A High-Quality Genome Assembly of Johnsongrass Illuminates Relationships Between Polyploidization, Crop-Wild Hybridization, and Reproductive Biology

Abstract

Johnsongrass [Sorghum halepense (L.) Pers.] is an allopolyploid, rhizomatous, perennial grass species and one of the most troublesome weeds in global agriculture. We assembled the first Johnsongrass genome to clarify poorly understood genetic factors influencing variable rates of crop-wild hybridization with cultivated sorghum [S. bicolor (L.) Moench]. The draft genome assembly has a total size of 3.26 Gb and BUSCO completeness of 95.3%. We also report the first evolutionary analysis of INHIBITION OF ALIEN POLLEN (IAP), the only known cross-(in)compatibility locus in the genus. Our results reveal an evolutionary history of genome instability, including the loss of distinct parental subgenomes, and suggest that Nebraska accession “J-37,” the genome donor, is a segmental allotetraploid that may function as a diploid or aneuploid during meiosis. Genome instability could explain observations of variable ploidies in Johnsongrass and facilitate ongoing hybridization with sorghum where gamete ploidies and IAP alleles match. Given this information, we provide a suggested research framework for studying evolution and gene expression in the Sorghum genus where crop-wild hybridization occurs and for predicting the potential for hybridization between specific crossing partners. Collectively, this work will bolster efforts to study and manage reproductive biology in other crop-wild polyploid complexes.

Significance.

Johnsongrass is a globally important weed that can hybridize with cultivated sorghum and serve as a conduit for crop genes to escape into wild populations. We report the first publicly available Johnsongrass genome, which is also the first polyploid genome to be published in the Sorghum genus. Our results suggest that variation in a major cross-(in)compatibility locus as well as a segmental allopolyploid genome structure may collectively influence the ability of Johnsongrass to hybridize with sorghum. This body of work will accelerate population genetics research in Johnsongrass and support improved systematics in the Sorghum crop-wild complex, especially when developing biocontainment strategies to prevent unintended gene flow.

Introduction

There are 25 described species in the Sorghum genus (Dillon et al. 2007), and of them, only sorghum [S. bicolor (L.) Moench; 2n = 2x = 20] is cultivated. Several species occur as weeds in farm fields where they exhibit variable rates of crop-wild hybridization (Ohadi et al. 2017). Identifying ways to limit crop-wild hybridization between Sorghum species is one strategy to reduce the ecological risk of deploying modern biotechnologies in sorghum to make it a more competitive crop with less heat- or drought-tolerant species. Those investments would pay off by adding resilience to food, feed, and biofuel systems under lower-input conditions that will become more prevalent in the coming years.

The allopolyploid Johnsongrass [Sorghum halepense (L.) Pers.; reported as 2n = 4x = 40] is a perennial wild relative of sorghum and one of the most economically important weeds in global agriculture (Paterson et al. 1995, 2020). It is often found growing in sorghum fields where it competes for space, light, water, and nutrients through the production of above-ground biomass and aggressive rhizomes. Commercial sorghum seed lots with excessive weed contamination are discarded. This necessitates labor-intensive crop scouting throughout the growing season in regions where Johnsongrass is prevalent, as there are no herbicides that are selective for Johnsongrass in sorghum. Given the widespread occurrence of Johnsongrass and the need for biotechnology in sorghum improvement, our long-term research interest in Johnsongrass has focused on its potential as a conduit for the escape of crop genes into the environment.

Many groups have studied the risk of sorghum traits transferring to and perpetuating within perennial weed populations (Paterson et al. 1995; Arriola and Ellstrand 1996; Morrell et al. 2005; Andow and Zwahlen 2006; Adugna et al. 2013; Korres et al. 2015; Ohadi et al. 2017; Hodnett et al. 2019). Outcrossing rates between sorghum and Johnsongrass range from 0% to 7% at 5 m (Arriola and Ellstrand 1996), and from 9.6% to 16.2% at 5 m among Johnsongrass populations (Maity et al. 2022). However, the lack of a Johnsongrass reference genome to support population genomics studies has hampered progress. It is clear that the genetic causes and effects of crop-wild hybridization between these species are complex (Arriola and Ellstrand 1996; Morrell et al. 2005; Kong et al. 2020; Paterson et al. 2020) and require updated methods to quantify and predict.

Recent studies have provided some clarity as to why this is. First, a density-dependent matrix model was developed to evaluate the effect of pollen-mediated gene flow from herbicide-resistant sorghum on the evolution of resistance in Johnsongrass (Holmes et al. 2022). The sorghum was assumed to be homozygous for a single, dominant resistance allele to acetolactate synthase (ALS)-inhibiting herbicides. Results showed that the degree of dominance of the herbicide-resistance trait and ploidy level have a much larger effect on the predicted speed of resistance evolution compared to the rate of gene flow. Given that initial hybrid formation and repeated backcrossing to crop or wild parents results in a wide range of ploidies in hybrid sorghum × Johnsongrass populations (Ohadi et al. 2017; Sias et al. 2023), and that the copy number of ecologically beneficial alleles (influenced by ploidy level) is a bigger driver of evolution than the frequency of gene flow, this model may at least partially explain the persistence of crop genes within geographically diverse Johnsongrass populations even in the absence of recent exposure to sorghum (Morrell et al. 2005; Ohadi et al. 2017).

Second, unreduced (2n) sorghum ovules appear to be the primary route through which initial hybridization events with Johnsongrass occur in the southern Great Plains (Hodnett et al. 2019), and genotypic variation for aneuploidy rates vary substantially among common US sorghum cultivars (Hodnett et al. 2019; Sias et al. 2023). The genes influencing aneuploidy rates (likely regulators of meiosis) in sorghum have not yet been identified.

Third, genetic variation exists among Sorghum species and cultivars at the INHIBITION OF ALIEN POLLEN (IAP) locus, which is the only known pollen–pistil cross-(in)compatibility locus in the genus (Hodnett et al. 2005, 2010; Price et al. 2006; Bartek et al. 2012). IAP was first discovered when researchers attempted to cross sorghum with wild relatives from Australia (Hodnett et al. 2005; Price et al. 2006). We previously identified five genes at IAP using published data (Gill 2014) and a subsequently updated version of the sorghum reference genome (v3.1) (McCormick et al. 2018), but efforts are ongoing to validate their specific functions during cross-pollination (Yerka 2023; Yerka and Cooper 2023; Cooper 2024).

The ancestral genomes of Johnsongrass are reported to include S. bicolor and S. propinquum (Kunth) Hitchc. (a rhizomatous weed native to Southeast Asia; also 2n = 2x = 20) (Paterson et al. 1995, 2020; Vercellino et al. 2023). However, one study of US populations of Johnsongrass has reported a greater abundance of S. bicolor-derived sequences than S. propinquum-derived sequences (Paterson et al. 2020), suggesting that the S. bicolor subgenome may be dominant over the S. propinquum subgenome. One plausible explanation for the over-representation of S. bicolor sequences in US accessions of Johnsongrass is that ongoing gene flow (backcrossing to a parental species) among cultivated sorghum and Johnsongrass accessions has occurred in farm fields. The availability of a high quality, functionally annotated Johnsongrass reference genome may clarify the relative contributions of each parental genome to its evolution and sexual and asexual reproductive biology.

Despite the negative impacts of crop-wild hybridization, there are instances when it benefits agriculture, particularly when superior abiotic stress-tolerance traits are transferred to crops. Research is actively underway in several breeding programs in the United States to improve the sustainability of sorghum grain and biomass production by converting it to a perennial crop (Nabukalu and Cox 2016; Jessup et al. 2017a, 2017b; Cox et al. 2018; Ryder et al. 2018; Kaler et al. 2025). These programs focus on combining the rhizome production of Johnsongrass with the grain or biomass yield and reduced weediness of sorghum. Hence, another benefit of having a Johnsongrass reference genome is that it would enhance molecular breeding in those materials for improved biotic and abiotic stress resistance by enabling the development of novel interspecific pangenomes for functional genomics, marker-assisted selection, physiological network modeling, and genomic prediction.

