Genome assembly and annotation of L. panicea

Based on k-mer frequency analysis, the L. panicea genome had an estimated size of about 286.51 Mb and a heterozygosity level of 0.047% (Figure S2). The estimated genome size was close to the 280 Mb determined using flow cytometry. We generated a total of 32.73 Gb (133×) PacBio HiFi sequences with a read N50 length of 16.60 kb (Table S1, Figure S3). These HiFi reads were de novo assembled into 348 contigs with an N50 size of 21.96 Mb (Table S2). Using Hi-C data of approximately 97× coverage (Table S1), a total of 235.12 Mb of assembled contigs were clustered into 10 pseudomolecules with sizes ranging from 15.55 Mb to 32.52 Mb (Figures 1a, S4), of which 7 and 3 had telomeric sequences (5′-TTTAGGG-3′) at both and single ends, respectively (Table S3). Evaluation using BUSCO (Simão et al., 2015) indicated that 98.7% of plant conserved orthologs were fully captured by the L. panicea assembly (Table S4). Illumina paired-end reads were mapped back to the assembly, resulting in an overall alignment rate of 99.23%. The LAI score of the assembly was 17.02 (Figure S5). Together, these results suggested the high quality of the L. panicea genome assembly (Table 1).

A total of 33 481 gene models were predicted in the L. panicea genome, of which about 99.36% (33267) were assigned to the 10 chromosomes (Table S5). In addition, 2751 tRNA genes, 8114 rRNA genes, 114 microRNA genes (miRNAs), 95 small nuclear RNA genes (snRNAs) and 362 small nucleolar RNA genes (snoRNAs) were predicted in the L. panicea genome (Table 1). BUSCO assessment indicated that 99.2% of plant conserved orthologs were completely covered by the predicted genes (Table S4).

Among the 33 481 predicted proteins, 30 842 (92.12%) were annotated by GenBank nr, 28 957 (86.49%) by eggNOG mapper, 20 638 (61.64%) by Swissprot, 11 023 (32.95%) by the KEGG database and 13 780 (41.16%) were assigned with Gene Ontology (GO) terms. The average GC content of the CDSs of L. panicea (56.06%) was similar to that of L. chinensis (56.90%), O. sativa (55.19%), Sorghum bicolor (55.51%) and Setaria viridis (56.01%), but higher than that of Arabidopsis thaliana (44.17%) (Table S6). The GC content and GC3s (GC of silent 3rd codon position) of CDSs showed a bimodal distribution in L. panicea, consistent with those found in other grasses (Figures S6, S7).

In the L. panicea genome assembly, we identified 66.24 Mb (28.21%) repetitive sequences (Table S7). The long-terminal repeat retrotransposon (LTR-RT) was the most abundant type of repetitive sequences in L. panicea, spanning 21.11 Mb (8.98%) of the genome. A total of 226 intact LTR-RTs were classified, including 129 Gypsy-type and 74 Copia-type LTRs. The largest LTR-RT superfamily Gypsy, comprising ~6.34% of the genome, was concentrated near the putative centromeres (Figure S8A). The other superfamily of LTR-RT, Copia, comprised ~1.37% of the genome (Figure S8B). Other interspersed repeats such as LINEs (Long Interspersed Nuclear Elements) and SINEs (Short Interspersed Nuclear Elements) occupied 1.86% and 0.08% of the genome, respectively (Figure S8C,D).

In addition, centromere regions were identified in nine of the 10 L. panicea chromosomes (except chromosome 6) using Tandem Repeat Finder (Benson, 1999) (Figure S9). Furthermore, we detected the top eight TE subfamilies, including four LTR/Gypsy, two unknown, one LINE/L1 and one LTR/cassandra type subfamilies, which together comprised over 10.65% of the L. panicea genome (Figure S10). Density of these TE subfamilies along the 10 chromosomes showed that only the longest unknown-type rnd-3_family-356 subfamily (6.44% of the genome) was enriched near centromeres but absent from the rest of the genome. These results are consistent with results from Tandem Repeat Finder and complement the identification of the centromere position on chromosome 6 (Figure S9). Overall, we predicted potential centromeric regions on all 10 chromosomes.

