Comprehensive Two-Dimensional Gas Chromatography as a Powerful Strategy for the Exploration of Broas Volatile Composition

Broa is a Portuguese maize bread with characteristic sensory attributes that can only be achieved using traditional maize varieties. This study intends to disclose the volatile compounds that are mainly associated with the baking process of broas, which can be important contributors to their aroma. Twelve broas were prepared from twelve maize flours (eleven traditional maize varieties and one commercial hybrid). Their volatile compounds were analyzed by GC×GC–ToFMS (two-dimensional gas chromatography coupled with time-of-flight mass spectrometry) for an untargeted screening of the chemical compounds mainly formed during baking. It was possible to identify 128 volatiles that belonged to the main chemical families formed during this stage. Among these, only 16 had been previously detected in broas. The most abundant were furans, furanones, and pyranones, but the most relevant for the aroma of broas were ascribed to sulfur-containing compounds, in particular dimethyl trisulfide and methanethiol. Pyrazines might contribute negatively to the aroma of broas since they were present in higher amounts in the commercial broa. This work constitutes the most detailed study of the characterization of broas volatile compounds, particularly those formed during the Maillard reaction. These findings may contribute to the characterization of other maize-based foodstuffs, ultimately improving the production of foods with better sensory features.


Introduction
Bread is considered one of the most important foodstuffs worldwide [1]. In Portugal, whole maize flour is used to produce a traditional bread known as broa [2], considered one of the 50 world's best breads by CNN Travel in 2019 [3]. Over the last centuries, several traditional open-pollinated maize varieties have been developed to produce high-quality broas with improved flavors and aromas that are not possible to achieve using the presently available more productive commercial maize hybrids [2,4].
The aroma of bread plays a key role in its acceptance by consumers [1,5,6]. The nature of aroma is very complex [7,8], and the contribution of an individual compound to the varieties and contribute to a better knowledge of the Maillard reaction, which is essential to design foods that present sensory attributes demanded by the consumers [15,31].

Results and Discussion
Eleven broas (B1 to B11) prepared from eleven traditional maize varieties and one broa (B12) prepared from commercial hybrid maize flour were studied (Table S1 and Figure  S2) [27]. A contour plot of the total ion current chromatogram of a broa sample (B1) under study is presented in Figure 2. In a first approach, the presence of volatiles belonging to the chemical families mostly associated with baking was identified in the chromatograms of all studied samples (n = 12). A total number of 128 volatile compounds belonging to these families were detected and are described in Table 1. More than 80% of the compounds have been identified according to the criteria described in Section 3.3.
The peak areas corresponding to the studied volatiles were measured, and their average in the twelve studied broas is presented in Table 1. The ratio of the peak area to odor threshold values in water (OT, in the Log10 form) was used as a screening method to identify the most relevant volatiles to the aroma of broas. Calculation of the odor activity value (OAV, ratio of the concentration to OT) is often carried out to evaluate the most aroma-active compounds in a food product. A compound might be sensed when OAV > 1 (or > 0, when the Log10 form is considered) [18,30,32,33]. Although differences in peak areas among different compounds do not give direct information about their relative concentration, they greatly varied (sometimes more than 10.000-fold), suggesting they were present in very distinct concentrations. Thus, the higher the ratio, the higher the probability of a compound contributing to the aroma of this bread. The table also includes some examples of foods where the compounds have been detected (visually represented in Figure S1 from Supplementary Materials).

Results and Discussion
Eleven broas (B1 to B11) prepared from eleven traditional maize varieties and one broa (B12) prepared from commercial hybrid maize flour were studied (Table S1 and Figure S2) [27]. A contour plot of the total ion current chromatogram of a broa sample (B1) under study is presented in Figure 2. In a first approach, the presence of volatiles belonging to the chemical families mostly associated with baking was identified in the chromatograms of all studied samples (n = 12). A total number of 128 volatile compounds belonging to these families were detected and are described in Table 1. More than 80% of the compounds have been identified according to the criteria described in Section 3.3. The results will be presented as follows: firstly, the advantages of the analysis of broas volatile compounds by GC×GC-ToFMS will be discussed by providing examples that allow the identification of unreported compounds. Secondly, the compounds will be explored according to their chemical families, and those which may be more relevant to the aroma of broas will be suggested. Finally, the results from the analyzed samples will be compared, and the correlations among their volatiles will be studied to elucidate which  n/a n/a n/a n/a n/a n/a n/a n/i   The peak areas corresponding to the studied volatiles were measured, and their average in the twelve studied broas is presented in Table 1. The ratio of the peak area to odor threshold values in water (OT, in the Log 10 form) was used as a screening method to identify the most relevant volatiles to the aroma of broas. Calculation of the odor activity value (OAV, ratio of the concentration to OT) is often carried out to evaluate the most aroma-active compounds in a food product. A compound might be sensed when OAV > 1 (or >0, when the Log 10 form is considered) [18,30,32,33]. Although differences in peak areas among different compounds do not give direct information about their relative concentration, they greatly varied (sometimes more than 10.000-fold), suggesting they were present in very distinct concentrations. Thus, the higher the ratio, the higher the probability of a compound contributing to the aroma of this bread. The table also includes some examples of foods where the compounds have been detected (visually represented in Figure S1 from Supplementary Materials).
The results will be presented as follows: firstly, the advantages of the analysis of broas volatile compounds by GC×GC-ToFMS will be discussed by providing examples that allow the identification of unreported compounds. Secondly, the compounds will be explored according to their chemical families, and those which may be more relevant to the aroma of broas will be suggested. Finally, the results from the analyzed samples will be compared, and the correlations among their volatiles will be studied to elucidate which compounds may be responsible for the differences in their aroma.

