Food Monitoring by Untargeted Screening with Semi Quantification — Feasibility Study on Analysis of Cereals and Honey

Trust of food is of great importance to our society. In order to ensure the food safety for now and for the future, the analytical strategy should be developed in parallel with the growing health concern and increased complexity in chemicals occurring in food. Many chemical compounds can be found in food, and a large number of them are unwanted.

It is hard to include too many unwanted compounds into one specific targeted method. Therefore there has been growing interests to identify and quantify not only targeted and known compounds but also untargeted or “unknown” compounds in food that are missed in those targeted analysis. There is clearly a need for a robust and reliable untargeted analytical methods with wide scope that can cover more compounds, even standards are not available.

In this project, cereal and honey were selected as cases for the untargeted analysis feasibility study. The perspectives of this PhD project is to deal with the massive and complicate data come out of high resolution mass spectrometry (HRMS) and develop a generic, wide scope, and efficient analytical strategy for untargeted analysis, in particularly to get quantitative data from these when standards are not available.

In the end, providing more data for food monitoring and risk assessment with less efforts. Mainly four parts were carried out in my project, namely, analytical method optimization, data processing method optimization, identification, and semi-quantification method development and application. In analytical method optimization part, to achieve better sensitivity, ion source parameters and the eluent compositions were studied and optimized.

Ion source optimization was first operated in flow injection mode on APCI and ESI. As no obvious advantage in APCI was found compared with ESI from the results, ESI was selected for the further study. The effects of four key source parameters in ESI on common food contaminants and residues were also studied with chromatography separation (Manuscript 1).

A 40% increase in positive mode and 20% increase in the negative mode of response factor compared with central points was achieved by ion source optimization. In eluent composition optimization, pH 3.0 and pH 3.3 with ammonium formate buffer achieved the best performance for all test compounds. In data processing method development part (Manuscript 1), the commonly used metabolomics approaches were optimized and compared regarding peak detection capacity and accuracy.

These approaches were also compared with the performance of suspected screening. Developed and highly automated workflow based on XCMS package was setup and achieved highest mass accuracy, highest detection rate (96%) and made a clear distinction between the control and spiked groups by multivariate analysis in a true untargeted way (metabolomics-like approach), even for the concentration of 5 μg/kg.

Such approach could be a good complement to routine targeted analysis in a view of rapidly detecting potential contaminated food products without prior information and sacrificing too much accuracy. In identification parts, a generic database were constructed including common chemical residues and contaminants according to EU regulations, standards available from our collection, identified or tentatively identified mycotoxins and metabolites from literature.

In addition, the predicted logP/logD values were also registered into the database to minimize the number of false candidates. The tentative identification of compounds was also demonstrated by correlating fragmentation spectra from QToF-MS to in silico generated spectra from database to find the compounds with the best match.

The constructed database was applied in the screening in both manuscript 1 and 2. Semi-quantification method development and application took the largest proportion of my study. Quantitative prediction model (QPM) was applied for pyrrolizidine alkaloids (PAs) in honey, and pesticides and mycotoxins in cereals without using standards.

In honey study, a targeted method was proposed that can be extended to a general strategy to achieve estimation of quantitative data when standards are unavailable (manuscript 2). The QPM by using multiple linear regression (MLR) was successfully quantified eight PAs without standards the first time and only 50.8% prediction error was achieved on QToF-MS.

Another QPM using random forest was applied for cereal matrices and quantified 134 pesticides and 5 mycotoxins without standards (manuscript 3 and 4). This model was fist time validated on different instruments from different labs in this study, a high correlation was also achieved (R2=0.86). Less than 4 times prediction errors were achieved for all test compounds, which is quite promising to be used in future food monitoring.

The untargeted strategy proposed in this study shows the semi-quantitative untargeted analysis is valuable tools for helping ensure food safe in the future, while the quality of identification and semi-quantification can always be further developed until it is more close to truth, but we moved one step further on this way.