Kichawi Kill™, a Maize Seed–Coating Technology Delivering Virulence-selected Fusarium for Striga Biocontrol

Herbicide resistance and concerns about the toxicity of synthetic herbicides are two of the reasons for developing bioherbicides in high-income countries such as the United States. However, in Kenya in 2007, an unusual set of reasons were presented regarding S. hermonthica, a widespread invasive, parasitic weed across sub-Saharan Africa. Farmers losing 20% to 100% of their crop yield due to this parasitic weed struggled with a range of conditions that prevented their ability to it. These conditions included the lack of synthetic chemical herbicides available at the village level, farmers’ concerns about the potential human toxicity of herbicides due to hand planting, and a lack of available funding for farm inputs. With most farmers working a hectare or less, they were seemingly overlooked by the primary input companies, and few local input distributors existed at the time.

The challenges surrounding striga demanded a new approach—one that was safe, affordable, effective, and, ideally, not committed to the introduction of problematic synthetic chemical herbicides. The use of a host-specific, endemic fungal pathogen for striga control was considered as an alternative. As defined by Gressel (Reference Gressel2024), the four pillars for successful mycoherbicide commercial development are: (1) must have enhanced virulence; (2) must be cost-effectiveness; (3) shelf life must accommodates distribution; and (4) and must be biosafe without toxic impact beyond the target pest. Previous researchers were unable to achieve significant commercial success using endemic bioherbicides. Our hypothesis was that endemic, wild-type host-specific fungi would not be virulent enough to act as an effective bioherbicide. As noted by Ejeta and Gressel (Reference Ejeta and Gressel2007), mortality against your only host is not a good evolutionary strategy. Given this situation, could a host-specific pathogen with higher virulence could be selected so that it could effectively, economically, and sustainably control its target host?

There have been many attempts to use pathogens that produced phytotoxins, and Toothpick’s approach aims to avoid this approach to biocontrol of weeds in that it might be merely replacing one phytotoxin with another. The work of Steinberg (Reference Steinberg1952) interested us, because he demonstrated that essential amino acids were inhibitory to tobacco (Nicotiana tabacum L.) and that this seemed to be the MOA of two common soil-borne bacteria causing the frenching disease of tobacco. This approach appears to have important implications for its broader use in biocontrol of weeds, as most weeds surveyed so far, in fact most plants, are inhibited by at least one essential amino acid at some concentration. Given this knowledge, amino acid analogue selection was used to obtain plant pathogens that excreted select amino acids. These selections resulted in strains of Fusarium oxysporum f. sp. with enhanced virulence (Nzioki et al. Reference Nzioki, Oyosi, Morris, Kaya, Pilgeram, Baker and Sands2016). While striga is inhibited by three amino acids, leucine, tyrosine, and threonine, maize (Zea mays L.) was tolerant to leucine and tyrosine, but sensitive to threonine; therefore, F. oxysporum strains excreting leucine and/or tyrosine were selected. Peacock and Muirhead found that methionine was converted to ethylene by soil microbes, and that it served as a general stimulator of soil seedbank germination (Peacock and Muirhead Reference Peacock and Muirhead1974). For this reason, F. oxysporum strains that were triple excreters of leucine, tyrosine, and methionine were selected to target striga and its soil seedbank.

Delivery of the selected fungal strains to farmers in rural Kenya was the next challenge. Striga seed germination is triggered by hormones emitted by the host-crop seed germination, allowing the hemiparasite to germinate and attach to the crop within 72 h. Our inclination was that a live, fresh inoculum was needed to inhibit the fast-germinating and growing weed. These selected strains of Fusarium hermonthica f. sp. strigae could be grown on toothpicks and distributed to farmers as primary inoculum. The farmers could then grow their own fresh secondary inoculum by placing the toothpicks in cooked rice to cultivate the fungus over 3 d. The farmer could then co-plant a pinch of this new inoculum in each planting hole with their seeds. This approach was successful, improving crop yield by 42% to 56% in 500 paired-plot trials over two seasons and demonstrating similar restorative outcomes in years of trials and farmer reports (Nzioki et al. Reference Nzioki, Oyosi, Morris, Kaya, Pilgeram, Baker and Sands2016; Figure 1).

Figure 1.

(A) Untreated striga-infested field on Kisumu-Busia Road, Busia County, Kenya. (B) Striga-infested field treated with Kichawi Kill™ on Kisumu-Busia Road, Busia County, Kenya. Photos: Geoffrey Wanjala, Farm to Market Alliance, World Food Program.

Through the social enterprise, Toothpick Company Ltd., the product received regulatory approval for commercial use in 2021. However, while effective in the field, the product had some issues, including a 10-d shelf life, risk of contamination, and the high cost for the rice substrate. Growing the fungal strains on a wood powder rather than toothpicks meant the powder could be used as a seed coating. This seed coating demonstrated efficacy in the field and received regulatory approval in June 2023 (Lüth et al. Reference Lüth, Nzioki, Sands Baker and Sands2024). This iteration reduced the price by 60% and increased the shelf life to 3 mo at room temperature and a year in the freezer. This shelf stability allows new distribution channels, including through agrovet shops and distribution companies. In both internal surveys and studies conducted by a large distribution company, farmers are reporting average yield increases of 12.3 to 14.8 90-kg bags ha⁻¹. Overcoming the striga barrier results in economic growth for the household, and because the product is manufactured locally in western Kenya rather than imported, it also boosts the local economy. The details of the obstacles to commercialization of this product are provided by Baker et al. (Reference Baker, Sands and Nzioki2024).

The next steps for the Toothpick Project are to expand across sub-Saharan Africa, starting with Uganda, Ethiopia, and Cameroon. Regulatory protocol for biocontrol registration for commercial use varies from country to country. However, this is a growing area of regulatory development, with harmonized protocols emerging in the East African Community, for example. With this breakthrough in bioherbicide development, and keeping in mind Gressel’s four pillars necessary for successful bioherbicide development, it is worth noting that other species of weed pathogens might also be useful in biocontrol of weeds.

The Toothpick Project is an example of how a scientifically sound bioherbicide technology can be developed to match the needs of farmers. In 2007 in Kenya, the need was urgent and growing, but more recent awareness of threats within the chemical herbicide industry (litigation due to human and environmental toxicity and herbicide resistance of hundreds of weed species) has brought awareness of the need for more sustainable approaches to weed control globally across the range of agricultural sectors. Kichawi Kill™ represents one such host-specific bioherbicide selected for enhanced virulence.

ProFarm Strategy to Overcome the Challenges of Developing Microbial Bioherbicides

In seeking to develop a bioherbicide, ProFarm scientists have found it useful to distinguish between three paradigms, each operating in largely separate niches. Burn-through products include many organic, naturally sourced extracts, acids, soaps, and oils that rapidly compromise cuticular integrity at the point of contact. These products are most often sold in the home and garden markets. Regrowth from plant parts not directly exposed, especially roots, is a problem with these products. A second category of bioherbicides controls weeds by infecting them with a live pathogen; these products are necessarily highly specific in their spectrum of activity, they may operate relatively slowly and over long-term time frames, and they can be highly suited to situations that require targeting an invasive or parasitic weed species without damaging the surrounding flora. Because these pathogen-based products use live organisms, they face certain commercial and supply-chain disadvantages in fitting into the row-crop or large-scale agricultural markets (Duke Reference Duke2024; Duke et al. Reference Duke, Pan, Bajsa-Hirschel and Boyette2022) and are primarily developed for use on public lands by government agencies. By contrast, focusing on the deployment of natural product chemistry produced by plants or microbes can help a bioherbicide transcend some of the limitations imposed when a product relies on the action of live organisms or naturally sourced, burn-through extracts. In the case of microbial bioherbicides, once a fermentation cycle is complete and the desired metabolites have been produced, the microbes themselves can be inactivated, and the dead cells either included or removed from the final formulated product, thus negating the issue of spread of the metabolite-producing microbe to unintended plant species. Perhaps more importantly for this third category of bioherbicides, natural products can operate and be handled as phytotoxins, much like most conventional herbicides.

