A GA3-loaded hydrophilic–lipophilic diblock polymer acts as a nano-safener to alleviate herbicide-induced injury in rice by activating key metabolic pathways Authors: Xile Deng, Heng Qiao, Yijie Zhan, Jie Shen, Lianyang Bai, Shuo Yan Published: 2026 ABSTRACT AbstractAlthough herbicides are among the most effective tools for weed control in rice cultivation, they often cause phytotoxicity. Herbicide safeners can mitigate such damage; however, commercial safeners may have adverse effects on non-target organisms. Natural safeners such as gibberellin A3 (GA3) are considered safer alternatives, but new strategies are needed to improve the utilization efficiency of GA3 because its poor solubility and limited potency restrict its practical application. Here, taking advantage of the self-assembly capacity of a hydrophilic-lipophilic diblock polymer (HLDP) to formulate poorly soluble natural products, GA3 was assembled with HLDP to generate GA3@HLDP. The results demonstrated that GA3@HLDP was successfully constructed through hydrogen bonding and hydrophobic association. Its nanoscale particle size significantly improved permeability and safening activity, thereby alleviating pendimethalin (PND)-induced phytotoxicity in rice seedlings without affecting the herbicidal activity of PND. Compared with GA3, GA3@HLDP more effectively activated phenylpropanoid biosynthesis, glutathione metabolism, and starch/sucrose metabolism pathways, while also restoring the activities of multiple antioxidant enzymes. Furthermore, GA3@HLDP displayed favorable biosafety in various model organisms, and its application had negligible effects on soil microorganisms. This work shows that HLDP can comprehensively improve the performance of GA3 through multifaceted synergistic mechanisms, providing a feasible strategy to bridge the gap between the theoretical development and practical application of natural safeners.Key words: crop protection, gibberellin A3, nanocarrier, nano-safener, permeability Introduction Weeds pose major biotic threats to global agriculture, causing approximately 10% annual crop yield losses worldwide and estimated economic losses exceeding US$95 billion annually over the past decade (Guo et al., 2017; Li et al., 2023). Rice (Oryza sativa) is one of the world’s most important staple crops, sustaining more than half of the global population through its major contribution to caloric intake and essential nutrition (Quan et al., 2025). However, weeds severely reduce rice production by competing for water, space, sunlight, and soil nutrients, causing yield losses of 35% to 91% and reducing grain quality (Heap, 2014; Li et al., 2024; Patidar and Kaur, 2024). Herbicides are currently widely used in rice production because of their broad-spectrum effectiveness against annual grasses and broadleaf weeds (Powles and Yu, 2010). However, improper herbicide application can cause two major problems: unintended phytotoxicity in rice seedlings and the emergence of herbicide-resistant weeds, both of which increase yield losses and the economic burden on farmers (Qiao et al., 2024; Xie et al., 2024). During early growth stages, rice is difficult to distinguish from barnyard grass (Echinochloa crus-galli), making rice seedlings susceptible to herbicide injury (Wei et al., 2011). Herbicide damage disrupts key physiological processes in rice plants, resulting in stunted growth, impaired root development, substantial yield reduction, and, in severe cases, complete crop failure (Dass et al., 2017; Lu et al., 2025). Pendimethalin (PND) is a widely used dinitroaniline pre-emergence herbicide in many countries. It acts by inhibiting chromosome separation and cell-wall formation, thereby disrupting plant cell growth and division (Wang et al., 2025a, 2025b). PND is extensively applied for the control of annual weeds, especially barnyard grass, in paddy fields (Chen et al., 2025a, 2025b). However, PND application often causes severe phytotoxicity in rice seedlings, reducing the productive panicle rate by up to 34.7% and increasing yield loss by 52.0% (Ahmed and Chauhan, 2015; Becerra-Alvarez and Al-Khatib, 2025). Currently, no PND-resistant rice cultivars have been bred or engineered because of the complex mode of action of PND, whose effects are not easily attributable to a single molecular target (Wang et al., 2025a, 2025b). Although controlled-release technology can partially mitigate PND phytotoxicity, sustained release and insufficient root-zone targeting prevent complete elimination of crop injury (Fan et al., 2025). Therefore, practical strategies are needed to mitigate herbicide-induced injury while maintaining weed control efficacy. Herbicide safeners are agrochemicals that mitigate herbicide-induced crop injury while maintaining herbicidal efficacy for weed control (Zhao et al., 2024). Although hundreds of candidate safeners have been identified in initial screens, only 23 safeners have reached the market to date, including fenclorim, benoxacor, and AD-67 (Deng et al., 2026a). Safeners enhance crop tolerance to herbicides through multiple mechanisms, including metabolic detoxification and activation of antioxidant defense responses (Kraehmer et al., 2014). However, no commercial safeners have been reported for co-application with PND. Moreover, acute and chronic exposure studies indicate that existing commercial safeners can adversely affect soil, aquatic, and mammalian systems (Liu et al., 2021a, 2021b; Oloye et al., 2021; Lanasa et al., 2022; Simonsen et al., 2024). Therefore, PND-specific next-generation safeners with low toxicity, strong biological activity, and high environmental safety are urgently needed for sustainable weed management in rice cultivation. Natural products have emerged as promising safeners that offer a sustainable alternative to synthetic safeners. As phytohormones or signaling molecules, they generally show low mammalian toxicity and minimal environmental impact. Their unique structures and protective activities distinguish them from commercial safeners and offer new opportunities for eco-friendly crop protection (Leng et al., 2023). Gibberellin A3 (GA3), an endogenous plant hormone, regulates growth processes including α-amylase synthesis, cell elongation/division, and seed germination and emergence (Chen et al., 2024). GA3 functions as a safener that protects maize (Zea mays), sorghum (Sorghum bicolor), and rice from chloroacetamide herbicide-induced phytotoxicity (Nemat Alla and Hassan, 1998; Zhang et al., 2022; Wang et al., 2024a, 2024b). However, the protective efficacy of GA3 against PND-induced phytotoxicity in rice seedlings remains unexplored, and the underlying protective mechanisms are unknown. In addition, the poor aqueous solubility of GA3 complicates formulation and reduces its bioavailability under uneven field application. Nanocarriers have a large specific surface area and small size, giving them high affinity for target crops (Xia et al., 2025; Deng et al., 2026b). Numerous studies have shown that nano-agrochemical formulations improve solubility, wettability, and dispersion properties. Encapsulation of agrochemical molecules within nanocarriers can also protect them from physicochemical degradation, especially during prolonged exposure under field conditions (Wang et al., 2022; Deng et al., 2025a). Thus, nano-enabled safeners are expected to become powerful tools for modern weed management and crop protection (Lowry et al., 2024). In this study, we aimed to use a hydrophilic–lipophilic diblock polymer (HLDP) nanocarrier to develop an efficient nano-safener, GA3@HLDP, to mitigate PND-induced injury in rice seedlings. First, the loading efficiency of HLDP toward GA3 was examined using a freeze-drying method, and the self-assembly mechanism of GA3@HLDP was analyzed by isothermal titration calorimetry (ITC), molecular docking, and molecular dynamics (MD) simulations. The physicochemical properties of GA3@HLDP were then characterized using scanning electron microscopy (SEM), transmission electron microscopy (TEM), dynamic light scattering (DLS), Fourier transform infrared (FTIR) spectroscopy, and thermogravimetric analysis (TGA). Next, plant uptake and soil mobility of GA3@HLDP were investigated by high-performance liquid chromatography (HPLC), and the translocation mechanism of GA3@HLDP across the cell membrane was explored using MD simulations and hyperspectral imaging. Importantly, the safening activity of GA3@HLDP and the herbicidal activity of PND emulsifiable concentrate (PND EC) co-applied with GA3@HLDP were evaluated. The potential synergistic mechanism was investigated through integrated transcriptomic and metabolomic analyses and further verified by RT–qPCR, flow cytometry, and physiological and biochemical assays. Finally, the safety risk assessment of GA3@HLDP was performed using zebrafish (Danio rerio) embryos, fruit flies (Drosophila melanogaster), human skin fibroblast (HSF) cells, and the soil microbial community. Overall, our study developed an eco-friendly nano-safener with multiple synergistic mechanisms, providing an innovative strategy for designing efficient nano-safeners to mitigate the adverse effects of herbicides. Results and discussion Moderate GA3-loading capacity of HLDP HLDP was synthesized through a simple polymerization process, as shown in Figure 1A. Because the molecular weight of HLDP was much higher than that of GA3, unassembled GA3 could be removed by dialysis after excess GA3 was incubated with HLDP. The loading efficiency was