Characterization of the yerba mate (Ilex paraguariensis) volatile fraction using solid-phase microextraction-comprehensive 2-D GC-MS



The present research is focused on the use of a solid-phase microextraction-comprehensive 2-D GC methodology, in the analysis of the volatile fraction of yerba mate. Yerba mate is used for the generation of a tea-like beverage, widely consumed in South America. A rapid-scanning quadrupole mass spectrometer (qMS), employed as a detection system and operated at a 25 Hz scanning frequency, supplied high-quality mass spectra. The effectiveness of the 3-D comprehensive 2-D GC-qMS experiment was compared to that of GC-qMS analysis on the same sample. Peak identification, in both applications, was achieved through MS library matching, with the support of linear retention index data. Apart from a great increase in the number of analytes separated (approx. by a factor of 5) and identified (approx. by a factor of 3.5), the comprehensive 2-D GC-qMS approach enabled the determination of a high number of hazardous contaminants (aliphatic hydrocarbons, polycyclic aromatic hydrocarbons, and plasticizers), barely visible in the GC-qMS analysis.

1 Introduction

Yerba mate (mate) is a tea-like beverage widely consumed in South American countries as a tonic and as a stimulant to reduce fatigue. The drink is obtained through the infusion of the leaves and stems of the perennial tree Ilex paraguariensis St. Hil. (Acquifoliaceae family). Prior to use, the raw natural material is subjected to several processes: after harvesting, the leaves are blanched by flash heating over wood or propane fire (about 1 min), to deactivate specific enzymes (in particular, poliphenol oxidase), and then slowly dried often by using wood smoke. Some producers include an aging step for fermentation and a roasting one for flavour-development 1.

Mate is also used in popular medicine and is included in commercial herbal preparations. Indeed, several health benefits have been reported in the literature, such as hepatoprotective, central nervous system stimulant, hypocholesterolemic, antirheumatic, anti-thrombotic, anti-inflammatory, and antioxidant effects 2–5. Such beneficial effects have been correlated to the several bioactive compounds found in mate tea, such as polyphenols (chlorogenic acid), xanthines (caffeine and theobromine), flavonoids, amino acids, minerals (P, Fe, and Ca), and vitamins (C, B1, and B2) 6, 7. Despite these proven health benefits, epidemiological studies have reported negative effects related to mate consumption, correlated to the presence of several contaminants, in particular polycyclic aromatic hydrocarbons (PAH) 1. The volatile composition of mate, which is also characterized by the presence of antioxidant compounds, and obviously contributes greatly to the flavour of the beverage, has been rarely investigated. The first study, carried out in 1991 by Kawakami and Kobashi 8, investigated the composition of the steam-distilled fraction of green and roasted mate by using GC-MS, with a 50 m×0.25 mm id Carbowax column: considering both matrices, 196 compounds were identified (172 and 168 compounds, respectively). In 2006, Bastos and co-workers 9 studied the effect of roasting on the antioxidant activity of mate infusion, highlighting the presence of 30 volatile compounds in the distilled fraction by using GC-MS. Finally, in 2007, Araújo and co-workers 10 identified 70 compounds by using headspace (HS)-solid-phase microextraction (SPME)-GC-MS.

The full elucidation of the volatile composition of a natural food matrix can be a cumbersome task, using single-column GC-MS; although MS detection can be very useful in the reliable identification of overlapping peaks (single ion monitoring, extracted ions, deconvolution processes, etc.), it is certainly desirable to achieve a high-resolution GC separation step. Unfortunately, 1-D GC often fails to provide a satisfactory analytical result, even on moderately complex samples.

Comprehensive 2-D chromatography (GC×GC) is a very high resolution power GC methodology, widely used in food analysis 11–13. Briefly, in GC×GC peaks eluting from a primary conventional column are fractionated, re-concentrated and injected onto a secondary column of differing selectivity (usually a short micro-bore segment), by using a transfer device, defined as modulator. Modulation is carried out in a continuous manner, at short intervals, defined as the modulation period. The raw data obtained from a GC×GC analysis consist of a stream of high-speed 1-D chromatograms and, therefore, dedicated software is necessary to generate a 2-D space plane. The possibility of peak co-elution is greatly reduced by applying this innovative technology; furthermore, the combination of an MS detector generates a very powerful 3-D analytical tool 13.

