Enhancing Sensitivity in Headspace-Mass Spectrometric

Mar 6, 2007 - Environmental Forensic science. Jennifer Suggs , Diana Love ... Impressions (Barefoot, Ears, Lips). Andy Becue , C Champod , Pierre Marg...
0 downloads 0 Views 159KB Size
Download PDF
Anal. Chem. 2007, 79, 2997-3002

Enhancing Sensitivity in Headspace-Mass Spectrometric Determination of BTEX in Drinking Water Antonio Serrano, Mercedes Gallego,* and Manuel Silva

Department of Analytical Chemistry, Campus of Rabanales, University of Co´ rdoba, E-14071, Co´ rdoba, Spain

A way to extract useful chemical information from the volatile profile provided by a headspace-mass spectrometer (HS-MS) is developed in order to improve sensitivity in HS-MS analysis. The methodology is based on the selection of a narrow window in the volatile profile where the signal-to-noise ratio was maximal by combining the data acquisition time and scan rate. To test this approach, benzene, toluene, ethylbenzene, and p-xylene (BTEX) as well as their mixtures were quantified in drinking waters. Individual hydrocarbons were determined between 1 and 30 µg/L (mean RSD, 4.0% for 10 µg/L) while mixtures were quantified at a microgram per liter level by using the partial least-squares multivariate algorithm with a relative standard prediction error of under 3.5%. These results indicate that the method proposed is useful as a sensitive and selective tool for the determination of BTEX and surpasses other reported HS-MS alternatives. In addition, the proposed methodology can be extended to others that insert analytes from a sample directly into a MS, such as membrane introduction mass spectrometry among others. Current research and development activities, in analytical sciences in general and measurement process in particular, are aimed at improving existing methods and meeting the new chemical information demands made by clients of analytical laboratories. The demands for chemical information in a variety of fields, such as environmental studies, food, health, and industry, have increased dramatically in recent decades. The process used in routine analytical laboratories involves rapidly obtaining information on isolated compounds according to European guidelines. In addition, one of the main problems for routine laboratories of environmental analysis or quality control is the high number of samples that must be analyzed.1 Sample treatment is the most costly and time-consuming part of the process and one of the most frequent sources of errors. In this sense, the development of simple and fast analysis methodologies is currently of great interest since they could solve the above-mentioned problems.2 * Corresponding author. E-mail: [email protected]. (1) Valca´rcel, M.; Ca´rdenas, S. Trends Anal. Chem. 2005, 24, 67-74. (2) Valca´rcel, M. In Analytical Chemistry, 2nd ed.; Kellner, R., Mermet, J. M., Otto, M., Valca´rcel, M., Widmer H. M., Eds.; Wiley-VCH: Weinheim, 2004; Chapter 4, pp 37-60. 10.1021/ac070044r CCC: $37.00 Published on Web 03/06/2007

