Genetic analyses of leaf traits in an interspecific Zoysia japonica × Zoysia matrella F₂ population

Abstract

Zoysiagrass (Zoysia spp.) is an important warm‐season turfgrass cultivated across tropical, subtropical, and temperate regions of the world. The genus is characterized by the presence of salt‐secreting glands on the adaxial leaf surface, which contribute to its high salt tolerance. In this study, we analyzed an interspecific F₂ population, derived from selfing an F₁ from a cross between Z. japonica acc. Meyer and Z. matrella acc. PI 231146, for variation in adaxial salt gland density, leaf width, and vein count. Using composite interval mapping with a previously constructed genetic map as a framework, we identified three quantitative trait loci (QTL) for leaf width, two QTL for vein count, and two QTL for salt gland density. We complemented the QTL analysis with bulked segregant RNA‐seq (BSR‐seq) to identify shared genomic regions and candidate genes for leaf width and salt gland density. BSR‐seq identified four trait‐associated regions, but only a single region identified for leaf width on Chr08 overlapped with a QTL for the same trait. We highlight putative candidate genes underlying the leaf width and salt gland density QTL and discuss their potential roles in leaf development. Together, the QTL and candidate genes provide an important resource for breeding stress‐resilient Zoysia germplasm.

Plain Language Summary

Leaf anatomical traits such as leaf width, vein count, and salt gland density influence plant growth and stress tolerance, yet their genetic basis remains poorly understood in many grasses. Using an interspecific Zoysia japonica × Zoysia matrella F₂ population, we combined linkage mapping and gene expression analyses to find regions of the genome linked to trait variation. We identified seven chromosome regions associated with these traits, including two novel regions for salt gland density. These results provide valuable resources for Zoysia breeding and future functional studies.

Abbreviations:

BSR

Bulked segregant RNA‐seq

CIM

composite interval mapping

CNN

convolutional neural network

LW

leaf width

MIM

multiple interval mapping

NMD

nonsense mediated decay

QTL

quantitative trait loci

SEM

Scanning electron microscopy

SG

salt gland

SNP

single nucleotide polymorphism

TTG

Transparent Testa Glabra

1. INTRODUCTION

Zoysiagrass is a popular turfgrass species widely utilized in sports fields and urban landscapes due to its high stress resilience. The three primary species, Zoysia japonica Steud, Zoysia matrella (L.) Merr., and Zoysia pacifica (Goudswaard) (M. Hotta & S. Kuroki), are distinguished mainly by their leaf morphology (Patton et al., 2017). One of the key determinants of turfgrass quality is texture, primarily influenced by leaf width (LW). Finer textured (narrow‐leafed) cultivars are often preferred for their aesthetic value; however, wider leaves (coarse‐textured) offer functional advantages, such as improved wear resistance (Braun et al., 2021). Typically, Z. japonica has a LW >2.5 mm, while LW in Z. matrella ranges from 1.5 to 2.5 mm, and is <1.5 mm in Zoysia pacifica (Patton et al., 2017). The increased durability of coarse‐textured cultivars can be attributed to a higher number of vascular bundles in broader leaves, which enhances leaf tensile strength and mechanical durability (Wei et al., 2022). Broader leaves with well‐supported vascular structures can better withstand mechanical stress from foot traffic (Głąb et al., 2015). However, a negative relationship has been observed between LW and water use efficiency in species with PEPCK and NADP‐ME subtypes of C₄ photosynthesis (Cano et al., 2019; Pan et al., 2022). Zoysia has been classified as having the PEPCK subtype of C₄ photosynthesis (Gutierrez et al., 1974). Hence, understanding the genetic basis of LW will potentially contribute to breeding efforts for improving water use efficiency and, hence, adaptation in zoysiagrass (Hammer et al., 2020).

In addition to LW, the three Zoysia species differ in a range of other traits, including salt tolerance (Chandra et al., 2015; Marcum & Murdoch, 1994; Patton et al., 2017). Zoysiagrass’ salinity tolerance is closely linked to a genotype's ability to secrete salt through specialized salt glands (SGs) located on the adaxial leaf surface (Marcum & Murdoch, 1990). Zoysia matrella is typically considered the more salt‐tolerant species due to its higher Na⁺ secretion in tissue washing experiments (Marcum & Murdoch, 1990). Further, a study of 48 different zoysiagrass accessions by Yamamoto et al. (2016) found that, in general, Z. japonica has a lower SG density compared to Z. matrella. Although SG density and salt secretion levels are not always positively correlated, and considerable variation exists for both traits within a species, Yamamoto et al. (2016) classified Z. japonica as a salt accumulator and Z. matrella as a salt secretor. The genetic basis of SG density and function in zoysiagrass remains poorly understood.

Here, we mapped the variation observed in LW, vein count, and SG density in an interspecific F₂ population derived from a Z. japonica acc. Meyer × Z. matrella acc. PI 231146 cross (Pradhan et al., 2025). We found two quantitative trait loci (QTL) for SG density, two colocalizing QTL for LW and vein count, and one additional QTL for LW. Using bulked Segregant RNA‐seq (BSR‐seq), we identified single nucleotide polymorphisms (SNPs) differentiating bulks representing the phenotypic extremes for SG density and LW. Only one of the four trait‐associated regions identified by BSR‐seq overlapped with an identified QTL for the same trait. Annotated genes in select QTL intervals for SG density and LW were mined for putative candidate genes by integrating data on allelic variation between the parents and F₁ of the mapping population with differential expression data from the BSR‐seq. The identified QTL and associated genes could pave the way for marker‐assisted trait selection in turfgrass breeding programs.

2. MATERIALS AND METHODS

2.1. Plant material and growth

The Z. japonica acc. Meyer × Z. matrella acc. PI 231146 F₂ population was described in Pradhan et al. (2025). The parents of the population, Meyer and PI 231146, the F₁ plant selfed to generate the F₂ population (F₁‐19‐TZ‐14321), and the F₂ progeny were maintained in a greenhouse at the University of Georgia (UGA) under 26°C (day)/21°C (night). For SG measurements in the parents and F₁, which were carried out under both freshwater and saline conditions, plants were propagated from single nodes and grown in soil, consisting of three parts sand:one part pine bark mix (mix of 500 L fine‐grade pine bark, 2 L coarse vermiculite, 500 mL superphosphate, 250 mL potassium nitrate, 250 mL calcium nitrate, 250 mL gypsum, and 250 mL Micromax micronutrient blend), in Ray Leach cone‐tainers for a month. The freshwater irrigation solution consisted of 2 g per gallon Excel Plus and Bedding Special (13‐2‐13 nitrogen [N]‐phosphorus [P]‐potassium [K]) liquid fertilizer (Everris NA, G99120) and 0.6 g per gallon MgSO₄, while the saltwater irrigation solution consisted of 300 mM NaCl along with the fertilizer. A salt level of 300 mM NaCl induces a differential salt response between Meyer (more sensitive) and PI 231146 (more tolerant) without compromising survival. Plants were watered through bottom irrigation for 15 min every 2 h for 3 weeks with either freshwater or saltwater before SG density was measured.

2.2. Leaf surface imaging of representative lines using scanning electron microscopy

Fully expanded leaf blades from Meyer, PI 231146, F₁‐19‐TZ‐14321, and four F₂ genotypes representing the phenotypic extremes for LW and SG density (two F₂ progeny for each side of the distribution per trait) were excised with a razor blade and transported from the greenhouse to the Georgia Electron Microscopy facility wrapped in a damp paper towel inside a sealed Ziploc bag. Leaf segments of ∼1 cm² were mounted onto Conductive Carbon Adhesive Tabs (Electron Microscopy Sciences) affixed to aluminum stubs designed for mounting into a Hitachi High‐Tech SU3900 scanning electron microscope. The microscope was operated at 15 kV, and images of both the adaxial and abaxial surfaces were captured from 75x to 360x.

2.3. Cross‐section leaf anatomy

Leaf blades were collected as described above, and 1 cm² segments were fixed in 2% formaldehyde for 48 h. The segments were then dehydrated through a graded ethanol series at room temperature (10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, 95%, and 100%) for 25 min each and embedded in medium‐grade LR White Resin (Electron Microscopy Sciences) as described in Spiekerman and Devos (2020). Transverse sections of 2.5 µm thickness were prepared from the embedded tissues using a Reichert Ultracut ES microtome and stained with 0.1% toluidine blue. The sections were examined and imaged using a bright‐field light microscope.

2.4. Trait phenotyping

2.4.1. SG density

Four to six fully expanded leaves from each genotype were carefully excised and laid flat atop a plastic‐lined surface with weights applied to either side of the leaf to avoid leaf curling. Clear nail polish (Electron Microscopy Sciences) was uniformly applied to the adaxial surface of each leaf using a single stroke. After the nail polish had dried, a piece of clear adhesive tape was gently pressed onto the treated surface and then peeled off, capturing an impression of the leaf surface. This impression was placed on a microscope slide, with the imprinted side facing upward. One imprint was prepared per leaf per genotype. All imprints for a single genotype were placed on a single slide and secured with a cover slip. Imaging of the SGs was performed using a light microscope with digital camera at 100x magnification (10x objective and 10x optical zoom on the digital camera).

SG density of the two parents and F₁‐19‐TZ‐14321 was determined manually in 0.6 mm² regions taken from the midsection of the widest portion of fully expanded leaves (Figure S1). SG density was determined on newly emerged but fully expanded leaves from plants grown under both freshwater conditions and after 3 weeks of exposure to 300 mM NaCl.

