Introduction

In the year 2000, the Arabidopsis thaliana L. genome became the first plant genome to be completely sequenced and assembled1. Since then, many genomes from the plant kingdom have been sequenced, e.g. green algae2,3, bryophytes4,5, ferns6, gymnosperms7,8 and angiosperms9,10. In the grass family (Poaceae) the reference assemblies were primarily obtained for crops11,12,13 and model plants14,15,16. The advent of second-generation sequencing and the subsequent decreasing of the overall sequencing costs have enabled the determination of whole genome sequences in many non-model plant species17,18,19,20.

Recently, the 1KP project that was aiming to sequence 1,000 green plant transcriptomes21,22,23 has been followed by the 10KP project24. The later initiative intends to sequence complete genomes from more than 10,000 plants and protists. The project is supposed to be completed in 2023 and it presumes to provide family-level high-quality reference genomes, ideally with chromosome-scale assemblies. Nevertheless, the data at the level of genera may not be processed immediately24. In comparison to other kingdoms, plants have very large genomes13,25,26, high ploidy level27 and the abundance of repetitive sequences28,29,30. Currently, to face these issues, the third-generation sequencing has been applied. The so-called single-molecule real-time (SMRT) sequencing provided by Pacific Biosciences (PacBio)31 and nanopore sequencing by Oxford Nanopore Technologies32 afford a range of benefits, including exceptionally long-read lengths (20 kb or more), resolving extremely repetitive and GC-rich regions and direct variant phasing32,33.

In the fossil record Stipa L., or a close relative genus, is known from about 34 Mya of the upper Eocene34,35. For many decades, Stipa has been described as a genus with over 300 species common in steppe zones of Eurasia, North Africa, Australia and the Americas36,37. According to the recent studies based on both morphological and molecular data, the genus has been reduced and currently includes over 150 species geographically confined to Europe, Asia and North Africa38,39,40,41,42. Most species of Stipa are dominants and/or subdominants in steppe plant communities43,44,45 and can be used for their classification46. Moreover, some species are of economic importance mainly as pasture and fodder plants, especially in the early phases of vegetation36,47, they can be used for soil remediation processes48,49, in studies related to climate change50,51,52 and as ornamental plants (e.g. S. capillata L., S. pulcherrima K. Koch, S. pennata L.).

In recent years, large-scale molecular data began to appear for Stipa: de novo transcriptome assemblies of S. purpurea Griseb.50,53, S. grandis P. A. Smirn.54 and S. lagascae Roem. & Schult.52, whole chloroplast genomes for 19 taxa57 and raw genomic data available via the NCBI Sequence Read Archive (SRA) for S. capillata58 and S. breviflora Griseb.59. In addition, nucleolar organising regions (NORs) were sequenced for six Stipa taxa60. Nevertheless, no complete or draft genome assembly currently exists for any Stipa species. In order to fill this gap, here we aim to: (1) present for the first time a single-molecule long-read dataset (nuclear, mitochondrial and chloroplast genomes) generated using the SMRT sequencing on the PacBio Sequel platform; (2) demonstrate and discuss the potential usage of this data in further studies of Stipa.

For the goals of the study we chose to sequence the entire genome of S. capillata (Fig. 1) as it is the most widespread taxon within the genus, growing on sandy to loamy, nutrient poor soils in the dry grasslands of Eurasia61. Currently, this species is increasingly attracting the interest of conservation biologists due to its large distribution range, common occurrence in the Eurasian steppes and pseudosteppes, a limited number of refugia in Europe and both great morphological and genetic variability within its range62,63,64.

Figure 1
figure 1

A representative individual of Stipa capillata.

Full size image

Results

Assembled nuclear genome

The SMRT sequencing yielded in 23.16-fold genome coverage consisting of 25.84 Gb sequence data with an N50 read length of 17,096 bp (Supplementary Table S1). De novo assembling of PacBio reads using Flye v.2.465,66 resulted in a genome size of 1,004 Mb67 with a contig N50 of 351 kb and a GC level of 45.97%. On the other hand, another de novo assembly performed with FALCON v.0.2.568 demonstrated a smaller genome size of 773 Mb with a GC level of 46.04%. However, the Flye assembly has a better N50 of 350,543 that is almost three times bigger than for FALCON. In case of applying Purge Haplotigs v1.1.169, the final genome size was reduced by 177 Mb with an N50 of 381,155 (Table 1) and a GC level of 45.82%.

