Novel non-specific liquid fingerprint technology for wine analysis: a feasibility study



Background and Aims

A novel, rapid and simple liquid fingerprinting technology is described and demonstrated for wine identification and for quality control.

Method and Results

The wine sample, selected chemical modulators on the surfaces of an array, and a long lifetime luminescent europium label interact non-specifically providing a unique luminescence fingerprint that is highly wine specific. The technique was applied to 15 red wines of different vintages from four European vineyards. The fingerprint data, in addition to identification and after data processing, show a significant correlation with the results from existing Fourier transform infrared spectroscopic and spectrophotometric methods of wine analysis.


Identification of individual wines through specific luminescent fingerprints provides a simple and efficient tool to combat wine adulteration and fraud. The same principles combined with proper data processing can enable the monitoring of other parameters such as wine aging.

Significance of the Study

This study demonstrates a fast, affordable and rapid test platform for red wine analysis.


Wine is a complex of molecules that defines its taste, aroma and mouthfeel. These molecules vary from flavonoids to carbohydrates and phenolic substances to salts (Styger et al. 2011). Despite the advances in analytical technology, the human senses are often superior in detecting trace amounts of flavour-affecting compounds (Saenz-Navajas et al. 2012). Generally for quality control and regulatory purposes, several parameters, such as residual sugars, alcohol content, dry extract and acidity, are monitored. Official wine analyses – when needed – often follow the methods published by bodies, such as the Organisation Internationale de la Vigne et du Vin and European Union (EU). These analyses are typically time consuming, and therefore simpler non-specific methods, such as Fourier transform infrared spectroscopy (FTIR) or near infrared spectroscopy (NIR), are often used in practice (Gishen et al. 2010).

The main benefits of non-specific methods are minimal sample preparation and rapid measurements after calibration has been established (Anzenbacher et al. 2010). Sensor arrays combining multiple non-specific sensors relying on pattern recognition or fingerprinting have become popular in the research world during past 10 years (Lozano et al. 2006; Aleixandre et al. 2008; Hufnagel and Hofmann 2008). They are, however, not readily commercially available (Peris and Escuder-Gilabert 2009). The benefit of using an array of sensors is that it improves the sensitivity of the assay and also enables the usage of more specific measurement chemistries as part of the array. When the system is carefully designed, all the benefits of NIR or FTIR measurement are preserved.

This paper presents a luminescent fingerprinting-based method for analysis of wine. The method is based on the use of long luminescence lifetime europium chelate label and individual cross-reactive modulators in an array form. The europium label signal, by default, is highly sensitive to the environment of the label, enabling the detection of the weak interactions among the label, the modulators and the sample. The resulting time-resolved fluorescence (TRF) signals from the array are used to generate a digital fingerprint for a given sample. Figure 1 shows the principle of the fingerprinting assay. The obvious benefit of non-specific array strategies is their ability to react with many compounds with a limited number of sensors. For example, the human nose has approximately 350 different active olfactory receptor types, enabling us to distinguish well over 10 000 different smells (Breer 2003). The cross-reactivity of human olfactory sensors explains why it is possible to distinguish 30 times the number of smells than with a similar array of highly specific sensors.

Figure 1.

The principle of the fingerprinting method. The assay components and the sample interact in different ways: (a) the lanthanide chelate having no interaction with the sample is bound to the surface, and the luminescence is quenched; (b) the sample has higher affinity to the surface than the lanthanide chelate, and it outcompetes the chelate; and (c) the sample has high affinity toward chelate, and surface binding is prevented. Wine samples containing multiple sample components provide a degree of complexity higher than that presented here, leading to an increased number of potential interactions. UV, ultraviolet.

The luminescence of lanthanide complexes has been widely applied in bioaffinity assays and also in non-specific assays because of their extremely large dynamic measurement range, varying from 10−18 to 10−9 moles of label/reaction vessel (Weiss 1999, Härmä et al. 2001). The photoluminescence of europium complexes as well as other lanthanide complexes is long lasting as the lifetime of the luminescence lies typically in the millisecond range. Pulsed excitation and gated detection provide extreme assay sensitivity; autofluorescence and light scattering no longer have an impact on the gated (delayed) luminescence readout. Typically, the lanthanide complexes are excited at around 320–360 nm, and luminescence is monitored at 615 nm using >100 microsecond gating (Bünzli and Eliseeva 2011). The high Stokes' shift and narrow emission band further improve the signal to background ratio of the specific read-out signal.

