Screening, risk assessment, and rapid detection of mycotoxins in edible herbs based on UPLC-MS/MS and FT-NIR

Abstract

Mycotoxin contamination in edible herbs poses substantial health risks and economic losses due to its accumulation. This study employed a modified QuEChERS-based UPLC-MS/MS method for qualitative screening of 19 mycotoxins in 1000 batches of 77 kinds of edible herbs. For three kinds of edible herbs with high positive rates, Sojae semen praeparatum (SSP), Lablab semen album (LSA), and Allii tuberosi semen (ATS), the QuEChERS parameters were optimized to quantify the ten most common mycotoxins, while Monte Carlo simulation was used to assess human health risks. Moreover, nine classification models were applied to discriminate the content of sterigmatocystin (ST) in SSP and aflatoxin B₂ (AFB₂) in LSA. Qualitative screening results showed that 65.20% of the samples were contaminated with one or more mycotoxins, and alternariol monomethyl ether (AME), ST, and alternariol (AOH) were detected most frequently in all samples. Risk assessment indicated that long-term consumption of SSP contaminated with ST, LSA contaminated with AFB₂, and ATS contaminated with AME posed health risks. Optimal models for SSP and LSA were support vector machine (SVM) and k-nearest neighbor (KNN), respectively, and the accuracy, precision, recall, and F1-score were all 100% for their test sets. Regarding independent validation samples, the accuracy reached 90% for SSP and 100% for LSA. This study identifies a high prevalence of mycotoxins in edible herbs, highlights their non-negligible dietary risks, and demonstrates the promising application of Fourier transform near-infrared (FT-NIR) spectroscopy combined with machine learning for rapid detection mycotoxins.