Identification and Categorization of Volatile Sulfur Flavor Compounds in Roasted Malts and Barley

Abstract We report for the first time the application of HS-SPME-GC coupled with sulfur-specific pulsed flame photometric detection to sensitively analyze the volatile sulfur compounds (VSC’s) present in drum roasted malt and barley samples typically used in brewing. Twenty-five VSC’s were identified across a range of 9 roasted products produced from barley/malt. Thiophenes (n = 10) were a major class of heterocyclic sulfur compounds identified, along with thiazoles (n = 4), and thiofurans (n = 2). Quantitative (n = 18) and semi-quantitative (n = 6) data are reported for VSC’s across this product range. Principal Component Analysis (PCA) of data clearly identified (PC1) that heterocyclic sulfur compounds were formed in products processed at high temperatures (>170 °C) under dry conditions (roasted barley, chocolate and black malts). Whereas compounds such as methyl dithiolane and methyl propyl sulfide were associated primarily with lower temperature finished products (crystal, amber and cara malts). Pathways for the generation of observed VSC’s are considered alongside typical roasting conditions employed in the manufacture of these products. Concentrations of VSC’s identified will certainly contribute characteristic aromas to the roasted products themselves. The transfer of VSC’s from the grist into finished beer, and their sensory impact in a range of beer styles, remains to be determined. Supplemental data for this article is available online at https://doi.org/10.1080/03610470.2021.2003669 .


