As people, we have a complex relationship with food. We want it to be simple, cost-effective, and convenient. But we also savor gourmet foods prepared with the finest ingredients available. The quality of high-end ingredients depends on their authenticity, purity, and provenance, while our biological need for food requires it to be safe and of good quality. Raman spectroscopy can address both of these challenges, and the technique is gaining traction. Courtesy of iStock.com/Alessandro Cristiano. The food industry draws on a wide variety of analytical technologies to assess food quality and authenticity, including chromatography, nuclear magnetic resonance, stable isotope analysis, spectroscopy, and DNA-based methods. While all are well-established techniques, many are limited by the need for expensive instrumentation and skilled operators, require time-intensive analysis, or consume part of the sample. The speed and nondestructive nature of vibrational spectroscopy techniques such as NIR and Fourier transform mid-infrared (FT-MIR) spectroscopy have made them popular for online and at-line quality control in food production. But Raman offers even greater potential for deployment of rapid, nondestructive testing within the food chain. What makes Raman spectroscopy unique? Like FT-MIR spectroscopy, Raman is a fingerprinting technique, yielding a rich spectrum of sharp peaks that correlate to the fundamental molecular stretches and bends within a sample. As a result, it shares an ability to detect minor components that signal adulteration, as well as those that serve as unique markers of origin and authenticity. This is valuable information to have at the receiving dock, along with import/export points to preserve the integrity of the food chain. In contrast to other vibrational spectroscopies, Raman accesses information by illuminating the sample with a laser and detecting the shift in energy in the scattered light. Raman scattering occurs roughly once in every million photons, and only for vibrations that induce a change in the polarizability of the molecule. This process lends Raman spectroscopy a few unique advantages. In addition to being a nondestructive and noninvasive technique, its confocal nature enables sampling through glass or plastic packaging1. Further, Raman scattering is weak for water, making the technique largely insensitive to moisture in the sample or environment. Advancements in Raman instrumentation are also progressing quickly, making it more rapid, cost-effective, and portable than ever. When combined with the expanding chemometrics options for data analysis, these instrumentation advancements position Raman technology for tremendous growth. For the new user, however, Raman may appear daunting. When it comes to food analysis, the user needs to know: What instrumentation is available and which excitation wavelength should be used? How should the spectra be interpreted or analyzed? And perhaps most importantly, what kinds of problems can be addressed with Raman? Finding the right wavelength Raman spectroscopy reports its spectra as a frequency shift relative to the excitation laser, allowing it to be performed with any wavelength of laser, at least in theory. In practice, shorter excitation wavelengths generate a stronger Raman signal. But they also produce more background fluorescence, particularly in pigment-rich foods. As a result, the most widely used wavelengths for Raman analysis of foods are 1064 and 785 nm. The former offers a good degree of fluorescence suppression, even for complex samples such as wine, oils, and honey. This minimizes the need for pretreatment of spectral data, making it easier to extract information and reducing the risk of introducing artifacts into the chemometric models. Several instrument options are on the market for 1064-nm Raman, with many of the new modular and hand-held systems performing remarkably well compared to benchtop systems. Hand-held systems offer the convenience of point-and-shoot operation but come with onboard software designed for library matching rather than export of spectral data for chemometric analysis. Compact, modular spectrometers allow more flexibility in sample and software interface, and some offer the option to transition to a lightweight OEM equivalent for integration into application-specific or hand-held systems. With the variety of options available, it is important to choose a spectrometer designed specifically for Raman spectroscopy, as its performance is key to the output of consistent, accurate, and reliable data. Defined by data Food contains many different compounds, and the components that confer quality by virtue of geographical origin, purity, processing method, or species tend to be present in low concentrations. The markers of adulteration with subpar or unsafe substitutes may likewise be subtle. The elegance of Raman is that it captures them all in a single spectral fingerprint. This fingerprint, in turn, can be characterized using chemometrics as being either part of or outside of a specific sample group, offering a simple pass or fail to assess authenticity, purity, or safety. Accurate answers depend upon acquiring spectra with a good signal-to-noise ratio (SNR) and a high degree of reproducibility. This is because the spectral differences needed to distinguish between two similar food samples can be small and may emerge only from