method validation an - ACS Publications - American Chemical Society

Analysis Center], Bay St. Louis, MS; the extracts were analyzed at the EPA EnvironmentalMonitoring Systems Laboratory,. Research Triangle Park (EMSL-R...
1 downloads 0 Views 798KB Size
Anal. Chem. 1986, 58,46L-468

463

Determination of 2,3,7,8-Tetrachlorodibenzo-p -dioxin in Human Milk at the 0.1-10 Parts-per-Trillion Level: Method Validation and Survey Results R. G. Heath* U.S. Environmental Protection Agency, OPTSIOTS, E E D (TS798),Design and Development Branch, Washington, D.C. 20460

R. L. Harless* U.S. Environmental Protection Agency, ORD,Environmental Monitoring Systems Laboratory (MD-77), Research Triangle Park, North Carolina 2771 1

M. L. Gross* and P. A. Lyon Department of Chemistry, University of Nebraska-Lincoln,

Lincoln, Nebraska 68588

A. E. Dupuy, Jr.,* and D. D. McDaniel

U.S. Environmental Protection Agency, OPTSIOPP, BUD, COB, Environmental Chemistry Laboratory, Bay St. Louis, Mississippi 39529

Capillary column high-resolution gas chromatography/highresolutlon mass spectrometry (HRGWHRMS) and packed column gas chromatography/high-resolution mass spectrometry (LRGWHRMS) were evaluated for the analysls of 0.1-10 parts-per-trllilon (pptr or pg/g) of 2,3,7,8-tetrachlorodIbenzop-dloxln (2,3,7,8-TCDD) In l o g samples of human milk. Thls was done as a blind study In two laboratories. The regression equatlons representing the results are y = 0 . 7 9 ~ 0.34 and y = 0 . 6 5 ~ 0.06 for the HRGWHRMS and LRGWHRMS methods, respectively, where x Is the amount of TCDD added (parts per trillion) and y Is the amount reported (parts per trlliion). The limits of detectlon range between 0.1 and 0.6 pptr and 0.5 and 6.0 pptr for the HRGC/ and LRGCIHRMS methods, respectively. No false posltive and only two false negative reports were made for a total of 38 analyses of splked milk and three analyses of standard solutions. The methods were applied to the analysls of 103 samples of human milk from mothers residing In areas where herblcldes contalnlng 2,3,7,8-TCDD had been used. No 2,3,7,8-TCDD could be detected at an overall medlan detection limit of 2 pptr and a range of 0.1-6 pptr.

+

+

Intense interest has been focused for over a decade on the family of compounds known as chlorinated dibenzo-p-dioxins (CDDs). The 2,3,7,8-TCDD, one of 22 tetra isomers, is reportedly the most toxic member of the CDD family. Toxicological properties of 2,3,7,8-TCDD,including acute oral LD, ( I ) , teratogenicity (a),carcinogenicity (3),and fetotoxicity (4) in specific animal species, have been documented. 2,3,7,8TCDD is formed as a trace contaminant in the production of 2,4,5-trichlorophenol (TCP). The 2,4,5-TCP and its salts, contaminated with parts-per-billion (ppb or ng/g) levels of 2,3,7,8-TCDD, are used as intermediates in the preparation of various algicides, bactericides, fungicides, and such phenoxy herbicides as silvex, ronnel, and 2,4,5-trichlorophenoxyacetic acid (2,4,5-T). Recently, 2,3,7,8-TCDD and several isomers were identified in an o-chlorophenol product (5). TCDD isomers have also been identified in 2,4-D products ( 6 ) and in pentachlorophenol (7); however, 2,3,7,8-TCDD has never 0003-2700/86/0358-0463$0 1.50/0

been detected in either product. The manufacture, use, and disposal of these chemical products have provided a vehicle for 2,3,7,8-TCDD, other TCDD isomers, and other CDDs to enter the environment. In recent reports (8-14), it was also shown that CDDs, including 2,3,7,8-TCDD, may be emitted into the environment from specific combustion processes. Millions of kilograms of phenoxy herbicides contaminated with parts-per-billion levels of 2,3,7,8-TCDD have been used to control weeds and brush on rights-of-way, range land, in rice growing areas, and in forest management. Because of the wide usage of these herbicides and the extreme stability and toxicological properties of 2,3,7,8-TCDD, the risk of human exposure to and food chain contamination by 2,3,7,8-TCDD have been a major public concern. The presence of partsper-trillion (pptr or pg/g) levels of TCDD has been reported in a wide variety of sample matrices including fish (15-17), human tissue (18-20), soils and sediments (21-23), and various types of biological samples (24). The analytical methodology used to determine parts-pertrillion levels of TCDD and the results obtained have been the subjects of intense scrutiny by the scientific community during the past decade. These determinations are complicated by the presence of higher levels of naturally occurring compounds and chlorinated industrial pollutants, e.g., polychlorinated biphenyls (PCBs), dichlorodiphenylethylene (DDE), benzyl phenyl ethers, and hydroxydiphenyl ethers. Therefore, extremely efficient sample preparation procedures coupled with highly sensitive and specific gas chromatography/mass spectrometry detection methods are required for determining conclusively the presence or absence of partsper-trillion and sub-part-per-trillion levels of TCDD. The validity and credibility of TCDD determinations at low parts-per-trillion levels, especially those involving human subjects, are enhanced by incorporating method validation studies, stringent quality assurance programs, stringent analytical criteria for confirmation of TCDD, and multiple laboratory participation. We report here the results of a 1979-1980 study undertaken by the U.S.Environmental Protection Agency (EPA) and the University of Nebraska to investigate whether trace partsper-trillion residues of TCDD were present in the milk of nursing mothers residing in regions of widespread 2,4,5-T 0 1986 American Chemical Society