Comparison between the L. panicea and L. chinensis genomes

The high-quality genomes of L. panicea provides an opportunity to compare the two subgenomes (At and Bt) of L. chinensis (Wang et al., 2022) with the L. panicea genome to evaluate the effect of polyploidization in Leptochloa weeds. A total of 174.9 Mb syntenic regions between L. panicea and At, and 165.85 Mb between L. panicea and B_t were identified (Figure 1b, Table S8). Every chromosome of L. panicea was basically collinear with two homologous chromosomes of L. chinensis (Figure 1c), consistent with the diploid and tetraploid nature of L. panicea and L. chinensis, respectively. And each chromosome of L. panicea was basically collinear with one chromosome of O. thomaeum, a diploid species in Chloridoideae (Figure 1d).

SVs in gene body and promoter regions can impact gene functions and expression. We found that 6402 predicted genes in L. panicea and 5909 in At had at least one indel in their gene coding sequences or promoter regions, with 3792 in L. panicea and 3333 in At having indels in their coding sequences (Figure S11). In addition, 6144 predicted genes in L. panicea and 5662 in Bt had at least one indel in their gene body or promoter regions, with 3648 in L. panicea and 3271 in Bt having indels in their coding sequences (Figure S12). To explore whether there is a subgenome bias for genomic variation (SV), we found that 175 033 SVs existed only between L. panicea and At, 162162 SVs existed only between L. panicea and Bt, and 53 046 SVs shared by both At and Bt (Figure 2a). Moreover, we found that 9.09% of these SV sequences between L. panicea and At and 6.99% of these SV sequences between L. panicea and Bt were gypsy-like retrotransposons, compared to 6.34% of the entire genome, and 2.10% of these SV sequences between L. panicea and At and 1.74% of these SV sequences between L. panicea and Bt were copia-like retrotransposons, compared to 1.37% of the entire genome. The contents of other types of transposable elements were similar between the SV regions and the entire genome (Figure 2b), suggesting that SVs occurred more frequently in genome regions occupied by gypsy and copia-like retrotransposons.

Chromosomal rearrangements such as inversions and translocations have long been thought to play a critical role in adaptation and speciation (Dvorak et al., 2018). We identified 128 genome rearrangements (translocations and inversions) between L. panicea and At and 138 between L. panicea and Bt distributing across all 10 chromosomes (Figure S13). Among this, 77 inversions ranging from 2.11 Kb to 4.94 Mb were identified between L. panicea and At, and 74 inversions ranging from 2.03 Kb to 4.94 Mb were identified between L. panicea and Bt. Interestingly, we found two large inversions on chromosomes 2 and 9 shared by both At and Bt (Figures 2c, S13), which were further supported by Hi-C maps and/or Illumina read mapping (Figures 2c, S14–S17). GO term enrichment analysis of the 200 genes surrounding the breakpoint regions of chromosome 2 showed that biological processes such as ‘heterocycle metabolic process’, ‘DNA recombination’, ‘nucleic acid metabolic process’ and ‘DNA integration’ were significantly enriched (Figure 2d), and genes surrounding the breakpoint regions of chromosome 9 were mainly enriched with GO terms related to plant defence, including ‘defense response to fungus’ and ‘defense response to other organism’ (Figure 2d), which suggests that this inversion may mediate the difference in biotic stress tolerance between the two species.

Currently, it is unclear whether L. chinensis is an autotetraploid or an allotetraploid. In our previous study, we found that the two subgenomes of L. chinensis display neither fractionation bias nor overall gene expression dominance, suggesting a possible autopolyploid of L. chinensis (Wang et al., 2022). To further explore this issue, we broke the L. panicea genome into 100-mer fragments and mapped these 100-mers to the two subgenomes of L. chinensis. We found that two subgenomes showed substantially similar genomic similarity with the L. panicea genome (Figure S18), supporting an autopolyploid origin of L. chinensis.

