Reduction in Off-Flavors in Wine Using Special Filter Layers with Integrated Zeolites and the Effect on the Volatile Profile of Austrian Wines

Abstract

In the course of the present study, filter layers with embedded zeolites (patented layer of the company Filtrox (Zwingen, Switzerland) with the brand name Fibrafix^® TX) were tested for the elimination of the wine defects, cork taint (2,4,6-trichloroanisole), mouldy aroma (geosmin), and strong “animal” phenolic aromas (4-ethylphenol, 4-ethylguaiacol, 4-ethylcatechol). The test design allowed a comparison with conventional filter layers (sterile filter layers) in a small-scale trial (25 L) as well as in a large-scale trial (125 L). By means of gas chromatography-mass spectroscopy, not only the impact compounds of the wine faults were analysed, but also the loss of volatile substances such as 15 free monoterpenes, 34 ester compounds, and 24 wood flavours. Sensory analyses were carried out by means of expert panels. The Fibrafix^® TX layers were satisfactory with regard to the reduction in 2,4,6-trichloroanisole (the lead substance of cork flavour; reduction > 90%) and geosmin (the lead substance of mould flavour; reduction > 75%), but not with regard to the reduction in 4-ethylphenol, 4-ethylguaicol, and 4-ethylcatechol. However, the reduction in these off-flavours was generally not specific enough to prevent a loss of volatile compounds, especially various ester compounds and free monoterpenes. The wines obtained after filtration (in case of contamination with geosmin or TCA) were sensory faultless, but for some of the tasters, the overall quality of these wines did not correspond to the quality wine clasification.

1. Introduction

The quality of Austrian wines is highly praised internationally. The strict quality control of the wines also contributes to this. By law, Austrian quality wine must be sensory and analytically controlled (basic parameters); in the case of a positive decision, this wine receives a quality control number [1,2]. With an average annual harvest of about 2.4 million hl, an official quality number application was submitted for about 1.89 million hl in Austria in 2020. Out of 36,805 applications, 4669 (about 12.6%) were rejected [3]. Evaluations of reasons for rejections are rare, but [4] provided an overview of the reasons for rejection from the period from 1 January 2016 to 13 May 2016 from four Austrian control authorities. It was found that “mold, muffled, musty” was the most frequently mentioned reason for rejection, with 13.18% ahead of “lack of varietal character” (13.09%), “Böckser (reductive note)” (12.61%), and “tannin” (8.91%). Additionally, about 1% of the rejections were due to an “animal” phenolic taste.

Mould-associated compounds in Austrian wines can still play a negative role in the 21st century. While it was previously mainly the cork taint, primarily caused by trichloroanisole until the comprehensive changeover to screw caps or selected natural corks, that was most problematic, geosmin is nowadays the bigger problem synergistically with other mould aromas [5,6,7]. A new paper summarised a number of possibilities for the reduction in already existing mould aromas in wine and tested in a practice trial the depth filtration using filter sheets with embedded zeolites. In the process, an approximately 85% reduction in geosmin could be achieved [8]. The use of such filter layers with zeolite Y faujasites was already approved by resolution OIV-OENO 444-2016 in 2016 [9]; the principle behind this technique is a molecular sieve and zeolite Y faujasites synthesised minerals for flexible and stable design. [10]. The patent-pending depth filter sheet FIBRAFIX^® TX-R and has been developed by the company Filtrox group (Zwingen, Switzerland) for the adsorption of haloanisoles. The effectiveness of the product Fibrafix TX-R in the removal of 2,4,6-trichloroanisole (TCA) and 2,4,6-tribromoanisole from wine has been clarified for some time [11,12], but recent results show that, in addition to the findings regarding the activated carbon, which is damaging to the wine quality [13,14], these layers are also a good alternative for the reduction in mould aromas [8,15]. Some unconfirmed practical reports even attest to a certain reduction power of these layers against the leading substances of the wine defect “animal” phenolic character. The authors are not aware of any experiments in this regard.

Even though the nasal and retronasal perception of wine faults is very complex and, above all, depends essentially on the aroma buffer of the wines [16], sensory threshold values can help assess the effectiveness of methods for reducing wine faults. The determination of sensory and olfactory thresholds is not a trivial task from many points of view, as they depend on very many partly subjective parameters. Accordingly, there is a respective range of threshold values in the current literature. The range of threshold values for mould aroma (impact substance: geosmin) [17] in the literature is 17.0 to 90.0 ng geosmin/L [17,18,19]. The range for cork taint (impact substance: TCA) [20] is 1.4 to 210.0 ng TCA/L [21,22,23,24], and for animal phenolic aroma (impact substance: 4-ethylphenol (4-EP), 4-ethylguaiacol (4-EG), 4-ethylcatechol (4-EC)) [25,26] is 50–968 µg 4-EP/L [27,28,29,30,31,32], 33–135 µg 4-EC/L [27,28,29,30,31,32,33], or 0.9 to 33 µg 4-EG/L [32,33,34]. A more detailed definition of the threshold value or the method used for determination is usually not specified. A very important factor is the medium in which the substance is present, and in most cases, not all the data have been made available either. The problem with threshold values is that new techniques should not be finally assessed purely analytically but must include a sensory study. For this reason, the results of [8,15] should be verified since no sensory analysis was conducted in either study.

