Nontargeted Screening Approaches for Potential Food Adulterants and Contaminants

Despite a wealth of analytical methods existing for food safety screening, the vast majority of these methods focus on specific compounds or a defined set of compounds, leaving unseen contaminants. Ann M. Knolhoff, a researcher for the U.S. Food and Drug Administration, spoke to LCGC Europe about the development of nontargeted screening approaches for potential contaminants and adulterants in food, and the considerations around sample preparation, chromatography, mass spectrometry, and data processing workflows.

Q. Why is the development of nontargeted screening approaches for food adulterants and contaminants important?

A: Many analytical methods used for food safety monitoring are designed to identify a specific compound or compound class, such as pesticides. While these targeted screening methods are important, compounds that are not contained in this defined method will not be found. A noteworthy example is melamine, which is a nitrogen-rich compound; melamine was being used as an economically motivated adulterant in pet food and milk products to increase the measured signal for total protein content. This adulteration resulted in illnesses and deaths among infants and pets (1). However, melamine and related compounds were not previously monitored. Nontargeted screening methods can aid in identifying the presence of adulterants and potentially hazardous compounds that may be present.

Approximately 11% of the total US food supply is imported; specific examples include 51% of fresh fruit, 28% of vegetables, and 95% of seafood by volume (2). This globalization of the food supply results in increased complexity. For example, each country has its own regulations for what is or is not acceptable. There are also more complicated supply chains, processing can vary between countries, and more diverse sample types are widely accessible. This necessitates the development of analytical methods that can identify potential health hazards that may arise. Nontargeted screening methods will ideally be able to identify potential issues with the food supply and aid in quickly identifying responsible compounds if a negative health effect is observed. These methods need to be accurate and able to quickly identify problematic samples and compounds.

Q. Sample preparation is an important part of most analytical methodologies, however, what are the issues surrounding the development of effective sample preparation for nontargeted methods?

A: Different analytical techniques are used for nontargeted screening, such as spectroscopy and mass spectrometry, which are often complementary to one another. Each of these methodologies have different requirements for sample preparation and it would be incredibly difficult to have one sample preparation workflow that could be used with all of these instrument platforms. Mass spectrometry has the advantage of being able to detect many different compound classes-thousands of compounds can be detected in a single sample-and a large dynamic range can be measured which is useful for detecting both low- and high-levels of hazardous compounds. Because of these advantages, the discussed sample preparation challenges will be specific to mass spectrometry.

An optimal sample preparation method would extract compounds of interest, reduce potential interferents, and could be applied to different sample matrices without removing compounds of interest. Furthermore, compounds that differ in size, charge, acidity and alkalinity, and polarity would be reproducibly extracted (3). However, developing one method that fits all of these requirements is challenging, especially with sample matrices as complicated and diverse as food. To obtain the best compromise for different compound classes, implemented methods strike a balance between sufficient sample clean-up to prevent instrument contamination and extracting as much as possible from a single sample. It is also likely that nontargeted methods will not exhibit the same recoveries as targeted methods because they may not be as selective. Food samples are also diverse, where different sample preparation strategies may need to be implemented depending on the sample type. The diversity of food matrices and compound classes also makes developing a universal sample preparation approach that will be successful for all cases unlikely for nontargeted screening. However, by using traditional approaches, such as preparing sufficient sample replicates, extraction blanks, and matrix spikes that contain diverse analytical standards, methods can be examined to determine if they are fit-for-purpose and reproducible for the needed application.

Q. Why is a chromatographic step important in a nontargeted method with regards to data quality?

A: One of the major advantages of using mass spectrometry is that thousands of compounds can be detected in a single food sample. This is especially true when combined with good chromatography because it can reduce the measured sample complexity, resulting in a greater number of compound identifications (4). High-resolution mass spectrometry (HRMS) offers high mass accuracy and results in better separation of compounds that are similar in their mass-to-charge ratios (m/z). However, this resolution may not be sufficient in a complicated sample without a chromatographic step. We have observed matrix interferences at the 140,000 resolving power setting on an orbital trap instrument, despite using a long chromatographic gradient (50 min) (5). Using chromatography decreases the probability that these issues will occur. Increased mass accuracy errors can also be observed in orbital trap instruments when coeluting compounds of similar m/z are present, which can result in impaired molecular formula generation (4). Another consideration is that a significant number of food constituents may be measured in high abundance, which can result in ion suppression. Compounds with insufficient abundance will have higher isotopic ratio errors that can result in incorrect molecular formula generation (5). Likewise, quadrupole time-of-flight (QTOF) mass analyzers can be susceptible to higher mass accuracy errors at low and high abundance (5). These matrix effects can be reduced if eluting compounds are chromatographically resolved. Achieving optimal data quality is vital to ensuring high-throughput, automated data processing workflows can be successful and reproducible.

Q. Are there any downsides to the inclusion of a chromatographic step?

A: Some may argue that a benefit to not using chromatography can be time. It can be faster to screen samples without chromatography, where analysis times can be 30–60 min per sample depending on the gradient length. However, because the rate-limiting step in nontargeted screening workflows is analyzing these information-rich data sets, I would argue that by using chromatography the data quality increases, which leads to faster data processing and more reproducible and accurate results. This is especially true when analyzing chemically complex sample matrices and data sets.