Because of the labor‐intensive nature of image preparation for SG quantification, a subset of 188 progeny was randomly selected from the 530 F₂ individuals used for genetic map construction (Pradhan et al., 2025) to phenotype SG density. SGs were manually annotated on 25 individual images (0.6 mm²) derived from plants grown in freshwater by drawing a tight box around them using the VGG Image Annotator software. The annotated images were utilized for training a Convolutional Neural Network (CNN) machine learning model. A custom object detection model based on the ConvNeXt (Liu et al., 2025) CNN architecture that leveraged bounding box annotations of SGs from two independent annotators was developed. High‐resolution images were paired with object‐level annotations to generate training and validation datasets with data augmentation strategies including random cropping to 256‐pixel resolution, affine transformations, flips, brightness/contrast/gamma jitter, and random blurring. Object instances were encoded into a You Only Look Once‐style format (Redmon et al., 2016) using a grid size of 16 and five prediction channels (object classification, object center x/y, and object height/width) suitable for training deep learning models. The ConvNeXt model was then trained using the mean squared error objective function and the AdamW optimizer (Loshchilov & Hutter, 2017) with a learning rate of 0.001, a batch size of 32, and early stopping of five epochs. For inference, each test image was divided into overlapping tiles of size 256 pixels, which were fed to the trained model in a sliding‐window approach with a step size of 64 pixels. The model generated per‐cell bounding box predictions for each overlapping tile. To reduce redundant detections, non‐maximum suppression was applied using hyperparameters that were determined by a grid search to maximize bounding box intersection over union over the validation set. This strategy enabled efficient, high‐throughput detection of SGs across large image sets while maintaining spatial accuracy and reducing false positives. The SG density was recorded as the number of glands/image (0.6 mm²) on two to six imprints per genotype. The associated scripts can be found on https://github.com/jlager/zoysiagrass‐phenotyping/.

2.4.2. LW and vein count

A subset of 176 progeny that largely overlapped (174 out of 176) with the subset used for SG density analyses was selected for phenotyping of LW and vein count. Four to six fully open leaves per genotype were excised from the leaf sheaths and scanned in a flatbed scanner at 1200 dpi. Zoysiagrass leaf segmentations were used to determine both the widest LW (in mm) and the vein count from the scanned images (Figure S1). Vein counting was performed using classical image processing and machine learning. Images of zoysiagrass leaves were segmented using a simple decision tree classifier trained on pixel‐level red‐green‐blue (RGB) values. The resulting predictions were post‐processed such that minimum‐area bounding rectangles identified individual leaves for feature extraction. Leaf maximum width was computed using classical computer vision approaches, while the number of veins was estimated by analyzing grayscale intensity profiles over the cross‐section of each leaf at its widest point. For each column in a horizontally aligned leaf, mean intensity profiles were smoothed and local minima were detected using peak‐finding algorithms. The number of peaks was interpreted as the number of veins passing through a given cross‐section, with the final vein count taken as the maximum across all columns. Hyperparameters for the detection algorithm, including peak prominence and smoothing window size, were optimized via grid search against manually annotated vein counts using mean absolute error. This approach enabled automated, scalable vein quantification with close agreement to human annotations. Manual measurements of vein count, performed on 25 individual images using ImageJ software, were used to determine the efficiency of the model. Vein density was calculated as vein count divided by LW (in mm). To decouple venation patterning from leaf size, an “adjusted vein count” was used as a derived trait, calculated as the residuals from a linear regression of vein count against LW. A constant of 3 was added to these residuals to ensure nonnegative values as input for WinQTL Cartographer without altering the trait's phenotypic variance.

2.5. Trait correlations and statistical analyses

To determine and visualize correlations between the measured leaf traits, a pairwise correlation matrix was generated using the ggpairs function from the GGally package (Schloerke et al., 2023) in R. To minimize the influence of outliers on the correlation analysis, the interquartile range method was applied to identify and exclude outlier values for each trait. Significance levels for correlations were calculated using the cor.test function, with asterisks (*, **, and ***) indicating p‐values < 0.05, < 0.01, and < 0.001, respectively. The results were visualized using a ggpairs plot, displaying both the Pearson correlation coefficient (r) and the coefficients of determination (R ²) for each trait pair. One‐way analysis of variance (ANOVA) analyses with Tukey's honestly significant difference mean‐separation tests were conducted to determine statistical significance at the level of p < 0.05. The normality of the phenotypic data was assessed using the Lilliefors‐corrected Kolmogorov–Smirnov test via the nortest package in R (Thode, 2002).

2.5.1. Genetic map

The Meyer × PI 231146 F₂ genetic map used in the QTL analysis was described in Pradhan et al. (2025). The map consists of 8402 SNP markers distributed across the 20 Zoysia (2n = 4x = 40) chromosomes.

2.5.2. QTL mapping

The values for SG density, LW, and vein count were averaged across two to six good‐quality images per genotype (Table S1) and, together with vein density and adjusted vein count values, used as phenotypes for composite interval mapping (CIM) using Windows QTL Cartographer v2.5_011 (Silva et al., 2012) with a 10 cM window size and 1 cM walk speed. The cross type was set as SF2. Significance thresholds for the logarithm of odds scores (p < 0.05) were determined by conducting 500 permutations for each trait. The percentage variance explained by each locus was obtained from the WinQTL Cartographer output. Additionally, significant QTL were verified and tested for epistasis with multiple interval mapping (MIM) in R/qtl, using makeqtl and fitqtl functions (Broman & Sen, 2009; Roy et al., 2023).

2.6. Bulked segregant RNA‐seq (BSR‐seq)

2.6.1. Creation of bulks with contrasting phenotypes and genotypes

Two types of bulks were created that we refer to as phenotypic (P) bulks and genotypic (G) bulks. Phenotypic bulks were generated as an alternative approach to identify genomic regions associated with a trait and consisted of eight F₂ progeny from the Meyer × PI 231146 cross with the highest phenotypic values (bulk P_H) and eight plants with the lowest phenotypic values (bulk P_L). Phenotypic bulks were created for both SG density and LW. Genotypic bulks were generated to identify differentially expressed (DE) genes in specific QTL regions (Chr01 for SG density and Chr08 for LW) by pooling eight progeny with the highest trait value that also carried the allele from the parent with the highest trait value at the targeted QTL (bulk G_H), and eight progeny with the lowest trait value and carrying the allele from the parent with the lowest trait value at the QTL region (bulk G_L). If insufficient F₂ progeny were available with the target genotype over the entire QTL region, bulks were supplemented with progeny that were heterozygous (H) in the QTL region. The progeny included in the bulks, their phenotypes, and genotypes in the QTL regions analyzed are given in Table S2.

2.6.2. Total RNA extraction

Three to four fully expanded leaves from each genotype were collected in 2‐mL tubes, snap frozen in liquid nitrogen, and subsequently ground using a TissueLyser II bead mill. The ground tissue was homogenized in 1 mL of TRIzol reagent (Invitrogen) per 100 mg of tissue and allowed to rest for 10 min at room temperature. Following homogenization, the tubes were centrifuged at 12,000 g for 10 min at 4°C. The supernatant was transferred to a new tube, and 200 µL of chloroform was added. The mixture was shaken until it appeared milky, allowed to rest at room temperature for 5 min, and then centrifuged at 12,000 g for 15 min at 4°C. The RNA‐containing aqueous phase was transferred to a 1.5‐mL tube and mixed with an equal volume of isopropanol. After resting for 5 min, the mixture was centrifuged at 12,000 g for 15 min at 4°C to pellet the RNA. The RNA pellet was washed with 75% ethanol twice and dried. The RNA was dissolved in 20 µL nuclease‐free water and DNase‐treated using the TURBO DNA‐free kit (Thermo Fisher Scientific) according to the manufacturer's instructions. The RNA concentration was determined using a NanoDrop spectrophotometer, and the quality was assessed via electrophoresis on a 1% Tris‐Borate‐EDTA agarose gel.

2.6.3. Library preparation and sequencing

One microgram of RNA from individual genotypes was pooled into bulks according to the approach described above. For each bulk, 1 µg of RNA was used to generate barcoded stranded mRNA‐seq libraries utilizing the KAPA Stranded mRNA‐seq library kit (Roche) according to the manufacturer's instructions, with half reactions used for all steps except the final library amplification. The concentration of the libraries was measured using a Qubit Fluorometer with the Qubit 1X dsDNA High Sensitivity Assay Kit (Invitrogen). Subsequently, 40 ng of each library was pooled into a larger set of 24 libraries and loaded onto a single flow cell of an Illumina NextSeq 2000 for paired‐end (PE) 100 bp sequencing at the Georgia Genomics and Bioinformatics Core (GGBC) at UGA.

2.6.4. BSR‐seq read processing

Reads were demultiplexed by barcode at the GGBC and downloaded from the BaseSpace Illumina hub. The reads were trimmed for adapter sequences and low‐quality bases using Trim‐Galore v0.6.5 (Krueger, 2015; https://github.com/FelixKrueger/TrimGalore). The trimmed reads were aligned to the scaffold‐level zoysiagrass assembly (ZJN_r1.1.fa; Tanaka et al., 2016) (P and G bulks) and the chromosome‐level assembly (ZJN_r1.0.pseudomol.fa; Tanaka et al., 2016) (P bulks only) using HISAT2 v2.2.1 (Kim et al., 2019). HISAT2 was also used to index the reference genome. The resulting alignments, initially in Sequence Alignment/Map (SAM) format, were converted to Binary Alignment/Map (BAM) format and sorted using SAMtools v2.12 (Danecek et al., 2021). StringTie (Pertea et al., 2015) was utilized to assemble reads into transcripts for each sample, guided by the ZJN_r1.1.gff gene annotation file (Tanaka et al., 2016), producing individual .gtf files. The assembled transcripts for each sample were merged using StringTie, with the individual .gtf files listed in mergelist.txt, to generate a comprehensive merged.gtf file that included all nonredundant transcripts across all samples. Using the merged.gtf file as reference, StringTie was then employed to re‐estimate transcript abundances for each sample, generating new individual .gtf files within the respective sample directories. Finally, gene count files were generated using the “prepDE.py” Python script. To assess variance in expression profiles between the genotypic and phenotypic bulks, a principal component analysis (PCA) plot was generated using the plotPCA function of DESeq2 (Love et al., 2014).

