Coupled with Gas Chromatography/Mass Spectrometry for the Analysis of Environmental Samples

		Abstract			Results and Discussion
		Introduction			Determination of Values
		Experimental Section			Selection of Surrogates
		Vacuum Distillation Apparatus			Water Sample Analyses
		GC/MS Apparatus			Soil Sample Analyses
		Sample Preparation			Oil Sample Analyses
		Vacuum Distillation Procedure			Conclusion
					Acknowledgment
					References
					Tables and Figures

Abstract
A procedure is presented that uses a vacuum distillation/gas chromatography/mass spectrometry system for analysis of problematic matrices for volatile organic compounds. The procedure compensates for matrix effects and provides both analytical results and confidence intervals from a single sample analysis. Surrogate compounds are used to measure matrix effects relating to boiling point and relative volatility and to provide the information necessary to accurately determine analyte concentration. Relative volatility values (a) are experimentally determined for 114 organic compounds and are shown to be comparable to gas-water partition coefficients. These compounds include those with boiling points up to 245 °C and gas-water partition coefficients less than 15,000. Multiple samples are tested and the accuracy of determinations is shown to be within 5% for water, soil, and oil matrices. Method detection limits are below 1 ppb for most analytes studied.

Introduction
One of the major objectives of the analytical chemistry research program at the EPA's Characterization Research Division (National Exposure Research Laboratory) in Las Vegas is to broaden the array of pollutants that can be determined with conventional analytical instrumentation. The U. S. Environmental Protection Agency (EPA) has developed a vacuum distillation method for determining the concentration of volatile organic compounds (VOCs) in environmental samples¹ and identified the relationships controlling analyte recovery and the potential of surrogate-based matrix corrections.² The purpose of the present study was to incorporate a surrogate-based matrix correction in a general vacuum distillation/gas chromatography/mass spectrometry (VD/GC/MS) method to be used for routine environmental analyses. At the same time, the list of applicable analytes has been better defined and documented.

Suitable compounds would reflect the effects of a matrix on analyte recovery as functions of boiling point (b-effects) and relative volatility (a-effects). Through the analyses of multiple samples, the ability of the specified surrogates to predict matrix effects is demonstrated. The surrogate prediction routine is simple and provides accurate determination of analyte concentrations in aqueous and mixed-phase samples. The quality assurance needed to document a VD/GC/MS analysis is performed by simply reviewing the surrogate performance in the sample. This replaces the need for costly matrix spikes or standard addition analyses.

A vacuum distillation procedure for determining the relative volatility of analytes as a constant (a_K-values) is presented. The a_K-values for 114 compounds are experimentally determined and shown to be comparable to their gas-liquid partition coefficients (K). The determination of a_K-values in this study are an improvement over previously reported² a-values (normalized) as the analyst can use published K values to identify potential analytes and estimate their behavior. The boiling point and K of an analyte govern its performance and determine whether it is suitable for VD/GC/MS analyses. Suitable analytes include compounds with boiling points up to 245 °C and partition coefficients up to 15,000; note that this range includes compounds not normally considered as VOCs (e.g., nitrosamines, aniline, and pyridine). Through multiple analyses of various matrices, the accuracy that can be expected for determining an analyte is reported. Method detection limits for VOCs are in the low ppb range for water, soil, and oil matrices.

Experimental Section

Vacuum Distillation Apparatus. The vacuum distiller has been previously described.^1,2 In the current study, a Nupro toggle valve (0.172-in. orifice) was used as the sample chamber valve. A vacuum gauge was installed between the cryoloop and the vacuum pump to monitor the integrity of the apparatus under vacuum. The vacuum was considered acceptable at a pressure of 0.5 torr or less.

The condenser column was normally held at 5 °C during vacuum distillations and at 40 °C between distillations. Water was used to replace isopropyl alcohol as the temperature-controlling fluid in the condenser.

GC/MS Apparatus. A Hewlett-Packard mass spectrometer (Model 5972) and gas chromatograph (HP5890 Series II with Model MJSC metal jet separator) with a 60-m x 0.53-mm i.d., 3.0-mm film thickness, VOCOL capillary column (Supelco, Bellefonte, PA) was used for the determination of analytes from the vacuum distillation apparatus. Gas chromatograph operating conditions were: 3 min. at 10 °C, 50 °C/min ramp to 40 °C; 5 °C/min ramp to 120 °C; 20 °C/min ramp to 220 °C; and isothermal at 220 °C for 3.4 min, resulting in a total run time of 28 min. The jet separator was held at 210 °C and the transfer line held at 280 °C. The mass spectrometer was operated at 3.1-s scans of 38 to 270 amu. The injector was interfaced to the vacuum distillation apparatus by connecting the carrier inlet gas line to the cryoloop valve and back to the injector. The injector inlet temperature was 240 °C and the inlet pressure was 10 psi.

