First Attempt of Odorant Quantitation Using Gas Chromatography

Chemical Composition of French Mimosa Absolute Oil. Rodolphe Perriot , Katharina Breme , Uwe J. Meierhenrich , Elise Carenini , Georges Ferrando and ...
0 downloads 0 Views 88KB Size
Anal. Chem. 1999, 71, 5391-5397

First Attempt of Odorant Quantitation Using Gas Chromatography-Olfactometry Philippe Pollien, Laurent B. Fay, Marcel Baumgartner, and Alain Chaintreau*,†

Nestle´ Research Centre, NESTEC Ltd., P.O. Box 44, Vers-chez-les-Blanc, 1000 Lausanne 26, Switzerland

An aroma compound was quantitated for the first time by GC-olfactometry (GC-O) on the basis of the detection frequency of odorants by a panel of 8-12 persons. The method was previously optimized regarding the coincidence of olfactometric peak apexes and the repeatability of peak height and area over 4 months. The number of required calibration points and the confidence interval of the curve were investigated. This technique was then tested by quantifying a model solution of 1-octen-3-one. The standard addition method was found to be unsuitable in this context, but external calibration gave excellent results in the ppt range. GC-O was then challenged using one of the most sensitive and selective methods, GC/MS, to quantitate 1-octen-3-one in coffee, a complex aroma. Results showed performances comparable to GC/MS/MS for this odorant, or even better as the latter required 75500 times more sample to perform the quantitation. However, at such a low concentration, overestimation cannot be excluded with either technique because of possible coelution of odorants or isobaric ions, respectively. These results show that GC-olfactometry can compete with the most sensitive and selective techniques, such as MS, for determination of extremely intense odorants, because little sample preparation is required and there is no need for the synthesis of labeled compounds. Gas chromatography-olfactometry (GC-O) is a unique analytical technique which associates the resolution power of capillary GC with the selectivity and sensitivity of the human nose. This latter sometimes detects odorants that occur in extremely low amounts, much below the detection limit of any physical system (e.g., 2,4,6-trichloroanisole1). GC-O was first rationalized by Acree,2 who proposed the CHARM analysis. This consists of the GC injection of increasing dilutions of the aroma extract until no odor is perceived at the sniffing port. From individual panelist’s results, a computerized data treatment builds a global aromagram and assigns a greater peak height to odorants that are perceived at highest dilution. A similar and simplified procedure (AEDA) was later proposed by † Present address: Firmenich SA, 1 Route des Jeunes, CP 239, 1211 Gene ` ve 8, Switzerland. (E-mail) [email protected]. (1) Spadone, J. C.; Takeoka, G.; Liardon, R. J. Agric. Food Chem. 1990, 38, 226-233. (2) Acree, T. E.; Barnard, J.; Cunningham, D. G. In Analysis of volatiles; Schreier, P., Ed.; de Gruyter: New York, 1984; pp 251-267.

