Discovery of novel structures (Agrematch innovative approach)
The Agresense model generation platform enables specific functional modules to be developed by training with relevant positive and negative datasets. In most cases, it takes multiple training iterations that sometimes require new data generation to create a robust predictive module. Once trained, the predictive module can be used for the desired application where every use contributes to the accuracy of the prediction. Each model has different characteristics which depend on the complexity of the question and the domain it searches; each model run may take from milliseconds with a single computer to days with thousands of CPUs activated in the cloud.
A database of ~ 1.2B compounds was screened to create a library of 50 compounds, predicted by Agresense (Agrematch Artificial Intelligence system) to have herbicidal activity, based on a unique proprietary molecular representation of the screened compounds. The iterative process included testing the compounds identified by Agresense for their herbicidal activity in the lab and feeding back the results into the Agresense system to optimize it and to create additional iterations of compound libraries. These compounds were then analyzed by Agrematch’s MoA classifier algorithm to predict whether these compounds have a new MoA or belong to an existing HRAC class.
Plant material and growth
For herbicidal activity assays, Palmer amaranth (Amaranthus palmeri), Xanthium strumarium, Datura ferox, Solanum nigrum and Amaranthus blitoides were planted in pots with a potting mix and grown in either growth-chamber or Net-house. All plants were watered as needed.
Cucumber seedlings (cultivar Straight eight) were planted in trays with promix and grown in the greenhouse for up to 3 weeks to collect fresh cotyledons. These plants were discarded once the 2 true-leaf started to emerge. Palmer amaranth plants that were either sensitive (S) or resistant (R) to PPO inhibitors due to a glycine210 deletion 31–33 were grown in the greenhouse for 4 weeks under similar conditions. All plants were watered as needed. The use of plants in the present study complied with international, national and/or institutional guidelines.
Herbicidal activity
For pre-emergence assays, compounds were dissolved in DMSO and diluted in water to a final concentration of 10–100 mg/L with 2% DMSO. Compounds were sprayed using a VL-SET Paasche Airbrush at a 3,000 L/ha application volume.
For post-emergence assays, compounds were dissolved in DMSO and diluted in water to a final concentration of 25–250 mg/L with 2% DMSO. Break-Thru® S-240 (EVONIK) at a final concentration of 0.05% was added. Compounds were sprayed using a VL-SET Paasche Airbrush at a 1,000 L/ha application volume. Weeds were sprayed at the 4–6 leaf stage.
For post-emergence net-house assay, 300 mg/L of compound AGR001 was dissolved with 900 mg/L xylenes, 73 mg/L ethoxylated castor oil (Kolliphor® RH 40) and 48 mg/L calcium dodecylbenzene sulfonate (Rhodacal® 60be) in water. Break-Thru® S-240 (EVONIK) at a final concentration of 0.05% was added. Weeds were sprayed using a VL-SET Paasche Airbrush at a 1,000 L/ha application volume at the 4–6 leaf stage (2-leaf stage for D. ferox).
Herbicidal activity was assessed and scored 7 days after application by visual inspection of the weeds in comparison to untreated controls. Activity score was in the range of 0 to 100, where 0 represents no herbicidal activity like control weeds and 100 represents the maximal herbicidal activity (i.e., total death of the weed).
Electrolyte leakage
First, time-course experiments were conducted over 40 h to measure the effect of AGR001 or AGR002 on electrolyte leakage from cucumber cotyledons using a modified method of Dayan and Watson.34 For each compound, 36 discs (6 mm diam.) were cut from 7 to 15 day-old cucumber cotyledons and placed in a petri dish. The discs were floating over 5 mL of MES buffer (pH 6.5) with 2% sucrose with 100 µg/mL of either AGR001 or AGR002. This was done in low light intensity to prevent photodynamic damage.
Once the plates were prepared, the initial conductivity (a measure of electrolyte leakage) was measured using a FiveEasy Plus FP30 conductivity meter connected to an InLab 751-4mm microprobe (Mettler Toledo, Columbus, OH 43240). The plates were then kept in the dark at room temp for 16 h. Conductivity was measured after the dark incubation period and then the plates were moved into an LED-30L1 LED high intensity growth chamber (Percival, Perry Iowa 50220). Conductivity was measured 1, 5, 10, 24 h after exposure to light intensity (approx. 1050 µmol/m/s).
Second, dose-response curve experiments were conducted with AGR001 and AGR002 at 1, 3, 10, 30, 100 and 300 µM. Control treatment consisted of DMSO alone to determine the relative potency of these molecules. Conductivity was measured as described above after 16 h dark incubation followed by 24 h exposure to high light intensity.
Third, the effect of AGR001 and AGR002 was tested on biotypes of S and R Palmer amaranth plants. For this experiment, 60 leaf discs from 4-week old Palmer amaranth were used. Concentrations used were 10 µM AGR001 and 100 µM AGR002, conductivity was compared to controls with DMSO alone.
