香京julia种子在线播放

A total of 400 samples of commercially available beers from over 40 different countries were analyzed. The sample set is a cross-section representing all possible combinations of beer styles, fermentation types, raw materials, color impressions and alcohol contents available. Thus, metadata co-varying with the characteristic in focus could be excluded. The samples were purchased at local grocery stores between 2018 and 2020 and stored at −20°C prior preparation for analyses. High-resolution mass spectra were acquired on a Bruker solariX Ion Cyclotron Resonance Fourier Transform Mass Spectrometer (Bruker Daltonics GmbH, Bremen, Germany) equipped with a 12 T superconducting magnet (Magnex Scientific Inc., Yarton, GB) and a APOLO II ESI source (BrukerDaltonics GmbH, Bremen, Germany) operated in negative ionization mode. The sample preparation, measurement and data processing parameters were chosen as reported recently (Pieczonka et al., 2020; Pieczonka SA. et al., 2021). An average mass error of < ±0.15 ppm was reached within and between measurement batches. The resulting 7,700 unambiguous molecular formulae in the CHNOSPCl chemical space that occur in at least five samples were kept for further statistical analysis. An overview of the sample set is given in the Supplementary information (Supplementary Table 1).

Solid phase extraction (SPE) was applied to a sub-sample set including 100 beers. The SPE-parameters are given in Supplementary Table 2. The eluate was evaporated to dryness (25 °C, 1 mbar, 3 h, Christ Martin™ RVC 2-25 CD vacuum concentrator) dissolved in the starting conditions of the UPLC-gradient, vortexed and centrifuged (4 min at 14,000 rmp). The supernatant, five times concentrated compared to the initial beer sample, was used for UPLC-ToF-MS ESI-negative analysis on a Shimadzu LCMS-9030 Q-ToF-System (Shimadzu Deutschland GmbH, Duisburg, Germany) in randomized order. The parameters of the chromatography and ToF-measurements are given in Supplementary Table 3. A pooled QC consisting of all measured samples was used for system conditioning and measured after every 10th injection. On this basis, the batch was normalized (compensation for intensity fluctuations) by the LOWESS algorithm. A class QC, including all samples with the same carbohydrate source, was used for each of the barley, wheat, rice and corn classes. Features that occur in at least 33% of all samples belonging to the respective class were kept as potential maker features for statistical analysis. The data processing and extraction of chromatographic features was carried out with the open source MS-DIAL software (Tsugawa et al., 2015) after the export of the raw data to the centroided mzML-format within the LabSolutions™ 5.99 SP2 software (Shimadzu Corp., Kyoto, Japan). The data treatment parameters were optimized and are given in Supplementary Table 4.

To validate the origin of the statistically most significant features, 10 g of respective foodstuff (corn grits, corn flour, corn starch, corn oil, wheat grits, wheat flour, whole-wheat flour, wheat starch, rice grits, rice flour, rice starch) including typical grain adjuncts in the brewing industry, was extracted with 40 ml MeOH for 1 h on the shaker (250 min⁻¹). The suspensions were centrifuged (5 min, 14.000 rmp) and the supernatant was evaporated to dryness (25 °C, 1 mbar, 8 h). The residue was resolved in 2 ml starting conditions by vortexing and supersonification and syringe filtrated (0.2 µm) before UPLC-ToF-MS analysis. Furthermore, potential marker substances were measured in positive ESI mode to obtain another complementary fragmentation spectrum.

The aspartic acid conjugate of (6,6-d ₂ ) N-β-D-glucopyranosyl-indole-3-acetic acid was synthesized as described by Kai et al. (2007) and kindly provided by the latter authors. The standard was resolved in methanol (6 μg ml⁻¹) and added to a worked-up beer sample (sample 325) in equal volumes for co-chromatography.

We performed a supervised OPLS-DA analysis on both the FTICR and UPLC-ToF dataset-matrices consisting of metabolite features and intensities. Data-pretreatment included zero-filling, data normalization, scaling and transformation (Supplementary Table 4). The Hotelling’s T² test (95%) was applied to prohibit the influence of strong outliers on the models. The lists of the most important masses were defined choosing the highest loadings values. The top characteristic masses were selected within the 95th percentile (385 masses for each carbohydrate source for FTICR and 89 for UPLC-ToF respectively). Potential marker for the use of barley were neglected due to co-varying metadata (Supplementary Figure 1). The goodness of the fit and of the prediction were evaluated with the R² and Q² values. To exclude overfitting, we provide the p-value of the Cross-Validation Analysis of Variance (CV-ANOVA). With high values for the quality of prediction (Q²) that do not exceed those of the goodness of the fit (R²Y) and CV-ANOVA p-values far lower than 0.05 for the comparison of between-class against within-class variance, the significance of the models could be confirmed and overfitting excluded (Golbraikh and Tropsha, 2002; Westerhuis et al., 2008). Those elaborations were done in SIMCA 13.0.3.0 (Umetrics, Umeå, Sweden). The statistical parameters of the beer samples (Supplementary Table 1) and OPLS models (Supplementary Table 5) can be found in the Supplementary information. Eight samples with inadequate measurement quality were not integrated into the FTICR-MS statistical model. Three samples were excluded from the models because of information on the ingredient list contrary to their positions in the score plots (FTICR and UPLC-MS). Predicted score values were calculated.