10.1021/ac990367q CCC: $18.00 Published on Web 10/29/1999

© 1999 American Chemical Society

Grosch.3 CHARM and AEDA, called “dilution techniques”, have been widely used to determine aroma impact compounds of many foods.4-6 Another technique, OSME, is based on the continuous recording of the odor intensity that is perceived at the sniffing port.7 The panelist moves the cursor of a variable resistor as a function of the intensity of perception. The technique was improved by Guichard et al.8 Only a few number of applications have been published. At first sight, the objective of this work does not seem feasible or at least looks very risky because of the great variability in sensory detection. However, recent papers support the quantitative capabilities of olfactometric methods. The significance of the difference between two CHARM peak areas has been evaluated by computing their least significant difference (LSD).9 Using a replication of an OSME analysis with 10 different panelists, the variability of the peak height was calculated.10 Using results of the same paper, two panels of five people were simulated and standard deviations were calculated between the groups from their peak intensity averages.11 To improve the reliability of aromagrams, two research groups proposed to calculate peak detection frequencies from odors perceived by a panel of 6-10 assessors.12,13 Such aromagrams were shown to be repeatable by the same panel and reproducible by two different panels without training. An average LSD was calculated to compare detection frequencies of a given odorant occurring at two different concentrations.11 Peak heights and peak areas were called nasal impact frequency (NIF) and surface of nasal impact frequency (SNIF), respectively. (3) Ullrich, F.; Grosch, W. Z. Lebensm. Unters. Forsch. 1987, 184, 277-282. (4) Grosch, W. Flavour Fragrance J. 1994, 9, 147-158. (5) Blank, I. In Techniques for analyzing food aroma; Marsili, R., Ed.; M. Dekker: New York, 1996; pp 293-329. (6) Feng, Y. G.; Acree, T. E. Foods Food Ingredients J. Jpn 1999, 179, 57-66. (7) McDaniel, M. R.; Miranda-Lopez, R.; Watson, B. T.; Michaels, N. J.; Libbey, L. M. In Flavors and off-flavors.; Charalambous, G., Ed.; Elsevier Science B.V.: Amsterdam, 1989; pp 23-36. (8) Guichard, H.; Guichard, E.; Langlois, D.; Issanchou, S.; Abbott, N. Z. Lebensm. Unters. Forsch. 1995, 201, 344-350. (9) Acree, T.; Barnard, J. In Trends in Flavour Research; Maarse, H., Ed.; Elsevier Science B.V.: Amsterdam, 1994; pp 211-220. (10) Guichard, H.; Guichard, E.; Langlois, D.; Issanchou, S.; Abbott, N. Z. Lebensm. Unters. Forsch. 1995, 201, 344-350. (11) Pollien, P.; Ott, A.; Montigon, F.; Baumgartner, M.; Munoz-Box, R.; Chaintreau, A. J. Agric. Food Chem. 1997, 45, 2630-2637. (12) Ott, A., Pollien, P., Montigon, F., and Chaintreau, A. Improved headspaceGC-sniffing technique: screening of aroma impact flavorings. Fourth International Symposium on Hyphenated Techniques in Chromatography. Bruge (B). February 7-9, 1996. (13) Van Ruth, S. M.; Roozen, J. P.; Cozijnsen, J. L. Food Chem. 1996, 56, 343346.

Analytical Chemistry, Vol. 71, No. 23, December 1, 1999 5391

Table 1. SNIF and Concentration Variations of Various Impact Odorants between Milk and Yogurt Aromas 17 compound

∆(SNIF)

signifa

∆(concn) mg/L/%

acetaldehyde dimethyl sulfide 2,3-butanedione 2,3-pentanedione benzothiazole

+7006 -1300 +5492 +3549 +1707

+ + + -

+16,6/+∞ -0.005/-20% + 1.35/+∞ + 0.13/+∞ +0.09/+24%

a + and - indicate a significant or a nonsignificant SNIF variation, respectively, at 95% confidence, using the same panel.

Because of that, this detection frequency method was entitled GC-“SNIF”. Van Ruth et al.13 and Pollien et al.11 observed a linear and a sigmoid relationship, respectively, between the detection frequency and the logarithm of the odorant concentration. Since the central part of a sigmoid may be considered to be linear in a first approximation, both frequency approaches are similar, as recently discussed.14 The sigmoid, called a “psychometric law”, represents the total number of panelists detecting a given odor as a function of its concentration. This curve can be linearized by transforming its points into probits according to Bliss’ transformation.15 Probits correspond to the reduced value of the normal distribution law:

probit ) 5 +

log(C) - log(CT) SD

(1)

where C is the concentration of the odorant, CT is the mean threshold concentration of the odorant, and SD is the standard deviation of panelists around CT. In practice, CT and SD are unknown when the GC-O has just been performed. Therefore, they must be calculated in the following way: For a given NIF value, x, (between 0 and 1), the probit of x is z + 5, where z is such that

1 2xπ



z

-∞

e-y /2 dy ) x 2

(2)

