanopore, and Hi-C technologies were adopted for sequencing to accurately assemble its genome. Based on 61.87 Gb of Nanopore long reads corresponding to 110× coverage of the estimated ~520 Mb genome (Supplementary Table 1), we polished the raw assembled genome using NextPolish and performed deredundancy with purge_haplotigs, resulting in a final genome assembly with length of 556 Mb and contig N50 of 9.08 Mb (Supplementary Table 2). The Benchmarking Universal Single-Copy Orthologs (BUSCO) evaluation score was 97.6%, indicating a very complete and high-quality genome assembly (Table 1). Based on ~137 Gb of Hi-C data, we further connected 127 contigs onto 20 pseudochromosomes. In total, 99.80% (554.86 Mb) of the assembly was anchored and oriented on 20 pseudochromosomes (Supplementary Fig. 2, Supplementary Table 3). 91.37% of Illumina short reads could be properly paired mapped to the final genome assembly (Supplementary Table 4). These assessments indicate that the genome of O. kokonorica was assembled with high quality, completeness, and accuracy.

Table 1 Assembly and annotation summary of O. kokonorica.

Genome annotation and gene prediction

A total of ~327 Mb (58.81%) of the O. kokonorica genome assembly was identified as repetitive elements. The vast majority of repeats were classified as LTR retrotransposons, accounting for 36.02% of the genome, including approximately 28.71% Gypsy and 7.30% Copia retrotransposons (Supplementary Table 5), a higher percentage than that in C. songorica (26.54%). Analysis of dynamic evolution of LTRs indicated that LTRs of O. kokonorica were younger than those in C. songorica, with the former experienced a more recent expansion with a peak of 0.8 Ma (Supplementary Fig. 3). DNA transposons, long interspersed nuclear elements (LINEs) and short interspersed nuclear elements (SINEs) accounted for 8.92%, 3.56%, and 0.22%, respectively, of the genome assembly (Supplementary Table 5).

In total, a high-confidence set of 48,598 protein-coding genes was predicted using a combination of de novo, homology-based, and transcriptome-based approaches, and 48,521 (99.84%) were anchored into 20 pseudochromosomes (Table 1; Supplementary Table 6). With similar features of other Gramineae species, protein-coding genes in O. kokonorica were 4327 bp long and covered 6.18 exons on average. The lengths of exons and introns were highly conserved in all five investigated plant genomes (Supplementary Fig. 4, Supplementary Table 7), which further illustrated the reliability of the annotation results. In addition, BUSCO analysis of the protein set showed that the annotated genome contained 90.60% BUSCOs (Supplementary Table 8), suggesting good annotation completeness of protein-coding genes. Approximately 90.59% of O. kokonorica genes could be annotated by non-redundant nucleotides and proteins in the SWISS-PROT Protein Sequence Database, Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), Clusters of orthologous groups for eukaryotic complete genomes (KOG) and Non-Redundant Protein Sequence Database (NR) (Supplementary Table 9). We also identified 231 microRNA (miRNA), 1012 small nuclear (snRNA) genes, 903 transfer RNA (tRNA), 183 ribosomal RNA (rRNA) in the genome sequence (Supplementary Table 10). Characterization and features of the O. kokonorica genome are exhibited in Fig. 1b. The LTRs were mainly distributed across the pericentric regions, while genes were mainly enriched in the more distal chromosomal regions.

Allotetraploid origin of O. kokonorica

The karyotyping (2n = 4x = 40, Fig. 1a), the K-mer analysis in which two peaks were present with the smaller peak at the doubled multiplicity of the major peak (Supplementary Fig. 1), and the chromosome-scale pairwise syntenic relationships within O. kokonorica genome (Fig. 1b, Supplementary Fig. 5) indicated that O. kokonorica was a tetraploid. To determine whether the species is an allotetraploidy or autotetraploidy, we first used Smudgeplot27 to visualize the haplotype structure and to estimate ploidy of the genome. The result indicated an obvious peak of AABB type of K-mers (Supplementary Fig. 1b), implying the allotetraploid origin of O. kokonorica. This is further supported by the genome synteny and synonymous substitutions per synonymous site (Ks) values between O. kokonorica and two C. songorica subgenomes (Fig. 1c). Ks is assumed to be neutral, if O. kokonorica is an autotetraploidy, we would expect that the Ks values between the two duplicated chromosomes of O. kokonorica and one of their two syntenic C. songorica chromosomes are similar. However, our result showed that each C. songorica chromosome corresponded with a pair of O. kokonorica pseudochromosomes with apparently different Ks values, supporting the allotetraploid origin of O. kokonorica.

Based on genome synteny and Ks values between O. kokonorica and the two C. songorica subgenomes, we divided the O. kokonorica pseudochromosomes into A and B subgenomes. It should be noted that the two subgenomes of C. songorica are simply determined based on the phylogenetic distance between C. songorica and Oryza sativa17, which should be treated with caution. Nonetheless, this will not affect our comparative genomic analyses because aligning the chromosomes of the two species is prerequisite for such analyses. We found that high collinearity among the two genomes by visualizing the synteny blocks. For subgenomes A of the two species, each pair of chromosomes corresponded very well, while for subgenomes B, two pairs of large interchromosomal rearrangements were revealed between chromosomes B2 and B5, and between B1 and B8. When using O. thomaeum, a diploid close relative to the two allotetraploid species, as a reference, we found no interchromosomal rearrangement events between O. kokonorica and O. thomaeum (Fig. 1d), indicating that these large interchromosomal rearrangements occurred specifically in C. songorica.