The luminescence intensity is highly dependent on the chelating structure and the environment surrounding the lanthanide metal ion. Free radicals can destroy the ligand, and low pH promotes the dissociation of the ion from the ligand. Coloured compounds and metal ions can also affect the energy-transfer processes within or with the chelate if applied in sufficient concentration (Mathis and Bazin 2011). In addition, water molecules are known to quench the lanthanide luminescence – also larger macromolecules may have an impact by increasing (protecting) or decreasing (quenching) the luminescence quantum yield. The interaction of europium ion, ligand and modulator with the sample influences the level of the luminescence signal measured. These interactions are responsible for the creation of a sample-dependent fingerprint.

The aim of this study was to demonstrate the non-specific fingerprinting technique as a potential tool for identification of wines and for quantitative prediction of their measurable parameters.

Materials and methods

Red wine samples

Alko Oy (national alcoholic beverage retailing monopoly in Finland) provided 15 European red wines for testing. Wines of several vintages were selected on the basis of their availability (see Table 1 for details). The selected wines were stored horizontally to preserve the cork in a controlled environment during their lifetime at Alko Oy.

Table 1. Wine samples selected for the study
Alko's identifierWineVintageCountry of origin
2012-00236-1Chateau Kirwan1985France
2012-00237-1Chateau Kirwan1989France
2012-00238-1Chateau Kirwan1993France
2012-00239-1Chateau Kirwan1998France
2012-00240-1Travaglini Gattinara1995Italy
2012-00241-1Travaglini Gattinara1998Italy
2012-00242-1Travaglini Gattinara1999Italy
2012-00243-1Travaglini Gattinara2006Italy
2012-00249-1Argiano Brunello di Montalcino1998Italy
2012-00250-1Argiano Brunello di Montalcino2000Italy
2012-00251-1Argiano Brunello di Montalcino2001Italy
2012-00252-1Argiano Brunello di Montalcino2003Italy
2012-00253-1Muga Reserva2004Spain
2012-00254-1Muga Reserva2005Spain
2012-00255-1Muga Reserva2007Spain

Liquid fingerprinting modulators

The assay modulators were obtained from Aqsens Oy (Turku, Finland). Fourteen modulators (Table 2) were measured under eight conditions. Modulators were applied on standard microtiter plate plates (Nunc A/S, Roskilde, Denmark)) using a simple procedure: the modulators were dispensed to each microtiter well in a volume of 40 μL and subsequently dried under airflow for 16 h.

Table 2. Surface modulators coated on microtiter wells
Modulator (supplier, product number)Concentration
Polyethyleneimine (Nippon Shokubai, Epomin SP-003)5%
L-arginine (Sigma-Aldrich, #A5006),1%
L-Cysteine (Sigma-Aldrich, #W326305)1%
Poly(allylamine hydrochloride) (Sigma-Aldrich, #283215)0.1%
Poly(sodium 4-styrenesulfonate) (Sigma-Aldrich, #243051)0.1%
Polyvinylpyrrolidone (Sigma-Aldrich, #PVP10)0.1%
′Y-globulins from bovine blood (Sigma-Aldrich, #G7516)0.1%
Gold(III)chloride hydrate (Sigma-Aldrich, #50790)1 mmol
Copper(II)chloride (Sigma-Alrich #222011)0.1 mmol
Iron(III)Chloride (Sigma-Alrich #157740)10 mmol
New Fuchsin (Sigma-Aldrich, #72200)10 μmol
Delfia assay buffer (Perkin-Elmer, #1244)1:10 dilution in H2O
Rhodamine 800 (Sigma-Aldrich, #83701)15 μmol
Maxisorb Nunc C12 (Nunc, #437915)N/A

Sample preparation

Buffer solutions

The following buffer solutions were prepared:

phosphate buffer: pH 7.0, 10.4 g/L Na2HPO4 ⋅ 12H2O 3.3 g/L NaH2PO4 · 2H2O

carbonate buffer: pH 11.0, 5.3 g/L Na2CO3, pH adjusted with HCl, and

TRIS-HCl buffer: pH 8.0, 6.1 g/L, pH adjusted with HCl.