Table 1 Statistics of the nuclear genome assemblies.
Full size table

The subsequent analysis based on a benchmark of 4,896 conserved genes belonging to the Poales order (dataset poales_odb10) revealed that the Flye assembly has 4,557 (93.10%) completed BUSCO (Benchmarking Universal Single-Copy) genes and only 293 (6%) missing BUSCOs versus 2,765 (56.50%) and 1,945 (39.70%) for the FALCON assembly. The Flye assembly after Purge Haplotigs shows 4,304 (87.90%) completed BUSCOs and 512 (10.50%) missing BUSCOs (Table 2).

Table 2 BUSCO statistics.
Full size table

Scaffolding of contigs

Nearly all contigs of S. capillata genome can be assigned to the reference chromosomes of Brachypodium distachyon L., Hordeum vulgare L. and Aegilops tauschii Coss., whereas genomes of Oryza sativa L. and especially Triticum aestivum L., have much less homology to the feathergrass assembly. In particular, 95.16% contigs of S. capillata genome were assigned to seven chromosomes of A. tauschii genome, 94.68% to five chromosomes of B. distachyon, 94.20% to seven chromosomes of H. vulgare, 89.92% to 12 chromosomes of O. sativa and only 41.67% to 21 chromosomes of T. aestivum. The total length of non-assigned contigs was reasonably low for A. tauschii (48.59 Mb), B. distachyon (53.40 Mb) and H. vulgare (58.17 Mb), whereas for O. sativa and T. aestivum it was about 101.15 Mb and 585.40 Mb, respectively (Table 3). In addition, the RaGOO grouping confidence and orientation confidence scores per chromosome ranged from 57.81 to 76.11% and from 80.03 to 95.11%, respectively, indicating that the contigs could be placed on a chromosome with an acceptable level of confidence (Supplementary Table S2). The only exception is T. aestivum for which scores ranged from 30.49 to 47.76% for the grouping confidence score and from 57.81 to 70.19% for the orientation confidence score. Nevertheless, based on the location confidence score, the exact position of the contigs on a chromosome could not be accurately estimated, reflecting a low level of synteny to the reference genomes. In particular, the score was in a range of 31.30–43.66% for O. sativa, 26.06–39.13% for B. distachyon, 19.56–31.41% for H. vulgare, 17.47–24.15% for A. tauschii and 10.30–38.23% for T. aestivum.

Table 3 RaGOO statistics.
Full size table

Transposable elements and nuclear genome annotation

Identification of transposable elements (TEs) revealed that more than half of the S. capillata genome (57.68%) is occupied by repetitive sequences. Particularly, retrotransposons represent at least 16.12% and transposons are reaching no less than 7.22% of the genome. Nonetheless, 34.34% of TEs are currently unclassified. Among classified repeats, long terminal repeats (LTRs) were the most abundant elements within retrotransposons, whereas Tourist/Harbinger elements were more common amid DNA-transposons. In total, 114,826 sequences were identified as simple repeats and occupy 0.57% of the genome. In addition, rolling-circles (0.28% of the genome) and low complexity sequences (0.11% of the genome) were found (Table 4).

Table 4 Statistics of repetitive elements.
Full size table

The subsequent structural annotation of the masked genome revealed 53,535 nuclear genes (Supplementary File 1). On the other hand, the unmasked genome has 154,755 structurally annotated genes and 94,237 of them have BLAST hits in the NCBI non-redundant database. Nonetheless, among the 94,237 genes of the unmasked genome, 12,094 sequences are related to transposable elements. In particular, 2,925 genes associated with transposons, and 9,859 assigned to retrotransposons. In addition, 229 genes encode transposase-related proteins. Thus, except transposable elements the unmasked genome has 81,224 genes that can be associated with already known proteins (Supplementary File 2).

