A yield‐reducing introgression linked to sd1 in rice Authors: Flavia Furlan, Brijesh Angira, Maria Montiel, Jonathan Richards, Tommaso Cerioli, Christopher Hernandez, Roberto Fritsche‐Neto, John Platten, Jomar Punzalan, Susan McCouch, Kelly Robbins, Xueyan Sha, Adam Famoso Published: 2026 ABSTRACT AbstractThe semi‐dwarf (sd1) allele was central to the Green Revolution, but its widespread use in US tropical japonica rice (Oryza sativa L.) may have introduced unintended effects through linkage drag. This study examined the genomic region surrounding sd1 using genome‐wide association and haplotype analyses in a multiparent population (MP6/8) evaluated across environments and in a biparental population (MP‐D) for validation. Genomic and phenotypic evidence indicated that an indica‐derived haplotype upstream of sd1 persists in US tropical japonica and global indica semi‐dwarf germplasm and is associated with reduced grain yield. Genome‐wide association analysis confirmed sd1 as the primary determinant of reduced plant height and identified a quantitative trait locus at 36.69 Mb upstream of sd1 that was consistently associated with reduced grain yield (0.6–0.9 t ha−1) across populations and environments. Haplotype analysis identified an indica‐derived Presidio haplotype spanning this upstream region was associated with significant yield reductions in both populations. Pedigree analysis traced this unfavorable haplotype to the Taichung Native 1 donor lineage, representing one historically deployed sd1‐linked introgression class. Although its frequency has declined in modern cultivars, its persistence highlights the long‐term impact of linkage drag following major trait introgression. These findings emphasize the value of marker‐assisted selection and targeted recombination to separate favorable dwarfing alleles from linked yield penalties and sustain yield gains in US rice improvement programs. Abbreviations analysis of variance best linear unbiased estimate Diversity Arrays Technology Sequencing Genome Analysis Toolkit genomic variant call format genome‐wide association study honestly significant difference International Rice Research Institute Louisiana State University minor allele frequency principal component analysis quantitative trait locus single‐nucleotide polymorphism temperate japonica Taichung Native 1 tropical japonica whole‐genome sequencing INTRODUCTION The semi‐dwarf (sd1) allele was a central genetic innovation of the Green Revolution, enabling reduced plant height, increased harvest index, and improved nitrogen responsiveness in rice (Oryza sativa L.) (Ferrero‐Serrano et al., 2019; Hedden, 2003; Khush, 2001). The causal sd1 deletion allele originated from the Taiwanese landrace Dee‐geo‐woo‐gen and was independently incorporated into the indica cultivars Taichung Native 1 (TN1) and IR8, which subsequently served as major donors of semi‐dwarfism in global rice breeding. In the United States, where long‐grain rice is predominantly tropical japonica, the sd1 deletion allele was introduced through both TN1‐ and IR8‐derived backcrossing schemes beginning in the late 1960s, establishing two primary sd1 lineages in modern US germplasm (Dalrymple, 1980; Hargrove et al., 1980). Although TN1 and IR8 carried the same sd1 deletion allele, their distinct breeding histories led to different donor‐derived genomic segments flanking the sd1 locus. The TN1 lineage entered US breeding programs through the advanced backcross line IR659‐10‐8‐3 (Bluebelle*6/TN1), which was subsequently used in the development of Bellemont and Lemont (Dalrymple, 1987). In contrast, the IR8 lineage was introduced via a backcross to the California line R17 and following selection, leading to the variety L201 (Dalrymple, 1980). Introgression of sd1 into US long‐grain rice involved interspecific hybridization between indica donors and tropical japonica backgrounds, resulting in the transfer of large genomic regions of indica ancestry flanking the sd1 locus that persist in contemporary cultivars (Angira et al., 2019; Zhao et al., 2010). Such introgressions can impose linkage drag, defined as the co‐introgression of deleterious alleles tightly linked to a target locus under strong selection. Large introgressed haplotypes can persist for extended periods, especially when recombination is limited, potentially reducing genetic diversity and constraining long‐term breeding progress (Qian et al., 2017; Shi et al., 2022; M. Wang et al., 2020). In rice, genomic regions linked to sd1 have been associated with unfavorable agronomic effects in non‐US germplasm, including reduced spikelet number (Murai et al., 2002) and the loss of favorable drought‐related alleles (Vikram et al., 2015). However, in US long‐grain rice, the fine‐scale haplotype structure surrounding sd1 and whether linked donor‐derived segments contribute to yield penalties have not been systematically resolved. Considering the pivotal importance of sd1 in US rice cultivation and the historical reliance on two principal donor lineages, a detailed assessment of genomic variation associated with this locus is warranted. In this study, we integrated genome‐wide association analysis, high‐resolution haplotype characterization, and pedigree‐based ancestry reconstruction to investigate genomic variation surrounding sd1 in two complementary mapping populations. Our objectives were to (i) determine whether unfavorable alleles linked to sd1 persist in US long‐grain germplasm, (ii) resolve sd1‐associated haplotypes and quantify their effects on grain yield, and (iii) trace the historical origins and distribution of these haplotypes within US breeding programs and global elite germplasm. Together, these analyses clarify how historical sd1 introgressions shaped local genomic architecture and identify a donor‐derived haplotype associated with reduced yield in modern US long‐grain rice. MATERIALS AND METHODS Population development This study included two different populations: a multiparent group (MP6/8) and a biparental group (MP‐D). The MP6/8 population was derived from eight southern US long‐grain varieties selected for complementary agronomic traits, including semi‐dwarfism, grain quality, ratoon yield potential, and disease resistance. Founders included CL111, RoyJ, CL153, LaKast, CL172, Cypress, Presidio, and Catahoula, representing elite germplasm released between 1992 and 2017 by public rice breeding programs in Louisiana, Arkansas, and Texas. Detailed founder release dates and pedigree information are provided in Table S1. Population development followed a structured crossing pipeline beginning with four biparental crosses (MP2), which were subsequently intercrossed to form six four‐way families (MP4). F1 plants from MP4 families were intercrossed to generate the six‐ and eight‐way MP6/8 populations, as described by Cerioli et al. (2022). For validation, we evaluated the biparental MP‐D population derived from Presidio × Catahoula. This cross was segregated primarily for the key upstream haplotype identified during discovery analyses. Phenotypic data Field trials were conducted in 2021 and 2022 at two sites: the H. Rouse Caffey Rice Research Station (Crowley, LA; 30°14′ 30″ N, 92°20′ 46″ W) and the Rice Research and Extension Center (Stuttgart, AR; 34°28′ 34″ N, 91°25′ 09″ W). Louisiana trials included two planting dates per year (March 10 and April 5 in 2021; March 4 and March 10 in 2022) and followed an augmented design with unreplicated entries. Trials were established at two planting dates to capture variation within the Louisiana planting window and to evaluate the effect of planting date on agronomic performance. Planting date is a major environmental factor influencing yield, development, and grain quality in rice; therefore, using multiple planting dates allows for a more robust assessment of genotype performance across relevant field conditions. Arkansas trials used a randomized complete block design with two replications. Plots measuring 1.42 m × 4.12 m were drill‐seeded at a rate of 78 kg ha−1 and grown following recommended management practices from the Louisiana Rice Production Handbook (Saichuk, 2026). Plant height was measured from ground level to the tip of the primary panicle using a ruler and recorded in centimeters with the Fieldbook application (Rife & Poland, 2014). Grain yield (kg ha−1) was harvested using a Wintersteiger Delta Plot combine equipped with a Harvest Master Grain Meter (Juniper Systems) and adjusted to 120 g kg−1 moisture. A total of 316 and 174 MP6/8 lines were assessed in Louisiana during the 2021 and 2022 seasons, respectively, while 234 (2021) and 174 (2022) lines underwent evaluation in Arkansas. The MP‐D population included 100 lines phenotyped in 2020 and 55 lines phenotyped in 2021 using two planting dates. Genotyping The MP6/8 population was genotyped using DArTSeq (Diversity Arrays Technology Sequencing), producing 4865 initial markers. Quality control removed markers with minor allele frequency (MAF) < 0.05, missing data > 10%, or heterozygosity > 8%. Imputation was performed using the Wright method implemented in the snpReady package (Granato et al., 2018). Genotype calls were coded as 0, 1, or 2 for downstream analyses. Following quality assessment, a total of 1501 high‐quality markers were retained. The MP‐D population was genotyped using the LSU1200 SNP panel (where LSU is Louisiana State University) developed by the LSU AgCenter breeding program (Cerioli et al., 2022) and processed through Agriplex Genomics. Marker selection for the MP‐D panel was guided by polymorphisms identified in the MP6/8 dataset; however, because the two populations were genotyped using different platforms, the marker sets are not directly overlapping. A total of 33 markers from the MP‐D panel were selected to target the region of interest identified in the MP6/8 analyses. Whole‐genome sequencing (WGS) was conducted on the LSU380 panel, which includes the MP6/8 founder lines, to generate high‐resolution variant data in the region of interest. These data were used to define haplotypes, characterize founder polymorphisms, and support downstream haplotype and ancestry analyses. WGS reads were mapped to the Oryza sativa Nipponbare reference genome using Burrrow–Wheeler Aligner (Li & Durbin, 2009). Single‐nucleotide polymorphisms (SNPs) and insertion/deletions for each sample were identified with the Genome Analysis Toolkit (GATK) HaplotypeCaller (DePristo et al., 2011; McKenna et al., 2010; Poplin et al., 2018; Van der Auwera et al., 2013). Individual genomic variant call formats (gVCFs) from the LSU380 panel underwent combination and joint genotyping utilizing GATK's CombineGVCFs and GenotypeGVCFs commands, respectively. Variants were filtered to retain only markers with a MAF of at least 1% and no more than 40% missing data per marker. A permissive filtering threshold (MAF ≥ 0.01, missing data ≤ 40%) was applied to the LSU380 panel as a whole to accommodate its relatively low sequencing depth (∼4×) and to retain rare variants across the dataset. For downstream haplotype characterization of the eight MP6/8 founder lines within the target interval (31.5–41 Mb), subsequent filtering for haplotype characterization is described in Section 2.6. Statistical analysis All statistical analyses were performed using the R statistical computing environment (R Core Team, 2025) within the RStudio integrated development environment (Posit Team, 2025). Phenotypic data were analyzed using a two‐stage approach. In Stage 1, observations were analyzed separately within each environment using spatial mixed models implemented in the SpATS package (Rodríguez‐Álvarez et al., 2018). Genotypes were modeled as fixed effects to obtain best linear unbiased estimates (BLUEs), while spatial variation across the field was accounted for using two‐dimensional penalized splines and random row and column effects. The model was: y=Xβ+Xgg+Xsβs+Zss+Zuu+e where y is the vector of observed phenotypes. The term Xβ represents fixed effects, including the overall mean and design factors. Genotype effects (g) were fitted as fixed through the design matrix Xg to obtain BLUEs. Spatial variation across the field was captured using a two‐dimensional smooth surface, modeled as a combination of fixed and random components (Xsβs+Zss), allowing for gradual trends along rows and columns. To account for localized, discontinuous variation, row and column effects were included as random terms (u) through the design matrix Zu. The residual term e represents unexplained variation and is assumed to be independently distributed. Models were fitted separately for each environment, and genotype BLUEs were extracted for downstream analyses. In Stage 2, genotype BLUEs obtained from Stage 1 were combined across environments using a linear model including genotype and trial components: yij=μ+Gi+Ej+eij where yij is the BLUE of genotype i in environment j, μ is the overall mean, Gi represents genotype, Ej represents environment, and eij is the residual error. Trial‐adjusted marginal means were estimated using the emmeans package (Lenth, 2024) to account for unbalanced data across environments. Heritability was estimated using variance components from mixed models fitted in the sommer package (Covarrubias‐Pazaran, 2016). Models included fixed effects, random additive genetic effects, and a spatial component to account for within‐field heterogeneity. Narrow‐sense heritability (h 2) was calculated as the proportion of additive genetic variance relative to total phenotypic variance. h2=σa2σa2+σe2 where σa2 denotes additive genetic variance and σe2 represents residual variance. Residual diagnostics included the Shapiro–Wilk test for normality and Levene's test for homogeneity of variances, supplemented by visual inspection of residual plots. Correlation between plant height and grain yield was assessed using Pearson's correlation coefficient. Analysis of variance (ANOVA) followed by Tukey's honestly significant difference (HSD) test was used to separate means at α < 0.01. Data visualization was performed using ggplot2 (Wickham, 2016). Genome‐wide association study Genome‐wide association study (GWAS) was performed in the Genome Association and Prediction Integrated Tool (GAPIT) v3 using the FarmCPU model (Liu et al., 2016; J. Wang & Zhang, 2021) with BLUEs values from each line and the filtered SNP dataset. Population