Introduction
Roasted cereal products are used in brewing to introduce color and flavor into the grist [1] and can impart a range of desirable sensory characteristics to finished beers. [2,3] Roasting of cereal products generates flavor compounds through a host of reaction pathways, known collectively as thermal flavor generation. These pathways include classic Maillard chemistry, [4] caramelization of sugars, Strecker degradation, [5] lipid degradation and pyrolysis (thermal breakdown of molecules at higher temperatures). Not surprisingly therefore, the flavor chemistry of roasted products is complex and desirable roasted flavor characteristics such as 'coffee' , 'chocolate' or 'nutty' for example, are the result of interactions between multiple flavor compounds. This complexity defies simple control of thermal chemistry to generate specific flavors. However, maltsters have long-since learned to control the underlying chemistry to accentuate certain reaction pathways and sensory impacts in different product categories; caramel and chocolate malts are obviously named examples, albeit that the product ranges they exemplify contribute many other flavor compounds and that a chocolate malt will often smell as much of coffee and roasted character as chocolate.
Our group has studied the links between the barley or malt substrate, which is roasted, the roasting conditions and the formation of marker flavor compounds. [6,7] Yahya et al. [7] reported the time course of formation of key compounds through commercial roasting operations as a function of substrate type, moisture content and roasting conditions. This work involved sampling from commercial roasting operations during manufacture of roasted barley, crystal and black malt products. Parr et al. [6] modeled the formation of 20 key roasted product flavor compounds as a function of process time, temperature and substrate type. A laboratory-scale roaster was designed for accurate temperature control with batch sizes of 100 g. This minimized bulk effects leading to inconsistent conditions across a malt bed and also enabled isothermal studies of flavor formation with time, whereas in commercial roasting rapid temperature gradients are applied at key stages. This improved knowledge of the specific conditions under which certain types of thermally generated flavor compounds are formed. [6] Across these two trials, substrate moisture content, stewing time (for green malt roasted products) and the applied temperature gradient and finishing temperature were shown to be major determinants of flavor production. Stewing at moderate temperatures (e.g., 65 °C) at the start of roasting of green malts generates more Maillard reaction precursors (reducing sugars and amino compounds). Furthermore, subsequent heating to moderate finishing temperatures (e.g., 135-155 °C) from relatively high initial moistures (e.g., 40%) encourages 'Maillard reaction in the liquid phase' [8] which generates the classic caramel type flavors associated with these products. However, roasted pale malt products start with a substrate that is at around 5% moisture and as such are 'dry roasted' products, with finished flavors generated through 'Maillard reactions in the dry phase' [8] and with increased flavor contributions from pyrolytic reactions. High temperature dry roasted product flavors have been found to be heavily influenced by production of heterocyclic products of thermal flavor generation. These are cyclic structures featuring a non-carbon hetero-atom such as nitrogen (e.g., in pyrazines, pyridines). The models reported by Parr et al. [6] indicated, for example, that pyrazine formation accelerated greatly in roasted barley or pale malt at temperatures greater than 175 °C. Such compounds are thus expected to be accentuated in high temperature-finished dry roasted products such as chocolate and black malts.
In recent studies, [9] we attempted to characterize the key aroma active compounds in a range of roasted malt products using a GC-Olfactometry (GC-O) approach. This proved only partially successful. Firstly, due to the complex mixture of volatiles present, it was difficult to achieve a suitable chromatographic separation of compounds that facilitated sniffing at the odor port for a reasonable amount of time -even when splitting chromatographs between two assessors to avoid fatigue. Furthermore, whilst some odors could be clearly traced to underlying compounds identified by GC-MS, other odors reported by the panel were inconsistent with the known aromas of compounds eluting at a particular retention time. For others, no compound could be identified due to a lack of analytical sensitivity for compounds present that presumably had a low odor threshold. The nature of some of these roasted aromas that were evident at very low concentrations led us to believe that volatile sulfur compounds were most likely important in these systems, which prompted the present investigation.
Volatile sulfur compounds (VSCs) are found in a variety of foods, for example coffee, beer, wine, meat, cheese, onion and garlic. [10][11][12] In each of these products, VSCs contribute a range of flavor characteristics, depending on the combination of VSCs found, and their concentrations. Sulfur aroma compounds typically have low sensory detection thresholds, and so can be present in foods at very low concentrations whilst remaining flavor active. [13] As a result of this trait, VSCs contribute significantly to the overall aroma of many products. The off-putting reputation of VSCs (i.e., 'rotten egg' 'onion' , 'cabbage' , etc.) often results in the positive contribution of these compounds being overlooked, although sulfur compounds can also contribute to the aroma and flavor of foods positively. [13] Specifically, in roasted coffee, sulfur volatiles are known to make a significant contribution to the overall roasted aroma of the product. [14] Typically, high concentrations of VSCs result in more unpleasant off-flavors and aromas, whereas low concentrations of a combination of compounds can impart desirable savory characteristics to the overall aroma of a product.
The quantification of VSCs is particularly challenging due to their high reactivity and low concentrations in products. [15] Conventional methods of liquid injection with GC-MS are not sensitive enough to detect the presence of low-level VSCs. Preconcentration of the volatiles in the sample is common in VSC analysis. [12] In this investigation, the preconcentration step of headspace solid-phase microextraction (HS-SPME) was used in combination with gas chromatography, coupled with a pulsed flame photometric detector (PFPD) operated in sulfur mode. This detector is sulfur-specific and therefore more sensitive for the detection of VSC's at low levels. Fang and Qiang [10] found the quantification limits of SPME-GC-PFPD could be 0.5 µg/L for the most volatile sulfur compounds in wines, which is below, or within the range of most sensory detection limits. In comparison to other sample analysis techniques, headspace SPME reduces analysis duration and sample manipulation. [15] Hill and Smith [13] developed a successful analysis method (on which the one in this investigation is based) by using SPME-GC-PFPD for the analysis of beer. The key raw materials used in the brewing process (malt, roasted malt, and hops) contain a range of VSCs, some of which may remain, along with others that are formed, across the entire brewing process. [16] In this study, we apply HS-SPME-GC-PFPD to analyze sulfur compounds present in a range (n = 9) of barley commercial roasted products and report for the first-time quantitative data for the concentrations of 24 such compounds. The VSC contents of commercial roasted malts and barley have not previously been characterized systematically. Therefore, this investigation reports new knowledge of the role of VSCs in roasted product flavor and discusses the possible pathways for their formation and control during the roasting process.

Roasted products
Nine commercial roasted products were sourced from Boortmalt (Pauls Malt Ltd., Knapton, UK): roasted barley, black malt, chocolate malt, pale chocolate malt, dark crystal malt, medium crystal malt, light crystal malt, caramalt, and amber malt (20 kg sacks of each roasted product).
Portions of the samples were milled to produce a fine powder, using a Buhler Miag disc mill (Uzwil, Switzerland), 0.2 mm gap setting. The mill was cleaned between samples, to avoid cross-contamination.
Milled samples were vacuum-sealed in foil-lined pouches and stored at −80°C ready for subsequent analysis. Quantities of each unmilled roasted product were also vacuum-sealed in foil-lined pouches at −80°C for later use.

Chemicals
Authentic chemical standards were used to identify the sulfur containing compounds present in the roasted samples. Sourcing and purity of the volatile sulfur compound standards used in the study are reported in Supplementary Data Table 1.