Assay conditions

Eight different assay conditions grouped into three different categories were tested:

  • Varying buffer: phosphate, carbonate or TRIS-HCl

Phosphate, carbonate or TRIS-HCl buffer (4600 μL) was added to 250 μL of wine sample, and 150 μL of 1 μmol europium chelate 2,2′2”-{[4′-(4-aminophenyl)-2,2′:6′,2”-terpyridine-6.6”-diyl]-bis-methylenenitrilo)}-(tetrakis-acetic acid europium salt were added (Mukkala et al. 1993).

  • Chemical treatment: H2O2, NaOH, HCl

A sample of wine (250 μL) was mixed with 25 μL of 30% H2O2, 1 mol NaOH or 1 mol HCl. The mixtures were incubated for 30 min. Thereafter, 4575 μL of phosphate buffer and 150 μL 1 μM europium chelate were added.

  • Compound addition: epomin, bovine serum albumin (BSA)

In a tube, 250 μL of wine was mixed with 50 μL of 1 g/L polyethyleneimine (Epomin SP-003, Nippon Shokubai Co., Ltd., Osaka, Japan) or with 1 g/L BSA (Sigma-Aldrich, St. Louis, MO, USA, A7906) solution. After 30 min of incubation, 4550 μL of 50 mmol phosphate buffer and 150 μL of 1 μmol europium chelate were added.

The fingerprinting method

The prepared samples were added to microtiter wells in duplicate. The microtiter plates were read with standard TRF-reader Wallac Victor 1420 HTS (Perkin-Elmer, Wallac Oy, Turku, Finland) with its standard europium detection settings (excitation filter D340 nm, emission filter D615 nm, delay 400 μs, integration window 400 μs, cycle 1000 μs) after a 20-min incubation. All measurements were repeated twice using samples from the same bottle for all wines except for the 2006 Travaglini Gattinara that was measured five times.

Accredited methods

Reference analyses were performed by the Alcohol Control Laboratory (ACL) of Alko Oy, the official laboratory for alcoholic beverage analyses in Finland. Wine samples were analysed using accredited FTIR and UV-VIS methods for ethanol, total acid, volatile acid, glucose and fructose concentration; pH and dry extract were measured with a WineScan FT120 instrument (Foss, Hillerød, Denmark). Samples were diluted 1:20 with distilled water, and their absorbance at 280, 420, 520 and 620 nm was determined using an Arena 20XT automated photometric analyser (Thermo Scientific, Vantaa, Finland).

Data handling

Fingerprinting data were subjected to principal component analysis (PCA) with the R and FactorMineR package (Lê Sébastien and Husson 2008, R Core Development Team 2008). The data were mean centred and standardised before the PCA calculations. All the graphs in this publication have been created with R.

Aging and prediction of chemical parameters

To assess vintage, the data set was divided into training and testing sets of equal size (15 wines in both). The training and testing sets contained wines from all vintages. Training was undertaken on the known vintages utilising partial least squares (PLS) regression method of the Molegro Data Modeller (2009.2.1.0) (CLC bio, Aarhus, Denmark). On the test set (training set), the Pearson correlation was 97% (100%), and the Spearman rank correlation was 96% (99.6%).

Chemical parameters [volatile acidity (g/L), total acidity (g/L), dry extract (g/L), colour intensity, A450 nm/A520 nm, A280 nm, and alcohol content (% v/v)] were estimated by fitting a regression model to the data with the REG procedure in SAS (SAS Institute Inc., Cary, NC, USA) using the linear regression model with stepwise variable selection. Models were fitted separately for each chemical parameter, and individual variables of the fingerprint data were used as explanatory variables. All the variables (surfaces) in the models are significant at the P < 0.05 level.

Results and discussion

Liquid fingerprint method

A simple method was developed to demonstrate that wines can be identified and their parameters can be determined using a luminescence fingerprinting approach. The wine samples were selectively pretreated under given conditions using, for example, varying pH or an oxidising agent. The pretreatments changed the chemical composition of the samples, generating a more pronounced sample variation to enhance separation by fingerprinting. Pretreated and untreated samples were diluted with assay buffer, and europium chelate label was added. The dilution to an assay buffer generally improves the reproducibility of the results as pH is more controlled. The samples were incubated with different surface modulators on microtiter plates for 20 min, and each well was read in a standard TRF reader.