The potentialities of GC×GC-ToFMS in the Identification of Broas Volatile Compounds
In a previous study on broas volatile composition by GC-MS (1D) [27], only 16 out of the 128 volatiles described in the present work were detected (Table 1), which confirms that GC×GC-ToFMS provides a much higher potential for the detection and identification of food volatiles [63,64]. It should also be noted that, in this work, a relatively high signal-tonoise (S/N) threshold (100) was used, which limited the number of detected compounds. Figures 2 and 3 illustrate the advantages of GC×GC-ToFMS, which allowed the separation of analytes with similar volatility and common product ions through the secondary 2 D column. Considering the set of columns used (non-polar/polar), the decrease in volatility (high 1 t R ) is mainly related to the increase in the number of carbons [64,65]. In contrast, the increase in the 2 t R corresponds to increasing polarity [64,65]. For instance, 2-butyltetrahydrofuran (F31) was not detected in the previous study [27]. This compound was co-eluted with 2-nonanone ( 1 t R : 715 s), and both presented similar mass spectra, with identical product ions at m/z 71 and 43 ( Figure 3). However, since they present different polarities, their separation was achieved through the 2 D column. 2-Butyl-tetrahydrofuran (F31) exhibits a slightly lower Log P than 2-nonanone (Log P: 2.93 and 3.08, respectively), which supports its relatively high polarity, and, therefore, the higher retention time in the 2 D ( 2 t R: 0.77 vs. 0.56 s). Thus, despite the high similarity index of 2-nonanone with the library spectra (89%) obtained in the previous analysis by GC-MS [27], the compound 2-butyltetrahydrofuran also contributed to the peak area considered for 2-nonanone. Without the high resolving power of GC×GC, the separation, identification, and relative quantitation of these two compounds would have been a challenge.
Another example is the identification of methylpentylfuran (F27). This compound was not commonly detected in foods but was previously tentatively identified in broas by GC-MS [27]. The results obtained in the present work corroborate this earlier identification. The mass spectrum of F27 showed a high similarity index with the spectra from methylpentylfuran from the Wiley database (Table 1). Furthermore, this compound showed a very low retention time of 0.50 s in the 2 D column (Figure 4), suggesting its low polarity, which was confirmed by its calculated Log P of 4.43. Although methylpentylfuran usually refers to 3-methyl-2-pentylfuran, F27 can also correspond to other isomers, such as 2-methyl-5-pentylfuran or 4-methyl-2-pentylfuran. In addition, several other volatiles belonging to chemical families, which usually arise during baking, were eluted at this 1 t R (705 s), namely furfurylfuran (F28), 2-formyl-3-methylthiophene (Tp4), 2-acetylthiophene (Tp5) and furaneol (Fo20). Since they show different polarities (Log P of 2.52, 1.38, 1.28 and -0.33, respectively), they were separated through the 2 D column ( 2 t R : 0.95, 1.31, 1.56 and 4.02, respectively). As shown in Figure 4A, the peak intensities of methylpentylfuran (F27) and 3-nonanone were much higher than F28, Tp4, Tp5 and Fo20. Thus, it would not have been possible to identify these compounds without the orthogonal separation through the 2 D column [64,65]. The identification of volatile sulfur compounds ( Figure 4B) is particularly relevant since they are typically present in foods at extremely low levels, often at sub-parts-per-billion concentrations, but provide background sensory nuances to the flavor and are often considered "character-impact compounds" [13].
pounds. Figures 2 and 3 illustrate the advantages of GC×GC-ToFMS, which allowed the separation of analytes with similar volatility and common product ions through the secondary 2 D column. Considering the set of columns used (non-polar/polar), the decrease in volatility (high 1 tR) is mainly related to the increase in the number of carbons [64,65]. In contrast, the increase in the 2 tR corresponds to increasing polarity [64,65]. For instance, 2butyl-tetrahydrofuran (F31) was not detected in the previous study [27]. This compound was co-eluted with 2-nonanone ( 1 tR: 715 s), and both presented similar mass spectra, with identical product ions at m/z 71 and 43 ( Figure 3). However, since they present different polarities, their separation was achieved through the 2 D column. 2-Butyl-tetrahydrofuran (F31) exhibits a slightly lower Log P than 2-nonanone (Log P: 2.93 and 3.08, respectively), which supports its relatively high polarity, and, therefore, the higher retention time in the 2 D ( 2 tR: 0.77 vs. 0.56 s). Thus, despite the high similarity index of 2-nonanone with the library spectra (89%) obtained in the previous analysis by GC-MS [27], the compound 2butyl-tetrahydrofuran also contributed to the peak area considered for 2-nonanone. Without the high resolving power of GC×GC, the separation, identification, and relative quantitation of these two compounds would have been a challenge. Another example is the identification of methylpentylfuran (F27). This compound was not commonly detected in foods but was previously tentatively identified in broas by GC-MS [27]. The results obtained in the present work corroborate this earlier identification. The mass spectrum of F27 showed a high similarity index with the spectra from methylpentylfuran from the Wiley database (Table 1). Furthermore, this compound showed a very low retention time of 0.50 s in the 2 D column (Figure 4), suggesting its low polarity, which was confirmed by its calculated Log P of 4.43. Although methylpentylfuran usually refers to 3-methyl-2-pentylfuran, F27 can also correspond to other isomers, such as 2-methyl-5-pentylfuran or 4-methyl-2-pentylfuran. In addition, several other volatiles belonging to chemical families, which usually arise during baking, were eluted at  Figure 4A, the peak intensities of methylpentylfuran (F27) and 3-nonanone were much higher than F28, Tp4, Tp5 and Fo20. Thus, it would not have been possible to identify these compounds without the orthogonal separation through the 2 D column [64,65]. The identification of volatile sulfur compounds (Figure 4B) is particularly relevant since they are typically present in foods at extremely low levels, often at sub-parts-per-billion concentrations, but provide background sensory nuances to the flavor and are often considered "character-impact compounds" [13]. These results show the potentialities of GC×GC-ToFMS for the characterization of food volatiles and demonstrate the complexity of the volatile composition of broas. Moreover, this technique also allowed the identification of several compounds belonging to the characteristic chemical classes of the baking process, which have not been commonly described either in bread or maize-based foods. These results show the potentialities of GC×GC-ToFMS for the characterization of food volatiles and demonstrate the complexity of the volatile composition of broas. More-over, this technique also allowed the identification of several compounds belonging to the characteristic chemical classes of the baking process, which have not been commonly described either in bread or maize-based foods.