The values for z can be found in appropriate software (see the Experimental Section) or tables. Since the principle of the GC-SNIF method was published in detail,11 a first confirmation of its validity for quantitative comparisons was obtained. On the basis of peak areas (SNIFs) of milk and yogurt aromagrams, odorants generated by fermentation were differentiated from those originating from the milk.16 Products resulting from bacterial metabolism were assumed to only give increased SNIF values. For peaks above the FID detection limit, GC-O results were confirmed by quantitation of corresponding volatiles before and after fermentation17 (Table 1). Variations of SNIFs and FID areas were in the same direction, even when SNIF (14) Chaintreau, A.; Pollien, P. Actes des 17ie`mes Journe´es Internationales des Huiles Essentielles. Riv. Ital. EPPOS 1999, 258-266 (Numero Speciale Gennais 99). (15) Bliss, C. I. Statistics in biology.; McGraw-Hill: London, 1967. (16) Ott, A.; Fay, L. B.; Chaintreau, A. J. Agric. Food Chem. 1997, 45, 850-858. (17) Ott, A.; Baumgartner, M.; Chaintreau, A. J. Agric. Food Chem. 1999, 47, 2379-2385.

5392

Analytical Chemistry, Vol. 71, No. 23, December 1, 1999

differences were less than their least significant difference (LSD ) 2000 SNIF at 95% confidence level11). In the course of method development,11 the linear relationship of probits versus the concentration logarithm of 2-(Z)-hexenal was established. This suggested that such a curve could be used to calibrate the olfactometric signal and, therefore, to determine the concentration of a given constituent of an aroma. This paper aims to check this possibility in the case of a solution model and a real aroma. EXPERIMENTAL SECTION Material. Propanone and 1-octen-3-one were purchased from Merck (Les Acacias, CH) and Oxford Chemicals (Hartlepool, U.K.), respectively. The coffee blend was the same as previously used.18 The roast and ground beans were packed (5 g) into Nespresso capsules and brewed with 100 mL of hot water in a Nespresso coffee machine, model C-200-plus (Belmont/Lausanne, CH), to ensure good reproducibility. The water was a 2:1 mixture of Vittel “Bonne Source” and distilled water. The beverage was immediately chilled to 20 °C after brewing. Each GC-O run was performed with a freshly prepared coffee cup. Standard Solutions. A stock solution containing 1-octen-3one (100 mg/L) was prepared in propanone and diluted 10 times in the same solvent. Standard solutions were obtained by diluting 4 µL of the stock solution or 2.5, 5, or 10 µL of the daughter solution in 100 mL of the same water mixture as used for coffee brewing. The test solution to be quantified was made by diluting 8 µL of the stock solution in 1 L of water. The calibration curve of the standard addition method (Figure 6) was prepared with the same stock and daughter solutions as above. ATD-GC-Olfactometry. For all sniffing runs, the coffee aroma from 5 mL of coffee brew was collected by the static-andtrapped headspace technique (S&T-HS) as previously reported.11,18 To increase the sensitivity of MS determinations of 1-octen-3-one, the usual 160-mL headspace cell was replaced by a 4-L cell made of stainless steel, based on the same principle19 and using 10 mL of coffee brew. Thermal trap desorptions were achieved as reported elsewhere,11 with a 10:1 split between the thermal desorber and the GC column. This latter was a DBWax (J&W Scientific, MSPFriedli, Koeniz, CH), 60-m length, 0.53-mm inner diameter, and 1.0-µm phase thickness. The oven was kept at 20 °C for 5 min and then increased to 220 °C at 4°/min. The column outlet flow was split (1:1) between the sniffing port and the FID. A total of 1000 SNIF units correspond to a NIF of 100%, over a duration of 1 s.11 Calculations. The adjustment of retention times to that of the reference was made after transferring the raw ASCII data (matrix: time × intensities) from the GC software to EXCEL97 (Microsoft Co., Seattle, WA). Aromagram reconstructions and calibration curves were obtained with the same program (e.g., Figure 5). Probit values were calculated using the following formulas and EXCEL functions: (18) Pollien, P., Krebs, Y., Chaintreau, A. Assoc. Sci. Int. Cafe´ 1998, 191-196 (Paris). (19) Chaintreau, A.; Grade, A.; Munoz-Box, R. Anal. Chem. 1995, 34, 33003304.

probit of NIFs ) NORSMINV(NIF%) + 5 end time

probits of SNIFs )



Table 2. Standard Deviation (min) of the Retention Time of the Reference Peak

[probit(NIF)t ∂t]

start time

8 first panelists 12 panelists

solution

coffee

0.017 0.027

0.005 0.018

with ∂t the acquisition periodicity of the computer (0.1 s), from the start time to the end time of the GC-O peak, and (NIF)t the instant NIF value (%) at time t.