Sample Preparation. Aqueous samples were prepared directly in the 100-mL round bottom flask used in the vacuum distillation of the sample. Modified samples were prepared by weighing the amount of matrix modifier (i.e., glycerin, salt), adding water, and then adding the analytes of interest. The 2% soap samples were prepared by adding 0.1 mL soap concentrate (Micro Concentrated Cleaning Solution, International Products Corp., Burlington, NJ) to the water sample. Samples were spiked with a 10-mL methanol solution containing analytes at the concentrations listed in Table 1.

Two spike techniques were used to study the matrices not soluble in water. The first technique was to simply spike into a slurry of the sample and 5 mL water -- referred to as a water spike. The second technique was an attempt to maximize matrix contact and is referred to as a vacuum spike. The introduction of analytes to a soil sample contained in a vacuum has been reported to be a more difficult spike to recover and is potentially a more accurate spiking technique compared with the water spike.³ The vacuum spike in this study entailed several steps, the first of which involved weighing the soil in the sample flask and injecting the spike with a syringe onto the material. The flask was then attached to the vacuum distillation apparatus, and the soil plus spike was cryogenically cooled by immersing the flask in a liquid nitrogen bath (-196 °C). When the flask and sample were thoroughly cooled, the apparatus vacuum pump was used to lower the flask pressure to 0.5 torr. After the air had been removed (3-5 min) and the sample chamber valve closed to isolate the sample, the flask was warmed to 30 °C. After a 1-h equilibration period, the mixture was again cooled cryogenically, removed from the apparatus, 5 mL water added, and the sample reconnected to the apparatus.

Three different soils of varying water and organic contents were used in this study. Soil #1 and Soil #2 were garden soils that were 37 and 15% water and 21 and 16% organic matter, respectively. Soil #3 was a desert soil containing 3% water and 1% organic matter. Cod liver oil (Squibb Cod Liver Oil Mint Flavored, E. R. Squibb & Sons, Inc., Princeton, NJ) was used as the oil matrix. This material was selected as the oil matrix due to its low content of the analytes to be determined. This matrix was assumed to mimic both waste oils and the lipid content of biological samples in its effects on the analyte recovery.

Vacuum Distillation Procedure. Prior to a vacuum distillation, the condenser column was cooled to 5 °C. The sample contained in the 100-mL flask (normally at room temperature) was evacuated for 10 min, the water vapors were collected on the condenser column, and the distillate to be analyzed was collected in the cryoloop immersed in liquid nitrogen (-196 °C). The sample chamber valve was closed at the completion of the vacuum distillation and the cryoloop valve switched to allow the GC carrier gas to sweep the cryoloop. The cryoloop's liquid nitrogen bath was removed and replaced with a hot water bath (70 to 90 °C) to volatilize the distillate. The transfer of the distillate to the GC is completed after 2.5 min. After the sample was vacuum distilled, the condenser column was heated to 40 °C while being evacuated with the vacuum pump for 10 min; this removed the condensed water and potential contaminants. All measurements of analytes were performed by the GC/MS analysis of the transfered vapor.

Results and Discussion

The demonstration of a- and b-effects was demonstrated using GC/MS analyses of a methanol solution containing the vacuum distillates.² In this study the vacuum distillates were analyzed as vapor transfer between the vacuum distiller and the GC/MS. Interfacing the GC/MS and vacuum distiller simplifies analysis but adds another potential source of analyte loss that was evaluated. The next phase of this study was to determine the a-values for the analytes and to select surrogates for measuring the a- and b-effects. The final phase of this study was the analyses of a variety of matrices to test the GC/MS/VD method.

The transfer of vapors from the cryoloop to the GC was investigated for its affect on analyte recoveries. Cryoloop trapping efficiencies were investigated by comparing injections of analytes directly into the GC/MS with the injection of analytes into the vacuum stream just before the cryoloop. Figure 1 shows the cryoloop trapping efficiencies for injections of analytes at a low pressure (0.5 torr) and at a greater pressure (created by allowing air to be drawn into the cryoloop simultaneous with the injection on the cryoloop). The injection with air simulates the pressure within the cryoloop at the initiation of a vacuum distillation (sample flask is at atmospheric pressure), and the low pressure injection simulates the cryoloop internal pressure later in a vacuum distillation cycle when the air has been evacuated.