There are analogous examples of these bioherbicide paradigms in the biofungicide (FRAC 2023) and bioinsecticide (IRAC 2024) sectors, as well as many synthetically produced compounds that have been derived from natural products isolated from plants or microbes. In contrast to the insecticide and fungicide crop protection segments, there are hardly any examples of natural product–based or natural product–derived molecules available for control of weeds, with glufosinate—first isolated from several Streptomyces soil bacteria species—being the prominent exception. After decades of many well-conceived scientific efforts to develop a bioherbicide, it was posited that the persistent dearth of such products in the market may be due as much to technological and commercial challenges as it is to scientific challenges. ProFarm Group, a subsidiary of Bioceres Crop Solutions, is working to overcome some of these challenges in a pair of bioherbicide projects, each based on herbicidal compounds produced by soil-borne microbes. One is a project based on thaxtomin A (Figure 2), a molecule that belongs to Group 29 in the HRAC MOA classification system and is produced by the bacteria Streptomyces acidiscabies, which causes potato scab disease (Loria et al. Reference Loria, Bignell, Moll, Huguet-Tapia, Joshi, Johnson, Seipke and Gibson2008). The other draws upon a pair of molecules isolated from a strain of Burkholderia rinojensis (Bajsa-Hirschel et al. Reference Bajsa-Hirschel, Pan, Pandey, Asolkar, Chittiboyina, Boddy, Machingura and Duke2023; Owens et al. Reference Owens, Bajsa-Hirschel, Duke, Carbonari, Gomes, Asolkar, Boddy and Dayan2020), of which spliceostatin C is the principal active compound (Figure 2).

Figure 2.

Microbially produced herbicidal molecules.

Spliceostatin C interferes with the production of mRNA through alternative splicing by binding to a pair of sites on the spliceosome previously known to be targeted by other phytotoxic compounds but representing a potential novel MOA in the commercial herbicide landscape (Bajsa-Hirschel et al. Reference Bajsa-Hirschel, Pan, Pandey, Asolkar, Chittiboyina, Boddy, Machingura and Duke2023). Many of the genes that plants regulate through alternative splicing are involved in responses to abiotic stresses and often do not overlap with those genes regulated through steady-state gene expression (Martín et al. Reference Martín, Márquez, Mantica, Duque and Irimia2021); it has been suggested that these differences occur consistently across many plant species, underlying the possibility of a wide spectrum of activity. In tests of spliceostatin C in planta, many weed species in the Amaranthaceae and Brassicaceae families show high levels of susceptibility, yet grasses and some other broadleaves appear tolerant. Therefore, the spectrum of activity may not be strictly determined by the molecular target site but may rather also be governed by other factors, such as structural and metabolic plant defense mechanisms or the physiochemical properties of the molecule itself. Approaching the spectrum of activity as the product of the complex interplay of these factors rather than as a static property of one molecule may allow for the spectrum to be altered through formulation additives and other means. This is an example of how understanding the scientific reasons underlying a potential commercial limitation for a bioherbicide (in the form of a narrower than desired spectrum of activity) can point the way toward potential technological solutions.

Efforts to develop a microbial bioherbicide may face additional technical and commercial obstacles beyond the perennial challenges in formulating for stability and optimized activity of sometimes complex fermentation products (Table 1). Reaching the desired titer of actives produced through the fermentation process can itself be a major technical challenge particular to microbial production systems. Secondary metabolites can be costly for microbes to assemble, so it is unsurprising to find that internal feedback mechanisms may be in place to restrict the levels of production, even under the optimum growth conditions (Ruiz-Villafán et al. Reference Ruiz-Villafán, Cruz-Bautista, Manzo-Ruiz, Passari, Villarreal-Gómez, Rodríguez-Sanoja and Sánchez2022). Suboptimal levels of production may in turn dictate the need for downstream concentration steps to reach quantities of active compound(s) commensurate with expected field application rates; depending on the degree of concentration and the technology involved, the implications for cost of goods can be significant. For example, the amount of thaxtomin A that can be produced with the wild-type S. acidiscabies strain is limited to milligrams per liter, necessitating significant investments in concentration of the active molecule upon completion of the fermentation process. This in turn drove up the cost of production prohibitively. In spite of having been registered with the USEPA and possessing desirable attributes like pre- and postemergence activity, stability under a wide range of storage conditions, and an uncommon MOA (King et al. Reference King, Lawrence and Gray2001), the low production titer of thaxtomin A in the fermentation in wild-type strains of Streptomyces has impeded the development of a marketable herbicide product. However, subsequent development of higher-yielding strains, for example, via mutagenesis techniques, has led to increases in thaxtomin A production by orders of magnitude, allowing for the development of a much more cost-effective production process. In this example, the largest barrier for development and commercialization of the bioherbicide was not in the inherent herbicidal properties of the active molecule, but rather in the technical process through which that molecule could be produced. Experiences such as these suggest that efforts to overcome the technical challenges in developing a microbial bioherbicide will generally be more resource-demanding and interdisciplinary than efforts to discover and characterize the activity of the herbicidal molecule itself.

Table 1.

Nature of some of the challenges encountered in developing a bioherbicide

Ultimately, the most significant obstacles to the development of a widely utilized bioherbicide may be commercial, hinging on the degree to which users are willing to modify their expectations. While all pesticides must meet certain thresholds of efficacy, cost, stability, and art-of-use criteria if they are to successfully penetrate the market, the expectations surrounding herbicides seem to be particularly elevated. For the past few decades, much of the herbicide market has been dominated by highly active broad-spectrum products sprayed over crops with engineered herbicide-resistance traits. As cases of resistance of weeds to one or another herbicide MOA have emerged, numerous traits have been “stacked” into the crops to enable them to tolerate application of herbicide products that mix multiple highly effective MOAs. In this milieu, growers routinely expect near 100% weed control, a standard of performance that is often not imposed on fungicides or insecticides. A newly introduced bioherbicide will likely be compared with such premixes of multiple active molecules, making it difficult to gain traction. While some segments of the market will find inherent value in the perceived environmental and human safety of stand-alone bioherbicides, this should not preclude herbicidal actives produced by microbes from being incorporated into premixes, paired with traits for tolerance in selected crops, or serving as a springboard for the synthesis of chemical analogues. In the long run, external factors like tightening government regulations, increasing public demand for perceived environmentally friendly growing practices, and cascading incidences of herbicide resistance may all contribute to gradually generating more favorable conditions for bioherbicides to be sought out and accepted by the market.

INBIOAR Strategy to Develop Plant Extracts to Kill Weeds

Plant–plant interactions offer a unique opportunity to study the biochemical interactions between plant species under field conditions. As plants are sessile organisms that cannot move from stressful situations, they have developed a great number of physical and chemical defenses to prevent the invasion of their space by insects, diseases, and other plants (Andersen et al. Reference Andersen, Ali, Byamukama, Yen and Nepal2018; Callaway and Aschehoug Reference Callaway and Aschehoug2000; Hierro and Callaway Reference Hierro and Callaway2021; Jones et al. Reference Jones, Felton and Tumlinson2022; Kaur et al. Reference Kaur, Samota, Choudhary, Choudhary, Pandey, Sharma and Thakur2022; Yactayo-Chang et al. Reference Yactayo-Chang, Tang, Mendoza, Christensen and Block2020).

INBIOAR scientists have often observed a group of individual plants of one species, such as Prosopis alba for example, that grow together without plants of other species growing nearby or among them. Such intraspecies groupings somehow prevent the invasion of the same place by other plant species. “Monocultures” of the same plant species are also commonly found in large areas. How do such groupings occur?

The ability of a plant species to release a chemical that is toxic to other plant species competing for the same space may explain such vegetation groupings. Seeds from the species producing the chemical can germinate and grow, whereas other species will grow weakly or die. As a result, only one plant species will grow in an area. This phenomenon, known as allelopathy, refers to the effects of one plant species on another plant species through the release of chemical compounds called allelochemicals (Rice Reference Rice, Cooper-Driver, Swain and Conn1985). Different parts of a plant can deliver these compounds to the surrounding environment by leaching, root exudation, volatilization, organic matter decomposition, and other processes in both natural and agricultural systems.

Allelochemicals are key components of plant defenses against herbivory, microbial plant diseases, or other potentially competitive plants. Anticipating what plant species will be more successful through allelopathy is difficult, although observations of vegetation clustering, as described above, provide a rationale for collecting specific plant species for studies to search for novel phytotoxins (allelochemicals) that might be useful in weed management.

A Strategy to Survive Globally

Toothpickweed [Ammi visnaga (L.) Lam.] is a weed distributed worldwide. Originally, the species was studied for its medicinal properties. Its fruits have been described in pharmacopoeias as an antispasmodic, muscle relaxant, and vasodilator. Other uses in traditional medicine include treatment of mild angina symptoms, supportive treatment of mild obstruction of the respiratory tract in asthma or spastic bronchitis, and postoperative treatment of conditions associated with the presence of urinary stones (Gautam et al. Reference Gautam, Saklani and Jachak2007). This herb has also been used as a treatment for gastrointestinal cramps; as a diuretic; and as a treatment for vitiligo, diabetes, and kidney stones. Also, different extracts of this plant and their major constituents have antibacterial and antioxidant activities and prevent renal crystal deposition and cell damage caused by oxalate (Khalil et al. Reference Khalil, Bishr, Desouky and Salama2020). Thus, A. visnaga is rich in biologically active compounds.