© 2007 American Chemical Society

Direct sampling mass spectrometry (MS) methods have been developed for the analysis of environmental samples; these methods insert analytes from a sample directly into a mass spectrometer using a simple interface with minimal sample preparation and no prior chromatographic separation.3 Although there are four major types of inlets for direct sampling, the most popular is membrane introduction (MIMS); traditionally a semipermeable membrane, usually silicone rubber, is placed between the liquid or gaseous sample in the vacuum of the mass spectrometer. Analytes of interest pass through the membrane into the mass spectrometer in a three-step process called pervaporation. A detailed description of this process is beyond the scope of this article, but excellent references referring to this subject are provided.4,5 The recent development of a methodology based on the direct coupling of a headspace sampler with a mass spectrometry detector (HS-MS) has allowed a drastic reduction in analysis time, increasing the sample throughput.6 In this sense, the procedure is simple and involves, after the extraction by HS technique, the introduction of only the volatile compounds into the ionization chamber of the MS, avoiding chromatographic separation. The analytical signal generated is characteristic of the compounds present in the sample analyzed. Generally, data can be obtained by using the mass spectrum that represents the sum of intensities of all the ions detected during the data acquisition time.7 To date, a growing number of applications for this methodology have been reported that are related to quality control in foods such as olive oils8-12 or wines13-15 and hydrocarbon (3) Wise, M. B.; Guerin, M. R. Anal. Chem. 1997, 69, 26A-32A. (4) Bauer, S. Trends Anal. Chem. 1995, 14, 202-213. (5) Short, R. T.; Toler, S. K.; Kibelka, G. P. G.; Rueda Roa, D. T.; Bell, R. J.; Byrne, R. H. Trends Anal. Chem. 2006, 25, 637-646. (6) Pe´rez-Pavo´n, J. L.; del Nogal-Sa´nchez, M.; Garcı´a-Pinto, C.; Ferna´ndezLaespada, M. E.; Moreno-Cordero, B.; Guerrero-Pen ˜a, A. Trends Anal. Chem. 2006, 25, 257-266. (7) Martı´, M. P.; Boque´, R.; Busto, O.; Guasch, J. Trends Anal. Chem. 2005, 24, 57-66. (8) Pen ˜a, F.; Ca´rdenas, S.; Gallego, M.; Valca´rcel, M. J. Chromatogr., A 2005, 1074, 215-221. (9) Pen ˜a, F.; Ca´rdenas, S.; Gallego, M.; Valca´rcel, M. Anal. Chim. Acta 2004, 526, 77-82. (10) Pen ˜a, F.; Ca´rdenas, S.; Gallego, M.; Valca´rcel, M. J. Am. Oil Chem. Soc. 2003, 80, 613-618. (11) Marcos-Lorenzo, I.; Pe´rez-Pavo´n, J. L.; Ferna´ndez-Laespada, M. E.; Garcı´aPinto, C.; Moreno-Cordero, B.; Henriques, L. R.; Peres, M. F.; Simoes, M. P.; Lopes, P. S. Anal. Bioanal. Chem. 2002, 374, 1205-1211. (12) Marcos-Lorenzo, I.; Pe´rez-Pavo´n, J. L.; Ferna´ndez-Laespada, M. E.; Garcı´aPinto, C.; Moreno-Cordero, B. J. Chromatogr., A 2002, 945, 221-230. (13) Martı´, M. P.; Pino, J.; Boque, R.; Busto, O.; Guasch, J. Anal. Bioanal. Chem. 2005, 382, 440-443.

Analytical Chemistry, Vol. 79, No. 7, April 1, 2007 2997

contamination in soils and waters.16-19 Classification or regression approaches were used for screening or quantification purposes according to the analytical problem addressed. Similarly, when using MIMS, the positive identification of individual analytes in the complex matrix that contains compounds with overlapping molecular ions (e.g., benzene, toluene, ethylbenzene, and p-xylene, BTEX)4 requires, as in HS-MS, chemometric approaches since both methodologies use the same data acquisition in the MS. In spite of the number of applications reported to date, none of them has dealt with improving the sensitivity of the HS-MS methodology. This analytical property becomes crucial in environmental analysis (water, soil, etc.) where the maximum contaminant level (MCL) of a toxic compound is set at a very low level by authorities (e.g., benzene in drinking water, MCL 1 or 5 µg/L for EU or U.S. EPA, respectively).20,21 In a widespread study about volatile organic compounds carried out in drinking waters of California from 1996 to 2001, Williams et al. have reported that fewer than 1% of all sampled drinking water sources contained detectable levels of benzene, the mean values for each year ranging from 1.6 to 14.5 µg/L.22 Thus, the need to obtain very low limits of detection (LODs) in consonance with the legislation is, then, a crucial parameter. In the present work, a thorough study of the instrumental parameters related to acquisition data was carried out in order to enhance the sensitivity of the HS-MS methodology. For this purpose, benzene, toluene, ethylbenzene, and p-xylene were selected as model analytes taking into account their volatility and the low MCL established by European and American rulings.20,21 The methodology was also tested with excellent results for the resolution, without chromatographic separation, of mixtures of these hydrocarbons in drinking waters by using the classical partial least-squares (PLS) multivariate calibration algorithm. EXPERIMENTAL SECTION Chemicals and Reagents. BTEX was supplied by SigmaAldrich (Madrid, Spain), and methanol (gradient grade), nitric acid, and potassium chloride were purchased from Panreac (Barcelona, Spain). Individual stock standard solutions were prepared at concentrations of 1.0 g/L in methanol and stored in amber glass-stoppered bottles at 4 °C. These solutions remain stable for 3 months. More dilute individual or cumulative solutions were prepared in 50 mL of methanol and further diluted in ultrapure water or drinking water at the microgram per liter level. (14) Martı´, M. P.; Busto, O.; Guasch, J. J. Chromatogr., A 2004, 1057, 211217. (15) Martı´, M. P.; Boque´, R.; Riu, M.; Busto, O.; Guasch, J. Anal. Bioanal. Chem. 2003, 376, 497-501. (16) Del Nogal-Sa´nchez, M.; Pe´rez-Pavo´n, J. L.; Garcı´a-Pinto, C.; Ferna´ndezLaespada, M. E.; Moreno-Cordero, B. Anal. Bioanal. Chem. 2005, 382, 372-380. (17) Serrano, A.; Gallego, M. J. Chromatogr., A 2004, 1045, 181-188. (18) Pe´rez-Pavo´n, J. L.; del Nogal-Sa´nchez, M.; Garcı´a-Pinto, C.; Ferna´ndezLaespada, M. E.; Moreno-Cordero, B.; Guerrero-Pen ˜a, A. Anal. Chem. 2003, 75, 2034-2041. (19) Pe´rez-Pavo´n, J. L.; del Nogal-Sa´nchez, M.; Garcı´a-Pinto, C.; Ferna´ndezLaespada, M. E.; Moreno-Cordero, B. Anal. Chem. 2003, 75, 6361-6367. (20) Directive 98/83/EC of the Council of 3 november 1998, Official Journal of the European Communities 330, 05.12.1998. (21) http://www.epa.gov/safewater/contaminants/index.html#organic. Last update on November 28th, 2006. (22) Williams, P.; Benton, L.; Warmerdam, J.; Sheehan, P. Environ. Sci. Technol. 2002, 36, 4721-4728.