The fingerprint of a sample can be modified dramatically by choosing appropriate assay conditions. Thus, the optimisation of the modulators and measurement conditions is important to optimally address any given question. In the current study, our questions were related to the correlation of traditional quality-control parameters of wines. The initial selection criterion of modulators before the data analysis was the reproducibility of the measurement. This was assessed by calculating the standard deviation for repeated measurements of each modulator independently. All modulators having a coefficient of variation >5.0% were discarded (data not shown). This analysis led us to select 14 modulators and eight measurement conditions. Often, these luminescence signals differ more than 20% leading to a robust separation while the coefficient of variation is less than 5.0%.

The main difference between most e-noses and the developed luminescence fingerprint method is that the whole liquid is measured and not solely the gaseous phase or ion species. The non-specific sensors are typically constructed to monitor a broad spectrum of varying compounds or multicomponent samples. Detecting multiple non-specific parameters using the fingerprint approach can potentially give a highly specific detection means to a given multicomponent sample without any limitation to the size of the sample molecules – the specific fingerprint of the sample. This has been exploited in the current study to estimate classical wine quality-control parameters using statistical tools to correlate the liquid fingerprint data produced to that of the existing laboratory methods.

Control analyses

Wine quality-control parameters were analysed at the accredited ACL of Alko Oy by FTIR-based and spectrophotometric methods. The measurements were made on the same day as the liquid fingerprinting method. The recorded values are in the typical range of red wines. The results are summarised in Table 3. Variations in these chemical and physical properties alter the character of wine, e.g. in bouquet defined by a vast amount of different aroma compounds. These quality-control parameters are also the basis for regulation and legislation. For example, alcohol content is measured for consumer information and to meet legal demands (9–15% v/v). Within the tested red wines, the alcohol proportion by volume varied within the legal requirements for wines. The pH of wines was nearly constant. Red wines are most often dry, and the amount of residual glucose and fructose is thus low. Models predicting a certain characteristic of wine could be trained only if the property varied among wines and was in the optimal range for the control methods.

Table 3. Range of attribute values for the 15 European red wines analysed by the Alcohol Control Laboratory of Alko Oy, Finland
Measured attributesRange
Alcohol content (% v/v)11.96–14.48
Total acidity (g/L)4.6–6.40
Volatile acidity (g/L)0.56–1.02
Residual glucoseBelow detection limit
Dry extract (g/L)24.4–31.0
Colour intensity (A 420 nm + A 520 nm + A 620 nm)5.3–12.5
Colour nuance (A 420 nm/A 520 nm)0.88–1.35
Relative amount of phenolic substances (A 280 nm)52.7–70.8

Wine identification

Over the course of the experiments, multiple modulators and assay conditions were tested. The highest separation between different wine groups was obtained when data from two of the assay conditions were combined (wine diluted to phosphate buffer and BSA addition). The selection was based on PCA. Using selected conditions, wines were fingerprinted, identified and grouped independently of the vintage. The luminescence signal difference of four wine groups was high and clear in the raw data, for example, 6601 ± 1126 photon counts for Muga reserve vintages and 35 103 ± 5597 photon counts for Travaglini Gattinara vintages (surface AQ2 protocol 10). The PCA plot for the subset of the fingerprinting data is presented in Figure 2.

Figure 2.

Identification of wine groups: Argiano Brunello di Montalcino 1998–2003 (○), Chateau Kirwan 1985–1998 (△), Muga Reserva 2004–2007 (+), Travaglini Gattinara 1995–2006 (◇), and each individual wine is distinguished with selected surface modulators and assay conditions that are different from the combination selected for the vintage evaluation.

The first three PCA components were found to be significant in the analysis (data not shown) as the first two PCA components explained 89.2% of the variation in the data. This is in line with our study on 20 red wines reported elsewhere indicating that under selected assay conditions and modulators, wines can be identified and grouped according to their origin (Hänninen et al. 2013). Therefore, the liquid fingerprint technology can be considered as a potential method for wine identification. The results suggest that with further validation, the method can also be turned into a vehicle for monitoring wine authenticity. It is generally considered that independent of the vintage year, the similar soil, climate and grape cultivar as well as similar winemaking approaches create unique chemical characteristics of wines. Especially a holistic non-specific analysis provides a potential tool to link wine aging and origin in the authenticity analysis.

The EU collects yearly a wine databank by analysing a large number of wines mainly with isotopic methods (European Commission 2000, 2008). If a reference sample is not available, more sophisticated methods are required. These include a good experience of wines in question and extensive data bases of the chemical composition, and the use of statistical multivariate analysis, e.g. PCA or discriminant analysis. The developed liquid fingerprint method presents an alternative method to analyse wines with a low cost and user-friendly approach.