Exploring the Volatile Compounds Associated with Baking
The most relevant compounds of broas, selected based on their abundance and, more importantly, their OTs, will be discussed below.

Furans and Furanones
Furan and furanone derivatives are among the most common volatiles of heated foods [20,66], and give burnt, savory, sweet and caramel aroma to foods [9,15,20]. The major furans included 2-pentylfuran (F20) and 2-butylfuran (F14), which have been previously detected in broas [27]. 2-Pentylfuran (F20) was particularly abundant, and it is a potent odorant [48], giving floral-fruity notes [60], which may significantly contribute to their aroma. It has been reported as the most common aroma-active furan in wheat bread crumb [60] and a likely contributor to the total aroma and flavor of maize tortilla chips [43] and popcorn [33].
Several furans with oxygenated substituents, such as furfurals and furanones, were also identified in broas. Among all the compounds described in the present work, the highest peak areas were obtained for 2-furanmethanol (F13). Although it is not a very strong odorant [32,48], this compound may also be relevant for the aroma of broas due to its abundance. It is usually associated with pleasant, creamy and caramel notes [41] and may be relevant to the aroma of popcorn [33]. Other oxygenated furans include 5-methyl-2furanmethanol (F19), 3-furfural (F10), and isomaltol (F22), which had not been previously reported in broas [27]. Among these, 5-methyl-2-furanmethanol showed a relatively low OT and was confirmed as a key aroma compound of Chinese white bread [49]. It was not possible to confirm the OTs of 3-furfural and isomaltol, but they have been considered important compounds to the aroma of popcorn [33] and wheat bread [5,21], respectively.
Only two out of the twenty-six furanones detected in the present work have been previously detected in broas, namely N-caprolactone (Fo19) and γ-nonalactone (Fo26). Both are possible contributors to broa's sweet aroma [5,41] since they were present in high amounts in broas and are potent odorants [44,47]. γ-Octalactone (Fo23), detected in broas for the first time, may also contribute to their aroma, considering its abundance and low OT.

Pyrans and Pyranones
Pyrans have not been described in similar food products, but some may also be important odorants, specially tetrahydropyrans [48]. In contrast, pyranones occur in the volatiles of all heated foods [20], conferring sweet, burnt, pungent and caramel-like flavors and aromas [15]. The most relevant pyranones were maltol (Po7) and 3-hydroxy-2,3dihydromaltol (Po8). These compounds have been previously described as possible positive contributors to the 'taste and aroma' of broas [27]. Maltol imparts desirable, caramel-like, sweet and fruity characteristics to foods [14,20,21]. It can mask bitter flavors [14] and is considered a key odorant in cereal products [67]. Although it has a relatively high OT of 2500-9000 µg L −1 [14,43], it may be important for the aroma of broas due to its abundance.

