Chromosome-level Streptochaeta genome elucidates the ancestral karyotype and allopolyploid origin of grasses

Abstract

Grasses (Poaceae) are the most economically important group of plants. Reconstructing the ancestral grass genome and studying the evolutionary factors shaping major grass lineages are crucial for understanding the evolution of cereals, including rice, wheat and maize. Here, we report a chromosome-level genome assembly of Streptochaeta spicata, which represents the lineage without typical spikelets and the sister to the rest of Poaceae, and perform comparative genomic analyses across major grass lineages. We confirm the occurrence of the ρ-whole-genome duplication (ρ-WGD) in Streptochaeta, suggesting a shared allotetraploidization event for all extant grasses. Reconstruction of the ancestral grass karyotype (AGK) reveals a chromosome number of ante-ρ (x = 8) and post-ρ (x = 12), followed by chromosome number reduction by fusion events accompanying the diversification within major subfamilies. Biased subgenome fractionation and dominance are observed and inherited across Streptochaeta and major grass lineages. Comparative expression analyses of key floral genes in S. spicata indicate that the spikelet equivalent of this early-diverging lineage might represent a modified ancestral state of grass spikelet. Given its critical phylogenetic position, the Streptochaeta genome sheds light on the early karyotypic evolution of grasses, and the origin and key morphological innovation with respect to WGDs.

Early genomic and karyotypic evolution that underpinned grasses (Poaceae) diversification remains incompletely resolved. Here, the authors report genome assembly of Streptochaeta spicata and reveal the ancestral karyotype, subgenome dominance following genome duplication, and the origin of spikelet.

Introduction

As one of the most economically important plant groups, the grass family (Poaceae) provides the bulk of human nutrition and includes major cereals, such as rice, wheat, maize, sorghum, and barley. With ~12,000 species¹, the Poaceae dominates many terrestrial ecosystems and covers ~40% of the Earth’s land surface. These species are adapted to diverse environmental conditions^2,3 and demonstrate numerous morphological, physiological, and molecular innovations. Among them, the most striking is probably the spikelet (Fig. 1), serving as the basic unit of the inflorescence^4,5. Inflorescence structure and spikelet production are also important determinants of grain yield in cereals and have thus been a major target of human selection in domestication⁶.

Fig. 1. The time-scale framework of grasses, rho-WGD and the evolution of spikelet.

a Inflorescence (right) and spikelet equivalent (left) of S. spicata. b Estimation of divergence time across the major lineages of Poaceae (wild banana and pineapple as outgroup taxa) with hypotheses of ancestral grass karyotype and illustrations of monocot flower, grass spikelet and spikelet equivalent. Numbers associated with the nodes indicate million years ago (Ma).

Polyploidization or whole-genome duplication (WGD) is prevalent in the evolution of flowering plants^7–9, and is a genetic source of biological innovations. Three rounds of ancient WGD are recognized in grass genomes, with the most recent, termed ρ-WGD, likely coinciding with the origin of grasses^10–12. As a dramatic genomic event, ρ-WGD would firstly double the ancestral grass basic chromosome number, which was previously suggested to be five, six, or seven¹³, and lead to reorganization of the ancestral grass karyotype (AGK)^14–16 (Fig. 1b). This was subsequently followed by extensive multiple chromosomal rearrangements during the diploidization process. However, questions remain concerning the “basic” chromosome number of extant grasses, which has been inferred to range widely, from 2 to 18¹⁷. In addition to chromosome number reduction, massive losses of duplicate genes have also been inferred, and biased fractionation likely occurred with the preferential retention of gene copies from one parental subgenome in allopolyploids^18–21. This WGD event presumably contributed to the origin of the AGK and underlies morphological innovations in the grasses. The genomes of early-diverging grasses should harbor crucial insights into the genetic and developmental changes that arose early in the evolution of this important plant group^12,22.

However, as the only grass lineage without a typical spikelet, less attention has been given to generating and analyzing genomes of subfamily Anomochlooideae, which is sister to all other extant grass lineages^1,23. The four species from two genera in this subfamily are confined to tropical forests in Central and South America²⁴, and the chromosome number is 2n = 36 and 2n = 22 for Anomochloa and Streptochaeta²⁴, respectively, casting doubt on a reconstructed post-ρ AGK of x = 12 for grasses, as well as the currently recognized ante-ρ AGK of x = 7^15,25. In addition to questions concerning chromosome evolution, these taxa are also of great interest for their trait evolution. Species in Streptochaeta produce so-called “spikelet equivalents” that contain 11 or 12 spirally arranged bracts on a curving axis with six stamens and an enclosed gynoecium (Fig. 1a), which are spirally borne on the axis of the inflorescence²⁴. Homology between these bracts and glumes, lemma, palea, and lodicules in the typical grass spikelet is ambiguous based on morphological and anatomical evidence^24,26, although some suggestions are provided by expression analysis of certain MADS-box genes^27,28. Moreover, many of the loci known to control spikelet development were suggested to be present in a draft genome of S. angustifolia²².

In this paper, we present a chromosome-scale genome assembly for S. spicata using a combination of short-reads, Nanopore long-reads, and chromatin conformation (Hi-C) data. This genome, combined with large-scale comparative genomic analyses across grasses, enables us to pinpoint key genomic changes associated with the early evolution of the family, the origin of AGK, and the genetic basis of unique features present in the isolated lineage sister to all other grasses. In particular, we focus on the genomic nature of key morphological innovations as well as the karyotypic evolution of grasses.

Results

Chromosome-level genome assembly of S. spicata

Streptochaeta spicata was estimated to have a genome size of ~2.0 Gb from k-mer analysis (Supplementary Fig. 1), which is slightly larger than S. angustifolia at ~1.77 Gb²². We assembled the genome de novo through integrated sequencing of ~85× Nanopore long reads (171 Gb), ~89× MGI-SEQ short paired reads (179 Gb) and ~181× chromosome conformation capture (Hi-C) data (364 Gb) (Supplementary Table 1). The final assembly consists of 90 contigs totaling 1,845.6 Mb (N50 = 96.1 Mb, Supplementary Table 2). This assembly comprises eleven pseudochromosomes (Fig. 2a and Supplementary Fig. 2), consistent with a chromosome-level genome assembly of S. spicata. These pseudochromosomes contain 98.8% of the assembly span. Homologous chromosomal segments of S. spicata and rice, generally used as a reference in comparative genomics of grasses¹⁵, can be clearly identified (Fig. 2a). A comparison of the assembly to the genome of S. angustifolia highlights the high level of continuity of our chromosome-level assembly (Supplementary Fig. 3a). We mapped short-paired reads back to the assembly, revealing a mapping ratio of 99.7%; 97.7% of them could be properly paired. The long terminal repeat (LTR) assembly index (LAI) score representing assembly contiguity²⁹ was as high as 14.7, corresponding to reference genome quality (Supplementary Fig. 3b). And 96.6% of 1614 Benchmarking Universal Single-Copy Orthologs (BUSCO) genes could be identified in the assembly (Supplementary Table 3). Together, these results suggest a high-quality genome assembly for S. spicata.

Fig. 2. Genome characters of Streptochaeta spicata and comparative genomic analyses among grass genomes.

a The genomic landscape of S. spicata with the distributions of GC content, genes, Gypsy, Copia, and syntenic relationship to the rice genome. The inner circular representation shows inter-chromosomal synteny of S. spicata. b The comparison of density of short (less than 1000 bp) and long (equal to or more than 1000 bp) introns between the S. spicata and rice genomes. c The comparison of different repeat elements between short and long introns in the S. spicata genome. Source data are provided as a Source Data file.