This species grows preferentially under high sun exposure in clay soils, which are well drained and quickly desiccated on the surface, in the semiarid superior and subhumid bioclimatic zones (Sellami et al. Reference Sellami, Napolitano, Masullo, Smiti, Piacente and Pizza2013). In some regions, this plant has become an invasive weed of cultivated fields (Zandstra et al. Reference Zandstra, Particka and Masabni2004). Ammi visnaga plants were collected for extraction in a semiarid region of Argentina in an area of 900 by 30 m (Figure 3) completely covered by A. visnaga. This species is a successful invasive species. Native to the Mediterranean region of Europe, Asia, and North Africa, it is now found in Argentina, Brazil, Chile, Uruguay, North America, Southwest Asia, and some Atlantic islands.

Figure 3.

Ammi visnaga at the beginning of the colonization process in a natural area of Argentina.

Bioassay-guided isolation of phytotoxins from A. visnaga found that khelin and visnagin were the most active compounds (Travaini et al. Reference Travaini, Sosa, Ceccarelli, Walter, Cantrell, Carrillo, Dayan, Meepagala and Duke2016). These metabolites were of previous interest to the pharmaceutical sector. The bioassay-guided isolation of the active ingredient(s) consisted of seed germination and plant growth bioassays in a laboratory. Postemergence herbicidal effects of A. visnaga extracts and the two isolated compounds were determined in greenhouse trials. Visnagin was the most active, with contact postemergence herbicidal activity on velvetleaf (Abutilon theophrasti Medik.) and large crabgrass [Digitaria sanguinalis (L.) Scop.] at 2 kg ai ha⁻¹. Moreover, its effect at 4 kg ai ha⁻¹ was comparable to the commercial bioherbicide pelargonic acid at the same rate. The initial formulation included only Tween^® 20 (polisorbate 20) in water. Efforts are being made to improve the formulation of these active ingredients.

Khellin and visnagin may have more than one MOA. Their activity is not light dependent and involves effects on membrane stability, cell division, and cell viability in leaves and roots. These effects may not be related. Both compounds also reduce photosynthetic efficiency through indirect effects and induce oxidative damage under high light intensity.

Additionally, analogues of khellin and visnagin were synthesized, and their herbicidal activities were examined (Cantrell et al. Reference Cantrell, Travaini, Bajsa-Hirschel, Svendsen, Reichley, Sosa, Kim, Tamang, Meepagala and Duke2023). Acetate analogues of khellin and visnagin had more activity on lesser duckweed (Lemna paucicostata Hegelm.; syn. Lemna aequinoctalis Welw.) than visnagin, and the O-demethyl butylated visnagin analogue was the most active compound with an IC₅₀ of 47.2 µM. In additional herbicide bioassays, visnagin and the O-demethyl butylated visnagin analogue were the most active compounds and reduced the germination and early growth of lettuce (Lactuca sativa L.) and ryegrass species (Lolium spp.), translocated from the solution through the plant shoots of white mustard (Sinapis alba L.) and corn, and reduced the canopy cover of foxtail millet [Setaria italica (L.) P. Beauv.] and S. alba after postemergence application.

Progressive Effect of a Plant Extract

Most of the plant-based bioherbicides produce burning or contact effects. A weakness of “burndown/contact” herbicides, whether bioherbicides or synthetic, is that treated plants tend to regrow from meristems that did not come in contact with the herbicide. Therefore, INBOAR’s goal was to identify a phytotoxin with a systemic MOA.

a plant extract with a non-immediate and more systemic herbicidal effect was found by screening the flora in semiarid regions. This aqueous extract kills the weeds progressively after postemergence application (Figure 4). Within 10 d of plants being sprayed with an aqueous solution of the extract (20 mg ml⁻¹), high control (80% to 100%) of different species, with activity on both dicotyledonous and monocotyledonous species, was observed. Sorghum sp., S. italica, Sinapis sp., morningglory (Ipomoea sp.), Palmer amaranth (Amaranthus palmeri S. Watson), and soybean [Glycine max (L.) Merr.] were among the most sensitive species, while lettuce, corn, oat (Avena fatua L.), clover (Trifolium sp.), and Lolium sp. were affected moderately or not at all.

Figure 4.

A plant extract that kills weeds slowly and systemically when applied postemergence. Effects are 10 d after treatment.

The herbicidal efficacy was tested at different doses, and some weeds—green foxtail [Setaria viridis (L.) P. Beauv.], redroot pigweed (Amaranthus retroflexus L.), and wild poinsettia (Euphorbia heterophylla L.)—were still controlled (80% to 100%) at 15 and 7.5 kg ha⁻¹. The effect was improved with the use of different adjuvants. The active ingredient(s) of this extract have not been identified yet. However, the efficacy on weeds is consistent, and the purification of and identification of the active ingredient is in progress.

MicroMGx Strategy to Accelerate Phytotoxic Natural Product Discovery through Metabologenomics

Metabologenomics is a powerful tool driving innovation in applied science, particularly in the field of natural product discovery. By seamlessly integrating genome sequencing, genome mining, and state-of-the-art metabolomics techniques like mass spectrometry and nuclear magnetic resonance, this methodology empowers researchers to uncover novel bioactive compounds with real-world applications (Doroghazi et al. Reference Doroghazi, Albright, Goering, Ju, Haines, Tchalukov, Labeda, Kelleher and Metcalf2014; Goering et al. Reference Goering, McClure, Doroghazi, Albright, Haverland, Zhang, Ju, Thomson, Metcalf and Kelleher2016). At its core, metabologenomics offers a practical approach to navigating the complex landscape of microbial secondary metabolite biodiversity. Through systematic analysis of BGCs and metabolomics data, researchers can efficiently identify promising candidates for further investigation (Yan et al. Reference Yan, Liu, Zang, Yuan, Bat-Erdene, Nguyen, Gan, Zhou, Jacobsen and Tang2018). This targeted approach allows for the rapid identification and characterization of natural products with potential pharmaceutical, agricultural, or industrial uses. One of the key strengths of metabologenomics lies in its ability to prioritize the discovery of new secondary metabolites. By leveraging advanced computational pipelines and bioinformatics tools, researchers can pinpoint BGCs associated with previously unexplored chemical scaffolds. This targeted screening approach significantly accelerates the pace of natural product discovery, facilitating the development of novel therapeutics, agrochemicals, and biotechnological applications. Furthermore, metabologenomics enables researchers to tailor their searches for specific classes of natural products based on predefined criteria. Whether targeting polyketides, non-ribosomal peptides, phosphonates, or other specialized metabolites, this methodology provides a customizable framework for targeted screening and discovery. By harnessing the power of genomics and metabolomics, MicroMGx can unlock the full potential of microbial diversity and harness nature’s vast biochemical repertoire for practical applications.

Exploring the Potential of Phosphonate Natural Products in Agriculture

Phosphonate natural products represent a promising group of compounds for biocide discovery due to their potent bioactivities. Despite their significance (Ju et al. Reference Ju, Doroghazi and Metcalf2014; Yu et al. Reference Yu, Doroghazi, Janga, Zhang, Circello, Griffin, Labeda and Metcalf2013), this class of compounds remains relatively underexplored compared with other classes of natural products. The evolutionary mechanism underlying phosphonate inhibition highlights the strategy of molecular mimicry through the structural resemblance to phosphate esters and carboxylic acid metabolites (Figure 5).

Figure 5.

Structure and bioactivity of naturally occurring phosphonates and their molecular mimics. Bioactive phosphonates with their respective enzyme targets are shown with a light blue background. The enzyme’s native substrates are shown in a light red background. Phosphorus-carbon bonds are shown as bold red lines on the molecule, and commercially available compounds are underlined. Fosfomycin, clinically sold as Monurol^®, is used to treat difficult urinary tract infections, and phosphinothricin (also known as glufosinate) is the active ingredient for the broad-spectrum commercial herbicide sold under several trade names.

The chemically inert nature of the stable phosphorus–carbon bond in phosphonates confers their remarkable specificity and potency as enzyme inhibitors. By competing with their structural analogues for enzyme binding, phosphonates disrupt normal catalytic functions, offering a targeted approach to enzyme inhibition. Given the ubiquitous roles of phosphate esters and carboxylic acids in biological processes, the spectrum of potential cellular targets for phosphonate inhibitors is vast. In practical applications, phosphonates have demonstrated their efficacy as active ingredients in widely used herbicides such as glyphosate and glufosinate, showcasing their practical utility in agriculture.