Moisture content
The moisture content of roasted products was determined according to EBC Analytica Method 4.2. For each sample, three replicate determinations were carried out.

EBC color analysis
Roasted product color was analyzed according to EBC methodology. Firstly, a Congress Mash of each sample being analyzed was produced according to Method 4.5.1 of EBC Analytica using a 1-Cube Mash Bath (Havlíčkův Brod, Czech Republic). For the preparation of roasted malt mash, the milled roasted sample (25 g) was combined with pale malt (25 g) in a pre-weighed mash beaker. The malts were mixed together with 200 g water (at 45 °C). The contents were stirred in the mash bath for 30 min at 45 °C, after which a temperature ramp of 1 °C/min was applied for 25 min until reaching 70 °C, which was held for 60 min. The mash was then cooled to 20 °C and the contents adjusted to 450 g with water. The contents of the beaker were emptied fully into a filter funnel above a conical flask. The first 100 mL of filtrate was returned to the funnel. Filtered wort sample (50 mL) was immediately passed through a Millipore Millex HA 0.45 μm membrane filter (Merck KGaA, Darmstadt, Germany). The wort was established as clear by <0.02 absorbance at 700 nm when compared to water. Absorbance of the filtered wort at 430 nm was measured using a Biochrom Ultrospec 2100 pro Spectrophotometer (Biochrom, Holliston, MA, U.S.A.). Wort samples were diluted as appropriate to bring absorbance readings within the limits of the spectrophotometer.
To calculate the color units in °EBC (

Headspace sample preparation
Milled malt (4 g; Buhler Miag disc mill, 0.2 mm gap) samples were placed in individual 20 mL glass headspace vials and mixed with 8 mL water containing 5 µg/g benzothiazole as an internal standard. Duplicate samples were prepared for each roasted product. To quantify the analysis, external standard series were prepared by spiking known concentrations of authentic standards on to 4 g of milled raw barley. The raw barley was used in order to mimic the compounds' interactions with the matrix of the roasted samples, which influences volatile partitioning into the headspace. Stock solutions of each compound (Table 2) were prepared at 1000 µg/g in methanol (HPLC grade, VWR International Ltd.) with 5 µg/g benzothiazole as the internal standard. Stock solutions of the compounds were then diluted in water with internal standard (5 µg/g benzothiazole) to the following concentrations: 0.25, 0.5, 1, 5, and 10 µg/g. Standard solutions (8 mL) were added to milled raw barley (4 g; Buhler Miag disc mill, 0.2 mm gap) in individual 20 mL glass headspace vials. Duplicate samples were prepared for each of the five calibrant concentrations.

Solid phase micro-extraction (SPME) conditions
Volatile compounds in the headspace of each vial were pre-enriched onto a 23 ga divinylbenzene/carboxen/ polydimethylsiloxane (DVB/CAR/PDMS) SPME fiber (StableFlex, Sigma Aldrich, Dorset, UK). Each vial was initially incubated for 30 min at 30 °C with agitation at 500 rpm on a 2 s on, 4 s off sequence. The SPME fiber was then inserted into the vial through the septum and incubated to adsorb volatiles at 30 °C for 30 min, without agitation. Desorption of the volatiles was carried out in a PTV (Programmed Temperature Vaporizer) injector for 5 min at 240 °C, in split-less mode.

GC-PFPD operating conditions
The volatile compounds were separated using a Bruker Scion 456 Gas Chromatograph (Scion Instruments, Livingston, Scotland, UK) fitted with a ZB-Wax column (30 m × 0.25 mm ID × 1.0 μm film thickness; Phenomenex, Macclesfield, UK). The oven temperature was programmed as follows: 40 °C for 1 min, then a temperature ramp at 4 °C/min to 220 °C, with a final hold at 220 °C for 10 min. Helium carrier gas was used at a head pressure of 18 psi. The PFPD was operated at 210 °C, with settings as follows: multiplier voltage: 600 V; trigger level: 200 mV; gate delay: 6.0 ms; gate width: 20.0 ms; range: 9 (range 9 was chosen to reduce baseline noise and simplify peak detection. This reduced the area and number of the peaks detected slightly, in comparison to running the PFPD at range 10).