Based on ab initio prediction, evidence from the transcript dataset and homologous proteins of flowering plant genomes, we identified 30,772 protein-coding genes in the S. spicata genome (Supplementary Table 4). Similar to protein-coding genes in Pharus latifolius¹², the density distribution of GC3 content (G or C in the third codon position) in S. spicata is intermediate between the ancestral unimodal distribution and the bimodal distribution in the core grasses¹¹ (Supplementary Fig. 4). The coding sequences of S. spicata genes averaged 1139 bp in length with 5.04 exons, similar to those of other grass genomes (Supplementary Table 4), but the average intron length is the longest among genomes used in our comparative survey (Fig. 2c). Moreover, although no more than 16% of introns exceed 1 kb among the six grass genomes compared here, 18.9% of introns in the S. spicata genome are longer than 1 kb, with the longest 340.7 kb. We manually checked the gene with the longest intron plus 90% of 100 randomly selected genes with >1 kb introns; all were supported by our transcriptome data and orthologous sequence in S. angustifolia (Supplementary Table 5). For the long introns, repeat sequences, particularly LTR retrotransposons, comprise 71.0% of the total length, compared to 30.8% for introns under 1 kb (Fig. 2d). Additionally, orthologues were identified for 95.6% of 30,772 genes of S. spicata in genomes of 15 other grasses and the outgroup taxa of pineapple (Ananas comosus, Bromeliaceae, and Poales) and banana (Musa acuminata, Musaceae, and Zingiberales) (Supplementary Fig. 5). Compared to the core grasses, 1310 Streptochaeta-specific orthogroups were shared by S. spicata and S. angustifolia.

As expansion of intron size is caused mainly by insertions of transposable elements (TEs), we surveyed genome-wide TE abundance and identified 82% of the genome as TEs (Supplementary Table 6). This proportion is much higher than that of other diploid grasses with smaller genomes, such as sorghum (62.8%), rice (32.3%), and Brachypodium distachyon (28%). The most abundant TE types in the S. spicata genome are Gypsy (44.6%) and Copia (12.6%) retrotransposons with significant bursts at 0.9 and 0.7 million-year ago (Ma) (Supplementary Fig. 6), respectively.

Phylogeny and origin of the spikelet in grasses

Based on 295 single-copy orthologous genes, we reconstructed the phylogeny of seven Poaceae subfamilies where whole genome sequences are available at the beginning of the study (Fig. 1b). As expected, Streptochaeta, representing the subfamily Anomochlooideae, was sister to all other Poaceae¹, followed by the divergence of the spikelet clade, with Pharoideae sister to the core grasses that comprise two major clades, BOP and PACMAD. The emergence of the lineage leading to Streptochaeta was estimated to around 102.1 Ma, and the Pharus lineage at ~92.1 Ma, with major clades of Poaceae diverging ~18 million years later (Supplementary Fig. 7). Meanwhile, the molecular evolution rate was low in S. spicata (Supplementary Table 7), even lower than that of P. latifolius, which was then the lowest among grasses¹², making S. spicata the slowest evolving grass yet documented.

In addition, the topology of the phylogenetic tree is consistent with the first appearance of a typical spikelet in the ancestral grasses after the split of the Streptochaeta lineage (Fig. 1b). The number of stamens evolved from six and four in the ancestral monocots and early-diverging grasses (e.g., Streptochaeta, Anomochloa, and Pharus) to three and two in the majority of core grasses^24,26,30, pointing to the floral traits of spikelet equivalent in Streptochaeta as a possibly modified ancestral state of grass spikelet.

ρ-WGD just predated the divergence of the Streptochaeta lineage

Previous phylogenetic analysis using transcriptome data indicated that the Streptochaeta genome shared the grass-specific ρ-WGD event¹¹, an event that was initially identified from genomic analyses of rice³¹ and was also inferred phylogenetically from the S. angustifolia draft genome²². To reassess this using our newly assembled genome, the distribution of synonymous substitutions per synonymous site (K_s) for paralogous genes within the S. spicata genome and related genomes were plotted, as well as for orthologous genes between it and A. comosus, P. latifolius, and rice (Supplementary Fig. 8). Three significant peaks were evident in the K_s distribution of S. spicata (Fig. 3a), presumably corresponding to the three WGD events experienced by the grasses¹¹ with the most recent event (K_s = 0.48) likely to be the ρ-WGD event. Compared with the peaks of the K_s distributions of orthologous genes, this event occurred after the split of the Bromeliaceae lineage (represented by A. comosus) and the grass family, while appearing nearly simultaneously with the divergence of Streptochaeta and Pharus and after the divergence of Streptochaeta and rice. This was likely to be caused by the slow molecular evolutionary rate of Streptochaeta.

Fig. 3. Inference of whole genome duplication (WGD) and subsequent biased subgenome fractionation across S. spicata and other grasses.

a Distribution of synonymous substitution per site (K_s) of paralogues in syntenic regions of the S. spicata genome and of orthologues between S. spicata and Pharus latifolius, rice and Ananas comosus. The major peak (dark green) within the S. spicata genome could point to a WGD event shared by P. latifolius and core grasses. b Phylogenomic analysis of the WGD event in S. spicata and the other six grasses. Numbers associated with the branches indicate the number of gene families (after dash) from S. spicata with duplicate gene pairs (before dash) supporting the duplication events. And those numbers from other grass genomes are indicated in the heatmap. The WGD in the red star common to the S. spicata and all other grasses was the most frequently supported. All mapped duplication events have a bootstrap value 95% or higher in the gene trees. c Gene retention patterns among the two homoeologous copies of lineage group (LG) 11, 19, and 22 of A. comosus in S. spicata (blue) and rice (red). d Inference of gene loss after the shared WGD in S. spicata and three grasses, pointing to the presence of a less fractionated (upper) subgenome and a more fractionated one (below). The inference was performed in POInT by calculating the gene losses observed across the four species and along the phylogenetic tree depicted. The numbers above each branch indicate the number of genes in each subgenome surviving to that point, and those losses are in parentheses. Source data are provided as a Source Data file.

We further employed a phylogenomic approach by inferring the duplication history of paralogous gene pairs in synteny from each genome of S. spicata, S. angustifolia, and the other six grasses, to identify and place the ρ-WGD onto the grass phylogeny. The duplicate genes from eight taxa all coalesced most frequently on the common branch leading to Streptochaeta and other grasses (Fig. 3b), strongly supporting that Streptochaeta shared the ρ-WGD with the remaining grasses. Divergence time estimation indicated that the duplication happened around 104 Ma (Supplementary Fig. 9), just before or coincident with the split of the lineage leading to Streptochaeta (Fig. 1b).

Biased subgenome fractionation and allopolyploidization

Having confirmed the placement of the ρ-WGD in the evolution of grasses, we further investigated the evolutionary dynamics with respect to gene losses after duplication, using the genome of Ananas comosus as a diploid outgroup (Supplementary Figs. 10 and 11). A common, less fractionated subgenome in S. spicata and the rice genomes was identified, which has retained more overall genes than the homoeologous chromosomes derived from the ρ-WGD (Fig. 3c, Supplementary Figs. 10 and 11 for other chromosomes). The less fractionated subgenome retained about 14% and 18% more genes in S. spicata and rice, respectively. As the same fractionated subgenomes were identified in S. spicata and rice, it may be inferred that the biased fractionation occurred instantly after the ρ-WGD and was subsequently retained in the genomes of different grass lineages.