the chemometrics. “The changes are subtle but significant statistically, especially for meat, where they arise from the protein and fat,” said Keith Gordon, a researcher in analytical spectroscopy at the University of Otago (Figure 1). Figure 1. The mean Raman spectra of 30 samples each of lamb (black), venison (blue), and beef (red) appear very similar, yet contain statistically significant differences (a). This allows clear grouping of the species when viewed on an exploratory principal component analysis (PCA) scores plot (b). Adapted with permission from Reference 7. Courtesy of MDPI 2020, Keith Gordon/University of Otago. The ideal Raman instrument for field-deployable food analysis captures the maximum signal in the minimum time, with a high degree of reproducibility, to yield a high SNR and consistent spectra. Sampling optics with a high numerical aperture can help to achieve this, as can a high-throughput spectrometer design of modest spectral resolution. Sensitivity and repeatability are more important than resolution for the analysis of most foods because it is easier to maximize the significance of spectral differences when SNR is high. With the sensitivity of today’s compact Raman instrumentation, the spectra of many foods can be taken in a matter of seconds. To make the transition from lab to reliable field use in the food chain, a Raman instrument also requires good thermal stability and adequate calibration. For a given unit, this includes both factory calibration of wavelength response and a single-point daily calibration with a known Raman standard. This practice corrects for fluctuations in the excitation laser wavelength or spectrometer thermal drift. At a fleet level, all instruments utilizing the same chemometric models should also be calibrated for spectral response using a Raman emission standard prior to deployment to ensure consistent results from every instrument. Making the most of the data While Raman may not be the elusive “tricorder,” it can be trained to address specific problems, even complex ones, very well. One of the most crucial industry gaps lies in the authentication of foods, which often requires laboratory testing or an expert taster. Authentication is essentially a classification problem: Does the sample share the key characteristics of the genuine product, without evidence of adulteration or substitution? The ideal Raman instrument for field-deployable food analysis captures the maximum signal in the minimum time, with a high degree of reproducibility, to yield a high signal-to-noise ratio and consistent spectra. To assess this with Raman spectroscopy, it is necessary to build up a chemometric model to define the meaning of “authentic” for the particular food in question by using a sample set that represents the widest variety of bona fide product possible, as well as a representative range of fraudulent and/or adulterated versions. By taking a nontargeted screening approach that looks for differences in spectral patterns between authentic and fraudulent foods, rather than seeking to identify specific chemical constituents, it is even possible to flag adulteration or substitution when an unexpected substance is used. Luis Rodriguez-Saona’s group at The Ohio State University works extensively in food analysis, often comparing results for Raman, NIR, and MIR spectroscopy to find the best technique to solve particular problems. When looking at authentication, the researchers include a global data set with as much natural diversity as possible to help train the models for real-world use, and they often seize opportunities to assess food quality as well. “With Raman, we can go much further and try to understand minor components present in foods, by taking advantage of the fingerprinting that Raman provides,” Rodriguez-Saona said. His group’s Raman authentication studies of honey2, maple syrup3, extra-virgin olive oil4, and potato chip frying oils5 have leveraged the same spectra to develop quantitative models for specific sugars, oils, and other parameters used to benchmark quality and freshness, extending the benefits of Raman to other needs in the food industry (Figure 2). Figure 2. Raman spectra of olive oils show subtle but significant differences, allowing discrimination between extra-virgin olive oil (EVOO), virgin and general olive oils (VOO/OO), and EVOO adulterated with vegetable oils (EVOO + SO). The same spectra can also be used to quantify relevant parameters of olive oil quality, such as oleic acid content (inset). Adapted with permission from Reference 1. Courtesy of Elsevier 2020. Understanding your application When the questions to be answered and the sample set needed to answer them have been defined, users must go back to their application to understand its unique requirements. Where in the food chain is the answer needed, and will the user be an expert or a novice? Does the instrument need to be hand-held or merely portable, and will analysis be performed on board or in the cloud? What is the value of these answers to the target market? These factors influence the choice of instrumentation and software interface, as well as the viability of the chosen solution. VeriVin, a pioneer in the through-barrier analysis of complex liquids, has invested significant time in considering these questions in relation to a particularly enigmatic sample: wine. They’ve developed solutions for