Ann. occup. Hyg., Vol. 47, No. 6, pp. 503-517, 2003
© 2003 British Occupational Hygiene Society
Published by Oxford University Press
Biological Monitoring for Selected Herbicide Biomarkers in the Urine of Exposed Custom Applicators: Application of Mixed-effect Models
1 National Institute for Occupational Safety and Health, 4676 Columbia Pkwy, Cincinnati, OH, 45226, USA; 2 Department of Mathematical Sciences, University of Cincinnati, Cincinnati, OH 45221, USA
Received 28 February 2002; in final form 26 February 2003
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONBACKGROUNDMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
Metabolites and/or parent compounds of the herbicides atrazine, alachlor, metolachlor, cyanazine and the 2-ethylhexyl ester of 2,4-dichlorophenoxyacetic acid (2,4-D) were measured in the urine of 15 custom applicators who each provided from five to seven 24 h urine samples during a 6 week period (n = 87). Each applicator provided a pre-season urine sample and a reference population (n = 46) provided first-morning urine samples. Urinary biomarkers were measured by either immunoassay or gas chromatography. During the spraying season, the geometric mean amount of alachlor mercapturate equivalents (eq.), atrazine eq., 2,4-D and metolachlor mercapturate eq. excreted in 24 h was 17, 19, 110 and 22 nmol, respectively. Mixed-effect models were used to determine predictors of the amount of atrazine eq. and 2,4-D excreted in 24 h. The specific days of herbicide spraying associated with increased biomarker excretion varied for the two analytes, and included one or more days prior to urine collection. This confirms the importance of collecting covariate information on day(s) most relevant to the biomarker of interest. The within-worker variance component, expressed as a geometric standard deviation (WGSD range: 2.5–2.9), was substantially larger than the between-worker component (BGSD range: 1.3–1.5) for the modeled biomarkers. Alachlor mercapturate eq. and metolachlor mercapturate eq. were detected in more than half of the applicator pre-season urine samples. All biomarkers were detected infrequently in the reference population. Evaluation of non-spray exposure determinants was limited by inclusion of prior day spraying, adjustment for time and the small sample size.
Keywords: atrazine; alachlor; metolachlor; 2,4-D; cyanazine; biological monitoring; mixed-effect models; variance components; immunoassays
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONBACKGROUNDMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
In the spring of 1996, the National Institute for Occupational Safety and Health (NIOSH, Cincinnati, OH) conducted an exposure assessment study of custom (or commercial) applicators who applied pre-emergent herbicides to corn and soybean fields using ground boom application equipment. This study included multiple measures of exposure to the herbicides alachlor, atrazine, cyanazine, metolachlor and the 2-ethylhexyl ester of 2,4-dichlorophenoxy acetic acid (2,4-D EH). A description of the herbicide application process and the results of air, handwash, dermal patch and saliva sampling have been reported elsewhere (Denovan et al., 2000; Hines et al., 2001). The results of biological monitoring for selected urinary biomarkers of these five herbicides are reported here.
Biological monitoring can provide an accurate measure of the absorbed dose. Ideally, the biomarker should not be present in the urine at the start of the monitoring period and the applied dose, absorption, metabolism, distribution and elimination parameters of the parent compound should be known. In addition, all urine should be collected until the biomarker is nearly completely eliminated and no additional exposure should occur during the collection period. These conditions are difficult to achieve in many work environments, including the one described here. In these cases, it is often more feasible to estimate the urinary body burden of a substance over a specified time period.
In this study, the body burden of selected herbicide urinary biomarkers was characterized among 15 custom applicators during a 6 week pre-emergent spraying season. The influence of recent herbicide spraying on biomarker levels was evaluated using mixed-effect regression models. Since investigators could not control the timing of an applicator’s exposure and the biomarkers had varying elimination half-lives, a sampling strategy that was optimal for all herbicides was not possible. Instead, a systematic sampling scheme that provided information about body burden at multiple points in time over the 6 weeks was used (Hines et al., 2001). Applicator urine sampling was conducted concurrently with patch, air, hand wash and saliva sampling. Urine samples were also collected from a reference population for comparison to applicator samples.
