Long-term environmental risk assessment of glufosinate-tolerant GM zoysiagrass (Zoysia japonica Steud.) on non-target arthropod communities Authors: Ye-Jin Jang, Sung-Dug Oh, Dongmin Kim, Young-Kun Kim, Seong-Kon Lee, Ancheol Chang, Hyeon-Jin Sun, Hyo-Yeon Lee, Sang Jae Suh Published: 2026 ABSTRACT ABSTRACTEnvironmental risk assessment (ERA) of genetically modified (GM) crops is essential for evaluating their potential impact on non-target organisms under field conditions. In this study, a long-term field assessment was conducted to evaluate the ecological effects of glufosinate-tolerant GM zoysiagrass (Zoysia japonica Steud.; events JG21 and JG21-MS1) on non-target arthropod communities at two geographically distinct living modified organism isolation sites in South Korea (Jeonju and Seogwipo). A total of 528,796 arthropods were collected and classified into three functional groups: pests (45.73%), natural enemies (14.51%), and other arthropods (39.76%). Arthropod community composition exhibited clear regional and temporal variations. Linear mixed-effects model (LMM) analyses indicated that arthropod abundance and diversity indices were primarily influenced by the year and region, whereas treatment effects, defined as differences between GM events and non-GM controls, were not significant. Although occasional statistical differences were detected, post hoc comparisons confirmed that arthropod communities associated with GM treatments were ecologically comparable to those of non-GM controls within each region and year (p > .05). Non-metric multidimensional scaling (NMDS) ordination revealed clear clustering by region and year, with substantial overlap among treatments. Permutational multivariate analysis of variance showed that year explained the largest proportion of the variation (R2 = 0.640), whereas treatment effects were negligible (R2 = 0.006). Overall, these results provide robust field-based evidence that glufosinate-tolerant and male-sterile GM zoysiagrass does not adversely affect non-target arthropod communities, supporting its environmental safety in perennial agroecosystems.KEYWORDS: Environmental risk assessment, field study, genetically modified crops, herbicide tolerance, non-target arthropods, perennial ecosystem, Zoysia japonica Introduction The global adoption of genetically modified (GM) crops has transformed modern agriculture by improving productivity, reducing reliance on chemical inputs, and enhancing management efficiency.1,2 Despite these advantages, the environmental safety of GM crops remains under continuous scientific and regulatory scrutiny. A primary concern is the potential for unintended effects on non-target organisms (NTOs), which may influence ecosystem services, such as biological control and nutrient cycling.3,4 Accordingly, environmental risk assessment (ERA) frameworks must involve rigorous evaluation of ecological impacts under realistic field conditions.5,6 Non-target arthropods represent a major component of agroecosystems and include herbivores, predators, parasitoids, and decomposers that contribute to ecosystem functioning.7 Previous studies on insect-resistant and herbicide-tolerant crops have generally reported negligible or no significant adverse effects of GM crops on these communities.8,9 However, many of these studies have been limited in temporal and spatial scope. As arthropod communities vary substantially across environmental conditions, long-term, multi-location field studies are necessary to detect potential cumulative or context-dependent effects.10,11 Zoysiagrass (Zoysia japonica Steud.) is a dominant perennial turfgrass species widely used in East Asia for landscaping and creating playing surfaces in sports fields.12,13 Although it is valued for its durability and relatively low maintenance requirements, conventional management still relies on labor-intensive practices and repeated herbicide applications for weed control. The development of glufosinate-tolerant GM zoysiagrass, including events JG21 and JG21-MS1, offers a potential strategy to improve management efficiency. Unlike annual cropping systems, zoysiagrass represents a perennial ecosystem with a relatively stable vegetation structure, which may influence arthropod community dynamics. However, long-term field-based ecological risk assessments of GM zoysiagrass remain limited compared with those of major GM crops such as maize and soybean.12,13 In this study, we conducted a three-year (2023–2025) field-based environmental risk assessment of GM zoysiagrass at two geographically and climatically distinct