Introduction

Noble rot, produced by the infection of grape species such as Vitis vinifera, by the Ascomycete fungus Botrytis cinerea, is a widespread viticultural technique used to produce high-quality dessert wine styles. Microclimatic conditions determine whether B. cinerea infection of grape berries will result in noble rot, or detrimental grey rot (Fournier et al., 2013). When vineyard conditions are appropriate, with moist nights, morning fog, and dry, sunny days, grape berries undergo shrivelling, resulting in substantial changes in the concentrations of aroma compounds and their precursors, with some increasing significantly, and others decreasing (Negri et al., 2017). Descriptors attributed to the resulting wines often include honey, dried fruits, stone fruits (peach and apricot), and caramel (Magyar and Soós, 2016; Vannini and Chilosi, 2013).

Botrytised wines can be made with a range of suitable grape varieties. A large proportion of research on botrytised wines has focused on the renowned Sauternes wines, made primarily with Sauvignon blanc and Sémillon (Magyar and Soós, 2016), and Hungarian Tokaji Aszú, which can be made using six different grape varieties, including Furmint (Vyviurska and Špánik, 2020). Riesling is also commonly used to produce high-quality botrytised wines, with famed examples hailing from Germany (Magyar and Soós, 2016). Previous works have highlighted the extensive sensory and chemical differences between botrytised and non-botrytised wines. Compounds produced by the fungus or the grapes themselves, due to B. cinerea induced changes in berry metabolism, may be responsible for these changes. Studies have looked at the composition of both the grape must and the resulting wine, either through artificial induction of B. cinerea infection (Negri et al., 2017; Wang et al., 2017); by harvesting grapes at two-time points, standard and late harvest times, with B. cinerea infection (Genovese et al., 2007); or by comparing botrytised and non-botrytised wines made using visually-selected grapes from the same vineyard (Tosi et al., 2012).

By comparing Garganega wines made from grapes withered with and without the influence of B. cinerea, Negri et al. (2017) found that a range of compounds were present in higher concentrations in wines made from botrytised grapes, including benzaldehyde, 1-octen-3-ol, lactones (sherry lactone and γ-Nonalactone), and esters (such as ethyl phenylacetate), to name a few. These differences were attributed to the influence of B. cinerea and were primarily associated with “spicy”, “floral” and “mushroom” descriptors in these wines, compared to the “fresh fruit” attributes associated with the non-botrytised wine. Similar findings were reported by comparing wines produced from non-botrytised and artificially botrytised Chardonnay grapes. Compared to the non-botrytised wine, the botrytised wine exhibited higher concentrations of a range of aroma compounds including esters, lactones (including γ-Nonalactone), and thiols. A “dry apricot” aroma descriptor was attributed to the botrytised wine and found to be associated with several aroma compounds, including γ-Nonalactone (Wang et al., 2017).

Highlighted above, one potentially key aroma compound in dessert wine styles shown to be significantly impacted by noble rot infection is γ-Nonalactone. This compound, containing a five-membered lactone ring, is commonly found in a wide range of wines and is associated with coconut and stone fruit descriptors, with a relatively low odour detection threshold in wine (ODT, 30 µg L^-1) (Miller et al., 2022; Nakamura et al., 1988). A large survey of Australian red and white wines showed that significantly higher concentrations of γ-Nonalactone were present in noble rot wines, with a maximum of 59 µg L^-1 (Cooke et al., 2009).

One possible precursor to γ-Nonalactone in botrytised wines is linoleic acid. Analysis of the extracellular matrix of B. cinerea germlings found that linoleic acid constituted approximately 3.6 % of the acyl glycerides present (Cooper et al., 2000). Incubation experiments have shown that γ-Nonalactone is produced from linoleic acid during fermentation by Sporobolomyces odorus and Saccharomyces cerevisiae yeasts, via lipoxygenation, yielding its 9- and 13-hydroperoxide derivatives, which are subsequently further broken down to produce γ-Nonalactone (Garbe et al., 2001). It, therefore, follows that during fermentation, S. cerevisiae could be producing high levels of γ-Nonalactone in botrytised wines from the increased linoleic acid in botrytised grapes compared to non-botrytised.