464

ANALYTICAL CHEMISTRY, VOL. 58, NO. 2, FEBRUARY 1986

and/or silvex (2,4,5-trichlorophenoxypropionicacid) usage. In particular, we present a detailed statistical study of the method evaluation and the quality control testing incorporated in the study. Two analytical methods were employed and evaluated. Packed column low-resolution gas chromatography/ highresolution mass spectrometry (LRGC/HRMS) was used to screen all the actual samples of human milk. The method is specific for the family of TCDD isomers, but is not specific for the 2,3,7,8-isomer. Capillary column high-resolution gas chromatography/high-resolution mass spectrometry (HRGC/HRMS) was used as a confirmatory method. The latter had been demonstrated to be specific for 2,3,7,8-TCDD after this work was completed (see next section).

EXPERIMENTAL SECTION Samples of human milk for the study were donated by 103 nursing women, 72 of whom resided in western Washington, Oregon, and northwest California, where use of 2,4,5-T and silvex in forest management had been widely practiced for at least a decade. The remaining 31 women, who served as “controls”, resided in Alaska and southern California where any use of these herbicides was considered incidental. The method evaluation (ME) and quality control (QC) samples were prepared from pooled human milk donated by the LaLeche League of Bowie, MD. All samples-method evaluation (ME), quality control (QC), and survey-were prepared and extracted at the EPA Environmental Chemistry Laboratory (ECL) [formerly the Toxicant Analysis Center], Bay St. Louis, MS; the extracts were analyzed at the EPA Environmental Monitoring Systems Laboratory, Research Triangle Park (EMSL-RTP), NC, and at the University of Nebraska-Lincoln, The analytical methodology used for these investigations is fully described elsewhere (25, 26). Briefly, the sample preparation involved the following: fortification of 10-g human milk samples with 2.5 ng of 37C14-TCDD,the internal standard; saponificqtion with hot alkali followed by extraction with hexane; treatment with concentrated sulfuric acid; alumina column chromatography cleanup; and concentration of alumina column extract to 60 FL for shipment and analysis. The quantity of internal standard was chosen to minimize any ion signal at mlz 322 from incomplete labeling of 37C1-TCDD(25) and still permit recovery checks by using low-performance mass spectrometers. At the time of the study, this was done periodically so that recovery problems at the extraction laboratory could be identified before large numbers of samples were extracted and queued for high-resolution MS analysis. The cleanup procedure is appropriate for this type of relatively clean sample matrix. However, it is not suitable for detecting 2,3,7,8-TCDDat levels below 2-5 pptr in more complex samples such as certain fish or human tissue. The lack of suitability stems from other sample contaminants, which are not adequately removed in the cleanup. The contaminants yield ions that contribute to a general base line signal increase with concomitant increase of limit of detection for analyses done with sector mass spectrometers. Part of the base line increase is probably due to metastable ion decompositions. The extract aliquots (60 pL each) were shipped by ECL to the analytical laboratories in a blind fashion; i.e., the mass spectroscopist did not know the spiking levels of any of the ME or QC samples nor could he distinguish between extracts of survey and QC samples. The latter extracts were randomly interspersed with those of survey samples prior to shipment. HRGC/HRMS (25) and LRGC/HRMS (26) were used to determine sample preparation efficiency,the presence or absence of TCDD, and the minimum limit of detection for TCDD. Several survey samples were initially fortified with 10.0 ng of internal standard. The sample extracts were subjected to HRGC/HRMS multiple ion monitoring analysis a t EMSL-RTP and to LRGC/HRMS signal averaging analysis for TCDD residues at the University of Nebraska. The extract masses used in these determinations were m/z 327.8847, corresponding to 37C14-TCDD, the internal standard, m / z 319.8965, C12H40235C1,,and mlz 321.8935, C12H40235C127C1, corresponding to native TCDD. Because the samples spiked with 10 ng of internal standard gave signals at m / z 321.8935 corresponding to 1pptr of TCDD, the