Plant-associated Enterobacteria as Sources for Herbicidal Molecules

The escalating problem of pesticide resistance in agriculture poses a significant threat, leading to substantial losses in global crop yields, estimated at up to 50% (Gould et al. Reference Gould, Brown and Kuzma2018). Efficient and targeted discovery efforts rely heavily on access to diverse and reliable sources of new compounds. Host-associated enterobacteria emerge as promising candidates, offering a rich reservoir of bioactive natural products (Adnani et al. Reference Adnani, Rajski and Bugni2017; Pidot et al. Reference Pidot, Coyne, Kloss and Hertweck2014). These microorganisms boast attributes conducive to laboratory cultivation and genetic manipulation, leveraging well-established techniques developed for their close relative, Escherichia coli. Utilizing straightforward genetic engineering methods, such as promoter exchange (depicted in Figure 6, right), enables the native expression of BGCs. This approach simplifies the process of accessing and studying the bioactive potential of these compounds, circumventing the need for complex heterologous expression systems that frequently encounter inefficiencies and failures. Given the urgent demand for new phytotoxic agents and the well-established bioactivity of phosphonates, adopting a metabologenomics approach centered on plant-associated enterobacteria emerges as a viable strategy for pinpointing herbicides with potentially new MOAs (Figure 6).

Figure 6.

Discovery of novel herbicides from plant-associated enterobacteria (left). Novel phosphonate biosynthetic gene clusters (PepM BGCs) are identified from plant-associated enterobacteria through the metabologenomics platform (top, center). Native host expression of the PepM BGC is achieved by introduction of an inducible promoter, Ptac, through a simple promoter exchange method via homologous recombination (right). A bacterial recombinant with the exchanged promoter is then selected for using the antibiotic resistance gene (ARG). Spent media or purified compounds are then used for herbicide bioassays (bottom, center).

The application of metabologenomics to explore the bioactive potential of phosphonates from plant-associated enterobacteria represents a promising approach to addressing the urgent need for novel herbicides. This applied discovery approach holds promise for revolutionizing agricultural practices and mitigating the impact of pesticide resistance on global crop production.

Discovery of Pantaphos from a Plant-associated Enterobacterial Pathogen

Pantaphos, a novel phosphonate compound originating from strains of Pantoea ananatis, a common plant pathogen, was recently discovered. Pantaphos has remarkable phytotoxic properties, including inducing the characteristic lesions associated with onion center rot (Polidore et al. Reference Polidore, Furiassi, Hergenrother and Metcalf2021). Our investigations revealed potency that is comparable to widely used herbicides like glyphosate and phosphinothricin. With P. ananatis showing a wide host range and the prevalence of its BGC among pathogenic strains, the potential bioactivity of pantaphos across various plant species is noteworthy. To assess its efficacy, an efficient, scalable production method for pantaphos production was devised using the native host. After purifying the compound from spent media, it was formulated for greenhouse testing. The resulting product, MGX-1001, demonstrated significant growth inhibition in numerous crops and weed species, particularly broadleaf weeds (Figure 7).

Figure 7.

Growth impact assessment among common crop and weed species treated with formulated pantaphos. Data were compiled from five independent greenhouse trials with crop species in blue and weed species in red. The asterisks denote weed species with known glyphosate resistance. Horizontal axis is measured impact assessment based on overall growth yield compared with an untreated sample, and vertical axis distance is arbitrary. The growth phenotype of waterhemp [Amaranthus tuberculatus (Moq.) Sauer] after treatment with pantaphos is shown in the upper left inset. Rice (Oryza sativa L.) showed highly variable growth impact that was found to be dependent on the adjuvant used in the pantaphos formulation.

Intriguingly, MGX-1001 had minimal impact on key food crops like corn, oat, wheat (Triticum spp.), and barley (Hordeum vulgare L.), suggesting practical applicability in agricultural fields. Moreover, our findings indicate that MGX-1001 effectively suppresses the growth of herbicide-resistant weeds, hinting at a potential new MOA. While further investigations are warranted, initial experiments suggest a unique mechanism of action targeting broadleaf weeds specifically. Notably, our ongoing research includes the identification of a resistance allele that can be expressed in plants, offering insights into potential genetic modifications in plants for enhanced herbicide tolerance.

With its promising properties, including a suspected novel MOA and compatibility with broadleaf weed control, pantaphos emerges as a next-generation herbicide with significant agricultural implications. Our focus now shifts toward refining formulation techniques and advancing commercialization efforts, leveraging pantaphos’s natural origin and its anticipated impact on weed management. Also, the exploration of phosphonate natural products presents opportunities for scientists seeking innovative solutions in agriculture and beyond (Manghi et al. Reference Manghi, Masiol, Calzavara, Graziano, Peruzzi and Pavoni2021).

GALAXY, a High-Content Imaging Platform Enabling Novel Herbicide Discovery

Moa Technology (www.moa-technology.com) is a start-up company created in 2017 by a group of researchers at the University of Oxford (UK) to tackle the problem of weed resistance in crop production. Two main strategies have traditionally been deployed by the industry to discover herbicides with novel MOAs: phenotypic and in vitro targeted approaches (Hachisu Reference Hachisu2021). Moa Technology has developed an innovative and systematic large-scale in planta herbicide discovery platform that enables high-throughput discovery of novel classes of MOAs with unique and potentially powerful herbicide activity an order of magnitude faster than traditional methods.

More than three-quarters of a million synthetic and natural product compounds have been screened through GALAXY. Glasshouse performance in Moa Technology facilities for compounds displaying novelty has been extensively profiled and led to the identification of more than 60 moaNOVEL chemical areas (Figure 8). moaNOVEL areas are chemical areas empirically discovered to likely act with a new MOA by Moa Technology’s proprietary miniaturized plant-led high-content phenotyping assay used to screen an enormous diversity of natural and synthetic chemistries on microscopic whole plants and rapidly identify an abundance of new, never commercialized herbicide MOAs at unprecedented scale, speed, and cost.

Figure 8.

Graphical representation of moaNOVEL families identified by the GALAXY platform. Unique moaNOVEL families with more than 50 representative molecules are denoted with a bracket (>50). The number of each moaNOVEL molecule active in weeds is highlighted in green. moaNOVEL families with glasshouse activities are highlighted in yellow.

GALAXY answers the question “Is it a herbicide, and does it work in a new way?” by comparing the extensively profiled dose-dependent plant symptoms associated with tested molecules. The GALAXY platform discriminates between hundreds of molecules belonging to known MOA classes and novel molecules eliciting the unique, never-observed plant symptomologies that are specifically associated with the urgently needed moaNOVEL areas. Using high-content imaging, GALAXY characterizes nearly 100 individual parameters for each plant exposed in each well to unique compounds sourced for their diversity from vast compound libraries. A proprietary digital segmentation and categorization process for each plant then ensures the classification of every molecule in known or moaNOVEL categories, paving the way for the elucidation of their MOAs.

Once a novel molecule has been identified by GALAXY, the next step is to understand the MOA, the molecular target, and the pathways modulated by each molecule. To ensure a successful outcome, a combined constellation of powerful complementary screens is simultaneously mobilized. In this way, the new MOA, molecular target or pathway can be qualified, thus answering how the herbicide works and evaluating its probable safety and potential commercial opportunities. Triggered early to de-risk the next-generation herbicide discovery effort, the process delivers results in months rather than years, accelerating predictions of safety and optimization into farm-ready herbicides.

Moa Technology is advancing a growing pipeline of novel synthetic compounds and bioherbicides toward field trials at pace (Figure 9). Leveraging the suitability of GALAXY for testing minute amounts of compounds on whole plants, Moa Technology is shortening even further the path to market by screening natural compound libraries. Natural product herbicides have the potential to deliver commercial success more rapidly than synthetic counterparts, including natural products acting via known MOAs. Natural extract libraries derived from plants, fungi, bacteria, or marine organisms have already been successfully profiled, and selected candidates are advancing in further glasshouse testing. The speed and power of the approach taken for novel synthetic and natural product compounds means that international field trials in both the Northern and Southern Hemispheres are being initiated in 2024 with multiple lead molecules with new MOAs.

Figure 9.

Weed spectrum evaluation of two examples of molecules being shortlisted for field trials in 2024: (A) molecule with broadleaf spectrum and largely postemergence properties and (B) molecule with largely preemergence properties and broad spectrum. DFF, diflufenican; ALOMY, blackgrass (Alopecurus myosuroides Huds.); AMARE, redroot pigweed (Amaranthus retroflexus L.); ANTAR, chamomile [Anthemis arvensis (Wallr.) DC.]; APESV, common windgrass [Apera spica-venti (L.) P. Beauv.]; CENCY, cornflower (Centaurea cyanus L.); CHEAL, lambsquarters (Chenopodium album L.); ECHCG, barnyardgrass [Echinochloa crus-galli (L.) P. Beauv.] LOLMU, Italian ryegrass (Lolium perenne L. ssp. multiflorum (Lam.) Husnot; PHBPU, tall morningglory (Ipomoea purpurea (L.) Roth); POAAN, annual bluegrass (Poa annua L.); SETIT, foxtail millet [Setaria italica (L.) P. Beauv.]; STEME, common chickweed [Stellaria media (L.) Vill.]; VERPE, birdeye speedwell (Veronica persica Poir.). Blue indicates grass weeds; green indicates broadleaf weeds.