The FTICR-MS marker formulae were depicted in van Krevelen diagrams for each starch source. By plotting H/C versus O/C atomic ratios it is possible to depict common compositional patterns within observations’ markers (Hertkorn et al., 2008; Gougeon et al., 2009; Schmitt-Kopplin et al., 2019). A mass difference network (MDiN) was applied utilizing the NetCalc approach (Tziotis et al., 2011). The nodes, representing the annotated sum formulae, were connected by edges which represent compositional changes corresponding to 250 different (bio)chemical reactions.

The UPLC-ToF marker features were subjected to the open source Cyctoscape software environment (Shannon et al., 2003) to visualize a MS²-similarity-network based on similar fragments and neutral losses. The similarity cutoff was set to 0.65. Database search for matching fragmentation spectra was performed using the MS-FINDER (Lai et al., 2018) and MetFrag (Ruttkies et al., 2016) software tools. The entries in the HMDB, FooDB, ChEBI, LipidBlast, LipidMaps, KNApSAcK and PubChem databases were used to carry out a comparison of respective in silico fragmentation with our experimental data. The best five hits were examined for their plausibility. When possible, the hits were confirmed through experimental spectra of primary literature. The levels of identification were assigned as suggested by the Metabolomics Standards Initiative (Sumner et al., 2007).

The FTICR-MS and UPLC-ToF-MS data were compared with a mass tolerance of ±5 ppm. Isomeric compounds were merged with the same error tolerance for the UPLC-ToF-MS features. The overlaps were illustrated using pie charts.

In the first analytical step, we investigated the metabolome profile of a total of 400 bottled beer samples by direct-infusion Fourier transform ion cyclotron mass spectrometry (DI-FT-ICR-MS) using electrospray (-) ionization. The commercial beer samples covered the numerous facets of beer brewing and included beers manufactured in over 40 countries around the world. By that, we could exclude most co-variating metadata. Despite the abundance of different and combined brewing styles, the craft beer style including the step of dry hopping was found to co-variate with beers brewed with barley only (Supplementary Figure 1).

The non-targeted and holistic approach, renouncing discriminatory sample processing and chromatography, is capable to resolve the entire molecular complexity of beer within a quick (10 min) measurement. About 7.700 unambiguous molecular compositions could be assigned to exact monoisotopic masses spanning the mass range of m/z 100–1,000 (Figure 1A) within the sample batch. They reach from polar sugars, phosphates and sulfates over diverse secondary metabolites and peptides to non-polar lipids, hops bitter acids and highly unsaturated polyphenols and Maillard reaction end products (Pieczonka et al., 2020; Pieczonka SA. et al., 2021). We were able to resolve up to 40 monoisotopic mass features within one single nominal mass including several significant compositions regarding carbohydrate sources, as will be seen later (Supplementary Table 6). The molecular formulae were annotated in the CHNOSP chemical space and subjected to further statistical analysis.

We applied supervised orthogonal partial least-squares discriminant analysis (OPLS-DA) to the metabolite data resolved by DI-FTICR-MS, using the carbohydrate source as Y-variable. The classification power of the model was highly significant (Supplementary Table 5). The Q² value for quality of prevision (>0.6) and the R2Y value for the goodness of the fit (>0.85) prove the statistical relevance whereas overfitting was excluded by the p-value calculated after the CV-ANOVA (<<0.05) (Golbraikh and Tropsha, 2002; Westerhuis et al., 2008). The associated score plot (Figure 2A) showed a clear differentiation of barley, wheat, corn and rice beers. In the first principal component, beers brewed with barley only are separated from beers with an additional carbohydrate source. The second component separates beers brewed with wheat from that brewed with corn or rice (Figure 2A-I). Ultimately, the third component differentiates rice and corn beers (Figure 2A-II). Accordingly, a statistical model was achieved that uncovered the influence of all considered carbohydrate sources on the metabolic signature of beer. Metabolite features that drive the separation were extracted from the respective loadings plot (Supplementary Figure 2). Compositions causing the agglomeration of corn and rice beers in the first and second component are referred to as “corn and rice” features in the following.

