![[logo] US EPA](https://webarchive.library.unt.edu/eot2008/20090513152159im_/http://www.epa.gov/epafiles/images/logo_epaseal.gif){kind=logo}
Screening Processed Milk for Volatile Organic Compounds Using Vacuum Distillation/Gas Chromatography/Mass Spectrometry
Michael H. Hiatt
U. S. Environmental Protection Agency, National Exposure Research Laboratory,
Environmental Sciences Division, P.O. Box 93478, Las Vegas, Nevada 89193-3478
Phone: 702 798 2381. Fax: 702 798 2142. E-mail: hiatt.mike@epa.gov
Stephen Pia
U. S. Environmental Protection Agency, National Exposure Research Laboratory,
Environmental Sciences Division, P.O. Box 93478, Las Vegas, Nevada 89193-3478
Phone: 702 798 2102. Fax: 702 798 2142. E-mail: pia.stephen@epa.gov
[Note: minor content and formatting differences may exist between this web
version and the published version]
ABSTRACT
An adaptation of Office of Solid Waste and Emergency Response’ Test Methods for Evaluating Solid Waste Physical/Chemical Methods (SW-846) method 8261 to analyze milk for an expanded list of volatile organic compounds is presented. The milk matrix exhibits a strong affinity for organic compounds and the surrogate based matrix normalization described in method 8261 provided accurate results. This method had the sensitivity necessary to detect volatile organic analytes at or below maximum contaminant levels (MCLs) set by EPA for drinking water. In a survey of milk samples available in Las Vegas, Nevada 32 of 88 targeted volatile organic compounds (VOCs) were detected. Many of the detected VOCs have not previously been reported and a rationale for their presence in milk is presented.
INTRODUCTION
Cow’s milk is an important source of nutrition for infants and children. Milk production cannot be isolated from the environment and will be subject to potential contamination by organic compounds present in the air, food, and water consumed by cows. Studies of volatile organic compounds in milk have focused on milk quality (Urbach 1987, Imhoff and Bosset 1994). Studies that investigated contaminants in milk relating to environmental exposures are primarily pesticides and poly-chlorinated biphenyls (PCBs) (McLachlan 1996, Dowdy et al. 1996, Thomas et al. 1998). It has been reported that volatile aromatic compounds (Fabrietti et al. 2000, Imhoff and Bosset 1994) and disinfectant byproducts (Kroneld and Reunaned 1990, Miyahara et al.1995) are also present in cow’s milk. Our objectives were to investigate the presence of an expanded range of volatile organic compounds (VOCs) in commercially available milk samples and then to provide a rationale for the presence of those VOCs detected.
Some of the most prevalent of contaminants are VOCs and their mobility in the environment makes them detectable in most media. Monitoring drinking water for volatile organic chemicals that have potential health effects is done through EPA’s Office of Water (OW). The OW has published MCLs for these pollutants at http://www.epa.gov/safewater/mcl.html. We incorporated the detection of VOCs at or below drinking water MCLs as a method development goal for our study of milk. We made the assumption that it is important to detect VOCs in milk at concentrations required for drinking water as they are comparable sources of ingested VOCs.
The milk matrix presents far greater interferences compared to drinking water due to the organic and electrolyte content. In this work, we used the Office of Solid Waste and Emergency Response’ (OERR) Test Methods for Evaluating Solid Waste Physical/Chemical Methods (SW-846) method 8261 which is vacuum distillation coupled with gas chromatography/mass spectrometry (GC/MS). This method was selected for the analyses of milk samples due to its ability to compensate for severe matrix effects (including elevated organic and electrolyte content), and its application for a broad spectrum of volatile organic compounds (Hiatt 1995b, Hiatt 1997). Method 8261 incorporates a suite of surrogates to measure matrix effects that correspond to boiling point and relative volatility which provides the means to determine recovery of all analytes. Therefore the method was expected to mitigate the matrix effects presented by milk.
After developing the method performance data for milk samples, we tested the protocol performing analyses of milk available in Las Vegas Nevada retail outlets. The milk at these outlets were produced and processed in Arizona, California, Nevada, and Utah. Milk from throughout the southwest represented different processing plants and different dairy herds. Therefore the samples we obtained would be subject to differing VOC exposures from many sources as the milk moved from the cow to the container.