The plot of analyte recoveries versus boiling point (Figure 1) shows that trapping efficiencies vary closely in relation to an analyte's boiling point. The most volatile analytes are trapped least effectively for both injection pressures. The analyte efficiencies improve with increasing boiling point up to 220 °C. Analytes with boiling points above 220 °C also demonstrate lower efficiencies and most likely reflect a less efficient transfer from the cryoloop to the GC. The trapping efficiency drops when air is bled into the cryoloop. This indicates that efficiency drops with increasing pressure (mass transfer), and, therefore, during a vacuum distillation, the efficiency of the cryotrap will increase as air is evacuated from the sample flask. The loss of analyte occurring at the cryoloop is not easily distinguished from losses due to analyte condensation and therefore these cryoloop losses are included as a component of b-effects.

The boiling point-condensation relationship previously described² is minimized in this study as both the samples and standard solutions are vacuum distilled (previous study did not vacuum distill standard solutions) using the same vacuum distillation conditions. Because analyte condensation on the condenser, cryoloop trapping efficiencies, and the cryoloop-to-GC transfer are essentially the same for samples and standards, there is a normalization of the b-effects. The b-surrogates (surrogates to measure boiling-point effects) are now used to rectify any variation of boiling-point effects between the analyses of standard solutions and the analyses of samples (including the cryoloop-to-GC transfer).

The effects produced by varying the condenser column operating temperature were evaluated by comparing analyte responses obtained from direct GC-injection of analytes with those from the vacuum distillation of analytes using different condenser temperatures. Figure 2 presents the relationship of analyte recovery to condenser temperature using the surrogates naphthalene-d₈ and ethyl acetate-2C¹³ as representative analytes. The recoveries of both analytes are maximized when the condenser column is between -6 and 10 °C. As these compounds have greatly different boiling points, the b-effects (at the condenser) appear to be at a minimum when the condenser column temperature is between -6 and 10 °C. The minimizing of b-effect enhances the response of the higher boiling analytes and simplifies the correction of such effects.

The amount of water being collected on the condenser column was also a consideration. Preparation of the condenser column between distillations is made simpler if the amount of condensed water is minimized. Water collected as a function of condenser column temperature (Figure 3) shows that the amount of water collected decreases as the condenser column temperature is increased. A condenser operating temperature between -6 and 10 °C minimizes analyte condensation and between 5 and 10 °C minimizes water collected. The operating temperature of 5 °C was used in this study to allow for some temperature fluctuations with minimal impact on analyte recoveries.

Determination of Relative Volatility Values. The recovery of an analyte in the absence of observable b-effects depends on its relative volatility. Using the same operating conditions to perform each distillation within a set (a series of 10-min vacuum distillations required to completely evaporate a 5-mL sample) makes the influence of b-effects (within experimental variation) consistent. With the response of analytes comparable between distillations in a set, the rate of removing analytes by distillation (a-effects) can be measured. Therefore, the recovery of an analyte corresponding to its relative volatility is calculated to be its response in the initial vacuum distillation divided by the sum of responses for the set.

A cylindrical flask (15-mm i.d., 8-cm. length) replaced the standard 100-mL round bottom flask for this part of the study. The cylindrical flask produced a constant sample surface (2.7 cm²) in the flask and a more reproducible distillation rate of water (< 0.5 mL water per 10 min at 20 °C). The use of a 100-mL round bottom to contain the water sample would have resulted in a rapidly changing surface area (decreasing as water is vaporized) that would have to be addressed when comparing the vacuum distillations within a set. The cylinder flask also had the advantage of slowing the vacuum distillation of analytes (greatest impact on the least volatile analytes) and exaggerating the losses related to a-effects. While the cylinder would not be as desirable as the round bottom flask for routine analyses, using the cylinder made the distinction between analytes (related to a-effects) easier to observe and measure.

The cryoloop efficiency varies with pressure, and, to make distillation conditions the same within a set, it was necessary to remove air from the flask containing the sample prior to the first vacuum distillation. The evacuation of air was accomplished by cryogenically freezing the spiked sample in liquid nitrogen and then evacuating the flask for 3 to 5 min. The sample was then warmed to 20 °C and the first vacuum distillation was performed.

The first vacuum distillation of each set was performed while the sample was immersed in a 20 °C water bath. The sample temperatures used after the initial vacuum distillation were varied (20 to 45 °C using a water bath) to exaggerate the range of recoveries corresponding to a degree of relative volatility (a-values). When fewer vacuum distillations (higher sample temperatures) were required to evaporate the 5-mL sample, the recovery differences between the lower a-valued analytes were emphasized. More vacuum distillations (lower sample temperatures) emphasized the differences between the higher a-values.