calculated to be 22.11%, which was comparable to reported values for other compounds (Supplemental Table 1). For example, the loading efficiencies of HLDP toward metamifop, salicylic acid, and chitosan were reported to be 21.71%, 48.49%, and 51.21%, respectively (Yin et al., 2024; Deng et al., 2025b). The variation in drug loading content (DLC) among different molecules may result from differences in their chemical structures and interaction modes with HLDP. Self-assembly mechanism of GA3@HLDP. (A) Schematic illustration of the HLDP synthesis route. (B) Self-assembly of GA3 and HLDP into a stable nanocomplex driven by hydrogen bonding and hydrophobic association. (C) ITC titration of GA3 solution (1 mmol/l) into HLDP solution (0.1 mmol/l). The ITC raw data show heat release during the titration of GA3 into HLDP. (D) Gibbs free energy (ΔG), enthalpy change (ΔH), and entropy contribution (ΔS) during GA3@HLDP assembly. (E) The 100-ns interaction between GA3 and HLDP. (F) Time evolution of the binding process between GA3 and HLDP obtained from MD simulations. The system was modeled as a core-shell structure, with GA3 molecules (blue) encapsulated within HLDP (purple). Both GA3 and HLDP are represented using the stick model. Representative snapshots from the simulation trajectory at 0, 20, 40, 60, 80, and 100 ns illustrate the conformational dynamics over time. Upper row, surface mode; lower row, cartoon mode. Self-assembly mechanism of GA3@HLDP through hydrogen bonding and hydrophobic association The self-assembly of GA3 with HLDP was investigated using ITC, a widely used method for quantifying molecular interactions. As shown in Supplemental Figures 1C and 1D, the measured binding affinity was high, with a Ka value of 6.66E+05 M−1 and a correspondingly low dissociation constant Kd of 1.50E−06 M, confirming a strong and effective interaction. The negative ΔG value (−33.24 kJ/mol) indicated that the assembly process occurred spontaneously. The positive values of both ΔH (4,698 kJ/mol) and ΔS (15,869 J/mol⋅K) were consistent with an entropy-driven assembly process, indicating that hydrophobic association was the dominant mechanism underlying the interaction between HLDP and GA3. In general, polymers can interact with pesticides through various mechanisms, including electrostatic interactions, hydrogen bonding, and van der Waals forces. However, the predominant interaction in a given system largely depends on the chemical structures of the components, particularly the functional groups present (Bou-Rabee, 2014). Such diverse interactions with pesticides may expand the practical application range of HLDP in agricultural production. Molecular docking was then used to predict the binding mode of GA3 with HLDP (Figure 1B). The simulation revealed hydrogen-bonding interactions, with bond lengths of 2.5, 3.2, 3.3, and 3.4 Å, between the structural unit of HLDP and the hydrogen or oxygen atoms of GA3. The binding free energy was −21.35 kJ/mol. These non-covalent interactions contributed to the formation of stable GA3@HLDP, further indicating that hydrogen bonding was also a major stabilization mechanism. To comprehensively evaluate the binding stability and interaction characteristics of GA3@HLDP, all-atom MD simulations were performed for 100 ns to capture the conformational dynamics of GA3@HLDP under near-physiological conditions. Key parameters, including root-mean-square deviation (RMSD), radius of gyration (Rg), solvent-accessible surface area (SASA), and the number of intermolecular hydrogen bonds, were monitored to systematically assess structural integrity, flexibility, and interaction strength. These analyses further clarified the self-assembly mechanism of GA3@HLDP. RMSD is a primary indicator of the stability of a simulated system (Brüschweiler, 2003), with lower values generally corresponding to greater structural rigidity (He et al., 2023). As shown in Supplemental Figure 1A, the RMSD value of GA3@HLDP increased rapidly during the initial 10 ns, reaching approximately 4.3 nm before stabilizing. This pattern suggested an initial rapid assembly phase followed by the formation of a stable conformational state with minimal subsequent fluctuations, demonstrating substantial structural robustness. Rg is a key metric of structural compactness (Miu et al., 2008), with lower Rg values typically indicating a more tightly packed and stable molecular conformation (Davoudmanesh and Mosaabadi, 2018). As shown in Supplemental Figure 1B, the Rg of GA3@HLDP initially declined and then plateaued at approximately 2.8 nm after 40 ns. This decrease suggests increased compaction of the complex, likely owing to interfacial reorganization or strengthened internal cohesion. SASA quantifies the molecular surface area exposed to solvent and provides insights into structural compactness and the accessibility of hydrophilic or hydrophobic regions (Richmond, 1984). A lower SASA value often indicates a denser assembly with reduced solvent-exposed surface area, which may correlate with enhanced resistance to thermal denaturation (Castrosanto et al., 2023). Our data showed an initial decrease in SASA that stabilized after approximately 10 ns (Supplemental Figure 1C), consistent with the increased compactness observed in the Rg analysis. Hydrogen bonds are key non-covalent interactions that mediate atomic-level interactions with solvent molecules and facilitate specific ligand binding at active sites (Nazar et al., 2020; Zhou et al., 2023). Therefore, we quantified intermolecular hydrogen bonds to evaluate the stability of GA3@HLDP. As shown in Supplemental Figure 1D, the number of hydrogen bonds between HLDP and GA3 increased from approximately 10 at the start of the simulation to a stable range of 20 to 40. These results indicated the establishment of a stable polar interaction network, supporting the structural complementarity and consistent interaction pattern observed during the later phases of the simulation. Collectively, the trends in RMSD, Rg, and SASA indicated a convergent conformational trajectory for the self-assembled GA3@HLDP complex during the MD simulation, characterized by compaction and interfacial rearrangement. To visualize these dynamic structural changes and relative interchain positioning, representative 3D structures were extracted and rendered at 0, 20, 40, 60, 80, and 100 ns (Supplemental Figures 1E and 1F). In the surface-rendered snapshots, the molecules were initially randomly dispersed in the aqueous environment at 0 ns. Over time, the buried surface area increased, and the complex showed a pronounced compaction trend. These observations were consistent with the contraction behavior identified in the Rg and SASA analyses, indicating optimized interfacial reorganization that reinforced intermolecular interactions and led to the formation of a stable nanoparticle system. The simulation results at 100 ns were also consistent with the morphology observed by electron microscopy. Analysis of interaction energy over time showed that GA3@HLDP exhibited thermodynamically favorable binding, with a mean total interaction energy of −5299.20 kJ/mol (Supplemental Figure 1E). The dominant contributions of electrostatic (−1386.65 kJ/mol) and van der Waals interactions (−3912.54 kJ/mol) suggested strong charge complementarity and spatial adaptation at the interface, consistent with the ITC data. A gradual decrease in total, Coulombic, and van der Waals interaction energies over time further indicated the spontaneous tendency of GA3@HLDP to assemble into a stable polymeric nanoparticle system under physiological conditions. Reduced particle size and enhanced stability of GA3@HLDP As shown in Figure 2A, SEM and TEM images revealed that GA3@HLDP consisted of ellipsoidal or spherical particles with smooth surfaces and uniform size, in contrast to the irregular, block-shaped crystals of GA3. Notably, when the mass ratio of GA3 to HLDP decreased from 1:1 to 1:3.548, the particle size distribution became narrower, indicating that the amount of HLDP played a critical role in regulating the final morphology and uniformity of the nanoparticles. Furthermore, energy-dispersive spectroscopy (EDS) analysis demonstrated the uniform distribution of carbon (C), nitrogen (N), oxygen (O), and the characteristic element bromine (Br) within the particles, confirming the successful preparation of the GA3@HLDP nanocomposite (Figure 2B; Supplemental Figure 2). FTIR analysis provided further evidence for the molecular interactions between GA3 and HLDP (Figure 2C). The GA3@HLDP spectrum showed significant changes in the functional-group vibration region of 3500–3300 cm−1. The doublet of O–H stretching vibrations (3447 and 3373 cm−1) present in GA3 and the broad N-H/O-H absorption peak (3419 cm−1) of HLDP were completely merged in the GA3@HLDP spectrum, forming a stronger, broader, and red-shifted single peak. These results indicated strong intermolecular hydrogen-bonding interactions between the hydroxyl groups in GA3 molecules and the amino groups in HLDP. DLS showed that the average particle size of GA3 was 2079.85 nm, whereas that of GA3@HLDP was significantly reduced to approximately 120 nm (Figure 2D; Supplemental Table 2). Previous studies have shown that some polymers can be applied as pesticide adjuvants for nanonization (Zhu et al., 2023). For instance, complexation with HLDP can disrupt the self-aggregated structures of osthole and thiophanate-methyl and reduce their particle sizes from 289 to 18 nm and from 