Comparative genomic and phylogenetic analysis

A comparative genomic analysis was performed among 16 plant genomes, including 12 Gramineae species, 1 Bromeliaceae species, 1 Musaceae species, 1 Orchidaceae species, and 1 Brassicaceae species. For the three allotetraploid species (O. kokonorica, C. songorica, and Eragrostis tef), their subgenomes were used. In total, 34,019 O. kokonorica genes (70.00%) were clustered into 25,903 gene families, of which 8961 families were shared with the other four Gramineae species (O. sativa, C. songorica, Zea mays, O. thomaeum) and 1387 were unique to the O. kokonorica genome (Supplementary Fig. 6a, Supplementary Table 11). These unique gene families were predicted to have functions involved in ‘regulation of salicylic acid metabolic process’ (GO:0010337), ‘amino acid transport’ (GO:0006865), and ‘male-female gamete recognition during double fertilization forming a zygote and endosperm’ (GO:0080173) (Supplementary Fig. 6b).

We identified 495 expanded and 6400 contracted gene families in the A subgenome of O. kokonorica compared to the other plant species, and the corresponding numbers in the B subgenome was 549 and 6556, respectively. However, compared with O. kokonorica, the number of expanded gene families in C. songorica was larger (1234 in A and 1350 in B subgenomes), whereas the number of contracted families was much lower (1645 in A and 1491 in B subgenomes, Fig. 2a, Supplementary Table 12). GO analysis revealed that the expanded orthogroups of O. kokonorica were significantly enriched in the functional terms ‘programmed cell death involved in cell development’, ‘jasmonic acid biosynthetic process’, and ‘regulation of double-strand break repair’ (Supplementary Fig. 7, Supplementary Data 2). KEGG analysis of these expanded gene families revealed significant enrichment in the “DNA repair and recombination proteins pathway”, “flavonoid biosynthesis pathway”, and “environmental adaptation pathway” (Supplementary Fig. 8, Supplementary Table 13).

Fig. 2: Evolution of the O. kokonorica genome.
figure 2

a Phylogenetic relationships and divergence times among 16 species. The numbers at nodes represent the divergence times. Gene family expansion (+) and contraction (−) are indicated by red and blue, respectively. b Distribution of average synonymous substitution levels (Ks) between syntenic blocks. c Two possible scenarios of species formation for O. kokonorica and C. songorica, with the number in the lower left corner of each tree representing the number of gene trees that support the topology (Os: O. sativa; Oko: O. kokonorica; Cso: C. songorica; Otho: O. thomaeum). d Classification of gene duplicate origins in the Oko genome. e Gene Ontology (GO) enrichment analysis of the tandem duplicated genes in Oko. Circle color represents -log(FDR) (false discovery rate) in the hypergeometric test corrected using BH (Benjamini and Hochberg) method. Circle size represents the gene count of the GO terms.

The same 16 plant species were used to infer the phylogeny of O. kokonorica based on 381 single-copy orthologous genes. Subgenomes of the three allotetraploid species were used for the phylogenetic reconstruction and estimation of divergence times. As expected, O. kokonorica displayed the closest relationship with C. songorica. The diploid progenitors of subgenomes A and B (now likely extinct) of O. kokonorica and C. songorica diverged from their common ancestor ~18.19 Ma (12.99–24.31 Ma, Fig. 2a). The ancestor of O. kokonorica and C. songorica diverged from O. thomaeum ~ 20.86 Ma (15.21–27.33 Ma), which together diverged from E. tef ~ 26.19 Ma (19.82–31.33 Ma, Fig. 2a). The phylogenetic relationships among these 16 species were the same as those recovered from previous studies22.

The divergence times estimated by phylogenetic and Ks analyses between subgenomes of O. kokonorica and C. songorica suggested that the two species (or the two genera) may have shared one paleo-allotetraploid event. The A and B progenitors were estimated to diverge from each other ~18.19 Ma, corresponding to the overlapped Ks peaks between A-B subgenomes of the two species (OkoA–CsoB, OkoB–CsoA, Fig. 2b, Supplementary Data 1). The Ks analysis also showed that Ks peaks of OkoA–CsoA, OkoB–CsoB, and Oko–Cso overlapped, which indicated that the OrinusCleistogenes divergence, A subgenomes divergence, and B subgenomes divergence occurred at the same time, i.e., ~10 Ma (Fig. 2a). We further performed phylogenetic analysis to test this shared allotetraploidy event. Using the O. sativa genome as a reference, colinear genes of O. thomaeum, O. kokonorica, and C. songorica were extracted to build the gene trees. We found that most collinear gene trees (74.8%, Fig. 2c) well supported the shared allotetraploidy event between O. kokonorica and C. songorica. In other words, the two genera diverged from the same allotetraploidy ancestor, and the allotetraploidy event occurred between 18.19–10 Ma.