Q. What are the key considerations and potential pitfalls when generating the chemical formulae for unknown compounds?

A: Most available data analysis software programs that process HRMS data have functionality to generate molecular formulae. There are typically different settings that you can choose that will influence the output, such as minimum and maximum numbers of elements that can be used. As the molecular weight increases, so does the number of possible chemical formulas. Seven golden rules were established for increasing the probability of generating a correct molecular formula (6). Among these rules are thresholds that should be used for mass accuracy (<3 ppm) and relative isotopic ratio error (<5%). This is another reason why ensuring high quality data is important.

Q. Chemical database can be incredibly useful when identifying unknown compounds as well as in certain data analysis approaches such as those in food “omics”. However, incomplete databases can lead to issues. How would complete databases change the prospect of untargeted screening methods, and can they ever truly be complete?

A: Compounds in chemical databases can be referred to as “known unknowns”. Suspect screening using liquid chromatography (LC)–HRMS uses a large database of specified compounds using the m/z and isotopic pattern to determine the presence of a compound. This strategy is different than nontargeted screening because the data are still being screened against a targeted compound list. There is a lot of merit to this workflow; it can be useful in ruling out the presence of known adulterants in these information‑rich data sets and can complement nontargeted screening strategies.

The majority of compounds in foods are safe. From a food safety standpoint, all of these compounds do not require identification. If we could assume that databases are “complete”, known food compounds could be removed from the data and the remaining features could be identified. It would be useful if molecular databases would characterize compounds by being safe or hazardous; this information can be difficult to find but would be useful for high-throughput screening purposes to quickly highlight compounds of concern. A “complete” database in terms of the compounds that are present would also need to contain MS/MS spectra to identify and confirm generated molecular formulae. However, I don’t think that a database would ever be complete-I also don’t know what metrics could be used to define it as such. One of the ways, we’re trying to be less reliant on available molecular databases is to develop chemometric data analysis workflows to determine what compounds warrant identification or what sample requires further analysis (7).

Q. Automation is crucial in this type of screening to reduce costs and make the method viable for widespread use. Are there any special considerations that need to be taken to ensure a method can be automated?

A: The data analysis process is the most challenging part of a nontargeted LC–HRMS workflow to automate, especially if chemometrics methods are applied. Analyzing extraction blanks, a quality control mixture, matrix spikes, and sample replicates can help streamline this process (7). Analyzing extraction blanks enables the removal of features from a data set that are not inherent to the sample matrix. A quality control standard mixture can be used to monitor instrument performance by ensuring stable retention times, sufficient signal abundance, and measured mass accuracy errors are less than 3 ppm to promote correct molecular formula generation. Matrix spikes can indicate whether molecular features are being accurately extracted from the data set, if small chemical differences can be determined between sample groupings, and if the data processing workflow is effective. They can also be used to determine data quality. Sample replicates yield information regarding sample variability and should especially be used if applying chemometric methods for distinguishing sample groupings. However, processing workflows need to incorporate mechanisms to automatically process and report findings. Randomization of acquired samples can also limit the effects of instrumental differences on data output.

Q. Reproducibility has been a major issue across analytical chemistry. What steps would you recommend to ensure any developed nontargeted method can be reproduced?

A: In order for nontargeted screening methods to be reproduced, each part of the method will need to be examined, including the sample preparation, data acquisition, and data processing methods. This can be challenging if the available instrumentation and data analysis processing software differ from reported methods. However, the development of a standardized quality control standard mixture would help ensure that developed workflows are sufficient and that the same result could be obtained at different sites, on different instruments, and using different data processing software. This will also enable researchers to optimize each step of their own workflows. Additionally, incorporating the factors discussed with regard to automation in the previous question will also help establish workflows that will be easier to reproduce.

References

1. L.S. Jackson, J. Agric. Food Chem.57, 8161–8170 (2009).
2. https://protect-eu.mimecast.com/s/VzsQCYW99H6KXp0f00bST? domain=fda.gov
3. A. Armirotti, A. Basit, N. Realini, C. Caltagirone, P. Bossù, G. Spalletta, and D. Piomelli, Anal. Biochem.455, 48–54 (2014).
4. T. Croley, K. White, J. Callahan, and S. Musser, J. Am. Soc. Mass Spec. 23, 1569–1578 (2012).
5. A.M. Knolhoff, J.H. Callahan, and T.R. Croley, J. Am. Soc. Mass Spec.25, 1285–1294 (2014).
6. T. Kind and O. Fiehn, BMC Bio. 8, 105 (2007).
7. A.M. Knolhoff, J.A. Zweigenbaum, and T.R. Croley, Anal. Chem.88, 3617–3623 (2016).

Ann M. Knolhoff obtained her Ph.D. in chemistry at the University of Illinois at Urbana-Champaign, USA, working under the direction of Jonathan Sweedler. During her Ph.D., her research focused on analyzing different cell types within the brain using mass spectrometry, which included implementing metabolomic workflows. In 2011, she began working at the U.S. Food and Drug Administration in the Center for Food Safety and Applied Nutrition. As a research chemist, she develops nontargeted screening approaches using liquid chromatography and high‑resolution mass spectrometry for food safety applications.