The process of fatty acid biosynthesis in plants is similar to that in fungi, with the extension of fatty acid chains through the use of acetyl-CoA (Zhang et al., 2022), and fatty acids are also present in grapes, being essential to plant metabolism. Of the fatty acids found in grapes, linoleic acid tends to be the most abundant, with the highest concentrations detected in the grape skins (≥ 10 % total weight), as shown by a recent study of New Zealand (NZ) Sauvignon blanc and Pinot noir grapes (Sherman et al., 2023). Concentrations of linoleic acid in Pinot noir grapes were higher than in Sauvignon blanc grapes (Sherman et al., 2023), which may explain the higher γ-Nonalactone levels generally found in dry red wines compared to dry white wines (Cooke et al., 2009). Fatty acids in general can have a significant impact on wine aroma when used to supplement grape must (Pinu et al., 2019), however, many of the processes involved in this relationship are unclear. This may be due to the misconception that the hydrophobic nature of fatty acids limits their extraction during winemaking, but this has been proven false (Sherman et al., 2023). Further investigation is required to understand the complex relationship between fatty acids and grape aroma, and what vitivinicultural factors may impact this relationship.

Another compound speculated to be a γ-Nonalactone precursor is 4-oxononanoic acid, supplementation with which was shown by deuterium-labelling to result in the production of γ-Nonalactone in Merlot and Cabernet-Sauvignon (de Ferron et al., 2020). (2E,4E)-2,4-Nonadien-1-ol was also suggested as a downstream product of the 9-hydroperoxide derivative of linoleic acid (Garbe et al., 2001). Neither of these compounds nor the lipoxygenation products of linoleic acid have yet been linked to the influence of B. cinerea.

The impact of grape stress events on wine aroma and quality is well-documented. When grapes are stressed due to a variety of factors, including infection, water stress, extreme heat, and extreme cold, metabolic processes within the grapes are altered (Blanco-Ulate et al., 2015; Ma et al., 2021; Pons et al., 2017; Venios et al., 2020). A study of alterations in the metabolism of Sémillon grapes infected with B. cinerea via transcriptomic and metabolomic methods demonstrated an increase in fatty acid biosynthesis, suggested to lead to alterations in the resulting wine aroma (Blanco-Ulate et al., 2015). However, neither lactones nor linoleic acid were investigated specifically in this work, thus, further research is required in this area.

NZ is known worldwide for its famed Sauvignon blanc wines, with their characteristic tropical passionfruit aroma (Parr et al., 2007). In addition to dry table styles of Sauvignon blanc, NZ also produces high-quality noble rot white wines from this variety, often as a single variety but sometimes blended with Sémillon, as used in Sauternes (New Zealand Winegrowers Inc., 2023a). Although a significant amount of work has investigated the aroma profiles of dry styles of NZ Sauvignon blanc wines, little research has been conducted on sweeter wine styles.

Previous work by this group involved the design and synthesis of an isotopically labelled γ-Nonalactone analogue (²H₂¹³C₂-γ-Nonalactone, Figure 1), and its use for the quantification of γ-Nonalactone using a stable isotope dilution assay (SIDA) in a range of 12 NZ Pinot noir samples, via solid phase extraction-gas chromatography-mass spectrometry [SPE-GC-MS] (Miller et al., 2023). This robust SIDA γ-Nonalactone quantification method allows for very different wine styles, with a range of matrices (varying sugar and ethanol levels) to be analysed using a single method.

This current work aims to quantify γ-Nonalactone in a range of NZ botrytised and non-botrytised white wines, and to determine whether γ-Nonalactone concentrations are higher in noble rot wines compared to their non-botrytised counterparts. It is hypothesised that γ-Nonalactone concentration will not be significantly impacted by grape variety, region, nor vintage and that Botrytis influence will be the most important factor in determining γ-Nonalactone concentration.

Materials and methods

1. Chemicals and reference compounds

²H₂¹³C₂-γ-Nonalactone was synthesised previously according to the literature procedure (Miller et al., 2023). γ-Nonalactone (> 97 %) was purchased from Sigma-Aldrich (MO, USA). Sodium sulfate and all organic solvents were purchased from ECP Ltd (Auckland, NZ). Sartorius (Göttingen, Germany) Arium Pro ultrapure water system was used for Type 1 water production.