Tandem duplication contributed to adaptation to harsh plateau environment

Although Orinus and Cleistogenes shared the same paleo-allotetraploidy event, O. kokonorica has experienced severe gene loss (large number of contracted gene families) compared with C. songorica during its adaptation to the high plateau. We found only 47.82% (23,240) of gene duplications resulted from WGD/segmental duplication. There was a considerable number of duplicated genes generated from dispersed duplication (22.37%, 10,873), proximal duplication (5.10%, 2477) and tandem duplication (4.10%, 1993) (Fig. 2d). Tandemly arrayed genes are thought to be volatile after gene duplication, and therefore the retained tandemly genes may indicate functional importance28. Therefore, we investigated GO enrichment of the tandemly duplicated genes and their expressions under three stress treatments (cold, heat, and drought; see details in the Methods). We found that they were enriched in some GO terms related to adaptation to the plateau environment, such as ‘cellular response to oxygen levels’, and ‘leaf senescence’ (species at higher altitudes have longer leaf life than those at lower altitudes)29 (Fig. 2e). Furthermore, 755 (37.9%) out of 1993 tandemly duplicated genes were differentially expressed in at least one tissue or treatment, and 236 (26.2%) out of the total 902 tandem gene clusters had at least two differentially expressed copies, suggesting the importance of tandem duplicated genes in adaptation to harsh environmental conditions.

Genomic fractionation bias and subgenome dominance in O. kokonorica

Duplicated genes following polyploidization can be retained through subfunctionalization/neofunctionalization or reverted to a single copy through genome fractionation. We investigated genome fractionation in O. kokonorica and found that the two subgenomes of O. kokonorica retained a similar number of genes across the chromosomes, showing no biased fractionation (Supplementary Fig. 13).

Unequal expression of homoeologous genes in allopolyploids can be an important feature and consequence of polyploidization. We investigated homoeolog expression bias in five different tissues (mature leaf, young leaf, root, rhizome, and rhizome tip) and under three stress treatments (cold, heat, and drought; see details in the Methods). Of 11,861 identified homoeologous gene pairs between subgenomes A and B, 11,409 had homoeologous expression bias (HEB) in at least one tissue or treatment, and 1160 had biased expression in all sampled tissues and treatments. Pairwise comparisons between syntenic gene pairs showed a slight bias in expression toward the B subgenome across all five tissues and three treatments (Supplementary Fig. 10). Roughly 26.6% (3155) of the 11,861 pairwise comparisons across the five tissues and three stress treatments showed biased expression toward homoeologs in the B subgenome. Each tissue and treatment have ~6000 homoeologous gene pairs with differential expression, including 52.16% biased toward the B subgenome (Supplementary Fig. 10).

Genome-wide variation and differentiation in homoeolog expression patterns between O. kokonorica and C. songorica

Genome-wide variation, including SNPs, indels, duplications, and structural variations (SVs), was identified based on whole-genome alignment with O. kokonorica and C. songorica genome assemblies. A total of 9,522,801 SNPs, 44,618 indels, 2768 duplications, and 925 inversions were identified using NUCDIFF (Fig. 3a). Based on Assemblytics, we identified a total of 41,559 SVs, in which presence/absence variants (PAVs) accounted for 23.78%. Most SVs were detected in non-coding regions, while 26.57% of the SVs were present in exon regions (Supplementary Fig. 9), which could have affected gene function and led to divergence between the two genera. A total of 14,604 genes were categorized as SV-high-impact genes (i.e., the SV is assumed to have a high or disruptive impact on the protein by causing protein truncation, causing loss of function, or triggering nonsense-mediated decay), which were mainly enriched in ‘flower development’, ‘post-embryonic development’, and ‘regulation of gene expression, epigenetic’, and involved in ‘mismatch repair’, ‘cytochrome P450’, and ‘homologous recombination’ pathways (Supplementary Fig. 9).

Fig. 3: Genetic differentiation between O. kokonorica and C. songorica.
figure 3

a Distribution of genetic variations on chromosomes. From top to bottom are SNP, deletion, insertion, duplication, and inversion. b Summary of expression patterns of the orthologous DEG (differentially expressed genes) pairs in O. kokonorica and C. songorica under cold, drought, and heat treatments. c Expression patterns of the orthologous DEG pairs in O. kokonorica and C. songorica. Tracks from outside to inside correspond to 1) chromosomes; 2) DEGs in root under cold treatment; 3) DEGs in shoot under cold treatment; 4) DEGs in root under heat treatment; 5) DEGs in shoot under heat treatment; 6) DEGs in root under drought treatment; 7) DEGs in shoot under drought treatment; 8) DEGs synteny between chromosomes. For (b–g), blue color indicates downregulation, red color indicates upregulation.

To investigate the differentiation in homoelog expression patterns between O. kokonorica and C. songorica, we first performed transcriptomic analysis under the same stress treatments (i.e., light drought, cold, heat) in both shoots and roots of O. kokonorica as conducted for C. songorica. Hierarchical cluster analysis of transcriptomic data showed clustering of the four repeats for the shoot or root at each treatment, except for one root and shoot sample in the cold treatment group and one shoot sample in control group (Supplementary Fig. 11), which were discarded for downstream analysis. Differentially expressed genes (DEGs) were identified under different treatments by comparison with the control group. A total of 1246 and 2316 DEGs in shoots and roots, respectively, were identified under cold stress with 2-fold changes compared with the control sample. 338 DEGs in shoots and 420 DEGs in roots were identified under heat stress. However, under light drought stress, much lower DEGs (11 in shoots; 631 in roots) were detected compared to the other two treatments (Supplementary Fig. 12). The number of DEGs in roots were about 57 times greater than in shoots, indicating that drought stress in early stages does not considerably affect the shoots of O. kokonorica.