It was also observed that the relative volatility of several of the more soluble analytes (i.e., pyridine and the nitrosamines) would vary as a function of the initial vacuum distillation time. The relative volatility of these analytes would increase as the vacuum distillation times were increased, which seems to indicate some analytes exhibit a retention-time gap in the condenser or the formation of an azeotrope with the water collected in the condenser column. For this study the relationship of recovery to relative volatility for these analytes was determined using initial vacuum distillation times between 2 and 5 min.

The recoveries of selected surrogates (benzene-d₆, o-xylene-d₁₀, 1,2-dichloroethane-d₄, ethyl acetate-2C¹³, acetone-d₆, dioxane-d₈, and pyridine-d₅) were then used to plot the a-value versus recovery relationship. The values of their partition coefficients (25 °C) were assigned as their respective a-values so that interpolated a-values  (a_K-values) would correspond to partition coefficients. The a_K-values for 114 analytes were experimentally determined and are listed in Table 1. The assignment of the a_K-value for pyridine-d₅ was an estimation.

It was observed that when the sample boiled vigorously a large amount of water collected on the condenser, the recoveries of the less volatile analytes were greatly improved, and the recoveries of the more volatile analytes were diminished. This made the recovery versus a_K-value relationship less distinct and interpolation of values less precise. Therefore, a set was not used to determine a_K-values if the initial distillation boiled (the only distillation in a set prone to boil vigorously).

A listing of published partition coefficients values in Table 1 show the a_K-values are a good estimator of analyte partition coefficients. The exception would be for those analytes with a_K-values greater than 5,800 where the estimate of the partition coefficient (and a_K-value) for pyridine-d₅ could introduce error.

Selection of Surrogates. The recovery of an analyte is related to its a-value and its boiling point, and these relationships can be quantified using surrogates.² Table 1 identifies the surrogates used in this study. The a-surrogates represent ranges of a_K-values, and their recoveries are used to describe the a_K-value versus recovery relationship.  b-Surrogates represent ranges of boiling points, and their recoveries are used to describe the analyte recovery versus boiling-point relationship. Those surrogates identified as check surrogates in Table 1 were spiked into all samples to check analyses and calculations.

The analytes and a-surrogates are grouped in five divisions by their a_K-values (Table 2). Each of these groups has four surrogates to describe how the analytes of the group behave. The a-surrogates are selected so that there are two pairs of surrogates with similar a_K-values bracketing a group of analytes. The replicate calculations that can be made then provide the analyst with a means to assess confidence in the measurements. The surrogate pairings shown in Table 2 identify the combinations of surrogates used to calculate a-effects.

The boiling point groupings and their respective surrogate pairings used in b-effect calculations are also presented in Table 2. The b-surrogates represent seven increments of boiling recovery relationship. Surrogates are not used to address the recovery-boiling point relationship for the boiling points below 80 °C. The compounds with boiling points below 40 °C generally are not concentration-stable (losses due to evaporation) in a surrogate spiking solution and do not withstand routine handling. While there is a drop in cryotrap trapping efficiencies for those analytes with boiling points below 80 °C, a greater effort to profile the lowest boiling analytes is of limited value considering the cost of frequent replacement of a surrogate solution and the potential vaporization of the most volatile analytes prior to analysis. The effects of not having a pair of lower-boiling surrogates is seen in the poorer-quality analyses reported for the lowest boiling analytes.

The vacuum distillation of an actual sample can yield different values for analytes compared with the vacuum distillation of the same amount of analytes in a standard solution. In this study, the comparison is called relative recovery. The component of relative recovery of an analyte that is related to its a_K-value is referred to as a-effect or R_a.  The relative recovery related solely to its boiling point (vapor pressure) is referred to as b-effect or R_b and includes those relative losses that occur at the cryoloop and the condenser column. The relative recoveries of the a- and b-surrogate compounds predict how other analytes behave in the same sample matrix relative to the analyses of a standard solution (in distilled water). Once the relative recovery of analytes from a matrix is established, the responses of analytes in a standard solution are used to determine analyte concentrations.

The evaluation of a- and b-effects was undertaken with two approaches. The first approach was to initially measure the a-effects by using those surrogates with boiling points less than 150 °C. After the a-effects were predicted, the b-effects were calculated as the difference between the relative recoveries predicted using a-values and their measured recoveries. This approach worked well for aqueous samples as the b-effects were observed to be minimal adjustments to relative recoveries. This first approach was not adequate for soils where the b-effect could be significant for analytes that have boiling points below 130 °C (likely due to the partitioning between the solid phase and the vapor phase). There were too few unaffected surrogates to describe the a-effects when soils were analyzed. Therefore, as a second approach, the b-effects were determined prior to the a-effects. This second approach, however, still required an initial estimate of the a-effects on the b-surrogates before calculating the b-effects. This second approach was used for generating data reported in this study.