We further assessed the patterns of gene loss using the Polyploidy Orthology Inference Tool (POInT)³² with the genomes of S. spicata, P. latifolius, and two core grasses. POInT statistically assigned genes from each of the four grass genomes to the two subgenomes with high confidence, and 75% of the 4843 pillars (loci containing homologous genes) (Supplementary Data 1) had subgenome assignments with posterior probabilities >0.978 (Supplementary Fig. 12). We observed a clear signal of biased fractionation with an overall estimate of the ratio of genes retained from the less fractionated subgenome relative to the more fractionated subgenome at 0.3527, and the estimated value was the lowest for the branch immediately after the ρ-WGD (Fig. 3d). Moreover, the gene loss rate along this branch was higher than the other branches in the phylogenetic tree. There were 566 genes lost from the first subgenome along the stem branch before the divergence of S. spicata, and 2069 were lost from the second. We hence confirmed the presence of biased fractionation in the grass genomes after the ρ-WGD³³, resulting in greater gene content for a dominant subgenome.

Ancestral grass karyotype after ρ-WGD

Using genomic data from S. spicata, P. latifolius, and rice, we reconstructed the post-ρ ancestral grass karyotype (post-ρ AGK) at the macrosyntenic level (Supplementary Fig. 13). A high level of synteny along the whole chromosome among these genomes was revealed, with rearrangements mainly involving chromosomes corresponding to Chrs1, 7, and 12 in rice (Fig. 4a). Conserved synteny spanning the entire chromosome length was observed between S. spicata Chr1 and rice Chr1, indicating that both retain the ancestral karyotypic state, whereas the rearrangement between Chr1 and Chr7 in P. latifolius represents a lineage-specific derived event. Similarly, whole chromosomal collinearity between P. latifolius Chr12 and rice Chr12 confirms their shared ancestral state, in contrast to the derived configuration of S. spicata Chr7.

Fig. 4. Reconstruction of the ancestral karyotype ante-ρ and its evolutionary trajectory toward the post-ρ in Poales.

a Dotplot synteny comparison between S. spicata, P. latifolius, and O. sativa. b Reconstructed karyotype of the post-ρ AGK. c Inferred evolutionary steps from the ante-ρ to the post-ρ AGK. d Conserved synteny of rice Chr8 supporting ancestral chromosomes, illustrated with synteny dotplots of rice Chr4 and Chr9, as well as collinear regions in Jo. ascendens Chr17, Ju. effusus Chr2, A. comosus Chr20, and Sp. polyrhiza Chr12. e The synteny comparison between S. spicata and outgroup species (Jo. ascendens, Ju. effusus, A. comosus, and Sp. polyrhiza).

Based on these comparative analyses, we reconstructed the post-ρ AGK, which revealed a karyotype consisting of 12 protochromosomes (Fig. 4b). The results show that the post-ρ AGK is nearly identical to that of rice, as well as explaining the chromosome evolution in early-diverging grasses, probably with a reciprocal translocation (RT) followed by an end-end joining (EEJ) (Pρ7 and Pρ12) in Streptochaeta and one RT event (Pρ1 and Pρ7) in Pharus being derived (Fig. 4b), both involving the same ancestral chromosome Pρ7.

We further used A. comosus, representing the lineage sister to the rest of Poales, as an outgroup to validate the post-ρ AGK and the proposed fusion points involved in the Streptochaeta RT-EEJ event (Supplementary Fig. 14a). Rice Chr7 and Chr12 predominantly correspond to two (Chr9 and Chr18) and three (Chr7, Chr13, and Chr17) chromosomes in A. comosus, respectively. Meanwhile, Chr7 of S. spicata is also homologous to these five chromosomes in A. comosus. A key finding was that the entire A. comosus Chr9, which shows continuous synteny with one arm of rice Chr7, spans both fusion breakpoints of RT and EEJ events identified on Chr7. In addition, the supposed fusion breakpoints in S. spicata were supported by high coverage of long reads and a single assembled contig, and also validated by the Hi-C data (Supplementary Fig. 14b). This excluded the possibility of artifacts in the assembly of S. spicata Chr7. Together, these findings provide robust validation that the rice genome retains a karyotype highly similar to the post-ρ AGK, supporting the basic chromosome number of 12 in the most recent common ancestor (MRCA) of all extant grasses (Fig. 4b).

Ancestral grass karyotype before ρ-WGD

By tracing back from the post-ρ AGK, we reconstructed the ancestral grass karyotype prior to the ρ-WGD, termed ante-ρ AGK (Aρ-AGK, Supplementary Fig. 15). This resulted in a reconstruction of Aρ-AGK consisting of eight protochromosomes, which subsequently underwent WGD and formed an initial 16 chromosomes. These duplicated chromosomes then generally experienced one fission, two nested chromosome fusions (NCF) and three EEJ events, giving rise to 12 chromosomes of post-ρ AGK (Fig. 4c). Aρ2′ and Aρ3′ were joined via NCF to form Pρ2; Aρ4′ and Aρ7 underwent NCF and subsequently fused with Aρ5′ via EEJ to form Pρ3; Aρ6′ and Aρ7′ underwent EEJ to generate Pρ12. A fission event occurred in Aρ8′ with one of its segments fused with Aρ3 via an EEJ to give rise to Pρ4, while the remainder was retained as Pρ9 (Fig. 4c). This fission event is validated by conserved synteny, especially the ancestral continuity shown by rice Chrs8, 4, and 9 and collinearities in A. comosus Chr20, Joinvillea ascendens Chr17 (Joinvilleaceae of the clade sister to Poaceae), Juncus effusus Chr2 (Juncaceae, Poales), A. comosus Chr20, and Spirodela polyrhiza Chr12 (basal monocot order Alismatales) that together show Aρ8 (Pρ8) is ancestral rather than a fusion product (Fig. 4d).

Our analysis confirms that Aρ7 was an independent protochromosome (or the extra one compared to Murat et al.¹⁵ with a total of seven protochromosomes, while Aρ8 corresponding to the seventh one, as previously numbered, see details in “Methods”), in ante-ρ AGK, which consisted of eight protochromosomes. The primary evidence stems from the fate of its duplicated homeologs, which contributed to segments of Pρ3 and Pρ12 (Fig. 4c, orthologous to rice Chr3 and Chr12, respectively). These segments maintain extensive collinearity with entire chromosomes of Jo. ascendens (Chr18), Ju. effusus (Chr14), and A. comosus (Chr8) (Fig. 4e). This conserved synteny, coupled with the evolutionary trajectory traced from basal monocots like S. polyrhiza to the Poaceae, provides clear evidence for the existence and evolution of Aρ7 through subsequent multiple WGD events.

Karyotype evolution in major lineages of grasses

Using a bottom-up, micro-to-macro synteny approach applied to 26 diploid grass genomes (Supplementary Table 8), we reconstructed the ancestral karyotypes for key nodes within the Poaceae phylogeny and traced lineage-specific evolutionary trajectories.