Genome evolutionary of Leptochloa weeds

To investigate the genome evolutionary history of Leptochloa weeds, we first calculated Ks values for homologous gene pairs to identify the whole-genome duplication (WGD) events. According to the Ks peak, we dated the divergence time of L. panicea and L. chinensis at 11.6 million years ago (mya), and the two subgenomes of L. chinensis were separated from L. panicea simultaneously (Figure 3a). We found that L. panicea only experienced one ancient WGD event (ρ) shared by members of the grass family, while L. chinensis experienced an additional WGD event recently (the tetraploidization event), dating back to ~9.88 mya. The results indicated that the time of L. chinensis tetraploidization is very close to the divergence time of L. panicea and L. chinensis, and it is possible that this tetraploidization event had led to the L. chinensis speciation.

To further investigate the evolutionary relationships between Leptochloa weeds and other grasses, gene family clustering was carried out using Leptochloa weeds, seven other gramineous plants and A. thaliana. A total of 549 single-copy orthologs shared by these plants were identified and used for phylogenetic reconstruction and species divergence time estimation, which showed that O. thomaeum was the closest relative to the Leptochloa weeds, and the divergence of Leptochloa weeds and O. sativa was estimated to occur ~49.3 mya (Figure 3b).

In addition, we identified 21 086 shared gene families among Leptochloa weeds. A total of 901 gene families containing 2383 genes were unique in L. panicea, and 1479 gene families containing 2528 genes were in L. chinensis (Figure 3c). We found that for both shared and specific genes, the highest number of them were expressed in seeds, followed by roots, stems and leaves in L. chinensis. In L. panicea, on the contrary, the highest number of genes were expressed in seeds and stems, followed by roots and leaves (Figure 3d).

To detect the expansion and contraction of gene families, we used protein sequences of the aforementioned 10 species identified 30 737 orthologous groups (gene families). Through comparing gene families among these 10 species, 3432 gene families were found to be significantly expanded in L. chinensis and 3489 to be significantly expanded in L. panicea (Figure 3b). Functional analysis of the 3432 expanded gene families in L. chinensis revealed that a large number of them were involved in regulating various metabolic pathways such as ‘regulation of metabolic process’ (Figure 3e). Notably, metabolism is one of the main mechanisms of herbicide nontarget-site resistance (Gaines et al., 2020; Powles and Yu, 2010). However, functional analysis of the expanded gene families of L. panicea revealed no enrichment of such functions (Figure 3f). Next, we performed domain annotations of these 3432 gene families expanded in L. chinensis and identified 35 cytochrome P450, 25 ABC transporter, three glutathione S-transferase, 13 AP2 domain and six GRAS domain containing gene families that are known to be involved in the regulation of plant abiotic stresses, in particular, herbicide resistance (Figure 3g). These results indicated that expansion of these gene families may have provided a foundation for the adaptation of L. chinensis to the fields, especially in fields managed with herbicide applications. However, L. panicea does not have such basis for adaptation.

Genes loss and gain during the polyploidization process

To gain insights into the effects of polyploidization on herbicide adaptation, we first calculated gene family sizes by identifying protein domains in L. panicea, L. chinensis and S. italica. The results indicated that the sizes of the majority of gene families in L. chinensis were almost twofold of those in L. panicea (Figure 4a,c), and those in L. panicea and S. italica were almost same (Figure 4b,c). Furthermore, we calculated the sizes of several stress tolerance-related gene families in eight grasses and found that sizes of all the stress tolerance-related gene families in L. chinensis were larger than those in L. panicea and O. thomaeum, and sizes of most gene families in L. chinensis were larger than those in most grasses except NB-ARC and AP2 (Figure 4d).