Besides the effectiveness in terms of minimising wine defects, it is, of course, necessary to assess the effect of these filter layers on the overall quality of the products after filtration, especially with regard to the volatile profiles of the wines and overall sensory perception. Already [11] compared wines filtered with layers with the unfiltered wine regarding some main aroma compounds, mainly higher alcohols, carboxylic acids, carbonyl compounds, and a few ester compounds, and carried out sensory tests. It was found that the wines filtered with Fibrafix^® TX-R layers were sensory but also analytically only marginally distinguishable from the noncontaminated wines and that the lead substance of the cork flavour, 2,4,6-trichloroanisole, was sufficiently reduced. However, additional studies should be carried out because specific aroma groups, especially free monoterpenes, wood aroma substances, and a wide range of ester compounds, can significantly influence the aroma of the wines. From the results of [11], it can also be deduced that the four ester compounds investigated were reduced much more than the analysed higher alcohols, so a detailed look at ester compounds would make the most sense because it can be assumed that there is a stronger aroma loss. Little is known about the aroma loss in conventional depth filtration with sterilisation filter layers, so the comparison of Fibrafix^® TX-R sheets with conventional depth filter layers would make sense throughout. While Fibrafix^® TX-R sheets are made of purified and bleached cellulose with the natural inorganic filter aids zeolite and polyamidoamine (<3%), the conventional depth filter layers for sterilisation are made of purified and bleached cellulose with natural inorganic filter aids diatomaceous earth (DE, Kieselguhr) and perlite. Significant differences in aroma loss and reduction in bad taste aromas are to be expected already because of the different compositions of the used filter layers.

The aim of this study was to test the filter sheets of Filtrox (FIBRAFIX^® TX-R, Zwingen, Switzerland) for the reduction in mould aroma (geosmin) and cork taint (TCA) and, for the first time, for the reduction in the impact compounds of “animal” phenolic aroma (4-EP, 4-EG, 4-EC). Additionally, we planned to obtain an overview of the decrease in important volatile compounds such as a number of major and minor esters, free monoterpenes, and wood aroma compounds. A comparison with conventional depth filter sheets was also included in the study. These tests were aimed not only at laboratory scale but also at practice-relevant experimental designs, and not only analytical but also sensory tests were carried out. Filtrox also provided a newly developed filter sheet for the special reduction in “animal” phenolic aroma for testing. These additional layers have been examined in a practice-relevant experiment for their reduction power in relation to the leading substances of “animal” phenolic aroma, and the results have been put into context with the results of filtration with the patented Fibrafix-TX-R layers.

2. Materials and Methods

2.1. The Trial Wines

Four Austrian quality wines were used for the study:

Trial wine A: white, cuvèe 2019, origin Klosterneuburg (Austria), 12.5% vol, 5.3 g/L titratable acidity calculated as tartaric acid, total sugars nondetectable (n.d.), SO₂ free 38 mg/L, SO₂ total 98 mg/L;

Trial wine B: white, Grüner Veltliner 2018, origin Klosterneuburg (Austria), 12.5% vol, 5.4 g/L titratable acidity calculated as tartaric acid, total sugars 1.9 g/L, SO₂ free 47 mg/L, SO₂ total 101 mg/L;

Trial wine C: red, Zweigelt 2018, aged in wooden barrels, origin Klosterneuburg, integrated production, 13.0% vol, 4.6 g/L titratable acidity calculated as tartaric acid, total sugars n.d., SO₂ free 42 mg/L, SO₂ total 112 mg/L;

Trial wine D: red, varietal blend 2020, aged in wooden barrels, origin Klosterneuburg, integrated production, 12.8% vol, 4.6 g/L titratable acidity calculated as tartaric acid, total sugars <0.6 g/L, SO₂ free 50 mg/L, SO₂ total 110 mg/L.

The trial wines were filtered with a crossflow filter (Microza Hollow Fibre Module, type PWUSP-543, Asahi Kasei Corporation, Chiyoda, Japan) before the experiment started. Trial wine A was then treated with geosmin (target concentration: 32 ng/L; 97% purity, Sigma Aldrich, St. Louis, MO, USA) and trial wine B with 2,4,6-trichloroanisole (target concentration: 40 ng/L, 99% purity, Sigma Aldrich, St. Louis, MO, USA). The trial wine C was tested with 4-ethylphenol (target concentration: 700 µg/L, 99% purity, Sigma Aldrich, St. Louis, MO, USA), 4-ethylguaiacol (target concentration 100 µg/L, 98% purity, Sigma Aldrich, St. Louis, MO, USA), and 4-ethylcatechol (target concentration 350 µg/L, 95% purity, Sigma Aldrich, St. Louis, MO, USA). The trial wine D was tested with 4-ethylphenol (target concentration: 700 µg/L, 99% purity, Sigma Aldrich, St. Louis, MO, USA), 4-ethylguaiacol (target concentration 100 µg/L, 98% purity, Sigma Aldrich, St. Louis, MO, USA), and 4-ethylcatechol (target concentration 350 µg/L, 95% purity, Sigma Aldrich, St. Louis, MO, USA). For this purpose, stock solutions of the standards were prepared in ethanol (99% (AustrAlco Österreichische Alkoholhandels-GmbH, Spillern, Austria). Before the addition of the impact compounds of the wine defects, 10 bottles of trial wines A–C were filled for sensory and chemical analysis and stored at 4 °C until analysis. Trial wines B and C were used only for the 25 L scale trials, while trial wine A was also used for the large-scale 250 L scale trial. Only trial wine D was used for testing the prototype for the reduction in “animal” phenolic aromas.

