Sniffing Out Trouble: Scientists Detect Pest Attacks on Maize in Real Time by Reading Plant Odors

When a caterpillar starts chewing on a maize leaf, the plant does not stay silent. Within hours, it floods the air with a distinctive cocktail of volatile organic compounds, a chemical distress call that parasitic wasps have exploited for millions of years to find their prey. Now, a team of researchers from the University of Neuchâtel and collaborators in Switzerland, Japan, and elsewhere has shown that this ancient chemical language can be translated into a modern early-warning system for agriculture. In a proof-of-concept study published in Smart Agricultural Technology, the scientists demonstrated that sophisticated odor-sensing instruments can distinguish maize plants under attack by caterpillars or fungal pathogens from healthy plants, in some cases using just a single second of air sampling.

The stakes are considerable. Pests and diseases destroy up to 40 percent of potential crop yields worldwide, and the synthetic pesticides deployed against them are widely recognized as unsustainable, contributing to pollution, biodiversity loss, and climate change. Real-time monitoring that pinpoints exactly which plants are under attack, and by what, could allow farmers to intervene precisely and sparingly, slashing pesticide use while protecting harvests. Precision agriculture already offers acoustic sensors that listen for insect activity and imaging systems that track changes in plant appearance, but these approaches struggle with early detection and with telling different stressors apart under real field conditions. Plant odors, the researchers argue, could provide the missing layer of specificity, because different attackers trigger different signaling cascades inside the plant and therefore produce recognizably different volatile blends.

The study, led by Marine Mamin, Carla Arce, and Ted Turlings, put two very different sensing technologies through a gauntlet of increasingly realistic tests. The first was a compact electronic nose built around Membrane-type Surface Stress Sensors, or MSS, developed at Japan’s National Institute for Materials Science. This palm-sized module contains twelve microscopic channels, each fitted with a silicon membrane just three micrometers thick and coated with different receptor materials, including functional silica-titania hybrid nanoparticles and polymers. When volatile molecules adsorb onto and desorb from these receptor layers, they deform the membranes, altering the resistance of piezoresistors embedded in four sensing beams. A Wheatstone bridge circuit converts these resistance changes into voltage signals recorded one hundred times per second, and pattern recognition algorithms can then discriminate between odor profiles.

The second technology was chemical ionization time-of-flight mass spectrometry, or CI-TOF-MS, a heavyweight analytical technique that can measure trace volatiles in open air in real time. In the laboratory and semi-controlled outdoor experiments, the team used a Vocus S instrument with proton transfer reaction ionization, which ionizes volatile compounds by transferring protons from hydronium reagent ions. For the field trial, they switched to a newly developed portable version, the Vocus C, weighing roughly 30 kilograms and consuming only about 250 watts, powered by the battery of a hybrid car and configured with benzene cation ionization, a more selective chemical ionization approach that keeps mass spectra relatively simple.

The experimental design progressed in three stages. First, in the laboratory, maize seedlings of two varieties, Delprim and Aventicum, were enclosed in glass bottles with charcoal-filtered air flowing through, and the volatile emissions of plants attacked by beet armyworm and fall armyworm caterpillars, or infected with the anthracnose fungus Colletotrichum graminicola, were captured and analyzed. Gas chromatography-mass spectrometry revealed that undamaged plants released only trace amounts of compounds like linalool or indole, while attacked plants produced rich blends dominated by green leaf volatiles such as (E)-2-hexenal, monoterpenes like beta-myrcene, sesquiterpenes like (E)-beta-farnesene, the homoterpene DMNT, and aromatic compounds like indole. Fungal infection produced its own signature, distinguished by fatty aldehydes such as octanal, which were absent from caterpillar-induced profiles. Remarkably, the two closely related Spodoptera species elicited quantitatively distinct blends, suggesting the chemical fingerprints are nuanced enough to tell even sister pests apart.

Both sensing technologies rose to the laboratory challenge. The MSS module, sampling volatiles for just ten seconds at a time, generated distinct electrical response patterns for healthy, caterpillar-infested, and fungus-infected plants of both maize varieties, with permutational multivariate analysis of variance showing treatment effects explaining 77 to 82 percent of the variation in sensor responses. Exploratory Random Forest classification based on the sensor data correctly separated the treatments, including the two caterpillar species. The PTR-TOF instrument performed similarly well and even captured additional herbivory markers that GC-MS missed, including ions consistent with protonated aldoximes and other nitrogen-containing compounds particularly associated with beet armyworm damage, underscoring that the full volatile response of plants is richer than conventional analytical methods typically reveal.