|
BACKGROUND |
|---|
|
TOPABSTRACTINTRODUCTIONBACKGROUNDMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
Biomarkers that might serve as exposure markers were identified for each of the five herbicides using information from human studies whenever possible (Fig. 1). The acetanilide herbicides, alachlor and metolachlor, are primarily metabolized in humans to their respective mercapturic acid conjugates and excreted in the urine (Driskell et al., 1996; Driskell and Hill, 1997). In rhesus monkeys, 92–94% of an intravenous (i.v.) alachlor dose was excreted in the first 24 h, and 88% of a dermal dose within 48 h, with the major metabolite being alachlor mercapturate (Feng et al., 1994). In a second study, radiolabeled alachlor administered i.v. or dermally to rhesus monkeys was eliminated primarily in the urine within 48 h, with an estimated elimination half-life (et1/2) of 3.5–6.5 h by the i.v. route (Kronenberg et al., 1988).
|
Atrazine is primarily metabolized in humans to its mercapturic acid conjugate (Lucas et al., 1993; Brady et al., 1998; Buchholz et al., 1999). The mono-N-dealkylated metabolites of atrazine (deethylatrazine and deisopropylatrazine) and its di-N-dealkylated metabolite (2-chloro-4,6-diamino-1,3,5-triazine) have also been detected in varying proportions in human urine (Catenacci et al., 1993; Lucas et al., 1993; Buchholz et al., 1999; Mendaš et al., 2000). The bi-dealkylated metabolite degrades rapidly when urine is stored for an extended time or thawed and refrozen (Bradway and Moseman, 1982), and is less suitable for biomonitoring. In rhesus monkeys dosed with i.v. [14C]atrazine, atrazine mercapturate elimination in urine peaked at 12 h after dosing and decreased to nearly baseline levels by 24 h (Brady et al., 1998). In dermally dosed human volunteers, atrazine, measured as 14C-labeled urinary metabolites, had a et1/2 of 24–28 h (Gilman et al., 1998). Human or primate data on cyanazine metabolism were not found.
Esters of 2,4-D are primarily hydrolyzed to the free acid in humans (Sauerhoff et al., 1977). Reported urine elimination rates for 2,4-D vary across studies. In six human volunteers administered 2,4-D by i.v., 100% of the dose was excreted in the urine within 120 h (et1/2 =13 h), whereas following dermal administration, only 5.8% of the dose was recovered after 120 h (et1/2 not reported) (Feldmann and Maibach, 1974). In one oral ingestion study of 2,4-D in a human volunteer (Kohli et al., 1974), 75% of the dose was excreted unchanged in the urine after 96 h, while in a similar study of five human volunteers (Sauerhoff et al., 1977), an average of 95.1% of the dose was excreted in the urine, primarily as the free acid (82.3%), and to a lesser extent as a conjugate (12.8%), with an average urine et1/2 of 17.7 h. In 26 occupationally exposed mixers, loaders and applicators applying predominantly an amine formulation of 2,4-D by ground application, a urine et1/2 of 35–48 h was estimated after a single exposure event (Nash et al., 1982). An average of 4.5% of a 2,4-D dose applied to the hands of five human volunteers was recovered in the urine after 144 h, with an estimated et1/2 of 39.5 h (Harris and Solomon, 1992). Thus, depending on the route of exposure and possibly formulation, the urine elimination half-life of 2,4-D may range from as low as 13 h to >40 h.
In summary, the above evidence suggests that biomarkers of the target herbicides are eliminated in the urine within several hours to several days of exposure and that for some biomarkers, accumulation in the body is possible depending on the frequency and intensity of exposure.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONBACKGROUNDMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
Sampling strategy
Fifteen custom applicators in Ohio, all males, were sampled at ~4 day intervals, including weekends, over six consecutive weeks (6 May–13 June 1996) as described in Hines et al. (2001). Samples were collected on the applicator’s scheduled day whether the applicator sprayed or not, unless the applicator was given the day off. For each sampled and unsampled day in the 6 week period, information on the type and amount of all herbicides applied was obtained for each applicator by interview and from records. Other covariate information, such as glove use, was recorded on only the sampled day. For each applicator, at least three males residing in the same township who did not use pesticides at work were recruited as a reference population. In this paper, the term ‘herbicide’ refers to the active ingredient. Participation in the study was voluntary and informed consent was obtained. This study was approved by the NIOSH Human Subjects Review Board.