NIF (%) ) 100 × NORMSDIST(probit-5) The evaluation of confidence intervals was performed using S-PLUS, version 4.5 (Math Soft, Seattle, WA). The air-to-water partition coefficient of 1-octen-3-one was estimated to 29.9 × 10-3 (concentrations in g/L) using the Properties Plus software, version 9.3-1 (Aspen Technology, Cambridge, MA). Mass Spectrometry. The GC/MS experiments were carried out on a Finnigan SSQ-7000 mass spectrometer connected to a HP-5890 gas chromatograph (Finnigan MAT, San Jose, CA) equipped with a DB-1701 capillary column (J&W Scientific), 30 m × 0.32 mm i.d., film thickness 0.25 µm. Helium was used as carrier gas at a pressure of 10 psi. The samples were injected using an ATD400 thermal desorber under the same conditions as used for GC-O experiments, without splitting the column inlet flow. The oven program was 35 °C for 2 min, then 4 °C/min up to 200 °C, then 30 °C/min up to 250 °C, and hold for 20 min. The transfer line was held at 250 °C and the source at 185 °C. Negative chemical ionization was performed using ammonia as reagent gas. The [M - H]- ion at m/z 125 for 1-octen-3-one was recorded at 200 eV with a dwell time of 0.25 s. GC/MS/MS experiments were carried out on a Finnigan TSQ700 mass spectrometer connected to a HP-5890 gas chromatograph equipped with the same DB-1701 capillary column. Helium was used as carrier gas at a pressure of 10 psi. The samples were injected using the same thermal desorption conditions as used for GC-O experiments, without spliting the column inlet flow. The oven program was 35 °C for 2 min, then 4 °C/min up to 200 °C, then 30 °C/min up to 250 °C, and hold for 20 min. The transfer line was held at 240 °C and the source at 180 °C. Negative chemical ionization was performed using ammonia as reagent gas to generate [M - H]- ions with an electron energy of 200 eV. These ions were fragmented by collision-induced dissociation, and the daughter ion at m/z 69 was monitored. A collision energy of 20 eV in the laboratory frame was used and the pressure of the collision gas argon was set to 1 mTorr. RESULTS AND DISCUSSION Any quantitation requires optimization of the analytical procedure for the target compound to ensure the accuracy of measurements. The mean variability of the GC-SNIF method was previously evaluated for peaks of a whole aromagram.11 In the case of quantitation of a given odorant, a large confidence interval was suspected from our preliminary results.14 Therefore, the influence of four parameters was further investigated in this work: (1) the repeatability of retention times; (2) the repeatability of NIFs and SNIFs; (3) the number of calibration points; and (4) the difference between NIFs and SNIFs. Panelists who participated

Figure 1. Linearity of probit(NIF) versus log(C) with (×) and without (+) adjustment of retention times (eight panelists).

in the following evaluation were considered to be trained due to their participation in previous GC-SNIF analyses. Among them, only people giving reproducible detections were selected, without taking into account their olfactive sensitivity, as this method is based on the difference of detection thresholds between assessors. Retention Time Repeatability. A calibration curve may require several days to construct because repetitions of GC-O runs for the different calibration points and the unknown sample must be performed and small variations of the retention time (RT) may occur. This affects the coincidence of odor perception of panelists from one aromagram to another. After these traces are summed, the peak height (NIF) can be altered if square signals are shifted. To overcome these variations, the carrier gas flow exiting the chromatographic column was split to monitor an FID trace at the same time as the sniffing. One peak (an odorless artifact generated by heating of the Tenax) was chosen as a retention time reference. The GC-O was repeated by 12 panelists using a 1-octen-3-one solution and using the elution region of the same odorant present in a coffee brew. Calculating the standard deviation of the RT’s reference peak with the same 8 first panelists showed a lower value than with the 12 people in both cases (Table 2). This confirmed a drift of RTs over the experiment’s duration. Consequently, times of all eight individual aromagrams were corrected to make the apex of the reference peak’s RT in each GC-O run coincide. The Figure 1 shows the improvement of the calibration curve linearity by applying this adjustment. NIF and SNIF Repeatability of a Given Odorant Solution. To evaluate the NIF and SNIF repeatability of the compound to be quantified (1-octen-3-one naturally occurring in coffee aroma), the GC-O procedure was duplicated after an interval of 4 months using the same instrument, under the same operating conditions, with the same panelists. In both cases, a NIF variation corresponding to (1 panelist was observed (Table 3). Consequently, Analytical Chemistry, Vol. 71, No. 23, December 1, 1999