Allotetraploid genomes are notoriously challenging to assemble. This is especially true in feral plant accessions like Johnsongrass where heterozygosity and variation in genome structure due to repeated crop-wild hybridization across ploidies (Hodnett et al. 2019; Sias et al. 2023) can result in aneuploidy and complicate contig assembly and haplotype phasing. A recent study of the grape cultivar “Kyoho” highlighted the challenges of assembling allopolyploid genomes derived from diploid parent species with weak reproductive barriers (Zhang et al. 2026), which is a similar situation to Johnsongrass. Nevertheless, recent advancements in long-read sequencing technologies and increasingly sophisticated assemblers have enabled the construction of more complex genomes. Here, we report the first high-quality draft genome for Johnsongrass using de novo long-read assemblies and evaluate its evolutionary divergence from ancestral diploid species. We also report the first evolutionary analysis of the IAP locus (Hodnett et al. 2005; Price et al. 2006; Gill 2014; Trowbridge 2019), which is a key target for studying the genetic mechanisms underpinning rates of interspecific gene flow among Sorghum species. While the primary focus of this work was on developing genomics resources for studying crop-wild gene flow, the results will also assist with wide-ranging studies of weed science, ecology, systematics, synthetic biology, and plant breeding in the economically important Sorghum crop-wild complex.

Results

The S. halepense Genome

We assembled the de novo S. halepense genome of accession “J-37' (Fig. 1) collected from Buffalo County, NE in 2015. It is a partially self-fertile (based on % seed set) tetraploid [2n = 4x = 40; x = 10] based on a k-mer–based ploidy inference (GenomeScope P = 4) and Smudgeplot (Ranallo-Benavidez et al. 2020) dosage patterns (Table 1 and Fig. S1a). While flow cytometry was not conducted on this specific accession, it had been used previously to determine that a geographically close accession also from Buffalo County, NE (“J-36') (Werle et al. 2016), was a tetraploid (Supplementary File S2). J-37 was used as the genome donor because, like J-36, it is resistant to ALS-inhibiting herbicides; but unlike J-36, it grows well in the greenhouse. An ALS-resistant accession was desired to promote genomics-assisted investigations of herbicide-resistance evolution in Johnsongrass.

Fig. 1.

A representative plant a) and panicle b) at anthesis from the J-37 accession of S. halepense.

Table 1.

Draft genome assembly statistics

Nx metric	Size (bp)	Number of contigs
N50	4,263,274	111
N60	2,309,267	216
N70	1,370,518	398
N80	787,497	710
N90	322,655	1,358
N100	8,171	4,688
Assembly statistic	Count/percentage
% GC	43.73%
Number of gene models	177,328
BUSCO completeness	95.3%

Nx statistics summarize assembly contiguity, where the contig length (in bp) at each Nx threshold (e.g. N50, N60) indicates the minimum size of contigs that together cover x% of the total assembly. The corresponding number of contigs meeting or exceeding each size is also provided, reflecting assembly fragmentation. % GC refers the proportion of G and C nucleotides within the assembly. Number of gene models refers the number of predicted gene models within the assembly. BUSCO completeness refers to the proportion of Basic Universal Single Copy Genes found within the annotated genes.

The J-37 genome was assembled from 81.2 Gb of HiFi reads from five 8 M SMRT cells using Sequel II sequencing chemistry 2.0 and 30-h movies each on a PacBio Sequel II and Sequel IIe sequencers and a mean read length of 17.6 kb (see distribution in Fig. S1b). To facilitate scaffolding, we generated 68.2 Gb of PE150 Hi-C data. The heterozygosity rate of the genome was estimated to be between 2.53% and 4.81%. GenomeScope models the k-mer multiplicity spectrum from a set of computed k-mers from sequenced reads. For a tetraploid with per-haploid coverage C, the heterozygous and homozygous k-mers cluster near ∼1×C, ∼2×C, ∼3×C, and ∼4×C. In our data, the k-mer spectra with GenomeScope2 (k = 21) exhibited two major peaks and a minor single-dose peak at the expected tetraploid multiplicities, and the model fit returned P = 4 with low error, indicating tetraploidy rather than diploidy (Fig. S1c). To evaluate the inheritance mode, we quantified allele-dosage classes using Smudgeplot, which extracts heterozygous k-mer pairs from k-mer count databases to clarify genome structure by comparing the sum of k-mer pair coverages (CovA + CovB) to their relative coverage (CovB/(CovA + CovB)) (Ranallo-Benavidez et al. 2020). This approach allows the analysis of obscure genomes with duplications and various ploidy levels. The distribution showed strong 1:3 classes (AAAB/AAB) together with a distinct 2:2 (AABB) band. In a strict allotetraploid with disomic pairing and deeply diverged subgenomes, AABB typically dominates and would have a higher percentage, whereas in a strict autotetraploid, AAAB/AAB dominate. The observed mixed pattern of inheritance, AAAB/AAB > AABB, is most consistent with segmental allopolyploidy (patterns of mixed disomic–tetrasomic pairing during meiosis) where subgenomes are only moderately diverged, a mosaic of parental chromosomes is present rather than distinct subgenomes, and when some chromosomes or segments pair tetrasomically (Mason and Wendel 2020). Thus, while AAAB > AABB would argue against a strict disomic model, the presence of a substantial AABB band together with the 4× k-mer peak structure supports tetraploidy with mixed (predominantly tetrasomic, partially disomic) inheritance, consistent with an allopolyploid-derived genome that is undergoing partial rediploidization. This could explain observations of reduced fertility in the greenhouse. While all of the current allele-dosage work was done bioinformatically, our results suggest that future work on Johnsongrass genome assemblies should be paired with observations of meiotic behavior in addition to flow cytometry. Flow cytometry may estimate genome size, but it cannot explain chromosome pairing behavior.

Segmental allopolyploidy has been reported in some other grass tetraploids, including Urochloa (Vigna et al. 2016), Brachiaria, (Worthington et al. 2016), and rice [Oryza sativa L.] (Sun et al. 2017). In this context, our GenomeScope and Smudgeplot results show clear 4x peaks and both AABB and AAAB dosage classes, which is consistent with an allotetraploid exhibiting segmental behavior. Accordingly, we interpret Johnsongrass as a segmental allotetraploid with mixed chromosomal pairing behavior, rather than a purely auto- or purely allotetraploid. This is consistent with a scenario in which an allotetraploid has been backcrossed to one of its diploid progenitors and most, but not all, of the other diploid progenitor's genome has been lost. The average coverage of the sequencing was estimated to be ∼27x, assuming the genome size closely approximates the sum of its two progenitors (S. bicolor ∼730 Mb and S. propinquum: 700 to 800 Mb) (McCormick et al. 2018; Paterson et al. 2020; Tao et al. 2021). The assembled genome included 4,688 contigs with an N50 of 4.26 Mb and a total size of 3,255,881,808 bp (Table 1). The longest contig (53 Mb) covers most of Chromosome 4 of the S. bicolor B.Tx623 genome, spanning from position ∼2.1 Mb to the end. Additionally, there are 32 contigs longer than 20 Mb, with a total size of 1 GB, which are long enough to span approximately one chromosome arm. After aligning the HiFi reads to the contigs, we found that 663 contigs (14% of the total contigs, Fig. S1d) had coverage larger than the major peak (>35x). These contigs, which have read coverage above the average sequencing coverage, may have collapsed during assembly due to low variation among chromosomes.

Given that J-37 exhibits ALS resistance, we next examined copy number variation at the ALS1 large subunit gene (Sobic.004G155800). Four complete haplotypes were identified in the assembly (Fig. S2a), consistent with tetraploidy. Notably, mutations associated with ALS tolerance were detected in two of the four haplotypes. These included a one–amino acid deletion near the signal peptide (SP) region, a missense mutation (A₃₀G) within the SP, and a V₅₆₀I substitution in the TPP_AHAS domain, allelic variants corresponding to previously characterized ALS resistance mutations (Fig. S2b;  Yu et al. 2008). The presence of multiple gene haplotypes and segregating resistance-associated alleles further supports the tetraploid and heterozygous nature of the J-37 genome.

