Quantification of Polyfunctional Thiols in Wine by HS-SPME-GC-MS Following Extractive Alkylation

Analyses of key odorous polyfunctional volatile thiols in wines (3-mercaptohexanol (3-MH), 3-mercaptohexylacetate (3-MHA), and 4-mercapto-4-methyl-2-pentanone (4-MMP)) are challenging due to their high reactivity and ultra-trace concentrations, especially when using conventional gas-chromatography electron impact mass spectrometry (GC-EI-MS). We describe a method in which thiols are converted to pentafluorobenzyl (PFB) derivatives by extractive alkylation and the organic layer is evaporated prior to headspace solid phase microextraction (HS-SPME) and GC-EI-MS analysis. Optimal parameters were determined by response surface area modeling. The addition of NaCl solution to the dried SPME vials prior to extraction resulted in up to less than fivefold improvement in detection limits. Using 40 mL wine samples, limits of detection for 4-MMP, 3-MH, and 3-MHA were 0.9 ng/L, 1 ng/L, and 17 ng/L, respectively. Good recovery (90%–109%) and precision (5%–11% RSD) were achieved in wine matrices. The new method was used to survey polyfunctional thiol concentrations in 61 commercial California and New York State wines produced from V. vinifera (Riesling, Gewürztraminer, Cabernet Sauvignon, Sauvignon blanc and non-varietal rosé wines), V. labruscana (Niagara), and Vitis spp. (Cayuga White). Mean 4-MMP concentrations in New York Niagara (17 ng/L) were not significantly different from concentrations in Sauvignon blanc, but were significantly higher than 4-MMP in other varietal wines.

sensitivity. In one report, a detection limit of 30 ng/L for 3-MH was achievable using GC-EI-MS detection after an initial LLE of a 200 mL wine sample into pentane followed by back-extraction into alkaline aqueous solution, PFB derivatization, and headspace solid phase microextraction (HS-SPME), [13]. Quantification of 4-MMP and 3-MHA was not reported. By comparison, a 2 ng/L detection limit was achieved for a 10 mL wine sample using on-cartridge SPE derivatization followed by GC-NICI-MS [10]. Ethyl propiolate derivatization followed by GC-EI-MS detection has also been described but detection limits were 10-fold above sensory threshold for 4-MMP and 3-MHA [14]. A SPME-GC-MS/MS method for 4-MMP following methoxime derivatization is reported to have <1 ng/L detection limits [15], but requires the availability of tandem MS, and the method's suitability for other thiols is unclear. Promising methods based on derivatization of thiols with 4,4ʹ-dithiodipyridine [16] or o-phthaldialdehyde [17] were recently described, but requires the availability of an LC-MS/MS for quantification. In summary, current methods for polyfunctional thiol analyses in wine suffer from poor sensitivity or else require intensive sample preparation, large sample sizes, toxic reagents, and/or instrumentation that may not be routinely available.
This paper describes the optimization and validation of a new method utilizing SPME-GC-EI-MS for quantification of PFB derivatives of three polyfunctional thiols, 4-MMP, 3-MH, and 3-MHA, with detection limits at or near their respective sensory thresholds. Beyond the advantages inherent to using the more common GC-EI-MS, the method also simplifies extraction, derivatization, and clean-up steps based on extractive alkylation prior to evaporation of the organic layer and HS-SPME. We also demonstrate that improvements in HS-SPME recovery can be realized by addition of -NaCl solution to the dried organic residue. The new method was then used for measurement of the three thiols in California and New York State wines in V. vinifera wines as well as wines from interspecific hybrid grapes (Vitis spp. and V. labruscana).

Results and Discussion
Analysis of key wine polyfunctional thiols (3-MHA, 3-MH, 4-MMP) in underivatized form is challenging, and reports that have achieved detection limits below sensory threshold have required the use of organomercuric reagents to achieve a high degree of selective pre-concentration. Alternate quantification schemes for thiols involving derivatization prior to analysis exist [18]. We chose to develop an optimized method based on pentafluorobenzylbromide (PFBBr) derivatization, as this approach has previously been demonstrated to be effective for the key wine polyfunctional thiols [9][10][11]19]. Representative chromatograms for wine samples analyzed by the optimized method are shown in Figure 1.