Peak assignment to VSCs
Chromatographic peaks were assigned to compounds based upon two levels of validation. Firstly, by demonstrating chromatographic similarity with authentic VSC standards run under identical conditions. Secondly by calculation of the Kovat's linear retention index (LRI) measured under experimental conditions and comparison with literature sources reported on equivalent WAX phases. [18,19] To calculate LRI values, an alkane standards mixture (C8-C22) was run under chromatographic conditions and LRI's were calculated using the following formula: Where n = the number of carbon atoms in the preceding alkane peak; t n and t n+1 are the retention times of the preceding and following alkane peak and t i is the retention time of the compound of interest.

Statistical analysis
Principal Component Analysis (PCA) of the standardized volatile sulfur compound concentrations across the nine roasted products was carried out using XLSTAT version 2019.1.2 (Microsoft, Redmond, WA, U.S.A.).

Results and discussion
Example GC-PFPD sulfur-compound chromatographs of a roasted barley, pale chocolate malt and medium crystal malt are shown in Figure 1. It is evident from Figure 1 that high temperature 'dry roasted' samples, such as roasted barley and pale chocolate malt, contained a more complex spectrum of sulfur compounds at higher concentrations than was the case for crystal malt products.
Peaks were assigned to compounds by calculating the Kovat's retention index and comparing this with documented values for sulfur compounds reported on a similar GC column phase. Authentic standards were then purchased at GC purity and run under identical conditions to those used to analyze the roasted malts. Compound assignments were confirmed when chromatographic similarity under experimental conditions was demonstrated, and the calculated retention index was in the range reported in the literature. In all, it was possible to assign 24 sulfur compounds to peaks within roasted product chromatograms. For 18 of these, a satisfactory linear calibration was attained when an external standard series was run (R 2 values are reported in Table 2). For these compounds, quantitative data are reported (Table 3) as the concentration within the product (ng/g). For the remaining 6 compounds (thiophene, 2-methyl thiophene, diallyl sulfide, 2,5-dimethyl thiophene, 4,5-dimethyl thiazole and 2-acetyl thiophene) non-linear calibrations were obtained. Semi-quantitative data for these 6 compounds are reported as peak areas corrected against the internal standard ( Table 4). The lack of linearity may be due to the authentic standard compounds' interaction with the matrix of milled raw barley upon mixing.  [17] . § Satisfactory linearity of calibration not obtained.
Many of the sulfur compounds identified across the commercial roasted products were forms of thiophenes (n = 10), including thiophene itself. Thiophene has a parallel structure to furan, but with sulfur as the heterocyclic atom as opposed to oxygen. Substituted forms found in the roasted samples included alkylated, formylated, and acetylated thiophenes. Other classes of heterocyclic VSCs identified included: thiazoles (n = 4), thiophenones (n = 1), thiazolines (n = 1), thio-furans (n = 2), and methyldithiolane. Non-cyclic VSCs were also identified across the samples (n = 6).
Furthermore, a significant number of peaks within the samples remain unidentified, including notable peaks with LRI's of 1294 and 1428. Looking at the order of elution of VSCs under chromatographic conditions, it is tentatively suggested that these are different positional isomers of 2-methyl-5-ethyl thiophene (LRI 1294) and dimethyl thiazole (LRI 1428) for which authentic standards were not sourced.

Principal component analysis of sulfur volatile data
To depict the main sources of variation in the data set, the sulfur volatile compound concentrations (for 28 identified compounds across each of 9 roasted products) were standardized and analyzed using Principal Component Analysis (PCA). Figure  2 shows a biplot for PC1 and PC2, which accounts for 72.7% of the variation amongst the data set. Principal Component 1 (52.8% of variance) clearly separates the commercial products, with positive loadings for all high temperature, dry-roasted products and strong negative loadings for all other products, which were either roasted green malt products (produced at higher moisture/and relatively lower temperatures) or the amber malt sample. That a majority of sulfur compounds load positively on PC1 confirms that the higher temperature, dry roasting conditions strongly favored production of volatile sulfur compounds and in particular sulfur-containing heterocycles (n = 19). Thiophene, 3-methylthiophene and 2-methyl-5-(methylthio)furan had the strongest positive loadings on PC1, with several other substituted furans, thiophenes and thiazoles also loading strongly. Only two sulfur compounds -methyl dithiolane and methyl propyl disulfide -loaded negatively on PC1, indicating that their formation and retention was more associated with roasted green malt products or amber malt. PC2 (Figure 2) is mainly separating out differences between roasted barley and the high-end roasted pale malt products (pale chocolate, chocolate, black). Methyl allyl sulfide, 2-acetyl-2-thiazoline and 2-formyl-5-methylthiophene had the highest positive loadings on PC2 and were therefore strongly characteristic of the roasted barley product. Methional/dimethyl trisulfide (DMTS) loaded most negatively on PC2, indicating they were formed at much higher concentrations in roasted pale malt products than in roasted barley. Co-location of a product label and sulfur compound in Figure 2 indicates that the compound was closely associated with that product. Formation of the substituted furans (2-methyl-3-(methylthio)furan and 2-methyl-5-(methylthio)furan; ('coffee') and 2-formylthiophene ('bitter almond') were most closely associated with chocolate malt products. For roasted barley, the most characteristic sulfur compounds were 2-acetylthiophene, 4,5-dimethylthiazole, 2-methylthiophene and 4-methylthiazole, which together are responsible for roasted, nutty, vegetal and green aroma characteristics.