locations in South Korea (Jeonju and Seogwipo). We evaluated the effects of two GM events (JG21 and JG21-MS1) and a non-GM control on the abundance, diversity, and community structure of non-target arthropods using standardized sampling and multivariate analyses. The objective of this study was to provide robust empirical evidence regarding the environmental safety of herbicide-tolerant GM zoysiagrass in perennial ecosystems. Materials and Methods Plant Materials Herbicide-tolerant genetically modified (GM) zoysiagrass events JG21 and JG21-MS1 were evaluated in this study.14,15 These events were developed through gamma irradiation of a previously established GM Zoysia line carrying the bar gene derived from Streptomyces hygroscopicus, which confers resistance to glufosinate-ammonium via phosphinothricin acetyltransferase activity.16 The non-GM parental line, Z. japonica Steud, was used as the near-isogenic control. Study Sites and Experimental Design Field experiments were conducted over three consecutive years (2023–2025) at two locations in South Korea: Jeonju and Seogwipo. The overall experimental design and key experimental factors are listed in Table 1. The study focused on two GM events (JG21 and JG21-MS1) and a non-GM control (Zoysia japonica), arranged in a randomized complete block design with three replicates per treatment at each site. Each experimental plot covered approximately 200 m2 and contained 400 plants arranged at a spacing of 15 cm × 30 cm. Table 1.Summary of the experimental design for the field assessment of GM zoysiagrass effects on non-target arthropods.FactorDescriptionLocationsJeonju and Seogwipo (South Korea)Study period2023–2025TreatmentsJG21 (GM), JG21-MS1 (GM), Zoysia japonica (non-GM)Experimental designRandomized complete block designReplicationThree replicates per treatment per site Summary of the experimental design for the field assessment of GM zoysiagrass effects on non-target arthropods. Arthropod Sampling Arthropods were sampled eight times annually at 2-week intervals from mid-May to late September. Yellow sticky traps (350 × 250 mm; Greenagrotech, South Korea) were used because they are widely used to monitor plant-associated arthropods in agroecosystems.17 One trap was installed per replicate at a height of 20–30 cm above the turfgrass canopy. Each trap remained in the field for 2 weeks and was collected at the end of each sampling interval. The collected traps were sealed in zipper bags, treated with ethyl acetate to immobilize the specimens, transported to the laboratory, and stored at −17°C until analysis. All specimens were examined and identified using a stereomicroscope. Functional Group Classification and Data Processing Collected arthropods were categorized into three functional groups: pests, natural enemies (including predators, parasitoids, and spiders), and other arthropods (e.g., decomposers and neutral species).18 When possible, pest taxa were identified to the species level, whereas natural enemies and other arthropods were generally identified to the family level. Abundance data were aggregated by taxon and functional groups for each replicate. A community matrix based on taxon-level abundance was constructed for diversity and multivariate analyses.19 Diversity Indices Arthropod diversity was quantified using taxon richness (S), the Shannon – Wiener diversity index (H’), Simpson’s diversity index (D), and Pielou’s evenness (J’) (Magurran, 2004). These indices were calculated from taxon-level abundance data using the vegan package in R (version 4.5.3).20 Statistical Analysis Abundance and Diversity Analysis Differences in arthropod abundance (total and functional groups) and diversity indices were analyzed using linear mixed-effects models (LMMs).21 The models included Treatment (two GM events and one non-GM control), Region, Year, and their interactions as fixed effects. Replicates nested within the Region and Year were incorporated as a random effect to account for the hierarchical structure of the experimental design (Table 1). Estimated marginal means (EMMs) and associated standard errors were calculated using the EMMs package. Post hoc multiple comparisons were performed using Tukey’s honestly significant difference (HSD) test, with a significance level set at p < .05.22 Community Structure Analysis Variations in arthropod community composition were evaluated using Bray – Curtis dissimilarity matrices based on log(x + 1)-transformed taxon abundance data.23 Permutational multivariate analysis of variance (PERMANOVA) was performed using the adonis2 function in the vegan