Simulations Guide Magnetic COF Adsorbent That Extracts Trace Herbicides from Food

Chemists in China have shown that computer simulations can do the heavy lifting in designing a new material for detecting pesticide residues in food. In a study published in Food Chemistry: X, researchers used density functional theory (DFT) calculations and molecular dynamics simulations to select a covalent organic framework (COF) capable of capturing five phenoxy carboxylic acid (PCA) herbicides before ever synthesizing it. Guided by that molecular-level screening, they then built a magnetic composite adsorbent in just 30 minutes at room temperature and used it to pull trace amounts of the herbicides out of apple and cabbage samples for measurement by high-performance liquid chromatography.

The five target compounds—dicamba, 2,4-dichlorophenoxyacetic acid (2,4-D), 2,4,5-trichlorophenoxyacetic acid (2,4,5-T), 2-(2,4,5-trichlorophenoxy)propionic acid (2,4,5-TP), and 2,4-dichlorophenoxybutyric acid (2,4-dB)—belong to a class of hormone-like agrochemicals used widely as herbicides in agriculture and forestry and as growth regulators on fruits and vegetables. Because of their extensive use and environmental persistence, PCAs readily leach into soil and surface water, contaminating the environment and accumulating as residues in agricultural products. Previous studies have reported that certain PCAs and their derivatives show considerable biotoxicity, with long-term exposure linked in earlier literature to soft tissue sarcomas, endocrine disruption, and fetal developmental abnormalities. Detecting these compounds at trace levels in food is therefore a practical priority—but conventional chromatographic techniques often struggle to find them directly in complex food matrices, making sample pretreatment critical.

The team’s strategy addressed a familiar bottleneck in adsorbent development. Choosing the right COF building units and framework structure has traditionally depended on experimental validation and empirical judgment, meaning that a poor initial choice can force repeated synthesis and optimization cycles. Because the chemical structures of the building units directly determine the interaction sites available at the adsorption interface, the researchers reasoned that a computational pre-screening could identify promising candidates before any material was made. They selected the conventional amine monomer 1,3,5-tris(4-aminophenyl)benzene (TAPB) and the hydroxyl-functionalized aldehyde monomer 2,5-dihydroxyterephthalaldehyde (DHTA) to construct a TAPB-DHTA-COF, a combination chosen because TAPB offers extended π-electron-rich regions that can engage the aromatic rings of PCAs, while DHTA’s hydroxyl groups introduce polar sites capable of hydrogen bonding with the carboxyl and ether oxygen groups of the target molecules.

The DFT calculations, performed on a COF cluster model of one TAPB node and one DHTA linker at the B3LYP-D3(BJ)/6-31+G(d,p) level with subsequent single-point calculations using the SMD water solvation model, delivered a clear verdict. After basis set superposition error correction, all five PCA–COF systems showed negative adsorption energies in water, ranging from −13.55 to −10.19 kcal/mol, indicating energetically favorable adsorption dominated by multiple non-covalent interactions including hydrogen bonding, π–π stacking, dispersion, and electrostatic effects. The COF–2,4-dB system showed the most negative adsorption energy at −13.55 kcal/mol, possibly because its flexible butyric-acid side chain favors conformational adaptation on the surface, while dicamba showed the least negative value at −10.19 kcal/mol, perhaps because its carboxyl group attaches directly to a rigid aromatic ring. The narrow energy range suggested comparable molecular-level affinity toward all five targets. Complementary analyses—electrostatic potential mapping, QTAIM charge analysis, and independent gradient model based on Hirshfeld partition (IGMH) visualization—revealed predominantly green isosurfaces, indicating van der Waals interactions and π–π stacking as the major non-covalent forces, with only small charge transfers of 0.004 to 0.028 electrons from the COF to the PCA molecules, consistent with non-covalent adsorption rather than covalent bond formation.

Molecular dynamics simulations added the dimension that static quantum calculations cannot capture: motion and solvent. Running GROMACS simulations with two parallel COF layers, five PCA molecules, and 4,041 water molecules in a periodic box, the team performed three independent 60 nanosecond production runs at 298 K. The trajectories showed the PCA molecules initially dispersed in the box gradually approaching the COF surface, with the most pronounced adsorption early in the simulation followed by limited rearrangement, indicating stable PCA–COF configurations. Root-mean-square deviations, radii of gyration, and solvent-accessible surface areas all stabilized over time, and the number of hydrogen bonds between COF and PCAs remained within a steady range. After roughly 25 nanoseconds, the center-of-mass distances between the PCA molecules and the COF layers reached stable fluctuations, averaging 7.21 Å for dicamba, 4.99 Å for 2,4-dB, 6.26 Å for 2,4-D, 7.13 Å for 2,4,5-TP, and 4.94 Å for 2,4,5-T. Radial distribution functions showed a well-defined first adsorption layer at approximately 4.0–4.5 Å, with the strongest peak for aromatic carbon–carbon contacts at about 4.35 Å, supporting the dominance of π-related and dispersion interactions, while hydrogen bonding contributed comparatively little under the simulated conditions.