The initial approximation of a-effect on the b-surrogates is accomplished by using the a-surrogates, fluorobenzene and 1,2-dichloroethane-d₄ (boiling points of 85 and 84 °C, respectively), to approximate a-effects on the b-surrogates with the assumption that b-effects are minimal at 85 °C. The equation:

ln(R_a ) = a×a_K + c,      (1)

where R_a is the surrogate's relative recovery corresponding to its a_K-value, a and c are constants, and a_K is the relative volatility of the surrogate, describes the a-effect versus recovery relationship.²  The relative recoveries of the b-surrogates, toluene-d₈, chlorobenzene-d₅, bromobenzene-d₅ and 1,2-dichlorobenzene-d_4, are adjusted for their a-effects (R_b = measured recovery/R_a). The resulting relative recovery represents the component of the relative recovery related to b-effects.  Similarly, the a-surrogates 1,2-dichloroethane-d₄ and 1,4-dioxane-d₈ are used to solve eq 1 and to interpolate R_b for the b-surrogate 1-methylnaphthalene-d₁₀.

Using the b-surrogate R_bvalues, the R_b-boiling point relationship is described using the equation:

R_b = a×(bp-bp_o) + b        (2)

where R_b is the relative recovery corresponding to the boiling point, a and b are constants, bp is the analyte's boiling point, and bp_o is the lowest boiling point of the b-surrogate used in the solution.² The impact of a single b-surrogate relative-recovery measurement error is minimized by calculating three solutions to eq 2 for each analyte. The b-surrogate pairs used to solve eq 2 for groups of analytes by boiling point are identified in Table 2.  The average and standard deviation of three R_b (only two solutions for the 80 to 111 °C and 220 to 250 °C ranges) generates the predicted analyte relative recovery,`R<sub>b</sub> ± r<sub>a</sub>, corresponding to b-effects. The resultant`R_b for each a-surrogate is used to correct their measured relative responses (R_a = measured recovery/`R_b) to isolate the relative recoveries related to a-effects.

The a-surrogate corrections are performed by groupings of analytes with similar a_K-values. The a-effects exhibited by those compounds at the limits of a group are the best data to describe the a-effects for those analytes within these groups and therefore pairs of a-surrogates are selected to represent the extremes of each group's range of a_K-values (i.e., surrogates hexafluoro-benzene and fluorobenzene represent the lower and upper ends of the grouping of a-values between 0.07 and 3).

One lower-value a-surrogate and one higher-value a-surrogate are selected to calculate the relationship of relative recovery to a_K-values within the group. Using the four possible combinations of surrogates to solve eq 1, each analyte will have four a-effect measurements. The predicted relative recovery relating to a-effects for an analyte is`R_a ± r_a.  The predicted relative recovery that includes a- and b-effects is:

  _T =_a × _b    (3)

where`R_b is the average bias using eq 2 for the combinations of b-surrogates in the analytes boiling point grouping.

The associated variance term for eq. 3 is:

   r_T ² = r _a² + r _b ²      (4)

Accuracy and standard deviation associated with the surrogate corrections is calculated as

A_T ± a _T = (`R_T ± r_T)/(measured recovery)      (5)

where the measured recovery is the response of the analyte from a sample analysis compared with the analyte response in the vacuum distilled standard of the same concentration. Performing replicate vacuum distillations, the average and standard deviation of resulting determinations of A_T is identified as`A<sub>T</sub> ±`A_dev, and the average a_T is`a_T.

Water Sample Analyses. The list of analytes used in this study is long and therefore some grouping of data was necessary for presentation. Analytes are categorized as the following subsets: volatile gases (boiling points 30 °C), volatiles (boiling points > 30 °C and < 160 °C), soluble volatiles (a _K > 34), semivolatiles (boiling point > 160 °C), and basic semivolatiles. The marginal analyte subset was created for those analytes at the limits of the method. Table 1 identifies the analytes in each grouping. Analytical results (`A<sub>T</sub> ±`A_dev and`a _T) are provided as either the average or the range of the results for each of the analytes contained in a group. The individual analyte data are available as supplementary material.