Table I. Recovery and Limit of Detection for Method Evaluation Samples Analyzed Using HRGC/HRMSasb

sample no. HTSR-4 -3 -8 -2

-7 -10 -9 -5 -1

-6

native TCDD added reported 0 0.1

0.2 0.3 0.5 0.65 0.75 0.9 1.0 2.0 3.0

-11 -12

5.0

-13”

0

nd’ 0.2 nd 0.6 0.4 1.4 0.6

0.9 1.9 1.4 3.0 4.0 nd

limit of detectiond

recovery, %

0.3

64 68 51 72 64 52 73 68 50 84 72 100 50

0.1

0.2 0.2 0.2 0.3 0.4 0.4 0.3 0.2 0.5 0.5 0.2

“All samples except no. 13 (a method blank) are 10-g mothers’ milk samples; all values except percentages are in parts per trillion. *Internal standard was 2.5 ng of 37C1-TCDD/10g of sample. nd, not detected. 2.5:l.O S/Ncriterion. spiking level was reduced to 2.5 ng of internal standard. The signal at m/z 321.8935 was due to incomplete labeling of the “C14-TCDD (25). For HRGCIHRMS, a 30 m X 0.25 mm i.d. SE-30 WCOT glass capillary column was used. Three ions were monitored at a mass resolution of 5000-10000 (10% valley): m / z 327.8847, m / z 321.8935, and m / z 319.8965. Signals were obtained in a Sample-and-hold fashion at each of the masses cited above (no mass profiles were acquired). At the time of this work, a 2,3,7,8-TCDD isomer-specific analysis had not yet been reported principally because of lack of standards for all isomeric TCDD’s, but an HRGC/LRMS method was later described by Buser and Rappe (27). Although the capillary column chromatography used was not proven to be isomer specific for 2,3,7,8-TCDD, the lack of specificity is not an issue because no TCDD was detected in any of the survey samples (see Results and Discussion). The LRGC/HRMS method involved the use of a 183 cm X 0.64 cm 0.d. glass column containing 3% OV-3 on Supelcoport. Later on during the study, a 183 cm X 0.20 cm i.d. glass column containing 0.60% OV-17 + 0.40% Poly s-179 coated on 100% methyl silicone bonded to 8O/lOO mesh Chromosorb WAW was used in lieu of the OV-3 column. Although this chromatography is not specific for the 2,3,7,8-TCDD, the method of analysis is suitable for screening samples because of rapid sample throughput and adequate limits of detection as demonstrated in this paper. Compound class specificity was achieved by dual ion monitoring with one channel centered at m / z 327.8847 and the other at 321.8935. Complete mass profiles were acquired by scanning at a frequency of 2 Hz corresponding for each channel to a mass range of 300 ppm (0.096 atomic mass units). Criteria for confirming the presence of TCDD were reported elsewhere (25). Five criteria were used in this study: (1)correct retention time as determined using the internal standard retention time as a benchmark, (2) correct isotope ratio for m / z 320 and 322 (0.77 & 0.10), (3) correct accurate mass of mlz 321.8935, (4) correct response for coinjection of second aliquot of sample fortified with 37C1-TCDDand authentic 2,3,7,8-TCDD, and (5) signal-to-noise (S/N) response to TCDD of at least 2.5:l.O (limit of detection definition used here). The concentrations of TCDD were calculated by using the usual internal standard method (25,26). As such, the concentrations are corrected for losses in the extraction and cleanup as assessed by measuring the percent recovery of the internal standard. The percent recoveries are specified in all the data tables.

RESULTS AND DISCUSSION Method evaluation data for the HRGC/HRMS and the LRGC/HRMS methods are presented in Tables I and 11, respectively; QC data are given in Table 111. From these data we have calculated for each method (28, 29): (a) 1’inear regression statistics for “TCDD reported” on “TCDD added”;

ANALYTICAL CHEMISTRY, VOL. 58, NO. 2, FEBRUARY 1986

465

Table 11. Concentration of T C D D Recovery and Limit of Detection for Method Evaluation Samples Analyzed Using LRGC/HRMS”?*

sample no.

added

MMVS-10 -23 -30 -11 -26 -31 -24 -33 -28 -29 -25 -34 -27 -32

0 0 0 1.0 1.0 1.0 3.0 3.0 5.0 5.0 7.0 7.0 11.0 11.0

native TCDD reported

limit working value

of detectiond

recovery, %

2 2 2,2 6 1, 2 0.8,1 1 2 1 2, 1 1 1, 2 1 1.8,1

60 33 60 30 60 65 55 70 55 75 60 65 60 65

ndc nd nd, nd nd nd, nd

1.3,nd 2.5 2.3 3.7 2.5,2.3 6.0 4.5,7.1 7.0,7.2 6.4,6.0

“All samples are 10-g mothers’ milk samples; all values except percentages are in parts per trillion. *Internal standard was 2.5 ng of s7C1-TCDD/10g of sample. not detected. 2.5:l.OS/N criterion.