Our reconstructions revealed stable ancestral karyotypes along the backbone nodes of grasses, with little change since the post-ρ AGK in the common ancestor of the BOP and PACMAD clades, and through to the crown node of the Bambusoideae and Oryzoideae (Fig. 5a). However, distinct patterns of chromosomal rearrangements were observed accompanying the diversification following the divergence of major subfamilies (Fig. 5b). Consistent with Winterfeld et al.³⁴, the BOP clade initially maintained karyotype stability until the ancestral node of the Pooideae, with most early-diverging pooid lineages preserving an x = 12 karyotype close to the post-ρ AGK. However, the late-diverging lineages, such as Brachypodieae, Poeae, and Triticeae, underwent extensive chromosomal rearrangements. One EEJ and four NCF events occurred after the early divergence of the Brachypodieae, resulting in the 7 chromosomes at the ancestral node of the Poaeae and Triticeae. This reduction was accompanied by an increase in genome size, followed by an additional RT event in the Triticeae. Interestingly, we identified that Bd4 of B. distachyon resulted from two rounds of NCF involving three ancestral chromosomes: Pρ9 into Pρ12, and then Pρ11 into Pρ9- Pρ12 within Brachypodium. Bs5 and Bs10 of B. stacei evolved from a RT between Pρ11 and Pρ9- Pρ12, followed by an NCF of Pρ9 into Pρ12 (Supplementary Fig. 16).

Fig. 5. Karyotype evolution in grasses.

a Reconstructed ancestral grass karyotype (post-ρ AGK), with a 12-color code, that evolved into ancestral karyotypes of the common ancestor of core grasses, BOP, PACMAD, and other taxonomic categories within the grass family. The karyotype change and the chromosome number of the internal node where the fusion event occurred is shown in the large colored rectangle. Uniformly colored child nodes in small rectangles denote that the karyotype remains stable, with no significant large-scale chromosomal rearrangements compared to the parent node. The three types of chromosome fusion, namely end-end joining (EEJ), reciprocal translocation (RT), and nested chromosome fusion (NCF), are illustrated above as examples of karyotype alterations observed in S. spicata, P. latifolius, and B. distachyon. The genome structure of extant species is characterized by conserved syntenic blocks, represented in a 12-color code of protochromosomes. Within these blocks, specific dots denote collinear genes. b The heatmap and numbers show the frequency of chromosomal rearrangement events from post-ρ AGK to individual karyotype of 30 representative species of eight grass subfamilies, with the additions of three polyploid bamboos and a tetraploid, Phragmites australis (Arundinoideae).

Within the PACMAD clade, the crown node of Chloridoideae and Panicoideae exhibited a similar evolutionary trajectory, progressing from an RT to EEJ and an NCF event. This sequence of events resulted in a reduction of the chromosome number from 12 to 10. Although different partner chromosomes were involved, Pρ10 participated in both RT and EEJ processes, and Pρ9 was also involved in two NCF events in these two independent processes (Fig. 5a), indicating that the two chromosomes represented hotspots of rearrangement during the early evolution of PACMAD. Phragmites australis, an allotetraploid (2n = 4x = 48) in Arundinoideae, possesses two sets of chromosomes that are completely collinear with those of rice³⁵, which further corroborates the inference that the karyotype of the most recent common ancestor of the PACMAD clade is identical to the post-ρ AGK karyotype (Fig. 5b).

Genetic basis underlying the origin of the grass spikelet

Unlike typical grass spikelets, the S. spicata spikelet equivalent lacks obvious glumes, lemmas, paleas, and lodicules, and instead is composed of 11–12 spirally arranged bracts (Fig. 1a). These bracts could be matched to the bract-like organs of typical spikelets^27,28 and bracts 1–5, bracts 6–8 were further interpreted as glume-like and lemma/palea-like, respectively³⁶. To infer the homologies of them we investigated the evolution of related gene families for floral development in the S. spicata genome and explored their expression patterns across different tissues (Supplementary Fig. 17). Two FRIZZY PANICLE (FZP)-like genes of the APETALA2 (AP2) transcription factor (TF) family³⁷ were identified in the S. spicata genome (Supplementary Fig. 18) compared to two copies in the Pharus genome and only one for the core diploid grasses¹². The identified FZP-like genes were phylogenetically clustered into two groups with only one copy having the grass-specific protein motif (Supplementary Fig. 19). Nevertheless, expression for these two copies was scarcely detectable (transcripts per million (TPM) <1) in any of the tissues examined in S. spicata (Supplementary Fig. 20), including the floral tissue at early developmental stage represented by the young inflorescences of <2 cm here. We also identified and examined the expression of another 11 key TFs involved in regulating grass inflorescence architecture and establishment of floral meristem⁵ (Supplementary Fig. 20). Unlike the condition found for FZP, expression was detected for nearly all of them despite exhibiting different patterns, such as a broad expression ranging from vegetative to reproductive tissues for SUPERNUMERARY BRACT (SNB)-like and INDETERMINATE SPIKELET 1 (IDS1)-like genes and more specific expression for ABERRANT PANICLE ORGANIZATION (APO1)-like and DROPPING LEAF (DL)-like genes. However, the expression patterns of these genes across individual tissues in S. spicata were very different from their previously reported expression domains in rice (Supplementary Fig. 20), likely reflecting their differentiated functional roles between the two lineages.

We further considered the MADS-domain TFs, the increase and diversification of which is often associated with evolutionary novelties in plants³⁸. We reconstructed the evolutionary history of these TFs across land plants, which were further split into 23 subfamilies (Fig. 6d). A general picture was the constant increase of MADS-domain genes during the evolution of land plants, tightly related to the multiple rounds of WGD events, both at the clade- and the species-levels. Streptochaeta spicata shares all the major subfamilies and comparable 60 gene copies with the remaining grasses. This number is 56 in S. angustifolia and larger than those of species from other families in Poales due to the ρ-WGD event. In addition, a tandem duplication of SsMADS7 resulted in two copies.

Fig. 6. Genomic basis for the origin of the grass spikelet.

a Gene expression patterns of ABCDE MADS-domain genes from various tissues of S. spicata. The vegetative and reproductive tissues were clustered according to the expression of MADS-domain genes. Expression values were scaled by log2(TPM + 1), in which TPM represents transcripts per kilobase of exon per million mapped reads. RNA-seq data were generated from three independent biological replicates for each tissue. The flowering ABCDE model in the model grass rice (b) and S. spicata (c) that specifies reproductive tissues is proposed based on the gene expression values (bar heights) from (a). d The evolutionary history of MADS-domain transcription factors (TFs) across land plants is related to the multiple rounds of WGD/WGT events. The heatmap and numbers show the gene copy numbers of each MADS-domain gene family in individual genomes. Red and blue stars on the phylogenetic trees represent WGD and WGT events, respectively. Source data are provided as a Source Data file.