Analyses of distilled water and distilled water modified to simulate more extreme aqueous matrices were conducted to evaluate the accuracy of the VD/GC/MS method with the specified surrogate corrections. The`A<sub>T</sub> ±`A_dev and the`a _T for the vacuum distillation of various water samples are presented in Table 3. The surrogates accurately profile analyte performance in water matrices, and determinations are continually accurate to within 5% of the spiked values. For those analytes not normally considered as volatile (e.g., nitrosamines), the system is less accurate. The method detection limits (MDLs) were determined using the Resource Conservation and Recovery Act (RCRA) guidelines⁸ (see footnote in Table 3). The mass spectrometer response for most analytes indicated lower MDLs could easily be obtained should they be required in future needs.

Distilled water proves to be an ideal matrix as the`A<sub>T</sub> for the individual analytes are consistently within 5% of the spiked value for distilled water and the precision errors (`A_dev) anda _T are typically less than 5%. The addition of salt increases the relative recovery of most analytes, especially those analytes with greater a-values. The salting effects are as expected, but the important finding is that the a-surrogates correctly compensate for the increases in relative recoveries as`A<sub>T</sub> is consistently within 5% of the spiked concentrations. The addition of glycerin depresses the analyte response relative to their a <sub>K</sub>-values but again`A_T is within 5% of the spiked values for most of the analytes. Such high content of soluble organic compounds is certainly an unusual sample, but the accuracy of such analyses demonstrates the reliability of the surrogate corrections.

The addition of soap produces one of the most difficult aqueous matrices for VOC determinations as the foaming produces an irregular interface between vapor and liquid. Foaming with bubbles expanding to more than 25 cc was persistent throughout the vacuum distillations. However, the quality of the analyte determinations were still very similar to distilled water. Nitrosamines, as a group, had recoveries lower than those predicted but their behavior was not dissimilar from distilled water results.

The higher boiling analytes are susceptible to cross-contamination. It was reported that the vapors of analytes being distilled that have boiling points greater than 220 °C were more than 90% condensed on the condenser.² The removal of analytes between analyses was effective for most analytes (>99 %) with the exception of the highest boiling analytes, primarily N-nitroso-dibutylamine and 2-methylnaphthalene, where more than 2% could remain. While the percentage of analyte available for carryover is not normally significant, 2% remaining on the condenser may be a greater amount of analyte than that condensed in the cryoloop and therefore a relatively significant source of contamination. The condenser temperatures used to eliminate carryover is an apparatus limitation and evidently not sufficient to completely remove the highest boiling analytes. Heating the condenser column to higher temperatures would be expected to minimize carryover and generally improve the performance for n-nitrosodibutylamine and 2-methylnaphthalene.

A 20% methanol solution was previously shown to have minimal impact on the vacuum distillation of analytes when the cryoloop condensate was being analyzed by aqueous injection.² With the configuration used in this study, however, more than 20 µL methanol interfered with the measurement of many analytes. It appears that the interferences are a result of poorer chromatography of some analytes (higher a-values) and an attenuated mass spectrometer response to analytes that co-elute with methanol. Lower recovery of higher boiling analytes is also observed and likely related to less efficient transfer of those analytes from the cryoloop to the GC. The net effects on analytes is primarily a depressed recovery of those compounds with higher a_K-values, or higher boiling points, or those that elute with methanol. The presence of 50 µL of methanol in a sample affects the chromatography of most analytes. The impact of methanol on chromatography is critical and therefore the methanol content of prepared samples is limited to 10 µL.

Even small amounts of methanol would periodically affect (attenuated mass spectrometer response) those analytes that coelute with methanol. Diethyl ether consistently coelutes with methanol, and on occasion`A_T is biased high for this analyte. Trichlorofluoromethane and acrolein are affected to a lesser degree. Changing the chromatography conditions could improve these analyte accuracies. Adding isotopes of the analytes (especially for ether) to the surrogate mix would detect the occurrence of the effect and provide a means for correcting the attenuated responses (i.e., isotopic dilution).

Larger sample sizes can be used with minimal impact on the accuracy of analyte determinations. Analyses of 25-mL samples yield relative recoveries very similar to those for 5-mL samples. While`A_T for the higher a _K-value analytes begins to exceed 10% deviation from the spiked concentrations, including the confidence window generated by ± a _T would generate a window larger than A_T that includes the true analyte concentration. Only the nitrosamines indicated a bias in accuracy that could not be compensated by the a _T. There was also some variation in the mass spectrometer response to the lowest boiling analytes. The presence of naturally occurring gases in the samples (carbon dioxide) was apparently causing mass spectrometer pressure fluctuations.

Sample bath temperatures of 10 and 30 °C were used to investigate the sensitivity of vacuum distillation to ambient temperatures. The low end of this temperature range reduced the relative recoveries more than 50%. The higher end increased the analyte responses relative to the increasing a _K-value. The surrogate corrections remain accurate with few exceptions. The`A<sub>T</sub> for nitrosamine concentrations generally exceed the true concentrations by more than 10%, and the windows generated using`a_T are too small to compensate. When the nitrosamines are of particular concern, the sample analyses should be more constrained in temperature and sample-size variations. The response for the poorer performing analytes can be improved with the addition of salt but this normally should not be required.