2.6.5. SNP identification in BSR‐seq data from P pools

To perform SNP identification on the BSR‐seq data from the phenotypic (P) bulks, reads from the P_H and P_L bulks aligned against the Z. japonica pseudomolecule assembly (ZJN_r1.0.pseudomol.fa; Tanaka et al., 2016) were used for SNP calling using the Haplotype Caller function (version 4.4) within the Genome Analysis Toolkit (McKenna et al., 2010). The HaplotypeCaller tool was used in Genomic Variant Call Format (GVCF) mode, followed by joint genotyping using CombineGVCFs and GenotypeGVCFs. SNPs were filtered based on quality metrics (quality by depth ≤ 10), and retained variants were exported to a tabular format for downstream analyses. Then, SNPs were further filtered to retain only those with a reference allele frequency ≥ 0.05, total read depth ≥ 16, per‐sample depth ≥ 8, and genotype quality ≥ 99.

2.6.6. Calculation of ΔSNP index

The QTLseqr package (Mansfeld & Grumet, 2018) was utilized to perform the ΔSNP index analyses. For each SNP, the SNP index (the proportion of sequencing reads containing the alternative or non‐reference allele) was first calculated separately for the high and low bulks using the formula:

$$\mathrm{SNP}\mathrm{index}=\frac{\mathrm{Number}\mathrm{of}\mathrm{reads}\mathrm{with}\mathrm{alternative}\mathrm{allele}}{\mathrm{Total}\mathrm{reads}\mathrm{at}\mathrm{that}\mathrm{position}}$$

The ΔSNP index was then determined by subtracting the SNP index of the low trait bulk from that of the high trait bulk:

$$\mathrm{\Delta }\mathrm{SNP}\mathrm{index}=\mathrm{SNP}\mathrm{inde}{\mathrm{x}}_{\mathrm{high}\mathrm{bulk}}-\mathrm{SNP}\mathrm{inde}{\mathrm{x}}_{\mathrm{low}\mathrm{bulk}}$$

To identify candidate QTL regions, tricube smoothing was applied to the ΔSNP index values using a 250 kb sliding window. Confidence intervals (CIs) at 95% and 99% were estimated through 1000 simulations with the population type set as “F₂”.

2.6.7. Differential expression analyses on BSR‐seq data from P and G bulks

DESeq2 (Love et al., 2014) was used to conduct differential expression analyses between the genotypic (G) bulks assembled for the SG density QTL on Chr01 and the LW QTL on Chr08. Because the genotypic bulks were not replicated, P bulks for the same traits were used as replicates to identify genes DE between P_H/G_H bulks and P_L/G_L bulks for each trait. Genes with fewer than 20 reads across the four samples for each QTL were removed prior to analysis. Expression counts were normalized, and DEGs were identified based on contrasts between high and low bulks. Genes were considered DE if they exhibited an absolute log₂ fold change ≥ 1 and an adjusted p‐value < 0.05.

2.7. Mining QTL regions for candidate genes

Candidate gene mining was done only for the Chr01 QTL for SG density and the Chr08 QTL for LW. For LW, we focused on genes located in the QTL region with functional annotations in either Oryza sativa or Eleusine coracana involved in cell proliferation, cell expansion, growth regulation and/or hormone signaling as potential candidates. To identify candidate genes for SG density, we assembled a list of genes involved in trichome initiation or density in Arabidopsis and conducted homology searches to identify putative Zoysia orthologs located within the SG density QTL. For genes that were shortlisted as putative candidate genes, we assessed whether contrasting alleles had differential expression levels in the bulks. We also manually assessed the presence of nucleotide variants with likely impacts on protein function that differentiated the parents and were heterozygous in F₁‐19‐TZ‐14321 and hence segregated in the F₂ population.

3. RESULTS

3.1. Leaf traits measured in zoysiagrass

3.1.1. LW and vein count

Zoysia japonica acc. Meyer leaves measured, on average, 3 mm at their widest point, whereas Z. matrella acc. PI 231146 had narrower leaves at 1.3 mm (Figure 1a–f and Figure S2). The F₁ hybrid, F₁‐19‐TZ‐14321, displayed an intermediate LW of 2.4 mm. The distribution of LW across 176 F₂ progeny is shown in Figure 1g, and examples of leaves from narrow‐ and wide‐leaf F₂ progenies are shown in Figure 1h,i. The individual measurements are given in Table S1. Vein count also differed between the parents of the F₂ population, averaging 28 veins per leaf in Meyer, 11 veins in PI 231146, and 23 veins in F₁‐19‐TZ‐14321. The trait distribution across the 176 F₂ individuals is shown in Figure 1j, and the underlying values are given in Table S1. To validate the accuracy of the R‐CNN model used in counting the number of leaf veins, a manual evaluation was performed on 25 randomly selected images using ImageJ. Vein count measurements obtained manually and with the R‐CNN model were highly correlated (R ² = 0.80, p = < 0.001) (Table S3).

FIGURE 1.

Zoysia japonica acc. Meyer (a, c, and e) and Zoysia matrella acc. PI 231146 (b, d, and f) grown in the University of Georgia (UGA) Tifton nursery (a and b; images provided by Dr. Sameer Khanal) and UGA greenhouse in Athens (c and d). Subparts e and f show representative leaves of each accession. Distribution of leaf width (g) and vein count (j) in the F₂ progeny; both traits exhibited a normal distribution based on the Lilliefors‐corrected Kolmogorov–Smirnov test (p > 0.05). The blue (right) arrow indicates the value of Meyer, the red (left) arrow that of PI 231146, and the green (middle) arrow marks the value of F₁‐19‐TZ‐14321. The red circle on the left in (g) highlights the F₂ progeny with narrow leaves included in the P_{L‐leaf width (LW)} bulk, while the red circle on the right denotes the progeny with wide leaves included in the P_H‐LW bulk. Leaves from a narrow‐leaf individual, Tray_90_6 (h), and from a wide‐leaf individual, Tray_69_15 (i).

Vein spacing and bulliform cell width were measured in transverse sections of Meyer (Figure 2a), PI 231146 (Figure 2b), the wide‐leaf F₂ Tray_87_13, and narrow‐leaf F₂ Tray_29_16, and data are provided in Table S4. Tukey's post hoc test showed that bulliform cell size differed significantly between accessions with wide leaves (Meyer and the wide‐leaf F₂) and narrow leaves (PI 231146 and the narrow‐leaf F₂) (Figure 2c). Further, it revealed statistically significant differences between all samples except PI 231146 and the narrow‐leaf F₂ for vein‐to‐vein distance (Figure 2d).

FIGURE 2.

Transverse sections of leaves of Meyer (a) and PI 231146 (b). The red line in (a) shows the vein‐to‐vein distance, while the white line shows the width of the largest bulliform cell. Scale bar: 50 µm. Representative transverse images for all four genotypes are given in Figure S2. Distribution of the width of the largest bulliform cell (c) and vein‐to‐vein distance (d) measured in transverse sections of Meyer, PI 231146, the wide‐leaf F₂ genotype Tray_87_13, and the narrow‐leaf F₂ genotype Tray_29_16. Lowercase letters above each box plot denote statistically significant differences at p < 0.01, as determined by one‐way analysis of variance (ANOVA) followed by Tukey's honestly significant difference (HSD) mean‐separation test conducted between each genotype. The y‐axis is in µm in both c and d.

3.1.2. Location and distribution of SGs

Imaging using scanning electron microscopy (SEM) and dissecting microscopy of the leaf surfaces of Meyer and PI 231146 revealed that the distribution of glands differed between the adaxial and abaxial leaf sides (Figure 3). On the adaxial surface (Figure 3a,c), the SGs are positioned parallel to and above the leaf veins, whereas on the abaxial surface (Figure 3b,d), glands lie parallel to but between the veins. Further, clear evidence of salt secretion through the formation of salt crystals was observed on the adaxial leaf side, but not on the abaxial side.

FIGURE 3.

SEM images of the adaxial (a and c) and abaxial (b and d) leaf surface of Meyer (a and b) and PI 231146 (c and d). Adaxial salt glands are highlighted with red circles (a and c). Scale bars: 200 µm. Salt gland density distribution in the F₂ progeny (e). The distribution showed a minor but statistically significant skew based on the Lilliefors‐corrected Kolmogorov–Smirnov test (p = 0.007). Blue arrow (left) shows salt gland density in Meyer, red arrow (right) shows salt gland density in PI 231146, and green arrow (middle) shows salt gland density in F₁‐19‐TZ‐14321. The red circle on the left in (e) highlights the F₂ progeny with lower salt gland density included in the P_{L‐salt gland (SG)} bulk, while the red circle on the right denotes the progeny with higher salt gland density included in the P_H‐SG bulk.

No significant difference in adaxial SG density was observed between leaves from freshwater‐ and saltwater‐grown plants in either of the two parents (Figure S3). However, the two parents exhibit distinct adaxial SG densities, averaging 23 glands per image (0.6 mm²) in Meyer and 49 glands per image (0.6 mm²) in PI 231146 (Figure S3). F₁‐19‐TZ‐14321 has an intermediate density of 39 glands per image (0.6 mm²). SG density in the F₂ progeny was assessed on two to six leaves per line using an R‐CNN model. Manual counts performed on 25 randomly selected images correlated highly (R ² = 0.81; p < 0.001) with the automated counts for the same images, demonstrating the high accuracy of the model (Table S4). Nevertheless, if the SG density for a leaf deviated by >20% from the standard deviation of the average for that progeny, this value was considered an outlier. A total of 19 outliers (out of 707 measurements) were identified, and the SGs in the corresponding images were manually recounted and, if needed, corrected values recorded. Across the F₂ population, SG density varied from 15 to 65 and averaged 35 (Figure 3e). The underlying data are given in Table S1.