As expected, the ABCDE MADS-domain genes were mainly expressed in the reproductive tissues in S. spicata (Fig. 6a), with only a few, such as the FUL-like genes, expressed in vegetative tissues as well. We carefully dissected the spikelet equivalents of S. spicata into bracts 1–5, 6, 7–8, 10–12, stamen, and pistil to assess the expression patterns of these genes in more detail by RNA-seq. The bracts 1–5 and 6, 7–8 and 10–12 were more similar to each other, respectively, with respect to the expression patterns, while the stamen and pistil clustered together (Fig. 6a). Generally, their expression profiles in S. spicata agreed to some extent with their putative ascribed roles in floral organ patterning in the model grass rice (Fig. 6b, c) if we considered a simple correspondence between bract 6 and lemma, bracts 7–8 and palea, and bracts 10–12 and lodicule. For example, the B-class gene MADS4 was mainly expressed in the bracts 10–12 and lodicule in S. spicata and rice, respectively. Moreover, the expression of C- and D-class MADS-domain homologues was confined to the stamen and pistil, both in rice and S. spicata. However, the majority of MADS-domain TFs in S. spicata tended to exhibit wider ranges of expression across different floral regions than their counterparts in rice, especially for the A-, B- and E-class genes. The A- and B-class genes have been detected with a relatively high expression level in the bracts 1–5, pointing to a somewhat flower-like identity of these bracts.

To further investigate the transcriptomic differences between bracts and true leaves, Gene Ontology (GO) enrichment analysis was performed on the differentially expressed genes (DEGs). The results indicated that photosynthesis-related terms, including “thylakoid,” “chloroplast,” and “photosynthetic membrane,” were enriched among the downregulated DEGs (Supplementary Figs. 17 and 21). Additionally, in line with the aforementioned MADS-box gene expression, upregulated DEGs showed enrichment in terms, such as “specification of floral/plant organ identity” and “regionalization.” Collectively, these GO enrichment profiles, characterized by the downregulation of photosynthesis-related terms and the upregulation of floral identity-related terms, are consistent with the transition of these bracts toward reproductive accessory organs.

Discussion

The origin and early evolutionary history of grasses was accompanied by a series of ancient polyploidization events that have been difficult to unambiguously resolve, particularly the most recent one^10,11,22. Challenges include the antiquity of the ρ-WGD and the likely short interval between it and the initiation of species diversification in grasses¹², as well as the lack of high-quality reference genomes from the early-diverging grasses. We used the chromosome-level genome assembly of S. spicata, representing the Anomochlooideae, which is sister to the rest of extant Poaceae, and integrated multiple approaches based on chromosome-scale genomic data to resolve the history of ancient grass polyploidization. Our study thus provides a temporal framework for the occurrence of the ρ-WGD and the origin of grasses. The occurrence of ρ-WGD was confirmed to be shared by all the extant grasses, and we dated it to ~104 Ma, shortly before or coincident with the divergence of grasses around 102 Ma, consistent with estimates based on large-scale transcriptomic data of Poaceae²³. Moreover, concurrent dominant and recessive subgenomes with biased gene losses could be identified across representative grasses, which are usually interpreted to result from allopolyploidization rather than autopolyploidization^19,33. Thus, the ancestral grass genome was probably the product of an allotetraploid event. However, firm inference of the timing of ρ-WGD and the mode of polyploidization depends on the evolutionary relationships of Streptochaeta and Anomochloa; while conserved target capture sequence data support their monophyly³⁹, a high-quality genome of Anomochloa would be useful to confirm these results. In short, our study represents a rare example of accurately establishing the timing and evolution of an ancient WGD and lineage divergence in a major angiosperm family and could be adopted as a model system for studying paleo-polyploidization events in plants.

Well-conserved genomic synteny has long been recognized during grass evolution⁴⁰, with cereals being treated as a single genetic system^41–43. However, key aspects of their ancestral karyotype evolution remained largely unresolved. An ancestral chromosome number of 12 for grasses was proposed in 1931 by Avdulov and has long and widely been accepted^{15–17,34,44}, while numbers, such as 11, were also suggested based on the chromosome number of early-diverging grasses and Streptochaeta²⁵. By leveraging the S. spicata genome, which exhibits slower duplicate gene loss and fewer chromosomal rearrangements, together with the most comprehensive genomic sampling to date from across Poaceae and key outgroup taxa, such as Joinvillea⁴⁵ and Ananas⁴⁶, our study provides insightful results.

We demonstrated that the ancestral grass karyotype prior to the ρ-WGD or ante-ρ AGK consisted of eight distinct protochromosomes, rather than seven^15,47,48 or five to seven as previously suggested¹³. Following the ρ-WGD, the chromosome number doubled to 16 and was subsequently reduced to 12 in the most recent common ancestor (MRCA) of grasses through one fission, two nested chromosome fusion (NCF) and three end-end joining (EEJ) events, rejecting alternatives, such as 11^25,49. Our results partially validate previous hypothesis on chromosome rearrangements⁵⁰, providing a more detailed and enriched evolutionary trajectory for the early grass lineages after the ρ-WGD. An interesting finding is that an ancestral chromosome of Pρ7 became a hotspot of rearrangements in the early stages of grass evolution, likely due to the characteristics of this protochromosome itself.

By reconstructing the AGK evolution at the subfamily level of grasses using 26 genomes, our analyses revealed the general trend of lineage-specific evolution and massive chromosomal rearrangements accompanying grass diversification following the divergence of major subfamilies. Within the BOP clade, the pooid grasses experienced the most extensive rearrangements (Fig. 5b), while the rice and bamboo subfamilies maintained extremely conserved karyotypes^15,21,51,52. In contrast, Chloridoideae and Panicoideae of the PACMAD clade underwent moderate levels of change. This detailed evolutionary framework not only provides a holistic view of early karyotype evolution across the grass family but also consolidates the phylogenetic relationships of subfamilies from the point of view of karyotype evolution. Moreover, it paves the way for addressing intricate challenges in tracing karyotype evolution among more closely related taxa, e.g., resolving whether Bs5 and Bs10 in Brachypodium stacei arose through ancestral chromosome fission or RT of two ancestral chromosomes⁵³. As demonstrated by a recent study in the cotton family⁵⁴, the evolutionary framework of the grass karyotype can effectively elucidate phylogenetic relationships inferred from genomic sequences, supporting its use as a model for studying karyotype evolution in other major plant families.

Duplicate gene loss and karyotype evolution dynamics could have occurred at different rates across various grass lineages. However, we observed biased fractionation of subgenomes at the chromosome level, which was likely to be established shortly after the ρ-WGD and inherited by all descendent grass lineages. The biased evolution of grass subgenomes points to an allotetraploidization for this duplication event³³, and subgenome dominance could still have played a role in the evolution of grass genomes, whether experiencing additional WGD events or not^9,21.