Soil Sample Analyses. The evaluation of soil matrices requires some assurance that there is matrix interaction. Rather than performing soil sorption-time studies, two different spiking techniques were used (the water spike, and the more rigorous vacuum spike). It was found that the accuracy of the analyses was equivalent for the two techniques. While the recoveries of analyte were lower when using the vacuum spike, the surrogates were similarly affected and were accurate in predicting analyte recoveries. This indicates that while the a- and b-effects may differ by degree of matrix interaction, the matrix effects are still accurately described using the a- and b-surrogates.

In the same manner as the water analyses, the`A<sub>T</sub>,`A_dev and`a_T for the soil analyses are presented as subset results (individual analyte data available as supplemental material). The soil results are presented in Table 3 (vacuum spike Soil #3) are similar to results from water analyses. The MDLs reported in Table 3 are also very similar to those reported for water (less than 1 ppb for most analytes). The detector response to analytes indicated lower MDLs could easily be obtained but the levels being reported already exceed most needs. Analyses of Soils #1 and #2 provided a greater variation in analyte accuracy that indicate that the organic content of a soil will greatly influence analyte behavior. Soil #1 was a heavily mulched garden soil and was 21% organic matter and 37% water. Soil #2 was another garden soil that contained less organic matter (16% organic matter and 15% water). The analyses of the higher organic content soils demonstrated good accuracy for most analytes.

The list of analytes for the soil analyses is abbreviated due to apparent degradation of some analytes. Acrolein apparently degrades very rapidly and is poorly recovered from soil. Ethyl acetate-2C¹³ and ethyl methacrylate were similarly affected but could be recovered from Soil #3 when they were added using the water spike technique. These analytes were not considered as viable analytes in the soil samples studied. Therefore the surrogate, ethyl acetate-2C¹³, is not recommended as an -surrogate for soils but rather as a check surrogate to identify when such degradation occurs. The surrogate, 1,4-dioxane-d₈, is substituted for ethyl acetate-2C¹³ in the 20-150 a_K-value grouping and the surrogate, 1,2-dibromoethane-d₄, is a substitute for ethyl acetate-C¹³ in the 150-15000 a_K-value grouping.

The relative recoveries of acrylonitrile, methacrylonitrile, and propionitrile were at times unexpectedly low. This only occurred when the analytes were vacuum-spiked into Soils #1 or #2. It appears these analytes degrade in a soil matrix and may not be viable analytes for soil analyses.

Acetonitrile is not accurately measured due to recurring spectral interference. The manual integration of acetonitrile is required, and a different temperature program would likely improve this analyte's performance. An MDL of 100 ppb is reported for 2-picoline although the MDL was actually calculated to be 5 ppb. The higher MDL is listed to minimize the frequency of manual integrations (poorly defined chromatographic peak) that would be required at the lower value.

Oil Sample Analyses. Analyses of cod liver oil proved sensitive for most analytes despite a decrease in the relative recovery of most analytes. The organic content also produced spectral interferences that made some determinations difficult and the determination of several analytes impossible. The relative recoveries for the analytes are very similar for both water spike and vacuum spike techniques, suggesting the oil matrices are not as difficult as soil to spike. Both spike techniques produce `A<sub>T</sub>,`A_dev, and`a<sub>T</sub> results that are accurate and quite similar. Table 3 presents the`A_T,`A<sub>dev</sub>, and`a_T as group results (vacuum spike of cod liver oil).

The analytes, N-nitrosodibutylamine and 2-methylnaphthalene, were poorly recovered from the oil matrix and had relative recoveries below 1%. The`A_T for these analytes indicate the determination of analytes with boiling points greater than 240 °C in oil should not be considered quantitative. Warming the sample to improve recoveries would not be desirable because much of the matrix would also be vaporized and likely become system contamination.

The MDLs for analytes in an oil matrix are listed in Table 3. The limits are much improved compared with earlier work,¹ reflecting the ability of the surrogates to correct for matrix effects and produce more precise results. A 0.2-g oil sample seems well suited for waste oil analyses and the MDLs for the smaller sample sizes meet current criteria and the smaller sample minimizes matrix spectral interference. Of course, such interferences would vary by sample.