834 to 86 nm, respectively (Yan et al., 2021; Su et al., 2023). Additionally, the polydispersity index of all composite samples remained below 0.3, indicating excellent monodispersity of the system. Characterization of GA3@HLDP. (A) Comparative morphology of free GA3 crystals and GA3@HLDP nanoparticles at different mass ratios (1:1 and 1:3.548) observed by SEM and TEM. (B) EDS elemental mapping of GA3@HLDP nanoparticles, confirming the uniform distribution of C, N, O, and the characteristic Br element. (C) FTIR spectra of GA3, HLDP, and GA3@HLDP, used to identify molecular interactions through functional-group shifts. (D) Particle sizes of GA3, GA3@HLDP (1:1), and GA3@HLDP (1:3.548). Each treatment included three independent samples, and data are presented as mean ± SE. (E and F) Particle size and (E) zeta potential (F) of GA3@HLDP at 5°C, 15°C, 25°C, 35°C, 45°C, and 55°C. Each treatment included three independent samples, and data are presented as mean ± SE. (G and H) TGA curves (G) and differential thermogravimetry curves (H) of GA3, HLDP, and GA3@HLDP. To evaluate the stability of GA3@HLDP under practical application conditions, changes in particle size and zeta potential were measured over a temperature range from 5°C to 55°C (Supplemental Figures 2E and 2F). The particle size of GA3@HLDP remained highly stable at different temperatures, with no significant changes as the temperature increased. This indicated that the complex maintained good structural integrity within the tested temperature range, without noticeable aggregation, dissociation, or phase transition. Zeta-potential values further revealed the surface characteristics of GA3@HLDP. Consistent with the positive charge of GA3 (Supplemental Table 3), GA3@HLDP also exhibited a positive charge across the entire temperature range, and the zeta potential remained within a relatively high positive range. This further supported the stability of the surface chemical properties of the composite. Furthermore, TGA and differential thermogravimetry were performed to verify the thermal stability of GA3@HLDP (Supplemental Figures 2G and 2H). GA3 began to decompose rapidly at approximately 200°C, whereas HLDP exhibited a broader decomposition temperature range. Notably, the thermal decomposition curve of GA3@HLDP lay between those of GA3 and HLDP, and its onset decomposition temperature was delayed compared with that of GA3. These results indicated that encapsulation within HLDP provided an effective protective barrier for thermally unstable GA3 molecules, significantly enhancing their thermal stability. Stronger plant uptake and better soil permeability of GA3@HLDP Plant uptake of agrochemicals is directly related to their biological activity and depends on basic physicochemical properties, including particle size and surface charge (Chachalis et al., 2001). Compared with GA3, GA3@HLDP had a smaller particle size and a higher positive surface charge, making it more likely to adhere to roots and thereby enhancing uptake (Figure 3A). According to the standard curve of GA3 (Supplemental Figure 3), the GA3 content in seedlings treated with GA3@HLDP increased to 169.22 and 260.33 μg/g at 4 and 8 h after root application, respectively, compared with 71.95 and 98.67 μg/g in seedlings treated with GA3 alone. Plant uptake, soil permeability, and cell membrane penetration mechanism of GA3@HLDP. (A) Time-dependent accumulation of GA3 in rice seedlings at 4 and 8 h after root application. (B) Residual GA3 in soil treated with GA3, GA3@HLDP (1:1), or GA3@HLDP (1:3.548). (C) Schematic illustration of the soil thin-layer chromatography flat-plate assay used to track formulation migration. (D) Retention factor (Rf) values of GA3, GA3@HLDP (1:1), and GA3@HLDP (1:3.548). (E–G) Leaching of GA3(E), GA3@HLDP (1:1) (F), and GA3@HLDP (1:1 and 1:3.548) (G). Each treatment included three independent samples, and data are presented as mean ± SE. Different letters above the bars indicate significant differences, as determined by Tukey’s HSD test, p < 0.05. (H) Simulation snapshots showing the penetration of GA3@HLDP into the rice seedling cell membrane at 4, 8, and 12 ns. (I) Hyperspectral images of transverse sections of rice seedling roots, stems, and leaves after root application of GA3@HLDP. The soil mobility of GA3 and GA3@HLDP was determined using soil leaching and soil migration experiments (Xiao et al., 2025). Soil leaching and migration experiments conducted under indoor conditions can effectively reflect actual in situ soil migration (Zhang et al., 2020). The detailed physicochemical properties of the soil are shown in Supplemental Table 4. As shown in Figure 3B, the residual GA3 content in soil significantly decreased as the leachate volume increased. Specifically, the residual GA3 content in soil was 45.51% after GA3@HLDP treatment at a mass ratio of 1:1 and 41.57% after GA3@HLDP treatment at a mass ratio of 1:3.548, compared with 88.33% after GA3 treatment. As shown in Figure 3C, the soil mobility of GA3@HLDP was determined by tracking the distribution of the GA3 mass fraction at various distances from the application point. The mass fraction of GA3 decreased sharply within the first 3 cm, with almost no detectable GA3 beyond 9 cm, suggesting limited soil mobility due to strong adsorption to soil particles (Figure 3E). In contrast, GA3@HLDP at mass ratios of 1:1 and 1:3.548 showed a more consistent mass fraction between 3 and 9 cm, indicating greater soil mobility, likely because encapsulation slowed degradation and enhanced transport (Supplemental Figures 3F and 3G). Furthermore, the retention factor (Rf) value serves as a key indicator of soil mobility and potential availability of the tested agrochemicals (Xiao et al., 2025). As shown in Figure 3D, the significantly higher Rf values of GA3@HLDP at mass ratios of 1:1 and 1:3.548 compared with GA3 suggested that encapsulation greatly enhanced the soil mobility of GA3, likely owing to reduced particle size, weakened interactions with soil components, and enhanced particle stability. In the case of GA3, microscale aggregation (2079.85 nm) and exposed polar groups led to severe entrapment through both physical filtration and chemical adsorption onto soil organic matter. Self-assembly with HLDP not only optimized the particle size to the sub-200-nm range but also provided a hydrophobic shield that minimized contact between GA3 molecules and high-affinity sites in the soil matrix. This synergistic effect significantly reduced the energy barrier during transport, as reflected by the more consistent mass-fraction distribution observed in the soil migration experiment between 3 and 9 cm. The strong soil mobility of GA3@HLDP may provide more uniform coverage and enhance plant uptake, particularly in deep-rooted crops, thereby reducing the need for repeated applications. Strong translocation of GA3@HLDP across the plant cell membrane The translocation of agrochemicals across the cell membrane is a critical factor influencing their biological activity (Deng et al., 2025c). Safeners must enter the cytoplasm of plant cells to activate detoxification pathways that mitigate herbicide-induced damage. Therefore, we aimed to clarify the membrane penetration mechanism of GA3@HLDP. As illustrated in Figure 3H, membrane-penetration molecular simulations were constructed to investigate the transmembrane behavior and free-energy barrier of GA3@HLDP. GA3@HLDP molecules initially contacted the cell membrane at 4 ns, began to penetrate the lipid bilayer at 8 ns, and achieved complete translocation across the cell membrane within 12 ns (Figure 3H). To further quantify penetration, potential of mean force (PMF) curves were used to estimate the energy barrier for GA3@HLDP, indicating that GA3@HLDP could spontaneously undergo transmembrane transport (Supplemental Figure 4A). Changes in the relative distance between GA3@HLDP and the center of the cell membrane are shown in Supplemental Figure 4B, illustrating the dynamic process by which GA3@HLDP successfully penetrated the cell membrane. Dynamic videos are provided in Supplemental Video 1. Supplemental Video 1. Dynamic process of GA3@HLDP penetrating cell membran In addition to PMF analysis, key structural properties of the cell membrane, including lipid surface area, membrane thickness, and lipid order parameters, were evaluated to comprehensively assess the overall impact of GA3@HLDP (Supplemental Figures 4C–4F). The membrane lipid composition of plants is provided in Supplemental Table 5. Only minor fluctuations in lipid surface area were observed in the GA3@HLDP system (Supplemental Figure 4C). Specifically, the average area of each lipid molecule remained stable throughout the 12-ns simulation, fluctuating only within a very small range. Similarly, membrane thickness remained stable at approximately 4.50 nm (Supplemental Figure 4D). These data demonstrated that GA3@HLDP penetration was a non-destructive process that maintained the macroscopic structural stability of the plasma membrane. To explore local perturbations of the membrane at the atomic scale, we analyzed the lipid-tail order parameters (Supplemental Figures 4E and 4F). Compared with the pre-penetration state, the post-penetration phase showed a marked reduction in lipid-tail order parameters, confirming that GA3@HLDP not only crossed the membrane but also induced substantial membrane fluidization. Together, the MD simulation revealed a potential mechanism by which GA3@HLDP translocated across the cell