5393

Figure 2. Calibration curves of 1-octen-3 one using three and four different concentrations. Dotted lines represent 95% prediction intervals. Table 3. Repeatability of the NIFs and SNIFs at an Interval of 4 Months, Using the Same Panel of 8 and 12 Persons 8 panelists

12 panelists

NIF (%)

SNIF

NIF (%)

SNIF

first GC-O second GC-O NIF contribution of 1 panelist

37.5 50 12.5

0.4709 0.6880

41.7 50 8.3

0.6522 0.6696

NIF RSD (%)

20.2

12.9

1.9

a

26.5

Peak apexes of the RT standard have been adjusted.

the relative standard deviation decreased with a greater number of panelists. These results confirm, and extend to a longer time lapse (4 months), previous reproducibility measurements performed over 2 weeks.11 Number of Calibration Points. Usually, the confidence interval around a calibration curve is lowered by increasing the number of calibration points. In the present case, this confidence interval was not improved by using four instead of three standard solutions (Figure 2), because the latter already exhibited a very high correlation coefficient that could not be improved. Conversely, the calibration with four points was worse, due to the “error” of one panelist in the olfaction of the 500 ng/L solution. These intervals are only indicative, as the validity of statistics derived from three or four calibration points is limited. To better take into account that NIFs may only have discrete values, another evaluation of the calibration curve confidence interval was considered. According to repeatability tests of this work and the previous one,11 each calibration point may be determined at (1 sniffer contribution. This possible “error” ((100/n %) has a lower impact on the NIF values with an increasing number of panelists (n). For instance, considering that for each of the three NIF values the three-point calibration curve was determined with an uncertainty of (1 panelist, corresponding to (12.5 or (8.3 NIF% for a 8- or 12-member panel, the envelope 5394 Analytical Chemistry, Vol. 71, No. 23, December 1, 1999

Figure 3. Confidence interval of the calibration curve (probit(NIF) versus lg(C)) for 8 and 12 panelists.

of the straight line of probit(NIF) as a function of lg(C) can be determined (Figure 3). The confidence interval was reduced in the central part of the curve and enlarged close to its ends. It seems to better represent the reality, hidden by the probit transformation, than Figure 2, which is a pure statistical treatment of the regression after transformation of NIFs by Bliss’ procedure.15 This way of evaluating the confidence interval was not applied to SNIF as it would require determining the variability of the GC-O peakwidth. Use NIF or SNIF? Measuring peak areas (SNIFs) was expected to overcome the repeatability of retention times, which affects the accuracy of NIFs mentioned above. However, the coefficient of determination of the calibration curve was noticeably lower than that of the corresponding NIF calibration (Figure 4). When 1-octen-3-one peaks of the mean aromagram were compared, peak shapes differed from one standard to another (Figure 5), and peak widths at the baseline did not increase regularly with concentration. Some coalescence with erratic peaks of the background might occur and dramatically influence the area measurement. Consequently, the SNIF measurement was not used

Figure 6. Quantitation of 1-octen-3-one in coffee using a standard addition method (eight panelists).

Figure 4. Calibration curve of probit(SNIF) versus lg(C).