Pyridines and Pyrimidines
Pyridines and pyrimidines were detected in broas for the first time. Although they were not present in high amounts, they might still be relevant to the overall aroma due to their low OTs. Pyridines give roasted and popcorn odors to foods [14,43].
2-Acetyltetrahydropyridines (or 6-acetyltetrahydropyridines) are possibly the most common pyridines in similar foodstuffs. They are potent Maillard flavor compounds that contribute substantially to caramel, roasty, and bready aromas [51,71] in several bakery products [14,16,32,71], and are formed by the degradation of proline and hydroxyproline [20]. However, these compounds were not detected in broas. Instead, 1-acetyl-1,2,3,4tetrahydropyridine (Pd8) was identified as the main pyridine. Although it exhibits a similar mass spectrum to 2-acetyltetrahydropyridine, Pd8 showed a very intense peak at m/z 68 ( Figure S3), which is not expected in 2-acetyltetrahydropyridine [72]. This compound has also been identified in the crusts of wheat bread [42] and lupin protein isolate-enriched wheat bread as an important contributor to its aroma profile [42]. Thus, it might also be an important broa volatile, contributing to a nutty odor [42].
2-Pentylpyridine (Pd9) may also be relevant for broas aroma, contributing to roasted and nutty odors [51]. Although it was present in lower amounts than other pyridines, it shows a very low OT [48]. This compound has not been described in similar foods, probably due to its low concentration, but it has been described in several fried foods [73].
4-Methylpyrimidine (Pd2) was the only pyrimidine detected in broas. This compound has not been described either in bread or in other maize-based foods, and it was not possible to obtain any information regarding its odor characteristics.

Pyrroles, Pyrrolines and Oxazoles
Pyrrole derivatives are responsible for roasted odors [14]. Skatole (Py12) has a very low OT of 0.2 µg L −1 [16] and may be relevant to the aroma of broas. It is usually described as an off-flavor. However, in low concentrations, it may introduce a natural note of 'overmature flower' [51].
One of the most striking results to emerge from this study was the absence of pyrrolines in broas. These compounds are abundant bread volatiles, significantly contributing to their flavor [5,7]. Pyrrolines have been described as character impact odorants of the 'roasted' and 'popcorn-like' notes [7]. Important pyrrolidines described in cereal bread and maizebased foods are 2-acetyl-1-pyrroline and its precursor 1-pyrroline [7,14,20,59]. However, recent studies have shown that 2-acetyl-1-pyrroline varies greatly among different cereal bread [69]. For instance, 2-acetyl-1-pyrroline seems not to be so relevant to the aroma properties of rye bread [7], and it was not detected in gluten-free bread, which was analyzed by GC×GC [31]. By contrast, other authors have found that 2-acetyl-1-pyrroline was higher in gluten-free bread [56]. These differences can be explained by the presence or absence of precursors or interferents in the cereal flours. For instance, differences in the ornithine content of yeasts may play a role in forming pyrrolines since this amino acid has been ascribed as the most important precursor for forming 2-acetyl-1-pyrroline during baking [7,11,20]. 2-Acetylpyrroline may also be formed by the degradation of proline and hydroxyproline, similar to 2-acetyl-tetrahydropyridines [20], which were also not found in broas, as previously discussed. Thus, differences in the amounts of proline and hydroxyproline may also influence the production of 2-acetylpyrroline and 2-acetyltetrahydropyridines. In addition, phenolic acids may inhibit the production of 2-acetyl-1-pyrroline [74], and whole maize flours were used to prepare broas, which have higher amounts of phenolic compounds compared to other cereals or with refined maize flours [75]. Lastly, 2-acetyl-1-pyrroline is highly volatile and can be oxidized to 2-acetylpyrrole [14,20], one of the major volatile compounds detected in the present work.
Three oxazoles were detected in broas, but they probably have a low impact on their overall aroma, as reported for other foods [20]. They were present in broas in very low amounts compared to other compounds.