Pathways to volatile sulfur compounds in roasted products
The PCA biplot (Figure 2) depicts the relative formation of the identified volatile sulfur compounds across the nine roasted products and in doing so clearly separates the roasted products according to the process conditions employed during manufacture and the nature of the substrate loaded into the roasting drum. Between them, these factors determine the relative concentrations of VSC's produced in each case through thermal flavor generation pathways. In this regard, crucial factors are known to be: i) substrate type and whether malted or not; ii) substrate moisture content and iii) roasting time-temperature profile. [6,7] When foodstuffs are treated thermally, sulfur-containing compounds can either react or breakdown to form VSC's. In barley, key sources of sulfur include the sulfur containing amino acids cysteine and methionine as well as thiamine (vitamin B1). Under thermal processing, thiamine can degrade to form hydrogen sulfide. [20] S-methyl methionine, which breaks down to the volatile dimethyl sulfide on heating, is a further well-known example. [21] Key pathways of thermal flavor generation include: caramelization, Maillard reactions, Strecker degradation and pyrolysis. Volatile sulfur flavor compounds can be formed as a result of all of these pathways, principally through reaction of small sulfurous compounds with products of, or intermediates in, caramelization or Maillard reactions. Strecker degradation of methionine to form methional is a notable reaction pathway. Furthermore, Strecker degradation or hydrolysis of cysteine at elevated temperatures generates highly reactive compounds such as H 2 S and NH 3 [22] ( Figure 3). These small, highly reactive molecules may then undergo ring formation reactions with carbon fragments resulting from sugar degradation, to generate heterocyclic aroma compounds. [23] Depending on the carbon fragments that react together, various side chains are produced on the basic ring structures (e.g., methylated, ethylated, formylated, acetylated).

Thiophenes (n = 10)
Thiophenes (2-acetylthiophene, 3-acetylthiophene and 2-acetyl-3-methylthiophene) have been previously identified in roasted coffee, [24] so it is no surprise to see them as major VSC's in roasted barley and malt products. Seventeen thiophenes (including many of those identified here) have been reported to form as products of the thermal reaction between ribose and cysteine. [25] Mechanistic routes for thiophene formation involve the reaction of furfural with hydrogen sulfide [26] or the condensation of mercaptoacetaldehyde with α,β-unsaturated aldehydes. [27] Also, at 130 °C, furaneol can act as a precursor in the formation of thiophene by the exchange of the oxygen atom for sulfur from hydrogen sulfide. [23] The parent compound thiophene was identified in all roasted products (Table 4), with highest levels in the higher temperature dry roasted products, but with readily detectable amounts in crystal and amber malts as well. There was a tendency for roasted barley to contain the highest concentrations of the substituted thiophenes reported here (Table 3) with the exceptions being 2-formyl thiophene (highest in pale chocolate), 2-propyl thiophene (chocolate) and pentyl thiophene (black malt). This may reflect the impacts of the malting process on the relative concentrations of precursors . thermal breakdown of cysteine to yield highly reactive maillard intermediates (after mottram, 1998). [22]  methyldithiolane.
-  required to form substituted thiophenes with the various side chains. The key determinant of thiophene formation overall, however, was product finishing temperature and all of the high temperature dry roasted products contained sufficient precursors to generate a range of thiophenes at concentrations up to the mg/kg range in the product. Substituted thiophenes were either absent or detected at trace levels in the crystal malts, amber and caramalt.