3.1.3. Correlations between the leaf traits

A correlation analysis showed that LW and vein count were highly positively correlated (R ² = 0.81; p < 0.001) (Figure 4). LW and SG density showed a very weak negative correlation (R ² = 0.03; p < 0.05).

FIGURE 4.

Correlation matrix of leaf traits measured in the zoysiagrass F₂ population. Pairwise scatter plots are shown below the diagonal, histograms are shown on the diagonal, and the coefficient of correlation (Pearson's r) and coefficient of determination (R ²) for each pair of traits are shown above the diagonal (*p < 0.05, ***p < 0.001).

3.2. QTL mapping of leaf traits

3.2.1. Genomic regions associated with LW and vein count

LW and vein count are highly correlated traits (Figure 4), and CIM revealed colocalizing QTL on Chr08 and Chr17 (Figure 5a,b). A third QTL was identified for LW on Chr06 (Table 1). The QTL on Chr08 spanned 2.97 Mb, explained 12.6% of the variation in width and 16.6% of the variation in vein count (Table 1), and comprised 794 genes (408 with functional annotation) (Table S5). The QTL on Chr17 spanned 2.56 Mb, accounted for 13.6% of the variation in width and 10.3% of the variation in vein count (Table 1), and comprised 564 annotated genes (304 with functional annotation) (Table S6). The QTL specific to LW on Chr06 spanned 1.02 Mb, explained 9.1% of the phenotypic variation (Table 1), and comprised 532 genes (212 with functional annotations) (Table S7). MIM validated these QTL, and all loci combined explained a total of 34.4% and 26.4% of the variation for LW and vein count, respectively. No epistatic interactions were detected (Table S8).

FIGURE 5.

Graphs showing quantitative trait loci (QTL) for LW, vein count, and SG density on Chr08 (a); LW and vein count on Chr17 (b); and SG density on Chr01 (d). LOD, logarithm of the odds. Absolute Δ(SNP index) plot on Chr08 (c). The x‐axis shows the genomic position in Mb and y‐axis the absolute Δ(SNP index). The solid black line represents the tricube‐weighted abs Δ(SNP index). The colored lines represent the CI_95% (coral) and CI_99% (cyan) (where CI is confidence interval) under the null hypothesis of no QTL present. The red rectangle shows the leaf width/vein count QTL region from subpart a, and the black arrow shows the location of the six putative trait‐associated SNPs identified from the BSR‐seq analysis that cluster at position 17.2 Mb.

TABLE 1.

Quantitative trait loci (QTL) for leaf traits in the zoysiagrass F₂ population.

Trait	Chr	Marker at QTL peak a	Position (cM) of QTL peak	Position (bp) of QTL peak on pseudomolecule	Marker at left border of QTL region a	Position (cM) of left marker	Position (bp) of left marker on pseudomolecule	Marker at right border of QTL region a	Position (cM) of right marker	Position (bp) of right marker on pseudomolecule	LOD at QTL peak	LOD threshold (α < 0.05)	PVE%	Additive effect	Dominance effect	Phenotypic value of Meyer allele	Phenotypic value of PI 231146 allele	Phenotypic value of heterozygote
Leaf width	6	Tag_25481u	78.20	13,829,055	Tag_25309r	71.40	12,754,262	Tag_29480	83.30	15,028,804	6.15	4.09	9.08	0.01	0.00	1.97	1.92	1.96
Leaf width	8	Tag_21554u	91.90	17,204,444	Tag_17979ur	81.80	15,574,819	Tag_20547h	101.30	18,553,416	8.30	4.09	12.62	0.01	0.00	2.04	1.76	1.97
Leaf width	17	Tag_13660r	68.90	7,741,052	Tag_13938	58.00	6,359,463	Tag_13383r	77.70	8,923,706	9.09	4.09	13.62	0.01	0.00	2.09	1.82	1.94
Vein count	8	Tag_21538ur	92.20	17,303,630	Tag_17979ur	81.80	15,574,819	Tag_20547h	101.30	18,553,416	9.28	4.21	16.65	1.61	0.68	18.13	14.94	17.19
Vein count	17	Tag_13675	68.30	7,693,901	Tag_13938	58.00	6,359,463	Tag_13383r	77.70	8,923,706	5.98	4.21	10.31	1.20	0.35	18.48	15.59	17.00
Salt gland density	1	Tag_36921	49.10	5,767,822	Tag_13660h	40.80	4,220,785	Tag_1577r	60.70	12,576,203	5.61	4.11	10.66	−4.56	1.05	30.88	38.46	35.43
Salt gland density	8	Tag_21655	89.30	16,673,524	Tag_21730	86.70	16,313,544	Tag_21655	89.30	16,673,524	4.36	4.11	8.43	−5.28	−1.50	32.8	39.45	34.75

Abbreviation: LOD, logarithm of odd; PVE, phenotypic variation explained.

^a Marker information available from Pradhan et al. (2025).

In a second approach, we employed BSR‐seq to identify SNPs and genomic regions associated with LW. The number of PE reads for bulks P_H‐LW and P_L‐LW, which represent the extremes of the phenotypic distribution for LW, were 34.23 M and 24.26 M, respectively (Table S9). On average, 78% of the reads aligned to the Z. japonica cultivar Nagirizaki pseudomolecule reference genome (82% for P_H‐LW and 77% for P_L‐LW). SNP calling resulted in a total of 354,033 SNPs that were polymorphic between the bulks. This number was reduced to 110,788 after applying filters for quality, read depth, and reference allele frequency. Of the 7124 SNPs with a ΔSNP index above the 99% CI, 28 had a ΔSNP index value ≤ −0.8, 236 had a tricube‐weighted value ≤ −0.40, and six fulfilled all three criteria (Table S10). The six SNPs clustered at position 17.2 Mb on Chr08, which coincides with the QTL peak for LW and vein count obtained through CIM (shown in red rectangle in Figure 5c).

3.2.2. Genomic regions associated with SG density

CIM identified two regions associated with SG density, one on Chr01 (Figure 5d) and one on Chr08 (Figure 5a) (Table 1). The QTL on Chr01 encompassed 8.35 Mb and accounted for 10.6% of the phenotypic variation (Table 1). The Chr01 QTL interval comprised 357 genes (326 with functional annotations) (Table S11). The Chr08 QTL encompassed ∼359 kb, accounted for 8.4% of the phenotypic variation (Table 1), and comprised 102 genes (34 with functional annotations) (Table S12). These significant loci were reproduced in the MIM, explaining a total of 19.4% of phenotypic variation, and had no loci interactions (Table S8).

Similar to LW, we used BSR‐seq as an alternative approach to identify genomic regions associated with SG density. The total number of PE reads obtained for bulks P_H‐SG and P_L‐SG, representing the extremes of the phenotypic distribution for SG density, was 27.56 and 36.67 M, respectively (Table S8). On average, 75% of reads (77% for P_H‐SG, 74% for P_L‐SG) aligned to the Z. japonica cultivar Nagirizaki pseudomolecule reference genome (Tanaka et al., 2016). SNP calling identified a total of 359,858 SNPs that were polymorphic between the bulks. After filtering for quality, read depth, and minor allele frequency, the SNP number was reduced to 106,086. A total of 4323 SNPs had a ΔSNP index between the P_H‐SG and P_L‐SG bulks above the 99% CI. Of these, 119 had a tricube‐weighted ΔSNP index ≥ 0.40 and identified regions on chromosomes 6, 12, and 18. Across these three regions, seven SNPs (five on Chr12, one each on Chr06 and Chr18) had a ΔSNP index value ≤ −0.8 (Figure S4; Table S13) and were considered significantly associated with variation in SG density. None overlapped with the QTL identified using CIM.

3.3. Mining select QTL regions for candidate genes

3.3.1. Candidate genes for the LW QTL on Chr08

BSR‐seq conducted on genotypic (G) bulks with contrasting alleles in the LW QTL region on Chr08 yielded a total of 57.35 M reads for bulk G_H‐LW and 34.97 M reads for bulk G_L‐LW (Table S9). The percentage of reads aligning to the Z. japonica acc. Nagirizaki scaffold‐level reference genome (the scaffold‐level assembly was used because the gene annotation files are based on this assembly) was 87% for G_H‐LW and 84% for G_L‐LW. To identify DE genes within the Chr08 QTL region, we used genotypic (G) and phenotypic (P) bulks, consisting of contrasting trait phenotypes, as replicates. Although G and P bulks were composed based on slightly different criteria, 63% of the variation in expression between the bulks G_H‐LW, P_H‐LW, G_L‐LW and P_L‐LW could be attributed to the difference in LW between the G_H‐LW/ P_H‐LW and G_L‐LW/ P_L‐LW bulks (Figure S5). Genes with >20 gene counts across all four samples were included in the analyses. A total of 1775 genes (1277 with functional annotation) were DE (Log₂FC ≥ |1| and adjusted p‐value < 0.05) between bulks P_H‐LW/G_H‐LW and P_L‐LW/G_L‐LW (Table S14). Of these, 70 DE genes were located within the QTL region, which spanned 2.97 Mb and contained a total of 794 genes. When considering only functionally annotated genes, 51 (out of 408) were DE and 336 were expressed at similar levels across both sets of bulks in the QTL region.

Because the Chr08 LW QTL showed a sharp peak (Figure 5a), we only focused on candidate genes located within the peak region plus 500 kb on either side of the peak (region from 16.56 to 17.51 Mb highlighted in green in column A in Table S5) for candidate gene mining. Genes whose associated gene ontology (GO) terms indicated involvement in cell division (GO:0007049), growth (GO:0051260; GO:0007275), developmental process (GO:0032502), and hormone signaling (GO:0007165) were considered candidates for the LW QTL, irrespective of their expression status (column P in Table S5). Genes without any of these five associated GO terms but shown in previous studies to be involved in cell expansion or cell size were also considered as potential candidates. The candidate genes along with their annotated function, GO terms when relevant, and location in the Z. japonica acc. Nagirizaki pseudomolecules are given in Table 2. We examined whether the genes were expressed in leaves, DE between the bulks (Table 2), or carried heterozygous variants in F₁‐19‐TZ‐14321 that could potentially impact protein function (Table S15).