SSR markers

In total, 77,614 perfect repeat motifs were identified for the nuclear genome assembly using Krait75 (Supplementary File 3). Within those, di- and tri-nucleotides were the most common types, accounting 28,365 (36.55%) and 25,794 (33.23%) repeats, respectively. Tetra-nucleotide motifs were the third most abundant repeats with 9,777 SSRs (12.60%), followed by mono-nucleotides with 6,572 SSRs (8.47%) and penta-nucleotides with 4,629 SSRs (5.96%). Hexa-nucleotides were the rarest motifs with 2,477 SSRs (3.19%). Only four mono-nucleotide, four di-nucleotide and three tetra-nucleotide motifs were found in the mitochondrial and chloroplast genomes. However, a total length of those SSRs was in a range of 12–16 bp. In addition, in total 58 unique repeats present only in a single copy in a range 101–325 bp were retrieved from the analysis of TEs. Within those were four hexa-, 35 hepta-, nine octa-, five nona- and five deca- nucleotide motifs (Supplementary Table S3).

Divergence time of Stipa

The Bayesian phylogenetic reconstruction based on the five loci within NORs revealed the divergence time of Stipa from Brachypodium around 30.00–35.52 Mya and the putative origin of feather grasses about 2.90–6.02 Mya (Fig. 2). Although not all branches were well supported within the genus, the current analysis confirmed the monophyly of Stipa and the general grouping of the analysed species regarding their taxonomic positions. In particular, S. capillata and S. grandis represent the section Leiostipa Dumort; S. magnifica Junge, S. narynica Nobis, S. lipskyi Roshev. and S. caucasica Schmalh. belong to the section Smirnovia Tzvelev. The remaining three groups include (1) S. orientalis Trin. and S. pennata L., (2) S. richteriana Kar. & Kir., S. lessingiana Trin. & Rupr., S. heptapotamica Golosk. and S. korshinskyi Roshev, (3) S. lagascae and S. breviflora currently have a discrepancy between morphological and molecular data. In addition, the divergence time estimation indicates that the potential origin of the clade comprising S. capillata and S. grandis is in a range of 0.67–2.93 Mya while the sister clade has the 95% credibility intervals for that parameter in a range of 2.38–4.78 Mya. Furthermore, the lowest genetic divergence time was registered for S. lessingiana and S. richteriana (0.00–0.48 Mya) as well as for the split between S. heptapotamica and the two above-mentioned species (0.01–0.78 Mya). The divergence times for the rest of taxa are present in Table 5.

Figure 2
figure 2

Phylogeny and divergence time estimation by molecular clock analysis. Letters at each node refer to Table 5. Numbers in brackets represent the Bayesian posterior probabilities (BPP > 0.50 only). The blue rectangles on the nodes indicate the 95% credibility intervals (CI) of the estimated posterior distributions of the divergence times. The red circles indicate the presumed divergence time splits set as a reference. The scale on the bottom shows divergence time in Mya. The figure was created using Figtree v1.4.4, https://tree.bio.ed.ac.uk/software/figtree/.

Full size image
Table 5 Node ages, BPP and CI related to Fig. 2.
Full size table

Assembled mitochondrial and chloroplast genomes

The resulting Flye assembly contained four mitochondrial contigs with a total length of 438,037 bp76,77,78,79 represented by six edges and an entire 137,832 bp-long circular chloroplast genome combining a long single copy region (LSC) of 81,710 bp, a short single copy region (SSC) of 12,836 bp and two inverted repeats (IR) of 21,643 bp each (Fig. 3). However, after a manual checking in IGV v.2.8.680 the final size of the chloroplast genome was slightly reduced to 137,823 bp. In addition, an analysis using Cp-hap81 detected two structural haplotypes of the chloroplast genome: haplotype A82 (LSC—IR, reverse-complement (rc)—SSCrc—IR) and haplotype B83 (LSC—IRrc—SSC—IR). We also obtained one assembly using Unicycler v.0.4.884 resulted in 76 linear contigs from which 29 can be assigned to mitochondrial sequences with a total length of 1,668,569 bp. Due to the Unicycler assembly being more complex and none of the obtained contigs were likely to be circular in nature, for the downstream genome annotation we used the Flye assembly.

Figure 3
figure 3

Visualisation of the de novo mitochondrial and chloroplast genome assemblies using Bandage v.0.8.185. (a) Contigs representing mitochondrion. (b) Contig representing chloroplast. Different colours represent different contigs; length (in bp) and coverage (x) of edges within contigs are shown. The figure was created using Bandage v.0.8.1, https://rrwick.github.io/Bandage/.

