Genomic revolution of US weedy rice in response to 21st century agricultural technologies

Understanding how genomes adapt is a central question in evolutionary biology. Crop domestication provides a model for adaptive responses to selection, and agricultural weeds—especially those closely related to crops—offer additional insights due to potential gene flow with cultivars. Weedy rice (Oryza spp.), a de-domesticated form of cultivated rice (O. sativa), has independently evolved multiple times and is highly adapted to rice fields, causing substantial yield and quality losses worldwide and in the US. Two distinct morphotypes historically predominated in the southern US: strawhull (SH), descended from indica varieties, and blackhull-awned (BHA), derived from aus varieties. Despite proximity to US cultivars (tropical japonica), outcrossing rates among SH, BHA, and cultivars were historically <1%. In 2002, non-transgenic herbicide-resistant (HR) rice (Clearfield) was introduced, conferring resistance to imidazolinone herbicides via ALS mutations (initially G654E in CL121/CL141, later primarily S653N in CL161 and successors). HR rice adoption peaked at ~65% of acreage and remains ~35%. Concurrently, hybrid rice technology expanded (first commercialized in 2000; now ~50% of acreage), increasing volunteer rice and outcrossing opportunities with weeds via segregating volunteers that bridge gene flow. This created a two-decade natural experiment: historically selfing, genetically distinct SH and BHA weedy strains have experienced strong selection for herbicide resistance coincident with enhanced crop–weed hybridization. Early reports documented HR weedy rice soon after Clearfield adoption, with high resistance frequency by 2010. It remained unclear how weedy rice continued to evolve under sustained HR use and hybrid rice presence. This study uses whole-genome resequencing to assess how contemporary US weedy rice genomes have changed post-introduction of HR and hybrid cultivars. Specific questions: (1) How do contemporary genomes differ from historic SH and BHA strains? (2) After crop–weed hybridization (~50:50 genomes), does selection bias genomes toward weed or crop ancestry? (3) Does selection elevate known weed- or crop-specific alleles at predicted adaptive loci (e.g., ALS for herbicide resistance; Rc for dormancy)? Findings reveal a genomic revolution in US weedy rice over the past 20 years that has reshaped crop–weed dynamics and mechanisms of adaptation.

Sampling and sequencing: Seeds from 48 maternal weedy rice plants were collected during harvest (late August 2018) from five Arkansas fields representing key southern US rice production. Prior work indicates no geographic genetic structure in US weedy rice; thus, this sampling is considered representative. One seed per maternal plant was germinated; leaf tissue was harvested at seedling stage for DNA extraction (modified CTAB). Plants were selfed one additional generation in the greenhouse before DNA collection. Library prep and sequencing: Illumina libraries (Nextera DNA Flex with CD indexes) were prepared; pooled libraries were sequenced (paired-end) on HiSeq X10 by Novogene. Contemporary reads were combined with previously published datasets, yielding 146 total samples (48 contemporary weeds; 22 historical US weeds—11 SH and 11 BHA; 49 cultivated—10 aus, 5 aromatic, 12 indica, 12 temperate japonica, 10 tropical japonica; 27 wild) and ~19.34 million SNPs. Wild accessions were excluded from analyses of US weed evolution after confirming no contribution. Read processing and variant calling: Raw reads were quality-trimmed (Trimmomatic), aligned to the Oryza sativa Nipponbare MSU v7.0 reference (BWA). Alignments were processed with SAMtools; variants were called (bcftools mpileup) and filtered using VCFtools to remove indels, MAF <0.05, and sites out of Hardy–Weinberg equilibrium (p<1e-7). Analyses excluding wild samples used filtered VCFs with wilds removed. Population structure: PCA and ADMIXTURE (via PLINK and ADMIXTURE) were used to assess structure; visualization via pong. Cross-validation suggested K=6 optimal, but K=5 better reflected biologically meaningful groups. Contemporary weeds were categorized by predominant weedy ancestry into SH-like, BHA-like, a homogeneous BHA-derived subgroup (‘beta’), and one ‘complex’ SH–BHA–tropical japonica admixed accession. Genetic diversity and generations since hybridization: Genome-wide heterozygosity per accession (PLINK –het) was estimated. A custom script identified ancestrally informative SNPs (fixed differences between presumptive weed and crop ancestors) to estimate generations since hybridization, assuming selfing after a single outcross and ~50% reduction in heterozygotes per generation; estimates account for seed bank/dormancy in interpretation. Local ancestry: Loter inferred local ancestry tracts across the genome using historical weed (SH, BHA) and tropical japonica as references. A custom script quantified proportions of weed vs crop ancestry per chromosome and genome-wide, and mapped gene coordinates (ALS on chr2; Rc on chr7) to Loter bins for candidate adaptive regions. Differentiation and diversity scans: Sliding-window FST (Weir–Cockerham; VCFtools, 500 kb windows, 250 kb step) compared each contemporary group against its predominant historical weed ancestor and tropical japonica cultivars to identify weed-like vs crop-like regions (notably ALS and Rc). Nucleotide diversity (π) and heterozygosity patterns were also assessed. ALS haplotype analysis: Reads mapping to ALS were extracted (SAMtools), assembled to a reference ALS sequence (Geneious), manually phased to remove heterozygosity from consensus, and aligned for median-joining haplotype network construction (PopART). HR-associated mutations S653N and G654E were annotated; common ALS resistance mutations Ala122 and Trp574 were specifically checked. Herbicide resistance phenotyping: Seeds were grown in greenhouse pots (50:50 field soil:potting mix), 16 h daylength, 25–35 °C. At 3-leaf stage, plants received two imazethapyr applications (70 g ai ha−1) ten days apart (187 L ha−1 spray volume; two 800067 nozzles at 46 cm spacing). Each treatment had three replicates with non-treated checks. Visible injury scored 3 weeks after the second application on 0–100% scale (0 no injury; 100 dead). Samples were binned as highly resistant (0–32% injury), moderate (33–67%), susceptible (68–100%), or segregating. Data and code availability: New sequence data deposited under BioProject PRJNA847219 (SAMN28922700–SAMN28922747). Custom Python scripts available at https://doi.org/10.5281/zenodo.6954804.