The analyses of the various matrices and the use of specific compounds as surrogates have identified some additional matrix effects that should be monitored. The use of ethyl acetate-2C¹³ has evolved from an a-surrogate for water analyses to a check surrogate to detect potential degradation of analytes for soil analyses. The variation of accuracy predictions of the nitrosamines show that an additional surrogate might be beneficial if it contains an amine or nitroso functional group.

The check surrogates, methylene chloride-d₂, benzene-d₆, 1,1,2-trichloroethane-d₃, and bromofluorobenzene are effective in monitoring the behavior of the majority of analytes. The surrogate 1,2-dichloropropane-d₆ has spectral interferences from the higher hydrocarbon-containing sample matrices. The frequency of manual integrations that were necessary makes this compound undesirable for routine use. The surrogate, 1,1,2-trichloroethane-d₃, was also affected but to a lesser degree.

Analytes that have either boiling points or relative volatilities at the limits of the method should be scrutinized closely. The check surrogates acetophenone-d₅ (boiling point 202 °C and a_K-value 161) and naphthalene-d₈ (boiling point 217 °C and a_K-value 18) are very useful in identifying how the matrices are affecting the higher boiling analytes (to 220 °C) and those with higher relative volatility (a-value to 200).

While all the analytes studied perform very well in an ideal matrix such as distilled water, a high content of organic matter can greatly impact the accuracy of analyte determinations. The analyst, however, is warned when such effects occur (in addition to the performance of check surrogates) by reviewing a_T (or r_T) values for the analytes as they will normally exceed 50% when accuracy of a determination is outside a 70 to 130% window for the analyte.

Conclusion

This work demonstrates the accuracy and sensitivity of VD/GC/MS and surrogate-based matrix corrections. The list of analytes is not limited to VOCs but can be expanded to include compounds that have partition coefficients up to 15,000 and boiling points up to 245 °C. The prediction of matrix effects on analyte recoveries with confidence intervals provides the analyst and data users with a powerful tool to interpret data and eliminates the need for additional analyses, such as matrix spikes, to estimate such effects. The accuracy of VD/GC/MS allows the analyses of standard solutions (distilled water) to be used for the determination of analytes in different matrices as well as sample sizes without compromising the usefulness of the data.

Supplementary Material Available: Four tables (14 pages) containing individual analyte data for the matrices analyzed are available as supplementary material. Table 4 is titled "Percent Accuracy of Surrogate Predictions for Water Mixtures" and lists the`A<sub>T</sub>,`A_dev, and`a<sub>T</sub> for each analyte from the analyses of water, water/glycerin, water/salt, and water/soap matrices. Table 5 is titled "Method Detection Limits in Parts per Billion" and lists the MDLs for each analyte in water, soil, oil, and waste oil matrices. Table 6 is titled "Percent Accuracy of Surrogate Predictions for Soils" and lists the`A_T,`A<sub>dev</sub>, and`a_T for each analyte from the analyses of Soils #1-3 using both vacuum and water spikes. Table 7 is titled "Percent Accuracy of Surrogate Predictions for Oil using Water and Vacuum Spikes" and lists  the`A<sub>T</sub>,`A_dev, and`a_T for each analyte from the analyses of cod liver oil.

Acknowledgment

The EPA, through its Office of Research and Development (ORD), funded and performed the research described here. It has been subjected to the Agency's peer review and has been approved as an EPA publication. Mention of trade names or commercial products does not constitute endorsement or recommendation for use. The U.S. Government has the right to retain a non-exclusive, royalty-free license in and to any copyright covering this article.

References

1) Hiatt, M.H.; Youngman, D.R.; Donnelly, J.R. Anal. Chem. 1994, 66, 905-908.

2) Hiatt, M.H.; Farr C.M. Anal. Chem. 1995, 67, 426-433.

3) McDaniel, J.A. The Effect of Water Added to Soils on the Analysis of Volatile Organic Compounds; Book of Abstracts; Pittsburgh Conference: New Orleans, LA, 1992; Abstract 711.

4) U.S. Environmental Protection Agency, Handbook of RCRA Ground-Water Monitoring Constituents: Chemical & Physical Properties, Office of Solid Waste, Washington D. C. 1992.

5) Li, J.; Dallas, A.J.; Eikens, D.I.; Carr, P.W.; Bergmann, D.L.; Hait, M.J.; Eckert, C.A. Anal. Chem 1993, 65, 3212-3218.

6) Vitenberg, A.G. J. Chromatogr 1991, 556, 1-24.

7) Snyder, J.R.; Dawson, S.A. J. Geophys. Res. 1985, 90, 3797-3805.

8) U.S. Environmental Protection Agency, Test Methods for Evaluating Solid Waste, SW-846, Office of Solid Waste, Washington, D. C., 1992.