The variants are shown in Supplementary Material Table S1. All experiments were carried out in triplicate.

2.2. Experiment on a 25 L Scale (125 L per m² Filtration Surface)

In this experiment, the trial wines A-C were each transferred into 9 glass bottles (25 L) with a CO₂ overlay. Three of these bottles were used as control (variant 1), three bottles were subsequently filtered with standard layers from Filtra company (20 × 20 cm, SEITZ-EKS^®, Guntramsdorf, Austria) (variant 2), and three bottles were filtered with the filter layers to be tested with integrated zeolites from Filtrox group (20 × 20, Fibrafix^® TX-R, St. Gallen, Switzerland) (variant 3). A Schneider pump Reform B (Andreas & Thomas Schneider Maschinenbau GmbH, Bretzenheim, Germany) and a filter press type Pilot (Filtra, Guntramsdorf, Austria) were used. A total of 5 layers were used in each case. The filter area was therefore 0.20 m², and the filtration capacity per run was 125 L per m² of filtration surface. After each bottle, the filter layers were changed and watered again for 10 min. The recommended flow rate for the Fibrafix^® TX-R layers of 350 L/m²h was also maintained as best as possible for the normal layers. The filtration time of 21 min ± 2 min per glass bottle was achieved. After filtration, 10 bottles of 0.5 L each were filled with the air removed (CO₂ overlay) and stored at 4 °C until chemical and sensory analysis.

2.3. Experiment on a 250 L Scale (1250 L per m² Filtration Surface)

In this experiment, conventional sterile filtration layers (EKS layers) were compared with Fibrafix^® TX layers. This test was carried out with the same filtration device under the same conditions (threefold repetition, flow rate), but the total filtration capacity was increased to 250 L (1250 L per m² filtration surface), and samples were taken subsequently during the process for chemical analysis. Therefore the samples were filled into 25 L bottles during filtration, and five 0.5 L bottles were then filled from each 25 L bottle and stored at 4 °C until chemical analysis. Subsequently, a pooled sample was prepared from all 25 L bottles of one variant per replicate. Thus, there were six pooled total samples (three with Fibrafix^® TX layers and three with EKS layers).

2.4. Trial with a Prototype of the Company Filtrox Group (Zwingen, Switzerland) for the Reduction in Strong “Animal” Phenolic Aromas

In this experiment, 250 L of trial wine D was filtered with the special filter layers (40 × 40) (prototype layers with zeolites built-in from the company Filtrox (Zwingen, Switzerland)) and a total filtration surface of 1.16 m². This resulted in a total filtration capacity of 223 L/m². This test was carried out in triplicate. The samples were stored at 4 °C until analysis of the impact compounds. These samples were not sensorily analysed, and no volatile profile of the wines was determined.

2.5. Analysis of the Volatile Profile

The experimental wines (variants 0–3, plus samples from the large-scale trial, plus samples from the prototype trial) were analysed by means of gas chromatography-mass spectroscopy for the off-flavours geosmin (trial wine A), 2,4,6 trichloroanisole (trial wine B), and 4-ethylphenol, 4-ethylguaiacol, and 4-ethylcatechol (trial wine C, D). Additionally, the concentrations of positive volatile compounds, including 34 ester compounds (test wine A–C), 15 free monoterpenes (test wine A, B), and 24 subtle wood aromas (test wine C) (carbonyl compounds, lactones, phenols) were determined. A gas chromatograph from Agilent Technologies (Santa Clara, CA, USA) and a total of five methods were used for the analysis of the different volatile substances. The system, consisting of a 7890A GC system with a 5975C Inert MSD with Triple Axis Detector and a CTC Analytics Autosampler (Zwingen, Switzerland), was equipped with a ZB-5MS column (length: 60 m, I.D.:0.25 mm, df = 0.25 µm) from Phenomenex (Torrance, CA, USA). The quantification of geosmin and TCA was carried out according to the method developed by [5] and modified by [7], using headspace solid-phase microextraction and gas chromatography-mass spectrometry (HS SPME GC-MS) technique by multiple standard additions. The quantification of ester compounds was performed by the partial stable isotope (SIDA)-SPME-GC-SIM-MS method according to [35]. The determination of 15 free monoterpene compounds was performed by the HS-SPME-GC-SIM-MS method and was based on the method published by [36]. Information on the calibration and validation of these methods can be found in the Supplementary Material (Table S2).