2998 Analytical Chemistry, Vol. 79, No. 7, April 1, 2007

Figure 1. Flow diagram representing the whole analytical protocol.

Safety. BTEX compounds studied are suspected carcinogens, and caution must be exercise with their use. All handling of the solutions should be performed in a ventilation hood using latex gloves, and inhalation or skin contact should be avoided. Instrumentation. Experiments were carried out with a HS autosampler HP 7694 and an HP 6890 gas chromatograph (Agilent Technologies, Palo Alto, CA) equipped with an HP 5973 massselective detector. The autosampler included a robotic arm, a 44space autosampler carousel, and a HS generation unit, which combined an oven to heat the samples inside the vials and a sixport injection valve with a 3-mL loop filled with the HS fraction. The operating conditions for the HS autosampler were as follows: vial equilibration time, 20 min; oven temperature, 70 °C; vial pressurization time, 21 s; loop fill time, 9 s; valve/loop temperature, 110 °C. Helium (6.0 grade, Air Liquid, Seville, Spain), regulated with a digital pressure and flow controller, was used both to pressurize vials and to carry the headspace formed to the injection port of the chromatograph via a transfer line at 120 °C. Injection was done in the split mode (1:10 split ratio) for 1.0 min; an HP-5MS [5% phenyl- 95% methylpolysiloxane capillary column (30 m × 0.32 mm i.d., 0.25-µm film thickness), J&W] was used with an oven temperature of 200 °C and an helium constant flow rate of 2.0 mL/min to transfer volatiles directly into the detector. The second module was a quadrupole mass spectrometer detector operated in full-scan mode with a range from m/z 50 to 110 at 10.3 scans/s. The transfer line, source, and quadrupole temperatures were maintained at 120, 230, and 150 °C, respectively. Total ion current chromatograms were acquired and processed using G1701DA (rev. D.01.02) MSD ChemStation software (Agilent Technologies) on a Pentium IV computer that was also used to control the whole system. Twenty-milliliter glass flat-bottomed vials for headspace analysis with 20-mm PTFE/silicone septa caps and a crimped aluminum closure (Supelco, Madrid, Spain) were also employed. Vials and septa were heated at 100 and 70 °C, respectively, overnight prior to use. Analytical Procedure. The whole procedure followed in this work is schematically shown in Figure 1. For the determination

Figure 2. Volatile profile of a drinking water sample spiked with 10 µg/L of each BTEX compound provided by HS-MS: (a) whole profile; (b) and (c) profiles obtained by using time windows of 3 and 12 s, respectively. For other details, see text.

