For as long as wine has been made, it has been manipulated, adulterated and counterfeited. It was a huge problem in ancient Rome and it’s still a problem today with rough estimates indicating about 5 percent of wines sold globally are counterfeit.
While ancient Romans were scientific pioneers in their own right, they didn’t exactly have access to the advanced analysis techniques today’s modern laboratory affords. Now, a new study by Australian researchers further builds on these techniques by combining fluorescence spectroscopy and machine learning to develop a method that is simpler, cheaper and more accurate.
Over the years, researchers have developed a myriad of techniques to authenticate wine. Elemental profiling is seen as the most reliable as it can connect wines to a specific geography region by tying the soil’s elemental composition to that of the wine grapes grown and harvested. Isotope ratio mass spectrometry, amino acid profile, LC-MS, GC-MS and more have all been successfully used for elemental profiling. The most common technique, however, is inductively coupled plasma-mass spectrometry, or ICP-MS. ICP-MS is accurate but highly technical, requiring a good chunk of time and resources for sample preparation and analysis.
That’s why David Jeffery, associate professor of wine science at the University of Adelaide (Australia), and his team turned to fluorescence spectroscopy—in addition to an instrument normally used for water analysis. For the study, published in the journal Food Chemistry, the researchers looked at Cabernet Sauvignon from three different regions of Australia as well as Bordeaux in France.
“Fluorescence spectroscopy was used to record an excitation-emission matrix (EEM) based on the fluorescent compounds in wine, which then acts like a molecular fingerprint,” Jeffery explained to Laboratory Equipment. “The composition of wine is affected by the region where the grapes are grown and this information is embedded in the fingerprint of a wine.”
The researchers also used HORIBA Scientific’s Aqualog spectrofluorometer, which is built on the A-TEEM (absorbance-transmission and fluorescence excitation emission matrix) spectroscopic technique. While A-TEEM was developed for and has typically been used for the analysis of organic matter in water, Jeffery and his team were successful in applying it to the analysis of phenolics in wine.
After EEMs were recorded for each Cabernet Sauvignon, the data was processed and modeled with a machine learning algorithm. In the past, data analysis techniques have been used to classify wine according to geographic origin, but in this instance, Jeffery and team wanted to see if a machine learning algorithm proved more accurate. Their hypothesis was correct.
Machine learning with ICP-MS data yielded 98% accuracy, while combining the algorithm with fluorescence data resulted in perfect 100% accuracy.
With Cabernet Sauvignon on the board, Jeffery said his team intends to look into additional types of wines, including Shiraz and wine blends, which contain two or more grape varietals.
“We want to determine how sensitive the technique is in terms of different proportions of other grape varietals being blended in,” the associate professor said. “We also want to determine which molecules are responsible for differentiating the wines from different regions. It’s likely that they will have some relevance to wine sensory properties, so we can then consider how the underlying chemistry influences what consumers may perceive when drinking wines from different regions.”
Photo: Ph.D. student Ruchira Ranaweera and Associate Professor David Jeffery load a wine sample into the Aqualog spectrofluorometer. Credit: University of Adelaide.