Figure 5. Peak shape comparison of the mean aromagram between the four standard solutions of 1-octen-3-one (12 panelists).

in this study to quantify 1-octen-3-one. This single observation about SNIFs cannot be considered as a general rule, as our previous results showed similar mean repeatabilities when using NIF or SNIF for a whole aromagram of a complex mixture of odorants.11 However, to quantify a specific aroma compound, such a cause of variability should be overcome. This problem has not yet been solved. GC-SNIF Quantitation of a Solution Model. Three main quantitation techniques are applicable to headspace-gas chromatography: internal standardization, external standardization, and standard addition.20 The first two are less attractive as they require preparation of standards in the same matrix as the sample, but free of the target compound. If this is not achievable, air-to-sample partition coefficients must be measured to deduce the concentration in the sample from that in its headspace. The choice of a quantitation method would be simplified if an extract (instead of the headspace) was injected into the GC because the target compound recovery in the extract could be assumed to be quantitative and partition problems would be neglected. However, in such a case, losses of analyte are frequent21 and give rise to the same difficulties as with HS sampling. The standard addition looks more attractive as it uses the sample matrix and the target compound themselves as the solvent and the standard, respectively. (20) Chaintreau, A. In Encyclopedia of Analytical Chemistry; Meyers, R. A., Ed.; Wiley: Chichester, in press. (21) Masanetz, C.; Blank, I.; Grosch, W. Flavour Fragrance J. 1995, 10, 9-14.

Standard Addition. The insufficient coefficient of determination and the low slope showed in Figure 6 did not allow a satisfactory extrapolation to a probit value of 0. Coming back to the meaning of such an operation in the case of the NIF versus lg(C), it would be equivalent to extrapolate the sigmoid down to a NIF value of zero. This has obviously no sense as the intercept occurs when lg(C) tends to -∞. External Calibration. An aqueous solution model containing 800 ng/L 1-octene-3-one was used to test the method. According to the three- and four-point calibration curves in Figure 2, its NIF value (58.3%) corresponded to 797 and 702 ng/L, respectively, without taking into account the NIF variability of the test solution. Their 95% confidence intervals were 231-2743 and 169-2911 ng/ L, respectively (computed using the classical approximation for the inverse prediction model22). Using the model of Figure 3, intervals were thinner: 570-1539 and 669-1287 ng/L using a 8or 12-member panel, respectively. Despite such a broad interval, the concentration found for the test solution was surprisingly close to its real content. GC-SNIF Quantitation in a Complex Aroma. To our knowledge, 1-octen-3-one has never been positively identified in coffee. Its occurrence has been assumed on the basis of its odor and retention index after GC-O analyses.18,23,24 This is presumably due to its extremely low concentration and difficulties of isolation from a mixture of more than 800 other volatiles.25 Consequently its concentration in coffee has never been reported. Using the same external calibration curve as used for the model solution (Figure 2, left), 520 ng/L 1-octen-3-one was found (12 panelists). This result differs from that proposed from preliminary experiments14 because of the optimization of the standard measurement made in the present paper (the NIF value of 1-octen-3-one itself in coffee remained unchanged). This concentration does not take into account the difference of partition of 1-octen-3-one between air and water for standard solutions and between air and coffee for the brew. Measurement of these partition coefficients is in progress to make necessary corrections. (22) Neter, J.; Wasserman, W.; Kutner, M. H. Applied Linear Statistical Models, 3rd ed.; Irwin: Homewood, IL, 1990. (23) Holscher, W.; Vitzthum, O. G.; Steinhart, H. Cafe´ Cacao The´ 1990, 34, 205212. (24) Semmelroch, P.; Grosch, W. Lebensm.-Wiss. u.-Technol. 1999, 28, 310313. (25) BACIS.; VCF96: database of volatile compounds in food. TNO nutrition and food research, 1996-7.

Analytical Chemistry, Vol. 71, No. 23, December 1, 1999

5395

Figure 7. S&T-HS-GC/MS monitoring in CI mode of m/z 125 in the coffee headspace alone (A), and spiked with 500 ng/L 1-octen3-one (B).