Sulfur-Containing Compounds
Sulfur compounds had not been previously reported in broas [27], possibly due to their low concentrations. Usually, foods from cereals show very low contents in sulfur-containing compounds due to the low amounts of sulfur amino acids [20]. However, sulfur-containing Maillard odorants constitute the most powerful aroma compounds of foods, even though they are present at trace levels [14,15]. These compounds have traditionally been associated with unpleasant and noxious off-flavors [13,14]. Still, they contribute to positive flavor characteristics at low concentrations (<1 µg kg −1 ), giving tropical, fruity, and savory aromas to foods [13]. They are considered "character-impact compounds" of bread crust, popcorn and toasted cereal grains [13,15].
Dimethyl sulfide (S2), dimethyl disulfide (S3), and, in particular, dimethyl trisulfide (S8) may be extremely relevant for the aroma of broas. At low levels, they all show positive aroma characteristics. Dimethyl trisulfide (S8) has an extremely low OT of 0.01 µg L −1 [43] and, when present at low concentrations, is associated with tropical fruit and grapefruit aromas [13]. The most abundant sulfur compound detected in broas was dimethyl disulfide (S3), an important contributor to maize flavor [24]. Dimethyl sulfide (S2) is considered a flavor impact compound of sweet maize and conveys the typical flavor impression of canned maize when present at reduced levels [13]. Methanethiol (S1) was the second most abundant sulfur-containing compound and may also significantly impact the aroma of broas. It has been described as a possible contributor to the aroma of sweet maize [13].
Another sulfur-containing compound detected in broas was the very well-known methional (S7) [68], possibly also one of the most relevant volatiles for the aroma of broas. It is a potent [5] and desired odorant [21], with a characteristic potato-like aroma [7,59]. It has been considered a key odorant of fried foods [43], wheat, rye [7] and Chinese white [49] bread. It may be responsible in part for typical popcorn [33], maize tortilla chips [43] and taco shell [32] aroma characteristics. Furthermore, it has flavor-modifying characteristics since it has been shown to suppress flavor and aroma [59]. Other aliphatic sulfur compounds detected in broas were 1-methylthiopropane (S4) and methylthio-2propanone (S5). These compounds have not been described in similar foods, and their impact on the overall aroma of broas is unknown.
Regarding aromatic sulfur heterocycles derivatives, it was possible to detect three thiazoles and six thiophenes. Thiazoles usually confer green, vegetable-like, cocoa and nutty aromas to foods [20], while thiophenes contribute to sulfurous, nutty and fatty aromas [51]. Among them, 2-acetylthiazole (Tz2) may have high importance in the aroma of broas. It has been described as a key volatile of popcorn and maize flour extrudates [40].
It was possible to detect six thiophenes in broas. Still, they have not been so commonly described in similar foods, and it was not possible to find information regarding their OTs in the literature.
Another sulfur heterocycle compound detected in broas was an oxathiane derivative (S6), which may refer to 1,2-, 1,3-or 1,4-oxathiane. However, considering their polarities (Log P of 1.14, 0.43 and 0.36, respectively) and the 2 t R of S6 (1.10 s), it was tentatively identified as 1,2-oxathiane. This compound has not been described in foods, but oxathiane derivatives have been ascribed as important aroma compounds. The most relevant example is 2-methyl-4-propyl-1,3-oxathiane, a key aroma compound of passion fruit, with a low OT of 3 µg L −1 [13]. Recently, this compound and 2,4,4,6-tetramethyl-1,3-oxathiane have been detected in wines [76]. Although the impact of oxathiane on the aroma of foods is currently unknown, it may impact the aroma of broas, and this compound is worth further study.

An Overall View of the Volatiles Associated with Baking in Broas
Sensory analysis in a previous study [77] revealed that the broa prepared from the available commercial maize hybrid variety (B12) showed the lowest scores for 'smell and odor' and 'taste and aroma'. In contrast, the traditional broas (obtained from the traditional open-pollinated varieties) were poorly discriminated among them [77]. A subsequent study revealed that these differences could at least be partially explained by their volatile composition [27]. Thus, a PCA (principal component analysis) was performed to explore further these differences for an easy, rapid and global assessment of the main differences in the volatile composition among the studied broas. Figure 5 shows the projection of samples and variables in the space defined by the two principal components, corresponding to 59.2% of the total variance. A hierarchical cluster analysis combined with a heatmap representation was also constructed and shown in Figure S4.