authentication of high-value and high-volume vintages, batch testing, and safety; and they advise new Raman users to think about sampling optics when designing an experiment or product (Figure 3). Figure 3. Solutions provider VeriVin has developed a Raman-based instrument for the analysis of wine in-bottle, allowing authentication and validation of quality. Courtesy of VeriVin. “If measuring through-barrier, the glass signal and attenuation needs to be considered and understood, especially the variability between samples,” said VeriVin’s CEO Cecilia Muldoon. Interaction volume or measurement time may also need to be optimized to prevent subsampling, whether solid or liquid. Some users “raster” or scan the measurement beam to compensate for inhomogeneity. Distilling the data The Raman spectra of food offer a wealth of information that can be distilled into useful insights using chemometric techniques already widely applied in analytical chemistry. Principal component analysis (PCA) is a good first step because it simplifies the data set by identifying spectral patterns that differ between sample types (for example, pure versus adulterated). Each of these component spectra can then be examined for the Raman peaks expected in different sample types to ensure that the spectra make sense chemically. For example, specific amide and fat peak ratios differ between meats, and certain sugar peaks are associated with honey versus syrups. PCA helps to assess and visualize the natural grouping of the samples6, as Figure 1 shows. Well-defined groups predict success with more sophisticated chemometric methods. This sets the stage for classification, in which a model is trained (with known sample groups) to classify an unknown as “authentic,” “pure,” or “safe” based on its characteristic spectral features. The most common chemometric methods used for classification of foods include soft independent modeling by class analogy (SIMCA), partial least squares variants, support vector machines, and neural networks. While each has its nuances, none is the clearly superior modeling technique for Raman in foods. It is common to start with the most familiar technique, or one previously applied, and then explore others until the desired level of prediction sensitivity and specificity is achieved. Solutions provider Hone has fine-tuned this approach, combining the power of neural networks with other methods to develop predictive software and instrumentation based on spectroscopic methods for farmers, researchers, agronomists, supply chains, and food processing industries. The company has automated the optimization of data preprocessing methods and adapts its modeling methods to suit each sample type and goal. Recent successes include using Raman to assess the quality of tea tree oil production in the company’s native Australia (Figure 4). Hone has developed models for over 15 compounds relevant to production and usage, some of which comprise only 0.1% to 0.2% of the sample. Once growers saw Raman’s quantitative abilities and its limit of detection, they began using it to optimize parameters that once seemed out of reach. Figure 4. Tea tree oil farmers in Australia are beginning to use a Raman instrument from solutions provider Hone to quantify key active components such as 1,8-cineole, as well as several minor compounds comprising less than 2% concentration — such as aromadendrene, limonene, and ledene. Courtesy of Hone AG. Bringing Raman to fruition Like all great revolutions, the emergence of Raman in the food chain hinges on a convergence of need, opportunity, and timing. As the demand for rapid food testing grows, research groups are expanding the types of questions that Raman can answer. The availability of smaller and more sensitive instrumentation is facilitating deployment in the field to, in turn, speed evaluation and increase safety with minimal disruption to the food chain. And solutions providers are fine-tuning all of these advancements to deliver application-specific answers that enable financial and life-saving decisions. Raman’s time in the food industry has come, and it is just getting started. Meet the author Cicely Rathmell is vice president of marketing at Wasatch Photonics. She indulges her passion for photonics through research, product management, sales, and technical writing (her favorite). References 1. D.I. Ellis et al. (2015). Point-and-shoot: rapid quantitative detection methods for on-site food fraud analysis — moving out of the laboratory and into the food supply chain. Anal Methods, Vol. 7, Issue 22, pp. 9401-9414. 2. D.P. Aykas et al. (2020). Authentication of commercial honeys based on Raman fingerprinting and pattern recognition analysis. Food Control, Vol. 117, p. 107346. 3. K. Zhu (2019). Fingerprinting maple syrup by vibrational spectroscopy and pattern recognition. Doctoral dissertation. The Ohio State University. 4. D.P. Aykas et al. (2020) Non-targeted authentication approach for extra virgin olive oil. Foods, Vol. 9, Issue 2, p. 221. 5. S. Yao et al. (2021). Rapid authentication of potato chip oil by vibrational spectroscopy combined with pattern recognition analysis. Foods, Vol. 10, Issue 1, p. 42. 6. D. Sun, ed. (2008). Modern Techniques for Food Authentication. Academic Press. 7. C. Robert et al. (2021). Rapid discrimination of intact beef, venison and lamb meat using Raman spectroscopy. Food Chem, Vol. 343, p. 128441.