The determination of 4-ethylphenol, 4-ethylguaiacol, and 4-ethylcatechol was carried out by means of a liquid–liquid extraction. First, 1.4 g potassium hydrogen phosphate (≥98%; Merck KGaA, Darmstadt, Germany) was weighted into centrifuge tubes. Exactly 10.0 mL of wine was mixed in a volumetric flask with 100 µL of internal standard 2,4-dimethylphenol (target concentration 1 mg/L, Sigma Aldrich Company (St. Louis, MO, USA)), and 3 mL of this mixture was transferred into the centrifuge tube and vortexed until the salt was completely dissolved. These tubes were placed in the refrigerator for approximately 10 min; 200 µL of propionic anhydride was added and mixed intensively for 1 min, then extracted with 0.5 mL hexane, vortexed for 15 s, and centrifuged at 6000 rpm for 7 min. Then, 0.4 mL of the supernatant was removed, and 1 µL of it was injected into the GC system. The injector temperature was 250 °C, and the procedure was operated in splitless mode. The starting temperature in the oven was 50 °C with a hold time of 3 min. The temperature was increased at a rate of 20 °C/min to 135 °C, then at a rate of 1.5 °C/min to 166 °C, then at 5 °C/min to 205 °C, and subsequently at a rate of 40 °C/min to 270 °C, where the temperature remained constant for 12.6 min. The total run time was 49.942 min. The mass spectrometer was operated in selected ion monitoring (4-ethylphenol: quantifier: 107; qualifier: 122/178; retention time 19.058 min; 4-ethylguajacol: quantifier: 152, qualifier: 137/208, retention time 27.191 min; 4-ethylcatechol: quantifier: 138, qualifier 123/1974, retention time: 35.825). The calibration was performed in a calibration wine (red wine vintage 2018; sensory inconspicuous, 150 mg/L free SO₂) in the steps 12.5 µg/L, 25 µg/L, 250 µg/L, and 500 µg/L per compound.

Table 1. Calibration and validation data of the method for the analysis of the impact compounds of “animal” phenolic aroma.

Compounds	R2	LOD	LOQ	Reproducibility	Recovery
4-ethylphenol	0.997	0.2 µg/L	0.5 µg/L	2.36%	104.4%
4-ethylguaiacol	0.999	0.1 µg/L	0.3 µg/L	2.61%	100.9%
4-ethylcatechol	0.996	0.1 µg/L	0.4 µg/L	2.47%	92.7%

shows the calibration and validation data.

The limit of detection (LOD) and the limit of quantification (LOQ) were estimated from the peak area to noise ratio. Reproducibility was determined by repeating a red and a white wine four times, and the average relative standard deviation was calculated. The recovery considered the matrix effect and was calculated from five red and three white wines at a spiking level of 250 µg/L (four quality red and two Austrian quality white wines) and 500 µg/L (one quality red and one Austrian quality white wine), respectively, and the average relative recovery was determined. The wines were selected according to the availability from the official wine tasting (all wines were sensorially unremarkable). The red wines used had a native content of 4-ethylphenol of 5–8 µg/L, 4-ethylguaicol of 1–4 µg/L, and 4-ethylcatechol of 3–9 µg/L, and the white wines used had a native content of 4-ethylphenol of 4–14 µg/L, 4-ethylguaicol of 1–7 µg/L, and 4-ethylcatechol of 3–16 µg/L.

2.6. Sensory Tests

The organoleptic evaluation of the 25 L variants was carried out by means of triangular tests within a trial wine. Variant 3 (respective pool sample of the three repeats) was compared with the other variants (variant 0 to variant 2). The expert panel included 15 experts, all of whom had passed an official training for tasters, including an exam. The panel consisted of 11 men (age 18–62 years) and 4 women (age 31–54 years). The tasting took place in the tasting room at a laboratory accredited according to ÖVE/ÖNORM EN ISO/IEC 17025. In addition to the distinctness of the samples, it was also asked which sample(s) was better in terms of quality. The tasting was repeated in duplicate. This led to 30 results for each comparison and a total of three sets of 6 triple tests each. There was a break of 15 min between the sets.

The large-scale trial was also sensory tested using triangle tests, with the three replicates of the total sample (Fibrafix^® TX layers) and three replicates of the total sample (EKS) pooled and compared sensorially with variant 0 and variant 1. The panel consisted of 6 men (age 27–51 years) and 3 women (age 41–54 years). The tasting was carried out in three replicates. This led to 27 results for each comparison and a total of three sets of 6 triple tests each. There was a break of 15 min between the sets. In addition to the distinction between the samples, it was also asked which sample(s) was better in terms of quality. Finally, the nine persons were asked whether the four samples would achieve quality wine status for them in an official tasting and, if not, why they would not classify the wine as quality wine, analogous to the official tasting for the quality wine in Austria [1,2].