Figure 3. S/N ratio achieved at three different time windows (3, 12, and 60 s) using a scan rate of 15.3 for the most characteristic ions of the BTEX compounds (m/z 78, 91, and 106).

of BTEX, 15 mL of drinking water containing individual or cumulative concentrations of BTEX between 1 and 30 µg/L, 2.2 g of KCl, and 300 µL of 5 mol/L HNO3 (sample medium 0.1 mol/L HNO3) were placed in a 20-mL glass vial that was tightly sealed. Afterward, the sample was thermostated under constant mechanical stirring at 70 °C for 20 min in order to equilibrate the gas phase and enrich it with BTEX compounds from the water. Then, the headspace of sample in the loop of the injection valve (3 mL) was introduced into the injection port of the gas chromatograph. The chromatographic column (used as an interface between the HS sampler and mass spectrometer) was kept at 200 °C, so that volatile compounds could reach the detector at once and a total ion current (TIC) profile was obtained. Only a little fraction of this volatile profile obtained from 1.95 to 2.01 min at 10.3 scans/s was selected for the determination of BTEX compounds in order to obtain the higher sensitivity. The determination of each BTEX in mixtures was carried out by using PLS multivariate calibration

Figure 4. Effect of the scan rate on the abundance signal and S/N ratio for the m/z 91. Table 1. Comparison of the Sensitivity Achieved for the Quantification of BTEX Using HS-MS and HS-GC-MS Methodologies LOD, µg/L compound

m/zb

benzene toluene ethylbenzene p-xylene

78 91 91 91

a

HS-MS (A)a 0.26 0.23 0.17 0.20

0.80 0.75 0.60 0.60

HS-MS (B) 1.0 0.9 0.8 0.8

3.0 2.7 2.5 2.5

HS-GC-MS (C) 0.20 0.19 0.14 0.15

0.60 0.60 0.50 0.50

Proposed method. b m/z values used for quantification.

algorithm provided by the Pirouette 3.11 software from Infometrix Inc. (Woodinville, WA). RESULTS AND DISCUSSION Optimization of the Data Acquisition Time and Scan Rate. As the optimization of the HS conditions for the BTEX compounds had been carried out in a previous work,17 the present study was Analytical Chemistry, Vol. 79, No. 7, April 1, 2007

2999

Table 2. Calibration and Validation Results of the PLS Model for the Determination of BTEX in Water calibration step compound

factors

toluene ethylbenzene p-xylene

4 4 6

RMSEC (µg/L)

prediction step

RSEPC (%)

RMSECV (µg/L)

RSEPCV (%)

RMSEV (µg/L)

RSEPV (%)benzene40.111.90.111.90.153.1

1.8 1.7 3.5

0.10 0.12 0.21

1.9 2.1 3.5

0.14 0.19 0.23

2.5 3.5 2.4

0.09 0.10 0.20

Table 3. Comparison of the Quality Achieved for the Quantification of BTEX in Drinking Water Samples concentration (µg/L) benzene

toluene

ethylbenzene

p-xylene

sample

added

predicted

added

predicted

added

predicted

added

predicted

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

1.00 5.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 5.00 5.00 0.00 5.00 0.00 10.00 1.00 1.00 5.00 0.00 5.00 5.00 30.00 0.00 5.00 5.00 10.00 30.00 5.00 10.00

1.02 5.03 0.04 -0.06 -0.14 -0.04 -0.02 0.03 1.08 0.02 5.05 5.09 0.02 5.14 0.03 10.34 1.04 1.03 5.02 0.02 5.19 5.06 30.31 0.02 5.03 5.21 10.25 30.39 5.17 9.83

0.00 0.00 1.00 15.00 0.00 0.00 0.00 0.00 1.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 1.00 0.00 5.00 10.00 5.00 0.00 10.00 5.00 5.00 5.00 10.00 10.00 10.00