Full size image

In total, 112 and 133 genes were functionally annotated for mitochondrial and chloroplast genomes, respectively. The mitochondrial annotation resulted in 78 protein-coding genes, 4 ribosomal RNA genes and 30 tRNA genes. The chloroplast annotation contained 85 protein-coding genes, 8 ribosomal RNA genes and 40 tRNA genes. The chloroplast genome size of 137,823 bp generated with Flye and the number of annotated genes in the current study were similar to the known assemblies for S. capillata obtained by Illumina sequencing57. However, the previous genome assemblies were slightly longer, specifically 137,830 bp86 and 137,835 bp87.

DArTseq markers

The DArT pipeline analysis resulted in 61,328 Silico markers and in 52,970 sequences with SNPs. The BLAST process revealed 58,701 Silico markers and 52,252 sequences with SNPs that were successfully mapped to 4,361 and 3,935 genome contigs, respectively. Thus, the current genome assembly has 95.72% of Silico markers and 98.64% of sequences with SNPs that are represented in 73.52% (the total length of 969.30 Mb) and 66.34% (940.37 Mb) of the contigs, respectively. In addition, we established that 50,953 Silico markers and 47,181 sequences with SNPs were present only in a single copy in the genome. Finally, we identified 30 Silico markers and 10 sequences with SNPs aligned to the mitochondrial genome and only 2 Silico markers and 4 sequences with SNPs that were found in the chloroplast genome.

Discussion

The number of sequenced plant genomes is rapidly increasing year by year serving as a fundamental resource for various genomic studies. In the current work, we present a 1004 Mb genome with the 23 × coverage of the most widespread feather grass species, S. capillata, using SMRT PacBio sequencing. The current assembly comprises 5,931 sequences with a contig N50 length of 351 kb (Table 1). The BUSCO completeness score of 93.10% (Table 2), the observation of a large portion of TEs (57.68%, Table 4) and the presence of Silico (95.72%) and SNPs (98.64%) markers derived from the DArT platform indicate that the assembly is of high quality. Moreover, the proportion of TEs has been reported for the first time in the genus due to the previous de novo assemblies which were performed exclusively based on transcriptomic data50,52,54. In addition, here we also attempted to perform a reference-guided scaffolding of the assembled contigs. Nevertheless, although nearly all contigs of the S. capillata genome were assigned to the chromosomes of B. distachyon, H. vulgare and A. tauschii, it was not possible to estimate their proper position on the reference with an acceptable level of confidence (Table 3 and Supplementary Table S2). In general, in the absence of a high-density genetic linkage map the task of reconstructing pseudomolecules of chromosomes seems to be challenging. On the other hand, we believe that in order to improve the contiguity of the long-read assembly the high-throughput chromosome conformation capture (Hi-C)88 technique should be applied. Currently, many studies on non-model species successfully utilised a combination of long-read techniques and Hi-C data to perform assemblies at chromosome scale89,90,91. Moreover, an additional key for improving this genome assembly in the future is merely to get more sequencing reads. Recently, it was shown that contig length metrics are positively correlated with both read length and sequence coverage. Specifically, long-read assemblies in maize demonstrated that the highest contig N50 of 24.54 Mb was reached with a subread N50 of 21,166 bp and a 75-fold depth of coverage while the longest contig of 79.68 Mb was observed with the same subread N50 but with a 60-fold depth92.

The newly generated genome has a GC content of 45.97% that is similar to the known estimates for species in Stipa varying in a range of 46.61–49.05%93, and more broadly to grasses ranging from 43.57% in O. sativa to 46.90% in Z. mays94. Recently, it was shown that a higher GC content in monocots is associated with adaptation to extremely cold and/or dry climates95. The genus Stipa highly supports this hypothesis due to the fact that all feather grasses are adapted to temperate, dry climates36. In addition, a positive correlation between the GC content and genome size was established96 suggesting insertion of LTR retrotransposons as a potential driving force of genome enlargement97. Similarly, here we showed that the expansion of the S. capillata genome also resulted from insertions of repetitive sequences that occupy 57.68% of the genome including LTR retrotransposons (13.97%). However, among all repetitive sequences around 34.34% are currently unclassified (Table 4). Nonetheless, the total proportion of TEs in S. capillata in comparison to other species within the Poaceae family is close to Oryza minuta J. Presl (58.35%) and O. alta Swallen98 (57.54%), bigger than in B. distachyon99 (28.10%) and O. sativa100 (45.52%) and smaller than in O. granulata Nees & Arn.101 (67.96%), Avena sativa L.102 (69.47%) and T. aestivum103 (84.67%).