2.7. Statistical Evaluation

The statistical analysis was carried out using SPSS 26.0 (IBM, Armonk, NY, USA). It was tested for significant differences in the individual volatile substances and aroma groups between the variants. The sample sets were first checked for normal distribution and variance homogeneity and, if the criterion was met, tested for significance level p ≤ 0.05 using simple ANOVA and pairwise comparisons based on a Tukey B test. If the requirement was not met, a Kruskal–Wallis test with pairwise comparisons based on a Mann–Whitney U test was performed at the significance level of p ≤ 0.10. The sensory results of the triangle tests were evaluated according to the significance table, with the significance level p ≤ 0.05 (significant), p ≤ 0.01 (highly significant), and p ≤ 0.001 (very highly significant) from [37]. For the evaluation of the question of which of the samples is better, only the correct triangle tests were used. The large-scale tests were calculated by means of ANOVA with repeated measurements (10 measurement times) at the significance level of p ≤ 0.05. Normal distribution was assumed for this.

3. Results

3.1. Experiment on a 25 L Scale (125 L per m² Filtration Surface)

3.1.1. Reduction in the Wine Faults

Figure 1a shows the results of the reduction in geosmin, Figure 1b shows the reduction in 2,4,6-trichloroanisole (TCA), and Figure 2a–c the reduction in the impact substances of “animal” phenolic aroma. In the experiment on the reduction in mould aroma substance geosmin, a significant reduction below the sensory threshold for geosmin in white wine (17 to 90 ng/L) [17,18,19] occurred during filtration with Fibrafix^® TX layers. Filtration with normal EKS filter layers, on the other hand, did not lead to any significant reduction. Results concerning TCA show a slightly different picture. Starting from about 40 ng/L TCA, the concentration was significantly reduced below 30 ng/L TCA by simple depth filtration with EKS. Filtration with Fibrafix^® TX layers resulted in a significant reduction below the analytical detection limit for TCA (0.2 ng/L) and sensory threshold for trichloroanisoles of 1.4 to 210.0 ng TCA/L [22,23,24]. In the “animal” phenolic aroma reduction study, on the other hand, the impact compound 4-ethylcatechol was not significantly reduced, and the impact compounds 4-ethylguaiacol and 4-ethylphenol were significantly reduced, but for none of the lead compounds were the reductions decisively below the sensory thresholds of 50–968 µg 4-EP/L [27,28,29,30,31,32], 33–135 µg 4-EC/L [27,28,29,30,31,32,33], and 0.9 to 33 µg 4-EG/L [32,33,34] respectively.

3.1.2. Effect on the Volatile Profile of the Filtered Wines

Figure 3 shows the impact of filtration on the ester profile (sorted by ester families) of the wines, and the Supplementary Material (Tables S3–S5) shows the impact on the individual compounds.

For aromatic esters (Figure 3a), a significant decrease was observed by filtration with Fibrafix^® TX layers for all wines and for red wine, which generally has higher contents [35], and also by normal EKS filtration. For the acetate esters of higher alcohols (Figure 3b), there were also significant decreases by filtration with Fibrafix^® TX layers, while filtration of wine B by means of EKS layers did not lead to any significant decrease, and this was very much the case for trial wine C. The difference between the two wines filtered with Fibrafix^® TX layers was not significant. Only wine B showed a significant difference between the two filtered wines. The sum of ethyl esters is known to be higher in white wines than in red wines [35,38], which can also be seen in Figure 3c. EKS filtration did not lead to any significant reductions in this ester group, whereas filtration with zeolites led to a significant reduction. The reduction was even up to 64% for trial wine A. For the esters with higher alcohols and medium-long-chain carboxylic acids, the aroma loss due to filtration was lower. Only in the case of trial wine C a significant reduction was observed in these ester compounds when filtered with zeolites (Figure 3d); similarly, for the ethyl esters with carboxylic acids with an odd number of carboxylic acids (Figure 3e), a significant reduction was only observed in the case of trial wine B when filtered with Fibrafix^® TX layers. Regarding methyl esters, on the other hand, there was probably a higher loss of aroma. All three wines were affected by a significant loss because of the filtration with zeolites, and a significant reduction was also observed in trial wine A due to EKS filtration. Altogether, there was a significant loss of ester molecules due to filtration with zeolites.

Figure 4 and the Supplementary Material (Tables S3–S5) show the results of the free monoterpene analyses. There was a significant and clear reduction in the sum of these compounds in both test wine A and test wine B through filtration with zeolites, and there was also a significant reduction through filtration with EKS layers, but a significantly smaller one. This is the first study that also puts the content of free monoterpenes in the context of filtration with Fibrafix^® TX layers.

Figure 5 and the Supplementary Material (Tables S3–S5) show the results of the wood flavours. Only in the case of the lactones was a significant loss of aroma due to filtration with Fibrafix^® TX layers observed. The phenols and carbonyl compounds were hardly influenced by the filtration.

3.1.3. Sensorial Examination

The samples filtered with Fibrafix^® TX layers (variant 3) were compared with the other variants in a triangle test. The results can be found in

Table 2. Evaluation of the triangle tests in the small-scale experiment according to the significance table at the significance level p > 0.05 (not significant, n.s.), p ≤ 0.05 (significant), p ≤ 0.01 (highly significant), and p ≤ 0.001 (very highly significant) of [37].