Because we could not differentiate various stages of milk processing for this study we have grouped transportation, packaging, and storage in the general category, processing. Hydrophilic VOCs, such as ketones and alcohols that are present in all milk samples may be attributed to some biological process such as metabolism or degradation and we refer to them as natural components. The concentrations of hydrophobic VOCs detected would be then evaluated as being background exposure correlating to their concentrations in air.
The partitioning of analytes in the environment (including air, water, and biota phases) have been modeled effectively using fugacities (Mackay 1979, Mackay et al.1985, Clark et al. 1988). It has been reported that the less volatile a compound, the less likely its predicted partitioning content in cow’s milk will reflect equilibrium with air due to a kinetically limited uptake (McLachlan 1996). However, the uptake of the more volatile hydrophobic compounds is not kinetically limited and their content in vegetation and cows milk would be expected to reflect equilibration with air (McLachlan 1996, Hiatt 1999). Given that fugacity models predict vegetation and the surrounding air are at equilibrium for VOCs we can extend this prediction to include dairy feed also being at equilibrium with air. Therefore we can interpret cows’ exposure to hydrophobic VOCs as solely through air and the partitioning throughout the cow would reflect the equilibrium.
To test our hypothesis we compare concentrations of VOCs in air predicted by content in milk to their reported concentrations in air. The concentration of VOCs in air data that were available for this study were yearly averages for large regions posted at http://www.sdas.battelle.org/airtoxics and referred to as Air Toxics Database. Given that the concentrations of VOCs were regional and that VOCs at a given location can vary dramatically over time (Hiatt 1999) we were interested how close predicted concentrations would come to the reported concentrations for VOCs in air.
EXPERIMENTAL
Vacuum Distiller. Samples were analyzed using a vacuum distiller coupled with GC/MS. The vacuum distiller serves to remove analytes from a sample by volatilizing the analytes in a reduced pressure environment. A condenser column serves to condense most of the water that is also being volatilized at the reduced pressure. The remaining volatile fraction passing through the condenser was cryogenically trapped in the vacuum distiller cryotrap. After vacuum distillation was complete, the cryotrap was ballistically heated to volatilize the distillate while a helium carrier gas swept the trap transferring the analytes to the GC/MS via a transfer line. The direction of flow at the cryotrap was controlled by a 6-port valve.
The vacuum distiller used in this study has been described in detail (Hiatt and Farr 1995a). The operating conditions were controlled by a programmable microprocessor (Bitstream, Las Vegas, NV). The sample container (100mL RB flask) containing 25 ml of milk sample was spiked with the surrogate compounds (5 μl) and attached to the distiller. During vacuum distillation the sample boiled vigorously with 1 - 1.5 g of water distilled and condensed within the condenser column. Operation of the vacuum distiller was performed through a custom Windows® program. This interface controlled operation of the system in autosampler mode with the GC/MS. The temperature of the six-port valve was maintained at 200 °C (Valcon E rotor). All lines transferring vapor between the sample vessel and the cryotrap were heated to 90-95 °C. The transfer line between the vacuum distiller (connected at the 6-port valve) and the GC was heated to 200°C. The vacuum distillations were 7.5 min. During the vacuum distillation the internal condenser temperature was 5.0 ± 2.5 °C and the cryotrap held at -145 ± 5 °C. During the desorb cycle the cryotrap was warmed to 100 °C. The transfer of distillate from the cryotrap to the GC was completed in 5 minutes. Between distillations, the condenser was warmed to 95 °C and flushed with nitrogen gas (10 psi) as the cryotrap was heated to 200 °C. After 16 nitrogen flushes, consisting of 0.1 minute nitrogen pressurization and 1.0 minute of evacuation the system was evacuated for 10 minutes.