Next, we downloaded transcriptomic data under the same stress treatments for C. songorica and identified DEGs using the same method as described for O. kokonorica. It should be noted that although we have tried to keep the same environmental conditions as those reported in ref. 17, other factors (e.g., artificial factor and equipment performance) may generate effects on gene expressions. To reduce such effects, we only focused on the orthologous genes that were both DEGs in the two species (either in shoots, roots, or both) under each treatment and only examined the differential expression pattern of two orthologous genes in each pair. A total of 533, 99, 350 orthologous DEG pairs were identified under cold, drought, and heat treatments, respectively (Fig. 3b, Supplementary Data 35). We found that 53% (52) of orthologous genes had conserved functions (i.e., with the same differential expression pattern in the same tissue under the same treatment) in the two species under light drought treatment, consistent with the fact that both are drought resistance species. However, only ~2% and ~11% of orthologous genes under cold and heat treatments, respectively, had the same expression patterns in the two species (Fig. 3b, c). Most orthologous genes were differentially expressed either in different tissues or in different ways (i.e., up- vs. down-regulation) or both, which may be related to the distinct altitudinal distributions of the two species.

Genetic basis of differentiation in floral development and rhizome growth between Orinus and Cleistogenes

Floral development is effectively explained by the ABCDE model30, and a recent study has suggested that differential expression of the ABCDE genes may affect the development of the chasmogamous (CHs) and cleistogamous flowers (CLs) in Amphicarpaea edgeworthii31. Another study has revealed that overexpression of CsAP2_9 gene confers an abnormal palea in a spikelet and shows degenerated lodicules in flowers in C. songorica17. Therefore, to gain some insights into why dimorphic flowers fail to develop in Orinus, we focused on the ABCDE genes that had differential expression in CLs compared with that in CHs in C. songorica17 and compared them between C. songorica and O. kokonorica. We first examined and compared the ABCDE model genes (AMGs) in O. sativa, O. kokonorica, and C. songorica. Using 21 AMGs in O. sativa as references, we found that, after the shared allotetraploidy event, many copies were lost in O. kokonorica (31), while C. songorica gained more copies (46) (Fig. 4a). Second, based on transcriptomic data of C. songorica flowers, we found that 43.5% (20) of AMGs showed differential expression (twofold change) in CLs compared with that in CHs (Supplementary Table 14). Three orthologs of these AMGs (MADS13, MASD16, and MASD7) were lost in O. kokonorica, and multiple SVs, such as deletions or insertions upstream, intron, or even exon regions of genes, were detected in the remaining 17 AMGs between the two species (Fig. 4b, c). For example, two deletions were detected in the exon regions of Oko006142 compared with its ortholog (CsA201468) in C. songorica. Another gene Oko019761 gained two MYB-binding domains compared with its ortholog in O. kokonorica. These SVs detected in the orthologous AMG pairs may affect their expression and contribute to the differentiation of floral development in the two genera.

Fig. 4: Structural variations (SVs) of the AMGs between O. kokonorica and C. songorica.
figure 4

a The maximum likelihood phylogenetic tree of the AMGs identified in O. kokonorica, C. songorica, and O. sativa. Red stars represent the 17 AMGs with higher expression (twofold change) in CLs compared with that in CHs in C. songorica. Blue stars represent the 3 AMGs with lower expression (twofold change) in CLs compared with that in CHs in C. songorica. b Examples of the four types of SVs detected in the 20 orthologous AMG pairs identified in (a). (I) SVs in upstream or downstream regions of genes; (II) SVs in gene regions (including both intron and exon); (III) Domain gain or lost; IV) gene lost. c The detailed SVs in the 20 orthologous AMG pairs between the two species. “Type” are the SV types in (b).

Rhizomes are modified stems that grow horizontally underground in various perennial species, a growth habit that is advantageous for vigorous asexual proliferation. O. kokonorica has elongated rhizomes, while C. songorica has a cespitose habit or very short rhizomes. The suppression of leaf blade development is necessary for the underground growth of rhizomes, and BLADE-ON-PETIOLE (BOP) homologs are important in the regulation of rhizome growth32. Based on Arabidopsis. thaliana and O. sativa BOP genes, we identified 7, 6, and 3 BOP genes in O. kokonorica, C. songorica, and O. thomaeum, respectively, and found that each BOP in O. thomaeum had synteny to two in the former two species, except one extra tandem copy in O. kokonorica (Oko044410 and Oko044411, Fig. 5a, b). All 7 BOP genes were highly expressed in rhizomes and rhizome tips of O. kokonorica, especially the two copies of BOP1 (Oko010623 and Oko007168, Fig. 5c), which may promote the stiffness and sharpness of the rhizome tips in this species to further form a strong underground root system. In addition, we also detected extensive SVs in the BOP orthologous pairs between the two species (Supplementary Table 15).

Fig. 5: Characterization of BOP Genes in O. kokonorica.
figure 5

a A phylogenetic tree based on BOP/NPR protein sequences identified in O. kokonorica, C. songorica, O. sativa, and A. thaliana. The NPR genes are shown as outgroup. b Collinear gene blocks between O. kokonorica and C. songorica indicate that the BOP family has one more tandem duplicated gene in O. kokonorica. c The expression of the BOP genes in five tissues of O. kokonorica. Expression level is indicated by transcripts per million (TPM).