Importantly, the presented genome size is roughly twice smaller than the expected size of 2,355 Mb and twice bigger than the expected monoploid size of 589 Mb estimated using flow cytometry93. Considering that we were unable to remove redundant sequences due to possible heterozygosity and the number of duplicated BUSCOs (Tables 1 and 2), it may be presumed that the current genome assembly combines two very distinct genomes. To the current knowledge, the vast majority of Stipa species have 44 (2n = 4x) chromosomes and are supposed to be tetraploids41,104. In addition, recently it was shown that a single-copy region ACC1 and a low-copy nuclear gene At103 have two different copies in Stipa104,105. Thus, it may suggest that S. capillata, and the genus Stipa in general, has arisen through hybridisation between genetically distant diploid species (2n = 22) and the subsequent allopolyploidisation via whole genome duplication (WGD) rather than via one WGD event of an ancestral species. Well-documented examples of natural allopolyploid taxa in the Pooideae subfamily are Triticum turgidum L. (2n = 4x = 28, genome constitution AABB) and T. aestivum (2n = 6x = 42, AABBDD) formed through hybridisation and successive chromosome doubling of ancestral diploid species T. urartu (2n = 2x = 14, AA), Aegilops speltoides Tausch. (2n = 2x = 14, BB) and A. tauschii (2n = 2x = 14, DD)106. Moreover, in the tribe Stipeae based on the At103 gene allopolyploidy was reported for the genus Patis Ohwi (2n = 46, 48)105. Heretofore, at least three hypotheses were considered regarding the base chromosome number in Stipeae: x = 7107, x = 11108,109 and x = 12110. Recently, it was suggested that the latter two are more plausible41,104. Thus, in order to better assemble the S. capillata genome and verify if Stipa is an allopolyploid genus we suggest sequencing at chromosome level the close relative diploid species (2n = 22) from genera representing, e.g. Ptilagrostis Griseb., Achnatherum P. Beauv., e.g. A. calamagrostis L. (2n = 22 + 0‒2B), or Piptatheropsis Romasch., P. M. Peterson & Soreng (2n = 20, 22, 24)41,104.

In general, the number of genes in Poaceae varies from 28,835 in the smallest known genome, Oropetium thomaeum Trin. (2n = 20; genome size of 245 Mb)111, to 107,891 in T. aestivum (2n = 42; 14,547 Mb)112. Here, we reported 53,535 nuclear genes that were structurally annotated for the masked genome assembly. Such a number of genes was roughly 1.8 and 1.6 times smaller than previously determined for S. grandis (94,674 genes)54 and S. purpurea (84,298 genes)50, respectively. On the other hand, the annotation analysis of the unmasked genome resulted in 81,224 genes associated with already known proteins. In comparison, only 65,047 functionally annotated genes were reported for S. grandis while S. purpurea had 58,966. Nonetheless, as RNA-seq data is currently unavailable for S. capillata, we believe that the current version of the genome annotation demands a further investigation to properly characterise the genes sets when the appropriate information will be available.

SSR markers are widely distributed across the genome and they are commonly applied in establishing genetic structure in Stipa. Previously, polymorphic microsatellite primers were reported in populations of S. purpurea (11113, 15114 and 29115 loci), S. pennata (7 loci116), S. breviflora (21 loci117) and S. glareosa (9 loci118). In the present study, we identified 77,614 perfect SSR markers (Supplementary File 3) and 58 imperfect repeat motifs presented only in a single copy (Supplementary Table S3). Although we did not test them on the population level we are confident that such a number of new loci will be a valuable source for the farther development of SSR markers in S. capillata, and more generally in the genus Stipa. Additionally, the revealed loci could be used for the designing dominant inter simple sequence repeat (ISSR) markers119. Recently, the usefulness of applying ISSRs were shown for studies in S. bungeana120, S. ucrainica and S. zalesskii121, S. tenacissima122 and the hybrid complex S. heptapotamica123.