Within the genome, 61.93% was annotated or classified as transposons, which is similar to the transposon content of S. bicolor and S. propinquum (McCormick et al. 2018; Tao et al. 2021). Gypsy is the major transposon family (38.7% of the genome; Table S1). A total of 177,328 gene models with an average length of 1510 bp and an average of three exons per gene were annotated using the transcriptome data from the same Johnsongrass genotype used in the assembly and its progeny. By way of comparison, the mean gene length in this draft genome assembly is comparable to that of the S. bicolor B.Tx623 v1.4 release (1,473 bp) but shorter than that of the v3.1 release (1,835 bp). This difference may be due to the amount of gene expression data used in the gene prediction pipeline, as the v3.1 B.Tx623 annotation incorporated 47 deeply sequenced transcriptomes. The relatively high proportion of shorter gene models in our annotation likely accounts for the smaller average exon count—three exons per gene—compared with five exons per transcript in B.Tx623 v3.1.

The BUSCO analysis showed a high degree of completeness for both the genome and transcriptome annotation. When running BUSCO in genome mode, it scored 94.8% for completeness with 4.5% missing and 0.7% fragmented. The BUSCO results from the predicted CDSs showed 95.3% completeness with 2.8% missing and 1.9% fragmented. These results indicate that the S. halepense draft genome assembly is of high quality and able to accelerate genomics and genetics studies even if a chromosome-scale assembly was not possible due to some collapsed contigs. The assembly is a major step forward in Johnsongrass molecular genetics and Sorghum systematics, being the first polyploid genome in the genus to be assembled. The segmental allopolyploid genome structure made the assembly especially challenging. Additional efforts are needed to obtain a more complete, preferably telomere-to-telomere (T2T) assembly for this and other accessions of Johnsongrass, which will likely require ultra-long reads and sequencing many gametes to resolve haplotypes (Sun et al. 2022). Nevertheless, this first effort to construct the Johnsongrass genome and annotation provides a foundational resource for the research community.

Comparison of the Johnsongrass Genome With Its Parental Genomes

Round 1 annotation had ∼98% of S. halepense gene models with annotation edit distance (AED) < 0.5 and had the best BUSCO results at 95% (Table S2); therefore, the round 1 annotation was used for further filtering. The initial MAKER round 1 annotation had 181,271 predicted gene models. After applying the AED filter <0.5 and a minimum open reading frame length of 25 codons, the number of predicted gene models was reduced to 177,328, which we define as our full high-quality set. Filtering overlapped complex repeats reduced these gene models to 120,271 gene models, which we define as our working set used for evolutionary analyses. The justification for obtaining a working set of gene models for these analyses is that the full set of ∼177 k included those with overlaps in complex repeats where ab initio gene predictors can be incorrect. The working set had a BUSCO completeness score of 57.7%.

To investigate the synteny relationships between S. halepense, S. bicolor, and S. propinquum, we aligned the coding sequences (CDS) of their genomes. The analysis showed 50% collinearity between S. halepense and S. bicolor, and 53% collinearity between S. bicolor and S. propinquum. To further assess chromosome-level conservation, we examined syntenic block coverage relative to the S. bicolor reference genome. Six of the ten S. bicolor chromosomes showed syntenic block coverage above 99%, two exceeded 86% coverage, and chromosomes 1 and 3 exhibited coverage of 71% and 80%, respectively (Fig. 2a and Table S3). Given that BUSCO analysis indicates >95% completeness of the gene space, the relatively lower synteny coverage on chromosomes 1 and 3 likely reflects greater divergence between our Johnsongrass germplasm and the modern reference line BTx623, rather than issues related to assembly quality or representation of the original progenitor genome. Significant genome rearrangements have occurred in S. halepense. For example, as a large section of parental chromosome 7 occurs in three scaffolds of S. halepense, these were used in a separate synteny analysis to provide a more granular picture of relationships between S. halepense and its parental genomes. Results showed that there are indeed three haplotypes of S. bicolor chromosome 7 in S. halepense and highlight the loss of chromosome 7 from S. propinquum through recombination with, and dispersal throughout, the haplotypes inherited from S. bicolor (Fig. 2b). Trisomy of S. bicolor chromosome 7 should be confirmed by a more complete genome assembly and/or assembling the genomes of the genome donor plant's progeny. It is possible that a fourth haplotype of S. bicolor chromosome 7 exists in the S. halepense genome and that the contig collapsed during the present de novo assembly due to a lack of diversity. Nevertheless, our synteny analyses provide evidence of an evolutionary history of genome instability and the potential for production of aneuploid gametes in J-37. It is clear that distinct parental subgenomes no longer exist and that the S. propinquum-derived subgenome is widely and unevenly dispersed throughout S. bicolor-derived sequences. This is novel information about Johnsongrass genome biology and evolution that likely impacts its ability to hybridize with cultivated S. bicolor and its dedomesticated conspecific wild relative, shattercane.

Fig. 2.

Synteny analyses among S. bicolor (SB), S. propinquum (SP), and the J-37 accession of S. halepense (JG). Results highlight the loss of distinct parental subgenomes through recombination, resulting in a mosaic genome structure. a) The first and second tracks show synteny between chromosomes of SB and SP, respectively. The second, third, and fourth tracks show synteny among SP chromosomes, consecutive CDSs (scaffolds) of JG, and SB chromosomes, respectively. Accession J-37 has undergone extensive whole-genome rearrangements consistent with segmental allotetraploidy. b) Evidence for an evolutionary history of genome instability in the J-37 accession of JG (third track) using a synteny analysis of ancestral Chromosome 7 from SP (second track) and SB (first and last tracks) as an example. There are three haplotypes of a nearly intact chromosome 7 from SB in JG and none from SP, whereas gene order and synteny are conserved between SP and SB, each JG copy has different degrees of synteny with SP, and sites are unevenly dispersed across the three haplotypes.

When the working set of S. halepense gene models was aligned with the CDSs of the parental genomes using BLASTN with default parameters, 10,609 S. halepense gene models produced hits only in S. bicolor, 24 gene models produced hits only in S. propinquum, and 79,207 gene models produced hits in both parents. These results are consistent with (i) OrthoFinder results (more orthogroups grouped S. halepense proteins with S. bicolor than with S. propinquum proteins; Fig. 3), (ii) observations from our synteny analysis, and (iii) previous molecular evidence (Paterson et al. 2020) that S. propinquum-derived genes are under-represented in S. halepense relative to S. bicolor-derived genes.

Fig. 3.

Venn diagram summarizing OrthoFinder results. Orthogroup relationships among S. bicolor, S. propinquum, and S. halepense are based on the predicted protein sets of the three species. Each number represents the number of orthogroups containing proteins from each combination of species. For instance, 15,527 orthogroups contain proteins from all three species, whereas 164 orthogroups contain only S. propinquum proteins.

The two parental genomes (S. bicolor and S. propinquum) contain a very similar number of protein-coding genes (32,160 and 32,019, respectively). Given the GenomeScope and Smudgeplot results supporting an allopolyploid origin, the most likely explanation for the over-abundance of S. bicolor-derived genes is gene flow between S. halepense and S. bicolor. To explore this in more detail, we inferred the origin of genes across the 20 largest contigs (Fig. S3a). J-37 lacks distinct parental subgenomes and instead has a mosaic genomic structure, consistent with segmental allopolyploidy due to extensive introgressions from S. bicolor. The self-alignment of the J-37 genome using Mummer 4 package (Marcais et al. 2018) revealed a similar mosaic genome structure (Fig. S3b). Large contigs showed variable numbers and lengths of duplicated alignments, ranging from long continuous blocks to short fragmented regions. This uneven pattern differs from the uniform one-to-one correspondence expected for a balanced AABB genome and is consistent with differential retention of ancestral genomic regions following recurrent gene flow from S. bicolor. We note that J-37 was collected from a cultivated field in a historically sorghum-producing region of Nebraska, where frequent crop-wild gene flow was possible.

The remaining 30,431 S. halepense gene models produced no hits in either parental genome. This may be due to a combination of several factors: some genes may have evolved de novo in S. halepense, some may be rearranged and difficult to detect, some may be undetectable by BLAST for other reasons (e.g. they are too short), some may be spurious gene models (due to over prediction), some may have been lost in the parental genomes, and some may not be present, sequenced, and/or annotated in the accessions used to assemble the parental genomes.