Quantitative energetic analysis from the simulations reinforced the picture. MM-PBSA binding free-energy calculations based on the equilibrated 25–60 nanosecond trajectories yielded an average binding free energy of −14.0 ± 1.58 kcal/mol, confirming energetically favorable association in aqueous solution. Energy decomposition attributed −52.41 ± 4.92 kcal/mol to van der Waals energy and −13.98 ± 1.11 kcal/mol to electrostatic energy, identifying dispersion-type interactions as the dominant favorable contribution with electrostatics providing additional stabilization. With the theoretical case established, the researchers moved to fabrication. They first prepared carbon-encapsulated magnetic nanoparticles (Fe3O4@C) hydrothermally, coated them with polydopamine to provide amino- and catechol-rich anchoring sites, and then grew the imine-linked COF shell ultrasonically at room temperature using scandium triflate as a catalyst—a sonochemical route exploiting acoustic cavitation to complete the whole composite in 30 minutes, in contrast to conventional solvothermal COF synthesis that typically requires temperatures above 120 °C and reaction times of two to eight hours.

Characterization confirmed the design had translated into a real material. Transmission electron microscopy showed roughly 10 nm Fe3O4@C spheres retaining their spherical shape after coating, with an organic layer of apparent thickness around 4.7 nm. X-ray diffraction preserved the inverse spinel peaks of Fe3O4 and retained the characteristic (100) reflection of TAPB-DHTA-COF near 2.83°, while FT-IR spectroscopy displayed the 1655 cm−1 imine C=N band expected for the framework. Nitrogen adsorption measurements gave Brunauer–Emmett–Teller specific surface areas of 57.18, 108.73, and 99.63 m²/g for Fe3O4@C, pristine COF, and the composite, respectively—values lower than those of highly crystalline solvothermal COFs, which the authors attribute to the rapid room-temperature synthesis limiting crystal growth, though they note that extraction performance also depends on site accessibility and interaction strength rather than surface area alone. Vibrating sample magnetometry showed superparamagnetic behavior, with saturation magnetization falling from 37.26 emu/g for Fe3O4@C to 15.15 emu/g for the composite, yet the particles still separated from solution within about 30 seconds under an external magnetic field.

Optimized magnetic solid-phase extraction conditions emerged from systematic testing: pH 3, no added salt, ultrapure water as the extraction medium, 10 minutes of contact time, and 100 μL of methanol containing 3% ammonia as the eluent, which worked best because alkaline conditions deprotonate the PCA carboxyl groups and weaken their hydrogen bonding to the COF. Comparative experiments showed the composite outperformed Fe3O4@C and Fe3O4@C@PDA alone for all five analytes, demonstrating that the COF component drove the enrichment. Enrichment factors for the five target PCAs ranged from 114.5 to 127.3, well above the 39.6 to 87.2 observed for three structural analogues—MCPA, phenoxyacetic acid, and clofibric acid—suggesting that chlorine substitution patterns and carboxylic acid side-chain structure shape adsorption behavior. Adsorption isotherms fit the Freundlich model better than the Langmuir model, with R² above 0.98, pointing to multilayer adsorption, while kinetics followed a pseudo-second-order model with R² above 0.999. The material also proved durable: relative standard deviations stayed below 8.33% after one month of ambient storage, immersion in 1 M hydrochloric acid or sodium hydroxide, exposure to five organic solvents, and 30 minutes of ultrasonication, and no obvious loss of extraction efficiency appeared after five reuse cycles.

The analytical performance of the resulting MSPE-HPLC method was strong across the board. All five PCAs showed good linearity from 1 to 1000 ng/mL with correlation coefficients exceeding 0.9994, limits of detection of 0.21–0.43 ng/mL, and limits of quantification of 0.69–1.45 ng/mL. Intra-day precision ranged from 2.39% to 4.42% and inter-day precision from 5.43% to 7.93%, while three independently synthesized batches of the sorbent gave coefficients of variation of 4.73% to 6.80%, indicating reproducible manufacturing. Matrix effects were weak in both apple and cabbage, with absolute values below 10%, allowing accurate quantification with matrix-matched calibration. When applied to real apple and cabbage samples from a local market in Tianjin, none of the five PCAs were detected, suggesting concentrations below the detection limits, but spike-recovery experiments returned recoveries of 91.56% to 108.92% for apple and 92.54% to 106.37% for cabbage, with relative standard deviations all below 5.3%. Compared with previously reported methods, the approach required only 10 mg of adsorbent and a reduced volume of organic solvent, and its detection limits were comparable to or lower than those in the literature. The authors present the work as a demonstration that COF adsorbent development can shift from trial-and-error screening toward molecular-level evaluation before fabrication, saving time and resources while delivering a practical tool for monitoring pesticide residues in food.

Subject of Research: Simulation-guided design of a magnetic covalent organic framework adsorbent for extracting phenoxy acid herbicide residues from fruits and vegetables

Article Title: Mechanism-oriented design of magnetic COF adsorbents for efficient extraction of phenoxy acid herbicides: Insights from DFT calculations and molecular dynamics simulations

Article References: Hou, Y., Bao, Z., Yang, B., Liu, X., Fu, Z., Tian, F., Du, P., Zhao, J., Xie, C., Wang, J., Liu, E., Zhang, Z., & Han, L. (2026). Mechanism-oriented design of magnetic COF adsorbents for efficient extraction of phenoxy acid herbicides: Insights from DFT calculations and molecular dynamics simulations. Food Chemistry: X, 40, Article 104689. https://doi.org/10.1016/j.fochx.2026.104689

Image Credits: AI Generated

DOI: 10.1016/j.fochx.2026.104689

Keywords: covalent organic frameworks, DFT calculations, molecular dynamics simulations, magnetic solid-phase extraction, phenoxy acid herbicides, food safety, pesticide residues, HPLC, sonochemical synthesis, adsorption mechanisms, apple and cabbage samples, Fe3O4@C@PDA@COF