In this study, we report the first chromosome-scale genome assembly of O. kokonorica, and the first for the genus Orinus. The karyotyping result together with our comparative genomic analyses confirmed that O. kokonorica is an allotetraploid species, with all chromosomes confidently assigned to two subgenomes. Unlike most polyploid genomes in the grass family, the allotetraploid O. kokonorica has a relatively compact genome (~556 Mb), similar to other tetraploid species from the subfamily Chloridoideae, such as C. songorica (540 Mb)17, E. tef (576 Mb)21, and Leptochloa chinensis (416 Mb)22. The small genome size in these tetraploids is probably inherited from their diploid ancestors, such as that of O. thomaeum, a diploid species closely related to these four tetraploids, whose genome is only 245 Mb in size16. However, we found higher proportions of repetitive elements and a fewer number of genes in O. kokonorica than those in the other three species (Supplementary Table 16). The former is mainly resulted from the rapid expansion of LTR-RTs in O. kokonorica in the very recent past of approximately 0.8 Ma compared with C. songorica (Supplementary Fig. 3). This time frame is congruent with the largest Naynayxungla glaciation in the QTP, which reached its maximum between 0.8 and 0.6 Ma33. The severe environmental conditions during this period may have induced the bursts of TEs34,35, which may have contributed to the increase in the genome size of this species. Nonetheless, the enormous contraction in gene families in O. kokonorica (Fig. 2a) may have counteracted this contribution to genome size. Therefore, the O. kokonorica genome might have been maintained by a reciprocal offset between recent expansion of TEs and contraction in gene families.

Our collinear gene phylogenetic and Ks analyses indicated that the two closely related genera, Orinus and Cleistogenes, may share one paleo-allotetraploidy event between 18.19–10 Ma. This polyploidization event is older than the ones detected in their relatives E. tef (1.1 Ma)21 and L. chinensis (<10.9 Ma)22. Following polyploidization, duplicated genes can be retained through neofunctionalization/subfunctionalization or reverted to a single copy through genome fractionation. Biased genome fractionation may lead to a dominant subgenome, which has often been reported in the genomes of allopolyploids especially in allopalaeopolyploids36, such as in Arabidopsis, bread wheat, maize, and Brassica rapa37,38,39,40. However, we found that the two subgenomes of O. kokonorica have retained similar numbers of genes across the chromosomes (Supplementary Fig. 13) and display no significant global gene expression dominance, which is consistent with that found in C. songorica17. Such non-biased homoelog expression between subgenomes in allopolyploids seems not as rare as previously thought, as this pattern is also observed in other allopolyploid species, including E. tef21, pumpkin41, and E. crus-galli42.

Our analyses of comparative genomics and transcriptomics provide some new insights into the genetic basis of divergence between the two genera after paleo-allotetraploidization. First, we found two pairs of large interchomosomal rearrangements occurred specifically in B subgenome of C. songorica. Although no gene changes were found at the breakpoints and these translocation regions have high collinearity between the O. kokonorica and C. songorica genomes, these large genomic rearrangements may affect chromatin organization as revealed in two ecotypes of Medicago truncatula43, which may result in different epigenetic modifications and gene transcriptional activity in the two species44. Nevertheless, whether these genomic rearrangements are present in all Cleistogenes species and whether these translocations affect chromatin organization needs further investigation. Second, SVs constituted a large proportion of the genomic variation that results in phenotypic variation in organisms45. Using two methods, we built an overview of the genomic landscape of SVs between O. kokonorica and C. songorica, many of which may have contributed to the phenotypic variation and diversification of the two genera. For example, some highly-impacted-by-SV genes were enriched in “flower development.” We also found extensive SVs in some flower development and rhizome growth-related genes and their vicinity (<2 kb), which may play important roles in determining the morphological differentiation of flowers and rhizomes between the two genera (see more detailed discussion below). Finally, while both genera have strong resistance to drought, Orinus occurs in high altitude regions and has adapted to cold environments; Cleistogenes is tolerant to heat in low altitude regions25. Consistent with this, we found most orthologous DEGs (98% for cold treatment and 89% for heat treatment) in the two species under cold and heat treatments exhibited functional divergence by differential expression in different tissues or in different directions. By comparison, more than half of the orthologous DEGs under drought treatment have conserved functions in the two species (Fig. 3b). In addition, tandem copy increases in gene family members may have contributed to the adaptation of O. kokonorica to the plateau environment, as also indicated in many other organisms46,47. Therefore, genomic rearrangements, SVs and functional innovations of orthologous genes may have together played an important role in promoting genomic divergence and speciation between the two genera after polyploidization.