-

7

/.

-Regression Line

,/

/,

----95% Confidence Limits 6 - for Regression Line

-RegressionLine

------95% Confidence Limits for Regression Line

- - - - 95% Confidence Limits

/

for Individual Samples

./

,’

’.,

0

0 Method Blank/Standard

0,’

0 Method Evaluation A Quality Assurance

. .

X=TCDDADDED (ppt) Flgure 1. Reported concentration vs. concentration of TCDD actually added to human milk samples, analyzed by using capillary column gas chromatographylhigh-resolutionmass spectrometry (HRGC/HRMS). The theoretical line Is Y = X (not shown); the regression line is Y = 0.79X 4- 0.34; r = 0.975; n = 13. All lines are truncated at the mlnlmum level of detection (ca. 0.1 pptr).

(b) the accuracy and precision with which TCDD added a t concentrations in the range of 0.1-10 pptr was extracted and quantified; (c) range and percentage distribution of the limits of detection; (d) incidence of erroneous reports; (e) the inverse regression line for estimating true TCDD concentrations from reported concentrations (inverse prediction); and (f) probability estimates for detection of specific levels of TCDD in human milk. Calculations for each method are based on its combined ME/QC data. Regression of TCDD Concentrations ”Reported”on Concentrations “Added”. The least-squares linear regression TCDD lines and equations for ”reported” (Y) vs. “added” (X) concentrations are presented for the HRGC/HRMS and the LRGC/HRMS, respectively, in Figures 1 and 2. For the data in both figures, the line Y = X is the theoretical line (not shown) for perfect spiking, extraction, and quantification. Two sets of 95% confidence bounds are shown: the pair of lines closer to the regression line depicts the bounds for the line per se; and the pair more removed are predictive bounds for individual analyses of samples spiked at given levels. All lines

I

.

I

I

I

I

I

I

I

AI

1

1

I

I

,

I

I

I

I

I

I

I

I 2 3 4 5 6 7 8 91011 I213 X=TCDD ADDED (ppt) Flgure 2. Reported concentration vs. concentration of TCDD added to human milk samples, analyzed by using packed column gas chromatographylhigh-resolutionmass spectrometry (LRGClHRMS). The theoretical line is Y = X (not shown); the regression line is Y = 0.65X - 0.08;r = 0.915; n = 18. All lines are truncated at the minimum level of detection (ca. 0.6 pptr).

have been truncated near either axis at the point of the lowest reported limit of detection. Reports of “not detected” are shown, by spiking level, in the space below the X axis. The slope of 0.79 for the HRGC/HRMS regression line (95% confidence limits of 0.67 to 0.91) is somewhat less than the theoretical slope of 1. The calculated and theoretical lines intersect at X = 1.6 pptr, denoting a slight positive reporting bias below that spiking level and a negative bias above it, which increases to -1.1 pptr for samples spiked at 7 pptr. Precision can be expressed in terms of the confidence limits for the line itself and for predicted results of single analyses of individual samples. Confidence limits for the regression line deviate in a range from f0.2 pptr at X (2.1 pptr) to f0.6 pptr at X = 7.0 pptr. Those for individual analyses deviate from the regression line in a range from hO.9 pptr at X to fl.1 pptr at X = 7.0 pptr. For the results obtained by using LRGC/HRMS, the slope of 0.65 for the regression line (95% confidence limits 0.50-0.80) is also less than the theoretical slope of 1. The calculated and

466

ANALYTICAL CHEMISTRY, VOL. 58, NO. 2, FEBRUARY 1986

Table 111. Concentration of TCDD. Recovery and Limit of Detection for Quality Control Samples Analyzed Using either HRGC/HRMS or LRGC/HRMS”

packed column GC/HRMS sample no. Nebr RTP MMG-72 -137 -55 -86 -112 -126 -147 -66 -120 -50 -77 -129

MME-12

-7 -22

-10 -60 MMG-41 -116 -128 -142 -125 -144

MME-17 -2 -19

TCDD (working

native type of sampleb

TCDD

M M M M M M M M M M M M M M M B B

0 1.2 2 2 3 3 3 3.5 4 5 6 6 7 7 9 0 0 0 0 0 2 2

B B B S S

added

reported

value)

ndc 0.6, nd nd nd 1.5, 2.2 1.3 1.2, 1.3 2 3.3, 2.4 3.5

(0.6) (1.9) (1.3) (2.9)

ndd 2.9 7 nd nd

nd nd 0.9, 0.5, 1.5 1.0, 1.0

(1) (1)

limit of detection

carillarv . “ column GC/HRMS TCDD limit of recovery,

recovery, %

1 0.6, 1.0 3 3 0.5, 0.6 0.6

85 85 65 65 60 50

1.0, 0.6 1.6 0.6, 0.3 1 3 0.7

85 95 45 80 65 70

1 2 1 0.8 1

75 50 45 60 45

0.5, 0.2, 0.8 0.8, 0.4

60 75

reported

detection

ndd

0.6

100

3 3

0.5 0.5

62

6

0.5

loo+

nd nd

0.5 0.6

100 56

2

0.6

loot

%

loo+

“All values except percentage are in parts per trillion. Internal standard was 2.5 ng of 37C1-TCDD/10g of sample. bM, mothers’ milk; B, method blank: S, 2,3,7,8-TCDD standard. = not detected. dFalsenegative. e2.5:1.0 S/N criterion.