structure in the MP6/8 population was assessed using principal component analysis (PCA) based on genome‐wide SNP markers. PCA was conducted in R using the prcomp() function. The first principal components were examined to evaluate potential population stratification before GWAS. Because MP6/8 was derived from balanced founder contributions and showed no evidence of residual population structure (Figure S1), population‐structure covariates were not included in the final GWAS models. Inclusion of up to five principal components as covariates did not alter the location or significance of the major associations. Manhattan and quantile–quantile plots were generated using the CMplot package (Yin, 2024) with −log10(p) values displayed. A suggestive threshold of p = 1 × 10− 5 and a Bonferroni significance threshold of 0.05/m (where m = number of SNPs) were applied. Haplotype characterization of MP founders Haplotype definitions were based on WGS data from the eight founder lines. To characterize the haplotypes of the founder lines across the 31.5–40.0 Mb region on chromosome 1, variants were first filtered to include only coding‐region SNPs (missense, nonsense, and splice‐site variants) using existing functional annotations. This filtering step was applied to reduce dataset complexity and facilitate consistent haplotype assignment, rather than to infer functional causality. At the sequence resolution used here, excluding noncoding variants is not expected to alter haplotype structure but improves interpretability and robustness of founder‐defined haplotypes. After filtering, 389 SNPs were retained for haplotype characterization across the region that were polymorphic among the eight founders (Table 1). TABLE 1Founder haplotypes and distribution of MP6/8 individuals across subregions on chromosome 1. Whole‐genome sequencing (WGS) data from the eight MP6/8 founders were used to define haplotypes across five subregions (32.68–40.71 Mb) on chromosome 1. The table lists haplotype classes, the number of single‐nucleotide polymorphisms (SNPs) per region, parental lines carrying each haplotype, and the number of MP6/8 progeny assigned to each class. Haplotypes represent identical‐by‐state segment groupings and do not imply functional annotation.Subregion assignmentHaplotype region (Mb)SNPs (WGS)Haplotype classificationParentsCount (population)132.68–34.2157Hap1CL111, CL153, Cypress, Catahoula, CL172, Presidio232Hap2Lakast, RoyJ55234.22–38.11254Hap1CL111, CL153, Cypress, Catahoula, CL172257Hap2Presidio25Hap3Lakast, RoyJ20338.23–38.71 (sd1)15Hap1CL111, CL153, Cypress, Catahoula, CL172, Presidio294Hap2Lakast, RoyJ13439.24–40.1339Hap1CL111, CL153, Cypress, CL172169Hap2Catahoula, Presidio, Lakast, RoyJ106540.17–40.7124Hap1CL111, CL153, Cypress, CL172, Catahoula226Hap2Presidio, Lakast, RoyJ59 Note: The 32.68–40.71 Mb interval was selected based on patterns of LD decay and founder sequence divergence.John Wiley & Sons, Ltd. Founder haplotypes and distribution of MP6/8 individuals across subregions on chromosome 1. Whole‐genome sequencing (WGS) data from the eight MP6/8 founders were used to define haplotypes across five subregions (32.68–40.71 Mb) on chromosome 1. The table lists haplotype classes, the number of single‐nucleotide polymorphisms (SNPs) per region, parental lines carrying each haplotype, and the number of MP6/8 progeny assigned to each class. Haplotypes represent identical‐by‐state segment groupings and do not imply functional annotation. Note: The 32.68–40.71 Mb interval was selected based on patterns of LD decay and founder sequence divergence. Haplotype subregions were defined by identifying contiguous genomic intervals in which founders could be grouped into a limited number of distinct allele classes based on shared SNP patterns (Table 1). Subregion boundaries correspond to positions where haplotype groupings among founders changed, reflecting historical recombination breakpoints within the interval rather than predefined physical or linkage disequilibrium boundaries. Intervals were selected to produce subregions containing two or three major haplotypes, enabling balanced comparisons of haplotype effects. Founder‐defined haplotypes were subsequently projected onto population datasets using the existing lower density marker panels. In the MP6/8 population, DArTSeq markers located within each subregion were used to assign lines to founder haplotypes based on allele similarity to WGS‐defined founder profiles. A total of 86 DArTSeq markers were polymorphic among the eight founders and were used to assign haplotypes across the MP6/8 lines. In the MP‐D population, 33 SNPs from the LSU1200 panel were polymorphic between the Catahoula and Presidio parents and were used to classify parental haplotypes within the same subregions. In both cases, marker sets were selected to tag the WGS‐defined parental haplotypes consistently across datasets. Lines showing evidence of recombination within a subregion were excluded from analyses of that subregion to ensure unambiguous haplotype assignment. Mean grain yield among haplotype classes was compared using ANOVA followed by Tukey's HSD test. In the biparental population, single‐marker ANOVA (F statistic) was used to test for yield differences among marker genotypes within the target interval. Historical and ancestry analysis To reconstruct the origin and distribution of sd1‐linked haplotypes, we conducted (i) ancestry inference using global reference populations, (ii) haplotype and pedigree analysis in US germplasm, and (iii) evaluation in global elite breeding lines. Reference populations and ancestry inference WGS data from the LSU380 panel were used to characterize introgression patterns surrounding the sd1 locus and provided the sequence data on the eight MP6/8 founders used in Section 2.6. The LSU380 panel (referred to as the Native Trait Panel, in some other studies) represents a collection of modern US rice breeding germplasm and key diversity lines. Reference global germplasm representing major Oryza sativa subpopulations (indica subpopulation 1, indica subpopulation 2, aromatic, aus, temperate japonica, and tropical japonica) were used for ancestry inference. The reference germplasm sequences were obtained from previously published studies representing global rice diversity (Duitama et al., 2015; McCouch, 2013; Melandri et al., 2021; The 3,000 Rice Genomes Project, 2014; Xu et al., 2012). This germplasm consisted of 1062 global germplasm accessions with defined subpopulation assignments (Table S2). Local ancestry was estimated using a supervised, window‐based approach implemented in Gnomix (Hilmarsson et al., 2021). This method assigns genomic segments based on posterior probabilities derived from allele‐frequency similarity to predefined reference populations, generating genome‐wide ancestry profiles for each line. A set of 300 individuals with high‐confidence assignments to a single subpopulation was used to construct the reference panel for ancestry inference. This reference panel included representatives of the major rice subpopulations: aromatic (n = 50), aus (n = 48), indica‐1 (n = 45), indica‐2 (n = 26), indica‐3 (n = 15), indica admixture (n = 16), temperate japonica (n = 50), and tropical japonica (n = 50). Population structure in the LSU380 panel was evaluated using PCA and ADMIXTURE following linkage disequilibrium‐based SNP thinning (Alexander et al., 2009). PCA was conducted in R using the adegenet package (Jombart, 2008). In contrast to the PCA used in the GWAS analysis of the MP6/8 population, which was conducted to assess within‐population structure, this PCA was performed on the LSU380 panel to characterize broader population structure for ancestry inference. Ancestry proportions were estimated using software ADMIXTURE, with cross‐validation used to determine the optimal number of clusters (K). Following established standard practice, the individuals in the LSU380 panel were assigned to a subpopulation when ancestry exceeded 0.80; otherwise, they were classified as admixed (Garris et al., 2005). Analysis of US germplasm panel (LSU380) To investigate the distribution of sd1‐linked haplotypes in US breeding materials, we analyzed the LSU380 panel, which includes 387 accessions representing modern cultivars, advanced breeding lines, and historically important varieties developed across US rice programs. The panel spans major rice‐producing regions, including Louisiana (n = 131), Arkansas (n = 105), Texas (n = 84), Mississippi (n = 46), California (n = 10), and Missouri (n = 1), and includes long‐grain (n = 316), medium‐grain (n = 48), and short‐grain (n = 4) types. Lines were further classified by herbicide technology (Conventional, Clearfield, and Provisia) and breeding origin (Core, US Other, and Exotic). Most accessions exhibit predominantly tropical japonica ancestry with minor contributions from other subpopulations. Only lines carrying the sd1 deletion were included in downstream analyses. Pedigree reconstruction Pedigree data for all 179 US lines containing the sd1 allele were retrieved from a locally maintained BreedBase database. Lineages were traced to identify the presence of TN1, IR8, or both within each pedigree. When available, documented ancestor–descendant relationships and variety release histories were used to infer the most probable donor source of the sd1 allele. These pedigree‐based assignments were later compared with haplotype and ancestry results to evaluate consistency between genomic and historical evidence. Haplotype classification around sd1 To characterize introgression patterns flanking sd1, an initial 5 Mb region (approximately 35–40 Mb on chromosome 1) was surveyed using WGS data to identify transitions in ancestry and haplotype structure. Based on these transitions, two focal windows closest to the sd1 locus were selected: one upstream (35.68–35.77 Mb) and one downstream (41.89–41.96 Mb). SNPs within these regions were filtered to remove markers with more than 2.5% missing data, resulting in 11 high‐quality SNPs upstream and 13 downstream. Within each window, haplotypes were defined based on SNP similarity and grouped into biologically meaningful classes. Haplotypes were designated as follows: Cypress type, representing tropical japonica ancestry shared by Cypress, CL111, CL153, CL172, and Catahoula Presidio type, representing indica ancestry unique to Presidio Recombinant (Rec), representing lines with mixed haplotype patterns Cypress type, representing tropical japonica ancestry shared by Cypress, CL111, CL153, CL172, and Catahoula Cypress type, representing tropical japonica ancestry shared by Cypress, CL111, CL153, CL172, and Catahoula Presidio type, representing indica ancestry unique to Presidio Presidio type, representing indica ancestry unique to Presidio Recombinant (Rec), representing lines with mixed haplotype patterns Recombinant (Rec), representing lines with mixed haplotype patterns Lakast and RoyJ, which carry the tall (SD1) allele, exhibited predominantly tropical japonica ancestry across this region and were treated as a distinct haplotype class where relevant. For each accession, upstream and downstream haplotypes were assigned independently. Lines were then classified based on their haplotype combinations relative to the sd1 locus. These classifications were compared with pedigree‐derived donor assignments to assess concordance between genomic structure and historical introgression pathways. Global germplasm analysis To evaluate the broader relevance of sd1‐linked haplotypes beyond US germplasm, we analyzed a panel of 424 elite rice genotypes from the International Rice Research Institute (IRRI) and collaborating public breeding programs, including PhilRice and the Bangladesh Rice Research Institute. The dataset included breeding lines and released varieties, along with representative donor genotypes such as TN1, IR8, Cypress, and Presidio. Sequence data were derived from IRRI‐led sequencing efforts, the 3K Rice Genomes Project, and publicly available resources, including GenBank. Across the genomic region spanning 34.5–42 Mb on chromosome 1, 49 windows of approximately 100 bp were selected based on high allelic diversity. Within each window, alleles were grouped by sequence identity to define haplotypes. Allele frequencies were calculated across the panel, with particular attention to the prevalence of Cypress‐type (putatively favorable) and Presidio/TN1/IR8‐type (putatively unfavorable) haplotypes. These analyses were used to assess the global distribution, diversity, and frequency of sd1‐linked genomic regions in elite indica germplasm. RESULTS Phenotypic characterization of the MP6/8 population Plant height in the MP6/8 population ranged from 85 to 116 cm (mean 97 cm), with h 2 = 0.75 across environments. Grain yield ranged from 7.36 to 11.43 t ha−1, with a mean of 9.63 t ha−1 and an across‐environment h 2 of 0.53. Presidio showed the lowest yield among founders, whereas CL153 and LaKast were the most productive (Figure 1). Descriptive statistics and trial‐level heritability estimates are presented in Table S3. Phenotypic distribution of plant height and grain yield in the MP6/8 multiparent population. Distributions of best linear unbiased estimate (BLUE)‐based (A) plant height (cm) and (B) grain yield (t ha−1) across all environments for 316 MP6/8 lines and eight founder parents. Founder‐parent means are positioned along the distributions: Cypress, CL111, CL153, Catahoula, CL172, Presidio, Roy J, and LaKast. Genome‐wide association analysis for plant height and yield Genome‐wide association analysis was conducted on 324 MP6/8 lines (316 progeny and eight founders) evaluated across six environments over 2 years. Across‑environment analyses identified quantitative trait loci (QTLs) for plant height and grain yield on chromosome 1 (Table S4). A major QTL for plant height was detected at 38.33 Mb and explained 74% of the phenotypic variance in the MP6/8 population (Figure 2A). At the peak SNP, all six semi‐dwarf founders carrying the sd1 allele possessed the “C” allele, whereas the two tall founders, Roy J and LaKast, carrying SD1, harbored the alternative “A” allele. Additional minor‐effect QTLs for plant height were detected on chromosomes 3, 7, 8, and 10, each explaining less than 5% of the phenotypic variance. Genome‐wide association results for plant height