Optimization of Extractive Alkylation Conditions
A schematic overview of the thiol analysis method is shown in Figure A1, and steps subjected to optimization are underlined. To simplify sample preparation and reduce the risk of thiol oxidation before derivatization, we investigated the use of extractive alkylation, in which derivatization and extraction occur in a single step. Under these conditions, PFBBr derivatization of the thiol involves nucleophilic substitution by the thiolate in a heterogenous system, and the reaction typically proceeds more rapidly in the presence of phase-transfer catalysts like 18-crown-6-ether [20]. Although extractive alkylation with PFBBr is widely used [20], only one group has reported an (unsuccessful) attempt to perform extractive alkylation of thiols in wine but no details were presented as to why the procedure did not work [9].
As a first step, we optimized conditions for extractive alkylation. Optimization experiments are typically performed with the goal of maximizing analyte signals, for example, in determining optimal SPME extraction temperature and extraction time for analysis of 3-alkyl-2-methoxypyrazines in grape musts [21]. In our method, we observed up to tenfold variation in both analyte signal and background noise across runs for replicate analyses during optimization experiments. The reason for this variation was unclear, but because the signal intensity and noise intensity were well correlated, it may be due to variation in the drying step. To overcome this problem, we attempted to minimize the detection limit (LOD) in our optimization experiments, where LOD was defined as the minimum concentration necessary to achieve 3 times the noise limit.

Reaction Time and Organic Solvent Volume
The effects of reaction time and organic solvent volume on extractive alkylation were investigated individually. No significant differences in LODs for 4-MMP and 3-MHA were obtained for reaction times of 10, 25, and 40 min at room temperature ( Figure A2). For 3-MH, the LOD decreased significantly by about a factor of two when reaction time was increased from 10 min to 40 min (p < 0.05). Because the detection limit for 3-MH was below sensory threshold even without further optimization, we chose to use a 10 min reaction time for expedience.
The volume of the pentane:diethyl ether organic solvent was found to have no significant impact on LOD when varied over 8 to 16 mL ( Figure A2). This likely indicates that the PFB derivatives are sufficiently non-polar to partition nearly completely into the organic phase for the range of organic volumes used. A volume of 9 mL was selected for the optimized protocol for convenience, although it is possible that even less solvent could have been used.

Sample pH, Sample Volume, PTC Concentration
Optimal values for pH, sample volume, and PTC concentration were evaluated concurrently in a multifactor fractional factor model (center composite face, CCF), and results are shown for each parameter in Figure 2. LOD was significantly affected by pH (p < 0.01 for all thiols), with the lowest LOD values achieved under the most alkaline conditions (pH = 12). As expected, LODs also decreased with increasing wine volume from 10 to 40 mL (p < 0.01 for 4-MMP and 3-MH; p < 0.1 for 3-MHA). We also determined that signal would continue to scale up proportionally when 160 mL of wine was used instead of 40 mL (data not shown). For convenience, the smaller volume of 40 mL was chosen for the optimized protocol, but improved LOD are attainable by using greater sample volumes. Significant interaction terms (p < 0.05) were observed for pH × volume, pH × catalyst and catalyst × volume, although this information was not further used for method optimization. Surprisingly, the presence of the PTC (18-crown-6 ether) did not improve LODs for any of the thiols (Figure 2). It is not clear why the reaction proceeded efficiently even without a PTC, but as a result no PTC was used in the optimized protocol. Mateo-Vivracho et al. have previously reported an improvement in PFBBr derivatization following methoximation of 4-MMP [10]; however, because we observed acceptable sensitivity for 4-MMP without this step, we did not explore methoximation.