Trial Wine	Comparison	Number of Panelists	Number of Results	Number Correctly Distinguished (Number of Variant 3 Better than the Compared Variant)	Significant Difference	Significant Preference
trial wine A	variant 3–variant 0	15	30	16 (4)	p ≤ 0.05	p ≤ 0.001
	variant 3–variant 1	15	30	24 (20)	p ≤ 0.001	p ≤ 0.001
	variant 3–variant 2	15	30	20 (20)	p ≤ 0.001	p ≤ 0.001
trial wine B	variant 3–variant 0	15	30	14 (4)	n.s.	p ≤ 0.01
	variant 3–variant 1	15	30	25 (22)	p ≤ 0.001	p ≤ 0.001
	variant 3–variant 2	15	30	24 (22)	p ≤ 0.001	p ≤ 0.001
trial wine C	variant 3–variant 0	15	30	26 (0)	p ≤ 0.001	p ≤ 0.001
	variant 3–variant 1	15	30	11 (5)	n.s.	n.s.
	variant 3–variant 2	15	30	9 (4)	n.s.	n.s.

.

In the case of trial wines A and B, the wines filtered with Fibrafix^® TX layers were found to be highly convincing in comparison to variant 1 (trial wine with the addition of an impact compound) and variant 2 (trial wine with the addition of an impact compound and EKS filtration). The variants were not only highly significantly distinguished (p < 0.001), but variant 3 was highly significantly preferred. Compared with variant 0 (trial wine without addition), the results were not clear. For trial wine B (addition of TCA), no significant difference could be detected, but variant 0 was significantly preferred in terms of correct results. For trial wine A (addition of geosmin), the samples could be significantly distinguished from each other, and the trial wine without addition was also highly significantly preferred. For base wine C, variant 3 did not perform better than variants 1 and 2; the samples could not be distinguished. Variant 0, on the other hand, was highly significantly recognised and rated as better.

3.2. Experiment on a 250 L Scale (1250 L per m² Filtration Surface)

3.2.1. Aroma Analysis

In the large-scale experiment, samples were taken continuously for analysis (in each case after 25 L or 125 L of filtration capacity per m²). This test was only carried out on trial wine A.

Table 3. Reduction in geosmin and aroma loss of ester compounds and free monoterpenes in the course of a normal EKS filtration and filtration with Fibrafix^® TX layers in a large-scale test (up to 1250 L filtration capacity per m² filter area). Significance test by means of ANOVA with measurement repetition.

Filtration Capacity in Litres per m2 Filtration Surfaces		ANOVA *1	125	250	375	500	625	750	875	1000	1125	1250
Geosmin	EKS	F = 580.527 p < 0.001	9%	3%	3%	2%	5%	5%	5%	4%	4%	3%
	Filtrox		86%	86%	85%	83%	76%	76%	69%	70%	67%	61%
aromatic esters	EKS	F = 37.832 p = 0.004	10%	10%	10%	9%	8%	9%	8%	8%	8%	7%
	Filtrox		40%	40%	40%	38%	35%	35%	32%	33%	31%	28%
acetate esters of higher alcohols	EKS	F = 4.285 p = 0.107	14%	14%	14%	14%	12%	13%	12%	12%	11%	10%
	Filtrox		31%	31%	30%	29%	26%	27%	25%	25%	24%	21%
ethyl esters of carboxylic acids with an even number of C atoms	EKS	F = 903.656 p < 0.001	6%	6%	5%	5%	4%	4%	3%	4%	3%	2%
	Filtrox		63%	63%	62%	60%	55%	55%	50%	50%	48%	42%
esters of higher alcohols and medium-length chain carboxylic acids	EKS	F = 0.172 p = 0.700	5%	5%	5%	5%	4%	4%	4%	4%	4%	3%
	Filtrox		6%	6%	6%	6%	6%	6%	5%	5%	5%	4%
ethyl esters of carboxylic acids with an odd number of C atoms	EKS	F = 0.679 p = 0.456	19%	19%	18%	18%	16%	16%	15%	15%	14%	13%
	Filtrox		24%	24%	24%	23%	21%	21%	20%	20%	19%	17%
isoamyl ester of medium-long chain carboxylic acids	EKS	F = 183.627 p < 0.001	5%	5%	5%	5%	5%	5%	4%	4%	4%	4%
	Filtrox		33%	33%	33%	32%	29%	29%	27%	27%	26%	23%
methyl esters	EKS	F = 124.088 p < 0.001	14%	14%	14%	13%	12%	12%	11%	11%	11%	10%
	Filtrox		59%	59%	58%	56%	52%	52%	47%	48%	46%	41%
free monoterpens	EKS	F = 111.782 p < 0.001	14%	14%	14%	14%	13%	13%	12%	12%	11%	10%
	Filtrox		50%	50%	49%	48%	44%	44%	40%	40%	39%	35%

*¹ ANOVA with repeated measurement.

and the Supplementary Material Table S6 show the results of the relative aroma loss compared with the variant 0. It was considered, statistically, whether there was a constant significant difference over the ten measurement repetitions between the common filtration with EKS layers and the filtration with Fibrafix^® TX layers. It was found that there was a significant difference in geosmin in four out of seven ester groups and the free monoterpenes. In each of these cases, the aroma loss was higher in the variants with Fibrafix^® TX than in the variants with EKS layers. If we look at the individual compounds, this result becomes even clearer: in 23 of 36 detected aromas, the aroma loss was higher in the wines filtered with Fibrafix^® TX layers than in the wines filtered with EKS layers. It is noticeable that the aroma loss of all compounds decreased with increasing filtration volume.

