Customer Story

Safety in numbers: Multifactorial optimization improves monitoring of PAH in foods 

An analytical chemist at Denmark’s National Food Institute uses design of experiments in JMP® to develop a new, more comprehensive method to detect carcinogenic food processing contaminants 

Polycyclic aromatic hydrocarbons (PAHs) are chemical contaminants produced by the combustion of carbon-based materials that can enter food either through environmental contamination or during processing. PAHs are of concern to regulatory bodies worldwide, as they are known carcinogens and may also adversely affect liver and kidney function when consumed.

“When herbal food supplements are being prepared, sometimes plants are dried using an accelerated process that involves high amounts of heat. That can favor the creation or adherence of particles of PAHs,” explains analytical chemist Alin Ionas, PhD. “It’s important because with processing contaminants, as long as we do not focus on improving the processes, we will not be able to reduce the concentrations.”

In the European Union, regulatory guidelines specify acceptable thresholds for potential toxins like benzo[a]pyrene, a common PAH, with food manufacturers expected to adopt their own quality assurance programs. Ionas was part of a team at Denmark’s National Food Institute tasked with developing new methods to detect and reduce processing contaminants.

Situated just north of Copenhagen, the Institute is operated by the Technical University of Denmark, which also spearheads research for the European Union Reference Laboratory for Processing Contaminants (EURL-PC). Here, classical analysis methodologies are being updated to keep pace with modern needs and trends in the food safety space. Of particular concern are products for which low maximum PAH4 limits exist, such as infant formula, and challenging analyses of PAH4 compounds where the performance criteria require a limit of quantitation (LOQ) of 0.9 µg/kg for each single PAH4 compound.

A paradigm shift in the analytical process 

Researchers working in the food safety space have long used a standard monitoring process that involves both gas chromatography with mass spectrometry (GC-MS) and high-performance liquid chromatography with fluorescence detection (HPLC-FLD). As scientists’ understanding of food safety has evolved in recent years, however, the field has been challenged to accommodate a paradigm shift toward detector systems that record more information simultaneously, such as high-resolution mass spectrometry (HRMS). The result has been a gradual implementation of techniques such as HPLC-HRMS for this purpose. 

For PAH monitoring, atmospheric pressure photoionization (APPI) provides the best performance among LC-compatible ionization techniques that can be coupled with HRMS. With only an atmospheric pressure chemical ionization (APCI) available however, Ionas has worked to develop a more comprehensive analysis approach that could be used both in contaminant monitoring and potentially even chemometric approaches in food safety. 

“The challenge was to not only develop an analysis method that could meet the sensitivity requirements for PAH monitoring and allow for expansion into other PAHs and analytes of interest, but to do so on a platform that does not provide the best sensitivity for this purpose,” Ionas explains.

“The part of the analysis method that yielded the highest gains in sensitivity was optimizing the source parameters. But this process is challenging and slow to do manually, since it involves evaluating parameters one at a time and examining how they impact each analyte. Furthermore, some parameters are co-dependent; there can be more than one combination of parameters providing similar results. With the complexity of codependent parameters and more factors to consider than in previous studies… I would have to take every parameter individually, one by one.”

Design of experiments helps to overcome previous process limitations

Ionas explains that the team turned to a liquid-chromatography high-resolution mass-spectrometry (LC-HRMS) instrument using APCI ionization because of fitness-for-purpose and availability. But Ionas says APCI works best for compounds that have low to medium molecular mass and are in a medium polarity range, making the technique suboptimal for nonpolar chemicals like PAHs.

“We weren’t at the necessary level of performance to meet the requirements included in the EU legislation,” he says. “But it was the tool we had, so I wanted to make it work. I was trying to squeeze every last bit of performance out of that instrument using the existing ionization source.”

The challenge of using APCI is twofold, Ionas explains: For one, the sensitivity and performance associated with this ionization technique is not optimal compared to other analysis techniques like classical GC-MS and HPLC-FLD. A thorough optimization is therefore required to achieve appropriate levels of sensitivity; in practice, the team was unable to produce a high enough level of signal for their four key PAHs by doing a basic manual optimization.

Secondly, while necessary to reach the required thresholds for the four PAH compounds that are included in monitoring schemes, the team felt it was important to include additional PAH compounds in the method so as to obtain a more complete picture of PAH contamination overall. While not being included in the classical monitoring scheme for PAHs, some of these compounds have associated toxicological concerns, such as carcinogenicity, as outlined by the EU Scientific Committee for Food (SCF 2002). 