Optimization of HS-SPME Conditions
A previous report using GC-NICI-MS reported detection limits <10 ng/L for all three thiols following a 4-fold concentration by solvent extraction, PFBBr derivatization, and large volume (20 μL) injections [9]. Preliminary investigations of liquid injection of the organic phase by our group with roughly the same mass injected (25-fold concentration factor by evaporation of organic phase, 3 μL injection) and GC-EI-MS detection resulted in unacceptable detection limits, approximately 1 μg/L for each thiol (data not shown).
As an alternative pre-concentration approach to partial solvent evaporation, we next explored completely removing the organic solvent prior to SPME extraction. This approach had been successfully applied by another group in analysis of thiol-PFB derivatives using GC-NICI-MS [12]. A two-phase PDMS/DVB fiber was selected, as this was reported to give better peak shape characteristics than a three-phase polydimethylsiloxane/divinylbenzene/Carboxen (PDMS/DVB/CARB) fiber [13]. As a two centimeter PDMS/DVB fiber is not commercially available, a one centimeter PDMS/DVB fiber was chosen for the optimization study.

Reconstitution Volume
We investigated if the addition of aqueous NaCl to the dried-down SPME vials could increase signal and decrease detection limits, which has been previously reported for quantification of pesticide residues in soil [22]. We observed a significant decrease in LOD for all three thiols upon addition of aqueous NaCl (p < 0.1 for dry vial vs. 10 mL buffer), as shown in Figure 3. The largest effect was observed for 4-MMP, where the LOD decreased by five-fold when 10 mL of NaCl solution were added, as compared to the control with no added solution. A potential explanation for this effect is that the aqueous solution used for reconstitution can compete for active sites on the glass SPME vial, decreasing adsorptive losses of the thiol-PFB analytes. Alternatively, the presence of water could alleviate hot spots of the SPME vial during extraction, preventing thermal degradation of the analytes.

Figure 3.
Detection limits for thiols obtained for dry vials or for vials reconstituted with 5 or 10 mL of aqueous buffer. Error bars represent ± 1 standard deviation for two replicates.

HS-SPME Extraction Time and Temperature
Optimal values for SPME extraction temperature and time were evaluated concurrently in a multifactor CCF model, and results are shown for both parameters in Figure 4. SPME extraction time had a significant effect for all three thiols (p < 0.001), with the longest extraction time of 60 min decreasing detection limits by two to three orders of magnitude. Extraction temperature also had a significant effect on detection limits for the three thiols (p ≤ 0.1 for 4-MMP and 3-MHA; p < 0.0001 for 3-MH). The extraction temperature chosen for the optimized method was 70 °C.

Linearity and Limits of Detection
Linearity was evaluated by addition of standards to a model white wine at eight levels. The observed linearity was excellent for the three thiols (r 2 > 0.99, Table 1). The linear range for 3-MH and 4-MMP covered the full range of standards (6-20,000 ng/L). The linear range for 3-MHA (63-6325 ng/L) was more limited, as the two lowest concentrations were at or below the detection limit, and the highest concentration interfered with the deuterated standard trace.  Method detection limits were determined by standard addition of thiols to a model wine and are reported in Table 1. Signal-independent noise and limits of detection were calculated based on all analytical replicates using Pallesen's method [23]. The limits of detection for 4-MMP (0.9 ng/L) and 3-MH (1 ng/L) compared well to their respective sensory thresholds, 0.8 ng/L and 60 ng/L. The LOD for 3-MHA was somewhat higher than its sensory threshold (17 vs. 4.2 ng/L) although still below concentrations reported in many Sauvignon blanc wines (mean = 29-516 ng/L, depending on region [24]) as well as other white wines [3]. The high LOD of 3-MHA in comparison to those achieved for 4-MMP and 3-MH was due largely to the lack of characteristic, interference-free ions for 3-MHA-PFB.