theoretical lines intersect essentially at the origin, indicating a constant negative reporting bias approximately equal to 35% of the spiking concentration. The 95% confidence limits for the regression line deviate in a range from f0.4 pptr at X (5.3 pptr) to fl.O pptr at X = 11.0 pptr. Those for individual analysis deviated in a range from fl.9pptr at X to f2.1 pptr a t X = 11.0 pptr. Limits of Detection. Limits of detection for the ME and QC analyses are listed in Tables I, 11, and I11 and are summarized in Table IV. We observe (see Table IV) that limits of detection for the 16 samples analyzed using HRGC/HRMS (12ME and 4QC samples, see Tables I and 111) ranged from 0.1 to 0.6pptr TCDD; 15 of the 16 limits (94%) did not exceed 0.5 pptr, and eight (50%) did not exceed 0.3 pptr. Limits of detection of 38 analyses (there were 21 analyses of ME and 17 of QC samples including duplicate runs, see Tables I1 and 111) done by using LRGC/HRMS ranged from 0.3to 6 pptr TCDD 37 of the limits (97%)did not exceed 3 pptr, 34 (89%) did not exceed 2 pptr, and 24 (63%) did not exceed 1 pptr. From these results, we estimate that, approximately 90% of the time, limits of detection for human milk analyses for TCDD conducted under conditions comparable to those herein will range from 0.1to 0.5 pptr, using HRGC/HRMS, and from 0.5 pptr to 2 pptr, using LRGC/HRMS. It is not surprising that better detection limits are obtained by using capillary column GC introduction into the mass spectrometer. For the analyses conducted as part of this study, the capillary column method incorporated peak-top monitoring whereby the mass spectrometer was focused on the peak maximum during the time period that a given ion was present (25). The packed column procedure was designed to acquire mass profiles that had bandwidths of 100 ppm of the ion mass observed at 10% peak height (mass resolution of 10000) by using a narrow scan over 300 ppm of the ion mass (25). Thus, the duty cycle for the LRGC/HRMS method was significantly poorer than that for HRGC/HRMS. Furthermore, a lower instantaneous concentration of sample was introduced into the ion source over a longer period of time by the packed column (peak width of ca. 40 8). As a result, longer time signal

Table IV. Estimated Probabilities of Detectionnfor Given Concentrations of TCDD (pptr) in 10-g Samples of Human Milk

limit of reported detectionb n cumul

percentages

TCDD

obsd

concn

fitted

detection probability

Capillary Column GC/HRMS 0.1 0.2 0.3 0.4 0.5 0.6

1 4 3 2 5 1

6 31 50 63 94 100

1 5 8 10 15 16

5 22 50 78 94 99 +

0.20--0.25 0.26-0.35 0.36-0.45 0.46-0.55 0.56-0.65 >0.66

0.05 0.21 0.48 0.74 0.89 0.95

Packed Column GC/HRMS 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 2 3 4 5 6

1 0 1 5 1 1 0 15 10 3 0 0

1 1 2 7 8 9 9 24 34 37 37 37

1

38



3 3 5 18 21 24 24 63 89 97 97 97 100

1.0 3.5 8 14 21 29 36 57 87 97 99 99 + 99

+

0.40-0.45 0.46-0.55 0.56-0.65 0.66-0.75 0.76-0.85 0.86-0.95 0.96-1.5 1.6-2.5 2.6-3.5 3.6-4.5 4.6-5.5 5.6-6.5 >6.5

nAdjusted for false negative probability of 0.05. LOD is 2.5:l S/N.

0.01 0.03 0.08 0.13 0.20 0.28 0.34 0.54 0.83 0.92 0.94 0.94+ 0.95

Criterion for

averaging was required for the packed column experiment. The signal-to-noise ratio was thereby reduced (and the detection limit raised) because of the presence of chemical noise. Erroneous Reports. We recognize the possibility of three types of erroneous reports: a false positive (FP) is defined as a positive TCDD value reported for an unspiked sample; a “false positive” (fp) is a positive value reported when the stated limit of detection exceeds the spiking level; and a ‘‘false negative” (fn) is a report of “not detected” (nd) when the