In the present work, GC×GC in combination with a rapid-scanning quadrupole MS (qMS) was exploited to investigate the volatile fraction of yerba mate. Automated SPME was employed to extract and concentrate volatile compounds from the HS of a Brazilian commercial sample. The GC×GC separation phase enabled the resolution of over 1000 peaks; amongst the latter, more than 240 compounds were identified with a spectral similarity match of at least 90%. Linear retention indices (LRI) were also exploited to confirm peak assignment. The results attained using the 3-D approach were directly compared with those derived from an optimized SPME-GC-qMS method.

2 Materials and methods

2.1 Sample and SPME procedure

The C7-C30 alkane mixture was kindly provided by Sigma-Aldrich (Milan, Italy).

The mate sample was purchased from a local market in Rio Grande do Sul, RS, Brazil. The commercial product comprised a 70:30 mixture of ground leaves and twigs of Ilex paraguariensis St. Hil. The SPME triple phase 50/30 μm fiber (divinylbenzene/carboxen/polydimethylsiloxane) was purchased from Supelco (Milan, Italy), and was appropriately conditioned before use. A Shimadzu AOC-5000 autosampler (Kyoto, Japan) was used for the HS-SPME operations.

Briefly, 150 mg of dry mate sample were introduced in a 5 mL vial. The sample was heated at 80°C for 15 min (pre-incubation) and agitated (by shaking, using clockwise–anticlockwise alternate rotation) at 500 rpm. The fiber, previously cleaned by thermal desorption, was then exposed in the HS for 60 min at the same temperature and agitation speed. After this process, the fiber was thermally desorbed in the GC injection port for 1.0 min at 250°C in the splitless mode (after 1 min, a 100:1 split ratio was applied). A blank fiber analysis was previously performed, following the above described procedure, but using an empty vial.

2.2 Instrumentation

2.1.1 GC×GC-qMS analysis

The GC×GC-qMS applications were carried out on a Shimadzu GC×GC-MS system consisting of two independent GC2010 gas chromatographs and a QP2010 Plus qMS. The primary GC was equipped with an AOC-20i auto-injector, a split-splitless injector (280°C), and a cable extension for the MS connection (due to the presence of the second oven). The first column was connected by using an SGE SilTite mini-union (Ringwood, Victoria, Australia) to an uncoated tubing (2 m×0.25 mm id), that was passed through a heated transfer line (270°C) into the second oven, where a dual-stage loop-type modulator (under license from Zoex Corporation, Houston, TX, USA) system was installed, and was finally connected to the secondary column by using an SGE SilTite mini-union. Cryogenic modulation was applied every 6 s.

The first column was an SLB-5 ms 30 m×0.25 mm id×0.25 μm df column [silphenylene polymer virtually equivalent in polarity to poly(5% diphenyl/95% methylsiloxane)], while the second was an Equity-1701 1.5 m×0.10 mm id×0.10 μm film thickness [poly(14% cyanopropylphenyl/86% dimethyl) siloxane] (Supelco). Two different temperature programs were used: in GC1 from 50°C (hold for 2 min) to 270°C (hold for 15 min) at 3°C/min; in GC2 from 55°C (hold 2 min) to 270°C (hold 15 min) at 3°C/min. The carrier gas (helium) was delivered at an initial pressure of 165.9 kPa (constant linear velocity mode).

MS parameters: the sample was analyzed in the full scan mode with a scan speed of 10,000 amu/s, a mass range of 40–360 m/z and a sampling frequency of 25 spectra/s; interface and ion source temperatures were 250 and 200°C, respectively. MS ionization mode: electron ionization; detector voltage: 1.0 kV. Data were collected by the GCMS solution software (Shimadzu); bidimensional visualization was carried out by using the ChromSquare v.1.0® software (Chromaleont, Messina, Italy). The MS library used for spectral matching was the FFSNC 1.3 (Shimadzu).