The marker metabolites for corn described by Fenz (1991) could be confirmed with significant mass values equal to p-coumaroyl glycerol (C₁₂H₁₄O₅), caffeoyl glycerol (C₁₂H₁₄O₆) and hydroxyoxindoleacetic acid (C₁₀H₉NO₄). In addition to individual masses, the van Krevelen diagram of the respective compositions revealed characteristic patterns for the carbohydrates sources. Beers brewed with wheat featured a multitude of very polar phosphates (Figure 1B) and beers brewed with corn showed a specific pattern of lipids (Figure 1C). Many compositions characteristic for rice beers are located in the area where peptides are expected (Figure 1D).

Additionally, compositional mass difference networks (MDiN) have proven to be powerful tools to set significant compositions in relation. It can utilize the exact mass information FTICR-MS provides, where compositions are represented as nodes that are connected by edges representing distinct mass differences that describe (bio)chemical relations. Such a MDiN sets in relation the lipid pattern found specific to corn by (de)hydrogenation, hydroxylation, water or glycerol addition and chain elongation reactions (Supplementary Figure 3). Several derivatives of the lipid with the mass m/z 335.22278 (C₂₀H₃₂O₄) could be explained by e.g., hydrogenation (C₂₀H₃₄O₄) and hydroxylation (C₂₀H₃₂O₅) reactions. Accordingly, the characteristic composition (C₂₁H₃₄O₄) is connected to (C₂₁H₃₆O₄) and (C₂₁H₃₄O₅) by hydrogenation and hydroxylation respectively.

A second MDiN excerpt that sets wheat, corn and rice markers in biochemical relation was investigated in more detail (Figure 3). As reported before (Pieczonka et al., 2020), the metabolome wheat adds to the beer’s complexity specifically is characterized by compositions corresponding to benzoxazinone derivatives. The mass corresponding to possible blepharin (C₁₄H₁₇NO₈), a plant phytoanticipine, is connected to the related HMBOA-glucoside (C₁₅H₁₉NO₉) by methoxylation. A subsequent sulfatation gives (C₁₅H₂₁NO₁₂S). An equivalent pattern links the hydroxylated DHBOA-glc (C₁₄H₁₇NO₉), the DIMBOA-glc (C₁₅H₁₉NO₁₀) and the respective sulfate (C₁₅H₂₁NO₁₃S). Besides methoxylation, hydroxylation, sulfatation and water addition, several glycation reactions of described molecular formulae lead to a complex network of known and unknown compounds specific for wheat. Those reach from rather unsaturated compositions (e.g., C₁₀H₉NO₃ and C₁₁H₁₁NO₃) to very polar glucosides of the potential aglyca (e.g., C₂₃H₃₁NO₁₃, C₂₆H₃₉NO₂₀ and C₂₆H₃₇NO₁₉). A similar, but smaller, network is being build up for corn based on compounds likely arising from the indoleacetic acid (IAA) biosynthetic pathway. Chloride adduct formation of the respective aglycon hydroxyoxindoleacetic acid (C₁₀H₉NO₄) reinforces the presence of a carboxylic acid group of these compounds on the molecular structure level. Based on this annotation, known to be specific for the use of corn (Fenz, 1991), two glycation reactions lead to the respective derivatives (C₁₆H₁₉NO₉) and (C₂₂H₂₉NO₁₄). Again, several compositional changes equivalent to hydroxylation, hydrogenation, methoxylation or water addition form a network of masses specific to corn. In parallel, there is a similarly structured network starting from (C₁₀H₉NO₅) for rice. The composition could potentially be annotated to hydroxydioxindoleacetic acid, found in rice bran by Kinashi et al. (1976). The described biochemical relations again lead to compositions specific for rice (e.g., C₁₆H₂₁NO₁₁, C₂₂H₃₁NO₁₆, C₁₆H₂₃NO₁₁, C₂₂H₃₃NO₁₆), but also includes compositions characteristic for corn and rice (e.g., C₁₀H₁₃NO₇ and C₁₆H₁₉NO₁₀). Overall, secondary metabolites deriving from tryptophan dependent pathways drive the differentiation of the carbohydrate sources wheat, corn and rice for brewing. The metabolites cover a wide range of polarity, all accessible from direct infusion with FTICR-mass spectrometry. However, metabolites could only be annotated by exact masses and their biochemical relations (expressed as mass differences) in combination with database and literature data. For definite structural confirmation, a chromatography-coupled mass spectrometric approach including ion fragmentation is necessary (UPLC-ToF-MS).