package to assess the effects of Treatment, Region, Year, and their interactions.20 Homogeneity of multivariate dispersion was assessed using the betadisper function. Community structure was visualized using non-metric multidimensional scaling (NMDS), a widely used ordination method for ecological community analysis.17 The EMMs were calculated using the EMMs package.22 All statistical analyses were conducted in R. Results Arthropod Community Composition and Functional Structure The climatic conditions differed between Jeonju and Seogwipo during the study period (Supplementary Table S1), which may have contributed to the observed regional differences in arthropod community composition.24 A total of 528,796 arthropod individuals were collected over the 3-year study period (Supplementary Table S2). Pests, other arthropods, and natural enemies accounted for 45.73%, 39.76%, and 14.51%, respectively, of the total arthropods collected. Among all collected arthropods, Frankliniella occidentalis (17.17%) and Scirtothrips dorsalis (16.56%) were the most abundant pest species, followed by Aphis gossypii (4.06%) and Rhopalosiphum padi (3.06%). Natural enemies were primarily represented by hymenopteran parasitoids, including eulophid and ichneumonid parasitoids, as well as predatory coleopterans such as coccinellids and staphylinids. The “other arthropods” category predominantly included dipterans, particularly sciarids (23.88% of the total arthropods collected) (Supplementary Table S2). Distinct regional differences in functional and taxonomic compositions were observed (Figure 1). In Jeonju, the community was strongly dominated by pests (65.10%), followed by other arthropods (24.80%) and natural enemies (10.10%). Taxonomically, Thysanoptera (49.95%) and Hemiptera (14.03%) were the dominant pest orders, whereas Diptera (23.95%) comprised most of the “other arthropods” category (Figure 1(a)). Relative abundance (%) of arthropod functional groups and taxonomic composition across treatments in (a) Jeonju and (b) Seogwipo. The inner rings represent functional groups (pests, natural enemies, and other arthropods) and outer rings indicate taxonomic composition at the order level. By contrast, Seogwipo exhibited a community structure dominated by other arthropods (50.02%), followed by pests (32.43%) and natural enemies (17.55%). Diptera accounted for a substantial proportion of the total community (49.13%), while Thysanoptera (22.62%) and Hemiptera (7.74%) were the dominant pest groups. The relative abundance of natural enemy species was notably higher in Seogwipo than in Jeonju, with Hymenoptera (16.12%) representing the dominant group (Figure 1(b)). Arthropod Abundance Arthropod abundance was primarily influenced by temporal (year) and spatial (region) factors, whereas the treatment effects, defined as the differences between GM events (JG21 and JG21-MS1) and the non-GM control (Z. japonica), were generally negligible (Table 2). The mean abundance values for pests, natural enemies, other arthropods, and total arthropods across treatments, regions, and years are presented in Table 3. Subsequent post hoc comparisons confirmed that arthropod abundance did not differ significantly among treatments within each region and year (p > .05; Table 3). Table 2.Effects of treatment, region, and year on arthropod abundance based on generalized linear mixed models (GLMMs).Response variableEffectdfTest statistic(χ2 / F)p valueSignificancePest abundanceRegion12.630.105ns Year210.792.005** Treatment23.366.186ns Region × Year27.770.021* Region × Treatment20.864.649ns Year × Treatment44.737.315ns Region × Year × Treatment42.536.638nsNatural enemy abundanceRegion160.622.000*** Year224.118.000*** Treatment26.481.039* Region × Year244.730.000*** Region × Treatment29.512.009** Year × Treatment410.083.039* Region × Year × Treatment413.145.011*Other arthropodsRegion1198.958.000*** Year23.325.190ns Treatment20.158.924ns Region × Year229.264.000*** Region × Treatment20.645.724ns Year × Treatment42.401.662ns Region × Year × Treatment42.632.621nsTotal abundanceRegion158.456.000*** Year23.888.143ns Treatment20.221.896ns Region × Year214.006.001*** Region × Treatment20.805.669ns Year × Treatment42.401.662ns Region × Year × Treatment42.983.561ns− Abundance data were analyzed using GLMMs with a negative binomial distribution.