For the 79,207 S. halepense gene models with hits in both parental species, if the S. bicolor and S. propinquum genes were best reciprocal BLASTN hits (n = 68,283 trios), we generated two pairwise alignments (S. halepense–S. bicolor and S. halepense–S. propinquum) and computed d_N, d_S, and d_N/d_S. We then used the d_S values to classify S. halepense gene models as more similar to S. bicolor (n = 46,345), more similar to S. propinquum (n = 18,814), or equally similar to both (n = 3124). The S. halepense–S. propinquum d_N/d_S ratios from the 18,814 gene models that were more similar to S. propinquum were significantly higher (median: 0.218) than the S. halepense–S. bicolor d_N/d_S ratios from the 46,345 gene models that were more similar to S. bicolor (median: 0.194; Mann–Whitney's U test, P = 9.5 × 10⁻⁴³).

For each of the 68,283 S. halepense–S. bicolor–S. propinquum trios, we conducted a BLASTN search using the S. halepense gene model's CDS as query against all Zea mays (Zm-B73-REFERENCE-NAM-5.0) (Hufford et al. 2021) CDSs and retained the best hit. For the 52,185 trios for which we found a hit, we aligned all four sequences and generated a phylogenetic tree. In 27,078 of these trees the S. halepense gene model clustered with the S. bicolor gene (19,162 of them had a bootstrap support ≥ 80%), in 11,761 of the trees it clustered with the S. propinquum gene (5,495 with a bootstrap support ≥ 80%), and 13,346 trees contained a S. bicolor–S. propinquum clade (6,810 with a bootstrap support ≥ 80%). The S. halepense–S. propinquum d_N/d_S ratios of the 5,495 S. halepense gene models that cluster with S. propinquum genes are significantly higher (median: 0.189) than the S. halepense–S. bicolor d_N/d_S ratios of the 19,162 S. halepense gene models that cluster with S. bicolor genes (median: 0.165; Mann–Whitney's U test, P = 7.9 × 10⁻¹⁴). Collectively, these results are compatible with a scenario in which the S. halepense genome is a hybrid species from two original subgenomes, one derived from S. bicolor and one derived from S. propinquum, with the latter one being represented by fewer and faster-evolving genes. This is compatible with a scenario in which the S. propinquum-derived subgenome is undergoing fractionation.

The divergence time of S. bicolor and S. propinquum was estimated at 1 to 2 MYA (Paterson 2008). For each of the 19,162 S. halepense gene models that clustered with B.Tx623 S. bicolor genes, we estimated d_S separately for each of the branches in the gene tree. The median d_S was 0.000005 for the S. bicolor branch, 0.000006 for the S. halepense branch, and 0.010913 for the branch subtending both branches (Fig. 4a), indicating that the S. halepense sequences derived from cultivated S. bicolor diverged from S. bicolor very recently. We then conducted an equivalent analysis for each of the 5,495 S. halepense gene models that clustered with S. propinquum genes. The median d_S was 0.010785 for the S. propinquum branch, 0.009665 for the S. halepense branch, and 0.009143 for the branch subtending both branches (Fig. 4b), indicating that the S. halepense sequences derived from S. propinquum diverged from S. propinquum around 0.5 to 1 MYA.

Fig. 4.

a) Median d_S values for S. halepense gene models clustering with S. bicolor; b) Median d_S values for S. halepense gene models clustering with S. propinquum.

Expression Divergence of Homoeologous Genes Derived From Parental Subgenomes

To compare differences in gene expression between homeologous pairs of genes inherited from S. bicolor and S. propinquum, we first identified 5,575 high-confidence homeologous gene models (Fig. 5a). This number of genes is low with respect to the total number of genes in the assembly because the ancestral S. propinquum genome is vastly under-represented in general, so a more complete comparative analysis was not possible. Across the 10 RNA sequencing (RNA-Seq) samples, more S. bicolor-derived genes demonstrated up- or down-regulation compared with S. propinquum-derived genes (Fig. 5b). S. bicolor-derived genes showed higher overall expression and more frequent two-fold expression changes compared to S. propinquum-derived genes (Fig. 5c and d). This result suggests that S. bicolor-derived genes are dominant in overall gene expression. Interestingly, however, drought-stressed stems contained more highly expressed genes (1,848) derived from S. propinquum (1,514). This result may indicate that the S. propinquum progenitor was more tolerant of drought stress, and/or that S. bicolor drought tolerance alleles were lost during domestication. Gene function prediction analysis (Fig. S4) revealed that S. bicolor-derived genes have enriched activity relative to chromatin binding, positive regulation of RNA biosynthetic processes, and positive regulation of cellular biosynthetic processes. These results suggest a strong emphasis on transcriptional regulation, cellular biosynthesis, and overall gene expression control, which are important for growth and environmental sensing. On the other hand, S. propinquum-derived genes are enriched in separate, distinct processes, including root morphogenesis and responses to oxidative and abiotic stress, indicating a focus on structural reorganization and stress response mechanisms. This implies that S. propinquum-derived genes may have been retained in S. halepense to provide enhanced environmental resilience and cellular integrity under stress.

Fig. 5.

Comparative expression analysis of homeologous genes derived from S. bicolor (SB) and S. propinquum (SP) in the J-37 accession of S. halepense. a) Total homeologous pairs (black) compared with high-confidence homeologous pairs (green); b) Fold-change distribution of high-confidence homeologous pair expression between SB (green)- and SP (blue)-derived sequences; c) Comparison of total counts (SB_total, SP_total) and log2 fold-changes (SB_2FC, SP_2FC) of high-confidence homeologous pairs from different tissues and conditions. The green and blue bars represent total counts while the darker shades indicate fold-change values for each parental species across tissues or conditions. d) Distribution of log TPM values for high-confidence homeologous pairs in SB (green)- and SP (blue)-derived sequences. The violin shapes highlight gene expression variability across different tissues and conditions, with wider shapes representing higher variability within each parental species.

Evolutionary Analysis of Johnsongrass IAP Loci in the De Novo Draft Genome Assembly

The IAP locus is the only known prezygotic cross-(in)compatibility locus in the Sorghum genus. A loss-of-function mutation at IAP enables cross-incompatible pollen from some wild Sorghum species to reach the ovules of S. bicolor (Hodnett et al. 2005, 2010; Price et al. 2006; Bartek et al. 2012; Gill 2014). We previously identified the region encompassing the IAP locus (genes Sobic.002G023300 to Sobic.002G023700) in the reference S. bicolor genome (B.Tx623 v3.1.1; chromosome 2, positions 77,742,459 to 77,742,459) (Trowbridge 2019). For the present analysis, the IAP candidate gene number was expanded from five (our original estimate) to six genes (Table 2) when we determined that Sobic.002G023200 and Sobic.002G023300 are adjacent and encode proteins that are 59.4% identical to each other, indicating that they are the result of a gene duplication event. Subsequently, BLASTN searches were conducted to identify the homologous regions in the 12 S. bicolor accessions and the S. propinquum accession sequenced by Tao et al. (2021), and in our S. halepense assembly. We identified one IAP locus in each S. propinquum and S. bicolor accession and three IAP alleles in our S. halepense assembly (in scaffolds h1tg000292l, h2tg000435l, and h2tg000951l). In some of the identified IAP loci, some genes are absent (due to large-scale deletions) or exhibit signs of pseudogenization (i.e. premature stop codons and/or frameshift mutations, which in some cases might be due to limited sequence quality) (Table 2). In two S. bicolor accessions, we found a large copy plus a partial, overlapping one, probably indicating assembly errors or incomplete duplications.

Table 2.

Gene content of the IAP loci in different Sorghum genomes. Putatively functional genes (i.e., those without signs of pseudogenization) are depicted as “F”, absent genes as “A”, and pseudogenes (i.e., those with premature stop codons and/or frameshift mutations) as “P.”

Genome	Sobic.002G023200	Sobic.002G023300	Sobic.002G023400	Sobic.002G023500	Sobic.002G023600	Sobic.002G023700
S. bicolor B.Tx623	F	F	F	F	F	F
S. bicolor PI532566	A	A	F	F	F	F
S. bicolor AusTRCF317961 (Chr 2)*	A	F	F	F	P	F
S. bicolor AusTRCF317961 (scaffolF 3234)*	F	A	A	A	A	A
S. bicolor PI536008	F	F	P	F	P	F
S. bicolor 353, copy a**	P	A	A	A	A	A
S. bicolor 353, copy b	F	F	F	F	P	F
S. bicolor PI525695	A	F	P	F	F	F
S. bicolor IS19953	P	F	F	P	F	F
S. bicolor IS929	F	F	F	F	P	P
S. bicolor Ji2731	F	F	F	P	P	P
S. bicolor R931945-2-2	F	F	F	A	F	F
S. bicolor IS3614-3	F	F	F	F	F	F
S. bicolor IS12661	F	F	F	F	F	F
S. bicolor IS8525	P	F	P	P	F	F
S. propinquum S369-1	F	F	P	P	P	P
S. halepense (h1tg000292l)	F	F	F	F	F	F
S. halepense (h2tg000951l)	F	F	P	F	F	P
S. halepense (h2tg000435l)	F	A	A	F	P	F

^*The two sequences overlap and were combined into a single one.

