Visualization of the molecular complexity

A diverse set of 85 bottled beers from different countries and of different types was profiled as the first batch. To explore the compositional diversity and molecular complexity of each individual beer the samples were analyzed by flow-injection ESI(−) FTICR-MS. The chemical space of beer is as diverse as the variety of different raw materials and their treatment during the brewing process including malting, roasting, boiling, fermentation, and filtration. As an example, Fig. 1 shows the spectrum of a Pilsner beer. The macroscopic general view (Fig. 1a) shows the abundancy of (oligo)saccharide patterns. However, the detailed view of a single nominal mass (Fig. 1b) revealed up to 27 m/z values within the mass of 391, which could be assigned to molecular formulae with a mean error of <0.1 ppm (<1/10 of an electron mass, respectively). The molecular variety of the beer samples, which ranges from peptides [C19H28N4O5], carbohydrates [C13H24O11], fatty acids [C21H40O4] through their sulfates [C18H31O7S] to isotopologues of potential Maillard reaction products like desoxyfructosyl(iso-)leucine [13C1C11H23NO7], could be displayed in one single nominal mass by highly resolved FTICR measurements. In total, an average of 2800 compositions could be found in each beer spectra. Bearing in mind that distinct isomers exist for a given formula, the 27 molecular formulae in the spectrum excerpt represent 68 hits reaching from 0 to 11 isomers in common databases. Therefore, the FIA-FTICR-MS spectrum of a single beer can be considered as an instantaneous overview of several thousands of compounds present in various concentrations. All m/z values assigned to a molecular formula and present in at least 5% of all beer samples are depicted in a two-dimensional van Krevelen diagram (Fig. 2). Thereby the masses can be associated to chemical families like carbohydrates, peptides, organic acids, phenolics, lipids, nucleotides or even hops bitter acids and their corresponding derivatives29. Plotting in the van Krevelen diagram the 350 formulae, which were present in over 95% of the beers spectra, we can recognize that the beer matrix seems, in general, to be defined by carbohydrates and derivatives, peptides, but also the hops bitter acids. In contrast to this, lipids and phenolic compounds were more specific for the single beers or group of beers (Supplementary Fig. 1).

Fig. 1: FTICR-MS spectrum reveals the chemodiversity of a Pilsner beer and biochemical patterns therein.

The full-scale view (a) shows hexose condensation patterns and an excerpt of the nominal mass m/z 391 (b) illustrates the resolved chemodiversity of the beer inside one single nominal mass. Annotated sum formulae and mass errors are given above the mass peaks. Color code of the sum formulae: CHO blue; CHNO orange; CHOS green; CHNOS red. Adduct formation is expressed by +H2PO4 for dihydrogenphosphate and +Cl for chloride, respectively.

Fig. 2: Van Krevelen diagram (H/C vs. O/C) of beer compositions shows their diversity and associated compounds classes.

Annotations, which appear in at least 5% of all beer samples are shown. Areas specific for certain compound classes are marked with dotted lines. Color code: CHNO blue; CHNO orange; CHOS green; CHNOS red; P violet; Cl light violet. The bubble size indicates the mean relative intensities.

By displaying assigned elemental formulae in a mass difference network27 one can exploit the exact mass information provided by FTICR-MS and set the CHO, CHNO, CHNOS, CHOS, and P chemical spaces into relation. Figure 3a shows that the sulfur containing spaces were separated from a highly connected CHO/CHNO sphere. The same holds true for phosphate containing molecules, which were mostly connected to the other spaces by glycerolphosphate, phosphoethanolamine, hexosephosphate, and phosphorylation itself. Mass differences indicating mainly reactions with amino acids were the most dominant inside the CHNO chemical space and between CHO and CHNO spaces (~50%). Condensation of hexose and pentose species are the most abundant sugar-related reactions connecting (oligo)saccharides with their dedicated aglyca. Reactions regarding more specific metabolic pathways like prenylation (terpenoids) could be found besides the condensation of nucleic bases and glycerol. Overall, raw chemical-related reactions (roasting/malting/boiling) were represented on a par with biochemically driven reactions (raw material/fermentation). An extract of the frequencies of individual modifications can be found in Fig. 3b.

Fig. 3: Mass difference network of the beer samples’ annotations and their (bio)chemical connectivity.