A representative sub-sample set (100 samples) was treated by solid phase extraction (SPE) and subjected to reversed phase UPLC-ToF-MS. An average of 680 chromatographic features per sample were obtained after applying filter criteria. The peaks shared by at least one third of all beer samples within a carbohydrate source class (1750 peaks) were used for statistical analysis (OPLS-DA). The classification power of the model was highly significant (Supplementary Table 5). The Q² value for quality of prevision (>0.6) and the R²Y value for the goodness of the fit (>0.85) prove the statistical relevance whereas overfitting was excluded by CV-ANOVA (p-value <<0.05) (Golbraikh and Tropsha, 2002; Westerhuis et al., 2008). As for the FTICR-MS data, the associated score plot (Figure 2B) showed a clear differentiation of barley, wheat, corn and rice beers. In the first two principal components, beers brewed with barley only are separated from wheat beers. Again, beers brewed with corn or rice are agglomerated against the two others (Figure 2B-I).

The third component enables the differentiation of corn and rice beers (Figure 2B-II). Ultimately, a statistical model distinguishing all carbohydrate sources could also be achieved by the isomeric resolved UPLC-ToF-MS data of the pretreated sub-sample set. The metabolite features that drive the separation were extracted from the respective loadings plot accordingly (Supplementary Figure 2).

The available MS²-spectra of the potential marker compounds were utilized in a mass spectral similarity network (Figure 4). The fragmentation spectra of compounds that were examined in more detail can be found in the Supplementary information (Supplementary Table 7). Two clusters with similar fragmentation patterns could be observed for the wheat-specific compounds. The first cluster (Figure 4A) could be identified as an agglomeration of benzoxazinone derivatives, known to be phytoanticipines in the wheat plant and validating the findings with DI-FTICR-MS data. By database and literature research (Bonnington et al., 2003; de Bruijn et al., 2016; Pieczonka et al., 2020) four compounds could be identified as MOBA, HBOA-glucoside, DIBOA-glucoside and HMBOA-glucoside [identification level 2 (Sumner et al., 2007)]. For the second cluster (Figure 4B), only in silico fragmentation spectra of database entries were available. All matching spectra found for the five peaks indicate the potential compound class of N-acyl-glutamines (identification level 3). Their almost identical retention behavior (4.32–4.62 min) and proposed molecular formulae [(C₂₃H₃₈N₂O₅), (C₂₃H₄₀N₂O₅), (C₂₅H₄₀N₂O₄), (C₂₃H₃₈N₂O₆) and (C₂₃H₄₀N₂O₆)] support their close chemical relation.

For corn markers, two clusters of compounds with related fragmentation spectra and thus similar structure and origin could be observed. The cluster with the greater significance (Figure 4C) was studied in more detail. The nine compounds examined were found to be isomeric pairs of m/z-values matching the molecular formulae (C₂₀H₃₄O₄), (C₂₀H₃₂O₅), (C₂₁H₃₆O₄) and (C₂₁H₃₄O₅). The close retention time window between 6.0 and 7.1 min supports their close chemical relation. Together with the similar molecular composition and fragmentation spectra, it brings us to suggest a shared compound class of a non-polar character. The best hits with regard to in silico fragmentation spectra all agree on lipid-type structures for the mentioned compounds. Besides, the fragmentation pattern of (C₂₀H₃₄O₄) and (C₂₀H₃₂O₅) compounds show a great similarity to DiHEtrE and TriHETE fragmentations spectra respectively, when compared to literature data (Wheelan et al., 1996; Ferreiros et al., 2014). However, based on our data, the exact molecular structure and in particular the position of possible hydroxylation cannot be determined. Accordingly, the identification level of this group of corn-specific compounds was indicated to level 3, as suggested by Sumner et al. (2007). In addition to this not yet described cluster of lipid-type molecules, we were able to confirm the hydroxyoxindol-acetic acid as a marker substance for the use of corn by comparison of both in silico and literature fragmentation data (Fenz, 1991) (identification level 2).