Heurik™ and Swap™—Two Integrated Platforms to Optimize Crop Genes for Herbicide Resistance

CRISPR gene editing is a relatively new technique that enables the introduction of precise mutations in plants (and other organisms) that can include one or several nucleotide changes (Jiang et al. Reference Jiang, Zhou, Bi, Fromm, Yang and Weeks2013). This new tool is opening new opportunities to create crops with novel traits (Scheben et al. Reference Scheben, Wolter, Batley, Puchta and Edwards2017). Moreover, in many countries, gene-edited crops derived from CRISPR technology are not considered genetically modified organisms (GMOs) and can be commercialized as conventional crops (Duarte Sagawa et al. Reference Duarte Sagawa, Assis Rd, Zaini, Abd-Elsalam and Ahmad2024; Sprink et al. Reference Sprink, Wilhelm and Hartung2022). This avoids the expensive and time-consuming regulatory process associated with GMO crops and circumvents current public concerns over GMO foods (Ryan et al. Reference Ryan, Henggeler, Gilbert, Schaul and Swarthout2024).

In plants, mechanisms that provide herbicide resistance can be classified as target site or non−target site, depending on whether the resistance is related to the protein that is inhibited by the herbicide or caused by other factors (for a review, see Gaines et al. [Reference Gaines, Duke, Morran, Rigon, Tranel, Küpper and Dayan2020]). Both approaches have been used to develop herbicide-resistance traits either by engineering genetically modified plants that express genes from other organisms (transgenic or GMO) or by traditional mutagenesis of native genes in crops (Duke Reference Duke2015; Green Reference Green2014). Nonetheless, traditional mutagenesis is a slow process that introduces random mutations that are limited in type. On the other hand, bringing a GMO to the market takes more than 16 yr and costs more than US$100 million (AgbioInvestor 2024). BioHeuris has developed two technology platforms that integrate protein engineering (Heurik™) and gene editing (Swap™) to develop next-generation herbicide-resistant crops in under 6 yr and at a cost 50 times lower.

Protein engineering is based on the modification of protein sequences through substitution, deletion, or insertion of nucleotides in the encoding gene. This process can be used to identify mutations in plant genes that provide target-site or non−target site resistance without affecting the fitness of the plant. BioHeuris developed Heurik™, a high-throughput microbial platform, to discover and measure the level of herbicide resistance in plant genes carrying different mutations (Figure 10). In a few weeks, this platform can mimic experiments that would take hundreds of hectares and years of field trials if done with traditional mutagenesis. Heurik™ uses directed evolution and rational design as the two main strategies for protein engineering. The directed evolution strategy involves generating a library of millions of gene variants and screening for herbicide resistance in engineered microbes. The rational design strategy uses computational models to infer which amino acid changes could provide the desired resistance while retaining the protein activity or function.

Figure 10.

Heurik™ trait-discovery workflow developed to identify candidate mutations in plant genes that provide herbicide resistance.

The identified mutations are then introduced in crops by BioHeuris using Swap™, a proprietary gene editing platform. Swap™ uses different versions of CRISPR enzymes to introduce short deletions or nucleotide substitutions (Figure 11). This platform combines gene editing with in vitro tissue culture of elite lines and speed breeding to generate edited crops in 1 yr without leaving DNA from other species in plant genomes (Ghosh et al. Reference Ghosh, Watson, Gonzalez-Navarro, Ramirez-Gonzalez, Yanes, Mendoza-Suárez, Simmonds, Wells, Rayner, Green, Hafeez, Hayta, Melton, Steed and Sarkar2018).

Figure 11.

Swap™ gene editing workflow allows the creation of mutations in plant genomes to efficiently obtain herbicide-resistant elite varieties. NHEJ, non-homologous end joining; HDR, homology-directed repair; DSB, double-strand break.

A different approach to identifying mutations by protein engineering could be to use mutations previously discovered in herbicide-resistant weeds. However, if the herbicide is new, this source of resistance might not be present in natural populations. Also, empirical evidence suggests that replicating known mutations in target-site enzymes across different plant species might not provide the expected herbicide resistance in all of them. This was shown by introducing equivalent single amino acid changes in target-site enzymes from different crops and measuring not only different levels of resistance but whether some mutations had a positive effect only in certain species. The work was done by scientists from BioHeuris who replaced a microbial gene with the plant version of a target-site enzyme from soybean, sorghum [Sorghum bicolor (L.) Moench], and cotton (Gossypium hirsutum L.) (60% to 70% identity). They then introduced all possible (19) changes in conserved amino acid positions and measured growth inhibition of the microbial “chassis” species at increasing amounts of herbicides (Figure 12).

Figure 12.

Herbicide resistance provided by different amino acid changes in two conserved positions of an orthologous target-site enzyme from soybean (red), sorghum (blue), and cotton (yellow). Changes are indicated with the three-letter code for each amino acid. Values correspond to the resistant/susceptible (R/S) ratio, calculated as GI₅₀ of mutant variant by the GI₅₀ of the wild-type variant.

At least in some cases, predicting and identifying mutations that can provide resistance might be crop specific, calling for the development of high-throughput cost-efficient discovery methods. By combining the described synthetic biology platforms, BioHeuris developed herbicide-resistant sorghum and rice elite varieties and validated them in field trials in only 3 yr. Several countries, including Argentina, Brazil, the United States, Chile, and Colombia, already confirmed that these plants are not GMO. Pre-commercial trials are underway, and BioHeuris expects that it will only take 6 yr to bring these traits to farmers with a cost 50 times lower than a GMO event.

ATTUNE™—A Novel Platform Harnessing Targeted Protein Degradation for Agricultural Use

Novel, non−enzyme inhibition approaches are needed for the expansion of new herbicide MOAs. Inspiration can be drawn from the pharmaceutical industry, especially in the field of targeted protein degradation (TPD). TPD has expanded rapidly over the past two decades, with more than 40 companies exploring this technology. PROTACs, or proteolysis-targeting chimeras, represent the most advanced chemistry now progressing to commercialization (Békés et al. Reference Békés, Langley and Crews2022). Similar to the adoption of CRISPR technology, PROTACs are poised for use in agriculture with an emphasis on creating safe, precise, and environmentally friendly chemistries.

PROTACs emerged in the early 2000s as a groundbreaking concept pioneered by Craig Crews and colleagues at Yale University (Bondeson et al. Reference Bondeson, Mares, Smith, Ko, Campos, Miah, Mulholland, Routly, Buckley, Gustafson, Zinn, Grandi, Shimamura, Bergamini and Faelth-Savitski2015; Zengerle et al. Reference Zengerle, Chan and Ciulli2015). PROTACs are biorationally designed, modular compounds that target specific proteins within cells for degradation by harnessing the endogenous cellular machinery responsible for protein degradation (proteasome). These synthetic molecules consist of three main components: (1) a ligand binding to the protein target of interest (POI), (2) a ligand binding to an E3 ubiquitin ligase enzyme, and (3) a linker that covalently binds those two ligands (Figure 13A).

Figure 13.

(A) Illustration of a proteolysis-targeting chimera (PROTAC) with (1) its target ligand binding the target protein of interest, (2) its ligase-specific ligand to recruit a specific E3 ubiquitin ligase, and (3) its linker that covalently attaches those two ligands. (B) Illustration of the PROTAC mechanism of action (MOA): the PROTAC recruits a specific E3 ligase and a target protein to form a ternary complex, allowing several molecules of ubiquitin to be transferred onto the surface of the target protein. The ubiquitin-tagged protein is then transported to the proteasome for degradation. Figure reproduced with permission from Oerth Bio, LLC.

Upon binding to both the E3 ligase and target protein, a PROTAC-mediated ternary complex forms and facilitates the transfer of ubiquitin onto the surface of the target protein by the E3 ligase. This ubiquitination flags the protein for degradation and recycling by the ubiquitin proteasome system (Figure 13B). Once the ternary complex dissociates, the PROTAC molecule can be released to engage in further rounds of degradation, which is referred to as the “catalytic mechanism.” This groundbreaking catalytic effect enables much lower amounts of compound to achieve the desired efficacy (Bondeson et al. Reference Bondeson, Mares, Smith, Ko, Campos, Miah, Mulholland, Routly, Buckley, Gustafson, Zinn, Grandi, Shimamura, Bergamini and Faelth-Savitski2015). Additionally, PROTACs can typically achieve more sustained and complete efficacy compared with classical inhibitors, as they remove the POI from the cell entirely rather than merely blocking its function (Burslem et al. Reference Burslem, Smith, Lai, Jaime-Figueroa, McQuaid, Bondeson, Toure, Dong, Qian, Wang, Crew, Hines and Crews2018). Importantly, PROTACs do not require a traditional active site to initiate degradation, which opens up the protein landscape to targets previously considered “undruggable” (Liu et al. Reference Liu, Chen, Kaniskan, Xie, Chen, Jin and Wei2021).