0.01 0.01 0.97 14.91 0.03 0.05 -0.03 0.01 1.02 1.08 0.01 0.01 0.01 0.02 0.00 0.00 1.06 1.07 0.01 5.19 10.32 4.98 -0.01 10.15 5.20 4.98 5.13 10.14 10.22 9.86

0.00 0.00 0.00 0.00 5.00 15.00 0.00 0.00 0.00 0.00 5.00 0.00 5.00 0.00 5.00 0.00 1.00 0.00 5.00 5.00 0.00 30.00 10.00 10.00 5.00 30.00 5.00 10.00 10.00 5.00

0.02 -0.01 0.03 -0.24 5.10 15.26 0.04 0.02 0.00 0.00 5.12 0.01 5.13 -0.36 5.07 0.03 1.00 0.00 5.19 5.03 0.01 30.24 10.09 10.10 5.21 30.27 5.12 10.15 10.16 4.83

0.00 0.00 0.00 0.00 0.00 0.00 10.00 15.00 0.00 1.00 0.00 10.00 5.00 30.00 30.00 30.00 0.00 1.00 5.00 5.00 5.00 0.00 10.00 30.00 5.00 30.00 10.00 30.00 10.00 10.00

0.00 0.04 0.02 0.07 -0.06 -0.02 9.80 15.15 0.04 1.04 0.03 10.14 5.07 29.88 30.25 30.38 0.04 1.09 5.14 5.03 5.20 0.02 9.92 30.36 5.13 30.42 10.22 29.81 10.41 10.25

Table 4. Regression Parameters for the Plots of Predicted versus Added BTEX Concentration for the Validation Set Obtained by the PLS Model compound

intercepta

slope

regression coefficient

benzene toluene ethylbenzene p-xylene

0.02 ( 0.02 - 0.05 ( 0.07 - 0.03 ( 0.04 - 0.02 ( 0.07

1.012 ( 0.002 0.96 ( 0.01 1.000 ( 0.004 0.988 ( 0.005

0.9999 0.9966 0.9997 0.9995

a

In µg/L.

focused on the optimization of data acquisition time and the scan rate in order to increase the sensitivity of the HS-MS instrument. For this purpose, the data matrix, consisting of the signal abundance of all the mass-to-charge ratio (m/z) monitored, was obtained by the sum of all abundance values recorded during the data acquisition time or time window. Since m/z signal abundances, in absolute terms, varied according to the different scan rates or time windows, a relative parameter such as the signalto-noise (S/N) ratio was used to optimize its sensitivity. The S/N 3000

Analytical Chemistry, Vol. 79, No. 7, April 1, 2007

ratio was calculated as the quotient between the abundance values of the most characteristic ions of the BTEX compounds (m/z 78, 91, and 106) corresponding to a contaminated drinking water sample with 10 µg/L concentration of each analyte as compared to an uncontaminated one. Figure 2a shows the TIC profile provided by the HS-MS instrument (also called “volatile profile”) from 1.60 to 2.60 min, whereas panels b and c in Figure 2 show the m/z 78, 91, and 106 profiles. As can be seen, in all cases, the time corresponding to the maximum intensity of the profiles was obtained at ∼1.98 min, which was denoted by tm. Initially, different data acquisition times were studied, such as the time corresponding to the whole volatile profile (60 s) and several time windows defined as tm ( ∆t, in order to evaluate their influence on the S/N ratio. Two time windows were assayed: the first corresponding to 1.98 ( 1.5 s, in which the peaks of the most significant ions of the BTEX compounds appeared complete (see Figure 2b) and the second located at 1.98 ( 6 s, which virtually provides the same chemical information as the TIC profile (see Figure 2c). Figure 3 shows bar plots of the S/N ratios for the