According to the previous studies, based on three chloroplast loci124 and four chloroplast loci and one nuclear region105, it was shown that the origin of Stipeae can be estimated in a range of 30.60–47.30 Mya and 21.20–39 Mya, respectively. Here, based on the five loci within NORs we demonstrated that the potential split between Stipa representing the tribe Stipeae and Brachypodium (the tribe Brachypodieae) took place approximately 30–35.52 Mya that supports the previous findings105,124,125. The present results also suggest that the genus Stipa likely originated ca. 4.39 (2.90–6.02) Mya. On the other hand, one previous study indicated the origin of feather grasses at about 12.90 Mya124 while another one showed different estimates based on chloroplast loci (21.20 Mya, 13–22) and the At103 region105. Specifically, two copies of At103 had the following suggested ages: 15.78 (6.30–26.60) Mya for the Eurasian Stipeae lineage and 5.62 (0–6.50) Mya for the American Stipeae lineage105. Thus, the latter estimate is close enough to the origin-age calculated in the current study. In addition, our data on the divergence time among S. richteriana, S. lessingiana and S. heptapotamica (Fig. 2 and Table 5) conforms to the previous findings on the ongoing hybridisation among these taxa123 suggesting NORs as a useful tool for revealing species of putative hybrid origin. Nonetheless, we believe that the current and previous estimates regarding the origin of Stipa should be treated with caution. Firstly, to our knowledge, there is still no available fossil data for any Stipa species from the Old World that can properly calibrate the historical diversification in the genus. Currently, the earliest definite Stipa caryopses were found in central Poland and are dated ca. 4,000 BC126. Secondly, available data demonstrate incongruence between chloroplast and nuclear loci analyses. In further studies we suggest utilising single-copy nuclear genes derived from whole genome sequencing projects. Thirdly, different sets of species and parameters used for inferring diversification dates may result in different estimates127.

Finally, we report a 137,823 bp chloroplast genome that is similar to the known assemblies in Stipa and specifically in S. capillata57. Here we highlight the applicability of a long-read sequencing technology like PacBio for the straightforward assembling of plastomes using Flye67,68. In addition, due to the long-reads we were able to identify two haplotypes presented in S. capillata. This result supports the previous findings in Poaceae81 suggesting that plastome structural heteroplasmy can be a common state in feather grasses. Moreover, for the first time in the genus Stipa, here we present a 438,037 bp mitochondrial genome. The current size of this genome is close to Alloteropsis semialata (R.Br.) Hitchc. (442,063 bp)128, T. aestivum (452,526 bp)129, Sorghum bicolor L. (468,628 bp)130 and A. speltoides (476,091 bp)131. Nevertheless, the present version of the genome is constituted by four contigs rather than one circular sequence. Although the general acceptance among mitochondrial biologists is that plant mitochondrial genomes have a variety of configurations132,133,134, in order to verify if a more accurate assembly could be performed, we suggest reusing our data for a more comprehensive analysis of the mitochondrial structures within Stipa.

Materials and methods

Plant material and DNA extraction

Our research complies with relevant institutional, national, and international guidelines and legislation. A S. capillata sample from Kochkor River Valley, central Kyrgyzstan (Supplementary Table S4), was selected for genome sequencing. The sample was stored in silica gel at ambient temperature until DNA extraction was performed. Total genomic DNA was isolated from dried leaves after a six-month storage period using a CTAB large-scale DNA extraction protocol (Supplementary information S1, described in Supplementary File 6). DNA extraction was performed by SNPsaurus (USA). In addition, we isolated DNA from dried leaves using a Genomic Mini AX Plant Kit (A&A Biotechnology, Poland). Subsequently, quality check, quantification and concentration adjustment were accomplished using a NanoDrop One (Thermo Scientific, USA) and agarose gel electrophoresis visualisation. The concentration of the sample was adjusted to 50 ng/μL. The purified DNA sample (1 μg) was sent to Diversity Arrays Technology Pty Ltd (Canberra, Australia) for sequencing and DArT marker identification. Moreover, to test the phylogenetic power of NORs in Stipa, we supplemented the study with five specimens of S. richteriana Kar. & Kir, three of S. lessingiana Trin. & Rupr., four of S. heptapotamica Golosk. and four of S. korshinskyi Roshev. (Supplementary Table S4). The isolation of genomic DNA was performed from dried leaf tissues using a modified CTAB method135.