^**The sequence was not included in our phylogenetic analyses.

We obtained a phylogenetic tree showing the relationships among all the identified IAP loci (i.e. the genomic regions encompassing genes Sobic.002G023200–Sobic.002G023700 or their orthologs) in S. halepense, S. bicolor and S. propinquum. We searched for homologous IAP loci in related Zea and Setaria species to use as an outgroup, but were unable to identify a complete locus outside the Sorghum genus; only fragmented sequences were detected. Therefore, we were unable to include an appropriate outgroup. In the phylogenetic tree, the S. propinquum IAP locus clusters within the S. bicolor sequences, suggesting that the IAP locus of S. propinquum was recently replaced by a S. bicolor ortholog due to gene flow (Fig. 6). Two of the S. halepense alleles (located in contigs h1tg000292l and h2tg000951l, which are very similar to each other) cluster within the S. bicolor sequences, whereas the third one (located in contig h2tg000435l) clusters outside the S. bicolor clade. Our tree is compatible with a scenario in which S. halepense inherited one IAP locus (represented by the alleles located in contigs h1tg000292l and h2tg000951l) from S. bicolor and one IAP locus (located in contig h2tg000435l) from S. propinquum, and subsequently, the original IAP locus within the extant accession of S. propinquum used to assemble its genome was replaced by a S. bicolor orthologous sequence, most likely through hybridization with S. bicolor. Such a scenario highlights the complexity of studying the evolution of sympatric, cross-compatible species where gene flow occurs. The S. propinquum-derived IAP locus of S. halepense is represented by a single allele, probably because contig h2tg000435l is a collapsed contig. It should be noted, however, that alternative scenarios cannot be discarded—for instance, the Johnsongrass IAP copy that is not derived from S. bicolor could be derived from another wild relative donor (other than S. propinquum) not included in our analysis because no reference genomes are available. Future efforts constructing a more complete, and preferably a T2T genome assembly, would help clarify this point.

Fig. 6.

Maximum likelihood unrooted phylogenetic tree of IAP loci in the J-37 accession of S. halepense, S. propinquum, and 13 S. bicolor accessions. The J-37 genome contains two IAP loci, one that appears to be derived from S. bicolor (represented by two alleles, located in contigs h2tg00951 l and h1tg000292 l), and one that appears to be derived from another parental genome such as that of S. propinquum (represented by a single allele, located in contig h2tg000435 l, probably due to it being a collapsed contig). The tree was generated using IQ-TREE 3 (Wong et al. 2025), using the MFP option and 10,000 bootstrap replicates.

Expression of IAP Genes in Johnsongrass

The mechanism through which one or more genes clustered at the IAP locus modulate(s) pollen–pistil cross-(in)compatibility is unknown. To provide an initial understanding of the tissue specificity of individual IAP genes in S. halepense, uniquely mapped RNA-Seq reads were aligned to the S. halepense genome and S. halepense IAP locus genes were manually annotated and named based on homology to genes in the S. bicolor genome (v3.1.1) (Fig. 7) (Manual annotations are available in Table S4). Sobic.002G023200 and Sobic.002G023300 homologs appear to be tandem duplicates that are highly expressed in rhizomes and drought roots, with the Sobic.002G023300 homolog also showing increased expression in leaves and low expression in stigmas. In S. bicolor, they are functionally annotated to play roles in pathogen defense, which makes them strong candidates for self-recognition and/or mounting a defense against foreign pollen. Interestingly, the Sobic.002G023500, Sobic.002G023600, and Sobic.002G023700 homologs are highly expressed in stigmas and show varying expression levels across other tissues, although these genes are not functionally annotated in the S. bicolor genome. The remaining IAP locus gene, the Sobic.002G023400 homolog, may play a role in environmental or stress responses based on its higher expression in drought stems and roots compared to control stems and roots, respectively. Future efforts are needed to study pollen–pistil interactions between plants carrying different IAP alleles (Table 2) to clarify the functional genomics of this locus and to identify haplotypes that influence % seed set (a quantitative measure of the ability to hybridize).

Fig. 7.

Variance stabilized transformed (VST) expression levels based on uniquely mapped RNA-Seq reads to the J-37 accession of S. halepense IAP genes. S. halepense genes are named based on homology to the S. bicolor genome (v3.1.1). Color scale represents VST expression data; white indicating zero counts detected. The Sobic.002G023600 homologs on contigs h2tg000435 l and h2tg000951 l were not expressed in the studied tissues. The Sobic.002G023300 and Sobic.002G023400 homologs are absent from contig h2tg000435 l (refer to Table 2).

Discussion

S. bicolor-Derived Genes Are Vastly Over-Represented Relative to S. propinquum-Derived Genes in Accession J-37 of S. halepense

Previous putatively allotetraploid genome assemblies have identified distinct haplotypes representative of progenitor genomes (subgenomes). This was only possible at the gene level in the present Johnsongrass assembly, not even for whole contigs, due to the overwhelming abundance of S. bicolor-derived sequences relative to S. propinquum-derived sequences and the chromosomal rearrangements of S. bicolor-derived sequences. More Johnsongrass genome assemblies will be needed to determine if this is universally true of the species or more common in some geography (i.e. New World versus Old World accessions). Nevertheless, the J-37 accession of Johnsongrass appears to consist primarily of amplified and rearranged S. bicolor-derived chromosomes with surprisingly few sequences derived from S. propinquum dispersed inside them. We initially considered this result to be compatible with one of two scenarios for the formation of S. halepense as a species: a spontaneous whole-genome duplication in S. bicolor followed by limited hybridization with S. propinquum, or the fusion of two diploid gametes, one from each parental species, followed by repeated backcrossing to S. bicolor. Of these two scenarios, our results suggest that the latter is more likely for three reasons. First, our GenomeScope and Smudgeplots results showed clear 4x peaks (tetraploidy) and both AABB and AAAB dosage classes, consistent with an allotetraploid exhibiting segmental behavior. Second, the extensive chromosomal rearrangements in J-37 suggest cross-ploidy hybridization and not simply the admixture of two closely related diploid genomes. Third, the S. bicolor reference genome used in our evolutionary analysis is from B.Tx623, a modern cultivar, not a predomesticated S. bicolor accession that would have been present ∼1 MYA when S. halepense first emerged as a species. Since J-37 genes are poorly diverged from B.Tx623 genes, the simplest explanation is backcrossing with the domesticated crop and potentially shattercane, a dedomesticated, conspecific wild relative of sorghum. The greater evolutionary divergence between J-37 genes and S. propinquum genes compared to B.Tx623 genes [even though S. propinquum is a perennial wild relative and is therefore expected to evolve more slowly than the annual crop S. bicolor (Yue et al. 2010; Edwin and Thomas 2016; Wang et al. 2019)] further supports this view. Nevertheless, as de novo sequencing technologies advance and more Johnsongrass genomes are assembled, a clearer picture will emerge of how consistently the over-abundance of S. bicolor sequences occurs in other accessions, as well as specific chromosomal and whole-genome rearrangements.