Accuracy and Precision
Accuracy was evaluated by recovery spikes of the three thiols into a commercial Pinot grigio wine at high (500 ng/L) and low (60 ng/L) concentrations of each thiol. The native concentrations of thiols in the wine were: 3-MH (228 ng/L), 3-MHA (<LOD), and 4-MMP (2.1 ng/L). Recovery was acceptable, and ranged from 90%-109% of the expected values ( Table 2). The calibration curves were generated in model wine, indicating that the isotopically labeled standards effectively compensated for any matrix differences between the real and model wine systems. Precision was evaluated by spikes at low (10 ng/L 4-MMP; 25 ng/L 3-MH; 25 ng/L 3-MHA) and high concentrations (100 ng/L 4-MMP; 200 ng/L 3-MH; 100 ng/L 3-MHA) of each thiol into the same Pinot grigio. Relative standard deviations were 5%-7% for the high concentration spikes and 7%-11% for the low concentration spikes, with the worst performance for the low concentration 3-MHA spike.

Comparison of Current Method to Existing GC-MS Methods for Wine Polyfunctional Thiol Analysis
Literature methods that rely on a low selectivity extraction step (e.g., SPE, SPME) to pre-concentrate underivatized wine polyfunctional thiols (3-MH, 3-MHA, 4-MMP) prior to GC-EI-MS have not achieved detection limits below sensory thresholds [6,7]. Quantification of these thiols below sensory threshold can be achieved using organomercuric reagents and anion exchanges for selective pre-concentration [4], but the approach has been criticized for needing large sample volumes and hazardous reagents [10,11].
More recent reports have achieved detection limits below sensory thresholds by preparing PFBBr derivatives of the thiols prior to GC-NICI-MS detection. The detection limits of these methods are generally better than what we have achieved with our current report: for example Mateo-Vivaracho, et al. reported LODs of 0.1 ng/L, 2 ng/L, and 0.3 ng/L, for 4-MMP, 3-MH, and 3-MHA, respectively, as compared to 0.9 ng/L, 1 ng/L, and 17 ng/L in our work. Correcting for the fact that this earlier report used a 6 mL wine sample compared to 40 mL in our work, NICI-MS provides about a 1-3 order of magnitude improvement in detection limits compared to EI-MS for PFB derivatives, comparable to values reported for other compound classes [25].
However, an advantage of our current method is that EI-MS is more widespread than NICI-MS in laboratories. One other group has utilized GC-EI-MS for quantification of the 3-MH-PFB derivative, although they did not report characterization of 3-MHA and 4-MMP [13]. In that report, 3-MH was initially extracted into pentane and back extracted into aqueous alkaline buffer prior to derivatization. Similar to our approach, the 3-MH-PFB derivative was extracted by SPME prior to GC-EI-MS. We achieved a detection limit of 1 ng/L for a 40 mL wine sample, as compared to 30 ng/L for a 200 mL wine sample in the previous report. The most notable difference between the two reports is that our current work utilized a dry-down of the organic layer step prior to reconstitution and HS-SPME, as opposed to the back-extraction step in the other report. Potentially, this dry-down step removed lower boiling volatiles that could compete with the thiol-PFB derivatives for absorptive sites on the SPME fiber. While we observed a considerable enhancement in S/N when aqueous NaCl solution was added to our dried-down SPME vials, this should have been comparable to the conditions in the other report for alkaline back-extraction, and is thus unlikely to account for the differences. Other potential explanations for the enhanced sensitivity of our current report are more optimal derivatization or HS-SPME conditions, or decreased losses during sample preparation due to fewer work-up steps. GC-EI-MS has also been reported for quantification of ethyl propiolate derivatives of thiols, but detection limits were well above sensory thresholds for 4-MMP and 3-MH [14].
Beyond using EI-MS, another advantage of our extractive alkylation method is that it requires fewer clean-up steps than many other methods that have analyzed PFB derivatives. For example, one report describes an initial loading of the sample onto an SPE cartridge followed by multiple wash steps, introduction of a catalyst and derivatizing reagents, addition of mercaptoglycerol to remove excess PFBBr, and, finally, elution of the derivatized thiol with organic solvent [12]. The organic phase is then evaporated in a SPME vial prior to HS-SPME analysis, similar to our current work. Other reports also use multiple LLE or SPE clean-up steps, often prior to derivatization [9,10,13]. By comparison, the extractive alkylation described here can be performed in a single step.