Thiazoles (n = 4) and thiazolines (1)
Thiazole compounds have been reported across a range of cooked food systems, contributing a variety of characteristic aromas. Mono-substituted alkylthiazoles have green, vegetable aromas, whilst compounds such as 4,5-dimethylthiazole, 5-ethyl-2,4-dimethylthiazole and 2,4,5-trimethylthiazole have been described as nutty, roasted and meaty. [27] Their formation requires heating at elevated temperatures, hence they have been identified in fried, roasted, or grilled foods, such as cooked meats, coffee, roasted peanuts and potato chips.
A scheme for thiazole and thiazoline formation proposed by Mottram in 1998 [22] is shown in Figure 4A. This shows the importance of reactive Maillard species such as H 2 S, NH 3 and acetaldehyde (all of which can result from the Strecker degradation of cysteine, Figure 3) to this class of compounds. In the final step, the alkyl-3-thiazoline is reduced to the alkyl thiazole. Alternatively, formation of  et al., 1990). [28] c. formation of 2-acetylthiazole by the reaction of l-cysteine with the abundant maillard intermediate methylglyoxal (after mulders, 1973). [30] 4-methyl-and 4,5-dimethyl thiazoles through the direct hydrolysis of thiamine has been demonstrated [28] as illustrated in Figure 4B. One source cites hulled barley as containing 0.65 mg/100 g of thiamine. [29] In our study, roasted barley contained the highest levels of these two compounds, whilst chocolate malt contained the highest level of two thiazoles (2-ethoxy and 2-ethoxy-4-methyl-). The latter may have been as a result of the prior malting process yielding more precursors for these compounds' formation on heating.
Mulders [30] showed that 2-acetylthiazole was produced from Maillard reactions between cysteine and ribose and proposed a pathway via reaction of L-cysteine with methylglyoxal ( Figure 4C). The compound 2-acetyl-2-thiazoline was identified here, solely in roasted barley product and was likely formed via this pathway. Hofmann and Schieberle [31] investigated the key aroma compounds formed by reactions between cysteine and various carbohydrates (ribose, glucose and rhamnose). It was discovered that up to four times the flavor dilution (FD) factor of 2-acetyl-2-thiazoline was detected under dry heating conditions in comparison to when formed in an aqueous solution. [31] Thiophenones (n = 1) The compound 4,5-dihydro-2-methyl-3-thiophenone is another well-known flavor volatile reported previously as a thermal degradation product of thiamine. [32] It has a sulfurous, fruity-berry aroma and was identified here solely in the pale chocolate malt.

Thio-furans (n = 2)
The compound 2-methyl-3-(methylthio)furan was amongst the sulfur volatiles formed from a reaction mixture of xylose, cysteine and thiamine. [33] Cerny and Davidek [34] investigated the reaction mechanism of products of the ribose-cysteine Maillard reaction and concluded that whilst the furan ring of 2-methyl-3-(methylthio)furan logically derived from ribose, the methylthio group carbon derived both from the sugar and from cysteine, via a mechanism as yet unknown. Substantial quantities of this and the isomer 2-methyl-5-(methylthio)furan were identified in all high-end dry roasted products, with the chocolate malt containing the highest concentrations of each. Amongst other descriptors, these compounds impart coffee aroma characteristics.

Dithiolanes (n = 1)
Methyl dithiolane can be formed from the condensation of aldehydes, H 2 S, and mercaptoacetaldehyde. [35] It is a known product of thiamine or thiamine/cysteine reaction flavor systems. [36] Methyl dithiolane was only identified in the amber malt sample, which suggests it is not formed so efficiently in roasted green malt products at higher product moisture contents and does not survive high temperature dry roasting. The latter could be due to volatility, thermal decomposition or onwards chemical reactions.
The major non-heterocyclic peak that was present in all roasted product chromatograms eluted at LRI 1106. Under experimental conditions, this was found to be a peak that resulted from the co-elution of two compounds: DMTS and methional (the Strecker aldehyde of methionine). In subsequent investigations using varied chromatographic conditions (data not shown), we identified that both of these compounds were present in all samples, but that the major component was always DMTS, irrespective of the roasted product. For example, when the peaks were separated, the percentage of total peak area due to DMTS was 93.3% (chocolate malt), 94.1% (medium crystal) and 74.4% (roasted barley). In Table 3 this peak was thus quantified against a DMTS standard series, although it should be noted that reported concentrations are subject to some error, due to the low levels of co-eluting methional.
Methional can be formed by Strecker degradation of methionine (sulfur containing amino acid) during barley malting. [37] DMTS can be formed via methanethiol with methional as a precursor. [38] Granvogl et al. [39] reported that the Strecker degradation of methionine yields different product mixtures depending on moisture content. Under aqueous conditions, methional could be further broken down to yield methanethiol and dimethyl disulfide. However, under dry heating conditions, a 3-oxazoline intermediate was formed that could then break down in the presence of water to yield methional.