S. propinquum-Derived Genes are Disproportionately Upregulated During Drought Stress

Most S. propinquum-derived genes have been lost from the J-37 genome, but those that remain contribute disproportionately to generalized stress responses. A common consequence of crop domestication is the loss of ecological fitness outside of agricultural contexts. Plant breeders select for alleles that confer higher yield based on the assumption that growers will provide nutrients, pest control, and water. From this perspective, it is possible that the original S. bicolor progenitor of Johnsongrass had more stress-tolerance alleles that were replaced over time through repeated backcrossing to cultivated sorghum as the progenitors of J-37 spread across the United States to Nebraska in the 1800s and 1900s. Conversely, S. propinquum stress-tolerance alleles may have provided greater ecological fitness outside of farm fields (e.g. roadsides and ditches) and may have been retained through positive selection. Future research is needed to more fully characterize genotype-by-environment interactions that differentiate among tissue-specific expression levels of S. bicolor- and S. propinquum-derived genes. Such studies would prove insightful into this living natural experiment wherein a wild relative has been extensively backcrossed to a crop, yet retained predomestication alleles with demonstrated impact on stress tolerance. We find it intriguing that the near elimination of S. propinquum-derived genes was achieved in J-37 through only indirect human intervention (growing and disseminating sorghum) at a level comparable to the success that plant breeders achieve when backcrossing quantitative traits. This observation highlights the profound impact that human activity can have on naturalized plant populations, even in the absence of direct efforts.

Genetic and Copy Number Variation for the IAP Locus

This is the first study to identify IAP genes with their tissue-specific expression profiles. Most studies of Johnsongrass reproductive biology have focused on asexual reproduction through rhizomes and sexual reproduction through post hoc evaluations of hybridization events with S. bicolor, which occur primarily through unreduced gamete formation in S. bicolor ovules (Hodnett et al. 2019). It is not yet known to what extent variation in IAP alleles or copy number impacts a priori outcrossing with sorghum, as the functional annotations of some genes are unknown. The discovery of both parental copies of IAP (three alleles; one contig containing the IAP locus was collapsed and there may be a fourth allele) in J-37 is especially intriguing because so much of the S. propinquum genome has been lost. Fluorescence in situ hybridization (FISH) probes specific to IAP could be designed to confirm how many IAP haplotypes there are. It is possible that genetic variation at IAP may impact outcrossing rates with diverse S. bicolor genotypes (including dedomesticated or feral accessions like shattercane) and/or other Johnsongrass accessions which, like J-37, may carry herbicide-resistance traits (Maity et al. 2022). If so, rates of crop-wild gene flow would be subject to variation in the frequency of 2n sorghum gamete production (Hodnett et al. 2019) as well as IAP genes. The present draft genome assembly will enable weed ecologists to determine whether specific IAP alleles in Johnsongrass are associated with higher or lower rates of gene flow from sorghum. If so, molecular markers could be developed to breed sorghum cultivars with IAP alleles associated with the lowest possible risk of interspecific cross-compatibility to minimize the escape of crop genes into wild populations.

Opportunities for Functional Genomics Research in Johnsongrass

Unsurprisingly, S. propinquum genes were among those to disproportionately impact root morphology and stress responses in Johnsongrass. One of the earliest evidences that S. propinquum was an ancestor of Johnsongrass was the presence of rhizomes in both species, which are rare in the Sorghum genus as a whole (reviewed by Klein and Smith 2021). The present draft genome assembly provides opportunities for researchers to compare alleles influencing rhizome production across accessions, geographies, and pest control methods (genotype-by-environment-by-management interactions; G × E × M) to develop a better understanding of the functional genomics of weediness and invasiveness. Physiological modeling methods that predict plant growth and productivity based on environmental thresholds known to impact plant development could be applied to rhizome biomass accumulation and other invasive features related to weediness to optimize local weed control tactics. This information can also be leveraged to develop novel weed control methods (e.g. selective herbicides, gene drives) that reduce Johnsongrass's reproductive success without harming cultivated sorghum (perhaps by targeting rhizomes that do not exist in sorghum). Such methods will require a reference genome to develop, test, and validate (Wong et al. 2022; Croghan et al. 2023; Kumam et al. 2023). Conversely, an improved understanding of rhizome biology can be leveraged by plant breeders developing perennial versions of sorghum (Washburn et al. 2013; Jessup et al. 2017a, 2017b; Cox et al. 2018; Habyarimana et al. 2018; Nakasagga et al. 2022) to select for heritable, but not aggressive, rhizome production within the environmental constraints of a given region.

The Next Frontiers for Johnsongrass Genomics: Pangenomes, Interspecific Pangenomes, and Clarifying the Genetic Drivers of Crop-Wild Gene Flow

We observed extensive genome restructuring in J-37 relative to S. bicolor and S. propinquum based on synteny analyses and contiguous sequences within the longest possible individual contigs. Other researchers have documented wide variations in ploidy in sorghum-Johnsongrass hybrid populations (Ohadi et al. 2017; Hodnett et al. 2019; Sias et al. 2023). Genome instability, aneuploidy, and segmental polyploidy are well-documented in crop-wild complexes where hybridization across ploidies occurs (Gonzalo 2022; Deb et al. 2023; Zhang et al. 2024, 2026). These observations will undoubtedly complicate the assembly of pangenomes for Johnsongrass, or sorghum-Johnsongrass hybrid populations, to support diverse research objectives. Nevertheless, these are the sorts of challenges that must be addressed by breeders who use crop-wild relatives as donors of stress-tolerance alleles and by federal regulatory agencies who evaluate the risk of increasingly sophisticated biotechnologies escaping from crops to their wild relatives. In the short term, a workable strategy may be to assemble pangenomes within ploidy levels and local geographies where identical-by-descent relationships can be determined to maximize similarities in genome structure. In addition, conserved syntenic blocks could serve as reference sequences for initial read mapping, gene discovery, and molecular marker development.

Our results complement previous studies and suggest that the genetic mechanisms underpinning variable rates of gene flow between sorghum and Johnsongrass include variation in both species for aneuploid gamete production, variation in genome structure, which may influence meiotic recombination; and cross-(in)compatibility alleles (including, but not limited to, IAP). Therefore, a workable genetic risk-assessment framework for crop gene escape to Johnsongrass populations should account for the physiology and genome biology of both species. We propose the following model:

1. Characterize the genome structures of diverse local and regional accessions to identify similarities and develop groups. Assembling the genome of S. halepense accession(s) from nonagricultural regions of Africa, which would be associated with little to no crop-wild gene flow, could also reduce the risk of contig collapse due to recent introgressions of similar genomic regions. Methods should include cytology experiments (e.g. chromosome staining, flow cytometry).

2. Characterize the segregation of gamete (Lyu et al. 2021) or progeny genome structures within groups as above.

3. Characterize IAP alleles within groups.

4. Select a panel of sorghum crossing partners for the Johnsongrass groups. The panel should include sorghum genotypes with various IAP alleles and with different rates of diploid (2n) gamete production (Hodnett et al. 2019).

5. Perform reciprocal cross-pollinations, characterize pollen tube growth phenotypes in pistils [an excellent example is provided in Hodnett et al. (2005)], and determine % seed set.

6. Assess the % germination from reciprocal crosses and characterize hybrid genome structures as above.

7. Assess hybrid fertility and the segregation of genome structures and IAP alleles in the F₂ generation.

This genetic framework should be deployed in a way that manages potential variation in environmental and experimental methodology (e.g. temperature, pollen load; Sias et al. 2023), which could confound results. Rates of hybrid formation, conserved genome structures, and the frequency of crop alleles in F₂ populations, should provide the most biologically accurate information about the true genetic risk of crop-wild gene flow between specific sorghum and Johnsongrass genotypes. With well-characterized genomes for these parent plants and progeny, sorghum breeders could identify molecular markers for regulators of diploid gamete production, sites of meiotic recombination, and IAP alleles that provide the best barriers to outcrossing. They could select for beneficial alleles at candidate loci that serve as complementary, redundant barriers to crop-wild gene flow and minimize the risk of stable introduction of crop genes into Johnsongrass populations.

Conclusions

The present study reports the first high-quality draft genome assembly for Johnsongrass, a troublesome weed of major global economic importance. The J-37 accession was selected to provide relevant molecular information for managing Johnsongrass populations that are sympatric weeds of cultivated sorghum and carry de novo herbicide-resistance traits due to selective pressure by pesticides. We anticipate that the J-37 genome will complement efforts in the wider scientific community to assemble more Johnsongrass genomes that are representative of diverse evolutionary histories.