and grain yield in the MP6/8 population. (A) Manhattan plot (left) and corresponding quantile–quantile (Q–Q) plot (right) for plant height based on the FarmCPU model. A major‐effect quantitative trait locus (QTL) on chromosome 1 at 38.33 Mb is highlighted. (B) Manhattan plot (left) and corresponding Q–Q plot (right) for grain yield, showing a significant QTL upstream of sd1 at 36.69 Mb. In the Manhattan plots, the dashed red horizontal line indicates the suggestive significance threshold (p = 1 × 10− 5), and the solid red horizontal line indicates the Bonferroni‐corrected significance threshold. For grain yield, a QTL was detected upstream of sd1 at 36.69 Mb, explaining 45% of the phenotypic variance (Figure 2B). Presidio uniquely carried the “T” allele at this SNP, whereas all other founders carried the “C” allele. Additional yield‐associated QTLs were identified on chromosomes 2 and 3, with individual loci explaining between 4% and 9% of the phenotypic variance. Differences in yield between allele classes were significant (p < 0.01) in four of the six environments (Table S4). Despite the close physical proximity of these loci, grain yield and plant height were not correlated (r = 0.03; Figure S2), indicating that the yield QTL represents a locus genetically distinct from sd1, although closely linked. Overall, the GWAS results indicate that separate but closely linked loci on chromosome 1 are associated with variation in plant height and grain yield in this population. This separation enabled targeted haplotype analyses to resolve the specific upstream segment associated with the yield penalty. Notably, the major QTL for yield is located upstream of sd1 and appears to act independently of plant height under the conditions evaluated. Haplotype characterization identifies a specific haplotype associated with reduced yield To resolve the yield‐associated QTL, we examined founder WGS data across an 8 Mb region (32.68–40.71 Mb) encompassing the 36.69 Mb peak. Based on SNP patterns among the eight founders, the interval was subdivided into five contiguous subregions, each containing two or three haplotype classes (Table 1). DArTseq markers were used to assign these haplotypes to 316 MP6/8 lines, with recombinants within subregions excluded from yield analyses (Figure 3A; Tables S5 and S6). Haplotype structure and yield effects across an 8 Mb interval on chromosome 1. (A) Founder‐defined haplotypes and MP6/8 progeny assignments across five contiguous subregions (32.68–40.71 Mb). Colors denote subregions, while shading distinguishes haplotypes defined independently within each subregion; haplotype blocks are not drawn to exact scale. The thicker outline in Subregion 2 marks Presidio's distinct haplotype; the dashed vertical line shows the top genome‐wide association study (GWAS) single‐nucleotide polymorphism (SNP) for yield. In Subregion 3, the dashed line indicates the most significant GWAS SNP for plant height. (b) Boxplots showing grain yield (t ha−1) of MP6/8 lines grouped by haplotype class in Subregion 2 (34.22–38.11 Mb). Differences among haplotype classes were evaluated using analysis of variance (ANOVA) followed by Tukey's honestly significant difference test (α = 0.01). Different letters above boxplots indicate significantly different mean yields, whereas shared letters indicate no significant difference among haplotype classes. For each subregion, mean grain yield was compared among haplotype classes using ANOVA followed by Tukey's HSD test. Subregion 2 (34.22–38.11 Mb) showed the largest and most significant haplotype effect, consistent with the GWAS signal at 36.69 Mb. Lines carrying Hap1 (shared by five semi‐dwarf founders; n = 257) or Hap3 (LaKast and RoyJ; n = 20) had significantly higher yields than lines carrying Hap2 (unique to Presidio; n = 25), with an average difference of 0.92 t ha−1 (ANOVA, p = 8.5 × 10− 14; Tukey's HSD, p < 0.01; Figure 3B). At the subregion containing sd1 (Subregion 3), no significant yield differences were detected among haplotypes (Figure S3). Some smaller effects were observed in adjacent subregions where Presidio shared haplotypes with other founders, but when lines with the Presidio Hap2 in Subregion 2 were removed, these differences disappeared (Figures S3 and S4). This indicates that the negative yield effect is specifically associated with the unique Presidio haplotype in Subregion 2. Because the negative effect in Subregion 2 was detected consistently across MP6/8 environments, we next tested whether this haplotype effect persisted in an independent genetic background. To do so, we evaluated the Presidio × Catahoula biparental population, which segregates between 34.27 and 37.69 Mb and is fixed for the semi‐dwarf sd1 allele. Validation of the negative haplotype in the Presidio × Catahoula (MP‐D) population To validate these findings, we evaluated the MP‐D biparental population derived from Presidio × Catahoula. Yield data were collected on 100 lines across 2 years and three environments. The parents share the same haplotypes for regions upstream of sd1 except in Subregion 2 (34.27–37.69 Mb), allowing a direct comparison of alleles within this interval. To compare the yield effect of the Catahoula and Presidio haplotypes in this region, lines that were segregating in the region were omitted, and lines fixed for the Presidio haplotype were compared to the lines fixed for the Catahoula haplotype. MP‐D lines carrying the Presidio haplotype in Subregion 2 (n = 32) had significantly lower yields than those with the Catahoula haplotype (n = 34), with an average reduction of 0.6 t ha−1 across all trials (ANOVA, p = 3 × 10− 5; Tukey's HSD, p < 0.01; Figure 4). Single marker ANOVA across 33.95–43.09 Mb identified significant markers (p ≤ 0.01) between 34.67 and 37.69 Mb (Table S7). The two most significant markers (F (df = 2, 99) = 17.96, p = 2.22 × 10− 7) were located at 36.41 and 36.48 Mb, closest to the MP6/8 GWAS peak at 36.69 Mb (Figure 5). These findings confirm that the Presidio haplotype is linked to lower yield and further narrow the candidate interval. Yield performance of Presidio versus Catahoula haplotype classes in the MP‐D biparental population. Boxplots showing grain yield (t ha−1) across 2 years and three environments for MP‐D lines carrying either the Presidio or Catahoula haplotype in Subregion 2 (34.57–37.69 Mb). Differences between classes were evaluated using one‐way analysis of variance (ANOVA) and Tukey's honestly significant difference test at α = 0.01. Different letters above boxplots indicate significantly different mean yields, whereas shared letters indicate no significant difference among haplotype classes. Single marker association analysis in the MP‐D population near the yield quantitative trait locus (QTL) region. Association signal (F statistic) from one‐way analysis of variance (ANOVA) testing single‐nucleotide polymorphisms (SNPs) between 33.95 and 43.09 Mb on chromosome 1. Peaks at 36.41 and 36.48 Mb represent markers strongly associated with grain yield and correspond closely to the MP6/8 yield QTL peak at 36.69 Mb. Ancestry and pedigree analyses reveal the historical origin of the negative haplotype Admixture and