Mycotoxins are a critical food safety threat. Dietary exposure to mycotoxins is connected to health effects like hormone disruption, immune system compromise, liver and kidney damage, miscarriages and developmental harm, and increased cancer risk.

According to the European Environment Agency (EEA), at present, approximately 25 percent of crops exceed EU regulatory limits for mycotoxins, with contamination occurring at levels above detectable limits in up to 60–80 percent of crops. EU chemical human biomonitoring data suggest that 14 percent of the adult European population is exposed to just the mycotoxin deoxynivalenol (DON) at dangerous levels. These statistics are only expected to worsen as global temperatures warm, suggested EEA.

In their study, the Wageningen University researchers also cited statistics anticipating increased probabilities of DON contamination in wheat in Northern Europe by 2050, as climate change drives shifts in fungal distribution, contamination frequency, and co-occurrence patterns in cereal crops.

Considering Mycotoxin Risk Under Different Future Scenarios

In this context, the researchers sought to develop a framework for predicting mycotoxin contamination in European wheat under climate change. The framework integrated historical monitoring data with the latest climate projections from the Coupled Model Intercomparison Project Phase 6 (CMIP6) under multiple Shared-Socioeconomic Pathways (SSPs), which are different scenarios that describe how societal development could affect global climate change mitigation, adaptation, and impacts. Machine learning models trained to understand crop phenology (i.e., the timing of growth stages in crops) were used to analyze the data to produce future contamination risk maps for wheat across Europe.

The use of the framework was demonstrated with winter wheat as an example, focusing on contamination by six major mycotoxins: DON, zearalenone (ZEA), T2 and HT-2 toxins (T2+H2), aflatoxin B1 (AFB1), fumonisins (FBs), and ochratoxin A (OTA).

The study is the first to apply a predictive machine learning modeling framework to a large geographic area in Europe. Previous research has focused on a single country or a select few countries at a time.

Although the researchers used the modeling framework to predict mycotoxin contamination risk in winter wheat in Europe, they stress that the framework is broadly applicable to various food safety concerns.

Model Performance and Results Interpretation

The model demonstrated strong performance for low-contamination classes, while performance for medium- and high-contamination classes remained limited due to class imbalance. Results showed an overall accuracy greater than 90 percent.

The reduced sensitivity for elevated contamination levels reflects the limited number of observed high-contamination events, and implies that projections indicating increased contamination risk should be interpreted as signals of relative risk escalation rather than precise predictions.

The researchers emphasize that, given uncertainties associated with climate forcing, phenology modeling, and statistical learning, the projections are most informative for identifying relative shifts and emerging patterns in contamination risk under future conditions. Projected changes in mycotoxin contamination reflect conditional outcomes under specific climate and socioeconomic scenarios and should be interpreted comparatively across scenarios and regions rather than as quantitative forecasts.

Mycotoxin Risks Increase Across Scenarios

Under four climate change scenarios, the validated mycotoxin prediction model indicated contamination risk increases for DON, ZEA, and OTA.

DON was more frequently projected to reach the highest contamination class in 2050 and 2070, especially in coastal areas of Europe, the UK, and northern France.

The model showed slight increases in OTA and ZEA contamination by 2050 and 2070, indicating that these mycotoxins would not pose a high safety threat in wheat in the near future.

Interestingly, the estimated mycotoxin contamination risk varied from year to year, with the medium-contamination class peaking in the middle of the century under a low-emissions, sustainable future scenario. The high-contamination class saw a mid-century peak under the high-emissions scenario and in 2075 under the low-emissions scenario.

The model’s indicated increase in high contamination of DON and ZEA—with DON contamination classes higher than those of ZEA—can be attributed to environmental conditions, such as ambient temperature and high relative humidity, during the wheat flowering stage. The later decrease in the high class of DON and ZEA contamination was likely caused by the extremely high temperatures, low relative humidity, and rainfall. The high correlation between DON and ZEA could also be attributed to both mycotoxins being simultaneously produced by Fusarium graminearum.

OTA did not correlate with the presence of DON or ZEA.

Using the historical monitoring data, the mycotoxins T2+HT2, AFB1, and FBs showed not to be present in wheat grain, rendering the results null. FB and AFB1 are mainly seen in maize, rather than in wheat. T2+HT2 toxins are mostly seen in oats, rather than wheat.

Wheat Life Cycle Has Strong Impact on Contamination

The findings showed that crop phenology shifts under climate change, likely due to alternation in metabolic activities of nearby soil temperature. Despite the phenology dates shifts, the length of the period between the flowering date and the full maturity date of wheat remained unchanged in all scenarios, with a maximum period variation of two days under climate change.

For the variables of high precipitation, low rainfall, and high temperature, the flowering and full maturity wheat development stages showed the highest effect on the high mycotoxin contamination class.