2.1.2 GC-qMS analysis

GC-qMS applications were carried out on a GC2010 gas chromatograph and a QP2010 Plus qMS. An SLB-5ms 30 m×0.25 mm id×0.25 μm df column was used. The temperature program applied was: 50°C (hold 2 min) to 270°C (hold 15 min) at 3°C/min. The carrier gas (helium) was delivered at an initial pressure of 32.0 kPa (constant linear velocity mode: 30.8 cm/s).

MS parameters were the same as previously described except for the scan speed (769 amu/s) and sampling frequency (2 spectra/s). The library used was the FFSNC 1.3.

3 Results and discussion

Initially, an SPME-GC-qMS analysis of the commercial mate sample was carried out. Despite the generation of a visibly complex chromatogram (over 200 peaks were counted; data not shown), only 70 compounds were tentatively identified with a spectral similarity ≥90% (Table 1). Thirty-six compounds were identified with a ≥95% similarity, while many peaks remained unassigned, certainly due to extensive compound overlapping. The identification procedure was carried out by using a dual-filter, previously described 14, and based on the exploitation of: (i) spectral similarity and (ii) LRI range (±5 units). The final analytical result can be considered as a typical example of an insufficient GC separation step. As aforementioned, mass spectrometry can be very useful for the structural elucidation of overlapping volatiles. However, the reliability of MS identification is dependent on the degree of co-elution and, thus, the delivery of pure (or pure-as-possible) analyte bands to the ion source is always highly desirable.

Table 1. Compounds identified in the mate GC×GC-qMS and GC-qMS analyses; library-derived (LRI Lib) and experimental LRI (defined as LRI) values, LRI differences (Δ LRI), and spectral similarities (MS%)
  • a)

    a) LRI value derived from ref. 10.

1Propylene oxide41095   
32-Methylbutyl acetate59090   
4Ammonium acetate63099   
52-Butenal61598  97
6Acetic acid57699  97
71-Penten-3-ol68096  91
82-Ethyl furane702967020 
11Propanoic acid73997725−14 
15Butyric acid773977774 
23Isovaleric acid84292834−8 
27Furfuryl alcohol860968644 
29Pentanoic acid87593884995
38Pyrazine, 2,5-dimethyl-9129192210 
455-Methyl furfural96094963392
47Hexanoic acid979989823 
533-Hexenoic acid983969951293
552,4-Heptadienal steroisomer1005961000−598
62β-Ocimene cis10359010383 
65Benzyl alcohol10409510411 
66Artemisia ketone1042901041−194
68γ-, γ-Vinyl-valerolactone10359410438 
77Heptanoic acid10739510730 
79trans-Linalool oxide10739610763 
81cis-Linalool oxide10879510925 
97Benzoic acid115094116717 
984-Ethyl phenol1171941169−2 
104Butyl diglycol1211971192−19 
117Hydroxy methyl furfural1225931231693
119Carvacryl methyl ether12399112434 
122Linalyl acetate12509512511 
125Nonanoic acid12679412692 
1322-Methyl naphthalene1308931307−193
1352-Ethylhexyl butyrate13179313214 
1374-tert-Butylcyclohexyl acetate132294133917 
1391,6-Nonadien-3-ol, 3,7-dimethyl-, acetate13449313462 
140α-Terpinyl acetate13499313534 
142Propanoic acid, 2-methyl-, 3-hydroxy-2,4,4-trimethylpentyl ester13479113558 
144Decanoic acid13649613662 
1474-tert-Butyl-cyclohexanol acetate13689413735 
161Geranyl acetone1455931451−493
167Germacrene D14809114877 
169β-Ionone epoxide146095149232 
176Dodecanoic acid1565961563−2 
1793-Hexen-1-ol, benzoate15739515796 
180Propanoic acid, 2-methyl-, 1-(1,1-dimethylethyl)-2-methyl-1,3-propanediyl ester1605901591−14 
182Caryophyllene oxide158793159912 
185Isopropyl laurate161593162611 
196(2E, 6Z)-Farnesal17149017162 
204Tetradecanoic acid1769961765−4 
209Isopropyl tetradecanoate18269518260 
210Vetivenic acid181290182311 
2125,9,13-Trimethyl- 4,8,12-tetradecatrienal1855941838−17 
216Isobutyl phthalate1908971865−4397
219Farnesyl acetone1913961916396
221Hexadecanoic acid, methyl ester19259619261 
2241-Butyl 2-isobutyl phthalate1973901961−1290
225Hexadecanoic acid1977951964−13 
226Ethyl hexadecanoate19939319952 
228Hexadecyl acetate20039620096 
229Isopropyl hexadecanoate20239420252 
230Octadecyl acetate20849620928 
232Methyl oleate20859421021794
236Octadecanol acetate22099122090 
237Tributyl-citrate acetate2254912250−4 
239Para-methoxy-, octyl-cinnamate23219123311091