local ancestry analyses along chromosome 1 using WGS data and a global diversity panel showed that semi‐dwarf founders Cypress and Presidio carry indica introgressions at sd1 but differ in the extent of the surrounding indica blocks (Figure 6 and Figure S5). Presidio carries a long indica segment from approximately 34.5 to 39.8 Mb, whereas Cypress (also representing CL111, CL153, CL172, and Catahoula) showed indica ancestry from approximately 38.04 to 42 Mb, with tropical japonica ancestry upstream. RoyJ (representing LaKast) carries tropical japonica ancestry across the sd1 region, with indica ancestry only at the distal end of chromosome 1. Local ancestry along chromosome 1 in three founder varieties. Subpopulation ancestry assignments for Presidio, Cypress (also representing CL111, CL153, CL172, and Catahoula), and RoyJ (representing LaKast). Color bars denote posterior ancestry probabilities for aromatic (ARO), aus (AUS), indica (IND), temperate japonica (TEJ), and tropical japonica (TRJ). A dashed vertical line indicated the genomic position of sd1. Pedigree reconstruction for 179 semi‐dwarf lines in the LSU380 US rice panel indicated that all trace their sd1 introgression to IR8, TN1, or both (Table S8): 162 lines (including the eight MP founders) include both donors, 16 lines trace exclusively to IR8, and one line exclusively to TN1. Using WGS data in approximately 100 kb windows upstream (35.68–35.77 Mb) and downstream (41.89–41.96 Mb) of sd1, we defined Cypress and Presidio haplotypes and classified 179 LSU380 lines accordingly (Figure 7; Table S9). Upstream, 149 lines carried the Cypress haplotype and 24 the Presidio haplotype; downstream, 142 carried the Cypress haplotype and 19 the Presidio haplotype. Twelve lines carried the Presidio haplotype on both sides of sd1, and 132 carried the Cypress haplotype on both sides, and the remaining lines were recombinants. Lines with pedigrees tracing only to IR8 carried the Cypress haplotype, whereas the single line tracing only to TN1 carried the Presidio haplotype upstream and downstream. Haplotype classes flanking the sd1 locus in 179 US semi‐dwarf breeding lines. Haplotype assignments based on single‐nucleotide polymorphisms (SNPs) in upstream (35.68–35.77 Mb) and downstream (41.89–41.96 Mb) windows around sd1 (38.38 Mb). “Cypress type” haplotypes correspond to IR8‐derived tropical japonica segments; “Presidio type” haplotypes represent the Taichung Native 1 (TN1)‐derived indica segment. Numbers indicate counts (n) of lines carrying each haplotype combination. Black bars indicate analyzed windows upstream (∼90 kb) and downstream (∼70 kb) of sd1; distances are shown for reference and are not drawn to scale. The figure does not display the other combinations that include recombinants. Exploration of the sd1 linkage drag region in global semi‐dwarf germplasm To assess whether the upstream haplotype associated with reduced yield in US germplasm reflects broader patterns among sd1 donors, we first compared the original semi‐dwarf donor varieties TN1 and IR8 with representative US rice cultivars. WGS data from TN1, IR8, and eight US varieties were analyzed, and haplotypes were assigned using nonoverlapping 100 bp intervals spanning the region upstream of sd1. The US panel included non‐semi‐dwarf varieties (Drew and Katy) and semi‐dwarf varieties known or presumed to trace to either TN1 or IR8 lineages. Comparison of haplotype alleles across this region revealed three distinct patterns (Figure S6). The non‐semi‐dwarf US varieties Drew and Katy carried a single haplotype (Hap1) across all evaluated intervals. In contrast, the US semi‐dwarf varieties Catahoula, Cocodrie, and Cypress carried Hap1 from 35.15 to 38.13 Mb and then transitioned to a second haplotype (Hap2) beginning at 38.32 Mb, immediately upstream of the sd1 locus (38.38 Mb). TN1 and IR8 carried Hap2 across the entire upstream region, confirming that the two original donors share an identical sequence at and upstream of sd1. Notably, the US semi‐dwarf varieties Presidio, Jefferson, and Lemont also carried Hap2 across the full region, matching the TN1 and IR8 haplotype pattern. To evaluate the broader relevance of this region beyond US germplasm, we analyzed a dataset from the IRRI comprising TN1, IR8, and 424 additional indica semi‐dwarf breeding lines (Table S10). Haplotypes were evaluated across 49 intervals spanning 34.5–42 Mb on chromosome 1 (Figure 8A,B). More than 95% of the examined indica lines carried the same sd1 deletion allele at the sd1 locus. Across the upstream interval from approximately 35.0 to 38.3 Mb, TN1, IR8, and Presidio shared an identical haplotype (Figure S6). Frequency of major haplotypes in elite indica material. (A) Frequencies of the four most common alleles in the interval from 34.5 to 42 Mb on chromosome 1. Haplotypes were defined in 100 bp intervals. The most common (major) allele, present in Presidio, IR8, and Taichung Native 1 (TN1), is at frequencies over 90% in the yield quantitative trait locus (QTL) region and at sd1. (B) Frequencies of the Cypress and Presidio/IR8/TN1 alleles in the same interval. The Cypress haplotype alleles are low‐frequency upstream of sd1 and transition to high frequency at sd1, while the Presidio/IR8/TN1 haplotype predominates at all positions from 35 to 39 Mb. Positions of sd1 (yellow) and the yield QTL (red) are indicated. This pattern indicates near‐fixation of the TN1/IR8 upstream haplotype in elite indica germplasm. By contrast, the Cypress‐type haplotype, associated with higher yield in US germplasm, was rare in global indica breeding material, ranking fourth or lower in frequency across most upstream intervals and occurring in only 0%–5% of elite lines (Figure 8A,B). These results indicate extremely limited haplotype diversity in the genomic region flanking sd1 among global indica semi‐dwarf breeding lines. Together, these findings suggest that the extensive use of a single sd1‐linked haplotype in indica breeding programs worldwide may have resulted in widespread retention of linked genomic regions analogous to the linkage drag identified in US long‐grain rice. The introduction and systematic evaluation of alternative haplotypes upstream of sd1 represents an important opportunity to improve grain yield potential and enhance long‐term genetic gain in global rice breeding programs. DISCUSSION Our study provides clear evidence that linkage drag associated with sd1 persists in modern US long‐grain rice germplasm and has relevance beyond US breeding programs. Unlike previous work documenting negative effects linked to sd1 in Asian germplasm (Murai et al., 2002; Vikram et al., 2015), we show that a specific upstream haplotype linked to historical sd1 deployment is associated with a consistent yield penalty in US tropical japonica germplasm and that closely related haplotypes predominate in elite global indica semi‐dwarf material. This study provides an empirical example of how selective sweeps around a single beneficial allele can mask linked yield penalties for decades, consistent with previous reports for drought susceptibility (Vikram et al., 2015). A key insight from our comparisons is that the original donor varieties