TABLE 2.

Putative candidate genes underlying the colocalizing quantitative trait locus (QTL) for leaf width and vein count on Chr08.

Gene	GO term	Functional annotation a	Accession number	Percent identity	Query coverage (%)	Distance from QTL peak	Avg transcript count (GH‐LW/PH‐LW)	Avg transcript count (GL‐LW/PL‐LW)	Log2FC	adj p‐value	References
Zjn_sc00022.1.g05490.1.am.mk	GO:0032502	Os growth‐regulating factor 10	Q6EPP9.1	75.47	25	34 kb	NA d	NA d	NA d	NA d	Wu et al., 2014
Zjn_sc00022.1.g04180.1.am.mk b	GO:0032502	Os growth‐regulating factor 4	Q6ZIK5.1	59.42	22	490 kb	63.34 c	59.38 c	0.14	0.97	Horiguchi et al., 2005
Zjn_sc00022.1.g06450.1.am.mk b	No term	Os expansin B5	Q7XT39.2	41.96	88	490 kb	0	12.59	−6.19	0.11	Choi et al., 2003
Zjn_sc00022.1.g05410.1.sm.mkhc b	No term	At REVEILLE 6	Q8H0W3.1	49.62	83	8 kb	550.27	555.51	−0.01	0.99	Gray et al., 2017
Zjn_sc00022.1.g05360.1.am.mk b	No term	At Scarecrow‐like 8	Q9FYR7.1	40.55	51	18 kb	873.85	748.23	0.22	0.68	Slewinski et al., 2012
Zjn_sc00022.1.g07470.1.sm.mk	GO:0007275	Os YABBY 4	A2 × 7Q3.1	82.69	100	893 kb	21.47	101.43	−2.25	0.00	Zhi et al., 2022
Zjn_sc00017.1.g00530.1.am.mk	GO:0007275	At AP2‐like ethylene‐responsive transcription factor baby boom	Q6PQQ4.2	41.27	73	∼2 Mb	60.77	81.55	−0.42	0.81	Feng et al., 2020
Zjn_sc00022.1.g05470.1.sm.mkhc b	GO:0007165	At Rho GTPase‐activating protein 3	Q8GYY5.1	52.00	93	27 kb	562.35	486.45	0.20	0.76	Kost, 2008
Zjn_sc00017.1.g00800.1.cf.mkhc	GO:0007049	At Enhancer of rudimentary homolog	Q96319.1	74.00	97	∼2 Mb	1369.21	549.04	1.32	0.00

Abbreviation: LW, leaf width.

^a Based on homology in NCBI's SwissProt database.

^b Genes within 500 kb of the QTL peak that are discussed further in this study.

^c Combined reads of Zjn_sc00022.1.g04180.1.am.mk and Zjn_sc00022.1.g04190.1.am.mk, which are part of the same gene.

^d NA: Gene not expressed.

The protein encoded by Zjn_sc00022.1.g05490.1.am.mk (GO:0032502) had the highest homology in NCBI's SwissProt to rice growth‐regulating factor 10 (Table 2). The gene is located 34 kb from the QTL peak but is not expressed in any of the bulks. Further, variant analysis indicated that both Meyer and PI 231146 were heterozygous for a nonfunctional allele and that F₁‐19‐TZ‐14321 had inherited two inactive alleles. Hence, this gene was not further considered.

The protein encoded by Zjn_sc00022.1.g04180.am.mk had the highest homology in SwissProt to maize growth‐regulating factor 4 (Table 2). The gene was located 490 kb from the QTL peak. Zjn_sc00022.1.g04180.am.mk is expressed in both LW bulks but not DE between the contrasting bulks. Sequence analysis identified a total of eight heterozygous non‐synonymous SNPs in F₁‐19‐TZ‐14321 (Table S15).

The protein encoded by Zjn_sc00022.1.g06450.1.am.mk had the highest homology in SwissProt to rice expansin B5 protein (Table 2). The gene is located 490 kb from the QTL peak. It is not expressed in the wide‐leaf bulks and has relatively low read counts in the narrow‐leaf bulks. Two non‐synonymous SNPs were heterozygous in F₁‐19‐TZ‐14321 (Table S15).

The protein encoded by Zjn_sc00022.1.g05410.1.sm.mkhc had the highest homology in SwissProt to Arabidopsis REVEILLE 6 and is located very close (8 kb) to the QTL peak. This gene had high expression levels in leaves but was not DE. Sequence analyses between the parents and comparison with the orthologous gene in finger millet showed that both Meyer and the Z. japonica Nagirizaki reference genome carry a 1 bp deletion after position Zjn_sc00022.1: 2237862 in the penultimate exon that leads to a premature stop codon. F₁‐19‐TZ‐14321 was heterozygous for the 1 bp deletion (Figure S6).

The protein encoded by Zjn_sc00022.1.g05470.1.sm.mkhc had the highest homology to Arabidopsis Rho‐GTPase‐activating protein 3. This gene is located ∼27 kb from the QTL peak and is not DE in the contrasting LW bulks. This gene lacks a homoeologous copy in both Z. japonica acc. Nagirizaki and Z. matrella acc. Wakaba. The six non‐synonymous SNPs between the two parents are given in Table S15.

A final candidate, Zjn_sc00022.1.g05360.am.mk, encodes a protein with the highest homology in SwissProt to Arabidopsis SCARECROW‐like 8 (SCL8). The gene is also located very close to the peak of the QTL (18 kb). Eleven SNPs leading to amino acid changes that are heterozygous in F₁‐19‐TZ‐14321 are given in Table S15.

3.4. Candidate genes for the SG density QTL

BSR‐seq conducted on genotypic (G) bulks consisting of contrasting alleles in the QTL region on Chr01 yielded a total of 49.32 M reads for bulk G_H‐SG and 52.01 M reads for bulk G_L‐SG (Table S9). The percentage of reads that aligned to the Z. japonica acc. Nagirizaki scaffold level genome assembly was 85% for G_H‐SG and 83% for G_L‐SG. To identify genes that were DE in the Chr01 QTL region, genotypic (G) and phenotypic (P) bulks, consisting of contrasting phenotypes for the trait, were used as replicates. Expression differences between G_H‐SG/P_H‐SG and G_L‐SG/P_L‐SG bulks accounted for 94% of the variation (Figure S5). A total of 3512 genes were DE (Log₂FC ≥ |1| and adjusted p‐value < 0.05) between bulks P_H‐SG/G_H‐SG and P_L‐SG/G_L‐SG (Table S16). Sixty‐three DE genes were in the QTL region, representing 17.6% of the genes present. An additional 283 genes in the QTL region (72.3%) were expressed (more than 20 gene counts across all four samples used for the DESeq2 analyses) at similar levels in both sets of bulks (Table S11). The corresponding percentages for functionally annotated genes (343) are 17.7% (61 DE genes) and 75.5% (259 genes similarly expressed in the bulks).

To identify genes potentially involved in SG density, a list was compiled of genes involved in trichome development and trichome density in Arabidopsis because SGs are considered glandular trichomes (Dassanayake & Larkin, 2017), and are hypothesized to share similar developmental pathways. Table 3 presents a list of these genes, including their Arabidopsis IDs used for BLASTp searches, known functions, and corresponding Z. japonica ortholog IDs and genome locations. Three genes, two located in the Chr01 QTL region (Zjn_sc00010.1.g01190.1.sm.mkhc and Zjn_sc00046.1.g01300.1.am.mkhc) and one located close to the Chr08 QTL (Zjn_sc00022.1.g05300.1.es.mkhc), were selected for further evaluation (in bold in Table 3).

TABLE 3.

Genes from Arabidopsis involved in trichome initiation and density, and their homologs and genomic locations in Z. japonica.

Gene name	Gene used as query	Function	Zoysia ortholog	Location	References
GLABRA1 (GL1)	AT3G27920	Essential for trichome initiation	Zjn_sc00078.1.g02720.1.am.mk	Chr13; ∼11.3 Mb	Han et al., 2022
GLABRA3 (GL3)	AT5G41315	Promotes trichome development; Mutations lead to reduced trichome number	Zjn_sc00004.1.g07010.1.sm.mk	Chr12; ∼11.6 Mb	Esch et al., 2003
TT8	AT4G09820	Involved in trichome differentiation	Zjn_sc00030.1.g01160.1.am.mk	Chr04; ∼1.6 Mb	Matsui et al., 2008
ENHANCER OF GLABRA3 (EGL3)	AT1G63650	Act redundantly to GL3 to promote trichome development	Zjn_sc00030.1.g01160.1.am.mk	Chr04; ∼11.6 Mb	Song et al., 2015
AtMYC1	AT4G00480	AtMYC1 mutants have less trichomes	Zjn_sc00004.1.g07010.1.sm.mk	Chr12; ∼11.6 Mb	Zhao et al., 2012
TRANSPARENT TESTA GLABRA1 (TTG1)	AT5G24520	Interacts with GL1 and GL3/EGL3 to form a complex promoting trichome formation	Zjn_sc00019.1.g00730.1.am.mkhc; Zjn_sc00022.1.g05300.1.es.mkhc a	Chr02; ∼7.5 Mb; Chr08; ∼17.2 Mb	Maes et al., 2008
JASMONATE‐ZIM DOMAIN (JAZ)	AT1G19180	Negative regulators; Overexpression of JAZ proteins can suppress trichome formation	Zjn_sc00010.1.g01190.1.sm.mkhc; Zjn_sc00046.1.g01300.1.am.mkhc b	Chr01; ∼9.9 Mb; Chr01; ∼8.1 Mb	J. Li et al., 2021
ZINC FINGER PROTEIN (ZFP8)	AT2G41940	Involved in trichome initiation	Zjn_sc00066.1.g03030.1.am.mk	Chr08; ∼19.6 Mb	Zhou et al., 2013

^a Located near the Chr08 QTL for salt gland density.