“This is where being able to set a relative (numerical) importance for the optimization, as is possible with the implementation of design of experiments in JMP, proved invaluable,” Ionas says.

Having heard of JMP from colleagues at DTU – but never having tried it himself – Ionas set about exploring the extensive menu of ready-made design of experiments (DOE) functionality in the software. “It helped that there was a lot of information on the JMP website, including a video on how to set up the software quickly,” Ionas says. “So I could do a quick trial run and see how easy it is to implement JMP and how well it works…. It was just a matter of exploring various models and options I could use.”

A platform with ready-made tools for DOE and response surface modeling

Beyond simply finding a solution to the immediate problem at hand, Ionas’ broader goal was to explore how close a new optimization method based on DOE could get in sensitivity to the established methods – and to identify the extent to which sensitivity could be increased with techniques like response surface modeling. This approach achieved the most from the UHPLC-APCI/HRMS instrument by using a sequence of designed experiments to explore the relationship between multiple explanatory variables and one or more response variables.

Response surface designs use predictive modeling to identify better process operating settings. This method, Ionas explains, allows him to implement a combination of source parameters – including nebulizer pressure, vaporizer temperature, drying-gas temperature and flow, corona current and capillary voltage – as defined by the model on a mix of PAH analytical standards. He can then input the results back into the model to determine optimal parameters for the experiment.

Compared with one-parameter-at-a-time approaches, multivariate optimization promised significant time savings in addition to increased sensitivity. “I’ve worked on projects where we tried to optimize a general method for a few hundred compounds. And then we had to try to find a set of instrumental parameters that were good enough to get reasonable performance for most of the chemicals we were looking at. That’s a daunting task,” Ionas explains. “So I felt that using the DOE platform from JMP was the solution.”

Simple model, impressive results

Ionas set up a response surface model using the JMP Custom Design platform. By default, the platform uses the recommended optimality criterion based on whether quadratic effects are added in the model outline. After testing several suggested combinations of factors, Ionas created individual methods for each combination. Setting the mass spectrometer to perform injections automatically overnight and automating data extraction, he then plugged the data back into JMP to figure out how well it worked and which parameters to optimize further.

For the experiment’s first iteration, Ionas kept the factors as broad as possible so as to surface potential optimization opportunities. He included the four most important PAHs, plus 12 additional PAH chemicals, some of which are either known or suspected carcinogens by the EU Scientific Committee for Food (SCF 2002). Ionas assigned a numerical importance to each as a model input, and he says, “I found JMP to be straightforward to use and quite powerful.”

In the end, Ionas was able to optimize APCI source parameters to achieve a significant increase in PAH response. Compared to the instrument’s default set of source parameters, multivariate optimization increased signal by between 170 and 120,000 percent. The method produced higher increases for those analytes with a higher level of importance, he notes, and less for the others.

Among the advantages of multifactorial optimization in JMP, Ionas says time savings has been one of the most noticeable benefits. “Once you have some experience running these kinds of experiments, you can just set it and forget it,” he says. “You set up the experiment, let the instrument do its thing, validate the way you extract data, and make sure you get high-quality data from your data files. Then it’s not much work to put it back into the model and generate the next iteration.” This repeatability has several benefits: not only can a single lab run more experiments simultaneously, the results are more easily verified through replication when needed.

Having only first come across DOE as a postdoctoral researcher, Ionas sees significant potential for statistical approaches to gain more penetrance in the field of analytical chemistry. In his own work, he says he’ll plan to include many more analytes in future models. “Small-molecule analysis is moving toward analyzing a larger number of chemicals in one method,” he predicts. “So this will definitely be useful in the future.”

REFERENCES

SCF. Opinion of the Scientific Committee on Food on the risks to human health of polycyclic aromatic hydrocarbons in food. European Commission Health and Consumer Protection Directorate-General. SCF/CS/CNTM/PAH/29 Final 4. December 2002. http://ec.europa.eu/food/food/chemicalsafety/contaminants/out153_en.pdf

ACKNOWLEDGEMENTS

Dr. Ionas would like to acknowledge Peter Abrahamsson for valuable discussions on typical APCI source parameter ranges.