- Significance levels: p < .05 *, p < .01 **, p < .001 ***, ns = not significant. Effects of treatment, region, and year on arthropod abundance based on generalized linear mixed models (GLMMs). − Abundance data were analyzed using GLMMs with a negative binomial distribution. - Significance levels: p < .05 *, p < .01 **, p < .001 ***, ns = not significant. Table 3.Mean abundance (± SE) of arthropod functional groups per trap across treatments, regions, and years.RegionYearTreatment*Pests(mean ± SE)Natural enemies(mean ± SE)Other arthropods(mean ± SE)Total arthropods(mean ± SE)Jeonju2023Z. japonica3,870.0 ± 390.2752.7 ± 30.71,275.7 ± 64.35,898.3 ± 351.8 JG213,959.3 ± 230.9789.0 ± 125.01,371.3 ± 100.56,119.7 ± 302.7 JG21-MS13,681.3 ± 189.1654.7 ± 41.21,242.0 ± 71.45,578.0 ± 103.8 2024Z. japonica4,005.7 ± 167.8584.0 ± 55.62,244.7 ± 373.06,834.3 ± 581.8 JG214,514.7 ± 226.8628.3 ± 64.22,709.3 ± 208.67,852.3 ± 193.1 JG21-MS14,176.7 ± 299.3535.0 ± 14.22,309.0 ± 345.37,020.7 ± 630.8 2025Z. japonica7,540.0 ± 497.51,036.7 ± 57.02,155.7 ± 148.110,732.3 ± 534.3 JG217,377.3 ± 1,087.91,077.7 ± 52.62,248.7 ± 178.710,703.7 ± 1,305.3 JG21-MS17,567.0 ± 647.31,186.0 ± 142.42,233.7 ± 335.810,986.7 ± 1,099.4Seogwipo2023Z. japonica3,630.0 ± 359.11,658.3 ± 259.06,120.0 ± 638.311,408.3 ± 987.5 JG213,412.3 ± 148.61,718.7 ± 114.25,914.7 ± 272.711,045.7 ± 333.1 JG21-MS13,073.0 ± 430.12,093.7 ± 91.45,901.3 ± 178.211,068.0 ± 528.8 2024Z. japonica3,592.3 ± 287.72,509.3 ± 282.15,350.3 ± 564.611,452.0 ± 1,037.2 JG213,771.3 ± 168.52,709.7 ± 75.65,301.7 ± 491.011,782.7 ± 568.2 JG21-MS13,302.3 ± 133.52,368.3 ± 57.25,616.7 ± 212.211,287.3 ± 189.3 2025Z. japonica4,063.0 ± 315.41,797.3 ± 181.75,477.0 ± 507.511,337.3 ± 974.7 JG214,586.7 ± 284.91,876.3 ± 136.76,388.0 ± 119.912,851.0 ± 203.4 JG21-MS14,476.7 ± 52.31,610.0 ± 75.06,220.3 ± 118.812,307.0 ± 71.4- Values are presented as mean ± standard error (SE).- No significant differences among treatments were detected within each region and year (p > .05).- JG21 and JG21-MS1 represent genetically modified (GM) events, whereas Zoysia japonica represents the non-GM control. Mean abundance (± SE) of arthropod functional groups per trap across treatments, regions, and years. - Values are presented as mean ± standard error (SE). - No significant differences among treatments were detected within each region and year (p > .05). - JG21 and JG21-MS1 represent genetically modified (GM) events, whereas Zoysia japonica represents the non-GM control. Pest abundance varied significantly with year (χ2 = 10.792, p = .005) and the region × year interaction (χ2 = 7.770, p = .021), showing a consistent increase from 2023 to 2025 across both regions (Table 2). No significant treatment effects were observed for pest abundance (p = .186). Natural enemy abundance was significantly affected by region (χ2 = 60.622, p < .001), year (χ2 = 24.118, p < .001), and treatment (χ2 = 6.481, p = .039) (Table 2). However, despite the statistical significance of the treatment factors, pairwise comparisons revealed no consistent differences between the GM treatments (JG21 and JG21-MS1) and the non-GM control (Zoysia japonica) within individual regions or years (p > .05; Table 3), suggesting that the observed effect was not biologically meaningful. The abundance of other arthropods and total arthropods was strongly influenced by region (p < .001) and the region × year interaction (p = .001), with no significant treatment effects detected (Table 2). Overall, Seogwipo consistently exhibited a higher total arthropod abundance than Jeonju, largely due to greater contributions from natural enemies and other arthropods (Table 3). Diversity Indices All diversity indices were significantly influenced by year, whereas no treatment effects were detected for any metrics (Table 4). The mean values for taxon richness, Shannon diversity, Simpson diversity, and Pielou’s evenness across treatments, regions, and years are presented in Table 5. Post hoc comparisons confirmed that the diversity indices did not differ significantly among treatments within each region and year (p > .05; Table 5). Table 4.Effects of treatment, region, and year on arthropod diversity indices based on linear mixed models (LMMs).Response variableEffectdfTest statistic(χ2/F)p-valueSignificanceRichnessRegion11.855.182ns Year2115.740.000*** Treatment20.107.899ns Region × Year21.244.300ns Region × Treatment20.237.790ns Year × Treatment40.242.913ns Region × Year × Treatment40.092.985nsShannon diversityRegion10.056.814ns Year259.316.000*** Treatment20.150.861ns Region × Year20.402.672ns Region × Treatment20.658.524ns Year × Treatment40.361.835ns Region × Year × Treatment40.783.544nsSimpson diversityRegion10.079.781ns Year217.936.000*** Treatment20.130.878ns Region × Year23.332.047* Region × Treatment20.176.839ns Year × Treatment40.365.832ns Region × Year × Treatment40.454.768nsPielou evennessRegion11.530.224ns Year223.112.000*** Treatment20.157.855ns Region × Year20.922.407ns Region × Treatment20.398.675ns Year × Treatment40.409.801ns Region × Year × Treatment40.858.498ns− Diversity indices were analyzed using LMMs.