^b Not a homolog but involved in the jasmonic acid biosynthesis pathway.

Zjn_sc00010.1.g01190.1.sm.mkhc (on Chr01) encodes a jasmonate ZIM domain‐containing protein, which is a key regulator in the jasmonate signaling pathway, and is located 4.16 Mb from the QTL peak. Zjn_sc00010.1.g01190.1.sm.mkhc exhibited similar transcript levels in both SG bulks. Variant analysis in Meyer, PI 231146, and F₁‐19‐TZ‐14321 showed the presence of two non‐synonymous SNPs that were heterozygous in both Meyer and F₁‐19‐TZ‐14321 relative to the Z. japonica acc. Nagirizaki reference genome (Table S17).

Zjn_sc00046.1.g01300.1.am.mkhc (on Chr01) encodes allene oxide cyclase (AOX). While not previously associated with trichome initiation, we included AOX because it is a key enzyme in jasmonic acid (JA) biosynthesis. Zjn_sc00046.1.g01300.1.am.mkhc is located ∼2.40 Mb from the QTL peak. Expression of Zjn_sc00046.1.g01300.1.am.mkhc in the high‐density bulks (P_H‐SG and G_H‐SG) was nearly double the expression observed in the low‐density bulks (P_L‐SG and G_L‐SG). Variant analysis showed that AOX was highly conserved and only carried synonymous SNPs.

Intriguingly, Zjn_sc00022.1.g05300.1.es.mkhc, a WD40 protein with strong homology to Arabidopsis Transparent Testa Glabra1 (TTG1), was located ∼55 kb outside the QTL region and ∼450 kb from the QTL peak obtained for SG density on Chr08 (Tables 1 and 3). TTG1 plays a key role in the trichome developmental pathway (Maes et al., 2008). While Zjn_sc00022.1.g05300.1.es.mkhc (Zj_TTG1) was not DE in the SG density bulks, which were composed to target the Chr01 QTL, it had differential transcript levels in the high and low Chr08 LW bulks. The Chr08 SG density QTL partially overlaps with the Chr08 QTL for LW and vein count (Figure 5a), and Zj_TTG1 colocalized with the QTL peak for those traits. PI 231146, which has the higher SG density, has three non‐synonymous SNPs (Table S17). Further, the 3’ untranslated region (3’ UTR) contains an intron (positions 2,185,305–2,186,494) that is largely spliced out in the G_H‐LW bulk but is retained in the G_L‐LW bulk. PI 231146 also has an 11 bp insertion in the non‐intronic part of the 3’ UTR, which is heterozygous in F₁‐19‐TZ‐14321 (Figure S7). Finally, the annotation for Zj_TTG1 (Zjn_sc00022.1.g05300.1.es.mkhc) in Z. japonica acc. Nagirizaki (ZJN_r1.1.fa; Tanaka et al., 2016) does not accurately represent the gene structure, and a corrected annotation is provided in Figure S8.

4. DISCUSSION

4.1. The genetic control of LW and vein count is highly conserved across grasses

Two colocalizing QTL for LW and vein count, and a third QTL for LW were identified in a Z. japonica × Z. matrella interspecific F₂ mapping population. The phenotypic variation explained (PVE) by individual QTL was low (range 9.1%–16.7%; Table 1) with the three QTL identified, combined, contributing 34.4% of the total trait variation as shown by MIM (Table S8). The unaccounted 65.6% is likely due to the cumulative effects of many genes with small individual contributions not strong enough to be detected with the population size used. Previous mapping efforts have primarily focused on zoysiagrass F₁ populations (Huang et al., 2016; Tsuruta et al., 2010; F. Wang et al., 2015), both inter‐ and intraspecific, but only one Zoysia study has reported QTL for LW. Tsuruta et al. (2010) identified two major QTL and a minor QTL for LW in a Z. japonica × Z. matrella F₁ population genotyped with amplified fragment length polymorphism (AFLP) markers. While the number of QTL is similar as found in our study, a lack of sequence information for the AFLP markers does not allow us to compare the locations. All three leaf QTL in our study have positive additive effects, with the Meyer alleles on Chr08, Chr17, and Chr06 increasing LW by 0.28, 0.27, and 0.05 mm, respectively (Table 1). The fact that transgressive segregation is largely absent for LW (Figure 1g) suggests that most of the width‐increasing alleles segregating in the analyzed F₂ population are derived from Meyer, the wider‐leaf parent.

LW, measured at its widest point, is highly correlated with vein count (R ² = 0.81***; Figure 4). The two QTL identified for vein count on Chr08 and Chr17 co‐mapped with QTL for LW. This suggests that LW determines the number of veins present, or vice versa, as has been observed in other monocots (Hughes et al., 2019; Smillie et al., 2012). Trait mapping of LW and vein patterning has been explored in several grass species, with multiple associated loci identified in rice (Tian et al., 2015), maize (Strable et al., 2017), and sorghum (Zhi et al., 2022). Two QTL reported by Zhi et al. (2022) for LW in sorghum, qLW_dtf4.2 and qLW_dtf9, overlapped at least partially with the orthologous Zoysia regions associated with LW on Chr08 and Chr17, respectively, in our study. Similarly, a rice LW QTL (qTLW2) identified by Wen et al. (2020), two coinciding QTL (qFWn2‐2 and qFWr2‐3) identified by J. Wang et al. (2022), and FLW1 identified by Yue et al. (2024) all appeared to be orthologous to the Chr08 QTL mapped in Zoysia in the current study. In addition, the location of a flag LW QTL in rice (qFLW5; B. Zhang et al., 2015) and two regions associated with uppermost LW1 in maize identified by Zeng et al. (2025) were orthologous to the Chr17 QTL region identified here for Zoysia. The occurrence of orthologous QTL for LW in Zoysia and other grass species demonstrates that these QTL are robust and conserved contributors to LW variation in grasses.

4.2. Two novel QTL for SG density

QTL for SG density were identified on Chr01 and Chr08, explaining 10.7% and 8.4% of the variation, respectively (Table 1). Both QTL exhibit negative additive effects, with the PI 231146 allele increasing the SG density by 7.6 and 6.5 glands/0.6 mm² for the Chr01 and Chr08 QTL, respectively. Interestingly, the Chr08 SG density QTL partially overlapped with the Chr08 LW and vein number QTL, suggesting pleiotropy of, or, more likely, close linkage between, the causal genes. Bidirectional transgressive segregation was observed in the F₂ population (Figure 3e), suggesting that specific combinations of alleles from each parent contribute to phenotypes that exceed the parental extremes. QTL for which Meyer contributes beneficial alleles, however, were not detected. To our knowledge, no QTL associated with SG density have previously been reported in zoysiagrass. QTL for microhairs and trichomes, structures considered orthologous to SGs (Dassanayake & Larkin, 2017), have been identified in Triticum aestivum (Fan et al., 2023), Solanum lycopersicum (Mata‐Nicolás et al., 2021), Sorghum bicolor (Kiranmayee et al., 2016), and Zea mays ssp. parviglumis (Lauter et al., 2004). None of the trichome QTL mapped in the three grass species were located on chromosomes that were syntenic to Chr01 or Chr08 in zoysiagrass.

4.3. BSR‐seq versus CIM to identify trait‐associated chromosomal regions

We conducted both CIM and BSR‐seq to identify loci associated with leaf traits. Combining multiple QTL analysis approaches enhances genetic dissection by increasing confidence in QTL detection (Jia et al., 2023; Q. Li et al., 2020; Z. Wang et al., 2024). If we solely considered ΔSNP values above the 99% confidence interval (CI_99), BSR‐seq identified trait‐associated regions on most chromosomes. Applying an additional threshold of absolute tricube‐weighted ΔSNP values ≥ 0.40 refined these results, yielding three regions associated with SG density (compared to two with CIM), and two regions associated with LW (compared to three with CIM). Finally, adding the strict criterion for absolute ΔSNP values to be higher or equal to 0.8 retained the three regions on Chr06, Chr12, and Chr18 for SG density and the Chr08 region for LW. Of the four putative trait‐associated regions identified by BSR‐seq, only the Chr08 region for LW was identified using both BSR‐seq and CIM. The lack of overlap between the results of the two methods is on par with findings from other studies (Jia et al., 2023; J. Li et al., 2020; Z. Wang et al., 2024). A key limitation of BSR‐seq is its reliance on expressed genes, unlike whole‐genome sequencing, which captures variation across the entire genome. Genes that are not expressed at the time of tissue collection will be missed in BSR‐seq. Further, if a gene is not or very lowly expressed in one bulk, variants will be available only for the highly expressing bulk and, hence, filtered out. Another limitation here was that BSR‐seq was conducted on single bulks of only eight individuals from the extremes of the phenotypic distribution, which may have amplified sampling bias (C. Zou et al., 2016). Further, because BSR‐seq reads were aligned solely to the Z. japonica genome assembly, it is possible that trait‐associated regions unique to Z. matrella were missed. However, such regions are expected to be rare. In contrast, CIM is statistically more stringent because it uses models that account for background genetic effects. The stringency can filter out some smaller effect QTL detected by BSR‐seq but provides greater confidence in the loci it identifies (Edae & Rouse, 2019).

4.4. Bulliform cell size affects interveinal distance but likely is not the main driver of the LW variation between Z. japonica and Z. matrella

Because the Chr08 QTL for LW was jointly identified by BSR‐seq and CIM, we focused on this QTL, which co‐mapped with a QTL for vein count, for the identification of candidate genes.