ANALYTICAL CHEMISTRY, VOL. 58, NO. 2, FEBRUARY 1986

487

Table V. Analytical Results for 2,3,7,&TCDDin Human Milk Survey Samples

area designation

Use area (2,4,5-T/silvex used in forest management) control area survey summary

donors residence

n

limits of detection, pptr percentile median 90th (range)

TCDD detected

10th

nda

nd nd

0.4 0.5 0.5

0.9 2 2

0.9 2

3

37C1,TCDDrecovery, %

median

(range)

(0.2-2) (0.5-3) (0.4-6)

60 75 67.5

(35-100)

(0.1-3) (0.2-3)

75 62.5

(35-100) (30-85)

Washington, W Oregon, W California, NW

20 34

Alaska California, S

23 8

nd nd

0.3

use area

72 31

nd nd

0.5 0.3

2

4

1.5

3

(0.2-6) (0.1-3)

70 67

(35-100) (30-100)

103

nd

0.5

2

3

(0.1-6)

70

(30-100)

control area total

18



-

1

3 4 ,.

(40-100)

(45-100)

and = not detected. stated limit of detection is less than the level of spiking. By definition, a report of nd when the limit of detection is equal to the spiking level is classified as a valid report and not an fn. There were no FP reports obtained by using either method among a total of five control milk analyses (one HRGC/; four LRGC/HRMS) and seven method blank analyses (three HRGC/; four LRGC/HRMS). Similarly, there were no fp reports of analyses by using either method among a total of 38 analyses of spiked milk samples (15 HRGC/; 23 LRGC/ HRMS) and three analyses of standard solutions. One fn each was reported among the results obtained by using both methods (Table 111),a relative frequency of 0.067 (1/15) for HRGC/HRMS and 0.043 (1/23) for LRGC/HRMS. Because the differences in the two frequencies do not approach statistical significance by x2,the data have been pooled to estimate a “working” probability of an fn report, which is essentially 0.05 (2/38 = 0.053) under laboratory conditions equivalent to those during the study. The 95% confidence limita (by binomial expansion) for this estimate are 0.01-0.18. Thus, we expect the true mean frequency of fns to lie between 1 and 18 per 100 analyses (P = 0.05), the best estimate being essentially 5 per 100 analyses. Estimates of True TCDD Concentrations. By use of the regression statistics developed from the ME and QC analyses, a “best estimate” of the true level of TCDD in a human milk sample and the statistical confidence limits for that estimate can be derived from a reported level. The method is that of inverse predi:tion for estimating the value of the independent variable ( X , the true TCDD level) for a given measurement of the dependent variable (Y, the reported level). The regression equations for these estimates are X = 1.26Y - 0.43 pptr for HRGC/HRMS and X = 1.53Y 0.09 pptr for LRGC/HRMS. The 95% confidence bounds for HRGC/HRMS range from approximately 1.2 to 1.5 pptr above and below the predictive line for reported values from 1 to 6 pptr. For example, the true TCDD level in a sample reported as 3 pptr will lie between 2.2 and 4.5 pptr (X = 3.4 pptr) unless the analytical result deviates sufficiently to be expected less than 5% of the time. The 95% confidence bounds for LRGC/HRMS range from 3.0 to 4.0 pptr above and below the predictive line for reported values from essentially 1 to 8 pptr. For example, the true TCDD level in a ?ample reported as 5.4 pptr will lie between 4.8 and 11.0pptr ( X = 7.8 pptr) unless the result deviates sufficiently to be expected less than 5% at the time. Confidence bounds for the means of duplicate (triplicate, etc.) analyses of the same sample would be narrower. For both methods, reported concentrations tend to represent conservative estimates of actual concentrations, with the exception of HRGC/HRMS reports below 1.7 pptr.

+

Probability for Detection of Given TCDD Concentrations. The question has been posed “Given that TCDD is present in a sample of human milk, what is the likelihood of ita detection when analyzed using a given method?” Such estimates of likelihood, (Le., of probabilities of detection) ‘require knowledge of the frequency distribution of previous detection limits and of the frequency of false negative reports. The estimates must be expressed in terms of the true (but unknown) level of TCDD in the sample. The appropriate probabilities of detection by using HRGC/HRMS and LRGC/HRMS are presented in Table IV. Derivation of these probabilities assumes a 0.05 probability of a false negative and a report of “not detected” whenever the limit of detection is equal to the true TCDD concentration. The “fitted” cumulative percentages for each analytical method have been derived from lines fitted to the cumulative observed percentages for detection limit categories plotted on normal probability paper (log normal for data obtained by using LRGC/HRMS). The data obtained by using the HRGC/HRMS method were fitted by employing least squares after transforming percentages to probits. The log normal fit for the LRGC/HRMS data was virtually perfect (by eye) after clustering the seven reports ranging from 0.5 to 0.9 pptr at 0.7 pptr, the mid-range. The detection probabilities were calculated by converting the fitted percentage values to proportions (i.e., by dividing by 100) and multiplying each by 0.95, the probability an analysis will not result in a false negative. The probability estimates are for single analysis of individual extractions of 10-g human milk samples. The probabilities are strictly predictive. Once a sample has actually been analyzed, any TCDD present either has or has not been detected, depending on the detection limit for that analysis relative to the TCDD concentration and the possibility of a false negative outcome. The detection probabilities for the range of levels tested indicate that one can expect TCDD to be detected approximately 74% of the time in samples containing 0.5 (0.46-0.55) pptr, 90% of the time in those containing 0.6 pptr, and 95% of the time in those containing at least 0.7 pptr if HRGC/ HRMS is employed for the analysis. If LRGC/HRMS is used, one can expect TCDD to be detected approximately 54% of the time in samples containing 2 (1.6-2.5) pptr, nearly 85% of the time in those containing 3 pptr, and approximately 95% of the time in those containing 5 pptr or more. These detection probabilities are intended as a guide in predicting the utility of potential investigations. Because these estimates are approximate and contain an element of subjectivity, we have not attempted to present statistical confidence limits. Analytical Results for Human Milk Survey Samples. Extracts of all samples were analyzed at the University of