Quantification of Polyfunctional Thiols in Commercial Wines
Using the optimized method, the three polyfunctional thiols were measured in 31 California wines and 30 commercial New York State wines representing seven wine types: four V. vinifera (Gewurztraminer, Riesling, Sauvignon blanc, Cabernet Sauvignon), one non-varietal V. vinifera (rosé blends), one V. labruscana (Niagara), and one Vitis spp. (Cayuga White). Results are summarized in Table 3. The only thiol detected in all wine samples was 3-MH, and concentrations were significantly higher in the vinifera-based wines than the interspecific hybrid-based wines (Cayuga White and Niagara). However, 3-MH concentrations were not significantly higher in Sauvignon blanc wines than the other white varietal wines, unlike results reported elsewhere [3]. Mean 3-MH concentrations were lower in New York and California Sauvignon blanc (less than 1000 ng/L) than reports from other regions. For example, a survey of commercial Sauvignon blanc wines from Australia, New Zealand, France, and South Africa reported mean concentrations ranging from 1700 to 7100 ng/L [24]. Mean concentrations of 3-MH in Sauvignon blanc in our survey were also lower than what was reported in an earlier survey of US Sauvignon blanc, ~2000 ng/L for eight wines over 3 vintages [24], although the lack of specific information regarding wine region and large standard deviations (~100%) in the previous report makes direct comparison challenging. In contrast, mean 3-MH in New York Gewürztraminer and Riesling (373 and 579 ng/L, respectively) were in the middle of ranges reported for Australian sources of these varietal wines [26]. 3-MH in the California Cabernet Sauvignon was within the broad range reported for Bordeaux red wines previously (10-5000 ng/L) [27]. The contribution of volatile thiols like 3-MH to red wines has not been well studied, but the mean values reported in this study (396-765 ng/L) are comparable to other non-Sauvignon blanc white wines, both in the current study and elsewhere [26].
The reasons for lower concentrations of 3-MH in the California and New York State Sauvignon blanc in this study as compared to other reports is unclear, as several factors are known to affect final 3-MH concentrations. For example, wine grapes are typically hand-harvested in the Finger Lakes and Napa, and it is reported that hand harvesting grapes results in lower 3-MH precursor concentrations and eventually lower 3-MH in wines as compared to machine harvesting [28]. However, this argument would not be appropriate for California Central Coast where machine harvesting is common. Other factors reported to affect final 3-MH concentrations include vineyard nitrogen availability, yeast assimilable nitrogen, abiotic stresses, disease pressure, oxygen status during maceration, and post-fermentation oxidation [1], and a recent study has reported modest correlations of over 20 grape metabolites (amino acids, fatty acids, etc.) with wine 3-MH concentrations [29]. Discerning which of these factors would lead to low 3-MH concentrations in Sauvignon blanc and Gewürztraminer samples in this study is not possible.
It was found that 3-MHA was only detectable in two out of the 30 New York wines evaluated; one rosé wine, and one Sauvignon Blanc wine. Formed by partial enzymatic acetylation of 3-MH during fermentation [30], 3-MHA will decrease during storage due to acid-catalyzed hydrolysis [31]. As a result, average 3-MHA values are reported to be 10%-20% the concentration of average 3-MH values in Sauvignon blanc [29], and can be much lower [24]. Since the average 3-MH concentration across New York State wines in this study was 195 to 569 ng/L, and many of the wines were 2-3 years old at the time of analysis, it is unsurprising that 3-MHA was generally below the 17 ng/L detection limit of the method. Additionally, New York State grapes tend to be low in α-amino acids as compared to other regions [32], and several of these amino acids have been correlated with lower 3-MHA/3-MH ratios [29]. Furthermore, 3-MHA could not be quantified in the California Sauvignon blanc samples due to an unknown interference on the quantifier ion, which could not be identified by library search. However, 3-MHA could be detected in California Cabernet Sauvignon, and mean values across sub-regions (46-67 ng/L) were approximately 10% of 3-MH concentrations. Similar to 3-MH, there is a lack of analytical data regarding 3-MHA in red wines, but values are within the range estimated for Bordeaux red wines, 1-200 ng/L [27].
In contrast to 3-MH and 3-MHA, mean 4-MMP concentrations in both New York (27 ng/L) and California (44-50 ng/L) Sauvignon blanc wines were greater than the mean values (5-10 ng/L) reported for Sauvignon blanc from most other wine regions [24]. The highest reported mean value for 4-MMP in this previous survey was for the Wairarapa region of New Zealand (19 ng/L), but this region also had order of magnitude higher concentrations of 3-MH. In addition, 4-MMP was not detectable in other V. vinifera wines, except for trace concentrations (mean = 2.5 ng/L) in NY State Riesling wines.
To our knowledge, this is the first report of thiol concentrations in any American wines from specific wine regions, rather than the general category of US wine [24]. The study also included the first analysis of thiols in wines produced from interspecific hybrids-Niagara and Cayuga White. The former contains V. labrusca parentage, and the latter is a complex interspecific hybrid with mixed parentage. The 3-MH concentrations in Cayuga White were significantly lower than in vinifera wines, and the other two thiols were not detectable. However, concentrations of 4-MMP in New York Niagara (mean = 18 ng/L) did not differ significantly from New York Sauvignon blanc (p > 0.05), and were significantly higher than all other varietal wines (p < 0.05). The "foxy" smelling o-aminoacetophenone and methyl anthranilate have previously been implicated in contributing to Niagara juice and wines [33,34]. Because 4-MMP exceeded sensory threshold by an order of magnitude in the majority of Niagara wines, 4-MMP may also be important to the varietal character of Niagara wines, but confirmation by sensory studies was not part of this study.