While the vast majority of the PROTAC development is strictly focused on human health, Oerth Bio is the lone pioneer in developing PROTACs for agricultural use. Oerth Bio was founded in 2019 as a joint venture between Bayer CropScience and Arvinas. The company was created to develop the next wave of novel crop protection tools for agriculture (Speake et al. Reference Speake, Fourches, Catalano, Hunter, Kallus and Pfeiffer2023). Since its inception, Oerth Bio has developed a proprietary platform called ATTUNE™, which is the first agricultural PROTAC discovery engine. The backbone of ATTUNE™ is Oerth Bio’s E3 ligase discovery and ligand identification pipeline. As E3 ligases are essential for enabling PROTAC activity, Oerth Bio focuses on identifying kingdom- and phylum-specific E3s to dial-in applications specific to crop protection. Once a suitable E3 ligase is identified, two key steps are critical for making it PROTAC ready: (1) confirmation of its degradative abilities with one or several POIs using a dedicated biological assay and (2) identification of a small molecule ligand to bind and recruit that E3 ligase. Once those two key milestones are achieved, the Oerth team pairs the ligase with a compatible protein target for actual PROTAC development. Unlike traditional “spray and pray” or phenotype-driven discovery, PROTACs are rationally designed to ensure the specific degradation of the POI. This design encompasses a unique set of computational and biological tools in combination with a team of synthetic chemists. Once protein degradation is achieved, the PROTAC efficacy can be significantly enhanced through rapid cycles of design, build, and test to afford a high enough level of degradation potency.

With more than 1,000 E3 ligases in plants, there is a tremendous opportunity to develop novel herbicides (Mazzucotelli et al. Reference Mazzucotelli, Belloni, Marone, De Leonardis, Guerra, Di Fonzo, Cattivelli and Mastrangelo2006). PROTACs offer three main advantages over traditional MOAs: (1) improved safety and specificity through tunability, (2) lower dosing with sustained efficacy, and (3) protein target expansion.

The first key feature of PROTACs for herbicide discovery is related to their overall tunability. Because their chemistry is modular, each of the three components of a PROTAC can be individually tailored to dial-in unique properties, including safety and specificity. The first level of tunability is linked with the actual E3 ligase selection. For instance, using a solely plant-specific E3 ligase significantly decreases the likelihood of off-target effects. The second level of tunability is within the E3 ligase and the protein target chemistry. Because active site orthosteric ligands are not strictly necessary for PROTAC activity, a larger chemical space can be utilized for PROTAC design, especially when it comes to ligands with a fairly wide range of binding affinities. Ligands that bind at allosteric sites and other nontraditional surface pockets further increase this chemical flexibility, making the different E3 binder–linker–POI binder combinations nearly infinite. Finally, the PROTAC linker can be further fine-tuned for the safety, selectivity, and delivery of these molecules. The linker is not only essential for the formation of the ternary complex and protein degradation, but also allows for tuning the overall physical chemistry properties of the PROTAC (Atilaw et al. Reference Atilaw, Poongavanam, Svensson Nilsson, Nguyen, Giese, Meibom, Erdelyi and Kihlberg2021). Collectively, these multiple layers of tunability provide a path for a more precise and safe-by-design herbicide.

The second key feature concerns the fact that PROTAC-based herbicides have the potential to reduce the environmental impact for farmers by allowing for lower use rates compared with traditional chemistries. This is a direct consequence of the catalytic mechanism of PROTACs, which enables the degradation of multiple copies of the POI by a single PROTAC molecule (as opposed to traditional small molecule inhibitors possessing a 1:1 stoichiometry with the POI). This amplification effect has the potential to dramatically lower the use rate by 10 to 100 times. Additionally, the catalytic mechanism could provide longer-lasting effects, requiring fewer field applications. Combined with the tunability of these molecules, there is an additional opportunity to replace environmentally unfavorable chemistries.

The last key feature of PROTAC-based herbicides is their promising ability to expand the protein target landscape but also to rescue old herbicide molecular targets with known, effective herbicide inhibitors. Commercial herbicides target a very limited range of weed proteins, characterized by well-defined binding sites. Additionally, many research targets have never been commercialized due to selectivity, potency, and safety concerns. Furthermore, the overuse of existing chemistries has led to evolution of significant resistance in the field. PROTACs offer a variety of avenues to expand and overcome all these issues. First, designing PROTACs against existing commercial protein targets can allow for lower use rates compared with current active ingredients due to the catalytic effect, which can improve safety profiles as well as limit the formation of resistance. Additionally, PROTACs can overcome existing resistance mechanisms, including target-site and non−target site resistance. This is because PROTACs will be active, even with a reduced binding affinity of the POI to current commercial herbicides, and they are unlikely to be metabolized in a similar manner as current herbicides. Second, there is a big opportunity to rejuvenate shelved herbicide targets, including those that have not been marketed because of selectivity and safety concerns. The tunability of PROTACs offers new ways to address these concerns (e.g., targeting an allosteric site or recruiting a plant-specific E3 ligase). Finally, and perhaps most importantly, PROTACs offer the ability to expand into nontraditional POIs such as structural proteins and transcription factors. Therefore, fully untapped pathways and protein targets leading to herbicidal phenotypes could then become accessible using PROTACs, unlocking a treasure trove of commercial opportunities and applications.

The future of PROTACs in agriculture holds immense promise, offering a revolutionary approach to combating weeds, but also pests and crop diseases. This precise targeting not only will enhance efficacy but also reduce off-target effects, making it a safer alternative to traditional chemistries. Beyond crop protection, there is a bigger opportunity for PROTACs to help farmers grow better, healthier, and more resilient crops in the face of a changing climate. Utilizing plant-specific E3 ligases, PROTAC “sprayable traits” can be developed for on-demand use applications, including drought, heat, and salt tolerance. Overall, the PROTAC technology could thus have the potential to disrupt and reinvent the whole crop protection landscape for decades to come.

Projini’s Platform for Discovering New MOA Herbicide Leads Inhibiting Protein–Protein Interactions

Projini is a start-up company founded in 2019 as a spin-off from MIGAL Galilee Research Institute by multidisciplinary scientists with a strong protein chemistry background and industrial experience in drug design. Projini is dedicated to developing new target-site resistance recalcitrant small molecule herbicides that interfere with protein–protein interactions (PPIs) instead of resistance evolution prone herbicides that interact with internal binding pockets (Lu et al. Reference Lu, Zhou, He, Jiang, Peng, Tong and Shi2020; Nooren and Thornton Reference Nooren and Thornton2003). PPIs occur in signal transduction pathways and enzymatic cascades where more than one enzyme is required, such as electron transport chains or reactions where enzymes are coupled, and the product of one enzyme is the precursor of the second; where an enzyme has more than one peptide subunit; and where activators/modulators are bound to an enzyme by PPIs. The plethora of PPIs provides many new targets to attack. The main challenge was that these targets had been considered “undruggable,” because PPI sites are often too broad and diffuse compared with classic discrete enzyme pockets that bind most pesticides. Still, the pharmaceutical sector has been developing short peptides, as well as other small molecule drugs, that target PPIs (Shin et al. Reference Shin, Kumazawa, Imai, Hirokawa and Kihara2020), and the question had been asked why the herbicide discovery groups have not followed their lead (Gressel Reference Gressel2022).

Target-site resistance is unknown or very rare with quite a few herbicides: the old dinitrophenols, the arsenicals, the carbamates, and the thiocarbamates, as well as the very-long-chain fatty-acid elongase inhibitors, and probably many HRAC Group 0 herbicides. The reason for this that these herbicides probably inhibit more than one enzyme target site. To evolve such multi–target site resistance, a weed would have to simultaneously mutate resistance to each target to remain alive. The expected frequency of resistant individuals when a herbicide has multiple targets is the compounded mutation frequencies of resistance to each target. For example, if a herbicide inhibits two targets, A and B, and the mutation frequency to resistance at target A is 10⁻⁶ and to target B is 10⁻⁷, the initial likelihood of there being a resistant individual in a field is one resistant individual among 10¹³ susceptible ones, an exceedingly low frequency. A herbicide that disrupts PPIs would also require two simultaneous complementary mutations to maintain target enzyme activity while not binding the herbicide, and thus such herbicides would be recalcitrant to the evolution of resistance.