Library construction and sequencing

In total, 5 ug of S. capillata genomic DNA were used to construct a PacBio library according to the 20 kb PacBio template preparation protocol omitting a shearing step. The size selection cut-off was set at 15 kb. The library preparation followed by sequencing on three PacBio Sequel SMRT cells (Pacific Biosciences, Menlo Park, CA, USA) was carried out by SNPsaurus, LLC. Prior to the assembly, reads from each SMRT cell were inspected and quality metrics were calculated using SequelQC v.1.1.0136. A high-density assay using the DArT complexity reduction method for S. capillata was performed according to a previously reported procedure137.

For the rest of the specimens used in the current study, the quality control using a fluorometer (PerkinElmer Victor3, USA) and gel electrophoresis, library construction using a TruSeq Nano DNA Library kit (350 bp insert size; Illumina, USA) and sequencing using 100 bp paired-end reads on an Illumina HiSeq 2500 platform (Illumina, USA) were performed by Macrogen Inc. (South Korea).

Nuclear genome assembly and validation

The execution of this work involved using many software tools, whose versions, settings and parameters are described in Supplementary information S2 (available in Supplementary File 6). The de novo assembly of the PacBio data was performed using Flye v.2.465,66. The draft assembly was cleaned by running BLASTn v.2.10.0138 against the NCBI nucleotide database v.5, and subsequently sending each BLAST hit to the JGI taxonomy server (https://taxonomy.jgi-psf.org/) with a downstream step of keeping only plant contigs. Thereafter, Qualimap v.2.2.2139 was used to identify mean coverage for each contig. In the final assembly we kept only contigs with an average coverage of more than 10x. In addition, overrepresented contigs (> 60x) were BLASTed against the NCBI nucleotide database v.5 and sequences assigned to chloroplasts and mitochondria were removed.

Due to the final assembly performed with Flye v.2.4 being roughly twice bigger than an expected monoploid genome size of 589 Mb93, we accomplished an additional assembly with FALCON v.0.2.568 and applied Purge Haplotigs v1.1.169 to filter redundant sequences due to possible heterozygosity. The assemblies' statistics were analysed using assembly-stats v.1.0.1140. In addition, in order to assess the completeness of the genome assemblies, we investigated the presence of highly conserved orthologous genes using BUSCO v.4.0.6141.

Scaffolding of contigs

Due to there being no reference genome for any Stipa species, here we applied RaGOO v.1.1142 to verify if a reference-guided scaffolding can be performed for the draft genome contigs based on four genomes from the Pooideae subfamily (B. distachyon70, H. vulgare71, A. tauschii72, T. aestivum74) and one genome from the Oryzoideae subfamily (O. sativa73). The subsequent assessment of the scaffolding accuracy was based on three parameters: (1) location confidence score, (2) orientation confidence score and (3) grouping confidence score142.

Repeat prediction and nuclear genome annotation

The repeat prediction for S. capillata was performed using a de novo transposable element (TE) family identification and modeling package RepeatModeler v.2.0.1143 which includes three repeat finding programs; RECON144, RepeatScout145, and TRF146. The resulting TE library was supplemented by the transposable elements database (Release 19, http://botserv2.uzh.ch/kelldata/trep-db/).147 Subsequently, the genome assembly was masked for TEs regions by RepeatMasker v.4.1.0148 (http://repeatmasker.org) with the search engine RMBlast v.2.9.0 + 149 and the custom library created in the previous step. Next, gene and protein sequences were predicted using Augustus v.3.2.3 with the unmasked and v.3.3.3150 with the masked genome assemblies. The predicted protein sequences of the unmasked assembly were then BLASTed against the NCBI protein database v.5 and the subsequent BLAST hit descriptions were added to GFF (General Feature Format) files.

Genome-wide identification of microsatellite markers

The unmasked nuclear genome, chloroplast and mitochondrial genome assemblies were screened for perfect mono-, di-, tri-, tetra-, penta- and hexa-nucleotide repeat motifs using Krait v.1.3.375. We applied the following criteria: mono-nucleotide repeat motifs contain at least 12 repeats, di-nucleotide repeat motifs contain at least seven repeats, tri-nucleotide repeat motifs contain at least five repeats, tetra-, penta- and hexa-nucleotide repeat motifs contain at least four repeats.