2. Analytical procedure

γ-Nonalactone was extracted and quantified in wine samples following the stable isotope dilution assay-solid phase extraction-gas chromatography-mass spectrometry (SIDA-SPE-GC-MS) method used previously, with ²H₂¹³C₂-γ-Nonalactone as an isotopically labelled internal standard (Miller et al., 2023). This method was based on an additional method used for the quantification of four saturated linear aliphatic lactones in Australian wines (Cooke et al., 2009). Wine samples (50 mL) were spiked with internal standard (IS, 100 µL of 10,000 µg L^-1 ethanolic solution), in triplicate. Samples were then passed through Bond Elut ENV SPE cartridges (200 mg, 3 mL), previously conditioned with methanol (2 mL) and water (4 mL). Cartridges were then washed with water (5 mL), then an aqueous solution (20 mL) of methanol (40 % v/v) and sodium bicarbonate (1 % w/v). Cartridges were allowed to dry by passing air through them for 30 min, then extracts were eluted with CH₂Cl₂ (2.5 mL). This eluate was collected and then concentrated under a gentle stream of nitrogen to a final volume of approximately 100 µL.

For analysis of samples, an Agilent 6890N Network GC coupled to an Agilent 5973 GCMS Single Quad inert mass selective detector was used. Samples (1 µL) were injected directly onto the column (Agilent HP-INNOWax, 60 m length, 0.25 internal diameter, 0.25 µm film thickness) using an Agilent 77683B Series Injector, in splitless mode. The carrier gas was He, at a flow rate of 1 mL min^-1. The temperature programme of the GC oven was as follows: from a starting temperature of 50 °C, the temperature was increased to 60 °C at a rate of 1 °C min^-1, then to 250 °C at a rate of 10 °C min^-1. Finally, the temperature was held at 25 °C for 25 minutes, giving a total run time of 54 minutes.

Ion monitoring was conducted in SCAN mode. For quantification of γ-Nonalactone 1, the base peaks of γ-Nonalactone and ²H₂¹³C₂-γ-Nonalactone (m/z 85 and 89, respectively) were used. The relative responses of γ-Nonalactone and ²H₂¹³C₂-γ-Nonalactone were compared to the calibration curve, with concentrations of γ-Nonalactone up to 100 µg L^-1. Qualifier ions used for the identification of compounds were m/z 99 and 114 for γ-Nonalactone and m/z 103 and 118 for ²H₂¹³C₂-γ-Nonalactone.

Alcohol levels (% v/v) in commercial wine samples were analysed using an Alcolyzer Wine M from Anton Paar (Graz, Austria). pH and titratable acidity (TA) of commercial wine samples were determined using a Hanna Instruments (RI, USA) Wine Titrator, equipped with a Hanna Instruments pH meter. Residual sugar (RS), free (FSO₂), and total (TSO₂) sulfite levels were determined in commercial wine samples using the Megazyme (Bray, Ireland) D-Fructose/D-Glucose Assay Kit and Total and Megazyme Free Sulfite Assay Kit, respectively.

3. Wine Samples

Thirty-eight white wines were purchased from retailers within Auckland, NZ. Wine samples represented a wide range of NZ wine-producing regions (Central Otago, Gisborne, Hawke’s Bay, Marlborough, Waiheke Island, Waipara Valley, Wairarapa), vintages (2014, 2015, 2016, 2017, 2018, 2019, 2020, and 2021) and grape varieties (Riesling, Sauvignon blanc, Sémillon, and Sauvignon blanc-Sémillon blend) [Table 1]. Of the wine samples analysed, 18 were botrytised and 20 were non-botrytised, according to the information provided by the producers. All vitivinicultural data provided by winemakers, including vintage, grape variety, region, presence of Botrytis, lees ageing, oak use, and harvest method are shown in Table 1. All technical parameters measured in this work including RS, pH, TA, alcohol percentage, FSO₂, and TSO₂ are shown in Table 1.

4. Quantification of γ-Nonalactone in NZ white wine samples

As previously performed (Miller et al., 2023), calibration was conducted in model wine (12 % v/v ethanol in Type 1water, 5 g L^-1 tartaric acid, adjusted to pH 3.2 with NaOH). Model wine samples (50 mL) were spiked with ²H₂¹³C₂-γ-Nonalactone (100 µL, 10,000 µg L^-1) and a range of concentrations of γ-Nonalactone (eight concentrations, 0–100 µg L^-1), in triplicate. Relative peak areas of analyte and internal standard were plotted against γ-Nonalactone concentration to produce a calibration curve (R² = 0.9985) with a limit of detection (LOD) of 0.4 µg L^-1.