TN1 and IR8 share identical haplotypes at and upstream of sd1 in the focal region. Therefore, the differences observed among US sd1 semi‐dwarf lineages primarily reflect recombination events and backcross histories rather than inherent differences between the original donor alleles upstream of sd1. Although sd1 itself confers major agronomic advantages, our results demonstrate that an extended indica introgression in the upstream flanking region impacts yield performance and early recombination patterns in the historical backcross lines through which sd1 was subsequently introduced, subsequently shaping the genetic foundation of US semi‐dwarf germplasm. In particular, the extended TN1‐derived indica segment present in Presidio contrasts strongly with the predominantly tropical japonica haplotypes surrounding the IR8‐derived sd1 introgression represented by Cypress. These ancestry patterns were supported by concordant evidence from pedigree reconstruction, haplotype comparisons, and local ancestry inference, indicating two predominant sd1‐associated haplotype configurations in US breeding material. Using both a multiparent population and an independent biparental population, we identified an upstream indica haplotype associated with a yield reduction of approximately 0.6–0.9 t ha−1, depending on population and environment. Importantly, the yield effect was localized to a genomic region tightly linked to sd1 but distinct from the subregion harboring the sd1 locus itself, supporting a linkage‐drag model rather than an effect attributable to sd1. The persistence of this yield penalty within elite germplasm since the introduction of TN1 into US breeding programs illustrates how historical introgressions can impose long‐term constraints on genetic gain when recombination is limited, and selection pressure is strong (Voss‐Fels et al., 2017; H. Zhang et al., 2025). The candidate interval refined by GWAS and biparental validation (approximately 700 kb flanking the peak SNP at 36.6 Mb) contains 94 annotated genes spanning diverse functional classes, including hormone signaling, carbohydrate metabolism, chloroplast development, stress responses, and core developmental and metabolic processes, based on International Rice Genome Sequencing Project 1.0 reference genome annotations (International Rice Genome Sequencing Project, 2005) (Table S11). The annotation set includes both functionally characterized genes and loci currently described as hypothetical, conserved hypothetical, or otherwise uncharacterized. Many loci in the region are associated with abiotic stress tolerance and growth, defense tradeoffs, including components of ABA (Gupta et al., 2020; Liao et al., 2023; Yan et al., 2023) and jasmonate signaling (Kumar et al., 2025; X. Wang et al., 2023), while others are involved in starch metabolism, photosynthetic capacity, and organ development (He et al., 2018; Lv et al., 2017; Y. Zhang et al., 2022). A substantial proportion of loci encode proteins of unknown function or noncoding transcripts, underscoring the region's complexity and the challenge of pinpointing causal variants within an extended haplotype. Further fine mapping will be required to enable confident inference of the underlying causal gene(s). Within this interval, several genes represent particularly plausible candidates for the observed yield reduction because they affect yield formation through pathways that can operate independently of plant height. Notably, OsOLP1 (OCTOPUS LIKE 1) regulates grain size through brassinosteroid signaling, a pathway known to influence yield components (Anne et al., 2015; Y. Zhang et al., 2022) without necessarily altering stature, making it consistent with yield effects that are separable from the semi‐dwarf phenotype conferred by sd1. Linkage drag around major domestication and Green Revolution genes is a well‐recognized consequence of strong positive selection and selective sweeps. Previous studies have documented unfavorable effects associated with genomic regions surrounding sd1, including the loss of drought‐tolerance alleles and increased drought susceptibility in modern semi‐dwarf rice varieties (Vikram et al., 2015, 2016), as well as similar linkage‐drag effects around other major targets of selection such as Wx (Asano et al., 2011). The drought locus reported by Vikram et al. (2015) identified a drought‐susceptibility locus on chromosome 1 between 38.4 and 38.9 Mb, immediately adjacent to sd1, with the susceptible allele present in sd1 deletion lines. Our ancestry analyses indicate that the Cypress sd1 haplotype transitions from tropical japonica to indica ancestry before this interval and remains indica across the reported region, whereas Presidio carries an extended indica introgression spanning approximately 34.5–39.8 Mb (Figure 6). Thus, both major sd1‐associated haplotype classes identified in this study are likely to carry the same drought‐susceptible indica haplotype described by Vikram et al. (2015). Some ambiguity exists in Presidio, as the 39.8 Mb position is near the distal boundary of the interval near the recombination breakpoint where the Presidio introgression reverts to japonica ancestry. Future analysis will need to be made once the underlying drought gene is identified to compare to the Presidio allele at that gene. From a practical breeding perspective, the low frequency of the Presidio haplotype in modern US breeding lines suggest that breeders have largely, if unintentionally, selected against indica alleles in this unfavorable region. Nevertheless, desirable traits in Presidio, including strong ratoon performance and favorable grain quality, have maintained its value as a parent and are likely to have contributed to the persistence of the haplotype in specific lineages. Marker‐assisted selection focused on trimming or replacing the negative upstream segment represents a realistic strategy to remove the yield penalty while retaining the sd1 semi‐dwarf allele. Routine flanking‐marker screening during parent selection or early generation testing would be sufficient to eliminate the unfavorable haplotype. Importantly, our analyses also suggest potential global implications for IRRI and other indica breeding programs. An exploratory survey of elite indica semi‐dwarf lines from IRRI revealed limited haplotype diversity across the critical upstream intervals near sd1, consistent with previous observations that extensive utilization of sd1 during and after the Green Revolution produced selective sweeps and reduced variation in elite pedigrees (Chen et al., 2017). However, whether this haplotype confers similar effects in indica backgrounds remains unclear. The yield penalty observed here was identified in tropical japonica populations, where alternative haplotypes allow direct comparison. In contrast, the near fixation of the TN1/IR8 haplotype in elite indica germplasm limits our ability to evaluate its effect using the current data and may also constrain its detection. Additionally, the magnitude and direction of this effect may depend on genetic background, and the haplotype may have different effects in indica germplasm. As a result, its impact in indica breeding