With the evolutionary characters of ρ-WGD clarified, we could relate early genomic events to the emergence and elaboration of grass innovations. In particular, gene duplicates with their appearance due to WGD were suggested to underlie the evolution of the structure of the spikelet^55–57. Analyzing the evolution of key genes involved in spikelet development, along with MADS-domain genes, through expression studies of dissected spikelet-equivalent parts of Streptochaeta provided valuable insights into the origin of the spikelet^{5,28,38,57–60}. Intriguingly, neither of the two FZP gene copies could be confidently found as expressed at floral maturity, although we cannot rule out expression in the meristems of S. spicata inflorescence, and more work is needed to confirm this result. It is somewhat expected, as FZP is responsible for grass spikelet identity^5,61,62 while S. spicata does not have a typical structure of spikelet. One FZP copy may have gained a distinct expression pattern in the ancestral grasses after the split of the Streptochaeta lineage and before the emergence of Pharus, contributing to the origin of the spikelet. Subsequently, the other was lost after the split of the Pharus lineage¹². In addition, the expression patterns of other key regulating genes, such as DL and IDS1, during the grass floral development also vary greatly between S. spicata and rice, with a general trend of specialized expression in the latter. In contrast, the overall expression pattern of ABCDE MADS-domain genes in different parts of the S. spicata spikelet equivalent is similar to what is known in rice^63,64, supporting the homologous origin of the bracts 6–12 in the S. spicata spikelet equivalent and bract-like organs (lemma, palea, and lodicule) of grass typical spikelets^27,28,36. However, these genes in S. spicata tend to have broader expression domains, a phenomenon analogous to that observed in the early-diverging angiosperms, such as Nymphaeaceae⁶⁵. Together, these results indicate that the spikelet equivalent of this early-diverging lineage might represent a modified ancestral state of grass spikelet, rather than an intermediate evolutionary stage in the origin of true spikelet. This hypothesis could be supported by more evidence from comparative genomic analyses in the other spikelet equivalent genus, Anomochloa, in the future. Furthermore, single-cell omics technologies have been recently used in core grasses^66–69, delineating floral development at cellular resolution with more details, and revealing how relevant genes coordinate their spatial expression to regulate the developmental process of grass inflorescence and spikelet. The origin of the grass spikelet might thus be accompanied by the specific and coordinated expression of the floral-related genes, possibly as well as accompanied by neofunctionalization of duplicate genes in the evolution of grasses, as evidenced by the comparison of their expression patterns between S. spicata and rice above. The bulk RNA-seq dataset generated here, and particularly the single-cell transcriptomic atlas of the S. spicata spikelet equivalent in the future, will provide a valuable resource for comparison in further revealing the molecular dynamics underlying the origin and evolution of grass spikelets.

In summary, the chromosome-level reference genome of S. spicata and transcriptomes of dissected tissues of the spikelet equivalent represent resources for the exploration of the genetic principles and the evolution underlying the unique traits in the early evolution of grasses, providing a valuable addition to the toolkit for evolutionary and comparative genomic studies in cereals.

Methods

Plant materials, sequencing, and assembly

Living plants of Streptochaeta spicata were cultivated in the greenhouse at the Kunming Institute of Botany, Chinese Academy of Sciences. A voucher specimen has been deposited in the herbarium of Kunming Institute of Botany, Chinese Academy of Sciences (KUN), labeled as No. Sspi06. High-quality genomic DNA of S. spicata was extracted from young leaves of a single plant using the QIAGEN® Genomic kit. Tissues from root, stem, leaf, and different parts of the spikelet equivalent were sampled and immediately frozen in liquid nitrogen for RNA extraction and transcriptome sequencing. We further sequenced ~10 Gb of PacBio full-length cDNA data and ~200 Gb of RNA sequencing (RNA-seq) short paired reads from 33 samples of root, stem, leaf, and different parts of the spikelet equivalent. We employed three DNA sequencing strategies, including short paired-end reads, Nanopore long reads, and chromosome conformation capture (Hi-C) sequencing. Short-reads libraries with a 400 bp insert fragment size were constructed and then sequenced on an MGISEQ-2000 system with a PE150 strategy. A total of 179 Gb (~89×) raw reads were obtained and further used to evaluate genome size based on k-mer frequency distribution. The transcriptome and Hi-C libraries were prepared and sequenced on the Illumina NovaSeq platform to generate 150 bp read pairs. Nanopore long-reads were sequenced on the PromethION platform.

The de novo chromosome-level assembly comprised three steps. First, Nanopore long reads with low-quality and short length were discarded, and the remaining reads were assembled into contigs using NextDenovo v2.3.0 (https://github.com/Nextomics/NextDenovo.git), with the following parameters: read_cutoff: 1k, seed_cutoff: 33k. Second, the draft assemblies were initially corrected with Nanopore long reads using Racon v1.3.1 (https://github.com/isovic/racon.git), and MGISEQ paired-end reads were recruited to further polish the assembly with Nextpolish v1.2.4⁷⁰. Subsequently, high-quality Hi-C read pairs were aligned to the draft assembled contigs using Bowtie2 v2.3.2⁷¹, retaining only uniquely mapped reads with valid di-tags according to chromatin interaction results. Finally, the Hi-C corrected contigs were clustered into 11 pseudo-chromosomes using the LACHESIS software (https://github.com/shendurelab/LACHESIS) with parameters of “cluster_min_re_sites = 100, cluster_max_link_density = 2.5, cluster_noninformative_ratio = 1.4, order_min_n_res_in_trunk = 60, and order_min_n_res_in_shreds = 60,” followed by manual adjustment.

To assess the quality of the assembled genome, we first mapped the short read data to the assembly using BWA v0.7.12-r1039⁷² with default parameters. Then we used BUSCO v4.0.5⁷³ to examine the gene content with the Embryophyta odb10 database. In addition, the long terminal repeat (LTR) assembly index (LAI) score representing assembly contiguity using LTR retrotransposons²⁹ was evaluated for the chromosome-level genome sequences of S. spicata, with a comparison of rice, B. distachyon, Oropetium thomaeum, S. bicolor, and P. latifolius.

Repeat and gene annotation

The tandem repeats in the genome were annotated using GMATA v2.2 (https://sourceforge.net/projects/gmata/?source=navbar) and Tandem Repeats Finder v4.07b (http://tandem.bu.edu/trf/trf.html). Transposable elements (TEs) were identified using a combination of de novo predictions and homology-based approaches. Briefly, the LTR-RTs were first annotated using LTR_FINDER v1.07⁷⁴ and LTRharvest v1.5.10⁷⁵, and the identified TE candidates were then filtered with LTR_retriever⁷⁶. In parallel, miniature inverted TEs (MITEs) were annotated with MITE-Hunter⁷⁷. Both LTR and MITE libraries were passed to RepeatModeler v1.0.11 (http://www.repeatmasker.org/RepeatModeler/) with default parameters. Finally, the TE library together with Repbase⁷⁸ (http://www.girinst.org/repbase) was used to annotate and mask the genome with RepeatMasker v4.0.5 (http://repeatmasker.org/). The age of full-length LTR retrotransposons was estimated based on the evolutionary distance between the LTRs using the Jukes–Cantor model⁷⁹, which was further converted into a substitution rate. For rice, the rate was chosen as 1.3 × 10^–8 mutations per site per year⁸⁰, and corrected for different grass species based on the ratio of their respective WGD peak value compared to rice⁸¹.

Gene annotation of the S. spicata genome was performed by a combination of three independent approaches, including homology search, RNAseq-based prediction, and ab initio prediction. Protein-coding genes from related species, such as rice, B. distachyon, and S. bicolor, were used for homology-based annotation with GeMoMa v1.6.1⁸². RNA reads from the root, stem, leaf, and different parts of the spikelet equivalent were aligned to the reference genome using STAR v2.7.3a⁸³, and the transcripts were assembled using Stringtie v1.3.4d⁸⁴ with default values. PASA v2.3.3⁸⁵ was then used for RNAseq-based prediction. For ab initio prediction, we used Augustus v3.3.1⁸⁶ and GeneMark-ST⁸⁷ on the repeat-masked genome using model training based on the closely related species. To finalize the gene set, all predictions were combined into a nonredundant gene set using EvidenceModeler (EVM) v1.1.1, and these genes were named following the nomenclature rules of the rice genome (http://rice.plantbiology.msu.edu/analyses_nomenclature.shtml).