At this point, a GC×GC-qMS method was developed with the aim of unravelling, in a satisfactory manner, the mate volatile profile. The first experiments were carried out setting the same temperature program in both the GC ovens: such a configuration generated 2-D chromatograms characterized by a high degree of wrap-around, with the main analyte band crossing both the lower and upper x-axis (data not shown). This negative visualization aspect was resolved through the application of a +5°C temperature offset in the second GC oven. The rapid-scanning qMS instrument was operated at a 40–360 m/z mass range and generated 25 spectra/s, which was more than sufficient for qualitative aims and was in most cases nearly sufficient for proper peak re-construction (on average, 7–8 data points per peak were attained). Moreover, mass spectral variation was measured across several single 2-D peaks (peak skewing) and was found to be negligible.

The optimized SPME-GC×GC-qMS method enabled the separation of a greatly increased number of peaks: over 1000 were counted on the 2-D space plane (Fig. 1). As it can be seen, a considerable amount of the available 2-D space was exploited for analyte separation. Two hundred and forty-one compounds were tentatively identified, with satisfactory library matches and through the application of a rather wide ±20 LRI range. The same dual-filter process used in the GC-qMS analysis was applied in the GC×GC-qMS experiment; however, a less restrictive LRI range was applied to compensate the effects of the second polar column (in GC-qMS analysis, using an apolar capillary, a ±5 LRI window is commonly employed). The tentatively identified compounds are listed in Table 1: the peaks assigned are illustrated in three sequential chromatogram expansions reported in Fig. 2A–C, which also highlights the great complexity of the mate volatile fraction, visible, in particular, in the 25–50 min zone (Fig. 2B). The LRI values of the first seven compounds are missing because they elute before the first alkane (C7) included in the standard mixture. The percentage of compounds identified with a ≥95% similarity increased slightly with respect to the GC-qMS application (56.8% vs. 51.4%). The LRI values, obtained by performing a comprehensive GC-MS analysis of a mixture of alkanes (C7–C30), were matched with data reported in the MS library for the primary 30 m column. It must be added that the influence of a 1.5 m polar capillary, in terms of retention, was negligible for the less polar analytes, while the more polar components, on the contrary, had more intense interactions. With respect to the MS library LRI values, a medium variation of 7 units (absolute value) was observed in the GC×GC analysis, indicating a general good correspondence between library and experimental values; in fact, when plotting the latter LRI values in a graph, an interpolated curve with a slope of 1.0012 and a regression coefficient (R2) of 0.999 was obtained. With regards to the GC-qMS application, as expected, the average degree of LRI variation was better, corresponding to one unit. Only a few compounds, with a similarity over 90%, were deleted by the LRI filter. LRI values were calculated through data derived from untransformed GC×GC chromatograms as follows: the retention time of the central peak was considered in the case of an odd number of cuts, while in the case of an even number of cuts, the central retention time between the two most internal peaks was considered. A total of five compounds [2-butoxyethanol (34), phenol (50), β-ionone epoxide (169), dihydroactinidiolide (175), and isobutylphthalate (216)] with a high similarity (≥95%), but with an LRI outside the pre-defined range, were exceptionally tentatively identified on the basis of their presence in literature-derived mate data 8–10.

Figure 1.

TIC GC×GC-qMS chromatogram relative to the commercial mate sample.

Figure 2.

(A–C) Three expansions derived from the chromatogram illustrated in Fig. 1, reporting the identified peaks (peak identification is reported in Table 1).