GC/MS Apparatus. An Agilent mass spectrometer (Model 5973N) and gas chromatograph (6890N Series II with) with a 60-m x 0.25- mm i.d., 1.4-µm film thickness, Rtx®-VMS capillary column (RESTEK, Belefonte, PA ) was used for the determination of analytes vaporized from a sample during vacuum distillation. Gas chromatograph operating conditions were: 6 min at -25°C; 50°C/min ramp to 40°C, 5°C/min ramp to 120°C, 22°C/min ramp to 220°C, and isothermal at 220°C for 7.15 min, resulting in a total run time of 35 min. The transfer line between the GC and MS was held at 200°C. The injector was interfaced to the vacuum distillation apparatus by connecting the carrier inlet gas line of the GC to the cryoloop valve and then back to the injector. The injection was split 5:1 at a constant flow of 2 mL/min He. The injector was maintained at 220°C. The inlet pressure was 33 psi. The mass spectrometer was operated in the full scan mode (37 to 270 amu, scans/sec). The GC/MS operating system was WindowsNT® ChemStation (Agilent Technologies, Inc., Wilmington, DE).
Data Processing. The method used for the milk analyses was RCRA SW-846 method 8261. This method incorporates matrix normalization to masses measured by surrogate compounds in water. The normalization requires interpretation of matrix effects related to boiling point and relative volatility, αk corresponding to the partition coefficient Kwa. Surrogate compounds are grouped to allow interpretation of their recovery as a function over discrete ranges of boiling point or relative volatility values. By measuring the matrix effects with surrogates, the recoveries of non-surrogate compounds were predicted. The use of surrogates to measure matrix effects by these properties has been discussed elsewhere (Hiatt and Farr 1995a, Hiatt 1995b). View the surrogate groupings used in this study .
For this study, data processing was done using custom software developed and incorporated within the Agilent ChemStation software. This software provided calibration (5-point), check calibration, and report generation routines required by method 8261. This software follows method 8261 guidance and standard data format.
Standards. The analyte and surrogate solutions used in this study were obtained as three custom mixes from Supelco (Belefonte, PA). The first custom mix (20048982) contained the surrogate standards. This mix was diluted to 50 ng/uL of the compounds in methanol and used as our surrogate spike mix (5 μL). The analyte standard (spiking) mix was prepared from three separate mixes diluted to 1 mL with methanol. Fifty uL of analyte mix (20048345) which contained 75 compounds were added to the standard mix resulting in individual analyte concentrations ranging between 50 and 250ng/uL. Fifty uL of the nitrogen-containing analyte mix (20048346) consisting of nitrosamines, aniline, o-toluidine, pyridine and 2-picoline resulting in analyte concentrations of 1000 ng/uL were added. Twenty-five uL of the gases mix (48799-U) were added resulting in concentrations of these analytes at 50 ng/uL. View a complete list of these target volatile organic compounds . The nitrosamines: N-nitrosodimethylamine, N-nitrosomethylethylamine, N-nitrosodiethylamine and N-nitrosodipropylamine present in the standard mixture could not be reliably quantified due to poor chromatographic resolution. Propionitrile, also in the mix, frequently co-eluted with benzene-d6 and could not be reliably determined because the primary ions are common to both compounds. The reporting limits of some compounds were made higher than the method detection limits as spectral confirmation was not reliable due to spectral interferences from other compounds present in milk. A 5-point standard calibration curve was used to quantify the milk samples. The calibration was performed by vacuum distilling 5 mL water aliquots spiked with the analyte mix. The total masses of lowest concentration analytes in the calibration standard were 1, 20, 50, 150, and 250 nanograms (some analytes such as nitrosamines were 20 times more concentrated). Because the method 8261 normalizes the response to nanograms, sample calibrations were performed as mass per standard.
Milk samples. The survey of milk sources in Las Vegas, Nevada was conducted January 2002 through February 2002. Thirty-five samples were purchased off-the-shelf from eight local grocery stores. The samples were maintained at 4°C during storage, without preservatives, and the analyses were performed 1 to 5 days after purchase of the sample. The VOC data was generated using 25 mL aliquots of pasteurized whole, 2%, and 1% milk. The milk samples represents 8 processing plants (representing 8 brands) from Arizona, California, Nevada, and Utah.
RESULTS AND DISCUSSION
Method Development. The method development phase of this study began with the evaluation of surrogate-based matrix normalization to interpret results gathered from the vacuum distillation of milk samples. This approach, presented in method 8261 and documented in previous articles (Hiatt and Farr 1995b, Hiatt 1997) allows the use of instrument calibration using 5-mL water and the resulting calibration curve used for all other matrix types. Therefore we initially performed the vacuum distillation/GC/MS analyses on spiked 5-mL water and 25-mL whole milk and compared analyte responses to determine if differences corresponded to the matrix properties used in method 8261. The method measures matrix effects that correspond to a compound’s boiling point and relative volatility (water to air).