At least two long-marketed herbicides seem to act as PPI disrupters. The auxin mimic herbicides displace a suppressor peptide on the auxin binding site, setting off a cascade of reactions. Likewise, paraquat displaces the protein ferredoxin in binding to P700 in photosystem I in the electron transport chain, disrupting electron transfer. A mutation in ferredoxin would preclude paraquat binding, but then ferredoxin would no longer bind to P700. To retain normal electron transport activity, a simultaneous complementary mutation in P700 would have to occur. Indeed, there are no reported field-fit target-site resistances to auxin mimics or paraquat despite their wide use for more than half a century.

The ag-chem industry discovery groups seem not to have gone the pharmaceutical sector route for a variety of reasons. Herbicide discovery efforts were heavily curtailed during the heyday of the glyphosate era. It also would have been both very expensive and difficult to attempt elucidating compounds that disrupt PPIs. One first had to crystallize the proteins of interest, itself not easy, then perform X-ray crystallography on the proteins and then derive their 3D structures from the complicated diffraction patterns. Only then could one try to determine the structures of the “hot spot” surfaces where the proteins interact most strongly. Experts had claimed that “it will take longer than the age of universe” to compute the 3D folded protein structure from amino acid sequences, but AI changed that with AlphaFold and similar AI-driven software programs that solved the structures of thousands of proteins, replacing the cumbersome process via X-ray crystallography (Arnold Reference Arnold2023; Bryant et al. Reference Bryant, Pozzati and Elofsson2022; Jumper et al. Reference Jumper, Evans, Pritzel, Green, Figurnov, Ronneberger, Tunyasuvunakool, Bates, Žídek, Potapenko, Bridgland, Meyer, Kohl, Ballard and Cowie2021).

Previous work by Projini scientists demonstrated that they could design PPI small molecule inhibitors of plant cystathionine gamma-synthase (Bloch et al. Reference Bloch, Haviv, Rapoport, Cohen, Shushan, Dotan, Sher, Hacham, Amir and Gal2021). An additional example is the two-enzyme complex that converts serine to cysteine in a two-step process (Figure 14). The two enzymes must be in physical contact to perform the two reactions.

Figure 14.

The two-enzyme protein–protein interacting complex catalyzing the conversion of serine to cysteine. The complex is composed of two dimers of O-acetylserine-sulfhydrylase (OASS) and a hexamer of serine-acetyl transferase (SAT). Only the binding of the one monomer of SAT whose C-terminal end binds into the binding pocket of one of the OASS monomers is shown. The structures of OASS and SAT were obtained using AlphaFold2. Images and text courtesy of Elad Cohen.

This enzyme complex does not appear in mammals, which are incapable of synthesizing cysteine, and a specific inhibitor should have no mammalian toxicity. From the 3D structure, Projini determined coordinates of PPI hot spots where the interaction was strongest, and then used Projini’s proprietary computational tools to screen virtual libraries of 40 million chemicals in an iterative process: first to 400,000 hits and then to find 250 compounds that should fit, without synthesizing any chemicals, spraying plants, or using greenhouses, all in silico... Projini then used two complementary biophysical techniques, isothermal calorimetry and fluorescence polarization, to elucidate the binding affinity K_d and to validate the ability of each molecule to displace the recombinant O-acetylserine-sulfhydrylase (OASS) from a serine-acetyl transferase (SAT) peptide. This led to less than 100 leads being tested in vivo, first with an Arabidopsis seedling quick test, and then with a cuticle-free common duckweed (Lemna minor L.) to ascertain whether, if taken up, a compound is phytotoxic (Figure 15).

Figure 15.

An example of two lead compounds that were phytotoxic to duckweed at 1 wk after sowing four frond-clusters per plate.

The bleaching of the duckweed fronds was expected, as the sulfur from cysteine is used in the synthesis of the iron–sulfur proteins in the photosynthetic electron transport chain. Disruption causes electrons to oxidize chlorophyll and other cellular components. A more complete description of the Projini discovery process and the structure of each lead appears in Dotan et al. (Reference Dotan, Bloch, Gal, Cohen, Ben-Shushan and Amram2023). Projini’s expertise is in finding leads for new targets, as shown in the example above and those in the pipeline. Examples of the effects of some of the lead compounds in preemergence assays with various species appear in Figure 16. Such leads go to ag-chem industry specialists for structural optimization and formulation.

Figure 16.

Examples of preemergence activity of Projini leads. The structure of each accession is described in Dotan et al. (Reference Dotan, Bloch, Gal, Cohen, Ben-Shushan and Amram2023). The various compounds were applied at rates varying between 1.8 and 2 kg per hectare⁻¹.

As a highly conserved plant hot spot was chosen, it is expected that the lead compounds will be nonselective, unless they or analogues are metabolized by some weeds or crops. Projini can obtain crop selectivity of non-metabolized chemicals by redesigning the protein–protein hot spots to retain enzymatic activity while preventing herbicide binding by changing key amino acids on apposing peptides to generate transgenic or gene-edited resistant crops and improve herbicide selectivity.

In summary, Projini considers efforts to discover single-target herbicides to be passé due to rapid evolution of target-site resistance and has validated its discovery strategy shift to finding herbicide leads attacking PPIs, where two simultaneous mutations are required to evolve target-site resistance. New, sophisticated computational chemistry on docking sites, coupled with computational screening of virtual chemical libraries and elucidation of direct biophysical binding, renders the Projini process much more efficient than random screening.

ENKOMPASS™: A Platform for Target-based Discovery of Novel Crop Protection Chemistries

Enko Chem, Inc. (Enko) based in Mystic, CT, USA, is a target-specific crop protection discovery company that is accelerating the discovery and development of safer, registrable novel crop protection molecules. Incorporating a multitude of emerging research technologies, Enko discovers novel crop protection molecules for target pests faster and more effectively than traditional research and development methods.

Enko utilizes its own scalable, target-based platform, ENKOMPASS™, to efficiently identify novel chemical starting points and rapidly optimize them into lead candidates. The ENKOMPASS™ platform starts with DNA-encoded library (DEL) screening, which identifies novel molecules that act on specific molecular targets, and not on others, from large and diverse pools of chemistry (Gironda-Martínez et al. Reference Gironda-Martínez, Donckele, Samain and Neri2021). Deep exploration of DEL datasets is enabled by Enko’s AI-driven data analysis and machine learning capabilities. Enko then uses structure-based design (SBD) to optimize novel chemical starting points rapidly and precisely. Combined, these tools enable rapid optimization of hit chemistries into leads and ultimately product candidates.

DELs are created using pool and split combinatorial chemistry (Lindell et al. Reference Lindell, Pattenden and Shannon2009) with the synthetic history of each molecule recorded in an oligonucleotide “barcode” that is attached via a chemical linker (Figure 17). The size of a library created using this approach can number in the billions. A screen consists of incubating a specific target protein of interest with a DEL and identifying molecules that bind to the target by sequencing their DNA barcodes (Gironda-Martínez et al. Reference Gironda-Martínez, Donckele, Samain and Neri2021; Peterson and Liu Reference Peterson and Liu2023). Billions of sequence reads are produced in each DEL screen, requiring the use of complex algorithms and AI-powered workflows to deliver results that are translated into the chemical structures of binding molecules. Binding molecules are resynthesized and tested for functional activity in biochemical assays and bioassays, with active molecules becoming the starting points for optimization into products.

Figure 17.

Schematic representation of a DNA-encoded library (DEL) molecule bound to a target protein. A DNA barcode attached to the small molecule through a linker encodes information associated with the specific building blocks that were incorporated into the small molecule during library construction.

Enko has created a broad and deep pipeline within its herbicide, fungicide, and insecticide programs, with more than 50% of these programs representing novel MOAs. The success of one of Enko’s early projects, the discovery of a novel class of protoporphyrinogen oxidase (PPO)- inhibiting herbicides, screened also for safety to humans and the environment, illustrates the power of the ENKOMPASS™ platform. A DEL screen of more than 120 billion unique molecules, using both the wild-type and resistant variants of the PPO protein, produced 15 distinct chemotypes with activity on the PPO variants used in the screen. Utilization of SBD based on a proprietary, ligand-bound crystal structure guided improvements in in vitro and in vivo activity, leading to testing of a lead candidate in the field ∼18 mo from discovery of the original DEL hit. This molecule provides excellent pre- and postemergence control of PPO-resistant populations of Amaranthus spp. carrying commonly occurring PPO2 target-site mutations (i.e., ΔG210, R128G, and G399A) (Barker et al. Reference Barker, Pawlak, Duke, Beffa, Tranel, Wuerffel, Young, Porri, Liebl, Aponte, Findley, Betz, Lerchl, Culpepper, Bradley and Dayan2023b). The molecule has now been through 3 yr of field testing with extensive evaluation across 200 field trials in the United States. The candidate molecule has activity against more than 40 key weed species across major crops in both foliar and residual applications, differentiating itself from existing commercial PPO herbicides.