The J-37 genome is composed of a surprisingly high proportion of S. bicolor-derived sequences relative to S. propinquum-derived sequences, but it has retained both ancestral copies of the IAP locus, which could modulate its prezygotic sexual compatibility with other Sorghum species. The complete loss of distinct parental subgenomes strongly suggests that J-37 could function as a diploid or aneuploid during meiosis, although ploidy information for this accession is 2n = 4x = 40. The availability of this draft genome assembly will promote new studies of Johnsongrass sexual reproductive biology to complement the many existing studies of its asexual reproductive biology (rhizomes). It will promote gene discovery and functional genomics investigations of traits that are unique to Johnsongrass and may lead to novel methods for preventing crop-wild hybridization and/or breeding perennial grain, forage, and bioenergy crops.

Future efforts and ongoing technological developments are needed to make assembling complex plant genomes faster in order to manage weeds, to breed more resilient crops utilizing crop-wild relatives, and to prevent or mitigate unintended crop-wild gene flow, depending on research objectives.

Materials and Methods

Plant Materials

Tissue from a single plant of Johnsongrass accession, “J-37,” collected in Buffalo County, Nebraska was used in this draft genome assembly (Fig. 1). It is only partially self-fertile. While its anthers appear healthy and produce abundant pollen, only 5% to 10% of ovules set viable seed. The genome donor was grown from seed in the S₃ generation following collection. It is heterozygous for a target-site mechanism of tolerance to ALS-inhibiting herbicides (ALS-inhibitors) (Yu et al. 2008) that evolved de novo due to historical exposure, not crop-wild hybridization (Werle et al. 2016).

High Molecular Weight DNA Extraction

Long-read sequencing is optimized by using high molecular weight (HMW) genomic DNA (gDNA). A low-shearing HMW gDNA extraction protocol (Mayjonade et al. 2016) was modified to support the draft genome assembly (Supplementary Methods S1). The molecule size and quality of the HMW gDNA was evaluated at the University of California-Davis DNA Technologies Core (Davis, CA) using pulsed-field gel electrophoresis. Two duplicate samples (technical replicates) were submitted and the sample with the best quality was selected for sequencing (below).

Long-Read Whole-Genome Sequencing

A HiFi SMRTbell library was constructed from the superior HMW DNA sample (above) at the University of California-Davis DNA Technologies Core (Supplementary Methods S2). The resulting 21-kb average library was sequenced using five 8 M SMRT cells (Pacific Biosciences, Menlo Park, CA; Cat #101-389-001), Sequel II sequencing chemistry 2.0, and 30-h movies each on PacBio Sequel II and Sequell IIe sequencers. The circular consensus analysis was performed in SMRT® Link v9.0 to generate the HiFi reads. BBMAP's readlengths.sh was used to produce a histogram summary from the HiFi reads, which were then plotted in R.

Hi-C Sequencing

Young leaf tissue was provided to Phase Genomics (Seattle, WA) and used to prepare deep sequencing libraries with Proximo Library preparation reagents (Supplementary Methods S2). Library quality was assessed with qPCR and a Bioanalyzer before sequencing.

De Novo Draft Genome Assembly

GenomeScope 2.0 (Ranallo-Benavidez et al. 2020) was run with the 21 bp-k-mers called by FastK 1.1.0 (https://github.com/thegenemyers/FASTK) to estimate the genome characteristics including ploidy. The output for FastK was also used to run Smudgeplot 0.5 (Ranallo-Benavidez et al. 2020) in order to determine inheritance patterns. Hifiasm-0.15.5 (Cheng et al. 2021) was used to assemble the Hi-C and HiFi reads with the parameters “–n-hap 3 -s 0.45–hom-cov 30.” Contigs were mapped to the S. bicolor reference genome (v3.1.1) by Mummer 4.0 (Marcais et al. 2018) for the evaluation of assembly quality, and minimap2 (Li 2018) was used to align HiFi reads back to the contigs to evaluate the read coverage with BBmap (Bushnell 2014).

RNA-Seq Experiments

Water Deficit Study

To generate a wide variety of mRNA sequences suitable for validation of the de novo draft genome assembly, J-37 seedlings (S₄ generation; progeny of the accession used as the genome donor) were subjected to water deficit and control treatments (two environmental conditions) followed by destructive harvest, tissue subsampling, and total RNA extraction (Fig. S5). Prior to initiation of the experiment, the amount of water needed to confer water deficit and control conditions was identified as follows. Plastic pots with drainage holes (17.78 cm tall, 16.5 cm dia.) were filled with medium-grit sand (QUIKRETE®, Atlanta, GA). Saturation capacity of the pots (200 ml; the daily application rate for the control condition) was determined by applying a nutrient solution (Miracle Grow® All Purpose Plant Food dissolved in tap water at the daily recommended application rate) using a graduated cylinder to determine the volume necessary for the solution to run from the bottom of each pot. The water deficit condition was 25% of the daily application rate of the control (50 ml).

The experiment was initiated in early April 2019 in a greenhouse at UNR's Valley Road Experiment Station (Reno, NV) by sowing 10 seeds per pot in 25 pots. The number of seeds used was well in excess of the tissue needed to ensure selection of similar-sized, healthy plants for treatment imposition and RNA extraction. All seeds were sown 3-cm deep in sand that had been premoistened to saturation (described above). All pots were provided 200-ml nutrient solution daily at 09:00 for 1 week to ensure uniform emergence and growth. The plants were maintained under optimal conditions until the first leaf collar was fully expanded among most plants, which were then thinned to one plant per pot, selecting for similar-sized plants. Remaining plants were divided evenly into water deficit and control treatments and the two conditions were imposed for 3 weeks by adjusting daily nutrient solution applications (100% vs 25%). A total of three representative, similar-sized plants were selected for destructive harvest and tissue subsampling from both control and water deficit conditions (Fig. S5). Tissues harvested from each condition included the leaf, stem, and root (three tissues in two conditions; six samples total). Pollen, stigma + styles, seeds, and rhizomes (four samples total; pollen and stigma + style samples were composites containing many pollen grains and ∼15 stigma + styles, respectively) were harvested separately from mature plants grown under control conditions after 3 months of growth. The 10 samples were placed in separate MP Biomedicals (Santa Ana, CA) Lysing Matrix “A” tubes for plant tissues, flash frozen in liquid nitrogen, and used for RNA extraction.

RNA Extraction and Illumina Sequencing

Total RNA was extracted from tissues using a Sigma-Aldrich® Spectrum Plant Total RNA Kit (Darmstadt, Germany) with the following modification: Spectrum Plant Total RNA Kit lysis buffer with β-mercaptoethanol was added to Lysing Matrix A tube containing frozen plant tissues and placed in a bead beater for homogenization of tissues with RNA stabilizing lysis buffer. Optional DNase treatment was included in the extraction and total RNA was eluted in 30 µl molecular biology grade water. Absorbance ratios (A260/280 and A230/260) were checked on a Nanodrop® spectrophotometer (NanoDrop 2000; ThermoFisher; Waltham; MA; USA). Purified RNA was checked for quality on a Bioanalyzer (Agilent Bioanalyzer 2100; Agilent; Santa Clara, CA; USA) and 10 samples having RIN scores >6.8 were selected for sequencing at the University of California-Davis DNA Technologies Core as follows. Libraries were prepared by pooling mRNA from tissues from the following samples: mature leaf, stem, and root from both control and water deficit condition plants, seeds, rhizome, stigma + styles, and pollen. Poly-A enriched single-strand sequencing was conducted on one lane on the HiSeq platform using 150-bp paired-end reads. Additionally, a single, separate library was prepared by pooling all samples and conducting single-strand sequencing on one lane of a MiSeq using 300-bp paired-end reads.

Transcriptome Assembly

RNA-Seq reads were aligned back to the S. halepense assembly using HISAT2 v2.2.1 (Kim et al. 2019). Quality control (QC) analysis indicated high-quality results, with 90% to 96% of reads successfully mapped. Reads were filtered with fastp v0.12.4 (Chen et al. 2018), the resulting filtered reads were used for a TRINITY v2.13.2 (Haas et al. 2013) assembly as an input for the MAKER v3.01.03 (Campbell et al. 2014a; Campbell et al. 2014b) annotation pipeline below. Additionally, the mapped RNA-Seq reads BAM file was supplied to TRINITY for genome-guided assembly of transcripts utilizing individually mapped RNA-Seq sample BAMs merged into one large multisample BAM file. The genome-guided assembly mode was executed using the merged BAM file and the option “–genome_guided_max_intron 10000.”