468

ANALYTICAL CHEMISTRY, VOL. 58, NO. 2, FEBRUARY 1986

Nebraska by using the LRGC/ HRMS method; confirmatory analyses were performed at RTP for 28 (27%) of the samples by using HRGC/HRMS. Within replicates, the analysis with the lowest limit of detection was that used for the respective sample. Initially, samples (10 g) were fortified with 10 ng of 37C14-TCDDinternal standard; this level was subsequently reduced to 2.5 ng after it was discovered that injection of 1 ng of 37Cld-TCDD,the internal standard, onto either GC/MS used here gave a response a t m / z 321.8935 corresponding to positive results of approximately 1pptr for 10-g samples. This response was due to incompletely labeled 37C1-TCDD. For the 103 survey samples, 64 of the final analyses are based on 2.5-ng internal standard and 39 on 10-ng of fortification. No problems were encountered with the lower fortification level (corresponding to ca. 0.25 pptr native TCDD). Quantification of the lowest levels of TCDD encountered in any sample was done either by subtracting the response due to the internal standard at m / z 321.8935 or by using the m / z 319.8965 channel, which showed no detectable response due to internal standard. There were no confirmed detections of TCDD in any of the 103 human milk samples a t the indicated limits of detection (see Table V) (30). However, nine samples appeared to give positive results for TCDD at levels of 0.7 to 11 pptr during the f i s t round of analysis. The results were obtained by using LRGC/HRMS. Splits of these samples were then reextracted and reanalyzed using both HRGC/ and LRGCIHRMS without knowledge of the sample code. For all samples, no TCDD was found at detection limits lower than for the first analyses. The apparent positive detections may be due to handling errors in either the extraction or analysis laboratories or memory effects in the GC/MS. Their occurrence points to the need for careful interlaboratory validation of positive results obtained in low part-per-trillion level trace analysis.

ACKNOWLEDGMENT We thank N. C. A. Weerasinghe for his help in preparing the manuscript. LITERATURE CITED (1) Henck, I. M.; New, M. A.; Kociba, R. J.; Reo, K. S. ToxiCol. Appl. Pharmacol. 1981, 59, 405, and references cited therein. (2) Tuchmann-Duplessls I n "Accidental Exposure to Dloxins, Human Health Aspects"; Coulston, F., Ed.; Academlc Press: New York, 1983; p 201. (3) Wassom, J. S.; Huff, J. E.; Loprleno, N. Mutaf. Res. 1977/1978, 47, 141. (4) Poland, A,; Knutson, J. Annu. Rev. Pharmacol. Toxicol. 1982, 22, 517. (5) Harless, R. L.; Oswald, E. O., unpublished report, Memorandum to EPA Dioxln Project Coordinator, 1979. (6) Harless, R. L., unpublished Report, Memorandum to EPA Dioxin Project Coordinator, 1980. (7) Rappe, C.; Buser, H. R. I n "Chemical Hazards in the WorkplaceMeasurement and Control"; Choudhary, G., Ed.; American Chemical Society: Washington, DC, 1982; ACS Symposium Series, No. 149, p 319. (8) Buser, H. R.; Bosshardt, H. P.; Rappe, C. Chemosphere 1978, 7, 165. (9) Oiie, K.; Vermeulen, P. L.; Hutzlnger, 0. Chemosphere 1977, 6, 455.