Chloride adducts were converted into their dedicated [M–H] ions in silico. Color code compare Fig. 2. The area of hops bitter acid derivatives inside the mass difference network (a) is marked. An excerpt of (bio)chemical reactions with their dedicated mass and sum formula differences and the frequencies they occur in the network is given below (b).

Multivariate analysis

The hierarchical clustering analysis (HCA) showed a general overview of the similarities across the different samples revealed a cluster of typical lager beers samples (Fig. 4). The quality control samples, namely aliquots of one same lager beer, were correctly located in exactly this group and build an own sub cluster, which showed that the fingerprint of this beer is conserved through the different batches. Beers with special grains like roasted malt, oat, or gluten-free grain were grouped together as well as wheat beers and alcohol-free beers. Besides these clusters there was a group mainly but not exclusively consisting of craft beers and special Belgian beers. Some more conventional beers were also allocated inside this group, probably due to the overlap of specific molecular patterns. A detailed inspection of the dendrogram plot revealed two pairs of beer from one brewery (denominated “brewery A” in the following)—namely the brewery A’s lager and wheat beer with their corresponding alcohol-free versions. These pairings reflect the fact that the dealcoholization process in this brewery consists mainly of downdraft evaporation of the original alcohol containing beer. The brewing process itself stays the same, which makes these beers very similar.

Fig. 4: Hierarchical clustering arranges the beer samples’ FTICR mass spectra with regard to their beer type.

Color code of the observed clusters: lager beer blue; beer brewed with special grain red; wheat beer green; craft beer yellow; alcohol-free beer light blue. The cluster of QC lager beer samples is framed. The enlarged excerpt shows the cluster of one brewing site’s alcohol containing and alcohol-free beers. The samples’ order is stated below.

OPLS-DA model 1: beer type

The first OPLS-DA model distinguishes between the different beer types (Fig. 5). Wheat beers were separated from the other beer types in the first component (x-axis). In the orthogonal second component (y-axis) it was possible to differentiate between classical lager beers and craft beers. The fourth class, the traditional Belgian abbey beers, were located in the middle of the score plot, whereas the spontaneous inoculated geuze beers were excluded from the model as outliers. The detailed statistical (Supplementary Table 1), loading plots (Supplementary Fig. 2), and score-plot coordinates (Supplementary Table 2) for each model are given in the supplementary information.

Fig. 5: OPLS-DA model’s score plot for the beer-type observation.

The score plot is surrounded by the different observations’ van Krevelen diagrams (lager beers (I); craft beers (II); rich hopped beer types (III); wheat beers (IV)). Color code and bubble size compare Fig. 2. Samples included in the model calculation are depicted as circles, whereas predicted samples are represented as triangles. Craft and lager beers are summarized as hops rich beer types to reflect the separation of metabolites in the first component.

The first component revealed the most significant molecular pattern separating wheat beers from the lager and craft beers. Both the latter beer types feature a great amount of hops compared with wheat beers and thus can be denominated “hops rich beer types”. The masses with the most negative loadings reflected this characteristic of a strong hops profile. The Van Krevelen diagram of their formulae showed a specific cluster of CHO-molecules in the region of 0.2 < O/C < 0.4 and 1.2 < H/C < 1.6, respectively (Fig. 5i). As mentioned before, this area of the diagram is typical for terpenoids and more accurately hops bitter acids (terpeno-phenolics) in the beer matrix. This pattern was also observed in the mass difference network, showing an agglomerated cluster of hop-rich beer-type markers in a certain area (Fig. 3a). The annotation of the given masses in databases offered exactly those hops bitter acids. Therefore, it is possible to uncover the area of the mass difference network, where the chemistry of the hops bitter acids is located. A number of 58 marker substances for rich hopped beers could be determined as derivatives by their molecular formula, whereas only 20 of them (35%) were found to have equivalent structures in the databases and pertinent literature (Supplementary Table 3). As FTICR-MS is not capable of distinguishing isomers, the [C21H30O5] marker can represent humulone, adhumulone or iso-humulone, but most likely a mixture. Further already known precursor molecules like prenylphlorisobutyrophenone [C15H20O4] and prenylphlor-isovalerophenone [C16H22O4] as well as bitter acid derivatives like cohumulone [C20H28O5], deoxycohumulone [C20H28O4], dihydrohumulone [C21H32O5], or humulinone [C19H26O5] are surrounded by molecular formulae without suitable hits (Fig. 6a). A demethylation reaction of the potential cohumulinone [C20H28O6] leads to the molecule [C19H26O6], whereas a decarboxylation of [C20H30O7] leads to humulone [C21H30O5]. Overall, finding literature equivalents of oxygenated structures like [C19H26O6], which might indicate hydroxyl-, epoxy-, carboxy-, or peroxyderivatives, turned out considerably difficult. Furthermore, reduction/hydration and addition/elimination of water seem to be important reactions inside this excerpt network of marker substances. Pairs of marker molecules within the same nominal mass (e.g., C20H28O6/C21H32O5; C19H26O6/C20H30O5; C20H26O6/C21H30O5) underlined the necessity of high resolving analytical techniques. In the second component, more oxygenated bitter acid species as well as phenolic and polyphenolic compounds and their dedicated glycosides seem to be characteristic for craft beers due to the typical dry hopping process (Fig. 5ii).