Divergence time of Stipa

In order to estimate the divergence between S. capillata and other Stipa species we used the nucleolar organising regions. Firstly, we prepared a set of reference sequences including S. lipskyi Roshev.151, S. magnifica Junge152, S. narynica Nobis153, S. caucasica Schmalh.154, S. orientalis Trin.155 and S. pennata L.156. Secondly, we mapped raw reads of S. capillata, S. richteriana, S. lessingiana, S. heptapotamica and S. korshinskyi (Supplementary Table S2) as well as S. grandis55, S. breviflora59, S. lagascae157 to the reference set using Minimap2 v.2.17-r941158 with keeping only uniquely mapped reads by Samtools v.1.9159. Thirdly, the de novo assembly of the NORs was performed using Canu v.2.0160 for S. capillata and SPAdes v.3.14.1161 for the rest of Stipa species. Additionally, we added to the analysis B. distachyon162 as an ingroup member of the Pooideae subfamily and O. sativa163 as an outgroup representing the Oryzoideae subfamily within the Poaceae family. Next, all sequences were aligned using MAFFT v.7.471164. Subsequently, the aligned sequences were visualised in AliView v.1.26165 and divided in five loci: (1) 18S ribosomal RNA, (2) Internal Transcribed Spacer 1 (ITS1), (3) 5.8S ribosomal RNA, (4) Internal Transcribed Spacer 2 (ITS2) and (5) 26S ribosomal RNA (Supplementary File 4). Estimation of divergence times was performed in BEAST2 v.2.6.3166 using the 121,321 substitution model determined by bModelTest167. We used the following constraints for time calibrations: 38–48 million years ago (Mya) for the Brachypodium-Oryza split101 and 33–39 Mya for the potential origin and divergence of Stipa34,35. Then, the divergence time was estimated using the strict clock model and the Yule prior. In total, we ran the analysis three times independently, 50 million Markov chain Monte Carlo (MCMC) generations for each run. The log and tree files were combined using LogCombiner v.2.6.3 (a part of the BEAST package) with the first five million generations discarded as burn-in from each run. Next, Tracer v.1.7.1168 was used to check the log files regarding Effective Sample Size (ESS) values. As all ESSs exceeded 200, we summarised the final maximum clade credibility tree (Supplementary File 5) in TreeAnnotator v.2.6.3 (a part of the BEAST package). The final tree was visualised and edited using FigTree v.1.4.4169.

Mitochondrial and chloroplast genomes assembly, annotation and validation

Prior to assembly, we mapped raw reads to 11 reference mitochondrial genomes of species belonging to the Poaceae family (Supplementary Table S5) using Minimap2 v.2.17-r941158. Only uniquely mapped reads were kept by Samtools v.1.9159 for the next step. De novo mitochondrial assembly of the 4.08 Mb data was performed using Flye v.2.7.1-b1590.

In the next step, we BLASTed the resulting contigs against the NCBI nucleotide database v.5, and sequences assigned to mitochondria were kept. Then, the PacBio subreads were mapped onto the kept contigs using Minimap2, and only uniquely mapped reads were retained by Samtools. A new de novo assembly of the 15.51 Mb data was performed using Flye. In order to check if the mitochondrial contigs obtained by Flye could be merged into larger scaffolds we applied Circlator v.1.5.5170. However, the resulting sequences were identical to the Flye contigs. In addition, we used Unicycler v.0.4.884 with reads that were mapped onto the Flye contigs as a reference.

Further, to detect all possible structural haplotypes of the chloroplast genome we applied Cp-hap81. Next, we mapped raw reads onto the resulting mitochondrial contigs and the chloroplast genomes to manually check in IGV v.2.8.680 if any potential SNPs or indels are present. Eventually, annotations of the final mitochondrial contigs of 438,037 bp and the chloroplast genomes of 137,823 bp were performed using Geneious Prime v.2021.1.1 (https://www.geneious.com) based on 85% and 95% similarities to the reference genomes of mitochondria and chloroplasts, respectively (Supplementary Table S5).

In Silico mapping of DArT marker sequences

Since the DArT markers are designed to target active regions of the genome171, here we use them to validate the completeness of the nuclear genome assembly and improve the accuracy of data filtering in further genomic studies on Stipa. Two data types, Silico and SNPs markers, were mapped to the nuclear genome using BLASTn v.2.10.0. As a query we used trimmed DArT sequences in a range of 29–69 bp with the percent identity values to the reference genome of 95% or greater and removing alignments below 95% of a query.