Sample collection
A ‘pre-season’ urine sample was collected from each applicator in early April before herbicide spraying began. On each sampling day during the spray season, applicators collected all urine starting with that day’s first-morning void (day 0) through to the first-morning void of the next day (day 1). Urine was collected in five time periods starting on day 0: wake-up, wake-up to 12 p.m., 12 p.m. to 6 p.m., 6 p.m. to 12 a.m., and 12 a.m. to wake-up on day 1. A portable urine collection kit consisting of a 9.5 l cooler containing five 500 ml high-density polyethylene (HDPE) bottles labeled by time period, a refrigerant pack, pen, brown paper bags, time-piece and instructions were provided. Applicators were instructed to void into the appropriate bottle depending on the time of day and to write all void times on the label. Bottles were removed from the cooler as each collection period elapsed. Urine volume was measured with a clean 250 ml graduated cylinder, and aliquots were poured into 30 ml HDPE bottles. Aliquots were frozen and transported to the laboratory on dry ice, then stored at –80°C. A duplicate aliquot was collected from one of the applicator’s bottles each day for quality control (QC). A first-morning urine void was also collected from each person in the reference population. Applicator and reference population participants were compensated $200 and $10 each, respectively, for full participation.
Immunoassays: alachlor mercapturate, atrazine and cyanazine
RaPID Assay® enzyme-linked immunosorbent assay (ELISA) kits (Strategic Diagnostics, Newtown, PA) were used to determine metabolites of alachlor, atrazine and cyanazine in urine. Kits were used according to the manufacturer’s instructions. Calibration standards (0, 0.1, 0.5, 1.0 and 5.0 µg/l) were prepared by fortifying diluted pooled urine from anonymous volunteers with alachlor mercapturate (prepared in-house; Striley et al., 1999), atrazine or cyanazine (Chem Service, West Chester, PA). The alachlor response was greater for alachlor mercapturate than for alachlor parent (Biagini et al., 1995). Since alachlor mercapturate is a major human metabolite of alachlor, it was used as the calibration standard. Pooled urine used as calibration standard diluent was diluted with water 1:10 or 1:100 to match the dilution of the samples. Laboratory QC samples were prepared in neat urine so that when diluted 1:10 or 1:100 the concentration would be 1.0 µg/l.
The RAPID Assay® for metolachlor had little cross-reactivity to metolachlor mercapturate. Since metolachlor is excreted as the conjugated form in humans, an ELISA for metolachlor mercapturate was developed. Analyses were performed according to a previously published method (Striley et al., 1999). Calibration standards (0, 0.15, 0.25, 0.5, 1.0, 2.0 and 5.0 µg/l) were prepared by fortifying diluted pooled urine with metolachlor mercapturate (prepared in-house; Striley et al., 1999). Pooled urine used as calibration standard diluent was diluted with SuperblockTM, a buffer (Pierce Chemical, Rockford, IL) 1:10 or 1:100 to match the dilution of the samples. Laboratory QC samples were prepared in neat urine so that when diluted 1:10 or 1:100 the concentration would be 1.0 and 2.5 µg/l.
Calibration, QC and field samples were run in duplicate for all ELISA analyses. Field samples with an absorbance higher than the highest calibration standard were diluted 1:10 or 1:100 in water or SuperblockTM, and additional calibration standards in diluted pooled urine were similarly prepared. Since enzyme-labeled pesticide was in competition with unlabeled pesticide, the developed color was inversely proportional to the concentration of the unlabeled pesticide in the sample. The % B/Bo (bound/unbound) was computed by dividing each standard or sample absorbance by the absorbance of the blank. The standard curve was plotted as a logit % B/Bo versus the log concentration of the standard. Simple linear regression was performed and the regression parameters, the sample absorbance and the dilution factor were used to compute the final concentration.
Cross-reactivities to compounds structurally related to the metabolites were examined for each immunoassay (Table 1). When in vivo metabolism adds characteristics to a metabolite that make it more similar to the hapten–protein immunogen used to immunize animals, increases in affinity with resultant increases in sensitivity can be observed (Biagini et al., 1995). This appears to be the case for the anti-alachlor and anti-atrazine antibodies, which strongly cross-react with their respective mercapturate metabolites. Because of cross-reactivity and affinity issues, immunoassay results are expressed as analyte equivalents (eq.).
|
For all immunoassays, except metolachlor mercapturate, the least detectable dose (LDD) was based on the manufacturer’s reported LDD, taking into account sample dilution. The LDD for metolachlor mercapturate was defined as 90% B/Bo x 10 (the urine dilution) and interpolated mathematically from the coefficients of a four-parameter logistic equation. The immunoassay LDDs were 1.3, 2.0, 1.7 and 4.9 nmol/l, for alachlor mercapturate eq., atrazine eq., cyanazine eq. and metolachlor mercapturate eq., respectively. The LDD is used as the method limit of detection (LOD).