Although LW QTL have been identified in orthologous locations across different grasses, no strong candidate genes have emerged from these studies as candidates for the zoysiagrass Chr08 region. The suggested candidate genes for the corresponding QTL in rice by J. Wang et al. (2022) (LOC_Os02g54254) and Yue et al. (2024) (LOC_Os02g43010) mapped outside the 1 Mb Chr08 QTL region deemed most likely to carry the causal gene. We noted, however, that the peak of the Zoysia Chr08 LW QTL falls within a fine‐mapped region carrying the rice semi‐rolledleaf3 (SRL3) locus. The leaf rolling phenotype controlled by the rice SRL3 locus is associated with differences in bulliform cell area and number (Yu et al., 2022). The srl3 mutant also has narrower leaves.

Measurement of bulliform cell size in transverse sections of Zoysia leaves (Figure 2a,b) showed that the width of the largest bulliform cell was significantly larger in Meyer and a wide‐leaf F₂ progeny (Tray_87_13) than in PI 231146 and a narrow‐leaf F₂ progeny (Tray_29_16) (Figure 2c). As a consequence, Meyer and the wide‐leaf F₂ progeny had significantly greater interveinal spacing than PI 231146 and the narrow‐leaf F₂ individual (Figure 2d). While we did not quantify leaf rolling in the F₂ population, we did note during sampling and transverse sectioning that narrower leaves rolled more prominently (Figure S2). ROC5 (LOC_Os02g45250), a homolog of Arabidopsis GLABRA2 that has previously been associated with bulliform cell size and leaf rolling in rice (L.‐P. Zou et al., 2011), is located just outside the 1 Mb region surrounding the Zoysia Chr08 QTL peak. While it is possible that variation in ROC5 leads to bulliform cell size reduction and enhanced leaf rolling of narrow‐leaf individuals in the Zoysia F₂ population, we do not consider it a candidate for the Chr08 LW and vein count QTL mapped in our study. ROC5 is also located outside the fine‐mapped SRL3 region in rice (Yu et al., 2022), suggesting that another gene that controls bulliform cell size and leaf rolling and, potentially, leaf size is located in the rice SRL3 region and, assuming orthology, in the Zoysia Chr08 QTL region. It should be noted, however, that intervein distances and bulliform cell size in our study were measured on fixed tissues that were dehydrated, while the LW and vein count phenotypes were determined on fresh leaves. Vein density did not differ significantly between the parental lines Meyer and PI 231146 (p > 0.05), and while the trait was normally distributed across the F₂ progenies, no QTL for vein density was detected. The same was true when the “adjusted vein count”, calculated as the residuals from a linear regression of vein count against LW (derived trait), was used as a trait. We suggest that altered bulliform cell morphology and vein spacing is a secondary factor influencing LW but is likely not the primary cause.

4.5. Auxin‐related genes associated with LW/vein number QTL

The role of auxin in both leaf initiation and vein patterning is well established (Perico et al., 2022). PIN1‐mediated efflux is required to initiate leaves at the shoot apical meristem flanks, and coordinated action of auxin transporters, including the auxin influx carrier LAX1, commits ground‐meristem cells to pre‐procambial and then to procambial fates, laying the foundation for leaf venation. Given the established importance of auxin in regulating both the orderly formation and the number of veins (Robil & McSteen, 2023), we looked for auxin‐related candidate genes for the LW/vein number QTL identified in our study. The ortholog of OsPIN1, Zjn_sc00022.1.g01530.1.sm.mkhc, fell just outside the broader QTL region on Chr08. A homolog of Forked 1 (Zjn_sc00022.1.g06530.1.sm.mk), an auxin canalization gene required for PIN1 localization in developing veins (Hou et al., 2010), was located within the broader QTL region but 524 kb from the QTL peak and hence just outside the 1 Mb region most likely to carry the causal gene. Nevertheless, Zjn_sc00022.1.g06530.1.sm.mk may be worth further investigation as it lacked a homoeolog on Chr18, so mutations would likely lead to a phenotype, and carried several SNPs that led to amino acid changes in conserved domains between Meyer and PI 231146 with heterozygous status in the F₁ (Table S15). Zjn_sc00022.1.g06530.1.sm.mk was not expressed in mature leaves (Chr08 bulks) at the time of sampling.

While we focused on the Chr08 QTL for candidate gene mining, it should be noted that the ortholog of LAX1, Zjn_sc00012.1.g05710.1.sm.mk (the actual gene structure is a combination of three predicted genes Zjn_sc00012.1.g05690.1.sm.mk, Zjn_sc00012.1.g05700.1.sm.mk, and Zjn_sc00012.1.g05710.1.sm.mk), falls in the LW and vein count QTL region we identified on Chr17 (657 kb from the QTL peak) and hence could be a candidate for this QTL. This gene was lowly expressed in mature leaves and not DE between the two extreme bulks. Important to note here is that the genome assembly we used (Nagirizaki; Tanaka et al., 2016) had a string of ∼3000 Ns across this gene that covered two exons, so SNPs differing between the parents were identified by aligning Z. japonica acc. Meyer, Z. matrella acc. PI 231146, and F₁ whole genome sequencing reads (Pradhan et al., 2025) to a newer Z. japonica acc. “Compadre” assembly by Shen et al. (2025) (Table S15). A heterozygous non‐synonymous SNP (p.Y278C) was present in the conserved auxin influx permease domain in PI 231146 and the F₁. The ancestral amino acid (Meyer: p.278Y) is conserved across most grass species, indicating a potential functional effect of this mutation. Given that vein patterning is established early during leaf development, expression analyses in the parents and leaf bulks at the stage of leaf initiation could help determine whether this gene functions differentially at an early stage to influence vein count.

4.6. Genes involved in cell expansion or division as candidates for LW/vein count

We also analyzed the variation present in five candidate genes putatively involved in cell expansion or division and located within a 1 Mb region surrounding the QTL peak (500 kb on either side of the peak) (Table 2). Zjn_sc00022.1.g05490.1.am.mk was excluded from the discussion because it carried likely function‐inactivating mutations in both alleles in F₁‐19‐TZ‐14321. Zjn_sc00022.1.g06450.1.am.mk encodes a protein with high homology to rice OsEXPB5, an expansin implicated in altering cell size by modulating cell‐wall extensibility (Choi et al., 2003). However, cell size variation was observed only for bulliform cells, and prior work has not linked expansins to bulliform cell development. Further, the two SNPs that differentiated the parents and were heterozygous in the F₁ were not located in conserved domains. Zjn_sc00022.1.g06450.1.am.mk, therefore, was unlikely to be causal to the QTL.

Zjn_sc00022.1.g05410.1.sm.mkhc encodes a protein that has the highest homology to REVEILLE‐6 in Arabidopsis. REVEILLE‐6, along with its other homologs, reduces growth of leaves as well as overall biomass by regulating cell size in Arabidopsis, and has been linked to growth advantages in polyploids (Gray et al., 2017). The Meyer allele of REVEILLE‐6 carries a 1‐bp deletion resulting in a premature stop codon in the penultimate exon. This mutation is heterozygous in the F₁ individual F₁‐19‐TZ‐14321. In contrast, the PI 231146 allele appears functional (Figure S6). Functional inactivation in the line with the largest LW, Meyer, aligns with findings from Arabidopsis, where knockout of REVEILLE‐6 and its homologs led to increased leaf area (Gray et al., 2017). However, Zoysia is a tetraploid, and the homoeolog in Meyer appears functional. In Arabidopsis, functionally redundant REVEILLE genes needed to be knocked out to obtain a phenotype. Zjn_sc00022.1.g05410.1.sm.mkhc was therefore not considered a strong candidate.

Zjn_sc00022.1.g04180.1.am.mk, homologous to the growth‐regulating factor ZmGRF1, exhibited low expression levels in mature Zoysia leaves, consistent with data from maize showing that ZmGRF1 is primarily expressed in immature leaves and suggesting a role in cell proliferation (Horiguchi et al., 2005; Kim & Lee, 2006). ZmGRF1 expression has been linked to an overall reduction in leaf area (L. Wu et al., 2014). None of the 10 non‐synonymous mutations differentiating Meyer and PI 231146 were in conserved domains, and neither was the gene DE in the contrasting bulks. It is possible that this gene is differentially regulated in the meristematic zone during the leaf initiation stage, with expression gradually decreasing as the leaf matures, as observed in tea (Z.‐J. Wu et al., 2017). Expression studies in the parents and leaf bulks during leaf initiation are needed to determine whether Zjn_sc00022.1.g04180.1.am.mk is a viable candidate.

Zjn_sc00022.1.g05470.1.sm.mkhc, encoding a protein homologous to Arabidopsis Rho‐GTPase‐activating protein 3, lacks a homoeolog in both Z. japonica acc. Nagirizaki and Z. matrella acc. Wakaba. While mutations or expression changes in genes present as a single copy are more likely to have a phenotypic effect, none of the six amino acid substitutions differentiating the Meyer and PI 231146 proteins encoded by Zjn_sc00022.1.g05470.1.sm.mkhc are in conserved domains. Rho GTPases are known to regulate key cellular processes such as directional expansion, motility, and division (Kost, 2008) and play a central role in establishing cell polarity through cytoskeletal organization and vesicle trafficking. Further functional studies are needed to determine if Zjn_sc00022.1.g05470.1.sm.mkhc contributes to the genetic control of LW and/or vein count.