The influence of roasting substrate on VSC production
Whilst PC1 of Figure 2 makes it clear that finishing temperature is a key determinant of VSC formation in roasted products, PC2 also draws out differences between the roasting of barley, as opposed to pale malt in this regard. This could be due to changes in precursor concentrations brought about by the malting process. Cysteine and methionine are the two sulfur-containing amino acids found in proteins. Hulled barley has been reported to contain around 0.28 g/kg cysteine and 0.20 g/kg methionine. [29] Singh and Sosulski [40] investigated the amino acid composition of hulled Harrington barley over 8 days germination. Their data showed that methionine levels as a proportion of total nitrogen decreased from 2.0 to 1.4 g/16 g nitrogen over the first 2 days of germination, whereas the data for cysteine held constant at around 2.2 g/16 g nitrogen. Mikulíková et al. [37] reported that the methionine concentration in malt kilned at over 50 °C decreased significantly from 180 µg/g (at 50 °C) to less than 10 µg/g with increasing kilning/roasting temperatures above 120 °C. These studies together suggest that malt as a substrate contains less methionine than barley; however, due to prior thermal processing on the kiln, malt will contain Maillard intermediate forms of sulfur other than methionine, which yield distinct products on roasting. This is a likely source of the variation in VSC's observed in roasted barley versus roasted pale malt products. For example, 3-oxazoline intermediates formed in kilned malt from methionine [39] might break down during water sprays in the roasting drum to yield methional in black and chocolate malts at notably higher levels than in roasted barley.
Green malt stewed and roasted products contained far fewer VSC's and the heterocyclic thiazoles and thiophenes were almost entirely absent. The higher moisture substrate (starting moistures typically around 40% as-is) and intermediate final roasting temperatures doubtless both contributed to this fact. Thiophene itself was identified in all nine roasted products. However, the crystal/caramalt products still contained the lower thiophene concentrations, followed by amber malt and then all of the higher temperature dry roasted products that contained 2-4 times as much thiophene as the amber malt, indicating the key role of finishing temperature.
The two sulfur compounds most closely associated with the lower temperature finished products in Figure 2 (methyl dithiolane and methyl propyl sulfide) were at their highest concentrations in the amber and dark crystal malts. Crystal malts have finishing temperatures in the region of 150 °C and Boortmalt state that their dark crystal malt is finished at the same temperature as the medium crystal product, but with prolonged roasting. [41] This suggests that methyl dithiolane and methyl propyl sulfide production was favored by dryer conditions but dis-favored by very high temperature finishing, probably due to volatilization, thermal instability or onwards reactions at higher temperatures.

Influences of roasting conditions on VSC production
Whilst precise production conditions for the samples used in this research were not disclosed, outline details of the production conditions for the Boortmalt range of roasted products are available on the internet [42] and are referred to here.
The chocolate malt sample yielded the highest number of VSC peaks, many of which were at the highest concentrations of all of the roasted products. Inspection of the lower right quadrant of Figure 2 indicates that overall VSC concentrations generally decreased for the high temperature finished pale malt products in the order chocolate > pale chocolate > black malt. During roasting of chocolate malt, temperatures up to 225 °C [43] are reached. Care is taken throughout roasting to avoid scorching of the grains, and the color development is closely monitored. This prevents excessive roasting, thereby retaining the key flavor compounds formed during thermal flavor generation. Black malt is roasted to 230 °C, for a longer period of time. [44] Such prolonged roasting would result in the loss of VSC's due to volatilization, or by taking part in further reactions leading to the formation of high molecular weight melanoidins at the final stages of the Maillard reaction. [45] As a result of this, the aroma of black malt is diminished in comparison to malts roasted to a lesser degree, but it still imparts a smoky, astringent characteristic to beer. [2,3,44] The finishing temperatures of chocolate malt and black malt differ by only 5 °C (225 °C and 230 °C, respectively). [43,44] In addition to this, water sprays are commonly used in commercial roasting under severe roasting conditions in order to prevent charring, and to encourage color development. [7,44] It is likely that the use of quenching in the production of the very darkest roasts (i.e., black malt, and roasted barley) facilitated some incorporation of VSC's into melanoidins. [46] Amber malt is characterized by the dry, biscuit flavor it imparts to a brew, and the Boortmalt product is typically roasted to temperatures between 100 °C to 150 °C. [47] Seven VSCs were identified in the amber malt sample, two of which were exclusively in this sample: methyl propyl disulfide ('onion, radish, mustard, tomato' aroma), and methyldithiolane ('onion, rooty, vegetal' aroma). Excepting these two compounds, the concentrations of the VSCs identified within the amber malt sample were rarely at high concentrations relative to the other analyzed samples.