Gas chromatography: 2,4-D, deethylatrazine, deisopropylatrazine, parent triazines
A gas chromatography (GC) method with electron capture detection was used to determine 2,4-D in urine. Concentrated phosphoric acid (1.000 ml) and an internal standard solution (150 µl, 4676 ng/ml) of 2,4-dichlorophenylacetic acid in methanol were added to 6.0 ml of urine in a test tube and vortexed for 60 s. A 4.0 ml aliquot of the urine-internal standard mixture was loaded onto a C18 solid phase extraction (SPE) cartridge conditioned with propanol (5 ml) and water (5 ml). The loaded cartridge was rinsed with 25% acetic acid (5 ml) and water (3.0 ml). The cartridge was then dried under a flow of nitrogen for 10 min and eluted with acetone (5 ml). The acetone extract was collected in a tube containing pentafluorobenzyl (PFB) bromide (100 µl) and potassium carbonate (63 mg). The reaction mixture was vortexed for 15 min at room temperature and transferred to a tared test tube containing 120 mg of the quenching agent, hordenine (prepared from hordenine hemisulfate). The quenched solution was vortexed for 30 s and weighed. For each gram of solution, 0.4 M sulfuric acid (0.7 ml) was added and the solution vortexed for 15 s. This aqueous solution was loaded onto a second C18 SPE cartridge conditioned as above. The loaded cartridge was rinsed with 5 ml each of 0.4 M sulfuric acid and water, then dried under a flow of nitrogen for 10 min. After drying, the cartridge was eluted with toluene (4 ml) and the extract collected in a test tube. The extract contained the PFB derivatives of both 2,4-D and the internal standard. An aliquot of the toluene eluate (1.0 ml) was transferred to a GC vial. An internal instrument performance standard (0.1 ml of dinoseb methyl ether in toluene, 15 000 ng/ml) was added to the vial.
The final extract was analyzed using a Hewlett-Packard (HP) 6890 gas chromatograph with a 63Ni electron capture detector and an HP-5 fused silica capillary column (60 m x 0.325 mm i.d., 1.05 µm film thickness). The carrier gas was hydrogen at 2.0 ml/min. The injection volume was 2 µl and the injection and detector temperatures were 300°C and 325°C, respectively. The initial temperature was held at 80°C for 1.5 min, then ramped at 20°C/min to 220°C where it was held for 41.8 min, then ramped at 70°C/min to 325°C, where it was held for 15 min. Eight calibration standards (nominal range 15–200 µg/l) and two laboratory QC samples (30 and 200 µg/l) were made by fortifying blank pooled urine with 2,4-D standard. A water blank was also run with each batch. If a field sample did not have 6 ml, then the available sample volume was diluted to 6 ml with water and the dilution factor used in the final calculations. If the 2,4-D concentration exceeded the highest standard, the sample was diluted and rerun. Sample concentrations of 2,4-D were determined from the regression parameters of a linear curve of the calibration standards. The method LOD was 86 nmol/l (19 µg/l).
A GC method with mass selective detection (MSD) was used to determine atrazine, deethylatrazine, deisopropylatrazine, simazine, propazine and cyanazine (NIOSH, 1994). Urine (5 ml) was added to a test tube containing sodium chloride (0.7 g) and sodium bicarbonate (0.5 g), and then mixed thoroughly. The urine was extracted consecutively with ethyl ether and ethyl acetate (5 ml each). Samples were mixed for 15 min after solvent addition and centrifuged for 5 min at 3000 r.p.m. The organic layers were removed, combined and transferred to an anhydrous sodium sulfate filtration tube. The filtrate was evaporated to dryness under a flow of nitrogen. An internal standard (IS), phenanthrene-d10 (10 µl of a 100 µg/ml solution in methanol), and ethyl acetate (90 µl) were added to the extract. Salts and solvents were from Fisher Scientific (Pittsburgh, PA) while the IS and parent triazines were from ChemService (West Chester, PA).