An Overall View of the Volatiles Associated with Baking in Broas
Sensory analysis in a previous study [77] revealed that the broa prepared from the available commercial maize hybrid variety (B12) showed the lowest scores for 'smell and odor' and 'taste and aroma'. In contrast, the traditional broas (obtained from the traditional open-pollinated varieties) were poorly discriminated among them [77]. A subsequent study revealed that these differences could at least be partially explained by their volatile composition [27]. Thus, a PCA (principal component analysis) was performed to explore further these differences for an easy, rapid and global assessment of the main differences in the volatile composition among the studied broas. Figure 5 shows the projection of samples and variables in the space defined by the two principal components, corresponding to 59.2% of the total variance. A hierarchical cluster analysis combined with a heatmap representation was also constructed and shown in Figure S4. Since all broas were prepared following the same procedure and submitted to the same temperature of kneading and baking, it can be stated that the differences observed among the different broas were caused by differences in the corresponding maize flours, which affect the Maillard reaction. A matrix correlation analysis was conducted and rep- Since all broas were prepared following the same procedure and submitted to the same temperature of kneading and baking, it can be stated that the differences observed among the different broas were caused by differences in the corresponding maize flours, which affect the Maillard reaction. A matrix correlation analysis was conducted and represented by a heatmap to shed some light on the volatile compounds which might have been formed from the same precursors ( Figure 6 and Tables S4 and S5). The volatiles that strongly contributed to differentiating the samples along PC2 (Figure 5) are represented in group A of the correlation matrix ( Figure 6) and highlighted in Table S4. This group is characterized by a simple pattern of strong and positive correlations amongst several furans and furanones. In low-moisture starchy foods, such as broas, the Maillard reaction seems to be the main route of their formation [78,79]. Thus, higher amounts of the Maillard reaction precursors, such as amino acids and sugars, may have generated broas with higher amounts in furans. Alternatively, the presence of interferents, for instance, phenolic compounds [27], may explain these correlations since they may promote [27,80] or inhibit [74,80] the reaction. In particular, very strong correlations (R > 0.85, The volatiles that strongly contributed to differentiating the samples along PC2 ( Figure 5) are represented in group A of the correlation matrix ( Figure 6) and highlighted in Tables S4 and S5. This group is characterized by a simple pattern of strong and positive correlations amongst several furans and furanones. In low-moisture starchy foods, such as broas, the Maillard reaction seems to be the main route of their formation [78,79]. Thus, higher amounts of the Maillard reaction precursors, such as amino acids and sugars, may have generated broas with higher amounts in furans. Alternatively, the presence of interferents, for instance, phenolic compounds [27], may explain these correlations since they may promote [27,80] or inhibit [74,80] the reaction. In particular, very strong correlations (R > 0.85, p < 0.05) were observed among 2-methylfuran (F2), 5-methyl-2-furanmethanol (F19) and 2-acetyl-5-methylfuran (F25), which are usually produced by caramelization reactions and by the breakdown of the Amadori or Heyns intermediates in the early stages of the Maillard reaction ( Figure 1) [14,19,20]. Strong and positive (R > 0.7, p < 0.05) correlations were also found between dihydro-2-methyl-3-(2H)-furanone (Fo2) and sulfur-containing heterocycles. Higher amounts in dihydro-2-methyl-3-(2H)-furanone, which reacts with ammonia and hydrogen sulfide (produced from cysteine by hydrolysis or by Strecker degradation) [19,20,24,81], may directly cause higher amounts of thiazoles and thiophenes.
The volatile compounds that strongly contributed to differentiating the samples along PC1 ( Figure 5) are placed in group B of the correlation heatmap ( Figure 6 and Tables S4 and S5) and consisted, among others, of several pyrazines. The major route for the formation of pyrazines is thought to be from α-aminoketones, which are products of the condensation of a dicarbonyl with an amino compound via Strecker degradation during the Maillard reaction [20,24]. Very strong and positive correlations were found between Pz7 (2-methyl-5-vinylpyrazine) and Pz3 (2,6-dimethylpyrazine), both products from the reaction between leucine and fructose [59]. The broa prepared from the available commercial hybrid variety maize flour (B12) showed a higher content in pyrazines (p < 0.05) than all the traditional varieties (Table S2 and Figure 7), especially in 2-methylpyrazine (Pz2) and 3-dimethylpyrazine (Pz5) ( Table S3). The higher amounts of total ferulic acid in the traditional maize varieties [82] might have inhibited the formation of pyrazines [80] in the corresponding broas. Maillard reaction (Figure 1) [14,19,20]. Strong and positive (R > 0.7, p < 0.05) correlations were also found between dihydro-2-methyl-3-(2H)-furanone (Fo2) and sulfur-containing heterocycles. Higher amounts in dihydro-2-methyl-3-(2H)-furanone, which reacts with ammonia and hydrogen sulfide (produced from cysteine by hydrolysis or by Strecker degradation) [19,20,24,81], may directly cause higher amounts of thiazoles and thiophenes. The volatile compounds that strongly contributed to differentiating the samples along PC1 ( Figure 5) are placed in group B of the correlation heatmap ( Figure 6 and Table  S4) and consisted, among others, of several pyrazines. The major route for the formation of pyrazines is thought to be from α-aminoketones, which are products of the condensation of a dicarbonyl with an amino compound via Strecker degradation during the Maillard reaction [20,24]. Very strong and positive correlations were found between Pz7 (2methyl-5-vinylpyrazine) and Pz3 (2,6-dimethylpyrazine), both products from the reaction between leucine and fructose [59]. The broa prepared from the available commercial hybrid variety maize flour (B12) showed a higher content in pyrazines (p < 0.05) than all the traditional varieties (Table S2 and Figure 7), especially in 2-methylpyrazine (Pz2) and 3dimethylpyrazine (Pz5) ( Table S3). The higher amounts of total ferulic acid in the traditional maize varieties [82] might have inhibited the formation of pyrazines [80] in the corresponding broas. Lastly, the volatiles which negatively contributed to PC1 and PC2 are located in group C ( Figure 6). These volatiles show strong and negative correlations with several of the volatiles which belong to groups A and B. Furfural (F12) was strongly and negatively correlated to both benzofuran (F21) and furfurylfuran (F28). Thus, broas with higher amounts of furfural also showed lower amounts of benzofuran and furfurylfuran, and vice versa. Furfural is mainly generated in the initial stages of the Maillard reaction and participates in further reactions [14,24,83]; thus, differences in the precursors which reacted with furfural and generated benzofuran and furfurylfuran, or the presence of interferents of this reaction [27,80], may have influenced its extent. Similarly, negative correlations were found among several sulfur compounds and both furans and furanones, which have been described as important precursors of thiazoles and thiophenes [24], and between oxathiane (S6) and sulfur-containing compounds, such as dimethyl disulfide (S3), dimethyl trisulfide (S8), 2-acetylthiophene (Tp5) and benzothiazole (Tz3), which might have been precursors of the formation of oxathiane. A recent study in wines has proposed that thiols act as precursors of the formation of oxathiane [76].
Taken together, these results suggest that the main differences among the volatile compounds of the studied broas are likely due to the presence of specific precursors or Lastly, the volatiles which negatively contributed to PC1 and PC2 are located in group C ( Figure 6). These volatiles show strong and negative correlations with several of the volatiles which belong to groups A and B. Furfural (F12) was strongly and negatively correlated to both benzofuran (F21) and furfurylfuran (F28). Thus, broas with higher amounts of furfural also showed lower amounts of benzofuran and furfurylfuran, and vice versa. Furfural is mainly generated in the initial stages of the Maillard reaction and participates in further reactions [14,24,83]; thus, differences in the precursors which reacted with furfural and generated benzofuran and furfurylfuran, or the presence of interferents of this reaction [27,80], may have influenced its extent. Similarly, negative correlations were found among several sulfur compounds and both furans and furanones, which have been described as important precursors of thiazoles and thiophenes [24], and between oxathiane (S6) and sulfur-containing compounds, such as dimethyl disulfide (S3), dimethyl trisulfide (S8), 2-acetylthiophene (Tp5) and benzothiazole (Tz3), which might have been precursors of the formation of oxathiane. A recent study in wines has proposed that thiols act as precursors of the formation of oxathiane [76].
Taken together, these results suggest that the main differences among the volatile compounds of the studied broas are likely due to the presence of specific precursors or interferents which affect the extent of some reaction pathways of the Maillard reaction, as the formation of pyrazines and oxathiane, rather than the overall Maillard reaction. These results can partially explain the contradictory reports on the effect of phenolic compounds on the Maillard reaction [27,74,80]. Differences in precursors or interferents of certain pathways of this reaction in maize varieties can cause differences in broas aroma.
As previously stated, the analyzed commercial hybrid variety sample (B12) showed lower scores in a sensory analysis [77]. The results from the present study have shown that this sample showed a higher content in pyrazines (Figure 7), particularly in 2-methylpyrazine (p < 0.05). These results were somewhat surprising since pyrazines are strong odorants usually associated with positive sensorial characteristics, giving roasted and nutty aromas to bread [5,51]. However, pyrazines may contribute to aroma characteristics in broas that are not typical of this type of bread, giving rise to some of the negative comments associated with B12, such as a 'weak typical flavor', 'wheat bread flavor', 'with a weak maize flavor' and 'no history' [27]. Similarly, it has been reported that most pyrazines contributed negatively to the aroma of popcorn [33]. Besides the higher amount in pyrazines, a closer examination of Table S3 also showed that 2-butyltetrahydrofuran (F31) was present in B12 in higher (p < 0.05) amounts. However, this compound has not been described in similar foods, and it was not possible to infer its impact on broas aroma. Therefore, these data should be further explored for the characterization of broas aroma and fingerprinting [29], since both 2-butyltetrahydrofuran and 2-methylpyrazine may be potential biomarkers of the authenticity of broas prepared from traditional maize varieties.
Other compounds can contribute to the lower sensory attributes of B12, namely the higher contents in some pyridines and pyrroles, including the powerful skatole (Py12), often associated with off-aromas [51], and lower contents in maltol (Po7), associated with higher 'taste and aroma' [27]. Although the ANOVA showed that these results were not statistically different from some traditional varieties (p > 0.05) ( Table S3), combinations of volatiles can yield different characteristics than those expected from individual compounds [12,30], and therefore contribute to the lower sensory attributes of broas prepared from the commercial maize hybrid variety under study.