The function of annotated genes was assigned based on the best match derived from the alignments to NCBI nonredundant (NR), Kyoto Encyclopedia of Genes and Genomes (KEGG), Eukaryotic Orthologous Groups of proteins (KOG), Gene Ontology (GO), and Swissprot database using BlastP with parameters of -evalue 1e-5, -max_target_seqs 1. Motifs and domains were annotated using InterProScan v5.32-71.0⁸⁸ by searching against publicly available databases. In addition, for non-coding RNAs, tRNA genes were predicted by tRNAscan-SE v2.0⁸⁹, and the rRNA, snRNA, and miRNA identifications were performed using Infernal software⁹⁰ by searching against the Rfam database (http://rfam.xfam.org/).

Gene family analysis

To identify gene family groups, the amino acid and nucleotide sequences of 15 grass species representing seven subfamilies in the Poaceae and two outgroup taxa were downloaded from various sources: Ananas comosus (v3), Brachypodium distachyon (v3.1), Hordeum vulgare (v1), Oryza sativa (v7), Setaria italica (v2.2), Sorghum bicolor (v2.1), and Zea mays (v4) from Phytozome (https://phytozome.jgi.doe.gov/), Aegilops tauschii (v4.0), Leersia perrieri (v1.4), and Triticum urartu (ASM34745v1) from Ensembl (http://plants.ensembl.org/), Cenchrus americanus (v1.1) and Phyllostachys edulis from GIGAdb (http://gigadb.org/dataset/), Pharus latifolius from NCBI database, Musa acuminata (v2) (https://banana-genome-hub.southgreen.fr/), Olyra latifolia from BambooBase (https://www.genobank.org/bamboo), Oropetium thomaeum (v2) (https://genomeevolution.org/coge), and Streptochaeta angustifolia (v1) from CyVerse data commons (https://datacommons.cyverse.org/browse/iplant/home/aseetharam/Streptochaeta_v1_publication_release_2021-05-13). We identified orthologous groups and conducted comparative analyses among S. spicata, S. angustifolia, P. latifolius, and core grasses by OrthoFinder v2.3.3⁹¹ with default parameters. For gene families shared only by grasses and those specific to S. spicata, GO annotations were carried out using rice genes as a reference. Gene family expansion and contraction along the evolution of the grass family were identified using CAFE v5.0⁹², with the species tree inferred from the single-copy orthogroups as the input tree. This analysis used three local gene family evolutionary rates: one for the terminal and related internal branches of outgroup taxa, the second for the stem of grasses and the terminal branch of S. spicata and P. latifolius, and the third for all the branches within grasses. The Poisson distribution was used for root frequency (parameter -p), and 1000 re-samples were used to calculate P-values at a 0.05 significant level.

Phylogenetic analysis and divergence time estimation

We used 295 single-copy orthologous groups from all 15 species mentioned above and S. spicata to reconstruct the phylogenetic tree. MAFFT v7.407⁹³ was employed to align each gene with the E-INS-I algorithm⁹⁴, and then filtered and concatenated to create the final alignment. RAxML v0.9.0⁹⁵ was used to reconstruct a maximum likelihood phylogeny with the GTR + G model and 1000 bootstrap replicates. The relative evolutionary rates among S. spicata, S. angustifolia, P. latifolius, and the core grass species were computed with A. comosus and M. acuminata as outgroup taxa using the concatenated dataset of 295 genes.

The origin and divergence time of grasses were estimated using MCMCtree from the PAML package v4.9h⁹⁶. The phylogenetic tree and the four-fold-degenerate site data derived from the 295 gene alignments were used as input. A relaxed molecular clock with two fossil-based calibration points were used: one set at >65 Ma for the crown of Poaceae and the other at >47.8 Ma for the crown point of subfamily Pooideae⁹⁷. The root age of the tree was set to a maximum of 150 Ma. To ensure convergence of this analysis, we carried out two independent runs of 2,000,000 Markov chain Monte Carlo generations, sampling every 100 steps, followed by removing the first 10% of iterations as burn-in.

WGD and duplicated genes analyses

To explore the WGD history of S. spicata, genome-wide duplications within the genome were first investigated. The assembled genome sequences were self-aligned using the Large-Scale Genome Alignment Tool (LAST; http://last.cbrc.jp/). Syntenic blocks within the genome were analyzed and plotted using MCScan⁹⁸, with the quota set as one-to-one and using a reciprocal best hit strategy.

The K_S distributions for sets of paralogous genes were constructed using the wgd suite⁹⁹ following the approach used by Vanneste et al.¹⁰⁰. This analysis was performed for S. spicata, S. angustifolia, P. latifolius, the core grasses, including rice, B. distachyon, Ol. latifolia, Or. thomaeum, and S. bicolor, as well as for outgroup A. comosus. Sets of paralogous families for each genome were determined by all-against-all comparison using BLASTP with an E value cut-off of 1 × 10^–10. The Markov cluster (MCL) algorithm¹⁰¹ was used to cluster the gene families, and MUSCLE v3.8.1551¹⁰² was adopted to build alignments. K_a and K_S rates between gene pairs within a gene family were evaluated using the ML method with the model of Nielsen and Yang (1998)¹⁰³ in CodeML of the PAML package. K_S distributions for paralogs with collinearity were visualized using i-ADHoRe v3.0¹⁰⁴.

Several analyses were conducted to determine the placement of the WGD events over the phylogenetic tree. First, we test whether the WGD event of S. spicata was shared with S. angustifolia, P. latifolius, and five core grass species using PUG software (https://github.com/mrmckain/PUG). The identification of duplicate gene pair sets and the estimation of multiple gene trees and species tree for the PUG analysis followed the strategy described by McKain et al.¹¹. Briefly, duplicate gene pair sets were identified based on homologous gene families, and gene trees were reconstructed for individual duplicate gene sets. A species tree was generated from single-copy orthologous genes, and comparisons between gene trees and the species tree were used to determine whether duplicate gene pairs were consistent with a shared WGD history. At the same time, S. spicata, P. latifolius, and five core grass species used for the K_S analysis, together with outgroup taxa A. comosus and M. acuminata, were selected to identify gene families and gene duplication events using OrthoFinder v2.3.3⁹¹. The results from both strategies confirmed a shared WGD event between S. spicata, P. latifolius and core grasses, known as ρ-WGD (see Results for detailed information). Therefore, subsequent analyses were mainly based on OrthoFinder results. Briefly, gene families of duplicated genes with two copies in collinear regions from all seven grass species and one copy from A. comosus were selected for gene tree inference. Gene families that failed to contain at least one gene copy from M. acuminata were excluded from the analysis. Consequently, 123 gene families were retained after filtering. Each gene family sequence and the supermatrix concatenated from all 123 gene families were aligned using MAFFT, and then FastTree v2.1.10¹⁰⁵ was used for maximum likelihood tree inference with default values. The timing of the ρ-WGD event was estimated using MCMCtree with the same parameters for estimating grass divergence times. The phylogenetic tree and the four-fold-degenerate site data derived from the concatenated 123 gene family alignments were used as input.