membrane. Hyperspectral imaging enables the non-destructive identification and spatial mapping of microcapsules within plant tissues by capturing detailed spectral signatures across a broad wavelength range. Therefore, this approach was used to trace GA3@HLDP in rice seedlings (Deng et al., 2025c). GA3@HLDP showed intense light scattering around 600 nm, and its spectral characteristics were analyzed between 400 and 1000 nm. Using a spectral angle mapping algorithm, GA3@HLDP was detected mainly in roots, stems, and leaves, where it appeared as red pixels, further indicating that GA3@HLDP had efficient transmembrane capability in rice tissues (Figure 3I; Supplemental Figure 5). In addition, the hyperspectral signals were significantly stronger in root and leaf tissues, suggesting efficient vascular-mediated transport, potentially via sieve tubes. The uneven distribution of GA3@HLDP among different plant tissues further confirmed the dominant role of root absorption and transport. Confocal laser scanning microscopy in BY-2 cells served as a bridge between the theoretical MD simulations and the phenotypic results in rice. As shown in Supplemental Figure 6, robust Cy3 fluorescence was observed within the cytoplasm of BY-2 cells after treatment with fluorescent GA3@HLDP. The presence of fluorescent cargo inside the cells provided direct visual evidence that GA3@HLDP could efficiently cross the plant cell membrane and deliver its payload into the cytoplasm. Collectively, these multi-scale results, spanning atomistic simulations, cell imaging, and plant assays, provided robust validation of the efficient transmembrane delivery of GA3@HLDP. To elucidate the release behavior of GA3 after cellular internalization, in vitro release experiments were performed at pH 5.5 and pH 7.4 (Supplemental Figure 7). The results showed sustained release of GA3 from HLDP in both environments, with cumulative release reaching approximately 70% within 48 h. The release data were further fitted using various kinetic models (Supplemental Table 6). The Korsmeyer–Peppas model provided the best fit, with the highest correlation coefficients (R2 = 0.9922 at pH 5.5; R2 = 0.9880 at pH 7.4), indicating a diffusion-controlled release mechanism. This sustained release capability ensured that, once GA3@HLDP entered plant cells, it could provide a steady supply of active GA3 to trigger downstream signaling pathways. Enhanced safening activity of GA3@HLDP while maintaining herbicidal activity The herbicide PND inhibits cell division, leading to suppressed crop growth and reductions in plant height, root length, and fresh weight, which ultimately compromise crop quality and yield (Chen et al., 2025a, 2025b; Wang et al., 2025a, 2025b). As shown in Figures 4A–4D, the safening activities of GA3 and GA3@HLDP against PND EC-induced rice injury were evaluated under greenhouse conditions. Specifically, PND EC application caused significant herbicide injury, with plant height, root length, and fresh weight reduced by 10.33%, 20.33%, and 12.33%, respectively. Co-application with GA3 or GA3@HLDP effectively alleviated PND-induced injury in rice. The PND + GA3 treatment still caused seedling damage, with relative plant height, root length, and fresh weight values of 67%, 97.33%, and 49%, respectively. Notably, PND + GA3@HLDP showed the best safening performance, achieving relative growth values of 79.67% for plant height, 99.00% for root length, and 79.67% for fresh weight, thereby substantially alleviating PND-induced injury. In addition, application of HLDP alone did not significantly affect the plant height or root length of rice seedlings compared with the control. Conversely, the PND + HLDP treatment caused a marked reduction in both plant height and root length, including approximately 80% inhibition of plant height, with no significant difference from the PND-only group. These results indicated that HLDP could not mitigate PND phytotoxicity in the absence of GA3 (Supplemental Figure 8). Traditional commercial safeners often have serious environmental risks, and natural alternatives, such as gibberellin and melatonin, are also limited by poor solubility, low plant absorption, and low utilization efficiency (Liu et al., 2021a, 2021b; Oloye et al., 2021; Lanasa et al., 2022; Leng et al., 2023). In the current study, encapsulation of GA3 within HLDP effectively addressed these limitations by improving dispersibility and bioavailability, offering a promising strategy for developing efficient natural safeners. Safening and herbicidal activities of PND EC, PND EC + GA3, and PND EC + GA3@HLDP. (A) Photograph of rice seedlings treated with different formulations at 14 days after application. (B–D) Relative plant height (B), relative root length (C), and relative fresh weight (D) of rice seedlings after 14 days of treatment. Each treatment included three biological replicates, and data are presented as mean ± SE. Distilled water was used as the control. Different letters indicate significant differences as determined by one-way ANOVA followed by Tukey’s HSD test, p < 0.05. (E) Photograph of barnyard grass treated with different formulations at 14 days after application. (F and G) Fresh-weight control effect (F) and visual control effect (G) after 14 days of treatment. Each treatment included three biological replicates, and data are presented as mean ± SE. Distilled water was used as the control. The same letters indicate no significant difference as determined by one-way ANOVA followed by Tukey’s HSD test, p > 0.05. Herbicidal activity should remain unaffected after the incorporation of safeners (Jia et al., 2022). Therefore, the weed control efficacy of PND EC co-applied with GA3@HLDP was also evaluated under greenhouse conditions (Figures 4E–4G). Our results showed that the fresh-weight control effect and visual control effect exceeded 98% across all treatments, demonstrating that the addition of GA3 or GA3@HLDP did not compromise the herbicidal activity of PND against weeds. These data suggested that application of GA3@HLDP at the current dose might not induce detoxification enzymes, such as glutathione S-transferases (GSTs), in weeds. Thus, GA3@HLDP may act as an effective safener that maintains herbicidal efficacy while providing crop protection. Future studies will focus on multi-dose evaluations of GA3@HLDP + PND in field trials to validate its efficacy and safety under real-world farming conditions. Synergistic mechanism of GA3@HLDP through activation of phenylpropanoid biosynthesis, glutathione metabolism, and starch/sucrose metabolism GA3 can serve as a natural safener by activating antioxidant defenses, modulating the expression of herbicide-induced stress-responsive genes, and regulating phytohormone biosynthesis and signaling, thereby enhancing crop tolerance to herbicides (Nemat Alla and Hassan, 1998; Zhang et al., 2022; Wang et al., 2024a, 2024b; Chen et al., 2024). The quality of the sequencing data and the correlations among the three biological replicates are presented in Supplemental Table 7 and Supplemental Figure 9, respectively. Compared with GA3 treatment, GA3@HLDP treatment resulted in 3662 upregulated and 3251 downregulated differentially expressed genes (DEGs) in rice seedlings (Supplemental Figure 10A). Notably, functional categorization of DEGs revealed significant enrichment in processes associated with herbicide resistance (Supplemental Figure 10B). As shown in Supplemental Figure 10C, heatmap analysis demonstrated that GA3@HLDP treatment significantly upregulated key genes related to glutathione metabolism, peroxisomes, ABC transporters, endocytosis, phenylpropanoid biosynthesis, and fructose and mannose metabolism compared with GA3 treatment. This coordinated activation of genes involved in defense, detoxification, redox homeostasis, and energy metabolism suggested a synergistic mechanism for enhancing herbicide tolerance. In addition, 14 DEGs related to herbicide resistance were selected for RT–qPCR analysis, and the results were consistent with the transcriptomic data (Supplemental Figure 10D). GA3@HLDP treatment upregulated genes encoding glutathione hydrolase 3 (GH3), gamma-glutamylcyclotransferase 2 (GGCT2), tau class glutathione S-transferase 6 (GSTU6), superoxide dismutase (SOD), ABC transporter B family member 4 (ABCB4), and ABC transporter G family member 22 (ABCG22), which are known to regulate oxidative stress and protect plants from oxidative damage (Sytykiewicz, 2014; Li et al., 2017; Molina-Moya et al., 2025). Furthermore, the endocytosis-related gene Ras-related protein Rab-5A (RAB5A) was markedly upregulated in the presence of HLDP, which may benefit the cellular uptake of GA3. RAB proteins primarily function in cellular vesicular trafficking by cycling between GTP- and GDP-bound states, thereby acting as molecular switches that regulate vesicle budding, formation, transport, tethering, and fusion (Zerial and McBride, 2001). The upregulation of these genes suggested that GA3@HLDP did not rely exclusively on passive diffusion but may also trigger clathrin-mediated endocytosis for cellular entry. This mechanism was consistent with the dynamic internalization observed in BY-2 cell images and highlighted the role of HLDP as a transport-enhancing carrier that promoted the efficient delivery of GA3 into the cytoplasm, thereby activating downstream defense responses. Furthermore, the significantly enhanced expression of phenylpropanoid biosynthesis genes (phenylalanine ammonia-lyase [PAL], cinnamoyl-CoA reductase 1 [CCR1], 4-coumarate: CoA ligase 3 [4CL3], cinnamyl alcohol dehydrogenase 2 [CAD2]) in rice seedlings suggests improved phenylpropanoid biosynthesis. This pathway contributes to reactive oxygen species (ROS) scavenging and reduces oxidative damage, thereby enhancing resistance to abiotic stress (Dong and Lin, 2021). As an important plant growth-regulating hormone, gibberellin mainly promotes cell division and elongation, thereby regulating plant height (Chen et al., 2025a, 2025b). Our results showed that cell division-related genes (fructose-bisphosphate aldolase [FBA], fructose-bisphosphate aldolase 5 [FBA5], triosephosphate isomerase [TPI]) were also significantly upregulated. Flow cytometry analysis further supported this conclusion, as seedlings treated with GA3@HLDP showed a higher cell division rate than those treated with GA3, providing additional evidence that HLDP enhanced the bioactivity of GA3 by promoting cell proliferation (Supplemental Figure 11). Furthermore, HLDP-only treatment did not trigger significant transcriptional changes in key genes such as GSTU6 and SOD (Supplemental Figure 12). In contrast, the PND + HLDP group showed significant downregulation of these genes, further supporting that the transcriptomic changes observed in GA3@HLDP-treated rice were specifically induced by HLDP-delivered GA3. In addition, metabolomic profiling was performed to analyze differentially accumulated metabolites (DAMs) in rice seedlings treated with GA3 or GA3@HLDP. Principal-component analysis of replicate samples revealed that the first two principal components explained 51.80% and 17.00% of the total variance, respectively. The clear separation of treatment groups and tight clustering of biological replicates demonstrated high experimental reproducibility (Supplemental Figure 13). Compared with GA3 treatment, GA3@HLDP treatment resulted in 430 upregulated and 432 downregulated DAMs in rice seedlings (Figure 5A). Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis suggested that the DAMs were primarily enriched in pathways related to plant hormone signal transduction, biosynthesis of various plant secondary metabolites, flavone and flavonol biosynthesis, and ABC transporters, among other pathways (Figure 5B). These pathways contribute to plant resistance through mechanisms such as toxin efflux, growth regulation under stress, and enhanced cellular tolerance (Conte and Lloyd, 2011). Integrated transcriptomic and metabolomic analysis was subsequently performed, identifying 61 shared pathways through KEGG enrichment analysis, including brassinosteroid biosynthesis, isoquinoline alkaloid biosynthesis, carotenoid biosynthesis, glutathione metabolism, phenylpropanoid biosynthesis, and starch/sucrose metabolism (Figure 5C). This convergence was particularly evident in key pathways associated with herbicide resistance, such as glutathione metabolism, phenylpropanoid biosynthesis, and starch/sucrose metabolism, which together enhanced the tolerance of rice seedlings to PND. GA3@HLDP treatment significantly activated the expression of PAL and ferulate-5-hydroxylase (F5H), promoting the synthesis of downstream phenylpropanoids, such as ferulic acid and 5-hydroxyferulic acid, thereby reinforcing the cell-wall barrier and enhancing direct defense against PND (Figure 5D). In parallel, GA3@HLDP treatment regulated the multistep pathway from L-glutamic acid to glutathione synthesis, effectively improving oxidative stress scavenging and detoxification capacity in rice seedlings. The glutathione pathway confers herbicide resistance primarily through GST-mediated enzymatic detoxification and alleviation of oxidative damage (Cummins et al., 2009; Jo et al., 2011; Brazier-Hicks et al., 2018). Meanwhile, changes in sucrose synthase activity and related sugar phosphate intermediates in starch/sucrose metabolism provided the carbon skeletons and energy required for these stress-resistance responses (Figure 5D). Integrated metabolomic and transcriptomic analysis of the synergistic mechanism of GA3@HLDP. (A) Volcano plot analysis of DAMs in leaves treated with GA3@HLDP or GA3. Upregulated and downregulated metabolites are represented by red and blue dots, respectively. DAMs, differential accumulation metabolites. (B) KEGG enrichment analysis of DAMs. (C) Common pathways identified by integrated metabolomic and transcriptomic analysis. Representative common pathways are shown. (D) Effects of GA3@HLDP on phenylpropanoid biosynthesis, glutathione metabolism, and starch and sucrose metabolism pathways. DAM levels are shown as a heatmap using log2(fold change). Upregulated DEGs are marked in red. PAL, phenylalanine ammonia-lyase; F5H, ferulate-5-hydroxylase; GSS, glutathione synthetase; GST, glutathione S-transferase; GGT, gamma-glutamyl transferase; SUS, sucrose synthase; HK, hexokinase. To further elucidate the regulatory logic underlying the safening effect of GA3@HLDP, a gene-metabolite correlation network was constructed (Supplemental Figure 14). The network highlighted sucrose and cellobiose as pivotal metabolic nodes that showed strong correlations with a wide array of DEGs involved in fructose and mannose metabolism, glutathione metabolism, and phenylpropanoid biosynthesis. This architecture suggested that sugar metabolism served as a central engine, providing the carbon skeletons and energy required for downstream defense processes. Furthermore, the correlations of ferulic acid and glutamic acid with their respective biosynthesis genes supported the coordinated reinforcement of cell walls and activation of the antioxidant system. Interestingly, genes related to endocytosis were also closely associated with these metabolic hubs, suggesting that efficient internalization of the nanocarrier may have contributed to the widespread metabolic reprogramming observed in rice seedlings. In summary, GA3@HLDP systematically enhanced the overall tolerance of rice seedlings to PND stress by synergistically activating an integrated metabolic network encompassing physical barrier construction, antioxidant detoxification, and energy provision. Specifically, the safening effect of GA3@HLDP appeared to arise from coordinated metabolic reprogramming rather than the independent action of isolated pathways. Our integrated analysis revealed that starch and sucrose metabolism functioned as a primary metabolic hub, providing the carbon skeletons and energy required to fuel active defense processes. This energy supply directly supported phenylpropanoid biosynthesis for cell-wall reinforcement and glutathione metabolism for enzymatic detoxification of PND and ROS scavenging. This energy-driven, multilayered defense model provides a comprehensive explanation for the superior activity of GA3@HLDP compared with GA3. Enhanced antioxidant capacity in rice seedlings treated with GA3@HLDP Herbicide exposure triggers adverse effects, including suppressed defense-related biosynthesis and nitrogen metabolism, reduced photosynthesis, oxidative damage, and nutrient imbalance, all of which contribute to growth inhibition and yield loss (Moustakas et al., 2016; Wang et al., 2025a, 2025b). The activities of GST, SOD, POD, CAT, PAL, 4CL, C4H, glutamine synthetase (GS), NADH-GDH, Fd-GOGAT, Rubisco, and PEPC, as well as the contents of MDA, soluble sugar (SS), and chlorophyll (a, b, and total), were measured in rice seedlings treated with deionized water (control), PND, PND + GA3, or PND + GA3@HLDP (Figure 6). In the PND treatment, the activities of GST, SOD, POD, CAT, PAL, 4CL, C4H, GS, NADH-GDH, Fd-GOGAT, Rubisco, and PEPC were significantly inhibited compared with those in the control (Figures 6A–6L). After the addition of safeners, most of these enzyme activities increased. In particular, compared with the PND + GA3 treatment, the activities of GST, SOD, POD, CAT, 4CL, C4H, GS, NADH-GDH, Fd-GOGAT, Rubisco, and PEPC in the PND + GA3@HLDP treatment increased by 27.98%, 570.55%, 15.30%, 274.55%, 66.12%, 20.74%, 0.41%, 15.54%, 7.92%, 62.58%, and 277.82%, respectively, and returned to levels comparable to those in the control. Elevated GST activity promotes the conjugation of GSH to herbicides, significantly improving the crop’s capacity for herbicide detoxification (Li et al., 2017). Meanwhile, SOD, POD, and CAT form a detoxification network that converts highly reactive and toxic ROS into less reactive or non-toxic molecules, thereby maintaining cellular redox homeostasis of cells and supporting normal crop growth and development. In addition, the enhanced activities of PAL, 4CL, and C4H, which are key enzymes in the phenylpropanoid biosynthesis pathway, promote the biosynthesis of phenolic compounds, lignin, and other secondary metabolites, thereby strengthening direct crop defense against herbicides (Dong and Lin, 2021). Concurrently, improvements in nitrogen metabolism, governed by key enzymes such as GS, NADH-GDH, and Fd-GOGAT, together with enhanced photosynthetic carbon assimilation mediated by PEPC and Rubisco, further support crop growth and development. Physiological and biochemical parameters in rice seedlings treated with PND EC, PND EC + GA3, and PND EC + GA3@HLDP. Activities of GST (A), SOD (B), POD (C), CAT (D), PAL (E), 4CL (F), C4H (G), GS (H), NADH-GDH (I), Fd- GOCAT (J), Rubisco (K), and PEPC (L), and contents of MDA (M), SS (N), chlorophyll a, chlorophyll b (O), and total chlorophyll (P). Each treatment included three biological replicates, and data are presented as mean ± SE. Distilled