• Population structure and hybrid origins: PCA separated japonica vs indica lineages (PC1=22.8%) and distinguished SH/SH-like vs BHA/BHA-like (PC2=15.6%). All but four contemporary accessions fell between historical weedy strains and US cultivars, indicating hybrid origins. ADMIXTURE at K=5 showed 35/48 (72.9%) contemporary weeds admixed (>15% membership in ≥2 groups). Of the admixed accessions, 28/48 (58.3%) were BHA-derived and 6/48 (12.5%) SH-derived; 9/48 (18.8%) formed a homogeneous BHA-like ‘beta’ subgroup; 4/48 (8.3%) were indistinguishable from historical SH (non-hybrid); 1 accession had complex SH–BHA–tropical japonica ancestry.
• Generations since hybridization and diversity: Contemporary hybrid-derived weeds had elevated heterozygosity relative to crops; SH-like averaged higher heterozygosity than BHA-like; the beta subgroup showed significantly lower heterozygosity, consistent with more generations of selfing. Heterozygosity-based estimates placed most samples at ≥5 generations post-hybridization, with only eight <3 generations, likely conservative due to assumptions of return to strict selfing.
• Genome-wide local ancestry bias toward weed ancestry: Loter indicated contemporary genomes deviated from the neutral 50:50 expectation, showing mean weed-ancestry assignments of 74.1% (BHA-like) and 69.2% (SH-like) across the genome, with similar per-chromosome biases (Table 1), suggesting selection favoring weed genomic components or purging of crop alleles after hybridization.
• Targeted genomic regions under selection: Sliding-window FST revealed a crop-like region around ALS on chr2 in contemporary weeds relative to crops vs weedy ancestors, consistent with adaptive introgression of HR alleles. On chr7, a weed-like region around Rc (seed dormancy) was evident in BHA-like (but not SH-like) weeds. Loter identified a large crop-like haplotype block encompassing ALS in BHA-like weeds; Rc showed weed-like ancestry enrichment.
• ALS haplotypes and resistance mechanisms: The ALS haplotype network formed two divergent haplogroups: cultivar-derived (tropical japonica) and weedy. Most contemporary weeds carried the S653N mutation within the CL161-like haplotype background; two (E08, E09) carried the older G654E mutation consistent with early HR cultivars CL121/CL141. Four SH-like non-hybrid weeds showed resistance without hybrid ancestry: A01 and A08 carried G654E (standing variation previously detected at low frequency), while A05 and A06 carried S653N, indicating convergent evolution. No Ala122 or Trp574 mutations were detected.
• Phenotypic resistance: 34/48 (70.8%) were highly resistant; 4/48 (8.3%) moderate; 8/48 (16.7%) segregating; 2/48 (4.2%) susceptible. Overall, 46/48 (95.8%) displayed some degree of resistance. The two susceptible plants were hybrids from fields not using Clearfield, likely progeny of segregating parents.

The predominance of crop–weed hybrid derivatives among contemporary US weedy rice, despite historically low hybrid persistence, suggests that while F1 hybrids may have low fitness, subsequent selfing and segregation can assemble favorable allele combinations enabling persistence under strong herbicide selection. Local ancestry analyses show that, irrespective of initial hybrid origins (BHA-like, beta, SH-like), genomes have shifted toward their weedy ancestors, indicating selective maintenance of weed-adaptive genomic regions and/or purging of maladaptive crop alleles. This selective landscape has facilitated adaptive introgression at ALS from HR cultivars, while retaining or re-establishing weed-like haplotypes at loci such as Rc associated with seed dormancy. The detection of convergent evolution of ALS resistance (S653N) in non-hybrid SH-like weeds, and selection on standing G654E variation, underscores multiple genetic routes to herbicide resistance. These findings have implications for gene flow and allele escape (transgenic or otherwise) in agroecosystems and illuminate the genome-wide processes of adaptive introgression under contemporary agricultural practices.

Whole-genome resequencing of contemporary US weedy rice reveals a marked shift since the introduction of herbicide-resistant and hybrid rice technologies: most contemporary weeds are crop–weed hybrids whose genomes have subsequently evolved toward their weedy ancestors. Adaptive introgression has driven the spread of cultivated ALS resistance alleles, while convergent mutations and selection on standing variation also contribute to resistance. Genome-wide ancestry patterns and targeted differentiation at ALS and Rc illustrate how strong selection and increased gene flow have reshaped weedy rice evolution. These results point to a new era of weedy rice in the US, with widespread herbicide resistance and altered crop–weed dynamics, emphasizing risks of reliance on single control strategies.

• Greenhouse selfing generation: DNA was obtained from plants selfed one additional generation under greenhouse conditions; thus, sequenced genotypes may not perfectly match field-collected genotypes for segregating variants.
• Generation estimates: Heterozygosity-based estimates of generations since hybridization assume a return to complete selfing after a single outcross and ~50% reduction of heterozygotes per generation; estimates may be conservative.
• Sampling scope: Contemporary samples were collected from five Arkansas fields; while prior studies indicate minimal geographic structure in US weedy rice (supporting representativeness), geographic nuances outside the sampled region cannot be fully excluded.
• Reference panels: Wild rice accessions were excluded after analyses indicated no role in US weedy rice evolution; undetected contributions from unsampled lineages are unlikely but cannot be entirely ruled out.