m/z 78, 91, and 106 obtained for the three data acquisition times studied. As can be seen, the narrowest time window (3 s) provided the highest sensitivity for the three ions studied, whereas the other choices, as expected, yielded similar S/N ratios (from 25 to 60% of the signal abundance achieved for the time window of 3 s). The increase in sensitivity (S/N ratio) by using a time window of 3 s can be explained based on the fact that the TIC profile in this time interval is restricted to the region where the abundance values of the most characteristic ions of the BTEX compounds are maximal (see Figure 2b), and the contribution of the background noise is minimum due to the narrow time interval selected. In addition, the narrow time window selected (3 s) minimizes possible interferences (increasing the selectivity) since the data acquisition is focused on the period of time in which the BTEX compounds are detected. As stated above, another way to enhance sensitivity is based on the fact that the MS can operate at different scan rates. The effect of this parameter on the sensitivity of the HS-MS instrument was also evaluated by analyzing both spiked and blank drinking water samples at 5.8, 10.3, 17.1, and 25.5 scan/s. As expected, absolute signal abundances increased as the scan rate increased from 5.8 to 25.5 since higher number of scans were added to generate the data matrix. However, the S/N ratio provided opposite results (see Figure 4); concretely, at scan rates over 10.3 for the m/z 91 (base peaks and quantification values for toluene, ethylbenzene, and p-xylene compounds), there was a decrease in the S/N ratio. Thus, in spite of being able to obtain higher signal abundances at 25.5 scan/s, the best S/N ratio was achieved at e10.3. This behavior can be ascribed to the combination of two effects: (i) the blank signal abundance also increases on increasing the scan rate from 5.8 to 25.5; and (ii) at a low scan rate, the MS spends more time for recording each m/z ratio, and hence, the signal abundance provided in each scan will be higher than that obtained at high values. From the foregoing it follows that a 10.3 scan/s was selected as the optimal value. Quantitative Results for Individual BTEX Determination. In order to evaluate the performance of the proposed approach for improving the sensitivity of the HS-MS instruments, the LODs for the BTEX obtained by proposed method A were compared to those provided by using the full volatile profile (time window, 60 s at 15.6 scan/s, method B) and using the HS-GC-MS, method C.17 In all cases, drinking water samples spiked with individual standard solutions of each BTEX (from 1 to 30 µg/L) were analyzed by using the appropriate experimental procedure (see Analytical Procedure section and ref 17) and the m/z 78 for benzene and m/z 91 for the other BTEX for MS quantification. For this purpose, a calibration graph was constructed (6 points in triplicate) for each analyte and method (A-C). The standard errors of the estimate (Sy/x) of the corresponding calibration graphs were used to calculate the LODs, assuming the value as the relative standard deviation of the blank.23 Table 1 lists the LODs obtained for both HS-MS approaches (methods A and B) and for the HS-GC-MS one (method C). From this, it can be concluded that the LODs provided by the proposed HS-MS method were three or four times lower than those previously (23) Miller, N. J.; Miller, J. C. In Statistics and Chemometrics for Analytical Chemistry, 4th ed.; Prentice Hall: Upper Saddle River, NJ, 2000; pp 111152.

obtained by method B and similar to the HS-GC-MS method; thus, a significant improvement is reported in the sensitivity of the HSMS method. Resolution of BTEX Mixtures in Drinking Waters by PLS. The classical PLS multivariate regression technique was used to assess the usefulness of the proposed approach for the sensitive determination of BTEX compound mixtures in drinking waters. For this purpose, an overall set of 130 drinking water samples spiked with individual or binary, ternary, and quaternary mixtures of the BTEX compounds at concentrations between 1 and 30 µg/L were prepared as described in the Experimental Section and run together. From this, two randomly chosen sets were used to design the PLS model (100 samples) and for its validation (30 samples). From preliminary studies, six samples were detected as outliers (possibly due to operational errors) and therefore they were removed from the data matrix. Consequently, the subset used for the optimization of the model consisted of 94 spiked drinking water samples. Fourteen variables (signal abundance measured at m/z values) instead of a possible 60 (see Instrumentation section) were selected to construct PLS models in order to facilitate calculations and data manipulation. These variables included the most significant m/z values of the BTEX compounds and three m/z values that represent background noise: m/z 50, 51, 52, 63, 65, 75, 77, 78, 88, 91, 92, 100, 105, and 106. For the construction of the different PLS models, the crossvalidation method was used24 and the number of significant factors in each case were chosen as the lower number whose root mean standard error (RMSE) of prediction by cross-validation was not significantly different from the lowest RMSE value:

x

m

∑ (c - cˆ )

RMSE )