In triplicate for each wine, white wine samples (50 mL) were spiked with ²H₂¹³C₂-γ-Nonalactone (100 µL, 10,000 µg L^-1), prior to extraction and analysis by SIDA-SPE-GC-MS. γ-Nonalactone concentrations in these samples were determined by comparing the relative peak area of γ-Nonalactone and the internal standard with the aforementioned calibration curve (Miller et al., 2023).

5. Statistical analysis of γ-Nonalactone concentrations alongside technical data and winemaking parameters

R: A language and environment for statistical computing [version 4.3.1] (R Core Team, 2022), and accompanying RStudio [version 2023.09.1] (RStudio Team, 2020) were used for data processing and analysis. For data visualisation, the R packages ggplot2 (version3.5.0) and factoextra (version 1.0.7) were used. To carry out statistical analyses, the base stats package (R Core Team, 2022) randomForest (version 4.7-1.1) and mdatools (version 0.14.1) were used (Kassambara and Mundt, 2020; Kucheryavskiy, 2020; Wickham, 2016). Microsoft Excel (version 2402) was used for additional data processing.

The heatmap dendrogram was produced using an agglomeration method for hierarchical cluster analysis, whereby each object was assigned to its own cluster and then proceeded iteratively, joining the two most similar clusters and continuing until there was just one cluster. Distances between clusters were calculated using the Lance-Williams dissimilarity update formula.

The final PLS-R model (number of components = 1) was chosen to minimise the cross-validation RMSE. A series of random forest models were fit to the data, using an out-of-bag cross-validation approach. In the model fitting process, 500 trees were fit for each forest model and three variables were randomly sampled at each split.

Results and discussion

A previously verified analytical method was selected for the quantification of γ-Nonalactone in NZ white wines. Considering this diversity in sample matrices, stable isotope dilution analysis was used for the quantification of γ-Nonalactone. The stability of the internal standard, ²H₂¹³C₂-γ-Nonalactone, in wine, and the repeatability and reproducibility of this analytical method were verified, as described previously (Miller et al., 2023). Average γ-Nonalactone concentrations and associated standard deviation values in commercial wine samples are shown in Table 2.

Of the 36 wines analysed, 29 had γ-Nonalactone concentrations above the LOQ, and only one sample had no detectable γ-Nonalactone. All samples with γ-Nonalactone concentrations below the LOQ or LOD were non-botrytised. Therefore, the most notable factor related to the concentrations of γ-Nonalactone was associated with Botrytis influence; botrytised wines showed significantly higher concentrations of γ-Nonalactone, compared to non-botrytised wines (two-sample t-test, p-value = 3 × 10^-8) [Figure 2]. This result was in agreement with previous research which included seven botrytised wines in a large survey of 178 Australian wines (Cooke et al., 2009). The concentration of γ-Nonalactone in botrytised wines ranged from 4.1 µg L^-1 to 43.5 µg L^-1, four of which were above the ODT of this compound (30 µg L^-1). Meanwhile, the highest concentration of γ-Nonalactone in non-botrytised wines was 8.7 µg L^-1, which was well below the ODT of 30 µg L^-1 [Figure 2] (Nakamura et al., 1988). Again, these results are similar to those observed previously; Cooke et al. (2009), reported γ-Nonalactone concentrations ranging from 8.9–59.3 µg L^-1 in botrytised samples, while in the dry white wines analysed, the highest concentration of γ-Nonalactone was 4.7 µg L^-1 in a sample of Viognier.

The other winemaking-associated variables investigated in this study were region, vintage, grape variety, oak usage, and harvest method (hand or machine). Notably, all samples possessed the same bottle closure (screwcap), so bottle closure was not considered as a variable in this work. Oak usage and harvest method did not appear to be associated with γ-Nonalactone concentration (two-sample t-test, p-values > 0.05). The grape variety was generally not associated with γ-Nonalactone concentration, except for the non-botrytised Sauvignon blanc-Sémillon samples NSS1 and NSS2, which had concentrations of 8.67 and 7.39 µg L^-1, respectively. The γ-Nonalactone concentration in non-botrytised Sauvignon blanc-Sémillon blends was higher than the concentration in non-botrytised samples of non-Sémillon containing wines (p-value < 0.05, Figure 5); however, concentrations were still below the ODT of γ-Nonalactone (30 µg L^-1). This result suggests that there is either a varietal factor contributing to γ-Nonalactone concentration, with Sémillon grapes perhaps containing higher levels of precursors to γ-Nonalactone, or that the grapes used to make these wines had some noble rot or grey rot, which would therefore contribute to the higher concentration of γ-Nonalactone. The latter is more likely, given that Sémillon grapes tend to be reasonably susceptible to B. cinerea infection due to their thin cuticles, and it is not common for botrytis influence to be mentioned for dry white wines (Köycü et al., 2018; Paňitrur-De La Fuente et al., 2018). However, the number of samples of Sémillon analysed in this study was very limited, reflecting the low abundance of this grape variety in NZ winemaking (< 0.1 % of grapes in 2022, by mass) (New Zealand Winegrowers Inc., 2023b). Notably, γ-Nonalactone concentration was relatively low in the non-botrytised Sauvignon blanc samples analysed, suggesting that this aroma compound is unlikely to contribute to the acclaimed aroma of NZ Sauvignon blanc (Parr et al., 2007).