The rice specific compounds, few of which were found highly significant in the loadings plot already, did not cluster with regard to their fragmentation pattern. Two of those could be characterized by their molecular formula (C₂₄H₄₀N₆O₈) and (C₂₂H₃₅N₅O₁₁) as being in accordance with mass values found to be specific in FTICR-MS. A second pair of highly significant peaks could be described as potential Glu-Trp-Leu/Ile-Pro (C₂₇H₃₇N₅O₇) and a cyclic Asp-Ser-Val-Leu-Trp peptide (C₂₉H₄₀N₆O₈), respectively, by comparison of in silico fragmentation data (identification level 3). The fifth potential marker of highest interest could be assigned based on both matching fragmentation patterns and co-chromatography (identification level 1). Kai et al. (2007) reported an aspartic acid-conjugate of N-β-D-glucopyranosyl-indole-3-acetic acid to be found in rice with a matching fragmentation pattern. The respective d₂-standard, synthesized and provided by the mentioned authors, was used for co-chromatography. The rice secondary metabolite was identified by matching retention time and fragmentation pattern (Figure 5). We were able to detect the corresponding peak in the vast majority of rice beers and two beers brewed with corn. To confirm our findings and the origin of the potential marker compounds, we analyzed methanol extracts of food made from the appropriate grain raw materials. Grits, starch and flour of wheat, corn and rice and corn oil were screened for the presence of the specific respective compounds (Supplementary Table 8). Wheat benzoxazinones and potential acyl-glutamines were present in the wheat products. The exception is pure wheat starch, in which none of the compounds were found, as in beers brewed with merely wheat starch (Pieczonka et al., 2020). We were able to confirm the hydroxyoxindol-acetic acid in all corn products except for the oil. The isomeric pairs of lipid class compounds (C₂₀H₃₄O₄) and (C₂₁H₃₆O₄) were found in the corn oil, whereas the other specific oxygenated lipids might be formed during the brewing process. The same is suspected for the (C₂₉H₄₀N₆O₈) cyclic peptide in rice. The other rice metabolites, including the aspartic acid-conjugate of N-β-D-glucopyranosyl-indole-3-acetic acid, were confirmed by the analysis of rice products. An overlap of the potential markers between the carbohydrate groups was not observed. Interestingly, the coumaryl and caffeoyl glycerols described by Fenz et al. (1992) and confirmed by FTICR-MS were found and identified in the methanol extract of corn grits but not the beers. We assume that they were lost through the SPE sample processing of the beers.

The investigation of the influence of different carbohydrate sources on the metabolome of the beer end product was carried out using two different, complementary mass spectrometric methods. Even with different sample numbers (400 for DI-FTCR-MS and 100 for UPLC-ToF-MS), fundamental differences and commonalities between the analytical approaches could be observed. Based on the direct infusion approach without extensive prior sample preparation, compounds of all polarities (ionizable by ESI) are accessible with FTICR-MS. This is reflected in the MDiN of the secondary metabolites, which maps numerous glycosylation steps up to highly oxygenated compositions. This wide polarity range was not tangible by RP-HPLC-ToF-MS. The corn marker hydroxyoxindol-acetic acid was found with an early retention time (3.02 min), whereas glycosylated derivatives of the aglycone were lost by sample preparation and chromatography. The phosphate-structures found in FTICR-MS could not be found either and thus not be further characterized by fragmentation spectra. The average mass values also differ between FTICR-MS (m/z 409) compared to ToF-features (m/z 553), which could be attributed to different accessible mass ranges (100 to 1,000 Da for FTICR-MS, 50 to 1,500 Da for ToF-MS). The mass features showed a moderate overlap within a ±5 ppm range, in accordance with the different and complementary chemical spaces analyzed (Supplementary Figure 4). About 35% of the chromatographic features showed had a corresponding m/z-value in FTICR-MS, whereas less than 10% of FTICR-MS-masses were found with an equivalent peak in LC-MS. Here, the different numbers of samples should be emphasized again. The majority of the chromatographic peaks showed at least one isomeric compound (up to 16), which confirms the complementarity of the information obtained by coupling chromatography to mass spectrometry. We observed a similarly low m/z-overlap with regard to the potential marker features. In particular, it is the statistically most significant compounds, which were detected in large parts in both DI-FTICR-MS and UPLC-ToF-MS. This enabled a deeper characterization through exact mass values and fragmentation mass spectra. Only the group of potential acyl-glutamines and some rice peptide-like structures are not represent in the FTICR-MS-data.

Overall, with two complementary mass spectrometric approaches, we have uncovered deep metabolic signatures that the use of wheat, corn and rice in brewing entails. The majority of the decisive compounds can already be found in the corresponding raw materials and survive the entire brewing process. We were able to set the compositions in relation by mass difference network analysis and uncovered a whole network of secondary metabolites specific to the respective grains. By mass fragmentation, the compounds could be characterized in detail and known reported marker substances could be confirmed. Finally, we want to highlight that in the aspartic acid conjugate of N-β-D-glucopyranosyl-indol-3-acetic acid we report a potential marker for the use of rice in beer.