Samples
Eleven broas (B1 to B11) were prepared from eleven traditional open-pollinated maize varieties, described in Table S1. The traditional samples were chosen as representative of the Portuguese maize germplasm variability, taking into account their agronomic performance in field trials, basic nutritional quality and genetic diversity evaluated under the scope of the FP7 SOLIBAM European project. One broa (B12) prepared from a commercial maize hybrid variety was also studied.
Broas was prepared following a traditional recipe, as previously described [27]. The ingredients included 70% maize flour, 20% commercial rye flour (Concordia type 70, Portugal) and 10% commercial wheat flour (National type 65, Portugal). The dough was manually molded into 400 g balls and baked in a gas oven (Matador; Werner & Pfleiderer) at 270 • C for 40 min.

HS-SPME Methodology
The HS-SPME methodology was based on the previous work developed by Bento-Silva et al., 2021 [27]. Briefly, 4 g of broas (whole bread, including crumb and crust) were smashed manually and placed in a 20 mL vial. A 50/30 µm DVB/CAR/PDMS (divinylbenzene/carboxen/polydimethylsiloxane) fiber purchased from Supelco (Sigma-Aldrich, St. Louis, MO, USA) was used to concentrate the volatile compounds present in the headspace. Samples were extracted for 40 min in a thermostated bath adjusted to 60.0 ± 0.1 • C at 250 rpm. All samples were extracted in duplicate.