Residual sugar in the botrytised wines ranged from off-dry (7.2 g L^-1) in an unusual style of botrytised Riesling (BR4) to sweet (262.5 g L^-1), and alcohol percentage ranged from 7.8 to 14.0 % (v/v). The non-botrytised wines were dry or off-dry (0.6 to 12.7 g L^-1 RS) with alcohol levels ranging from 8.7 to 14.0 % (v/v). pH was, on average, higher in the botrytised wines compared to the non-botrytised wines (two-sample t-test, p-value = 3.6 × 10^-5). There was some evidence that TA was higher in botrytised samples (two-sample t-test, p-value = 0.068). These relationships were demonstrated via principal component analysis (PCA), hierarchical clustering analysis (HCA), partial-least-squares regression (PLS-R), and random forest methods. For a more comprehensive analysis of the data collected, to determine relationships between different factors, PCA was carried out (Figure 4). Principal components 1 and 2 accounted for 53.6 % and 13.9 % of variability in the data, respectively, indicating a notable portion of the variability in the wine profile could be captured when analysing these principal components. PCA showed that, as expected, RS and TA were positively correlated with γ-Nonalactone concentration. This association can be explained by the strong links between these parameters and Botrytis influence. Botrytised wines commonly have higher RS levels (fructose and glucose) due to the concentration of sugars during grape berry infection and resulting shrivelling (Magyar and Soós, 2016). TA tends to be higher in dessert-style wines, to balance the high sugar levels. Unexpectedly, in the PCA the pH and TA appeared to be weakly correlated, according to Figure 4. This relationship was further investigated with correlation tests (Pearson and Spearman tests), which showed that contrary to what is suggested by the PCA, pH, and TA were not significantly correlated (p-values = 0.2 and 0.3, respectively). It is also worth mentioning that PCA can indicate weak correlations between variables that are not necessarily present when all the data is considered, with principal components only accounting for a set amount of variation in the data. Lower alcohol levels were generally present in wines with higher γ-Nonalactone concentration, which is consistent with the expectation that higher-sugar botrytised wines do not tend to be fermented to dryness, and fermentation is often stopped early via the addition of sulfur dioxide (Magyar, 2011). However, the statistical significance of the relationship between alcohol % and γ-Nonalactone concentration was none-to-weak (Spearman correlation test, p-value = 0.09). γ-Nonalactone concentration was found to be significantly positively correlated to FSO₂ and TSO₂ in the wines analysed (Spearman correlation test, p-values = 5 × 10 ^-5 and 3 × 10 ^-7, respectively). This is not unexpected, as sweet wines are permitted to have higher levels of SO₂ in NZ (Giacosa et al., 2019).

Interestingly, year of vintage was shown to be negatively correlated with γ-Nonalactone concentration, with older samples having higher γ-Nonalactone concentrations. This result may be due to the limitations of commercially available wines at the time this study was carried out. Non-botrytised samples were mainly available from more recent vintages (mainly 2020–2021), compared to the older botrytised samples from 2014–2019. Therefore, it is hypothesised that this relationship is related to confounding of vintage and Botrytis influence, rather than reflecting an actual vintage-related trend. The possibility of wine ageing having an impact on the concentration of γ-Nonalactone has previously been evaluated in the scientific literature. Analyses of Malvasia and Moscato white wines before and after bottle storage for 18 months showed that the concentration of γ-Nonalactone decreased significantly in both wines (Del Caro et al., 2014). Further work to investigate the negative correlation between γ-Nonalactone concentration and vintage could include a longitudinal study, looking at dry and botrytised Sauvignon blanc wines over successive years and determining if there is a significant trend between vintages.