- Significance levels: p < .05 *, p < .01 **, p < .001 ***, ns = not significant. Effects of treatment, region, and year on arthropod diversity indices based on linear mixed models (LMMs). − Diversity indices were analyzed using LMMs. - Significance levels: p < .05 *, p < .01 **, p < .001 ***, ns = not significant. Table 5.Arthropod richness and diversity indices (mean ± SE) across treatments, regions, and years.RegionYearTreatment*Taxon richness (S)(mean ± SE)Shannon diversity (H′)(mean ± SE)Simpson diversity (1−D)(mean ± SE)Pielou’s evenness (J′)(mean ± SE)Jeonju2023Z. japonica30.33 ± 1.201.661 ± 0.0610.719 ± 0.0190.487 ± 0.016 JG2131.00 ± 0.581.675 ± 0.0600.729 ± 0.0140.488 ± 0.017 JG21-MS129.00 ± 1.001.607 ± 0.0390.722 ± 0.0140.478 ± 0.016 2024Z. japonica55.67 ± 1.862.023 ± 0.0690.730 ± 0.0200.503 ± 0.014 JG2159.67 ± 3.181.974 ± 0.0420.718 ± 0.0150.483 ± 0.004 JG21-MS154.33 ± 1.202.015 ± 0.0290.731 ± 0.0040.505 ± 0.008 2025Z. japonica62.67 ± 0.672.253 ± 0.0540.817 ± 0.0130.545 ± 0.013 JG2163.00 ± 0.582.292 ± 0.0570.809 ± 0.0110.553 ± 0.014 JG21-MS161.00 ± 2.082.271 ± 0.0200.819 ± 0.0070.553 ± 0.003Seogwipo2023Z. japonica27.00 ± 0.581.677 ± 0.0170.733 ± 0.0070.509 ± 0.008 JG2128.00 ± 0.001.690 ± 0.0070.733 ± 0.0070.507 ± 0.002 JG21-MS127.67 ± 0.671.711 ± 0.0400.740 ± 0.0130.515 ± 0.013 2024Z. japonica55.67 ± 3.381.961 ± 0.0630.780 ± 0.0100.488 ± 0.013 JG2157.67 ± 0.331.975 ± 0.0430.782 ± 0.0110.487 ± 0.011 JG21-MS155.67 ± 2.601.934 ± 0.0370.764 ± 0.0070.481 ± 0.003 2025Z. japonica56.33 ± 0.882.380 ± 0.0410.837 ± 0.0100.591 ± 0.011 JG2156.33 ± 0.882.368 ± 0.0300.833 ± 0.0060.587 ± 0.006 JG21-MS157.33 ± 0.332.316 ± 0.0290.828 ± 0.0080.572 ± 0.008-Values are presented as mean ± standard error (SE).- No significant differences among treatments were detected within each region and year (p > .05).- JG21 and JG21-MS1 represent genetically modified (GM) events, whereas Zoysia japonica represents the non-GM control. Arthropod richness and diversity indices (mean ± SE) across treatments, regions, and years. -Values are presented as mean ± standard error (SE). - No significant differences among treatments were detected within each region and year (p > .05). - JG21 and JG21-MS1 represent genetically modified (GM) events, whereas Zoysia japonica represents the non-GM control. Taxon richness (S) increased markedly over time (F = 115.740, p < .001), from approximately 27 to 31 taxa in 2023 to over 60 taxa in 2025. Similarly, Shannon diversity (F = 59.316, p < .001), Simpson diversity (F = 17.936, p < .001), and Pielou’s evenness (F = 23.112, p < .001) increased significantly over time (Table 2). No significant differences were observed among treatments for any diversity metric (p > .05; Table 2). A significant Region × Year interaction was detected for Simpson’s diversity index (F = 3.332, p = .047), indicating that temporal patterns of arthropod diversity varied between the regions. However, no consistent treatment-related differences were observed within individual regions and years. Community Structure Analysis The multivariate analyses indicated that region and year were the dominant factors structuring arthropod communities. Non-metric multidimensional scaling (NMDS) ordination based on Bray – Curtis dissimilarity revealed clear separation between Jeonju and Seogwipo samples, with distinct regional clustering (Figure 2). This separation was supported by non-overlapping confidence ellipses, indicating consistent differences in community composition between the regions. Non-metric multidimensional scaling (NMDS) ordination based on Bray–Curtis dissimilarity of arthropod communities. Points are colored by year (2023–2025) and shaped by treatment (JG21, JG21-MS1, and *zoysia japonica*). Ellipses represent grouping by region (Jeonju and Seogwipo). Stress = 0.066. Within each regional cluster, samples exhibited directional shifts across years, reflecting temporal changes in the community structure during the study period. These temporal trajectories suggest that interannual environmental variability, including climatic conditions and habitat dynamics, plays a key role in shaping arthropod assemblages. By contrast, no clear separation was observed among treatments. Data points representing the GM treatments (JG21 and JG21-MS1) and non-GM control (Z. japonica) largely overlapped