water was used as the control. Different letters indicate significant differences, as determined by one-way ANOVA followed by Tukey’s HSD test, at p < 0.05. GST, glutathione S-transferase; SOD, superoxide dismutase; POD, peroxidase; CAT, catalase; PAL, phenylalanine ammonia-lyase; 4CL, 4-coumarate: CoA ligase; C4H, cinnamate 4-hydroxylase; GS, glutamine synthetase; NADH-GDH, NAD-dependent glucose dehydrogenase; Fd-GOGAT, ferredoxin-dependent glutamate synthase; Rubisco, Ribulose-1,5-bisphosphate carboxylase/oxygenase; PEPC, phosphoenolpyruvate carboxylase; MDA, malondialdehyde; SS, soluble sugar. In addition to these changes in enzyme activities, the contents of MDA, SS, and chlorophyll (a, b, and total) also differed significantly among treatment groups (Figures 6M–6P). In the PND treatment, MDA content increased significantly, whereas the contents of SS and chlorophyll (a, b, and total) decreased significantly compared with the control, indicating severe physiological damage and metabolic disruption induced by PND stress. Compared with the PND treatment alone, the addition of GA3 or GA3@HLDP significantly alleviated these effects. Notably, compared with GA3, GA3@HLDP reduced MDA content by 19.02% and increased the contents of SS, chlorophyll a, chlorophyll b, and total chlorophyll by 67.82%, 556.74%, 66.41%, and 220.71%, respectively. These improvements significantly exceeded those observed with GA3 treatment and restored the measured parameters to levels comparable to those in the control. MDA is a clear marker of lipid peroxidation and reflects oxidative damage to cell membranes caused by herbicide-induced ROS (Abou-Zeid et al., 2020). MDA levels were significantly reduced in seedlings treated with GA3@HLDP, likely owing to the elevated activities of major antioxidant enzymes, including SOD, POD, and CAT. Moreover, restoration of chloroplast function can reduce herbicide-induced light damage and promote the recovery of photosynthesis, thereby reducing phytotoxicity and promoting crop growth (Wang et al., 2024a, 2024b). Further biochemical analysis revealed that HLDP treatment had no effect on the activities of core enzymes involved in antioxidant defense (e.g., GST, SOD, POD, CAT) or metabolism (e.g., PAL, GS, Rubisco) (Supplemental Figure 15). In the PND + HLDP treatment, MDA content significantly increased, whereas chlorophyll and SS levels decreased sharply, mirroring the physiological stress observed in PND-treated rice. This confirmed that the physiological safening effect was not mediated by HLDP alone. In summary, GA3@HLDP showed optimal synergistic effects by restoring detoxification metabolism and defense capacity, repairing nitrogen assimilation and metabolic processes, and improving photosynthetic efficiency and carbon fixation. Reduced acute toxicity of GA3@HLDP toward non-target organisms To be suitable for agricultural application, safeners should exhibit minimal toxicity toward non-target organisms (Deng et al., 2025c). In this study, we evaluated the acute toxicity of GA3 and GA3@HLDP using multi-level toxicology models, including zebrafish embryos, fruit flies, and HSF cells. The median lethal concentration (LC50) values of GA3 and GA3@HLDP toward zebrafish embryos were 17.28 and 26.68 mg/l, respectively, indicating lower toxicity of GA3@HLDP (Figure 7A; Supplemental Table 8). Meanwhile, no fruit fly mortality was observed after treatment with either GA3 or GA3@HLDP, confirming their safety toward fruit flies (Figure 7B). In HSF cells, GA3 exhibited an half maximal inhibitory concentration (IC50) of 10.39 mg/ml, whereas GA3@HLDP showed lower cytotoxicity, with an IC50 of 18.21 mg/ml (Figure 7C). Consistently, a previous study also reported that HLDP was non-phytotoxic to rice seedlings (Deng et al., 2025b). Collectively, these results demonstrated that GA3@HLDP could serve as a selective and eco-friendly nano-safener that minimizes adverse effects on non-target organisms. Biosafety and environmental compatibility of GA3@HLDP. (A) Mortality of zebrafish embryos treated with GA3 or GA3@HLDP. (B) Survival rate of fruit flies treated with GA3 or GA3@HLDP. Each treatment included three biological replicates, and data are presented as mean ± SE. (C) Cytotoxicity of GA3 and GA3@HLDP toward HSF cells at 24 h. (D–F) Alpha-diversity analysis (D and E) and beta-diversity analysis (F) of soil samples treated with GA3 or GA3@HLDP. (G) Differences in the phylum-level composition of soil samples treated with GA3 or GA3@HLDP. (H) Heatmap showing differences in genus-level composition of soil samples treated with GA3 or GA3@HLDP. (I) Functional Annotation of Prokaryotic Taxa (FAPROTAX) analysis of soil microbial abundance in GA3 and GA3@HLDP treatments. No adverse effects of GA3@HLDP on soil microbial communities As an important component of soil ecology, soil microorganisms are highly sensitive to environmental changes, making them reliable indicators of soil quality and fertility (Li et al., 2022). However, studies have shown that various nanoparticles can adversely affect soil ecosystems by reducing bacterial diversity and altering community structure (Ge et al., 2011; Ma et al., 2013). Therefore, this study systematically evaluated the ecological safety of GA3@HLDP in soil by analyzing microbial community structure, taxonomic composition, and functional potential. Our results showed that GA3@HLDP had no adverse effect on the soil microbial ecosystem. As shown in Figures 7D and 7E, there were no significant differences in the Chao1 or Shannon indices between the GA3 and GA3@HLDP treatments. Similarly, beta-diversity analysis indicated highly similar community structures, with no significant separation between treatments (Figure 7F). The soil microbial communities treated with GA3 and GA3@HLDP were also highly consistent at the phylum level, with no significant changes in relative abundance (Figure 7G). Furthermore, genus-level analysis showed significant enrichment of both Haliangium and Sphingomonas in the GA3@HLDP treatment (Figure 7H). Haliangium spp. perform various ecological functions, such as nitrogen fixation and chemoheterotrophy, which can promote the growth of rice seedlings (Chang et al., 2022). Similarly, Sphingomonas is commonly associated with organic matter degradation and plant-growth promotion (Asaf et al., 2020; Takeo et al., 2020). The enrichment of Sphingomonas therefore further suggested that GA3@HLDP may promote beneficial soil microorganisms. The enrichment of Haliangium and Sphingomonas provided a favorable microbial context for the observed safening activity. Haliangium is a core rhizomicrobiome genus in rice that facilitates the acquisition of nitrogen and metal nutrients. Meanwhile, Sphingomonas species function as plant-growth-promoting rhizobacteria that can enhance abiotic stress tolerance and promote organic matter degradation. Therefore, GA3@HLDP may improve the rhizosphere environment in a manner conducive to the growth of beneficial bacteria, thereby enhancing the physiological resistance of rice seedlings to PND-induced phytotoxicity. In addition, beneficial environmental metabolic functions, such as phototrophy, ureolysis, and methylotrophy, were predicted to be enhanced in the GA3@HLDP treatment, further indicating that GA3@HLDP had no negative impact on the soil environment during the 7-day observation period (Figure 7I). However, given the dynamic and complex nature of soil ecosystems, future long-term field experiments are warranted to validate the sustained ecological safety of GA3@HLDP and the persistence of these beneficial microbial shifts under practical conditions. Concluding remarks In this study, the nanocarrier HLDP was successfully synthesized and loaded with GA3 to generate an efficient and eco-friendly nano-safener, GA3@HLDP. HLDP assembled with GA3 through hydrogen bonding and hydrophobic association, and the spontaneous complexation of GA3 with HLDP formed nearly spherical nanoparticles. GA3@HLDP showed greater soil mobility and enhanced penetration across plant cell membranes, thereby significantly improving the protective effects of GA3 and alleviating PND-induced phytotoxicity in rice seedlings. Integrated transcriptomic and metabolomic analyses showed that complexation with HLDP primarily activated key metabolic pathways, including phenylpropanoid biosynthesis, glutathione metabolism, and starch and sucrose metabolism, while also restoring key antioxidant levels and related enzyme activities. Furthermore, GA3@HLDP showed favorable biocompatibility in model organisms, including zebrafish embryos, fruit flies, and HSF cells, and had no negative effects on soil microorganisms. In summary, our study used HLDP to comprehensively improve the performance of GA3 through multifaceted synergistic mechanisms. The application of GA3@HLDP substantially alleviated PND-induced injury in rice, supporting the translation of natural safeners from fundamental research toward field application. Methods Materials Tetrahydrofuran (THF), acetonitrile, dichloromethane, methanol (MeOH), acetone, and 5% sodium hypochlorite (NaClO) solution were obtained from Beijing Chemical Works (China). GA3 (98%) was purchased from Shanghai Yuanye Biotechnology (China), and 33% PND EC was obtained from Hailir Pesticides and Chemicals Group (China). HLDP synthesis HLDP was synthesized as shown in Figure 1A. First, PCL was prepared by ring-opening polymerization of ε-CL (6.16 g, 54 mmol), initiated by 1-butanol (100 mg, 1.35 mmol) and catalyzed by