To identify gene loss and gain during L. chinensis polyploidization, we calculated synteny retention ratios of diploid L. panicea genes in tetraploid L. chinensis (i.e. within a gene family, the percentage of gene members with a 1:(1:1) syntenic relationship among L. panicea, the A_t and B_t of the tetraploid L. chinensis). Across the genome, 58.18% of syntenic genes fit the 1:(1:1) retention ratio, whereas 19.15% and 22.67% fit the 1:(1:0) and 1:(0:1) retention ratios, respectively (Figure S19, Table S9). Herbicide resistance can be divided into target-site and nontarget-site resistances (Powles and Yu, 2010). It has been reported that amplification of herbicide target genes in weed genomes leads to their increased herbicide tolerance (Gaines et al., 2010). To investigate whether herbicide targets were amplified and retained during L. chinensis polyploidization, we identified five herbicide target enzymes, including acetyl-CoA carboxylase (ACCase), acetolactate synthase (ALS), 5-enol-pyruvylshikimate-3-phosphatesynthase (EPSPS), phytoene desaturase (PDS) and protoporphyrinogen oxidase (PPO), in L. panicea and L. chinensis. We found that these five herbicide target enzymes all conformed to the 1:(1:1) synteny retention pattern (Table S10). Further expression analysis indicated that genes encoding these target enzymes in both subgenomes of L. chinensis were expressed at a medium to high level, suggesting that these genes are functional after polyploidization-mediated amplification. The amplification of these target genes during polyploidization may have enhanced the tolerance of L. chinensis to herbicides (Figure 4e, Table S11).

Furthermore, we found that synteny retention ratios for most abiotic stress tolerance gene families were higher than 50%, such as ABC transporter (76.14%), GRAS (77.5%), glycosyl transferase (65.52%) and CYP450 (55.63%). However, GSTs just reached 39.53% and AP2 reached 49.15%. In biotic stress-related gene families, WRKY reached 64.56%, Legume lectin reached 55.56% but NB-ARC just reached 28.30%, indicating an obvious gene loss of NB-ARC (Figure 4f). An example of an NB-ARC gene that deviated from the 1:(1:1) synteny retention ratio is illustrated in Figure 4g. In relation to L. panicea, only one homeologous copy of this NB-ARC gene was retained in L. chinensis (within B_t). Interestingly, we found that the herbicide resistance gene ABCC8 (Pan et al., 2021), a member of the ABC transporter family which could transport glyphosate molecules into the vesicles to protect cells from toxicity (Figure 4j),within the same genome region as the NB-ARC gene, conformed to a 1: (1:1) synthetic retention ratio (Figure 4g,j). In addition, we found that CYP76C1 and CYP76C4 (Höfer et al., 2014), members of the cytochrome P450 family known to confer metabolic resistance to herbicides in plants (Figure 4k,l), also fit this conserved pattern during polyploidization (Figure 4h,i). We next investigated expression patterns of these three genes. Surprisingly, we found that LpABCC8 and LpCYP76C4 were expressed at the highest level in L. panicea seeds, and LpCYP76C1 was expressed at the highest level in stems, while they were all expressed the highest in L. chinensis leaves (Figure 4m). This shift of expression in the primary tissues may contribute to their resistance to metamifop and cyhalofop-butyl, common stem and leaf spray herbicides for L. chinensis weed control. Furthermore, we analysed transcriptome data of cyhalofop-butyl-resistant and cyhalofop-butyl-sensitive L. chinensis lines (Chen et al., 2021; Zhang et al., 2022b) and found that all five genes, except LcABCC8-2, were expressed at significantly higher levels in the resistant L. chinensis than the sensitive L. chinensis (Table S12).

