Speciation and milk adulteration analysis by rapid ambient liquid MALDI mass spectrometry profiling using machine learning

Cristian Piras,Christopher K Reynolds,Michael Morris,Oliver J Hale,A K Jones,Nick Taylor,Rainer Cramer

doi:10.1038/s41598-021-82846-5

Abstract

Growing interest in food quality and traceability by regulators as well as consumers demands advances in more rapid, versatile and cost-effective analytical methods. Milk, as most food matrices, is a heterogeneous mixture composed of metabolites, lipids and proteins. One of the major challenges is to have simultaneous, quantitative detection (profiling) of this panel of biomolecules to gather valuable information for assessing food quality, traceability and safety. Here, for milk analysis, atmospheric pressure matrix-assisted laser desorption/ionization employing homogenous liquid sample droplets was used on a Q-TOF mass analyzer. This method has the capability to produce multiply charged proteinaceous ions as well as highly informative profiles of singly charged lipids/metabolites. In two examples, this method is coupled with user-friendly machine-learning software. First, rapid speciation of milk (cow, goat, sheep and camel) is demonstrated with 100% classification accuracy. Second, the detection of cow milk as adulterant in goat milk is shown at concentrations as low as 5% with 92.5% sensitivity and 94.5% specificity.

Highlights

Abbreviations AP Atmospheric pressure enzymelinked immunosorbent assay (ELISA) Enzyme-linked immunosorbent assay ESI Electrospray ionization fatty acid (FA) Fatty acid HPLC High-performance liquid chromatography Linear Discriminant Analysis (LDA) Linear discriminant analysis LDI Laser desorption/ionization LOD Limit of detection liquid support matrix (LSM) Liquid support matrix MALDI Matrix-assisted laser desorption/ionization MS Mass spectrometry PCR Polymerase chain reaction trichloroacetic acid (TCA) Trichloroacetic acid Waters Research Enabled Software (WREnS) Waters research enabled software
The model obtained from LDA was used to classify 64 unknown samples, achieving a 100% classification accuracy for these samples, the software’s built-in ‘leave 20% out’ cross correlation testing of the data from the samples used for LDA model building resulted in an accuracy of only 98%
For detecting small amounts of cow milk in goat milk, a wider m/z range was used to take into account additional milk components such as proteins and their fragments that offer species-specific sequence information that lipids do not provide

Summary

Introduction

Abbreviations AP Atmospheric pressure ELISA Enzyme-linked immunosorbent assay ESI Electrospray ionization FA Fatty acid HPLC High-performance liquid chromatography LDA Linear discriminant analysis LDI Laser desorption/ionization LOD Limit of detection LSM Liquid support matrix MALDI Matrix-assisted laser desorption/ionization MS Mass spectrometry PCR Polymerase chain reaction TCA Trichloroacetic acid WREnS Waters research enabled software. One of the most frequent adulterations in the field of dairy products is related to the fraudulent addition of cow milk to milk of other species[2]. This represents a matter of food quality and authenticity but is a matter of safety for consumers with cow milk allergies[3]. Profiling of lipids by MALDI MS has been successfully used for the discrimination of coconut, soya and bovine m ilk[5] It is a rapid and inexpensive method for the detection of cow milk as an additive to coconut and soya m ilk[7,8] that could lead to life-threatening consequences[9]. Compared to proteins, lipids are relatively small molecules without a species-specific sequence, which can be utilized to identify underlying species

Objectives

Methods

Results

Conclusion