Not only the aroma loss in the course of filtration was examined, but also the final cuvèes were compared with each other. It was found that filtration with Fibrafix^® TX layers reduced the geosmin content by almost 75.9 ± 3% at a total filtration capacity of 1250 L per m² filter area. This compares to a small reduction of 4.4 ± 3% with EKS filtration. The aroma loss of ester compounds and free monoterpenes was higher (significant for four out of seven ester groups and the monoterpenes) and, in some cases, considerable in the filtration with Fibrafix^® TX layers compared with the EKS filtration and amounted to 55% for ethyl ester and 52% for methyl ester (Figure 6).

Figure 7 shows the aroma loss of the individual compounds, and the Supplementary Material (Table S7) shows the significance table. The most significant aroma loss (>80%) from filtration with the special layers was detected for compounds ethyl decanoate, methyl decanoate, and hexyl acetate. Overall, the aroma loss was higher for 21 compounds than for EKS filtration, and for 13 compounds was ≥50%.

3.2.2. Sensorial Examination

In the trial, the four variants were compared with each other by means of a triangle test (

Table 4. Evaluation of the triangle tests in the big-scale experiment according to the significance table at the significance level p > 0.05 (not significant, n.s.), p ≤ 0.05 (significant), p ≤ 0.01 (highly significant), and p ≤ 0.001 (very highly significant) of [37].

Comparison	Number of Panelists	Number of Answers	Number of Correct Answers	Significant Difference	Significant Preference
variant 0–variant 1	9	27	24	p ≤ 0.001	variant 0 p ≤ 0.001
variant 0–variant 2	9	27	24	p ≤ 0.001	variant 0 p ≤ 0.001
variant 0–variant 3	9	27	14	n.s.	variant 0 p ≤ 0.001
variant 1–variant 2	9	27	7	n.s.	n.s.
variant 1–variant 3	9	27	26	p ≤ 0.001	variant 3 p ≤ 0.001
variant 2–variant 3	9	27	20	p ≤ 0.001	variant 3 p ≤ 0.001

). The situation was similar to the small-scale trials. Variant 3 showed a better sensory profile than variants 1 and 2 because of the reduction in the off-flavour geosmin, but because of the loss of aroma, it was rated slightly worse than variant 0. No significant difference (the limit for significance with 15 correct answers) could be detected between variant 0 and 3, but in all 14 correct triangle tests, variant 0 was rated as the better one.

Furthermore, the quality wine status of the samples was checked sensoryaly in the same way as the official quality wine inspection [1,2]. The results can be found in

Table 5. The sensory testing of the quality wine status of the four variants of the large-scale trial according to [1,2] *¹.

Variant	Quality Wine Number of Decision: Yes	Quality Wine Number of Decision: No	Reasons for Rejection (Number of Decisions)
variant 0	9	0	-
variant 1	0	9	faulty (9)
variant 2	0	9	faulty (9)
variant 3	5	4	faulty (1), not corresponding to the overall quality required for quality wine. (3)

*¹ Difference from [1,2] due to the higher number of tasters. Six tasters are required by law.

. Variant 0 was accepted as quality wine (tasting result 9:0), while variants 1 and 2 were declared faulty (tasting result 0:9). Variant 3 was barely accepted as quality wine (tasting result 5:4). It is interesting to note that the reason for the refusal “faulty” was chosen only once, while three tasters rated the wine as not corresponding to the overall quality required for quality wine.

3.3. Trial with a Prototype of the Company Filtrox Group (Zwingen, Switzerland) for the Reduction in Strong “Animal” Phenolic Aromas

Filter layers with integrated zeolites were tested for the reduction in the lead substances of the “animal” phenolic aroma. The first preliminary tests published here show that the lead substances 4-ethylphenol and 4-ethylguaiacol were significantly reduced, but to an insufficient extent. While the conventional Filtrox layers led to a reduction of 10% (4-ethylphenol), 5% (ethylguaiacol), and 4% (4-ethylcatechol) (Figure 2), the reduction by these new layers tested (Filtrox prototype) was 15% (4-ethylphenol), 16% (ethylguaiacol), and 5% (4-ethylcatechol). The reduction capacity was still not sufficient to decrease the impact substances below the sensory thresholds of 50–968 µg 4-EP/L [27,28,29,30,31,32], 33–135 µg 4-EC/L [27,28,29,30,31,32,33], and 0.9 to 33 µg 4-EG/L [32,33,34] (Figure 8).