Improvement of herbicide adaptation during L. chinensis polyploidization

To reveal the mechanism by which L. chinensis polyploidization confers greater herbicide adaptability, we conducted homology 3D reconstruction and molecular docking experiments on above three pairs of polyploidization-retained proteins with cyhalofop acid (the active molecules of herbicide cyhalofop-butyl) and metamifop, two common herbicides used to control Leptochloa weeds in paddy fields, to investigate whether these genes play a role in the herbicide environment. For the LcCYP76C4 gene pair (LcCYP76C4–9 and LcCYP76C4–10), we found that metamifop and cyhalofop acid molecules could bind to the binding pocket of LcCYP76C4–9 with the binding energy of −8.48 kcal/mol and −7.07 kcal/mol, respectively. Metamifop molecules formed strong hydrogen bonds with ILE-208 (bond length: 2.9 Å) and ARG-232 (bond length: 2.1 Å) of LcCYP76C4–9 in the binding pocket (Figure 5a), and cyhalofop acid molecules formed strong hydrogen bonds with TRP-114 (bond length: 2.4 Å), SER-99 (bond length: 2.2 Å), PRO-370 (bond length: 3.3 Å), LEU-367 (bond length: 1.9 Å) and PRO-366 (bond length: 3.2 Å) in the pocket (Figure 5a). Both molecules could also bind to the binding pocket of LcCYP76C4–10 with the binding energy of −8.76 kcal/mol and −7.2 kcal/mol, respectively. Metamifop molecules formed strong hydrogen bond interactions with ALA-295 (bond length: 2.8 Å) and ASN-102 (bond length: 2.1 Å and 3.5 Å) of LcCYP76C4–10 in the binding pocket (Figure 5b), and cyhalofop acid molecules formed strong hydrogen bond interactions with ALA-295 (bond length: 2.8 Å) and ARG-384 (bond length: 3.5 Å and 2.1 Å) in the pocket (Figure 5b). The LcCYP76C1 gene pair (LcCYP76C1–5 and LcCYP76C1–6) also had the similar function. We found that metamifop and cyhalofop acid molecules could bind to the binding pocket of LcCYP76C1–5 with the binding energy of −7.39 kcal/mol and − 6.18 kcal/mol, respectively. Metamifop molecules formed strong hydrogen bond interactions with PRO-343 (bond length: 2.3 Å) and ILE-345 (bond length: 2.0 Å) of LcCYP76C1–5 in the binding pocket (Figure S20A), and cyhalofop acid molecules formed strong hydrogen bond interactions with ARG-394 (bond length: 1.7 Å) and ARG-106 (bond length: 2.1 Å) in the pocket (Figure S20A). Metamifop and cyhalofop acid molecules could bind to the binding pocket of LcCYP76C1–6 with the binding energy of −12.02 kcal/mol and −7.63 kcal/mol, respectively. Metamifop molecules formed strong hydrogen bond interactions with ARG-33 (bond length: 2.5 Å) and THR-186 (bond length: 2.1 Å) of LcCYP76C1–6 in the binding pocket (Figure S20B), and cyhalofop acid molecules formed strong hydrogen bond interactions with ARG-33 (bond length: 2.0 Å) in the pocket (Figure S20B). The ABCC8 gene pair (LcABCC8-1 and LcABCC8-2, members of the ABC transporter family) could also be bound by the two herbicide molecules. Metamifop and cyhalofop acid molecules could bind to the binding pocket of LcABCC8-1 with the binding energy of −6.69 kcal/mol and − 6.26 kcal/mol, respectively. Metamifop molecules formed strong hydrogen bond interactions with HIS-495 (bond length: 2.3 Å) of LcABCC8-1 in the binding pocket (Figure S20C), cyhalofop acid molecules formed strong hydrogen bond interactions with ARG-494 (bond length: 1.8 Å) and HIS-495 (bond length: 2.3 Å) in the pocket (Figure S20C). Metamifop and cyhalofop acid molecules could bind to the binding pocket of LcABCC8-2 with the binding energy of −5.8 kcal/mol and −5.0 kcal/mol, respectively. Metamifop molecules formed strong hydrogen bond interactions with ARG-130 (bond length: 2.1 Å) and THR-241 (bond length: 3.6 Å) of LcABCC8-2 in the binding pocket (Figure S20D), and cyhalofop acid molecules formed strong hydrogen bond interactions with ARG-364 (bond length: 2.0 Å), MET-368 (bond length: 3.5 Å and 2.4 Å), GLN-375 (bond length: 3.3 Å) and ARG-274 (bond length: 2.7 Å and 1.7 Å) in the pocket (Figure S20D). We also performed homology 3D reconstruction and molecular docking experiments for LpABCC8, LpCYP76C1 and LpCYP76C4 genes from L. panicea and found that LpCYP76C1 and LpCYP76C4 had lower binding energies to metamifop/cyhalofop acid than their homologues in L. chinensis and that LpABCC8 had a similar binding energy compared with LcABCC8-2, but a much lower binding energy than LcABCC8-1 (Table S13). These results suggested that these genes in L. panicea may possess weaker herbicide resistance functions than those in L. chinensis and doubling of these genes in L. chinensis due to polyploidization could result in a much greater herbicide adaptation plasticity and buffering capacity in L. chinensis.