Observing the chromatograms illustrated in Figs. 1 and 2, it appears that peak distribution is rather chaotic with no hint of the presence of a typical GC×GC feature, viz., group-type patterns. If the 1-D and 2-D retention times, of a series of different homologous compounds (alkanes, aldehydes, ketones, alcohols, and carboxylic acids) identified in the mate sample, are plotted in an Excel-generated graph, then the situation changes, as can be seen in Fig. 3: linear hydrocarbons are situated in a distinct band in the apolar zone of the second dimension, while carboxylic acids are aligned in a band located in the polar 2-D zone. As to be expected, the aldehyde, alcohol, and ketone bands are located within an intermediate zone defined by the alkanes and acids.

Figure 3.

Distribution of homologous series (aldehydes, acids, alcohols, ketones, and hydrocarbons) of compounds in the 2-D plot.

Obviously, all compounds identified in the preliminary SPME-GC-qMS experiment were found using GC×GC; moreover, a direct comparison was made with previous SPME-GC-MS mate work 10: apart from methylsalicylate, (Z)-nerolidol, and 6,10,14-trimethyl-2-pentadecanone, the presence of all the other mate constituents was confirmed.

Despite the high GC-qMS similarity matches observed for 6-methyl-3,5-heptadien-2-one (87) and nonanal (86), these compounds underwent near complete co-elution, as can also be derived from the similar LRI values (1102 and 1104, respectively). Identification was carried out deriving the experimental spectra from the left and right peak shoulder (Fig. 4A), and then subtracting the background. Araújo et al. 10 identified only 6-methyl-3,5-heptadien-2-one, stating that unambiguous identification was not possible due to a probable co-elution with nonanal. Using GC×GC, the two compounds were completely resolved and the presence of other three (previously overlapping) minor components appeared (Fig. 4B). To highlight the benefits of the GC×GC analysis on spectrum similarity, the subtracted mass spectra (a spectrum showing the differences between the target spectrum and the library spectrum) relative to peak 86, in the GC-qMS and GC×GC-qMS analysis, are reported in Fig. 5A and B. The subtracted spectrum of the modulated peak (Fig. 5B) is much cleaner compared to that of the monodimensional analysis (Fig. 5A). The x-axis in Fig. 5A is characterized by a different scale due to the presence of heavier fragments, not present in the GC×GC spectrum.

Figure 4.

(A) Expansion derived from the GC-qMS chromatogram (for peak assignment see Table 1); (B) untransformed GC×GC-qMS chromatogram expansion relative to a single modulation (peaks labelled as a, b, and c were not identified). For peak assignment see Table 1.

Figure 5.

(A) Subtracted spectrum of peak 86 derived from the GC-qMS analysis; (B) subtracted spectrum of peak 86 derived from the GC×GC-qMS analysis. For peak assignment see Table 1.

The heating process (drying and/or roasting), to which the mate leaves are subjected before consumption, leads to a large amount of degradation products and enhances flavour. The furan (e.g. 8, 22, 25, etc.) and furanone (e.g. 19, 36, 44, 48, etc.) compounds identified could contribute to the sweet and smoky flavour of the beverage and are probably formed from carbohydrates or amino acids. Carotenoid degradation products, such as 2,6,6-trimethyl-cyclohexanone (64), β-cyclocitral (115), dihydroactinidiolide (175), α- and β-ionone (157, 169), and epoxy derivatives (165, 169), were identified. Many aliphatic alcohols (e.g. 7, 10, 14, 78, 100,127, etc.), acids (e.g. 6, 11, 15, 23, 29, 47, 53, etc.), and aldehydes (e.g. 5, 9, 18, 24, 27, 33, etc.), related to fatty acid oxidation, were found too. Caffeine (215), indicated in Fig. 1, was the most abundant constituent.