Repeat Analysis

RepeatModeler v2.0.3 (Flynn et al. 2020) was employed to predict repeats de novo. RepeatMasker v4.1.2 (Smit 2015) ran in repeat identification mode, utilizing the RepBase (Bao et al. 2015) library to identify previously identified S. bicolor repeats, as well as de novo predicted RepeatModeler repeats. These were merged to create a complex_mask version, which was hard-masked.

Gene Annotation

To enrich the annotation analysis, a training set was created from closely related species. Data for protein (pep) and CDS files from Ensembl Plants (Yates et al. 2022) were downloaded for the following species (Assembly Name): S. bicolor (Sorghum_bicolor_NCBIv3), Oryza sativa (IRGSP-1.0), Zea mays (Zm-B73-REFERENCE-NAM-5.0), Arabidopsis thaliana (TAIR10), and Setaria italica (Setaria_italica_v2.0).

The repeat-masked genome served as the basis for genome annotation with MAKER-P (Campbell et al. 2014a; Campbell et al. 2014b). The transcriptome assembly and the combined_training_set fasta files for both CDSs and peptides were used as inputs for annotation. MAKER-P uses ab initio gene finding, which was performed using SNAP (v2013_11_29) (Price and Carbone 2005) and AUGUSTUS v3.4.0 (Hoff and Stanke 2019). Following the initial annotation, SNAP and AUGUSTUS were retrained using 1000-bp padded predicted transcripts and MAKER-P was run subsequent times. Two additional annotation rounds were conducted with retrained parameters, resulting in a total of three rounds of annotations.

QC was conducted for each annotation round, involving BUSCO v5.4.4 (Manni et al. 2021) using the poales_odb10 database for score and AED assessment. The authors of MAKER (40) describe AED <0.5 as good annotations and AED <0.3 as high-quality annotations. AED distributions were analyzed to evaluate annotation quality and gene models with AED values of 0.3 or better were retained. Filtering based on AED < 0.3 and < 0.5 was performed to create two high-quality annotation sets (Table S2).

Evolutionary Analysis of the S. halepense Genome

Sequence Alignments

For each Johnsongrass protein-coding gene model, we obtained pairwise and multiple-sequence CDS alignments with the closest orthologs in the S. bicolor B.Tx623 reference genome, S. propinquum (Tao et al. 2021) and Z. mays (B73 v5; NCBI RefSeq Assembly GCF_902167145.1; annotation Zm-B73-REFERENCE-NAM-5.0). Alignments were obtained by translating in silico the CDSs to protein sequences, aligning the protein sequences using PRANK v.170427 (Löytynoja 2014), and then using the resulting alignments to guide the alignment of the CDSs.

Phylogenetic Trees

For each S. halepense–S. bicolor–S. propinquum-Z. mays CDSs alignment, we obtained a phylogenetic tree using RAxML, version 8.2.10 (Stamatakis 2014). We used the GTR + Gamma model, set Zea mays (Hufford et al. 2021) as the outgroup, and performed 100 rapid bootstrap searches per tree.

Synonymous and Nonsynonymous Divergence Estimation

For each S. halepense–S. bicolor and S. halepense–S. propinquum pairwise CDS alignment, we estimated a d_N, d_S, and d_N/d_S value using the M0 model of the codeML program of the PAML package, version 4.4d (Yang 2007). For each S. halepense–S. bicolor–S. propinquum-Z. mays CDS alignment, we estimated one d_S value for each branch using the free-ratios model of the codeML program of the PAML package, version 4.4d (Yang 2007).

Synteny Analysis

Synteny analyses among S. halepense and its parental genomes were conducted using MCScanX (Wang et al. 2012, 2024). This program uses protein sequence alignments among proteins, as well as their annotation locations within the GFF files, to determine syntenic blocks by collinearity of genes. A three-way comparison of S. halepense, S. bicolor, and S. propinquum was conducted by converting the protein sequence IDs in each species to shorter names (SH, SB, and SP, respectively) to run BLAST. The GFF file chromosomes/scaffold names were converted to two-letter codes with a number (e.g. SB1 for S. bicolor chromosome 1) for MCScanX to properly run.

OrthoFinder Analysis

We ran OrthoFinder v.2.5.5 (Emms and Kelly 2019) using the entire S. halepense, S. bicolor, and S. propinquum proteomes as input. We then classified the resulting orthogroups into seven categories according to the species represented in each of them. A Venn diagram was built using InteractiVenn (Heberle et al. 2015).

IAP Locus Phylogenetic Tree

All IAP locus sequences (i.e. the genomic regions encompassing genes Sobic.002G023200–Sobic.002G023700 or their orthologs) were aligned using Clustal Omega (Sievers et al. 2011). We edited the alignments manually using BioEdit (Hall 1999) and generated a Maximum Likelihood tree using IQ-TREE 3 (Wong et al. 2025) (using the MFP option and 10,000 bootstrap replicates). We restricted our analysis to the 16,708 nucleotide positions that were present in all 17 sequences.

Differential Gene Expression Analysis of S. bicolor- and S. propinquum-Derived Sequences in the J-37 Accession of S. halepense

RNA-Seq data from the ten samples used for gene annotation was analyzed to investigate differential gene expression between sequences derived from the two parental genomes. High-quality reads were aligned to the S. halepense draft genome using the STAR alignment tool (v2.7.11b) (Dobin et al. 2012) with default parameters. Following alignment, the expression level of each homeologous gene derived from both parental genomes was computed using FeatureCounts from the Subread package (v2.0.6) (Liao et al. 2013) across all predicted gene models in the S. halepense assembly. DESeq2 (v1.44.0) (Love et al. 2014) was used for differential expression analysis, applying its median of ratios normalization method to account for differences in sequencing depth across samples. High-confidence homeologous gene pairs were defined as pairs based the alignment above where (i) both genes can be aligned to the same genes in both S. bicolor and S. propinquum and (ii) each gene shows higher similarity to one of them. These pairs were identified using an in-house R script based on the alignments mentioned above. Differentially expressed homeologs were identified using an adjusted P-value threshold of <0.05 (FDR) and a log2 fold-change > ± 2. The results were visualized in R (version 3.5.0). Gene Ontology (GO) enrichment analysis was performed using the PANTHER Gene List Analysis tool (Mi et al. 2013). PANTHER compares the frequency of GO terms in the input gene set with their frequency in a reference set (here, S. bicolor; B.Tx623 v3.1.1), identifying terms that are significantly over- or under-represented. The over-representation test was performed using a binomial test, and a Bonferroni correction was applied to adjust for multiple comparisons across GO terms. Only GO terms with Bonferroni-corrected P-values < 0.05 were considered statistically significant, ensuring that the identified enrichments were unlikely to result from random chance and reflected meaningful biological insights.

Supplementary Material

Supplementary material is available at Genome Biology and Evolution online.

Author Contributions

M.K.Y., Y.J., A.J., and J.L. obtained funding. M.K.Y., Y.J., D.N., D.A.P., H.V.G., J.P., and J.N.W. designed and conducted the studies. M.K.Y., Y.J., D.N., D.A.P., H.V.G., and J.P. wrote the manuscript. All authors read and approved the final manuscript.

Funding

This work was supported by startup funding provided by the University of Nevada, Reno to M.K.Y.; startup funding provided by Texas Tech University to Y.J.; and partial funding from DuPont (now Corteva AgriScience) to A.J. This work was additionally supported by a grant from the National Institute of Food and Agriculture (2017-33522-27086) to J.L., M.K.Y., and A.J.; and by a grant from the National Institute of General Medical Sciences to J.P. (GM103440).

Data Availability

Seeds from the J-37 accession used in this draft genome assembly are being made publicly available through deposit into the USDA National Plant Germplasm System (NPGS; https://npgsweb.ars-grin.gov/gringlobal/search). Small aliquots of seeds are also available upon reasonable written request to M.K.Y. at myerka@unr.edu. The genome sequencing reads, RNA-Seq, genome assembly and gene annotation are available under NCBI BioProject ID PRJNA1201239 and are accessible with the following link: https://www.ncbi.nlm.nih.gov/sra/PRJNA1201239. In addition, the genome assembly, gene annotations, and all other data presented herein are managed by Y.J. and can be downloaded at https://www.depts.ttu.edu/igcast/Genomic/Johnsongrass/index.php.