Despite being closely related, Orinus and Cleistogenes have two distinct morphological differences. Cleistogenes has a dimorphic flowering mechanism, while Orinus has elongated rhizomes, both of which ensure reproductive success under harsh environmental conditions48,49,50. We investigated the genetic basis of differentiation in floral development and rhizome growth between the two genera. Using AMGs in O. sativa as references, we found that many duplicated AMG copies have been lost in O. kokonorica, while C. songorica has gained more copies. The increase in copy number of this gene family followed by functional innovations might have played an important role in the dimorphic flower development in Cleistogenes17,51. We further focused on the genes with differential expression in CLs than in CHs in C. songorica. We first found one copy of MADS13, one copy of MASD7 and one copy of MASD16 were lost in O. kokonorica (Fig. 4, Supplementary Table 14). All the three genes are suggested to be involved in determining floral organ identities in O. sativa52,53,54. The differential expression in CLs compared with that in CHs of the three gene copies in C. songorica may contribute to the development of CLs. Furthermore, we detected extensive SVs between the orthologous pairs of these 20 DEGs, especially in the regulatory regions (Fig. 4c). SVs in regulatory regions or introns may result in dosage variations in gene expression, which has been shown to play a crucial role in the variation of plant traits, especially for floral organ identity55,56. Many studies have suggested that changes in expression of a single ABCDE-class gene can easily shift the boundaries between different types of floral organs57,58. Therefore, extensive SVs in the AMGs may have played an important role in the differentiation of flower development between the two species. Similarly, we detected more copies of BOP genes that are all highly or moderately expressed in rhizomes of O. kokonorica and extensive SVs between BOP orthologous genes in O. kokonorica and C. songorica. The copy number variation and SVs of BOP genes may have contributed to the differentiation of rhizome length in the two species. However, our present data are not appropriate to gain detailed mechanistic explanations of the two morphological differentiations. Further studies involving more comparative genomic analyses associated with experimental validation would greatly enhance our understanding of the molecular mechanisms of flower and rhizome development in the two genera.


Plant materials, karyotyping, and estimation of genome size

The plant materials of O. kokonorica were collected from east side of the Qinghai Lake, Qinghai province, China. Bleach-sterilized seeds were germinated in a glasshouse. Root tips with a length of 1–2 cm were harvested from germinated seeds and pretreated in ice-cold water at 4 °C for 20–24 h. Root tips were then fixed in ethanol: glacial acetic acid (3:1, v/v) for 4 h at room temperature. Each root tip was squashed in a drop of 45% acetic acid and observed using a phase contrast microscope (Olympus Bx40). The slides containing metaphase chromosomes were frozen at −80 °C for more than 30 min and cover slips were removed with a razor. Chromosomes were counterstained with 4’,6-diamidino-2-phenylindole. Images were captured with a cooled charge-coupled device camera using a fluorescence microscope (Olympus Bx63).

Fresh leaves of O. kokonorica were used to extract genomic DNA using a DNA Secure Plant Kit (Tiangen Biotech, Co., Ltd., Beijing, China). Paired-end libraries with insert sizes of 150 bp were constructed and sequenced using the Illumina HiSeq X Ten platform. We used K-mer frequency distribution analysis to estimate genome size59. To calculate and plot the K-mer frequency distribution, we used 35.29 Gb of Illumina short reads (Supplementary Table 1) to determine the total number of K-mers of length 27 using Jellyfish v. 0.9.060. The selection of 27-mers to plot the frequency distribution was based on genome characteristics and the pattern of the Poisson distribution (Supplementary Fig. 1).

Genome sequencing and assembly

Total genomic DNA was further fractionated into 10–50 kb fragments with BluePippin to construct libraries following the Nanopore library construction protocol. The libraries were sequenced at the Nextomics Biosciences Company (Wuhan, China) using the GridION X5 sequencer platform (Oxford Nanopore Technologies, UK). Quality-controlled reads were used for de novo assembly using Nextdenovo v. 2.3.051 ( We corrected sequencing errors with NextCorrect, assembled with NextGraph, and polished with NextPolish v. 1.2.42461. At this stage, the short and long reads were used three times for genome correction. Finally, purge_haplotigs62 was used to remove allelic haplotigs, resulting in the final genome assembly. Benchmarking Universal Single-Copy Orthologs (BUSCO) v. 2.02663, with 1614 genes from Embryophyta odb10, was used to evaluate the completeness and accuracy of the assembled genome.

Chromosome-scale assembly with Hi-C data

Hi-C libraries were constructed from ~2 g of fresh leaves following64, which included chromatin extraction and digestion and DNA ligation, purification, and fragmentation. Then, these libraries were submitted for sequencing using the Illumina HiSeq X Ten platform (Illumina, CA, USA). The generated clean Hi-C data were mapped to the draft genome using BWA v. 0.7.175365. The uniquely mapped Hi-C data were retained, clustered, ordered, and placed onto the 20 pseudochromosomes using LACHESIS66. A heat map of the interaction matrix of all pseudochromosomes was plotted with a resolution of 100 kb.

Genome annotation

Repetitive elements, including tandem repeats and transposable elements (TEs), were predicted in the O. kokonorica genome based on a combination of homology-based and de novo approaches. Tandem repeats were annotated using Tandem Repeats Finder v. 4.0967. TEs were identified using RepeatMasker v. 4.1.055 and RepeatProteinMask68 with Repbase v. 22.1169 as the query library. Next, RepeatModeler v. was used to construct a de novo repeat library for the identification of TEs that were not found in the Repbase library. We used LTR_FINDER70 to search the full-length long terminal repeat (LTR) retrotransposons in O. kokonorica, C. songorica. We extracted the 5′ and 3′ LTR sequences based on the LTR_FINDER annotation results, which were subsequently aligned using MUSCLE v. 3.8.3171. The genetic distance (K) between the 5′ and 3′ LTR sequences was calculated using DNADIST, a program within PHYLIP v. 3.69672. The insert time (T) of each LTR-RT was calculated using the formula: T = K/2r, where r is the nucleotide substitution rate estimated by BASEML (see below).