The significance of VSC's to roasted product and beer flavor
The customary way to appraise the likely sensory impact of volatile aroma compounds is to express their concentrations relative to their sensory threshold concentrations as an odor activity value (OAV). For many of the VSC's identified here, accurate sensory threshold data are not available; furthermore, in complex mixtures, such as those resulting from Maillard reactions, sensory thresholds are usually lower than when determined with individual compounds, so individual OAV's have a limited meaning. Additionally, we have determined the concentrations of these compounds in roasted products themselves and not the concentrations they would release orthonasally or retronasally during consumption. However, bearing in mind the very low sensory thresholds of this class of compound generally, it is obvious that many of the VSC's analyzed here would contribute to the sensory perception of roasted products as in many cases they were found to be present at mg/kg (parts per million) levels in the roasted barley, chocolate and black malts. What remains to be determined is the extent to which these compounds transfer through the brewing process, or are transformed by it, to determine the flavors imparted to dark beers brewed with higher grist percentages of these products. In an earlier investigation, [2] we reported the sensory characterisation of roasted products and of beers brewed incorporating them. Descriptors such as 'treacle' 'burnt' , 'coffee' , 'smoky' , 'medicinal' and 'sulfury' were used to describe these beers. The chocolate malt beer (8% chocolate malt in grist) had the highest sensory scores for 'smoky' , 'coffee' , 'burnt' and 'treacle' aromas. It is highly likely that these sensory attributes are in part due to the compounds characterized in the present paper. Likewise, the beer brewed with amber malt incorporation (20% of grist) was scored most highly for 'sulfury' aroma and flavor; this could in part be due to compounds such as thiophene, 1-heptanethiol and methyl dithiolane, which were identified here in the amber malt sample. However, further studies are required to determine which VSC's survive into beers in the greatest amounts and are most important for the flavor of beers brewed with a range of roasted products.

Conclusions
The majority of the sulfur volatiles identified in this study were formed as a result of Maillard chemistry (including Strecker degradation) and/or pyrolysis reactions originating from sulfur contained in the amino acids methionine and cysteine, or other organic forms such as thiamine. Roasting parameters (roasting temperature and roasting time) had the greatest influence on the occurrence and concentration of VSCs in the samples, in addition to the substrate being roasted. Higher roasting temperatures, and for longer periods of time, resulted in more VSCs being identified overall, and typically in higher concentrations. The major exception to this general trend was the black malt sample, which contained lower concentrations of compounds identified in similar samples roasted to a slightly lower temperature, and possibly for a shorter period of time (chocolate malt and pale chocolate malt). It is likely that prolonged roasting at high temperatures may result in the loss of aroma volatiles due to the formation of melanoidins as a product of the final stage of the Maillard reaction.
In addition to the differences created by a variety of roasting parameters, an inherent compositional difference between malted barley and unmalted barley was highlighted. The conversion of methionine over the course of malting to more volatile VSCs may be a major source of the differences in identified VSCs between roasted malts and roasted barley.
Dark crystal malt exhibited the highest concentrations of identified VSCs of all the crystal malt samples, confirming that increased degrees of roasting favor the formation of VSCs to higher concentrations. Although a total of eight VSC peaks were identified across the range of crystal malts, it is clear that the roasting of green malt does not facilitate the formation of VSCs to high concentrations. The sensory characteristics of these crystal malts do not correlate with the expected savory, meaty aromas of sulfur volatiles, indicating their low sensory activity in crystal malt products.
The sulfur volatiles identified in this study contribute to the overall aroma of the roasted products. The effect of VSCs on the pronounced aroma of roasted malts and roasted barley may be reduced (or on occasions enhanced) upon subsequent processing steps in beer production due to volatilization, further flavor generation reactions, or assimilation or biotransformation by yeast.