GC×GC-ToFMS Analysis
The GC×GC-ToFMS experimental parameters were adapted from Costa et al. (2020) [65]. The equipment used was a LECO Pegasus 4D GC×GC-ToFMS system (LECO, St. Joseph, MI, USA) consisting of an Agilent GC 7890A gas chromatograph (Agilent Technologies, Inc., Wilmington, DE, USA) with a dual-stage jet cryogenic modulator (licensed from Zoex) and a secondary oven, and a mass spectrometer equipped with a time-of-flight (ToF) analyzer.
The SPME fiber was manually introduced into the port at 250 • C for analyte desorption. In an initial approach, the total ion chromatograms were processed using the automated data processing software ChromaTOF (LECO) at the S/N threshold of 100. The obtained GC×GC total ion chromatogram contour plots exhibited more than 1200 peaks. As the present study focused on the compounds mainly produced during baking, in a second approach, all the compounds belonging to the characteristic chemical families associated with the Maillard and caramelization reactions were selected. In addition, the volatile compounds described in the literature as key aromas of bread or maize-based foods and which were not detected using the automated data processing (S/N < 100) were searched on the chromatograms based on extracted ion chromatogram contour plots of the characteristic ions, and the corresponding peak areas were also included.
For identification purposes, the mass spectrum and retention times ( 1 D and 2 D) of the analytes were compared with the mass spectral libraries, namely an in-house library of standards and two commercial databases (Wiley 275 and US National Institute of Science and Technology (NIST) V. 2.0-Mainlib and Replib). Additionally, the linear retention index (LRI) was experimentally determined according to the van den Dool and Kratz LRI equation [84]. A C 8 -C 20 n-alkane series was used for LRI determination (the solvent n-hexane was used as the C 6 standard). These values were compared with those reported in the literature for chromatographic columns similar to the 1 D column mentioned above. A positive identification was considered when the experimental spectra shared at least an 80% similarity with spectra from the software libraries and the LRI deviation was less than 5%. This difference in LRI takes into account that (i) the literature data are obtained from a large range of GC stationary phases (several commercial GC columns are composed of 5% phenylpolysilphenylene-siloxane or equivalent stationary phases), and (ii) the modulation causes some inaccuracy in the first dimension retention time, and the majority of the values reported in the literature were determined in a 1D-GC separation system [64]. The Log p values were calculated using ALOGPS 2.1 to confirm the identification of some compounds which did not match all the criteria described above [85].
The DTIC (deconvoluted total ion current) GC×GC peak area data were used to estimate the relative content of each volatile component in the different samples.

Statistical Analysis
The peak area data of all studied compounds were extracted from the chromatograms and used to build the full data matrix consisting of 12 broas samples and 128 variables. After data normalization, hierarchical cluster analysis (HCA) and Pearson's coefficient correlations, both combined with heatmap visualizations, were applied for this dataset using the MetaboAnalyst 3.0 (web software, The Metabolomics Innovation Centre (TMIC), Edmonton, AB, Canada). Independent sample t-tests, ANOVA followed by post hoc Tukey tests and principal component analyses (PCA) were obtained using the software SPSS version 21 (IBM, Armonk, NY, USA). The limit of significance was set at p < 0.05.

Conclusions
This study aimed to extend previous research on the volatile compounds of broas, focusing on those mainly associated with the baking process. Almost 90% of the compounds identified in this work have not been previously detected in broas. One of the most relevant study findings was the absence of pyrrolidines and the lack of pyrazines in broas, especially when prepared from traditional open-pollinated maize varieties. Thus, the absence of pyrrolidines may contribute to the distinctive aroma of broas, and the high amounts of pyrazines may confer negative characteristics associated with the commercial hybrid maize broa analyzed. Sulfur compounds, such as dimethyl trisulfide and methanethiol, were identified as the most likely contributors to the aroma of this ethnic bread. Some volatiles, such as oxathiane, have not been previously reported in similar foods, and their relevance to the overall aroma of foods is currently unknown. The data obtained in this study can be further explored for the purpose of fingerprinting traditional broas since some compounds (2-methylpyrazine and 2-butyltetrahydrofuran) were present in significantly higher amounts in the broa prepared from the analyzed commercial maize hybrid variety. In conclusion, this work represents the most detailed study on the volatile composition of broas. It may contribute to disclosing possible volatiles of other bread and maize-based foods. These findings may have a number of important implications for future practice since better knowledge of the volatile compounds produced along the Maillard reaction may help to achieve the production of foods with sensory characteristics more appreciated by consumers.
Supplementary Materials: The following supporting information can be downloaded at https://www. mdpi.com/article/10.3390/molecules27092728/s1; Figure S1: Visual representation of the volatile baking compounds reported in the literature for several bread and maize-based foods (nodes), Figure S2: Picture of broas, Figure S3: Electron ionization mass of Pd8 (1-acetyl-1,2,3,4-tetrahydropyridine) and mass spectra of 1-acetyl-1,2,3,4-tetrahydropyridine and 2-acetyl-1,4,5,6-tetrahydropyridine from the Wiley library, Figure S4: Heatmap and hierarchical cluster analysis representation of the 128 volatiles identified in broas. The content of each metabolite is illustrated through a chromatic scale, Table S1: Maize flours and broas identification and description., Table S2: Average total peak areas and % of chromatogram area for the families of chemical compounds studied in broas, Table S3: Peak areas obtained for each sample (average of duplicates) and considered for the statistical analysis, Tables S4 and S5: Spearman correlation coefficients among the 128 studied broas volatile compounds.