4. Discussion

In the course of the present study, the reduction in cork taint, mould, and “animal” phenolic aromas, as well as the loss of other volatile compounds such as esters, free monoterpenes, and wood aromas in Austrian white and red wines, were tested using the patented filter sheet Fibrafix^® TX in comparison with conventional sterile filter layers. Furthermore, a novel prototype of filter layers with embedded zeolites for the reduction in “animal” phenolic aromas was tested. The patented, commercially available filter layers were able to reduce the cork taint 2,4,6-trichloroanisole as well as geosmin, the lead substance of the mould aroma, to a sufficient extent, whereas the lead compounds of “animal” phenolic aromas were not reduced to a sufficient extent. The results thus not only confirm the statements of the work of [11,12] that TCA can be reduced but also the statements of [8,15] with regard to the positive, reducing power towards geosmin. The unconfirmed assumptions, findings from practical application, that the Fibrafix^® TX layers filter layers can also be used to reduce strong “animal” phenolic taste, therefore, cannot be confirmed. The new prototype for the specific reduction in these flavours also could not substantially reach the goal of pushing the existing aromas below the perception threshold, even though these layers achieved a higher reduction (in a different trial wine) compared with the Fibrafix^® TX layers.

By filtration with the Fibrafix^® TX, the loss of ester and monoterpenes was partly considerable and very heterogeneous (from 4–90% reduction in single compounds at the beginning of filtration), but wood aroma compounds were less affected. This is in contradiction to the overall conclusion of [11], which summarises that filtration with Fibrafix^® TX layers does not result in any significant loss of volatile compounds. However, in this study [11], only main aroma compounds, mainly higher alcohols, and carboxylic acids, were considered, but also some esters, including isoamyl acetate. With this compound, an aroma loss of about 20% was found, whereas, in the present study, the loss was 24% for this compound in the large-scale test. This means that the present study suggests another overall conclusion than [11], but looking in detail, there are some similarities between the two studies.

However, there was a specific reduction in individual compounds, while other compounds were less reduced. According to the theory of zeolites acting as molecular sieves, the molecular size should be responsible for the exclusion [10]. The actual principle of elimination cannot be fully clarified. A significant correlation with the molecular weight of the compounds could not be established, although the molecular weight should tend to play a role after all. Compounds with molecular weights between 182 (geosmin) and 211 (TCA) were reduced by an average of 63.3% in the large-scale test, while compounds with molecular weights higher than 211 were reduced by only 42.6%, and compounds smaller than 182 by only 32.7%, although there were outliers declared in all three groups.

The aroma loss was most significantly higher in comparison with conventional filter sheets. The fact that conventional filtration with sterile filter layers also leads to aroma loss is known from practical experience but has not been specifically studied to date. The fact that compounds such as ethyl dodecanote were reduced by up to 29% (comparison with Fibrafix^® TX: 56%) in the large-scale test is a bit surprising. In the majority of compounds, however, the loss of volatile compounds through sterile filtration was moderate (far below 10%). A significant reduction in TCA by up to 25% by conventional depth filtration with sterilisation filter layers shows a certain potential. The functional group of the compound could also be a crucial factor for the reduction, which would also explain why the impact compounds of the “animal” phenolic aromas were not reduced to the same extent as other compounds with similar molecular dimensions (smaller 182).

With the exception of the trial wine with increased concentrations of “animal” phenolic aromas, the products showed a better sensory profile after filtration with Fibrafix^® TX than before filtration. In comparison with the variants without the addition of off-flavours and without filtration, the loss of aroma could be detected significantly by some of the tasters and rated as poor quality; other tasters could not distinguish the wines. In a quality wine test, the one trial wine in the large-scale trial was barely accepted as a quality wine after filtration with Fibrafix^® TX. This could be explained by the fact that a large loss of volatile compounds does not automatically have to have a large sensory impact. The number of flavour-active compounds in wines is high, and the nonlinear perceptual interactions between these compounds suggest that sensory spaces are more dependent on relative concentrations than on the total amount of flavour compounds [16].

The costs are a major factor in the filtration with Fibrafix^® TX. With a price of 50 euros per layer (40 × 40), this is more than ten times higher than that of a normal sterile filter layer [39,40]. Depending on the filtration performance, 0.20 to 0.50 euros per litre of wine can be expected, which is stated as extremely high. If the filtered and thus defect-free wine can be sold at a high price, such filtration is profitable, but not for bulk wine prices.

5. Conclusions

The filter layers with integrated zeolites (patented layer of the company Filtrox Fibrafix^® TX) tested in the present study were effective with regard to the reduction in 2,4,6-trichloroanisole (impact substance of cork taint) and geosmin (impact substance of mould aroma), but not with regard to the reduction in 4-ethylphenol, 4-ethylguaicol, and 4-ethylcatechol (impact substances of “animal” phenolic aromas). However, the reduction in these off-flavours is generally not specific enough to prevent a loss of quality-determining volatile compounds, especially ester compounds and free monoterpenes. The products obtained after filtration are sensory faultless, but for some tasters, the overall quality of these wines does not correspond to a quality wine, so the question of economic viability arises, which must be assessed separately for each individual case due to the extremely high costs for the special filter layers.