The presence of 2-butoxyethanol (34), an additive used in several petroleum-derived products (e.g. paints, inks, acrylic resin formulations, asphalt release agents, oil spill dispersants, etc.), was confirmed and has also been highlighted in previous work; however, because its presence was found in several samples, it was considered as a specific product of the mate plant 8. It is noteworthy, though, that several other contaminants were found in the present mate sample; in particular, a significant presence of hydrocarbon isomers, with high similarity matches (>95%), was determined (many alkanes were not assigned due to the lack of related standards): a classical petrochemical-like hydrocarbon GC×GC band is present in Fig. 1. The contemporary presence of several PAHs (identified in Table 1 as: 106, 132, 150, 159, 163, 200) supports the hypothesis that the sample had been contaminated by mineral oil. In the GC-qMS application, only 106 and 132 amongst the PAHs and 152, 170, and 183 amongst the alkanes (eighteen were identified in the GC×GC run), were identified with a minimum 90% similarity, even if a characteristic hump, usually related to mineral oil contamination, was present. The main concern of this kind of contamination is related to the petroleum origin, which is associated to a high amount of PAHs and in particular alkylated PAHs, which have a demonstrated carcinogenic activity 15. Considering the low volatility of PAHs and the unsuitability of HS-SPME as PAH extraction method, the presence of these compounds (the lighter ones) in the mate sample analyzed probably means that very high amounts could be present. The drying process can be another important source of PAHs (in particular parent PAHs), if combustion smoke enters directly in contact with the matrix. Several epidemiological studies 16–18 have reported a correlation between mate consumption and cancer; furthermore, high quantities of PAH metabolites in mate consumers, as well as intact PAHs in mate samples, have been reported 19–22. In particular, Ziegenhals et al. found that, amongst different types of tea, mate was the most contaminated one 21, while Zuin et al., using an extraction method more suitable to PAH analysis (stir bar sorptive extraction followed by HPLC with fluorometric detection), found all the parent PAHs indicated as priority pollutants by the US Environmental Protection Agency, in 11 commercial samples of Brazilian mate tea. Kamangar et al. 22 determined very high quantities of 21 PAHs in eight samples of commercial mate analyzed by using pressurized liquid extraction followed by GC-MS; benzo[a]pyrene ranged from 8 to 53 μg/kg. A mate infusion, derived from two of these samples, was prepared in the traditional way and the passage from the raw product to the final beverage was investigated. Considering the per capita mate consumption in South America, the intake of PAHs was comparable to a packet of cigarettes a day. Although further investigation on a higher number of samples would be necessary to confirm the presence of mineral oil (and related PAHs) in the matrix analyzed, the literature information confirms this suspicion, allowing the supposition that a systemic source of contamination is probably present in mate processing and that the natural origin of 2-butoxyethanol should be further investigated. Finally, many plasticizers, such as phthalates (162, 181, 216, 224) and N-butyl-benzenesulfamide (205) were tentatively identified (only 216 and 224 were identified in GC-qMS). Although phthalates are known to be ubiquitous contaminants and no particular analytical precautions were applied, their absence in the blank analysis can lead to the (sufficiently reliable) conclusion that they truly derive from the mate sample.

4 Concluding remarks

The GC×GC-qMS method, developed in the present research, proved to be a very suitable alternative in the application described, with a great improvement achieved in terms of separation and number of identified peaks. Although a more than tripled number of unknowns was identified, several co-elutions still occur. In fact, a series of mass spectra were characterized by insufficient spectral quality denoting the presence of interfering analytes. It must be observed, though, that this factor could be linked to the lack of the relative mass spectra in the MS library employed.

In particular, the authors feel that more extensive research should be directed to risk assessment studies related to mate production, considering all steps from harvesting to commercialization. As was seen, several harmful contaminants were found in the sample, some of which previously reported. The presence of high quantities of aliphatic hydrocarbons and several light PAHs would lead to a strong hypothesis of mineral oil contamination.


The Project was funded by the Italian Ministry for the University and Research (MUR) with a PNR 2005–2007 Project n. RBIP06SXMR ‘Sviluppo di metodologie innovative per l'analisi di prodotti agroalimentari’. The authors gratefully thank Shimadzu and Sigma-Aldrich/Supelco Corporations for their continuous support. R. Assis Jacques thanks the CNPq (Conselho Nacional de Desenvolvimento. Científico e tecnológico) for the scholarship.

The authors have declared no conlict of interest.