Polyploidization-retained herbicide resistance genes under herbicide selection

In a previous study, we have demonstrated that during its spread in China from the southern/southwestern provinces to the middle and lower reaches of the Yangtze River, L. chinensis has developed significantly increased herbicide resistance, accompanied by the selection of numerous genes involved in herbicide resistance (Wang et al., 2022). Interestingly, we found that both copies of LcCYP76C4 were under selection during the malignant spread of L. chinensis in China (Figure 5c,d), where strong herbicide selection has occurred. Their nucleotide diversity in L. chinensis of the southern border of China (not yet subject to herbicide selection) was significantly higher than that in the middle and lower reaches of the Yangtze River (subject to herbicide selection) (Figure 5e,f). These results indicate that the amplification of LcCYP76C4 driven by polyploidization may mediate the herbicide adaptation evolution of L. chinensis in China. We further identified 2738 genes that were not only under selection, but were also retained during the polyploidization process (Figure 5g), including 22 CYP450, seven GST, 11 AP2, 13 ABC transporter and four GRAS genes from families known to be involved in herbicide tolerance (Figure 5h).

We selected four pairs of herbicide resistance family-related genes whose two copies were both under herbicide selection and analysed their expression patterns. Interestingly, the LcCYP709B2 (Lc_Chr5.g15716) gene was highly expressed in leaves (Figure 5i), suggesting that this gene may contribute to the resistance to metamifop and cyhalofop-butyl, common stem and leaf spray herbicides for weed control. We further confirmed the homologous conservative relationships of this gene during polyploidization (Figure 5j) and found that LcCYP709B2–5 may be under more strongly selection than LcCYP709B2–6 (Figure 5k,l).

To test whether this polyploidization-retained LcCYP709B2 gene has the potential to help L. chinensis degrade the commonly used herbicide metamifop and cyhalofop-butyl, we conducted homology 3D reconstruction and molecular docking experiments on this protein. We found that the metamifop molecule could bind to the binding pocket of LcCYP709B2–5 with the binding energy of −10.51 kcal/mol and formed strong hydrogen bond interactions with TRY-338 (bond length: 3.1 Å), PHE-340 (bond length: 2.4 Å) and GLN-341 (bond length: 1.8 Å) in this binding pocket (Figure 5m). The cyhalofop acid molecule could also bind to the binding pocket of LcCYP709B2–5 with binding energy of −7.14 kcal/mol and formed strong hydrogen bond interactions with ASN-78 (bond length: 3.0 Å), TYR-338 (bond length: 3.2 Å and 2.2 Å), PHE-340 (bond length: 2.6 Å and 2.3 Å) and GLN-341 (bond length: 1.9 Å) in the binding pocket (Figure 5n), indicating that in addition to LcCYP76C4, LcCYP76C1 and LcABCC8 mentioned above, the polyploidization-retained LcCYP709B2 gene is also likely to degrade herbicides and plays a role in the malignant spread of L. chinensis in China. Taken together, these results suggest that amplification and retention of herbicide resistance genes mediated by polyploidization in L. chinensis may be the genomic basis that allows the L. chinensis rapid adaptation to field environments with herbicide stress (Figure 6).

Plant material

The L. panicea plant used in this study was grown in the greenhouse of Hunan Academy of Agricultural Sciences. The root tip was taken for flow cytometry analysis to determine the genome size of the plant.

DNA/RNA extraction, library construction and sequencing

DNA for genome sequencing was extracted from the young leaves of L. panicea using the cetyltrimethyl ammonium bromide (CTAB) method. DNA libraries for single-molecule real-time (SMRT) PacBio genome sequencing were constructed following the standard protocols of the Pacific Biosciences Company and sequenced on the PacBio Sequel II platform using the circular consensus sequencing (CCS) approach. The same DNA was used to construct an Illumina paired-end library with insert sizes of ~400 bp using the NEBNext Ultra DNA Library Prep Kit following the manufacturer's instructions, and the library was sequenced on the Illumina NovaSeq 6000 platform. A Hi-C library was constructed using the young fresh leaves of L. panicea following the proximo Hi-C plant protocol (Phase Genomics) and sequenced on the NovaSeq 6000 platform.