Fig. 6: Detailed excerpts of the mass difference networks for selected hops rich beer-type markers and wheat grain markers.

The nodes represent the annotated ions with given sum formulae or molecule names. They are connected by edges representing the sum formula differences for the hops rich beer-type markers (a) and the biochemical reaction for the wheat grain markers (b), respectively. All nodes depicted are considered marker substances. Wheat grain markers are additionally characterized by UPLC-MS2 of wheat beer sample 41 with literature matching retention time order and MS2-spectra showing respective fragmentation and mass difference pattern (c)32,39.

It was possible to confirm the calculated profiles of the beer types by the vicinity of the different types between the model and prediction sample sets (Fig. 5). Only the position of two samples in the score plot defy the cluster. Sample numbers 100 and 119, both brewed in a certified abbey and, therefore, characterized as abbey beer, were located inside the craft beer region. Besides this origin, the actual brewing technique of these beers is described as amber ale and triple ale, both in agreement with craft beer styles including dry hopping. Therefore, not the brewing location itself, but the molecular signature of the brewing process stands in the foreground. A second group of beers, that could not be assigned precisely, were craft beers brewed with wheat and Belgian wit beers made with raw wheat. These beers share the signature of craft beers (ale yeast; preferably strongly hopped) and the signature of wheat beers (wheat grain as ingredient), for which reason they were located between those beer types. The organic wheat beer (sample 109) differed slightly as well. These findings suggested that the compounds with the most positive loadings define the molecular pattern of wheat. For the investigation of specifically the wheat signature a second model was created.

OPLS-DA model 2: grain

The second OPLS-model was created to extract the influence of the ingredient wheat on the beer’s metabolome (Supplementary Fig. 3). All beers brewed with some amount of wheat were defined as wheat containing beers, regardless of their beer type and other brewing parameters. These stood against beers brewed exclusively with barley. Notwithstanding, that the model sample set consisted of beers with a plurality of various characteristics, it was possible to perform the separation based on the grain used without any ambiguous assignments. In addition to the intended separation it could be remarked that the wheat containing craft beers (sample 53, 54, 73), which were brewed with ale yeasts and dry hopped, were separated by the orthogonal information by the second component (y-axis). In the loading plot, several highly significant wheat grain markers, such as [C14H17NO8], [C14H17NO9], and [C15H19NO9], are separated by simple biochemical reactions (e.g., hydroxylation; methylation) and most likely belong to the family of benzoxazinoid hexosides. The intensity distribution of the mentioned markers is given in the supplementary information (Supplementary Fig. 4). These compounds are described to be specific phytoanticipines for wheat30 compared with barley and partially described in wheat beer31. Again, an excerpt of the mass difference network of the wheat marker substances revealed six masses corresponding to benzoxazines, which were already characterized by de Bruijn et al.32, and a plurality of potential derivatives (Fig. 6c). Against this background, the sulfatation reaction of the HMBOA-hexoside to the respective sulfate appeared especially promising. These secondary metabolites and their dedicated derivatives seemed to be a crucial part of the metabolic signature of wheat containing beers.

The prediction model (Supplementary Fig. 3) showed that the typical German wheat beers containing malted wheat were as well recognized as the Belgian wit beers, which contain unmalted wheat. In contrast, the metabolic pattern of the wheat grain in wheat containing craft beers (sample numbers 100, 114, 118) was recognized less strongly. The comparatively low amount of wheat was opposed by the contrary heavy hops signature. For beers brewed with merely wheat starch no wheat signature could be observed. These findings confirmed the applicability of the calculated pattern and advice to identify certain specific marker substances to detect even low amounts of wheat metabolites.