Prediction of wine chemical parameters using the fingerprint data

By definition, the interactions between the sample and the non-specific analysis components are unknown and impossible to explain. Multivariate statistics are extremely helpful to link luminescence fingerprints and specific qualities of the sample. Here, we used linear regression methods for the fingerprint data to demonstrate that quantitative information can be extracted to estimate the chemical quality of the wine. All the parameters having a meaningful extent of variation measured with the traditional analysis methods were predicted from the fingerprint data with nearly equal performance (Figure 3a–g). All linear regression models have F values ≥60 corresponding to P values <0.0001. Parameter estimates for surfaces and intercept are presented in Supporting Information Table S1. In general, the results were obtained with multiple modulators and assay conditions, and the combination of these modulators and conditions varied to predict each parameter. Because the pH and the amount of residual sugars varied insignificantly or were below the detection limit with the FTIR method, no reliable prediction model could be built.

Figure 3.

Correlation data of the fingerprint and the reference methods for (a) total acidity (g/L), (b) alcohol content (% v/v), (c) volatile acidity (g/L), (d) dry extract (g/L), (e) absorbance 280 nm, (f) color nuance and (g) colour intensity. The chemical parameters were predicted using linear regression, and (i) the vintage was predicted with partial least squares (PLS) regression model (h,j). The raw data for a selected set of surface modulators show a relatively good correlation to colour intensity and vintage.

The assay repeatability was found to be equal with the data measured at ACL, which operates under the quality standards SFS-EN ISO/IEC 17025 and is accredited by Finnish Accreditation Service. The repeatability results are presented in Table 4. Although high precision was shown, the prediction of ethanol concentration was somewhat compromised because the accuracy was lower than that of the reference method.

Table 4. The repeatability of the fingerprint (Aqsens) and the reference assays (Alko) for the 2006 Travaglini Gattinara red wine. The accuracy has been calculated from five independent repeats for the fingerprint method
Volatile acidity (g/L)0.91±0.020.95±0.06
Total acidity (g/L)5.7±0.035.7±0.16
Dry extract (g/L)27.9±1.4028.2±0.83
Colour intensity (A420 nm + A520 nm + A620 nm)5.31±0.015.17±0.31
A450 nm/A520 nm1.24 1.25±0.02
A280 nm63.6±0.6463.4±0.59
Alcohol content (% v/v)13.5±0.0313.5±0.36

Wine aging

Wine is an extremely complex medium having multiple aging processes (Von Baer et al. 2008). Some wines mature within a year whereas some wines are optimally mature in 20 years and are still drinkable after 50 years (Chira et al. 2011). Traditionally, these changes have been monitored by UV-VIS spectrophotometry by measuring anthocyanin equilibria, total phenolics, free and molecular SO2 (Somers 1971, Somers and Evans 1977). The main problem with UV-VIS methods is that the reactions that are responsible for the transformations are sensitive toward conditions during storage (temperature, oxygen and light) (Esparza et al. 2004). Initial concentration and the exact species of anthocyanins vary depending on the grape cultivar and winemaking methods. The use of colour and anthocyanins as global estimates for wine vintage is extremely difficult because they reflect the storage conditions and chemical age of wine.

The fingerprinting method described here is sensitive toward changes in colour and can perform equally with current methods. The raw data for specific fingerprint surfaces result in a relatively good correlation with the expected colour intensity and vintage values for a group of unrelated wines (Figure 3h,j). Fitting a PLS regression model to the entire data set further improves the predictability (Figure 3i). This result suggests that there could be unknown global changes in wine that would permit the determination of the vintage with fingerprinting (Del Alamo et al. 2004). The main limitation here is that in the present work, the sample set is small, and a larger set of wines needs to be investigated to confirm the findings.


Novel analytical methods add value and can provide winemakers key parameters for critical assessment of winemaking processes from grape to end product as well as their authenticity, especially with respect of their age and eventually their origin. Gathering long-term information together with the winemaker's valuable hands-on experience supports the decision-making process for improved control of the wine processes and chemical quality of the end product. Our data suggest that wine quality-control parameters can be potentially monitored using the developed rapid, affordable and simple liquid fingerprint technology, finding application as a routine analytical test platform during the wine production process and the lifecycle of wines at the wine production sites.