programs cannot be directly confirmed; however, its widespread prevalence suggests that linked yield‐limiting alleles may persist and warrants further investigation of alternative upstream haplotypes to explore opportunities for improving yield potential. Overall, our findings underscore that beneficial alleles derived from major historical donors can carry linked deleterious regions that remain concealed for decades under strong selection. Addressing legacy linkage drag through targeted recombination, haplotype‐guided selection, and functional analyses can help maximize the long‐term benefits of key agronomic genes, such as sd1. More broadly, this work illustrates how modern breeding programs can proactively monitor linkage drag around major introgressed loci and use genomic tools to decouple beneficial targets from unfavorable linked variation. Given the central role of sd1 in elite pedigrees worldwide, expanding haplotype diversity around sd1 may represent a broadly applicable route to unlocking incremental yield gains while preserving the agronomic advantages of semi‐dwarfism. AUTHOR CONTRIBUTIONS Flavia Furlan: Conceptualization; formal analysis; investigation; writing—original draft. Brijesh Angira: Data curation; investigation; writing—review and editing. Maria Montiel: Data curation; writing—review and editing. Jonathan Richards: Data curation; formal analysis; writing—review and editing. Tommaso Cerioli: Conceptualization; data curation; investigation; writing—review and editing. Christopher Hernandez: Data curation; investigation; writing—review and editing. Roberto Fritsche‐Neto: Formal analysis; writing—review and editing. John Platten: Data curation; formal analysis; investigation; writing—review and editing. Jomar Punzalan: Data curation; methodology; writing—review and editing. Susan McCouch: Investigation; writing—review and editing. Kelly Robbins: Investigation; supervision; writing—review and editing. Xueyan Sha: Investigation. Adam Famoso: Conceptualization; funding acquisition; investigation; methodology; project administration; supervision; writing—original draft. CONFLICT OF INTEREST STATEMENT The authors declare no conflicts of interest. Supporting information Figure S1. Principal component analysis (PCA) of the MP6/8 population based on genome‐wide SNP markers. The first two principal components show no evidence of residual population structure within the MP6/8 population. Figure S2. Relationship between grain yield plant height across environments. Each point represents a genotype, with a fitted linear regression line shown in red. Pearson's correlation coefficient (r = 0.03, P = 0.60) indicates no significant linear association between the two traits. Figure S3. Grain yield comparisons among haplotypes in subregions 1, 3, 4, and 5 within the 32.68–40.71 Mb interval. Statistical differences were tested using ANOVA followed by Tukey's honestly significant difference test at α = 0.01. Figure S4. Yield effects across haplotype classes in subregion 4 following removal of Presidio Hap2 (from subregion 2). Yield differences observed in adjacent subregions disappear after excluding lines carrying the Presidio haplotype in Subregion 2. Statistical differences were tested using ANOVA followed by Tukey's honestly significant difference test at α = 0.01. Figure S5. Local ancestry inference along chromosome 1 in founder varieties (CL111, CL153, Catahoula, CL172, Lakast). Ancestry assignments based on whole‐genome sequence data illustrate indica and japonica introgression patterns surrounding the sd1 locus. Figure S6. Haplotype structure upstream of the sd1 locus in selected U.S. rice cultivars and donor varieties. Haplotypes were defined using whole‐genome sequence data across non‐overlapping genomic windows upstream of the sd1 locus on chromosome 1. Genomic positions (Mb) are shown along the top. Blue cells indicate Hap1 and red cells indicate Hap2. TN1 and IR8 represent the original semi‐dwarf donor varieties. U.S. cultivars carrying the Presidio haplotype show extended Hap2 blocks upstream of sd1, whereas Cypress‐derived cultivars show recombined haplotypes. Table S1. Founder lines used to develop the MP6/8 population, with breeding program, release year, and pedigree information. Table S3. Summary statistics and narrow‐sense heritability (h2) for grain yield (t ha− 1) and plant height (cm) within individual trials and across trials. Genotype counts are reported for each trial and for the combined analysis across trials. Summary statistics include mean, median, standard deviation (SD), minimum (Min), and maximum (Max). Table S4. Genome‐wide association study results for plant height and grain yield. Significant and suggestive QTL were detected across environments, including marker positions, significance levels, and variance explained. Table S2. List of 1,062 global rice accessions used as reference germplasm for ancestry inference analyses. Table S5. WGS variants in the chromosome 1 candidate interval (32.68–40.71 Mb). Each row represents a variant (ID and position), with gene model (LOC), predicted consequence, allele calls for Cypress, CL111, CL153, CL172, Catahoula, Presidio, Lakast, and RoyJ, and subregion assignment based on allelic transitions within founder‐defined haplotypes. Table S6. Haplotype assignments of MP6/8 lines based on DArTSeq markers spanning the chromosome 1 candidate interval. The table reports, by subregion, marker chromosome and position, haplotype calls, and marker allele calls for each line. Lines were assigned to founder‐defined haplotypes for each subregion; recombinants were identified and excluded from subregion‐specific analyses. Table S7. Single‐marker ANOVA results for grain yield in the MP‐D biparental population. Markers spanning 34.27–43.10 Mb on chromosome 1 were tested across environments. Marker positions are reported in bp and Mb with corresponding F‐ratios and P‐values. The strongest signals were at 36.41 and 36.48 Mb (P = 2.22 × 10− 7). Table S8. Pedigree and metadata summary for LSU380 lines, including breeding program, grain class/type, herbicide technology, and sd1 functional allele. The presence/absence of TN1 and IR8 in each pedigree is indicated, and semi‐dwarf lines in the panel are categorized by inferred donor lineage (TN1, IR8, or both). Table S9. Classification of sd1‐flanking haplotypes in 179 U.S. rice lines from the LSU380 panel using whole‐genome sequencing data. Haplotypes were assigned independently for windows upstream (35.68–35.77 Mb) and downstream (41.89–41.96 Mb) of sd1 based on diagnostic SNPs, and lines were categorized as Cypress‐type, Presidio‐type, or recombinant for each window. Table S10. Elite indica germplasm analyzed for haplotype frequencies surrounding the sd1 locus. Lines from IRRI and collaborating breeding programs were evaluated across the target interval on chromosome 1. Table S11. Annotated genes within the chromosome 1 candidate interval associated with grain yield upstream of sd1. Gene models, transcript IDs, genomic coordinates, and functional annotations for the 94 genes located within the candidate interval were obtained from Ensembl Plants using the IRGSP‐1.0 Oryza sativa reference genome.