To challenge quantitative performance of GC-O, these results were compared with those of GC/MS, one of the most selective and sensitive quantitation techniques. Quantitation Verification by GC/MS. GC/MS and GC/MS/ MS were used to confirm the presence of 1-octen-3-one in the roast coffee beans previously detected and quantified by GC-O. First, experiments were carried out using a 160-mL internal volume headspace cell, previously described for S&T-HS. No peak corresponding to the target compound was found. To increase the sensitivity, a bigger cell was employed (4 L of internal headspace volume). Because of the need for high selectivity, negative chemical ionization was chosen, leading to formation of the [M - H]- ion base peak (m/z 125) of the 1-octen-3-one spectrum. Indeed this even-electron ion is efficiently produced only from compounds containing electronegative elements, and therefore, this confers some selectivity to the analysis of such a complex mixture as the headspace of coffee brew. Unfortunately, this selectivity in the ionization associated upstream with the chromatographic resolution and downstream with the selected ion monitoring of the ion at m/z 125 ([M H]-) did not permit unambiguous detection of the compound by GC/MS. Only a small shoulder on a much bigger peak was detected at the retention time of 1-octen-3-one (Figure 7). The selectivity was increased one step further using GC/MS/ MS, to monitor the daughter ion at m/z 69 produced by collisioninduced dissociation from the parent [M-H]- ion, obtained after negative chemical ionization. The resulting chromatograms (Figure 8) showed the presence of 1-octen-3-one in the headspace of coffee brew. This compound was estimated at about 240 ng/L using standard addition experiments. Considering the low level of detection of this compound and, therefore, the errors related to this measurement, we can consider this value in agreement with the 500 ng/L determined by GC5396

Analytical Chemistry, Vol. 71, No. 23, December 1, 1999

Figure 8. S&T-HS-GC/MS/MS monitoring in CI mode of the daughter ion at m/z 69 from m/z 125 in the coffee headspace alone (A) and spiked with 500 ng/L 1-octen-3-one (B).

O. In both approaches, an overestimation of the quantity due to a coeluting peak remains possible: (1) In MS/MS, the baseline is not clean enough, despite the expected high selectivity of this technique. Even the use of a labeled standard could not overcome this difficulty. (2) In GC-O, 1-octen-3-one elutes close to two other potent coffee odorants, which might partially overlap.18 It must be noted that the chromatogram of Figure 8 was obtained without any split of the carrier gas flow and a 25 times greater headspace volume. According to a previously reported equation19 giving the headspace concentration in the sampling cell, from that initially present in the solution, this size increase multiplied the amount of 1-octen-3-one in the headspace by a factor m′g/mg:

(

)

m′g V′g 1 + kv(Vg/Vl) ) mg Vg 1 + kv(V′g/Vl)

(3)

where mg, m′g are the masses of 1-octen-3-one in the headspace of two cells having different inner sample and gas volumes, Vg, V′g and Vl, V′l, respectively). Using an estimated value of 29.9 × 10-3 for the air-to-water partition coefficient of 1-octen-3-one, m′g/mg was about 3.8. The real m′g/mg value with coffee must be greater (3.8-25) because of a smaller partition coefficient (e.g., between 0 and 29.9 × 10-3) due to complexation in the matrix. This indicates that the sensitivity of GC-SNIF was 75-500 times higher than that of GC/ MS, for a quantitative procedure. CONCLUSION The quantitation of an odorant appears to be feasible using the GC-SNIF method and NIF measurements were repeatable at up to 4 months. As the technique is time-consuming, it can be

applied when the concentration is below the sensitivity of physical detectors or when the odorant coelutes in an inseparable mixture of less odorant volatiles. However, little sample preparation is required. This may compensate for GC-O’s duration, as a classical trace quantitation requires, e.g., labeled standards that must be previously synthesized. The quantitation seems satisfactory for a simple solution of volatiles. A rougher estimation is obtained for a very complex aroma. However, very potent odorants, such as 1-octen-3-one, that occur in trace amounts (ppt range) also appear to be hardly evaluated by sophisticated analytical methods. The present MS/ MS quantitation required a 75-500 times increase of the sample size, for a limited selectivity.

Therefore GC-SNIF does not exhibit more drawbacks than classical quantitation methods for substances occurring in trace amounts, but the target odorants must have unimodal detection thresholds as a function of the logarithm of their concentration. ACKNOWLEDGMENT We gratefully acknowledge Dr. D. Pre`tre for calculation of the air-to-water partition coefficient, and Dr E. Prior for reviewing this paper. Received for review April 7, 1999. Accepted September 2, 1999. AC990367Q

Analytical Chemistry, Vol. 71, No. 23, December 1, 1999

5397