To assist gene predictions, RNA-sequencing (RNA-Seq) was performed using tissues from root, stem, leaf and seed. Total RNA was extracted from each tissue using the TRIzol reagent based on the recommended protocol (Invitrogen, Carlsbad, California, USA). Strand-specific RNA-Seq libraries were constructed using the Illumina TruSeq RNA Sample Prep Kit and sequenced on the NovaSeq 6000 platform. In addition, RNA from all samples was equally mixed, and the mixed RNA was used to construct one PacBio Iso-Seq library. Briefly, cDNA was first synthesized using the Clontech SMARTer® cDNA Synthesis Kit, and then purified using the AMPure PB beads. The Iso-Seq SMRTbell library was then constructed from the purified cDNA with the SMRTbell Express Template Prep kit 2.0 and sequenced on the PacBio Sequel II platform.

Estimation of genome size and heterozygosity

Genome size of L. panicea was estimated using k-mer frequency distribution derived from Illumina short reads with the program Jellyfish (Marçais and Kingsford, 2011) (version 1.1.10). GenomeScope (Vurture et al., 2017) was used to estimate the heterozygosity level of the L. panicea genome.

Genome assembly

Hifiasm (Cheng et al., 2021) was used to assemble HiFi reads into contigs with default parameters. Redundant contigs/sequences in the assembly were removed using Purge Haplotigs with parameters ‘-l 0 -m 40 -h 175’ (Roach et al., 2018). Hi-C data were then used to scaffold the final assembled contigs into pseudomolecules using the Juicer pipeline (Durand et al., 2016). Potential misassemblies were manually checked and corrected based on Hi-C map, genome synteny and read mapping information.

Genome assembly quality evaluation

BUSCO (Simão et al., 2015) was used to assess the completeness of genome assembly and predicted protein-coding genes. Illumina short reads were mapped to the genome assembly using BWA-MEM (version 0.7.17) (Li and Durbin, 2009), and the mapping rate was calculated. LTR Assembly Index (LAI) (Ou et al., 2018) was also used to evaluate the assembly quality.

Repetitive sequence analysis

The repeat library of L. panicea genome was ab initio constructed using RepeatModeler (version 2.0.1) (http://www.repeatmasker.org/RepeatModeler). The consensus TE sequences generated by RepeatModeler were combined with RepBase and used as repeat library in RepeatMasker (version 4.1.0) (http://www.repeatmasker.org) for repetitive element identification. A preliminary list of candidate LTR-RT was generated using LTR_FINDER (Xu and Wang, 2007) and LTR_harvest (Ellinghaus et al., 2008). The identification of high-quality intact LTR-RTs and the calculation of insertion age for intact LTR-RTs were carried out using LTR_retriever (Ou and Jiang, 2018) with default parameters. Tandem repeats were detected using Tandem Repeats Finder (Benson, 1999). Locations of centromeres and telomeres were inferred from the output generated by Tandem Repeats Finder.

Genome annotation

Protein-coding genes were predicted from the L. panicea genome assembly using an integrated approach. RNA-Seq reads from different tissues were aligned to the genome assembly using HISAT2 (v2.0.4) (Kim et al., 2019) and then assembled into transcripts using Cufflinks (v2.2.1) (Trapnell et al., 2012). Open reading frames (ORFs) in the Iso-Seq transcripts were predicted using PASA (v2.0.1) (Haas et al., 2003), and potential full-length cDNA sequences were then extracted and used as the training dataset for ab initio gene predictors, including AUGUSTUS (v3.03) (Stanke et al., 2006), SNAP (Korf, 2004), GlimmerHMM (Majoros et al., 2004) and GeneMark-ET (Lomsadze et al., 2014). Protein sequences from Leptochloa chinensis, Oryza indica, Setaria italica and Oropetium thomaeum were aligned to the L. panicea assembly for homology-based gene prediction using GeMoMa (v1.4.2) (Keilwagen et al., 2016). Finally, ab initio, homology- and transcript-based predictions were integrated using EVidenceModeler (Haas et al., 2008) to generate a consensus model for each gene. Functional annotations of the predicted genes were performed by comparing their protein sequences against the GenBank non-redundant (nr), InterPro, KEGG and eggNOG databases.