The critical test came when the plants left the laboratory. In open-air measurements outside a university building, with no enclosure to concentrate the odors, volatile concentrations plummeted. Median indole levels dropped more than 200-fold compared with the enclosed headspace conditions, and sesquiterpenes fell 26-fold, reaching only low parts-per-trillion levels. Under these diluted, fluctuating conditions, the MSS electronic nose failed: principal component analysis showed its signals were dominated by sampling day, temperature, and humidity, and machine-learning models trained on the data performed no better than random guessing. The equilibrium-based sensor, the authors suggest, simply cannot accumulate enough signal fast enough when concentrations are this low and the air is constantly moving.

The PTR-TOF mass spectrometer, however, proved remarkably resilient. Sampling air just one to two centimeters from maize leaves for sixty seconds, and focusing on thirty-one targeted ions identified in the laboratory phase, the instrument detected significantly elevated levels of indole, sesquiterpenes, isobutyraldoxime, DMNT, and jasmone in caterpillar-damaged plants. Even more striking, a machine-learning classifier, an L1-regularized logistic regression validated with repeated nested cross-validation, could distinguish damaged from undamaged plants using the data from a single second of measurement, achieving a balanced accuracy of 87 percent. Averaging over the full sixty seconds pushed balanced accuracy to 96 percent, with perfect specificity, meaning not a single healthy plant was falsely flagged. A permutation test with 250 label shuffles confirmed the result was not a statistical fluke, and the model even generalized reasonably well to entirely unseen sampling days. Indole and sesquiterpenes emerged as the workhorse predictors, selected in nearly every cross-validation model and carrying by far the largest influence on classification.

The final stage took the technology into a genuine maize field in Mathod, Switzerland, where 129 plants were measured in a single day with the portable Vocus C instrument. Because live caterpillars could not be deployed, herbivory was simulated by mechanically wounding leaves and applying fall armyworm regurgitant, a well-established proxy that elicits volatile emissions similar to real caterpillar feeding. The results were encouraging but sobering: with total ion current normalization, the classifier reached a balanced accuracy of about 67 percent, significantly better than chance according to a block-level permutation test, and spectral features matching the nominal masses of indole and sesquiterpenes were consistently selected across models. Yet the discrimination was partial, with wide confidence intervals reflecting substantial uncertainty, and variable wind direction prevented the consistent downwind sampling that had helped in the simpler outdoor setup.

The authors are careful to frame the field trial as an early proof of concept rather than evidence of field-ready diagnostics. The single-day, single-plot experiment cannot capture the full variability of agricultural environments, the absence of a wound-only treatment means a general wound response cannot be excluded, and the lack of independent reference measurements limited the diagnosis of misclassifications. Still, the trajectory is clear: from 82 percent treatment effects in the lab, to 87 to 96 percent accuracy in open air, to a genuine signal in the field, all achieved with instruments that already exist. The researchers envision odor-based monitoring complementing robotic weeders and selective sprayers, guiding targeted pesticide application, and directing biological control agents to infested hotspots. Real-time mass spectrometers remain costly and bulky for now, but the volatile markers they have validated, led by indole and sesquiterpenes, could guide the development of next-generation portable sensors, and because many crops share overlapping herbivore-induced volatile profiles, the lessons from maize may travel far beyond it.

Subject of Research: Real-time odor-based detection of pest and pathogen attack in maize using plant volatile emissions

Article Title: Odor-based real-time detection of maize plants under attack by pests: proof of concept

Article References: Mamin, M., Arce, C. C., Röder, G., Kanagendran, A., Degen, T., Defossez, E., Rasmann, S., Akiyama, T., Minami, K., Yoshikawa, G., Lopez-Hilfiker, F., Bansal, P., Cappellin, L., Li, Y., & Turlings, T. C. (2026). Odor-based real-time detection of maize plants under attack by pests: proof of concept. Smart Agricultural Technology, 15, Article 102604. https://doi.org/10.1016/j.atech.2026.102604

Image Credits: AI Generated

DOI: Not provided

Keywords: maize, plant volatiles, pest detection, mass spectrometry, electronic nose, precision agriculture, Spodoptera, herbivore-induced volatiles, chemical ecology, crop monitoring, indole, sesquiterpenes