We determined the relative response of each analyte (response of analyte vacuum distilled from 25 mL whole milk compared to the response of analyte vacuum distilled from 5 mL water). The relative responses of analytes from the milk sample are shown in Figure 1 to correspond closely to the boiling point variable. While it has been found that relative volatility of octanol-air may be more appropriate in organic media (Hiatt 1997), that parameter is not as effective as boiling point as demonstrated in Figure 2. Therefore we did not pursue the use of the octanol-air relative volatility for this study.
This verification that boiling point and water-air relative volatility properties would be applicable to milk allowed the testing of milk samples spiked with both the surrogates and analytes. Studies were then conducted to evaluate method detection limits and recovery data for whole and 2% butterfat milk. We found the surrogate-based matrix normalization described in the method was effective in compensating for matrix effects and this approach proved viable for the analyte suite.
Method detection limits (MDLs) were generated using 25 mL aliquots of whole pasteurized milk in an abbreviated OERR approach. The OERR approach is to perform 7 replicate runs on 3 non-consecutive days where the concentrations of analytes are 3 times their estimated values. The MDLs are then calculated as 3 times the standard deviation. In our study the MDL values were determined as 3 times the standard deviation of 8 replicate runs on a single day using concentrations that were approximately 3 times the estimated MDL. Our approach required performing this routine at different concentrations to address all analytes in the study. If a resultant MDL was greater than the analyte concentration the MDL, the reported MDL was determined with a higher concentration replicate where the MDL was less than the analyte concentration.
View method performance by analyte. The subset of analytes investigated in this study that were detected in milk samples are presented in Table 1 . We found that the method performed well for all analytes of interest with a sensitivity needed to detect analytes with low MCLs (including 1,2-dibromoethane with an MCL of .05 and MDL of 0.03 ng/mL and 1,2-dibromo-3-chloropropane with a MCL of 0.2 and MDL of 0.03 ng/mL).
Milk analyses results. After method 8261 performance criteria were determined for milk, a survey of milk (whole, 1% and 2% fat) obtained locally was conducted. Those analytes that were detected are listed in Table1. Of the 35 milk samples analyzed, 32 of 88 analytes investigated were detected. No samples were found to contain an analyte above its drinking water MCL.
System blanks indicated most analytes were less than 1 ng/sample, with the exception of acetone and 2-butanone, having 10 and 5 ng/ total sample background respectively. Some of the analytes being detected are ubiquitous and could not be eliminated from our system background. For these compounds blank corrections to results were performed.
The polar compounds (natural components) acetone and 2-butanone (ethanol was also detected but this compound was not quantified) were detected in all samples and their presence in milk have been previously reported (Urbach 1987, Imhoff and Bosset 1994). Acetonitrile was also detected in all samples but not previously reported. 4-methyl-2-pentanone and methyl-tertiary butyl ether (MTBE) were detected frequently at very low concentrations and not previously reported. The hydrophobic compound, p-isopropylene was detected in all milk samples (as was limonene). This compound has been reported in elevated concentrations in foods due to its content in essential oils (Heikes et al. 1995) and likely present due to cows’ dietary intake.
Compounds that are likely due to processing include the disinfectant byproducts chloroform, bromodichloromethane, dibromochloromethane, and bromoform, were detected in all samples and was previously detected (Kroneld and Reunaned 1990, Miyahara et al. 1995). Additionally, styrene and methyl methacrylate were detected in some samples.
Due to the small population of the lower butterfat milk samples, comparisons of results by butterfat content is limited. However, it is obvious that the content of the more polar compounds (acetone, acetonitrile, and 2-butanone) is fairly constant and independent of butterfat content. The content of disinfection byproducts also appear to be independent of butterfat content. The naturally occurring essential oil component, p-isopropyltoluene, correlates with butterfat content as would be expected from uniform dietary uptake. Comparisons of analyte content suspected to be the result of environmental exposures (i.e., airborne compounds) are not supportable due to our small population of lower butterfat milk samples. In our study we did not restrict sample collection to sources when whole, 2%, and 1% milk were available and therefore the results in Table 1 reflect different populations (temporal and spatial). The concentration of these analytes exhibit a wide variation within a butterfat grouping demonstrating that environmental exposures are likely to vary greatly. Without sampling to minimize variations in environmental exposure, identifying those effects relating to butterfat content is obscured.