Genome comparison

Genome comparison between L. panicea and the two subgenomes of L. chinensis (Wang et al., 2022; available at the Genome Warehouse of the BIG Data Center under accession number GWHBJVB00000000) was performed via whole-genome alignment using the MUMmer package (Kurtz et al., 2004). MCScan (Wang et al., 2012) (Python version) was used for pairwise synteny region search.

Phylogenetic analysis and divergence time estimation

To investigate the evolutionary history of genus Leptochloa, two subgenomes of Leptochloa chinensis, nine grass subfamilies, Brachypodium distachyon, Oryza sativa, Setaria viridis, Setaria italica, Sorghum bicolor, Zea mays, Oropetium thomaeum, Leptochloa panicea and one dicot plant Arabidopsis thaliana were used for gene family construction using OrthoFinder (Emms and Kelly, 2015) (version 2.3.12) with default parameters. Protein sequences of 549 single-copy orthologs from the 10 species were concatenated for the species tree construction. Protein sequences of each single-copy orthologous group were aligned with ClustalW2 (Larkin et al., 2007). Maximum likelihood tree was constructed using FastTree (Price et al., 2010) with 1000 bootstrap replicates. The divergence time was estimated using MCMCTree (Yang and Rannala, 2006) with branch lengths estimated by BASEML in the PAML package (Yang, 2007) and the independent rate model for time estimation.

To detect whole-genome duplication events in the genus Leptochloa, we calculated nonsynonymous substitution (Ks) values for syntenic homeologous gene pairs using WGDI (Sun et al., 2022). We used the nucleotide substitution rate of 6.5 × 10⁻⁹ mutations × bp⁻¹ × generation⁻¹ as a molecular clock (Molina et al., 2011).

Gene family expansion and contraction analysis

Gene-family expansion and contraction in Leptochloa genomes were determined using CAFE (version 4.2.1) (De Bie et al., 2006). The gene family size for each species used in CAFE was calculated with OrthoFinder (Emms and Kelly, 2015) (version 2.3.12). The gene birth and death rate was estimated with orthologous groups that were conserved in all species. To better understand the potential functional category of each gene family, we used KinFin (Laetsch and Blaxter, 2017) (v1.0), along with gene functional annotations assigned by InterProScan and gene ontology, to derive rich annotations for gene families.

RNA sequencing data analysis

RNA-Seq reads were processed to remove adapters and to trim low-quality bases using fastp (Chen et al., 2018) (version 0.21.0). Cleaned reads were mapped to the L. panicea and L. chinensis genome using HISAT2 (Kim et al., 2019) (version 2.1.0) with default parameters. The Stringtie software (Pertea et al., 2015) was then used to calculate the TPM (transcripts per million) values of genes.

Diversity analysis

The nucleotide diversity (π) and population differentiation (F_ST) values were calculated using VCFtools (version 0.1.13) (Danecek et al., 2011) based on the high-confidence SNPs identified from 89 L. chinensis accessions reported in our previous study (Wang et al., 2022). The π value for each SNP was calculated, and the nucleotide diversity level was measured using a 100-kb window with a step size of 10 kb for each L. chinensis population.

Homology modelling

The NCBI BLAST server was used to select the three PDB proteins with the highest protein similarity as templates. The homology of amino acid sequence was aligned. Homology modelling was carried out using the Modeller (Webb and Sali, 2016) and assigning to the heme subsequently (in case of cytochrome P450 proteins).

Molecular docking

Molecular docking experiments were performed to investigate the binding mode between the proteins and cyhalofop acid/metamifop using Autodock4 (Morris et al., 2009). The 2D structure of the ligands was drawn by ChemBioDraw Ultra and was optimized by the MM2 method using ChemBio3D Ultrasoftware to obtain the 3D structure. The AutoDockTools 1.5.7 package (Morris et al., 2009) was employed to generate the docking input files. For docking, the default parameters were used if it was not mentioned. The best-scoring pose as judged by the AD4 docking score was chosen and visually analysed using PyMoL 1.7.6 (www.pymol.org).