Protoporphyrin IX accumulation
The effects of AGR001 and AGR002 on protoporphyrin IX (proto) levels were measured in cucumber cotyledons exposed to 300 µM of either compound and compared to DMSO control after 16 h dark incubation. Proto extraction and analysis followed a protocol described by Dayan et al.35 Approximately 0.2 g of cotyledonary tissue was ground to a powder in liquid nitrogen and homogenized in 2 mL of extraction solvent (methanol:0.1 M NH4OH, 9:1) and centrifuged at 10,000 × g for 15 min. The supernatant was saved and the pellet rehomogenized in 1 mL of extraction solvent, then centrifuged again at 10,000 × g for 15 min. Supernatants were pooled and then filtered through a 0.2-µm nylon syringe membrane filter before quantification with the LC-MS/MS system. Proto was separated in a biphenyl column (100 by 4.6 mm, 2.6 µm, 40 C) at a flow rate of 0.4 mL min− 1 using a linear gradient of methanol (B) and 10 mM ammonium acetate (A): 0 min, 50% B; 8min, 70% B; 11 min, 90% B; 13 min, 90% B; 13.5 min, 50% B; 17 min, 50% B. The MRM was optimized to 340.10 > 227.95.36 A standard curve generated with serial dilutions of authentic protoporphyrin IX (MilliporeSigma, St. Louis, MO) was used for quantification. Limit of detection (LOD) and limit of quantification (LOQ) for proto were 0.05 ng/µL and 0.15 ng/µL.
Protoporphyrinogen oxidase activity
Protoporphyrinogen oxidase was obtained by expression and purification of the Amaranthus tuberculatus wild type isoform as described by Dayan et al.32 Briefly, the cell line was cultured overnight at 37 C in 250 mL of LB with ampicillin, which was diluted into 1 L of LB with antibiotic and grown for 1 h before induction with 1 mM IPTG. After induction the culture was grown at 25 C for 5 h. Cells were harvested by centrifugation at 2,000Íg and washed with 0.1% NaCl. Cells were lysed by sonication (Model 120 Sonic Dismembrator with a Model CL-18 1/8 inch probe, Thermo Fisher Scientific, Waltham, MA, USA) in 3 x 30 s bursts with 60 s on ice in between in 50 mM sodium phosphate pH 7.5, 500 mM NaCl, 5 mM imidazole, 5% glycerol and 1ug/mL leupeptin. After lysis 1U of benzonase (Millipore Sigma, Burlington, MA, USA) and 1 mM PMSF were added. Debris were removed by centrifugation for 30 min at 2000Íg. Proteins were purified on a HisPur Ni-NTA Spin Column (Thermo Fisher Scientific, Waltham, MA, USA) as per the instructions with elution at 20 mM sodium phosphate, 300 mM sodium chloride 250 mM imidazole, pH 7.4. Protein was desalted on a PD-10 column (GE Healthcare Bio-Sciences Corp., Piscataway, NJ, USA) equilibrated with 20 mM sodium phosphate, pH 7.5, 5 mM MgCl2, 1 mM EDTA and 17% glycerol. Pure PPO which was stored at -80 C until use.
Protoporphyrinogen was prepared by reducing proto with sodium amalgam as described by Jacobs and Jacobs.37 Assays were conducted with 20 µg of protein per replicate as described by Dayan et al.32 with specific modifications for our spectrophotometer as follows: enzymatic activity was measured on an spectrofluorometer (Synergy H1, Agilent Technologies, Wilmington, DE USA ). Excitation and emission wavelengths were set to 395 and 633 nm, respectively. The assay was carried in black microplates (Costar 3915) out under kinetics conditions over 10 min. Enzyme activity was measured based on the linear portion of the curve. Proto amounts were calculated based on a calibration curve of proto standard (Sigma-Aldrich, Inc., St. Louis, MO 68178) at concentrations ranging from 8 nM to 2 µM (Supplemental Fig. 1). AGR001 and AGR002 were tested at concentrations ranging from 0.3 to 333 µM and compared to untreated control that received the same volume of solvent. Statistical analysis was performed using the R software (v 4.1.0). Dose responses were fit using the DRC package.38 Regression curves were imported into Prism 9.1.1 (GraphPad, San Diego, CA 92103).
Docking study
All herbicide structures were downloaded as 3-D.sdf files from PubChem 39. The two experimental compounds were build using a molecular modelling and computational chemistry application (Spartan18, Wavefunction, Inc. Irvine, CA 92612). The bond angles and length were corrected, and the atom energies were calculated by submitting all the molecules to geometric minimization using density function theory calculations (wB97X-D 6–31*). The optimized structures were saved as mol2 files along with their electrostatic charges.
The crystal structure of spinach PPO was obtained from 1sez.40 Prior to use for docking studies, the pdb file was modified to replace the seleniomethionine residues with methionine residues. Also, the atom types of the FAD cofactor were corrected, and the ligand was converted to its oxidized form using Spartan18.
All the herbicides were docked into the catalytic domain of PPO using a Autodock (AutoDock version 4.2, Scripps Institute, San Diego CA, USA).41,42 Additionally, the guanidino group of arginine 98 (atom id = 3782) was designated as important in the interaction between one of the propionate groups (coordinates of this proton are x = -43.390, y = -1.054 and z = 31.750). A grid box was used to delimitate the region of the catalytic domain according to the software. The gridbox dimensions were set to 38Í34Í34 points with a spacing to 0.375. The box was centered on the following coordinates: x = -40, y = -6, and z = 29. PPO was set as a rigid structure. The algorithm was set to generate 100 docking poses and the top clustered was selected as optimal conformation for the docking of each ligand. Interactions between the ligand and the catalytic domain of PPO were identified using the protein–ligand interaction profiler (PLIP).25