We next investigated those analytes that were found in milk and were suspected to be present due to environmental exposure. In the following section we perform simple two-phase modeling (butterfat and air) to determine if their concentrations could be related to ambient air contamination.
Modeling milk results to ambient air. Should VOCs in air and milk be at equilibrium it would be possible to predict the concentration of VOCs in one media if known in another. To perform this task we obtained available VOC analyte in air data for the regions where the milk was produced, determined what the analyte concentration in air should be if equilibrium between air and milk was obtained, and compared the results. For those comparisons, given the speculation that the reported air data was reflective of air at the dairies, we deemed an analyte would be the result of environmental exposure if the reported and predicted air data was within an order of magnitude. For this interpretation we did not use VOCs that we identified above as likely from a source other than air exposure.
The first step to identify a possible equilibrium condition is to apply an established model for the movement of analytes through the environment (including air, water, and biota phases). Modeling the movement of compounds in the environment is done effectively using fugacities (Mckay 1979,Mackay et al. 1985, Clark et al. 1988). Comparing fugacities in a system at equilibrium, the relative concentrations between phases can be described by their partition coefficients. Therefore, in a static environment the relationship between air, water, and fat can be described as the partition coefficients between octanol and air, water and air, and octanol and water. These partition coefficients labeled respectively, Koa, Kwa, Kow.
If we assume that in the environment that VOCs are near equilibrium between air and biota (i.e., milk, feed, grass) we can use available air content data to predict the content of the compounds in milk. While this predicted milk content data would be for an ideal occasion, we felt that if our survey results were within an order of magnitude of the predicted results we would associate the presence of those analytes as a reflection of air exposure.
The prediction on how a compound partitions into milk begins with an examination of how the compound should behave within milk. The compound is assumed to partition between the butterfat and aqueous phases according to the partition coefficient Kow which defined by:
| Kow = cf ∕ cw | (1) |
where cf is the concentration of the compound in the butterfat (ng/mL), and cw is the concentration of the compound in the aqueous phase (ng/mL). These results can be considered as:
| cm = cfff+cwfw | (2) |
where cm is the concentration of a compound in milk (ng/mL), ff is the fraction of butterfat, and fw is the fraction of the milk that is the aqueous phase. The concentration of the compound in the butterfat can then be expressed as
| cf = cm ∕ (ff + (1-ff )/Kow ) | (3) |
A compound in equilibrium between air and octanol (or in our case, butterfat) is described as
| Koa = 1000cf / ca | (4) |
where ca is the concentration of the compound in air (ng/L). For this study we are assuming static conditions and STP. Given the concentration of a compound in milk, the concentration of a compound in air is determined by
| ca = 1000cm ∕ Koa(ff + (1-ff)/Kow ) | (5) |
The most current air concentration data (yearly) for the counties where the plants reside was obtained from the Air Toxic Database. The data obtained from this database did not contain all the analytes we detected and there were not a consistent number of analytes for each region. The available data are listed in Table 2 .
The concentrations of analytes in air during milk processing were taken to be those reported for the region where the processing plants reside. Using equation 5 and the measured amount of analyte in air we predicted the concentration of the analytes in milk for the region. Predicted concentrations were compared to their measured concentrations of their respective counties. Table 3 presents the comparison (by city where plants were located) of the calculated air concentration to the measured concentrations. This table indicates which calculated concentration were within the range (R) of measured air concentrations. The table also presents the ratio of the calculated air concentration to the mean of the measured air concentrations. For example the calculated benzene in air concentration fell within the measured range for all samples 83% of the time and the mean of the calculated air concentrations were within an order of magnitude of the mean of the measured air concentration. A compound’s presence in milk was considered in equilibrium with its surroundings when that compound’s calculated air concentrations agreed with the measured air concentration presence in milk. Compounds that appeared to be in equilibrium we classified as ambient compounds. Analytes included in the ambient grouping were primarily analytes that were considered typical fuel components. This included MTBE (marginally), benzene and the alkyl aromatic compounds.