within each region and year (Figure 2), indicating that the introduction of herbicide-tolerant traits did not alter the overall community composition. The PERMANOVA results showed that year explained the largest proportion of the total variation (R2 = 0.640, p = .001), followed by region (R2 = 0.278, p = .001). Although treatment effects were statistically significant (R2 = 0.006, p = .007), their explanatory power was minimal (<1%), indicating negligible ecological relevance. The PERMDISP analysis detected significant differences in dispersion only for year (p < .001), but not for region (p = .192) or treatment (p = .979). These results suggest that variation in community composition was primarily driven by temporal differences rather than differences in within-group dispersion. Overall, these findings indicate that the arthropod community structure was strongly influenced by environmental and temporal factors, whereas the effects of GM zoysiagrass were minimal. This pattern is consistent with previous reports of limited effects of GM crops on non-target arthropod communities.6,10 Discussion In this study, we evaluated the ecological effects of glufosinate-tolerant GM zoysiagrass events (JG21 and JG21-MS1) on non-target arthropod communities through a 3-year field experiment conducted at two locations. Across all analyses, including abundance (Table 2), diversity indices (Table 4), and multivariate community structure (Table 6; Figure 2), arthropod assemblages associated with GM treatments were ecologically comparable to those observed in the non-GM control. Unlike previous environmental risk assessment studies of GM zoysiagrass that were conducted over shorter periods or under more limited environmental conditions, the present study integrated multi-year and multi-location field monitoring data encompassing more than 500,000 arthropod individuals. This large-scale assessment provides a more comprehensive evaluation of long-term ecological effects in a perennial turfgrass system and contributes valuable evidence for environmental risk assessment of GM turfgrass. Table 6.Effects of treatment, region, and year on arthropod community structure based on PERMANOVA and PERMDISP analyses.(A) PERMANOVAFactordfSum of squaresR2Pseudo-F / Fp-valueSignificanceRegion11.5110.278347.0880.001**Year23.4790.640399.5890.001**Treatment20.0310.0063.5900.007**Region × Year20.1800.03320.6860.001**Region × Treatment20.0320.0063.6200.008**Year × Treatment40.0200.0041.1340.346nsRegion × Year × Treatment40.0250.0051.4150.155nsResidual360.1567320.02884–––(B) PERMDISPFactordfSum SqMean SqF valuep-valueSignificanceRegion10.0230.0231.7470.192nsYear20.0250.01222.4890.000***Treatment20.0000.0000.0210.979ns− Significance levels as above.- Based on Bray – Curtis dissimilarity.- Significance levels: p < .05 (*), p < .01 (**), p < .001 (***), ns = not significant. Effects of treatment, region, and year on arthropod community structure based on PERMANOVA and PERMDISP analyses. - Based on Bray – Curtis dissimilarity. - Significance levels: p < .05 (*), p < .01 (**), p < .001 (***), ns = not significant. The absence of biologically meaningful differences across functional groups between glufosinate-tolerant GM zoysiagrass (JG21 and JG21-MS1) and the non-GM control (Zoysia japonica) is consistent with the findings of previous field-based studies, showing negligible or no significant effects of herbicide-tolerant (HT) crops on non-target arthropod communities.4,7,10 Although a statistically significant treatment effect was detected for the natural enemy abundance (Table 2) and PERMANOVA results (Table 6), their explanatory power was extremely low (R2 = 0.006). This finding suggests that statistical significance did not correspond to ecologically meaningful change and likely reflected natural field variability rather than transgene effects. Moreover, the observed differences were not consistently detected across years, regions, or analytical approaches, further supporting the conclusion that these statistically significant results do not represent biologically relevant ecological impacts. Such patterns, in which statistically detectable differences are ecologically negligible, are well documented in environmental risk assessment studies of GM crops.6,11 A key finding of this study is that arthropod community structure was primarily influenced by temporal and regional factors. Both NMDS ordination (Figure 2) and PERMANOVA (Table 6) identified year (R2 = 0.640) and region (R2 = 0.278) as the primary drivers of community variation, whereas treatment effects, representing