Two-way analysis of variance (ANOVA) was carried out to explore the relationships between γ-Nonalactone concentration and the variables Botrytis, vintage, oak, and harvest method. Two-way effects between these variables were also assessed. None of the two-way effects on γ-Nonalactone concentration were statistically significant, (p-values > 0.05), and nor were the main effects of oak or harvest method (p-value > 0.05). The only significant effects were observed for vintage and Botrytis (p-values = 3 × 10^-6 and 0.002, respectively), which was in line with observations from the PCA in Figure 4.

To explore how the winemaking variables and γ-Nonalactone concentration were related to other wine parameters, the PCA scores plots showing the samples were coloured according to Botrytis influence or lack thereof, grape variety used, oak usage, and harvest method (Figure 5A-D). Ellipses to represent 95 % confidence levels associated with these groups are shown. Figure 5A shows the non-botrytised samples are much more tightly clustered compared to the botrytised samples. While there is an overlap of the 95 % ellipses for the groups, the botrytised wines were almost exclusively to the left of the PCA scores plot and the non-botrytised to the right. Based on the loadings for the PCA, it can be said that the botrytised wines were from later vintages, as discussed above, and had higher γ-Nonalactone concentration, pH, TA, and RS.

The other variables visualised, grape variety (Figure 5B), oak usage (Figure 5C) and harvest method (Figure 5D) do not show distinct separate clusters. In literature, grape variety has been demonstrated to have a significant impact on lactone concentrations, particularly in red wines, evidenced in studies directly comparing volatile profiles of different red grape wines (de Ferron et al., 2020; Fan et al., 2010; Martínez-Pinilla et al., 2013). Varietal differences are likely due to differences in concentrations of precursors in the grapes (de Ferron et al., 2020). However, as dry white wines generally tend to have low concentrations of γ-Nonalactone, this relationship appears to be less prevalent in white wines, as evidenced in work by Cooke et al. in the analysis of Australian white wines (Cooke et al., 2009). Alternatively, or perhaps in addition to this factor, different treatments of grapes during winemaking, such as extended maceration commonly used in red wine production, may enable increased extraction of γ-Nonalactone precursors, resulting in higher concentrations of this compound in the finished wine.

The use of oak during winemaking is a widespread practice within the wine industry and is known to impart aroma compounds to the wine, including oak lactone, which is similar in chemical structure to γ-Nonalactone (Piggott et al., 1995). Perhaps due to this structural similarity, γ-Nonalactone has been suggested to be extracted from oak during wine ageing (Jarauta et al., 2005). Based on this theory, it might be expected that the wines in this study which have had oak exposure during winemaking, either through fermentation in barrel and/or maturation, would have higher concentrations of γ-Nonalactone. During winemaking, there are also many different options for oak usage, such as using new oak or old oak, duration of oak exposure, the portion of wine exposed to oak, the oak source (American and French), oak species, or the level of toast the oak has (Garde-Cerdán and Ancín-Azpilicueta, 2006). Unfortunately, the information on oak use in the production of the wines analysed in this study was very limited, and the data was consequently reduced to a simple “yes” or “no” concerning the use of oak. The current results suggest that oak is unlikely to influence γ-Nonalactone concentration in wine, but further study using controlled types and amounts of oak during winemaking would enable a better assessment of whether this is truly the case.

The harvest method was classified in this work as either hand harvesting or machine harvesting. Generally, botrytised grapes are hand-harvested and visually selected depending on the extent of B. cinerea infection, and the desired wine style (Magyar, 2011). On the other hand, non-botrytised grapes are commonly machine-harvested in NZ, particularly for varieties such as Sauvignon blanc. This technique is chosen because it is faster than hand harvesting, allows grapes to be processed more quickly, and retains positive fruity attributes that are characteristic of the variety (Parr et al., 2013). The machine-harvested samples in this study are more tightly clustered and situated more to the right of the scores plot (Figure 5D), compared to the hand-harvested samples. This is likely indicative of the typicity of non-botrytised machine-harvested New Zealand Sauvignon blanc.