The majority of analytes we detected are ubiquitous and are reported as being continually present in air samples collected from the southwest. While our sampling is small we were able to speculate that their concentrations were likely a reflection of their concentration in air. These compounds were typically components of fuel including MTBE and the aromatic VOCs.
The concentrations of a few compounds in milk that were being investigated for correlation to their concentration in air greatly exceeded what could be attributed to their concentration in air. This could be due to the fact that the cows’ exposure reflects a more localized environmental exposure than our regional data. These compounds could also have been introduced from a multitude of sources during processing. The solvents trichloroethene, tetrachloroethene, and 1,4-dichlorobenzene were among this group and were detected in most milk samples. The compounds, 2-chlorotoluene and 4-chlorotoluene were detected in only a few samples and the samples were from the same processing plant.
We have found that by using method 8261 for the analyses of milk samples we can obtain important data. This information can be baseline information that is indicative of ambient exposures or may identify instances when there are unexpected exposures. A much more intensive study combining air sampling with sampling of milk from cow to user would be required to identify sources in greater detail.
ACKNOWLEDGMENT
The U.S. Environmental Protection Agency (EPA), through its Office of Research and Development (ORD), funded and performed the analytical research described. This manuscript has been subjected to the EPA’s review and has been approved for publication. Mention of trade names or commercial products does not constitute endorsement or recommendation for use.
REFERENCE
(1) Clark, T., Clark, K., Paterson, S., Mackay, D., Norstrom, R. (1988) Wildlife monitoring, modeling and fugacity. Environ. Sci. Technol., 22:120-127.
(2) Dowdy, D., McKone, T., Hsieh, D. (1996) Prediction of chemical biotransfer of organic chemicals from cattle diet into beef and milk using molecular connectivity index. Environ. Sci. Technol. 30:984-989.
(3) Fabietti, F., Delise, M., Piccioli Bocca, A. (2000) Aromatic hydrocarbon residues in milk: preliminary investigation. Food Control 11:313-317.
(4) Heikes, D., Jensen, S., Fleming-Jones, M. (1995) Purge and trap extraction with GC-MS determination of volatile organic compounds in table-ready foods. J. Agric. Food Chem. 43:2869-2875.
(5) Hiatt, M., Farr, C. (1995a) Volatile organic compound determination using surrogate-based correction for method and matrix effects. Anal. Chem. 67:426-433.
(6) Hiatt, M. (1995b) Vacuum distillation coupled with gas chromatography/mass spectrometry for the analysis of environmental samples. Anal. Chem. 67:4044-4052.
(7) Hiatt, M. (1997) Analyses of fish tissue by vacuum distillation/gas chromatography/mass spectrometry. Anal. Chem. 69:1127-1134.
(8) Hiatt, M. (1999) Leaves as an indicator of exposure to airborne volatile organic compounds. Environ. Sci. Technol. 33:4126-4133.
(9) Imhof, R., Bosset, J. (1994) Quantitative GC-MS analysis of volatile flavour compounds in pasteurized milk and fermented milk products applying a standard addition method. Lebensm. Wiss U Technol. 27:265-269.
(10) Kroneld, R., Reunaned, M. (1990) Determination of volatile pollutants in human and animal milk by GC/MS. Bull. Environ. Contam. Toxicol. 44:917-923.
(11) Mackay, D. (1979) Finding fugacity feasible. Environ. Sci. Technol. 13:1218-1223.
(12) Mackay, D., Paterson, S., Cheung, B., Neely, W. (1985) Evaluating the environmental behavior of chemicals with a level III fugacity model. Chemosphere 14:335-374.
(13) McLachlan, M. (1996) Bioaccumulation of hydrophobic chemicals in agricultural food chains. Environ. Sci. Technol. 30:252-259.
(14) Miyahara, M., Toyoda, M., Ushijima, K., Nose, N., Saito, Y. (1995) Volatile Halogenated hydrocarbons in foods. J. Agric. Food Chem. 43:320-326.
(15) Thomas, G., Sweetman, A., Lohmann, R., Jones, K. (1998) Derivation and field testing of air-milk and feed-milk transfer factors for PCBs. Environ. Sci. Technol. 32:3522-3528.
(16) Urbach, G. (1987) Dynamic headspace gas chromatography of volatile compounds in milk. Journal of Chromatography 404:163-174.
 |
Click image to view full size. |
 |
Click image to view full size. |