The repeat masked genome was used for predicting subsequent protein-coding genes with a combination of three complementary methods: de novo, homology-based, and transcriptome-based prediction. Augustus v. 3.3.25773, GlimmerHMM v. 3.0.45874 and Genscan75 were used for de novo predictions. GeMoMa v.1.3.16176 was used for homology-based predictions, with protein sequences from A. thaliana, Eragrostis curvula, E. tef, O. thomaeum, O. sativa, Prunus persica, Sorghum bicolor, Triticum aestivum, Z. mays. For transcriptome-based predictions, we first sequenced the RNA libraries generated from five tissues (i.e., root, rhizome, rhizome tip, young leaf, and mature leaf) and assembled the RNA-seq reads into transcripts using Trinity v. 2.1.16277 with default parameters. We also used all the RNA-seq reads to assess genome assembly quality by mapping to the final assembled genome using PASA v. 2.1.06378 Finally, all predictions of gene models yielded by the above approaches were integrated using EVidenceModeler (EVM) v. 1.1.179 to generate a consensus gene set.

The predicted protein-coding genes were functionally annotated by searching against databases. We used Interproscan v. 5.3680, including Gene Ontology (GO) database annotations, protein motifs and domains, functional classifications, protein family identification, transmembrane topology, and predicted signal peptides, to obtain a comprehensive annotation of the predicted protein-coding genes. We used a custom Perl script to get the annotation information. Then, KOBAS ( was used to search the Kyoto Encyclopedia of Genes and Genomes (KEGG) database81 for orthologs. Finally, we used BLASTP to search against the Swiss-Prot82, NR83, and KOG databases with an e-value cutoff of 1e−5. All of the best hits of these database searches were integrated to obtain the final functional annotation result.

Transfer RNAs (tRNAs) were identified by tRNAscan-SE84 with eukaryotic parameters. miRNA and snRNA genes were predicted using INFERNAL85 by searching against the Rfam database86. rRNA genes were predicted by aligning reads to Arabidopsis template rRNA sequences using BLASTN with an e-value of 1e−5.

Polyploidization analyses and subgenome identification

We combined three approaches for polyploidization analyses. First, Smudgeplot was used to visualize the haplotype structure and to estimate ploidy of the genome. Second, protein sequences within and between genomes were aligned using BLASTP with an e-value cutoff of 1e−5. Syntenic blocks were detected based on homologous gene pairs using MCScanX28 and MCScan ([Python-version]). Finally, WGD and speciation events were inferred by calculating Ks values using MCScanX downstream analysis program (Perl). O. kokonorica and C. songorica are both tetraploids belonging to the same subtribe of Chloridoideae, and C. songorica has been revealed to be an allotetraploid species with 20 chromosomes assigned to two subgenomes17. Therefore, we partitioned the O. kokonorica genome into subgenomes A and B based on dotplot with syntenic blocks and synonymous nucleotide substitutions (Ks) values against the C. songorica subgenomes using WGDI87.

We further used the collinear gene phylogenomic analysis to check if WGD events are shared between O. kokonorica and C. songorica. Collinear genes were extracted by WGDI with “-at” parameters and the maximum likelihood trees were inferred using IQ-TREE88 with automatic selection of the best-fit substitution model by (-m MFP) and 1000 ultra-fast bootstrap replicates (-bb 1000). The frequency of gene trees supporting shared WGD event was calculated using ASTRAL89. For example, in two species A and B (assuming each has one WGD event), the collinear genes extracted in the two species were named as A1, A2, B1, and B2. Thus, the topology supporting shared WGD event in A and B is ((A1, B1), (A2, B2)), while the topology of ((A1, A2), (B1, B2)) supporting independent WGD events in A and B. We used the ASTRAL with “-t2” parameters and the specific gene tree ((A1, B1), (A2, B2)) to calculate the number of gene trees that support shared WGD event.

Gene families and phylogenetic analysis

OrthoMCL was used to identify the orthologous groups among 12 Gramineae species (O. kokonorica, C. songorica, O. thomaeum, E. tef, S. bicolor, Z. mays, Digitaria exilis, Panicum hallii, Setaria italica, Brachypodium distachyon, Hordeum vulgare, O. sativa), two Commelinids species (Ananas comosus, Musa acuminata), one Monocot species (Phalaenopsis equestris) and one Rosid species (A. thaliana). For the three tetraploids (i.e., O. kokonorica, C. songorica, and E. tef), both subgenomes were used for orthologous group construction and phylogenetic analysis. The dynamic evolution of gene families in these 16 species was predicted using CAFÉ v. 3.190, and the significantly expanded or contracted gene families were determined based on p-values (p < 0.01). We then completed GO enrichment and KEGG analyses on the expanded gene families in O. kokonorica.