(10) Bumb, R. R.; Crummett, W. 6.; Cutie, S. S.; Gledhill, J. R.; Hummel, R. H.; Kagel, R. 0.; Lamparksl, L. L.; Luoma, E. V.; Miller, D. L.; Nestrlck, T. J.; Shadoff, L. A.; Stehl, R. H.; Woods, J. S. Science 1980, 210, 385. (11) Eiceman, G. A.; Clement, R. E.; Karasek, F. W. Anal. Chem. 1979, 51,2343. (12) Harless, R. L.; Lewls, R. G. Workshop-Impact of Chlorlnated Dioxins and Related Compounds on the Environment, Rome, Italy, October 22-24, 1980. 13) Harless, R. L.; Dupuy, A. E.; McDaniel, D. D. "Human and Envlronmental Risks of Chlorlnated Dioxins and Related Compounds"; Tucker, R. E., Young, A. L., Gray, A., Eds.; Plenum: New York, 1983. 14) Weerasinghe, N. C. A.; Gross, M. L. I n "Dioxins in the Environment"; Kamrin, M. A., Rodgers, P. N., Eds.; Hemisphere Publishing Corp.: Washington, DC, 1985; pp 133-151. 15) Hariess, R. L.; Lewls, R. G. "Chlorinated Dioxins and Related Compounds, Impact on the Environment"; Hutzinger, O., Ed.; Pergamon: Oxford and New York, 1982. 16) Harless, R. L.; Oswald, E. 0.; Lewis, R. G.; Dupuy, A. E.; McDaniel, D. D.; Tai, H. Chemosphere 1982, 11, 193. 17) Stallings, D. L.; Petty, J. D.; Smlth, L. M.; Dubay, G. R. I n "Environmental Health Chemistry"; McKinney, J. D., Ed.; Ann Arbor Science: Ann Arbor, MI, 1981; p 177. IS) Leng, M. L.; Ramsey, J. C.; Brun, W. H.; Lavy, T. L. I n "Pesticide Resldues and Exposure"; Plimmer, J. R., Ed.; American Chemical Society: Washington, DC, 1982; ACS Symposium Series 182, p 133. 19) Ryan, J. J.; Williams, D. J.; Lau, B. P.-Y.: Sakuma. T. I n "Chlorinated Dloxins and Dibenzofurans in the Total Environment 11"; Keith, L. H., Choudhary, 0.. Rappe, C., Eds.; Butterworths: Stoneham, MA, 1984; p 205. (20) Gross, M. L.; Lay, J. 0.; Lyon, P. A,; Lippstreu, D.; Kangas, N.; Harless, R. L.; Taylor, S. E.; Dupuy, A. E. Envlron. Res. 1984, 33, 261. (21) Petty, J. D.; Smith, L. M.; Bergquist, P. A.; Johnson, J. L.; Stalling, D. L.; Rappe, C. I n "Chlorlnated Dioxins and Dibenzofurans in the Total Environment I"; Keith, L. H., Choudhary, G., Rappe, C., Eds.; Butterworths: Stoneham, MA, 1983; p 203. (22) Czuczwa, J. M.; Hltes, R. A. Envlron. Sci. Technol. 1984, 18, 444. (23) Smlth, R. M.; O'Keefe, P. W.; Aldous, K. M.; Hilker, D. R.; O'Brien, J. Environ. Sci. Technol. 1983, 17, 6. (24) Stanley, J. S. "Methods of Analysis for Poiychlorlnated Dlbenzo-p-Dioxins (PCDDs) and Polychlorinated Dibenzofurans (PCDFs) in Blologicai Matrices-Literature Revlew and Preliminary Recommendations"; Final Report to EPA, Contract No. 68-01-5915, Office of Toxic Substances, Washington, DC, 1984. (25) Harless, R. L.; Oswald, E. 0.; Wilkinson, M. K.; Dupuy, A. E.; McDanlel, D. D.; Tai, H. Anal. Chem. 1980, 52, 1239. (26) Gross, M. L.; Sun, T.; Lyon, P. A.; Wojinski, S. F.; Hilker, 0.R.; Dupuy, A. E.; Heath, R. G. Anal. Chem. 1981, 53, 1902. (27) Buser, H. R.; Rappe, C. Anal. Chem. 1980, 52, 2257. (28) Snedecor, G.; Cochran, W. "Statistical Methods", 7th ed.; Iowa State University Press: Ames, IA, 1980. (29) Feller, W. "An Introductlon to Probability Theory and Its Applications", 3rd ed.; Wiley: New York, 1968; Voi. I . (30) US. E.P.A. Office of Public Awareness [A-1071, Washington, D.C., U . S . Envlron. Prof. Agency J . 1980, 6, 3.

RECEIVED for review May 20,1985. Accepted October 7,1985. The research at UN-L was supported by the US.Environmental Protection Agency (CooperativeAgreement CR806847) and by the Midwest Center for Mass Spectrometry, an NSF Regional Instrumentation Facility (CHE-7818572 and CHE8211164). Although the research described in this article was funded in part by the United States Environmental Protection Agency through Cooperative Agreement No. CR806847 to M.L.G., it has not been subjected to Agency review and therefore does not necessarily reflect the views of the Agency and no official endorsement should be inferred. Mention of trade names or commercial products does not constitute endorsement or recommendation for use.