Perhaps the most promising candidate, Zjn_sc00022.1.g05360.1.am.mk, encodes a likely SCL8 protein. Its rice ortholog, LOC_Os02g45760, was identified as one of the 14 genes tightly linked to the semi‐rolled leaf mutant srl3 in a fine‐mapping experiment (Yu et al., 2022). SCARECROW (SCR)‐like genes have high similarity in the carboxy‐terminal region to SCR (Pysh et al., 1999), a key regulator of monocot leaf development (Hughes et al., 2023). In maize, a Mutator transposon insertion in the first exon of the ZmSCR1 gene caused significant alterations in leaf morphology, including abnormal vein development, the formation of additional bundle sheath layers, and reduced vein density due to minor vein termination (Slewinski et al., 2012). Hughes et al. (2019) further demonstrated that SCR1 and its homoeolog SCR1h have redundant functions in maize and regulate cell division in the innermost ground meristem layer of leaf primordia, determining the number of mesophyll cells that develop between veins. The functions of SCR‐like genes, however, are diverse with some being similar to (Long et al., 2015) and others differing from those of SCR (Cui et al., 2014; Z.‐L. Zhang et al., 2011). Although the function of SCL8 is not known, in Arabidopsis, it has been shown to be a direct target of the GLABRA1/GLABRA3 complex, which regulates trichome formation (Morohashi & Grotewold, 2009). This is intriguing, considering the close association of SCL8 with the SG QTL on Chr08 as well as the observation that SGs are located above the veins on the adaxial leaf side (Figure 3). Zoysia SCL8 (Zjn_sc00022.1.g05360.1.am.mk) does not have a homoeologous copy in Z. japonica acc. Nagirizaki. It was highly expressed, but not DE, in the contrasting P_H‐LW/G_H‐LW and P_L‐LW/G_L‐LW bulks. Five of the 12 non‐synonymous SNPs that differentiate PI 231146 from Meyer and are heterozygous in F₁‐19‐TZ‐14321 (Table S15) occur in the conserved GRAS domain (Figure S9; Table S15). Three of these lead to nonconservative amino acid substitutions (p.S264F, p.W536R, and p.R529G) and may have functional implications by affecting protein‐protein interactions or protein folding (Teng et al., 2010). The ancestral amino acids, two of which are present in Meyer (p.264S and p.529R) and one in PI 231146 (p.536R), are conserved across most grass species, further supporting a potential functional effect of the mutations. Whether SCL8 indeed plays a role in vein count, LW, and/or potentially, bulliform cell shrinkage will need to be determined experimentally.

4.7. Zj_TTG1 is a strong candidate for regulating SG density in zoysiagrass

SGs have been proposed to be specialized glandular trichomes (Dassanayake & Larkin, 2017; de Luna et al., 2020; Guo et al., 2023). Based on the known genes involved in trichome formation, we identified two potential candidate genes for the SG density QTL. The first candidate gene, Zjn_sc00046.1.g01300.1.am.mkhc, is located in the Chr01 QTL and encodes a likely allene oxide cyclase 3 (AOC3). AOC3 catalyzes the conversion of unstable allene oxide intermediates into 12‐oxo‐phytodienoic acid, a critical step in the biosynthesis of JA (Zdyb et al., 2018). JA regulates stress responses and promotes glandular trichome development in different species, including tomato, sunflower, and tobacco (Ahn et al., 2015; Qi et al., 2011; Yan et al., 2017). High AOC3 expression leads to higher JA levels (Zdyb et al., 2018). JA alleviates the inhibitory effects of JAZ genes and activates key transcription factors such as GLABRA1 (GL1), GLABRA3 (GL3), and TTG1, which are major regulators of trichome initiation (Boughton et al., 2005; Traw & Bergelson, 2003; Yang & Ye, 2013). By extension, higher JA levels may thus lead to a higher SG density. The AOC3 gene has a twofold higher expression level in high SG density bulks (P_H‐SG/G_H‐SG) compared to low‐density bulks (P_L‐SG/G_L‐SG). However, the homoeolog of Zjn_sc00046.1.g01300.1.am.mkhc, Zjn_00108.1.g01180.1.sm.mkhc, also has a twofold higher expression level in high versus low SG density bulks, but no corresponding QTL was identified on Chr02, which is homoeologous to Chr01 (Pradhan et al., 2025). The variation in AOC3 transcription may therefore be a secondary effect rather than causal to the SG density variation mapped to Chr01.

Zj_TTG1 (Zjn_sc00022.1.g05300.1.es.mkhc) is located 55 kb outside the SG density QTL on Chr08 but warrants further investigation. SG density is not uniform across the entire leaf, and while every attempt was made to be consistent in the selection of the leaf area for phenotyping, SG counts and, hence, the precision of the QTL location may have been influenced by sampling variation. TTG1 genes are well‐known regulators of trichome development (Bouyer et al., 2008; F. Zhang et al., 2003) that can both positively regulate density, as for trichomes in Arabidopsis, or negatively regulate it, as observed for SGs in Limonium (Yuan et al., 2022) and glandular trichomes in cucurbits (Feng et al., 2021). Examination of the structural variants in Zjn_sc00022.1.g05300.1.es.mkhc between Meyer and PI 231146 identified three non‐synonymous SNPs, one of which (Zjn_sc00022.1:2184456; p.S69A) was in a conserved WD‐repeat. The Ser to Ala change is heterozygous in F₁‐19‐TZ‐14321. Ala is the ancestral amino acid and is conserved across most grass species.

TTG1 in Zoysia contains an intron in the 3′ UTR. Interestingly, the 3’ UTR intron was spliced out in the wide‐leaf genotypic bulk (G_H‐LW) but largely retained in the narrow‐leaf bulk (G_L‐LW) (Figure S7). Because the Chr08 QTL for SG density and LW overlap, the Chr08 LW genotypic bulks also represent alternative alleles for the genes involved in SG density at that locus. The presence of an intron in the 3’ UTR may significantly enhance nonsense‐mediated decay (NMD) (Kerényi et al., 2008). However, Kerényi and colleagues only assessed the effect of the presence of an intron per se and not the effect of splicing of this intron. A recent study focused on understanding the impact of splicing on NMD in colon cancer found that intron splicing in the 3’ UTR of genes regulates transcript expression and could rescue a transcript from NMD (Riley et al., 2025). Our data similarly showed that the bulk (G_H‐LW) in which the 3’ UTR intron was spliced out had the higher transcript levels. Because this is the bulk with the largest LW that also carries the alleles for lower SG density, it may be that, as in the halophyte Limonium bicolor (Yuan et al., 2022), TTG1 negatively regulates SG development in zoysiagrass. The underlying cause for the differential splicing of the 3’ UTR intron is not clear. Both Meyer and PI 231146 carry consensus intron splice donor (GT) and accepter sequences (AG) for the 3’ UTR intron in Zj_TTG1. There are, however, variants in the intron, and it is possible that the branch point, which, in Arabidopsis, is located ∼19–40 nucleotides upstream from the 3’ end of an intron (Tolstrup et al., 1997), is mutated in PI 231146, the parent with the higher SG density. Because the branchpoint sequence is not well conserved in plants (consensus WWCTRAW in Arabidopsis; Tolstrup et al., 1997), it is difficult to ascertain whether any of the intron variants are located within the branch point. In PI 231146, we also identified an 11 bp insertion in the 3′ UTR immediately downstream of the intron region (Figure S7). Its effect, if any, is unknown.

5. CONCLUSION

In this study, we combined linkage mapping and BSR‐seq to investigate the genetic regulation of select leaf traits in an interspecific Z. japonica × Z. matrella F₂ population. The application of machine learning approaches for measuring LW at a constant location (widest point) and counting veins and SGs allowed for processing of larger sample numbers and detection of QTL of relatively small effects. A total of seven QTL were identified by CIM, two for SG density, three for LW, and two for vein count. ΔSNP index analysis from BSR‐seq data, after filtering, revealed four loci potentially associated with SG density and LW, with only the locus for LW on Chr08 being common with the CIM results. Mining of the common QTL region on Chr08 for candidate genes based on GO terms, position within the QTL, functional information from other species, differential expression between contrasting bulks for the phenotype, and structural variation between Meyer and PI 231146, identified SCL8 as the strongest candidate causal to the trait. For SG density, focusing on genes known to regulate trichomes, TTG1 emerged as the most likely causal gene to the Chr08 QTL. Interestingly, variation in transcript levels of TTG1 in the bulks and, by extension, Meyer and PI 231146 may be driven by differential splicing of an intron in the 3’ UTR.

These findings provide a foundation for future validation and functional characterization of the candidate genes identified. For SG density‐associated genes, an essential next step is to establish the extent to which allelic variants are associated with variation in salt tolerance. Phenotyping for salt tolerance requires specialized set‐ups in a greenhouse setting, and application of marker‐assisted selection for SG density, an important component of salt tolerance in zoysiagrass (Marcum et al., 1998; Yamamoto et al., 2016), would allow breeders to eliminate plants that lack the desired alleles at early growth stages under nonsaline conditions. Further, trait‐associated markers are integral to enhancing the predictive ability in genomic selection (Winn et al., 2022). This application extends to traits that, in isolation, can be easily assessed visually, such as LW, but for which trait‐associated markers will be invaluable to holistically select for genotypes with enhanced turf quality tailored to target growth environments and turf applications. The simultaneous stacking of stress‐resilient traits into elite, fine‐ or coarse‐textured backgrounds is needed to contend with water limitations leading to increased use of recycled water on US golf courses (Gelernter et al., 2015) and, potentially, in commercial and residential landscapes. Further, insights gained from our study not only advance Zoysia improvement but also have broader relevance for other Chloridoideae species, especially for the improvement of salt tolerance of related glycophytic food crops such as finger millet.

AUTHOR CONTRIBUTIONS

Shreena Pradhan: Conceptualization; formal analysis; methodology; writing—original draft. Thomas H. Pendergast IV: Formal analysis; validation. John J. Spiekerman: Conceptualization; methodology. John H. Lagergren: Methodology; software. Daniel A. Jacobson: Software; supervision. Katrien M. Devos: Conceptualization; funding acquisition; supervision; writing—review and editing.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Supplementary Material

Supplementary Material

Pradhan, S. , Pendergast IV, T. H. , Spiekerman, J. J. , Lagergren, J. H. , Jacobson, D. A. , & Devos, K. M. (2026). Genetic analyses of leaf traits in an interspecific Zoysia japonica × Zoysia matrella F₂ population. The Plant Genome, 19, e70282. 10.1002/tpg2.70282

DATA AVAILABILITY STATEMENT

Data supporting the findings of this work are presented within the main text, and as Supporting Information. The RNA‐seq reads for the bulks have been deposited in NCBI‐SRA (Project PRJNA1235172).