UHPLC-ToF-MS: marker identification

To support the interpretation of the FTICR-MS data and verify the predicted structures, we performed UHPLC-ToF-MS2 measurements on selected samples. The marker substances for a rich hops profile and the wheat metabolome were investigated in depth.

The marker substances of beers with a rich hops profile in the Van Krevelen region of 0.2 < O/C < 0.4 and 1.2 < H/C < 1.6, respectively (Fig. 5i), were proposed as hops bitter acid derivatives. The UHPLC-MS measurements of a hops rich beer revealed mass traces fitting to 46 of the 58 sum formulae (80%) of the mentioned markers (Supplementary Fig. 5). This is a notably high rate because only 35% of the markers were found to have structural equivalents in mentioned databases or cited pertinent literature (Supplementary Table 3). Moreover, the LC-dimension gave a better idea of how complex the structures behind these masses are as up to 21 peaks could be found for one single formula, all being eluted in the chromatogram region, where hops bitter acid derivatives were found (3.5–7.0 min). The 22 detected isomeric compounds for humulinone [C21H30O6] stood in contrast to other formulae like [C19H26O4] (cohulupone), which were represented by only one chromatographic peak (Supplementary Fig. 5). By tandem mass spectrometry we were able to identify twelve hops bitter acid derivatives like cohumulinic acid [C14H20O4], hulupinic acid [C15H20O4], cohulupone [C19H26O4], (ad)humulone[C21H30O5], tricyclocohumol [C20H30O6], or tetracyclohumol [C20H30O6] on level two33 by comparison of fragmentation patterns and intensities with literature data (Supplementary Table 4). Opposing a wheat beer, which does not feature a rich hops profile, shows, that the corresponding mass traces are decisively higher in hops richer craft and lager beers verifying their discriminating character (Supplementary Fig. 5). It is worth noting that more than 100 MS2 spectra did not lead to hits in databases or literature, and therefore are considered level 3 identifications (Supplementary Table 5).

Benzoxazinoidic phytoanticipines of the wheat plant were proposed as specific wheat grain markers in the beer matrix (Fig. 6c). Again, the marker formulae of the FTICR-MS models were transferred into a preference list to selectively acquire tandem mass spectrometric spectra. By comparison with literature known MS2 fragmentation, eight HBOA-derivatives could be identified in wheat beer (level 2) (Supplementary Table 6). The retention time sequence of the HBOA-, DHBOA-, DIBOA-, and HMBOA-hexoside coincides with the one described by de Bruijn et al.32, whereas the predicted HMBOA-hexosesulfate was eluted earlier than the corresponding hexoside due to the polar sulfate group. The MS2-spectra of the monohexosides are compared in Fig. 6c. The cleavage of the hexose group from the HBOA-hexoside (1) [M–H]-ion [C14H16NO8] results in an m/z value of 164.0348 [C8H6NO3]. The additional hydroxygroup of the DIBOA-hexoside (2) leads to the 180.0299 m/z ion [C8H6NO4]. Replacing the hydroxygroup by a methoxygroup, the m/z value of 194.0455 [C9H8NO4] can be found for the HMBOA-hexoside (3). The same pattern holds true for the 136.0399 [C7H6NO2], 118.0283 [C7H4NO], and 108.0438 [C6H6NO] fragment ions of the HBOA-hexoside. It was not possible to extract complex fragmentation pattern of the HMBOA-hexosesulfate (4) as it was a minor component with a peak intensity about 30 times lower than the respective hexoside. However, the loss of the sulfate group from the quasi-molecular ion 436.0554 [C15H18NO12S] to the dedicated HMBOA-hexoside (3) [M–H]-ion 356.0993 [C15H18NO9] could be observed. Hereupon both compounds share the loss of the hexose sugar. The dihexoside DHBOA-, DIBOA-, and HMBOA-equivalents showed several closely eluting isomeric peaks and were detected with lower retention times as they are more polar. All the substantiated compounds were only observed in wheat beer and none of them is present in beer exclusively brewed with barley, which confirms the assumption that benzoxazinoidic phytoanticipines are suitable specific compounds for the use of wheat grain. To our knowledge, the existence of a HMBOA-hexosesulfate has not been described before. However, for definite identification the synthesis of a corresponding standard would be needed.


Source link

One thought on “Decomposing the molecular complexity of brewing”

Leave a Reply

Your email address will not be published. Required fields are marked *