Sn(Oct)2 (310 mg) at 90°C for 9 h. Dichloromethane was then added, and the polymer was isolated by precipitation in cold MeOH. Subsequently, PCL-Br was obtained by reacting PCL (3 g, 0.58 mmol) with 2-bromoisobutyryl bromide (3 g, 13.04 mmol) and triethylamine (4 g, 40 mmol) in THF, followed by termination with MeOH and collection of the precipitate by filtration. Finally, HLDP was synthesized by polymerizing dimethylaminoethyl methacrylate (DMAEMA) (0.87 g, 6.07 mmol) using PCL-Br (100 mg, 0.04 mmol) as the macroinitiator, with CuBr (15 mg, 0.1 mmol) and N,N,N′,N′′,N′′-Pentamethyldiethylenetriamine (PMDETA) (37 mg, 0.21 mmol) in THF at 65°C. The structure of HLDP and the degree of polymerization of each block were previously confirmed by 1H NMR spectroscopy (Li et al., 2025). Preparation of GA3@HLDP GA3@HLDP was prepared according to the loading capacity, which was measured using a freeze-drying method. Specifically, HLDP (41.86 mg) was incubated with GA3 (50 mg) in 5 ml of ethanol for 15 min at room temperature. The 5-ml solution was transferred into a regenerated cellulose dialysis bag with a molecular weight cutoff of 1,000 Da and immersed in 1,000 ml of ethanol for dialysis for 24 h to remove unassembled GA3. GA3@HLDP was then freeze-dried, and the total mass of GA3@HLDP was weighed. The DLC was calculated using Equation 1: ITC assay, molecular docking, and molecular simulation for analyzing the self-assembly mechanism of GA3@HLDP For the ITC assay, GA3 and HLDP were prepared in methanol/double-distilled water (ddH2O) (1:49, v:v). HLDP (1,000 μl, 0.1 mM) was titrated with GA3 solution (250 μl, 1 mM) using a Nano ITC system (TA Instruments Waters, USA) at room temperature. The heat change associated with each injection was derived by integrating the corresponding titration peaks using Origin 20 software (OriginLab, USA). Finally, the Gibbs free energy (ΔG) of the interaction was calculated from the resulting data using ΔG = ΔH – TΔS. Molecular docking was performed using AutoDock Vina 1.1.2 to elucidate the molecular interaction modes among the components (Eberhardt et al., 2021). In this study, GA3 and the main fragment of HLDP were constructed as ligands and optimized in Sketch mode using the Tripos force field and Gasteiger–Hückel charges, along with other default parameters. The appropriate protonation states (protomols) of these molecules were generated using the automatic docking mode, and interaction scores were assessed using the total score to evaluate molecular interaction modes. These ligands were docked into the protonated states to clarify their molecular interactions, and the interaction modes were visualized using PyMOL 3.7 software (Schrödinger, USA) (Seeliger and de Groot, 2010). All-atom MD simulations of 100 ns were further performed using GROMACS v2022.03 to investigate the self-assembly process between GA3 and HLDP (Essmann et al., 1995; Hess et al., 1997; Martínez et al., 2009; Jo et al., 2014; Abraham et al., 2015; Donnelly et al., 2021). The initial systems were built with PackMOL and contained 50 HLDP molecules, 50 GA3 molecules, and 8000 water molecules, using the transferable intermolecular potential with 3 points (TIP3P) water model and generalized amber force field (GAFF) force-field parameters generated via Sobtop. The detailed simulation method is provided in the supplemental information. Characterization of GA3@HLDP Infrared spectra of GA3, HLDP, and GA3@HLDP at a mass ratio of 1:3.548 were recorded using a PerkinElmer Fourier transform infrared (FTIR) spectrometer. Next, the characteristics of GA3@HLDP at mass ratios of 1:1 and 1:3.548, including polydispersity, particle size, and zeta potential, were measured using a Zeta Sizer Nano Series instrument (Malvern Instruments, UK) at room temperature. Furthermore, the particle size and zeta potential of GA3@HLDP at a GA3 mass ratio of 1: 3.548 were measured at various temperatures to assess its stability. For morphological observation, the morphologies of GA3 and GA3@HLDP were observed using SEM (JSM-6700F, Japan) and TEM (Hitachi HT7700, Japan). Elemental mapping of GA3@HLDP was performed by EDS using a JEM-F200 TEM (JEOL, Japan). Thermogravimetric analysis (TGA) of GA3, HLDP, and GA3@HLDP at a GA3 mass ratio of 1:3.548 was conducted using a TGA/DSC3+ instrument (Mettler-Toledo, Switzerland). To evaluate the release behavior of GA3, in vitro release experiments were conducted under two different physiological conditions (pH 5.5 and pH 7.4). Detailed experimental procedures are provided in the supplemental information. Plant uptake assay of GA3@HLDP Rice seedlings (Xiangwanxian No. 10, a conventional indica rice cultivar) were harvested at 4 and 8 h after root application of GA3 (0.1 mg/ml) or GA3@HLDP at a GA3 mass ratio of 1:3.548 to determine the plant uptake of GA3. Leaves (0.5 g) were harvested and ground in 1.8 ml of acetonitrile/ddH2O (10:1, v/v) to extract GA3 for HPLC analysis. The mobile phase for GA3 was acetonitrile/ddH2O (35:65, v/v) delivered at a flow rate of 1.0 ml/min with the column temperature maintained at 35°C, and detection was performed at 220 nm. A standard curve was constructed using GA3 solutions ranging from 5 to 160 μg/ml prepared in methanol. Mobility assay of GA3@HLDP in soil The specific method used for the soil migration experiment is provided in the supplemental information. Soil samples collected from a depth of 0–20 cm were first dried at 25°C and then passed through a 2-mm sieve. The processed soil was packed into plastic columns (2.8 cm in diameter and 8 cm in height), with glass wool at the base. The top of each column was sequentially layered with quartz sand and filter paper. After 1 h of pre-wetting, 200 μl of either GA3 or GA3@HLDP solution (at a concentration of 25 mg/ml, with GA3 ratios of 1:1 or 1:3.548) was applied to the surface. Deionized water was then used for elution, and the resulting leachates were collected separately. Each leachate was analyzed by HPLC according to the method described above. The amount of residual GA3 was calculated using Equation 2: where N1 represents the total amount of GA3, and N2 represents the amount of GA3 eluted from the soil column. Molecular simulation of GA3@HLDP cell membrane penetration All-atom MD simulations were performed using GROMACS v2022.05 to investigate the penetration process of GA3@HLDP across the asymmetric plasma membrane of rice plants (Abraham et al., 2015). The simulation system was constructed using PackMOL (Martínez et al., 2009) and contained one HLDP molecule, one GA3 molecule, 75 sodium ions, 47 chloride ions, and 15 846 water molecules modeled with TIP3P water. Topologies and parameters for HLDP, GA3, and membrane components were generated using the CHARMM-GUI web server based on the CHARMM36 force field (Jo et al., 2014). To obtain a stable simulation system, energy minimization was first performed using the steepest-descent algorithm (Donnelly et al., 2021). All simulations were conducted under periodic boundary conditions. Short-range electrostatic and van der Waals interactions were treated with a cutoff distance of 1.2 nm, whereas long-range electrostatic interactions were computed using the particle mesh Ewald (PME) method (Essmann et al., 1995). Bond constraints were applied to all hydrogen atoms using the LINCS algorithm (Hess et al., 1997). System pressure was maintained at 1 bar using the Berendsen barostat, and temperature was controlled using the V-rescale coupling method (Berendsen et al., 1984). The procedure consisted of a 2-ns pre-equilibration stage to optimize the initial conformation of the compound, followed by a 10-ns equilibration in the NVT ensemble to gradually heat the system to the target temperature. Subsequently, a 10-ns equilibration in the NPT ensemble was performed for pressure equilibration. Finally, a 10-ns production MD simulation was carried out in the NPT ensemble at 298.15 K and 1 bar. The integration time step was set to 1 fs, and trajectories, including coordinates, velocities, energies, and forces, were recorded every 5 ps for subsequent analysis. To evaluate the stability and conformational evolution of the self-assembled complex, umbrella sampling was performed with a pulling velocity of 1 nm/ns and a force constant ranging from 200 to 500 kJ/mol·nm2. Key stability parameters were extracted from the trajectories for further assessment. Safening activity assay of GA3@HLDP in rice seedlings Based on prior herbicide efficacy studies, which established 300 g/ha PND EC as the application rate achieving more than 95% control efficacy against barnyard grass, pre-experimental optimization of the minimal protective dose of GA3 (4 mg/l) against 300 g/ha PND-induced phytotoxicity, and loading efficiency tests, in which the 1:3.548 GA3@HLDP ratio was optimized for delivery, the following concentrations were used in the safening activity assay: 300 g/ha PND EC, 4 mg/l GA3, and GA3@HLDP at a mass ratio of 1:3.548. Treatments with 300 g/ha PND EC, 300 g/ha PND EC + 4 mg/l GA3, and 300 g/ha PND EC + 4 mg/l GA3@HLDP (1:3.548) were applied post-emergence, and distilled water was used as the control. In addition, to differentiate the effect of the nanocarrier from that of the active ingredient, two additional control groups were included: HLDP only, in which rice seedlings were treated with HLDP at a concentration corresponding to that in the GA3@HLDP treatment; and PND EC + HLDP, in which seedlings were co-treated with PND EC and HLDP. Finally, plant height, root