GC-MS Conditions
Quantification of the derivatized thiols during method optimization, method validation, and analysis of commercial samples was carried out by gas chromatography-electron impact-mass spectrometry (GC-EI-MS). Samples were analyzed with an Agilent 6890N GC coupled to an Agilent 5973N mass spectrometer (Santa Clara, CA, USA).
Optimization of extractive alkylation and HS-SPME conditions (Section 2.3) was performed on an Agilent DB-5 column (30 m × 0.25 mm × 0.5 μm) column. Samples were injected in splitless mode at 250 °C, and the split turned on after two minutes. The GC was operated at a constant pressure of 17 psi and helium was used as a carrier gas with an initial flow rate of 2 mL·min −1 . The temperature program was: starting temperature of 50 °C with an initial ramp of 5 °C·min −1 up to 200 °C; then 20 °C·min −1 ramp to 250 °C, five minute hold.
Because of intermittent interferences and peak tailing on the DB-5 column for real wine samples, an Agilent DB-FFAP (30 m × 0.25 mm × 0.25 μm) column was used for determining figures of merit and analysis of real wine samples (Sections 2.4 and 2.5). The GC was operated at a constant pressure of 10 psi and helium was used as a carrier gas with an initial flow rate of 1.2 mL·min −1 . Samples were injected in splitless mode at 250 °C, and the split turned on after two minutes. The temperature program was: starting temperature of 50 °C with an initial ramp of 5 °C·min −1 up to 225 °C; then 20 °C·min −1 ramp to 250 °C, five minute hold.
The MS data was collected using selective ion monitoring (SIM) for specific time intervals. The quantifying ion, qualifying ions, and retention time for each compound and its corresponding deuterated standard are listed in Table 4. Data processing was carried out by Agilent Enhanced ChemStation v. E2.01 software.

Overview
A schematic overview of the thiol analysis method is shown in Figure A1, and steps subjected to optimization are underlined. All optimization experiments were performed on a model wine matrix. Descriptions of the optimized parameters, along with the values investigated, are listed in Table 5. Because of the large number of parameters optimized, it was necessary to fix values for some parameters while optimizing conditions for other parameters. These are listed in Supplementary Table A1. Thiol concentrations in the model wines (3-MH, 3-MHA, 4-MMP) were adjusted among optimization experiments to achieve acceptable signal-to-noise ratios for each experiment; these concentrations are also listed in Supplementary Table A1. Optimal conditions were defined as those that resulted in the lowest detection limit. Limits of detection for each sample were calculated as 3 × noise, where the noise was estimated as the peak-to-trough height of a baseline interval adjacent to the peak by ChemStation (Agilent).