2

i

i

i)1

(1)

m

where ci and cˆi are the experimental and calculated concentrations, respectively, and m is the total number of calibration samples. The accuracy of each model was assessed in terms of the relative standard error of prediction (RSEP) for the results obtained for each data set, that is, RSEPC for the calibration set, RSEPCV for the cross-validation set, and RSEPV for the validation set:

RSEP )

100 jc

x

m

∑ (c - cˆ )

2

i

i

i)1

m

(2)

where jc is the average of the true (spiked) concentrations. Table 2 shows the results obtained for the optimization of the PLS model as well as those provided by the validation set to quantify each BTEX in drinking water samples. As stated above, the cross-validation method, and concretely the leave-one-out procedure, was used for the construction of the PLS models; however, taking into account the high number of samples included in the calibration set (94 samples), different K-fold cross-validations were carried out to assess the robustness of the model and also (24) Osten, D. W. J. Chemom. 1988, 2, 39-48.

Analytical Chemistry, Vol. 79, No. 7, April 1, 2007

3001

to decrease computation time. Thus, PLS models were crossvalidated leaving out K ) 1, K ) 5, and K ) 20 calibration samples. The similar results obtained when leaving out 1 or 5 samples (see Table 2) and the slight increase observed in the RMSE for K ) 20 (e.g., RMSE for benzene ranged from 0.11 to 0.13 µg/L) testify to the robustness of the proposed PLS model. Consequently, the optimum number of factors in Table 2 for each BTEX was obtained by using 5-fold cross-validation in which the data set was divided into 5 subsets, and the holdout method was repeated 5 times. Regarding accuracy, the model provided quite good results for determining the concentration of each hydrocarbon in drinking waters: the RMSE values for the calibration and validation sets ranged from 0.09 to 0.23 µg/L whereas RSEP values varied between 1.8 and 3.5%. Table 3 shows the results for the synthetic samples of the validation set containing variable amounts of BTEX. As can be seen, samples can be accurately resolved with relative errors of less than ∼ (5%, except for some samples containing a 1.0 µg/L concentration of BTEX (relative error e10%). In addition, Table 4 shows the regression parameters of the plots of predicted versus added concentrations for each BTEX in the validation set. These results demonstrated that the predictive ability of the PLS model is very good for the determination of BTEX in drinking water samples. Only in the case of toluene was a slightly systematic negative error detected, taking into account the value of the slope of its regression plot. These values testify to the greater accuracy achieved by the proposed method even for complex samples such as quaternary mixtures of BTEX, which can be ascribed to the possible enhancing effect on the selectivity associated with the narrow time window used to acquire the chemical information provided by HS-MS. In summary, the proposed method is a useful choice to quantify BTEX in this kind of sample with a high degree of sensitivity and accuracy. CONCLUSIONS As shown in this study, the appropriate selection of the data acquisition time and the scan rate for extracting useful chemical

3002

Analytical Chemistry, Vol. 79, No. 7, April 1, 2007

information from the HS-MS volatile profile provides a straightforward method for sensitive determination of individual and mixtures of BTEX in drinking waters. The proposed HS-MS method (A) provided a sensitivity similar to that of the HS-GCMS one (C) being faster and simpler since no chromatographic step is required. The ensuing method compares favorably with HS-MS alternatives in terms of sensitivity because the quantitative determination of these hydrocarbons in the samples analyzed can be carried out at the microgram per liter level whereas poorer ranges are achieved by using the methods reported, concretely at the milligram per liter level.8,9,16,18,19 In addition, the proposed method permits the resolution of mixtures of these hydrocarbons at concentration ratios similar to that achieved by the sole HSMS methodology reported in the literature,19 also based on the construction of PLS models. The accuracy was also similar in spite of dealing with water samples with lower BTEX concentrations (at µg/L level) in the proposed method, and even better as in the case of the quantification of xylene. Finally, the advantages obtained by the approach proposed in the HS-MS methodology for BTEX compounds can be also extrapolated to other methods that insert analytes directly into a MS like MIMS as well as for other compound families such as trihalomethanes, aliphatic hydrocarbons, etc. ACKNOWLEDGMENT The authors thank the DGI of the Spanish Ministry of Science and Technology for its financial support awarded in the form of Grant CTQ2004-02798.

Received for review February 5, 2007. AC070044R

January

9,

2007.

Accepted