length, and fresh weight were measured 14 days after treatment and calculated according to Equations 3, 4, and 5: A, B, and C represent the plant height, root length, and fresh weight after treatment with the different formulations, respectively. A0, B0, and C0 represent the plant height, root length, and fresh weight after treatment with distilled water, respectively. Herbicidal activity assay of PND co-applied with GA3@HLDP Each pot was sown with 10 barnyard grass seeds along a predefined track. The following formulations were applied post-emergence: 300 g/ha PND EC, 300 g/ha PND EC + 4 mg/l GA3, and 300 g/ha PND EC + 4 mg/l GA3@HLDP (mass ratio of 1:3.548). Distilled water was used as the control. After treatment, the seedlings were air-dried and maintained in a greenhouse under the following conditions: temperature, 28°C ± 1°C; humidity, 80%; light intensity, 1,100 lux; and a 15-h photoperiod. The herbicidal activity of the above formulations was evaluated 14 days after treatment using a visual percentage scale. Each experiment was replicated three times. A value of 0% indicated no damage, and a value of 100% represented complete destruction of the aboveground biomass. In addition, the fresh weight control effect was calculated according to Equation 6: C and T represent the fresh weights of barnyard grass in the control and treatment groups, respectively. Transcriptomic and metabolomic analysis of the synergistic mechanism of GA3@HLDP Rice seedlings were treated with either 300 g/ha PND EC + 4 mg/l GA3 or 300 g/ha PND EC + 4 mg/l GA3@HLDP as described above. Rice seedlings (200 mg) were harvested for transcriptomic and metabolomic profiling 7 days after treatment. For RNA sequencing analysis, total RNA was extracted using the TIANGEN DP411 kit (TIANGEN Biotech, China), with three biological replicates prepared per treatment. Following library construction and quality assessment, sequencing was performed on an Illumina NovaSeq 6000 platform (Illumina, USA). Adapter sequences and low-quality bases (Q < 20) were trimmed from the raw reads using Picard v1.41 and SAMtools v0.1.18. SNP calling was conducted with GATK2 v3.2, and the resulting clean reads were aligned to the rice reference genome (ASM90020540v1). Gene expression was quantified using fragments per kilobase of exon model per million mapped fragments (FPKM) normalization. Differential expression analysis was performed with DESeq2 using thresholds of a false discovery rate < 0.01 and |log2fold change| ≥ 1. KEGG pathway enrichment was carried out using the KEGG database. The entire bioinformatic analysis was conducted on the BMKCloud platform (http://www.biocloud.net). After metabolomic profiling, integrated transcriptomic and metabolomic analysis was performed on the seedling samples. Gene expression verification Based on the transcriptomic analysis, 14 DEGs were selected for validation by RT–qPCR. Total RNA was isolated from seedling leaves using TRIzol reagent (TIANGEN, China). Next, 1 μg of extracted RNA was reverse transcribed into first-strand cDNA using the High RNA Tissue Kit (MilliporeSigma, Germany). This cDNA was then used as the template for RT–qPCR analysis. To normalize the amount of cDNA template in each reaction, the glyceraldehyde-3-phosphate dehydrogenase gene (GAPDH; NM_001402381.1) was used as an internal reference (Santos et al., 2018). Gene expression levels were calculated using the 2−ΔΔCt method. The corresponding primers are provided in Supplemental Table 9. Flow cytometry analysis An approximately 2-cm2 piece of rice leaf tissue was excised, minced thoroughly with a blade, and added to 2–3 drops of nucleic acid extraction buffer. Subsequently, 1,000 μl of extraction buffer was added, and the mixture was incubated for 1 min to complete the extraction. Nuclear DNA content was then measured using the CyStain PI Absolute P reagent kit (Sysmex, Japan) according to the manufacturer’s protocol. Ploidy levels in the leaves of rice seedlings treated with GA3 or GA3@HLDP were assessed using a FongCyte C2060 flow cytometer (Beijing Challen Biotechnology, China). Analysis of physiological and biochemical parameters in rice seedlings treated with GA3@HLDP The activities of GST, SOD, peroxidase (POD), catalase (CAT), PAL, 4CL, cinnamate 4-hydroxylase (C4H), GS, NADH-dependent glutamate dehydrogenase (NADH-GDH), ferredoxin-dependent glutamate synthase (Fd-GOGAT), ribulose-1,5-bisphosphate carboxylase/oxygenase (Rubisco), and phosphoenolpyruvate carboxylase (PEPC), as well as the contents of malondialdehyde (MDA), SS, and chlorophyll, were measured in rice seedlings 7 days after treatment with the different formulations described above. The treatments included deionized water (control), 300 g/ha PND EC, 300 g/ha PND EC + 4 mg/l GA3, and 300 g/ha PND EC + 4 mg/l GA3@HLDP. Leaf tissue (0.10 g) from each treatment was homogenized on ice using a high-speed homogenizer (Hunan Xiang Yi Laboratory Instrument Development, China). The homogenate was prepared in the appropriate extraction buffer and centrifuged to collect the supernatant for enzymatic assays. All enzyme activities were determined using commercial ELISA kits (Suzhou Grace Biotechnology, China) according to the manufacturer’s protocols, with three biological replicates per treatment. Safety evaluation of GA3@HLDP in zebrafish, fruit flies, and HSF cells The acute toxicity of GA3 and GA3@HLDP at a mass ratio of 1:3.548 was examined in zebrafish embryos. Experiments were performed to determine the median LC50 values of GA3 and GA3@HLDP at 96 h. The concentrations of GA3 and GA3@HLDP were set at 15, 20, 25, 30, and 35 mg/l. The acute toxicity of GA3@HLDP was further evaluated in adult fruit flies. For the assay, adults were provided with a 1% sucrose solution containing either GA3 or GA3@HLDP at a GA3 mass ratio of 1:3.548, whereas a group receiving ddH2O served as the control. Mortality was recorded 96 h after the start of oral feeding. Each treatment group consisted of 20 adults, and the experiment was performed in triplicate. The viability of HSF cells incubated with GA3 or GA3@HLDP at a GA3 mass ratio of 1:3.548 was evaluated using the CCK-8 assay. Cell viability was calculated according to Equation 7: Soil microbial community analysis after application of GA3@HLDP Rhizosphere soil samples harvested 7 days after application of 4 mg/l GA3 or 4 mg/l GA3@HLDP were subjected to bacterial DNA extraction. The extracted DNA was further purified, and its quality and quantity were evaluated by agarose gel electrophoresis. The V3–V4 region of the bacterial 16S rRNA gene was amplified by PCR using primers 338F (5′-ACTCCTACGGGAGGCAGCA-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3′). Paired-end sequencing was performed on the Illumina MiSeq platform using the MiSeq Kit v3 with a 2 × 250-bp read length. To evaluate similarities among rhizosphere bacterial communities in the different treatments, non-metric multidimensional scaling analysis was performed. Statistical analysis For comparisons between two groups, t-tests were performed using IBM SPSS Statistics 22.0 (IBM, USA), with statistical significance denoted as ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.001. For comparisons among multiple groups, one-way ANOVA followed by Tukey’s honestly significant difference (HSD) post hoc test was performed using IBM SPSS Statistics 22.0 (IBM, USA). In these analyses, significant differences among groups (p < 0.05), are indicated by different letters. Data and code availability The transcriptomic data (BioProject: PRJNA1476714) and microbiome data (BioProject: PRJNA1476723) generated in this study have been deposited in the National Center for Biotechnology Information (https://www.ncbi.nlm.nih.gov/). Funding This work was supported by the National Key Research and Development Program of China (2024YFC2607600), the Yuelushan Laboratory Breeding Program (YLS-2025-ZY01003), the National Natural Science Foundation of China (32372631), the Science and Technology Innovation Program of Hunan Province (2023RC3208), and the 2115 Talent Development Program of China Agricultural University. Acknowledgments We gratefully acknowledge Shenzhen Shuli Tech Co., Ltd. for technical assistance with MD simulations. We also thank eceshi (www.eceshi.com) for SEM, TEM, FTIR, and TGA analyses. The authors declare no competing interests. Author contributions S.Y., L.B., X.D., and H.Q. designed the experiments. X.D. performed most of the experiments. H.Q. performed the plant transcriptomic analysis and ITC experiments. Y.Z. and J.S. provided material resources. S.Y., X.D., and H.Q. wrote the manuscript. All authors edited the manuscript. Published: June 12, 2026 Footnotes Supplemental information is available at Plant Communications Online. Contributor Information Lianyang Bai, Email: lybai@hunaas.cn. Shuo Yan, Email: yanshuo2011@foxmail.com. Supplemental information Document S1. Figures S1–S15, Tables S1–S9, and supplemental methods Document S2. Article plus supplemental information References Associated Data Supplementary Materials Supplemental Video 1. Dynamic process of GA3@HLDP penetrating cell membran Document S1. Figures S1–S15, Tables S1–S9, and supplemental methods Document S2. Article plus supplemental information Data Availability Statement The transcriptomic data (BioProject: PRJNA1476714) and microbiome data (BioProject: PRJNA1476723) generated in this study have been deposited in the National Center for Biotechnology Information (https://www.ncbi.nlm.nih.gov/).