Single-copy orthologous genes from the orthologous clustering results were extracted and aligned using MAFFT v. 7.158b91. Then, Gblocks v. 0.917192 was used to delete regions with poor alignment or large differences after multiple alignments. A maximum likelihood phylogenetic tree was reconstructed based on the single-copy orthologous gene data set using RAXML v. 8.1.1793 with the PROTGAMMAJTTX model and 1000 bootstrap replicates. Nucleotide substitution rate and divergence time were calculated by four-fold degenerate sites (4DTv) of single-copy orthologous genes. The nucleotide substitution rate was estimated using BASEML v. 4.8a, a program within PAML v. 4.094. The program uses a maximum likelihood method with tree file with topology in Newick format and DNA sequence alignment in Phylip format as input files to calculate an average nucleotide substitution rates over the entire tree. Species divergence time was inferred by MCMCtree in the PAML program, based on the calculated substitution rate and known approximate divergence times for A. comosus and M. acuminata (102–120 Ma), and A. thaliana and P. equestris (148–173 Ma) from the TimeTree database (

Identification of tandem duplicated genes

Tandem duplicated gene pairs were identified using duplicate_gene_classifier, a program of MCScanX. Functional classification of Gene Ontology (GO) categories was performed on the tandem duplicated genes using Blast2GO95. Fisher’s exact test was used to calculate the statistical significance of enrichment (p < 0.05) with Benjamini–Hochberg’s false discovery rate adjustment.

Genome fractionation and subgenome dominance

Genome fractionation in O. kokonorica was investigated using homoeologous gene pairs and singletons (whose duplicated copies were lost) in the two subgenomes that also had syntenic orthologs in O. thomaeum. Analysis of homoeolog expression bias was preformed following21 for all five tissues (mature leaf, young leaf, root, rhizome, and rhizome tip) and samples under three abiotic treatments (see below). The five tissues were collected from the same place where plants were collected for genome sequencing.

Genomic basis of divergence between O. kokonorica and C. songorica

In order to investigate the genomic basis of divergence between O. kokonorica and C. songorica, we compared the two species at the whole-genome level and the differentiation in gene expression under diverse abiotic stresses. To detect genomic variation between the two species, we first used NUCDIFF96 to locate and categorize differences between the two species. The results were uploaded to RectChr ( for visualization. We also performed whole-genome comparisons using the nucmer (nucmer -maxmatch -l 100 -c 500) function from the MUMmer package. Structural variations (SVs) were called using Assemblytics97 based on the output of nucmer. A custom Python script98 was used to reformat the results from Assemblytics, and the effects of SVs (including high-, moderate-, low-, and modifier-impacted SVs) were annotated using SnpEff99. The genes categorized as highly-impacted-by-SV were subjected to GO enrichment and KEGG analyses.

To evaluate the differentiation in gene expression under drought, heat and cold treatments between the two species, seeds of O. kokonorica were collected from the same place where plants were collected for genome sequencing. Bleach-sterilized seeds were germinated and grown in a glasshouse under the same controlled conditions as for C. songorica used in ref. 17. Four-week-old seedlings were transplanted into individual pots with the same growth medium. Each pot was irrigated with 100 mL Hoagland nutrient solution every three days. We used the same conditions described in ref. 17 for the heat and cold treatment. Nine-week-old plants were treated under heat and cold (40 °C and 4 °C). For drought treatment, the soil water content was decreased to 10–20%, corresponding to light drought stress in ref. 17. Shoots and roots of each plant were collected 24 h after treatment, immediately immersed in liquid nitrogen, then stored at −80 °C. Four biological replicates were used for each sample for transcriptomic analysis.

Total RNA was extracted using the Tiangen DP411 kit following the manufacturer’s instructions. Messenger RNA (mRNA) was separated from the total RNA by Oligo (dT) and cleaved into short random fragments. Quality cDNA libraries were constructed by PCR enrichment and sequenced by BGI-Shenzhen Company (Wuhan, China) on the MGI2000 platform by 2 × 150 bp pair-end mode. Adapters, reads containing poly-N and lower-quality reads (<Q30) in the raw sequencing outputs were removed to obtain clean reads, which were then mapped to the O. kokonorica genome using HISAT2 v. 2.0.4100. We used StringTie v. 1.3.1101 to calculate transcripts per million (TPM) of mRNA in each sample. TPM of genes was computed by summing the TPMs of transcripts in each gene group, and the DESeq2 R package102 was used to perform differential expression analysis. Differentially expressed genes (DEGs) were defined as those having at least a twofold change in expression compared with those in controlled samples (false discovery rate, FDR < 0.05).

Genomic basis of differentiation in floral development and rhizome growth between Orinus and Cleistogenes

To investigate the genomic basis of differentiation in floral development between O. kokonorica and C. songorica, we identified genes related to floral development (ABCDE model genes) in these two species based on the ABCDE models of A. thaliana and O. sativa. ABCDE genes of A. thaliana and O. sativa were downloaded from TAIR ( and RICEDATA (, respectively. All genes of O. kokonorica and C. songorica were searched against the A. thaliana and O. sativa ABCDE model genes using BLASTP with an e-value cut-off of 1e−5. We then identified domains of candidate genes using HMMER v3.2.1103 by searching the Pfam database104. Phylogenetic analysis was performed based on the ABCDE genes identified in O. kokonorica, C. songorica, and in A. thaliana and O. sativa. Only genes grouped with A. thaliana or O. sativa in the same clade were selected as orthologs.

We searched BOP genes in O. kokonorica, C. songorica, and O. thomaeum using the same strategy mentioned above with A. thaliana and O. sativa BOP genes. Domain confirmation and phylogenetic analyses were performed to identify orthologs.

Statistics and reproducibility

The functional enrichment analysis was performed using the ClusterProfile package, wherein the statistical significance of GO and KEGG terms was evaluated using Fisher’s exact test combined with FDR correction for multiple testing (p < 0.05). For RNA-seq, four biological replicates were used for each tissue and each treatment.