To investigate the fractionation bias of duplicated gene pairs following the ρ-WGD event, the combined S. spicata and rice genome sequences were aligned against the A. comosus genome using LAST in the Comparative Genomics (CoGe) platform (https://genomevolution.org/r/1i4tg). We merged neighboring syntenic blocks using the “Quota Align Merge” option, with the maximum distance between two blocks set to 20 genes. Syntenic depth was estimated with the “Quota Align” algorithm, setting the ratio of coverage depth for the combined S. spicata and rice genome to the A. comosus genome at 4:1. Fractionation bias and gene retention proportions were then calculated and visualized using FraBias¹⁰⁶ with a window size of 100 genes. To trace the history of gene loss after ρ-WGD, we conducted Polyploidy Orthology Inference Tool (POInT)³² analyses using S. spicata, P. latifolius, rice, and S. bicolor genomes, following the methodologies described by Emery et al.³³ and Hao et al.²⁰. Briefly, duplicated syntenic regions from the ρ-WGD event were organized into collinear gene blocks. Synteny information and a likelihood-based framework were used to infer duplicate gene relationships and reconstruct post-polyploidization gene retention and loss patterns.

Ancestral grass karyotype and chromosome evolution

We reconstructed the ancestral grass karyotype (AGK) following the ρ-WGD by comparing it with existing genomes to reveal chromosomal rearrangements across different subfamilies within the grasses. Two different strategies were employed to reconstruct the ancestral karyotypes (Supplementary Fig. 13). The first is a macro-synteny strategy using WGDI¹⁰⁷ with genomic data from only three species: S. spicata, P. latifolius, and rice, yielding the post-ρ AGK. Subsequently, the ante-ρ AGK was reconstructed by comparing this post-ρ AGK to itself and to several outgroup taxa (Jo. ascendens, Ju. effusus, T. angustifolia, A. comosus, and S. polyrhiza) to detect more ancient syntenic structures. Our reconstruction is broadly consistent with that of Murat et al.¹⁵, with one notable exception: the chromosome designated as Chr6 in their ante-ρ AGK model is split into two separate chromosomes in our version. Therefore, while we have largely retained their numbering system, we assigned the newly identified chromosome as number 7 and accordingly renumbered the previously designated chromosome 7 as chromosome 8.

The second strategy applied a bottom-up, micro-to-macro synteny approach to reconstruct the AGK. Initially, we performed a thorough survey of all sequenced genomes within the Poaceae to identify representative diploid species at the genus level with chromosome-level genome assemblies, resulting in a total of seven subfamilies, 24 genera, and 26 species (See detailed information in Supplementary Table 8). To identify syntenic orthologs (syntelogs) among diploid species, we performed pairwise alignments between all diploid genome pairs and used a clustering algorithm SynPan¹⁰⁸ for synteny-based gene family analysis to group approximately 1,093,721 genes from 26 genomes into 593,902 distinct syntelog groups (Supplementary Data 2). Of these, 55,124 non-private groups were shared by at least two species, each with at least one syntelog. The alignment of each non-private syntelog group was conducted by MAFFT, and the phylogenetic tree was reconstructed by TreeBeST (https://github.com/Ensembl/treebest). Utilizing AGORA¹⁰⁹, a parsimony-based algorithm, we reconstructed the gene content and order of the Poaceae common ancestor using non-private syntelog groups and the phylogenetic tree of 26 extant species. This resulted in 15,200 ordered ancestral genes (referred to as protogenes) within 358 continuous ancestral regions (CARs) (Supplementary Data 3). Based on the results of the macro-synteny strategy, we aligned all 358 CARs to the 12 protochromosomes. Syntenic blocks and breakpoints were identified between protochromosomes and extant species using PhylDiag¹¹⁰ and JCVI¹¹¹, defining 35 conserved ancestral genomic blocks (GBs) (Supplementary Data 4). Karyotypes of all 26 diploid species were built using these 35 GBs. Finally, by considering the phylogenetic topology and the karyotypes of 26 extant species representing seven grass subfamilies, we reconstructed the post- and ante-ρ AGK and traced the evolutionary trajectory within Poaceae by inferring ancestral karyotypes for all internal nodes. Chromosomal rearrangement events from post-ρ AGK to individual karyotype of 30 species of eight grass subfamilies were detected, with the additions of three more polyploid bamboos²¹ and Phragmites australis, representing the eighth subfamily Arundinoideae³⁵.

Identification of AP2 and MADS-box gene families and gene expression analysis

Putative gene sequences of AP2 (PF00847 domain) and MADS-box (PF00319 domain) gene families were detected using hmmsearch in HMMER v3.2.1¹¹² with parameters -E 1e⁻¹⁰ –domE 1e⁻¹⁰, by searching the protein sequences of S. spicata genome against the sequences of functionally known genes extracted from the Arabidopsis thaliana and Or. sativa genomes. To confirm the presence of AP2 and MADS-box domains in the predicted gene sequences, we used InterproScan v5.2-45.0¹¹³ to analyze and remove false-positive hits. The obtained protein sequences were aligned with MAFFT under default settings. Phylogenetic trees for AP2 and MADS-box alignments were reconstructed with FastTree¹¹⁴ using default settings. Subsequently, the collinear relationships of the chromosomal regions for these identified genes between different grass genomes were plotted and checked. MEME v5.0.5¹¹⁵ was employed to identify conserved motifs for the FZP-like genes, with the parameters of zero or one occurrence per sequence, maximum number of motifs = 18, minimum width = 6, and maximum width = 50.

RNA-seq data sequenced from root, stem, leaf, and different parts of the spikelet equivalent of S. spicata were analyzed using the nf-core/rnaseq pipeline built in Nextflow (https://github.com/nf-core/rnaseq)¹¹⁶. After quality trimming, alignments sorting and transcripts assembly, gene expression of all S. spicata genes from all samples was quantified. To evaluate the global transcriptomic variance and biological reproducibility among samples, principal component analysis (PCA) was performed using the prcomp function in R based on the normalized expression matrix.

Differentially expressed genes (DEGs) between bracts and leaves were identified using DESeq2¹¹⁷ (adjusted P < 0.05, |log₂ Fold Change| ≥ 1). GO enrichment analysis of these DEGs was then performed with clusterProfiler¹¹⁸.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

D.-Z.L., P.-F.M., and Y.-L.L. conceived and designed the project. P.-F.M., D.-Z.L., H.W., M.W., and S.-Y.G. collected and prepared the samples. Y.-L.L., S.-Y.G., C.G., and D.-J.F. performed bioinformatics analyses and analyzed data. E.A.K., A.D.T., and P.M.H. contributed valuable suggestions to the analyses and interpretation of results. Y.-L.L., P.-F.M., C.G., and S.-Y.G. wrote the paper with input from D.-J.F., A.D.T., P.M.H., E.A.K., and D.-Z.L. All authors read and approved the paper.

Peer review

Peer review information

Nature Communications thanks Clinton Whipple and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This study is supported by the National Key R&D Program of China (grant 2023YFF0805804), the 811 Project of Shandong Agricultural University (539604 to D.-Z.L. and Y.-L.L.), the Ten Thousand Talent Program of Yunnan Province (YNWR-QNBJ-2020-297 to P.-F.M.), a Special Grant for National Leading Talents from the Shandong Provincial S&T Department (2025GJJLJRC-104 to D.-Z.L.), and is partially supported by China’s National Wild Plant Germplasm Resource Center (NWPGRC).

Data availability

The raw sequencing data and genome assembly files generated in this study have been deposited in the NCBI database under BioProject accession PRJNA1393216. The genome assembly and annotation files are available from the National Genomics Data Center (NGDC) under accession GWHJIFL00000000.1 [https://ngdc.cncb.ac.cn/gwh/Assembly/129527/show] and Figshare [https://doi.org/10.6084/m9.figshare.32101309], respectively. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-76979-2.