differences between GM events (JG21 and JG21-MS1) and non-GM controls, were negligible. The pronounced temporal shifts observed from 2023 to 2025 likely reflect ecological succession following turfgrass establishment. As a perennial system, zoysiagrass provides relatively stable and continuous habitat conditions, facilitating gradual community assembly and stabilization over time, unlike the recurrent disturbances characteristic of annual cropping systems.12 Similarly, the significant Region × Year interaction observed for Simpson’s diversity index suggests that temporal patterns of arthropod diversity differed between regions. This finding indicates that local environmental conditions and annual variations may influence diversity dynamics, highlighting the importance of considering spatial and temporal heterogeneity when interpreting long-term field assessment data. Nevertheless, no consistent treatment-related effects were detected within individual regions and years, indicating that the observed interaction was not associated with the GM trait. The clear separation between Jeonju and Seogwipo observed in the NMDS ordination (Figure 2) highlights the importance of environmental context in shaping arthropod communities. The warmer, more humid maritime climate of Seogwipo supported a more balanced functional structure characterized by a higher proportion of natural enemies and dipteran taxa, whereas Jeonju was dominated by pest taxa (Figure 1). Such geographical variation is widely recognized as a major determinant of arthropod community assembly in agroecosystems,7,18 and our findings further emphasize the importance of incorporating environmental heterogeneity into ERA frameworks. The present study relied on yellow sticky traps, a widely used tool to monitor flying and plant-associated arthropods in field ecosystems. Although these traps effectively capture dominant aerial taxa and facilitate standardized comparisons among treatments, they may underrepresent ground-dwelling arthropods because sampling efficiency varies among arthropod groups and trapping methods.25 Therefore, the present study results should be interpreted primarily in the context of the arthropod taxa most effectively sampled using yellow sticky traps. Nevertheless, because these traps were used across all treatments and sites throughout the study period, the comparative assessment of GM and non-GM zoysiagrass remains robust and appropriate for environmental risk assessment purposes. The inclusion of the male-sterile event JG21-MS1 is particularly important from a biosafety perspective because this trait was developed to mitigate pollen-mediated gene flow, a major regulatory concern for GM turfgrass. Our results demonstrated that this biological containment strategy did not exert additional ecological effects on non-target arthropod communities. Furthermore, the consistent dominance of key taxa, including F. occidentalis, S. dorsalis, and dipteran groups, such as sciarids, across all treatments (Supplementary Table S2) suggests that baseline arthropod assemblages in zoysiagrass ecosystems are primarily shaped by seasonal and regional environmental factors rather than host plant genetic background. In conclusion, this long-term field study provides strong empirical evidence that glufosinate-tolerant GM zoysiagrass is ecologically comparable to its non-GM counterpart. Variations in arthropod community composition were driven predominantly by temporal succession and regional environmental conditions, with negligible contributions from GM treatments. These findings support the environmental safety of HT GM turfgrass and provide a strong scientific basis for regulatory risk assessments and sustainable management of perennial GM cropping systems. Supplementary Material Supplemental Material Acknowledgments We are grateful to the Cooperative Research Program for Agriculture Science & Technology Development (Project No. PJ01672602) and Rural Development Administration, Republic of Korea, for supporting this study. This study was supported by 2026 the RDA Fellowship Program of National Institute of Agricultural Science, Rural Development Administration, Republic of Korea. Funding Statement The author(s) reported there is no funding associated with the work featured in this article. Disclosure Statement No potential conflict of interest was reported by the author(s). Supplementary Material Supplemental data for this article can be accessed online at https://doi.org/10.1080/21645698.2026.2704431. References Associated Data Supplementary Materials Supplemental Material