The ENKOMPASS™ platform continues to evolve and expand, incorporating innovations in DEL, data analysis, and other emerging technologies to continuously accelerate the discovery and development of novel crop protection chemistries that set new industry standards in safety, sustainability, and environmental impact.

AI4AI™—An AI Platform for the Discovery of Novel Chemistries

In the age of the AI explosion (Barbosa et al. Reference Barbosa, Lehmitz, Maksimainen, Mori, Richardson, Roth, Tai, Thakker, Wells and Zemmel2024), Agrematch is a unique company harnessing the capabilities of modern data technologies for the benefit of the agriculture industry. The company leadership, all industry veterans, have a strong belief in AI as the most likely disruptive technology in an industry that still focuses on the use of intensive biological methods for discovery. The current screening methods were developed before the introduction of the computational technology that enables AI to navigate the vast chemistry space still untapped by conventional methods. This paradigm change (Sadybekov and Katritch Reference Sadybekov and Katritch2023) could happen only with the convergence of data science and cloud computing that Agrematch harnesses for the discovery and development of novel products that empower farmers globally and contribute to the production of healthier and safer food for consumers.

The company’s focus on discovering the next generation of crop protection products is exemplified by its proprietary machine/deep learning compound discovery platform—AI4AI™, or Artificial Intelligence for Active Ingredients. The initial system focused on addressing critical demands in weed control, resulting in two closely related predictive platforms, each identifying lead compounds for both natural and safe synthetic herbicides.

AI4AI™ is a perfect example of generative AI (Viswa et al. Reference Viswa, Bleys, Leydon, Shah and Zurkiya2024), where the system has been learning the chemical language and its application space to deliver an efficient functional chemistry discovery approach that is rationally designed to cater to the needs of all phases of the product development process from the initial in silico screening. The system excels in accelerated early-stage active ingredient identification, evaluation, and selection by providing predictive in-depth insights into compound characteristics. This methodical, predictable, and fast screening approach enables early decision making, effectively mitigating product development and registration downstream risks.

At the heart of Agrematch’s computational platform lies the prediction of compound functions, where the definition of a function is very broad, from specific biological activity such as disease control, to many individual toxicity qualities, and even to the potential cost to produce the compound, to name a few. This is gen-AI, where new insights and knowledge are achieved through a harmonious integration of advanced machine learning/deep learning tools with a massive proprietary database and a rigorous validation process. The system architecture of AI4AI™ has several layers, each providing an interface that enables the efficient creation of many different functional models in many unrelated aspects of the product development cycle. The only common entity is the molecule; hence the method’s name: “compound-based.” The architecture’s effectiveness is evident in its various functional modules already trained and used, encompassing product efficacy (plant stress relief, growth stimulation, weed control, and insect control, to name a few), identification and classification of molecule MOA, many toxicity and environmental safety characteristics, selectivity to certain crops, formulation and delivery, production cost, and many more. This modular computational system ensures a comprehensive and tailored approach to predicting the best compound candidates for a well-defined product concept (Figure 18).

Figure 18.

The “compound-based” approach provides early prediction of many critical characteristics and functions for compounds to become products.

After constructing and training the functional modules within the AI4AI™ system, Agrematch effectively employs these modules to sift through the extensive chemical landscape in search of certain desired functionality. The Lipinski law estimates 10⁶⁰ compounds theoretically capable of biological activity, based on size and chemical composition. This hard to imagine large number of different candidate compounds highlights the immense potential of chemical exploration as well as the need for methods that can sort through vast data in reasonable time. Of this staggering number, only around 100 million compounds have been catalogued, tens of millions have been synthesized, and tens of thousands are being used for human applications, underscoring the vast untapped potential within the chemical universe. The bedrock of Agrematch’s technological capabilities lies in its proprietary compound database, boasting more than 6 billion synthetic, a million natural, and as many as desired theoretical compounds. Additionally, Agrematch possesses its own compound generator model, empowering the company to conceive and produce almost any desired compound. The automation tools integrated into the database support suite play a pivotal role, facilitating processes such as data acquisition, manipulation, standardization, predictive filling, and preparation for resource-intensive applications. This holistic approach emphasizes the platform’s capacity to deliver comprehensive and profound compound functional analysis.

In the realm of bioactive chemical computational screening and discovery, a dominant approach was initially developed in drug discovery. This approach assumes a disease agent or a certain protein as a druggable target for the product’s active ingredient to interfere with. This target-based approach typically relies on understanding the structure and function of specific target proteins, often utilizing tools like molecular docking to identify potential ligands that interact with these proteins. This method emphasizes the role of proteins as the central players in biological processes, and its success depends heavily on the availability of high-quality protein structures.

Agrematch divorced its process from the need for a target protein and named its methodology the compound-based approach. This method shifts the focus to the chemical compounds themselves, rather than specific target proteins, allowing screening and predicting the bioactivity of compounds based on their inherent characteristics, utilizing computational models and machine/deep learning algorithms. Instead of targeting a predetermined protein, the compound-based approach explores a broader chemical space, considering synthetic, natural, and theoretical compounds.

The advantage of the compound-based approach lies in its versatility and efficiency. Target-based methods are often constrained by the availability of the target protein structures and the knowledge of their full behavior, limiting the scope of the search. In contrast, a compound-based approach allows for a more extensive exploration of the chemical universe, considering a multitude of compounds without being confined to specific protein targets. Agrematch’s compound-based approach, powered by its proprietary AI4A™ platform, enables rapid and cost-effective screening, providing a holistic understanding of compound bioactivity and many other critical factors required to become a product. This methodology is particularly advantageous when dealing with novel applications; when detailed information about protein structures is lacking; or when the functionality itself is not protein dependent, as is the case for environmental stability or leaching, movement within the plant, or cost of production. By considering the compounds themselves as the starting point, the compound-based approach offers a more comprehensive and adaptable strategy in the quest for bioactive chemicals.

The Agrematch compound discovery process is initiated with a well-defined research question, supported by training datasets and a compound database. Agrematch’s platform showcases the unique ability to design novel compounds in response to specific partner requirements or its own internal pipeline needs. The integration of a validation process adds a layer of rigor to the iterative process, ensuring the reliability and convergence of the computational predictions. In most cases, it takes multiple training iterations that sometimes require new data generation to create a robust, predictive module. The iterative process includes testing the compounds identified by the system for their function, being for example herbicidal activity or MOA, and feeding the results back into the computational system to optimize it (Figure 19).

Figure 19.

An illustration of the Agrematch platform iterative process of in silico screening, laboratory validation, and feedback to the computational system to generate advanced functional compounds libraries.

Agrematch utilizes two proprietary herbicide platforms in house: the Natural Herbicide Platform, designed to create a highly effective and consistent approach to identify herbicides derived from natural sources, and the Safe Synthetic Herbicide Platform, focused on identifying herbicides with new MOAs with optimized safety for crops, humans, and the environment. Notable results from these platforms are the identification of novel inhibitors of PPO (Barker et al. Reference Barker, Geva, Simonovsky, Shemesh, Phillip, Shub and Dayan2023a) and the ongoing development of a natural herbicidal compound that cannot be divulged yet because of intellectual property issues.

The iterative process of AI4AI™ resulted in a family of synthetic compounds with herbicidal activity and a PPO-inhibiting MOA. In an iterative process of computational prediction, lab validation, feedback, and computational optimization, a screening of an ∼1.2 billion compound database resulted in a library of 50 compounds with similar molecular structure. Out of which, 14 compounds had good herbicidal activity in lab assays, which was later confirmed in greenhouse assays. Surprisingly, the MOA classifier algorithm predicted these compounds to belong to HRAC Group 14 (inhibitors of PPO), even though classical similarity methods showed extremely low similarity scores to the structures of known Group 14 herbicide that was used to train the system. Despite the lack of structural similarity with known PPO inhibitors, these compounds indeed acted by inhibiting PPO, causing the expected light-dependent loss of membrane integrity, photobleaching, accumulation of protoporphyrin IX, and inhibition of PPO (Barker et al. Reference Barker, Geva, Simonovsky, Shemesh, Phillip, Shub and Dayan2023a). More recently, a natural compound identified by the AI4AI™ system as having herbicidal activity has been tested in lab and greenhouse assays that confirmed its efficacy as a contact herbicide, surpassing pelargonic acid by 50 times in potency. It boasts a distinctive MOA and stands out for its low production cost.