Welcome to the Food Analysis Revolution

Food flavor – i.e., the combination of taste and aroma – determines whether we appreciate a food product. Understanding flavor formation and flavor defects is, therefore, the number one research priority in the food industry. Indeed, every research project will have a flavor angle to it. To give a few examples: in the transition to plant-derived proteins we encounter bitterness issues; recycling packaging materials holds the risk of flavor transfer; and more sustainable, easier to recycle packaging materials generally have poorer barrier properties with more air and water reaching the packaged product; reducing temperatures in almond roasting to save energy; reducing salt and sugar levels to make a product healthier – these will all have flavor consequences. There is no change in raw material, production process, or packaging that comes without risks for the flavor of the food – and hence consumer appreciation.

However, flavor is very complex and far from fully understood. Whether or not you like a given food is a multisensory process, complemented with past associations, the subconscious, other aspects of psychology, marketing, and so on. Even if we focus exclusively on the chemistry – the interaction of molecules with receptors – it is complex. Compounds can have very different odor thresholds; two compounds together can have a smell completely different from that of the two individual species; response is highly non-linear; there can be masking issues…

And that is why the proof of the pudding will always remain in the eating! Taste panels are an indispensable tool in flavor research. They are often the only correct way to detect problems and quantify the extent of the issue. Panels describe what they taste and smell, analysis shows which compounds are present and for many compounds, odor descriptors are known. The direct combination of GC with human sniffing and MS, known as GC-MS-olfactometry (GC-O/MS), enables the identification of individual odorant molecules. However, GC-O/MS is a challenging technique and is typically considered a last resort when other methods are unsuccessful.

So, sensory paneling is here to stay. But human panels are notoriously inaccurate, subjective, and tremendously slow and expensive, which is why we need to minimize their usage. Some food smells are always wrong – rancidity is one of these smells. But for many other smells it is just the level that is not correct – or the sample it comes from. A woody smell and taste might be nice for an oaked Chardonnay wine, but the same woody sensation from an ice cream would generally be seen as an indicator for the use of a low-quality wood stick. Peas taste “green” as everyone knows. But if a plant-based pea milk tastes green that is seen as a flavor defect. Using analytical chemistry, we can identify the cause of a flavor and trace back its origin.

This is easier than ever before, thanks to advances in our understanding of food flavor. For most common foods we now know the key odorants, so we know what to measure. Individual foods have (only) up to around 30 key odorants; and with sensitive, modern GC-MS systems, they can be targeted and analyzed rather easily. Around 10 years ago, a meta-analysis of food flavors demonstrated that if you analyze hundreds of foods for their main odorants, you only find 226 compounds (1). Apparently, there is a massive overlap in the key odorants of even very different foods and only 226 compounds describe the entire aroma space. This led us to the idea for a multi-targeted method that measures these 226 compounds. But this is not without challenges: these 226 compounds are present amidst thousands of other volatile compounds. Moreover, odor intensities can be very different meaning that there can be 10 orders of magnitude difference in the concentrations at which they are relevant.

It is only thanks to very recent developments in chromatography-MS systems, in particular the much-improved peak capacity and sensitivity, that we can now think about quantifying 226 compounds in one run. With this targeted 226 compounds analysis, we cover the entire aroma space, so we have all the information to correlate compounds to a product’s flavor. The improved peak capacity of modern systems helps to avoid overlap and incorrect data. Those compounds that correlate with the off-flavor intensity, color, instability, and so on could potentially be relevant. But one should always carefully evaluate whether there is a true causal relationship, and what the underlying explanation is. Did you know that there is a perfect correlation between ice-cream sales and people drowning in the sea? This one is obvious, but in flavor-composition correlations that is not always the case.

Performing measurements is probably the simplest part of the job as an analytical chemist. It is the decision what to measure and how to measure it that is difficult. As is setting up the system and making sure it provides reliable data. Measuring everything using just one generic system is a safer approach. There is no risk of measuring the wrong compounds, you are sure the system works, no start-up issues, no expected errors, and so on. But zooming out too much holds the risk of missing the detail. Finding the right balance between hypothesis-driven testing and exploratory learning is what makes the task challenging. And we work in an environment where methods and instruments are getting more complex, numbers of people are getting fewer and experience is undervalued because information is everywhere. In our lab, we solve this by running every new sample on the broadest sample prep technique we have: arrow solid phase microextraction on a mixed fiber. For the separation, we use the GC-MS system with the highest peak capacity and best sensitivity, comprehensive GC×GC-ToF MS. This together gives us the broadest possible compound coverage with the best possible resolution and sensitivity. And, as a result, the best chance of a first-time-right approach.

Evolution to AI revolution?
 

Developments in separation sciences and food chemistry are evolutionary rather than revolutionary. We innovate in small incremental steps. Researchers continuously refine techniques, optimize processes, and build upon existing knowledge. However, occasionally, when different research lines intersect, they can merge and lead to transformative breakthroughs that revolutionize the entire field. This is what we have seen in food analysis. 

As separation methods have improved, so too has our understanding of what determines the flavor of a food. We have learned that out of the tens of thousands of food volatiles, there are only 226 that are really important. And due to the improvements in separation science, we can now measure hundreds of compounds reliably and rapidly. So, we no longer have to worry about which compounds to measure when food flavor questions arise – just measure the 226 and think retrospectively. Small evolutions in separation sciences and in understanding of flavor came together and caused a revolution.

Looking to the future of food analysis, it is clear to me that sample preparation, chromatographic separation, MS detection, and data evaluation are no longer our bottlenecks. The weakest link is translation of the MS data into a reliable peak table, for all compounds, or for our selected set of 226 compounds. I once jokingly said that we measure samples on Monday and spend the rest of the week checking peak integrations and compound assignments. Here, we need AI to help us. 

AI can help us to get better data faster, but, more importantly, AI can also use these data to unlock qualitative design insights and use these to develop better products faster. Consumer food-preference in the end is determined by the molecules present. It is now about using AI to develop predictive models connecting our massive databases of compositional information with the even larger collections of marketing data on consumer preferences and trends.