An additional method, hierarchical cluster analysis (HCA), was employed to explore the measured wine properties and the white wines analysed in this research. HCA allowed the construction of a heatmap from the known numerical data and the grouping of samples based on the similarity of their observed properties (Figure 6). In the HCA, as shown in Figure 7, four distinct clusters were formed. Except for samples BR7, NSS1, and NSS2, it can be seen that all of the botrytised samples were clustered separately from non-botrytised clusters. Closer examination of these samples reveals that BR7 had the lowest concentration of γ-Nonalactone of all the botrytised samples (4.04 µg L^-1) which may be contributing to its clustering with non-botrytised samples with similarly low γ-Nonalactone concentrations. BR7 also has a relatively low RS concentration (35 g L^-1) compared to the mean (139 g L^-1) and median (132 g L^-1) RS concentrations in the botrytised samples. Generally, in agreement with previous analyses, the red, purple, and orange groupings (primarily botrytised samples, shown in Figure 6) exhibited higher concentrations of γ-Nonalactone, while the latter two groups exhibited higher concentrations of FSO₂, TSO₂, and RS, based on the heat map. The orange and purple groups appear to be separated based on their RS, γ-Nonalactone concentration, and TA to some extent. The results of these analyses further demonstrate the main grouping within the samples is due to the presence or absence of B. cinerea during winemaking.

PCA and HCA are unsupervised statistical techniques that do not take into account any response (i.e., γ-Nonalactone). As an additional statistical approach, the supervised modelling techniques, partial least-squares regression (PLS-R), and random forest analyses were also carried out. These models were created to understand variables that were most closely linked to γ-Nonalactone concentration, by having γ-Nonalactone as the response and the range of explanatory variables being alcohol %, FSO₂, pH, RS, TA, TSO₂, and vintage. The PLS-R model appeared to fit the data well, with excellent agreement between the actual and predicted values (Figure S2D). In the developed PLS-R model, RS and pH were the strongest numerical explanatory variables for the response variable (γ-Nonalactone concentration), whilst vintage and both free and total SO₂ were correlated with higher γ-Nonalactone concentration based on their variable importance (VIP) and selectivity ratios (Figure S2E and S2F). Random Forest models were also developed with γ-Nonalactone concentration as the response. Several random forest models were fit, with these generally corroborating the importance of the parameters highlighted in the PLS-R model, with pH and RS again being the strongest predictors (Figure S3).

Conclusion

This work provides the foundation for further research into γ-Nonalactone origins in wines and has corroborated the trend of higher concentrations of γ-Nonalactone being present in botrytised white wines compared to non-botrytised white wines, while the region of wine production, grape variety, oak usage, and harvest method do not appear to have a significant impact. This research also showed that NZ botrytised and non-botrytised white wines show similar trends in γ-Nonalactone concentration, compared to those produced in other countries, such as Australia (Cooke et al., 2009).

The mechanisms behind the possible relationship between γ-Nonalactone concentration and noble rot have yet to be fully investigated. The higher concentrations of γ-Nonalactone in dry red wines compared to dry white wines found in other works suggest the presence of some precursors in grapes which are more highly concentrated in red grapes (Cooke et al., 2009; Miller et al., 2023) or are extracted to a greater extent during the production of red wines. The presence of B. cinerea may alter the concentrations of precursors present in white grapes, either through the production of precursors, such as linoleic acid, by the fungus itself, and/or effected alterations in grape metabolism, leading to increased concentrations of precursors (also possibly including linoleic acid) or γ-Nonalactone itself. Additionally, physical changes induced by B. cinerea infection of grapes may induce greater extraction of γ-Nonalactone precursors whilst grapes are still on the vine.

It is worth considering whether there is any relationship between γ-Nonalactone and grape stress in future studies. In this case, the identification of γ-Nonalactone precursors in grapes could be useful markers of grape stress for winemakers. Particularly with the prospect of climate change impacting wine aroma by creating more stressful events for grapevines, it will be important to have a thorough understanding of the mechanisms by which this may occur.

Acknowledgments

This research was undertaken with the assistance of the University of Auckland Doctoral Scholarship, and the Frederick Douglas Brown Postgraduate Science Research Scholarship (G.C.M). The Faculty Research Development Fund (FRDF) was also received for this work (R.C.D.). Thanks also to Jennifer Muhl for assistance with collecting technical data for the commercial white wines analysed in this work.

Author contribution

G.C.M.: Conceptualisation, methodology, formal analysis, investigation, writing-original draft, writing-review & editing, visualisation. D.B.: resources, writing-review & editing, supervision. L.I.P.: Methodology, resources, writing-review & editing, visualisation, supervision. R.C.D.: Conceptualisation, resources, writing-review & editing, project administration, writing-review & editing, supervision, funding acquisition.