Optimization of Sample Preparation Conditions
Five derivatization parameters were investigated: agitation time, organic solvent volume, sample pH, sample volume, and concentration of the 18-crown-6 ether phase transfer catalyst (Table 5). Three factors (sample pH, sample volume, phase transfer catalyst [PTC] concentration) that were suspected to have interaction terms were investigated using a center composite face-centered (CCF) model. Following determination of optimal conditions for these three factors, the remaining two factors-agitation time and solvent volume-were optimized individually because the former was not expected to interact with other parameters and the latter was too challenging to optimize in a CCF model because of its effects on evaporation time and vial size.
After optimization of derivatization conditions, three HS-SPME parameters were optimized-buffer reconstitution volume, extraction time and extraction temperature. Buffer reconstitution volume was optimized independently first, and then HS-SPME extraction time and temperature were examined by a CCF model.

Optimization of Sample Extraction Method
In a 60 mL screw cap glass vial, 40 mL of wine was spiked with 80 µL of a d10-4-MMP standard solution 10 [H2]-4-mercapto-4-methyl-2-pentanone (0.27 mg/L in 2-propanol), 80 µL of a d8-3-MH standard solution of 8 [H2]-3-MH (0.22 mg/L in 2-propanol), and 80 µL of a d5-3-MHA standard solution (0.20 mg/L in 2-propanol). The pH of the mixture was adjusted using a 2 M solution of NaOH to a pH of 12 (±0.1) and 30 µL of the derivatizing agent (100 µL of 100% pentafluorobenzyl bromide (PFBBr) in 5 mL of 2-propanol) was added. The organic solvent (1:3 v/v pentane:diethyl ether, 9 mL) was added, and the sample was agitated on a shaker table for 10 min at room temperature (The Belly Dancer ® Hybridization Water Bath, Stovall Life Science, Greensboro, NC, USA). To break the resulting emulsion, the mixture was transferred to a centrifuge tube and centrifuged for 5 min at 13,000 rpm. The organic layer was transferred to a 20 mL SPME autosampler vial. Avoiding transfer of the aqueous layer or any emulsion was critical to avoid lengthy dry down times.

Optimization of HS-SPME Analysis Conditions
The organic solvent was evaporated under N2 at room temperature. Near complete removal of solvent was critical, as residual hexane resulted in a loss of signal. Prior to HS-SPME, 10 mL of H2O and 2 g of NaCl were added to each vial. HS-SPME analyses were performed by a CombiPal system (Cohesive Technologies, Alpharetta, GA, USA). A 1 cm, 65 µm, SPME fiber (PDMS-DVB) was used for all

Statistical Analyses
For the optimization experiments, CCF design-of-experiment and linear regression were performed on JMP version 9 (SAS Institute, Cary, NC, USA). For comparison of polyfunctional thiol concentrations among different wines, ANOVA and Tukey multiple-means tests were performed using JMP 9.

Conclusions
We have developed an optimized method for quantification of PFB derivatives of key polyfunctional thiols in wine (4-MMP, 3-MH, and 3-MHA). The method achieves good linearity and precision, and detection limits at or approaching the sensory thresholds for each thiol while utilizing commonplace GC-EI-MS instrumentation. Sample preparation was streamlined and detection limits improved by extractive alkylation followed by evaporation of the organic layer and back-addition of aqueous NaCl prior to HS-SPME extraction. We also report the first characterization of these thiols in wines from specific US wine regions (California and New York), as well as the first characterization in wines from interspecific hybrids, and generally observed lower concentrations of 3-MH/3-MHA and comparable or higher concentrations of 4-MMP as compared to the same varietal white wines from other regions.