The extract was analyzed using an HP5971A GC/MSD (Agilent, Santa Clara, CA) in the selected ion monitoring mode. An SPB-5 (Supelco, Bellefonte, PA) fused silica capillary column (30 m, 0.20 mm i.d., 0.20 µm film thickness), with a 1.5 m guard column of the same material was used. Helium was used as the carrier gas (1.5 ml/min). The injection volume was 1 µl, the injector and detector temperatures were 280°C and 285°C, respectively. The initial temperature (50°C) was held for 1 min, then ramped at 50°C/min to 160°C, ramped at 3.5°C/min to 230°C, and finally ramped at 50°C/min to 280°C where the temperature was held for 2 min. Ions at m/z 172 (deethylatrazine) and m/z 158 (deisopropylatrazine) as well as ions for all the parent triazines were monitored. Seven calibration standards (nominal range 9–4200 nmol/l) were prepared in ethyl acetate. Two laboratory QC samples (deethylatrazine: 78 and 253 nmol/l; deisopropylatrazine: 86 and 282 nmol/l) were made by fortifying blank pooled urine with the appropriate standards. A water blank was also run with each batch. If a field sample did not have 5 ml, then the available sample volume was used and noted in the final calculations. Sample concentrations of the six analytes were determined from the regression parameters of a linear curve of the calibration standards. The method LODs for deethylatrazine and deisopropylatrazine were 27 and 49 nmol/l, respectively. The method LODs for simazine, atrazine, propazine and cyanazine were 20, 28, 24 and 47 nmol/l, respectively.
Creatinine analysis
Creatinine in urine was determined using a BM/Hitachi 717 Analyzer (Boehringer Mannheim, Indianapolis, IN) and a reagent based on the Jaffé method. The method LOD was 30 mg/l. Creatinine was used to evaluate completeness of urine collection.
Statistical analysis
Arithmetic and/or geometric means and standard deviations were computed for individual measurements of urinary creatinine, volume and herbicide biomarkers. The LOD divided by two was imputed for samples below the detection limit, except where the percentage of samples below the detection limit was >50%, in which case only the range of detectable values was reported (Hornung and Reed, 1990). For 2,4-D, the laboratory also provided estimated values for samples below the LOD when an instrumental response was present at the retention time of 2,4-D. Therefore, 2,4-D data could be examined as being censored at either the LOD or the minimum instrumental response. In the latter case, the lowest instrumental response divided by two was imputed for samples below the minimum instrumental response for 2,4-D. Use of estimates below the LOD is considered appropriate if the intended use of the data is for statistical analysis (Analytical Methods Commitee, Royal Society of Chemistry, 1987). For comparison purposes, statistical estimates for 2,4-D are reported separately for these two censoring points. All dependent variables were highly skewed to the right and a natural log transformation was applied.
Mixed-effects linear regression modeling was used to examine the influence of spraying on the amount of herbicide biomarker excreted in 24 h in the applicator population. This 24 h total was computed by summing the number of nanomoles of biomarker in each urine sample collected after the first morning urine sample on day 0 through to the first-morning urine sample on day 1. This summed 24 h biomarker total was used as the dependent variable. For each summed 24 h dependent variable, the number of cases where all four samples were below the LOD was evaluated as an indicator of censoring in the summed dependent variable. Two variations of the dependent variable, computed biomarker concentration in 24 h (nmol/l) and the amount of biomarker excreted in 24 h divided by the applicator’s body weight (nmol/kg body wt), were also evaluated in the models.
PROC MIXED in SAS v. 8.12 (SAS Institute Inc., Cary, NC) was used to build linear models. Subject was treated as a random effect. Three covariance structures—compound symmetry, first-order autoregressive and spatial power—were evaluated for each dependent variable to address the correlation of measurements within subjects (Symanski et al., 2001). A spatial power structure was included as measurements were not equally spaced with respect to date. Structures were compared using maximum likelihood estimation in a model consisting of terms for spraying the target herbicide (yes/no) on day 0 and the 2 days immediately before day 0 (day –1 and day –2) and kg of target herbicide sprayed on days 0, –1 and –2. Akaike’s Information Criterion was used to compare the model fit.
Plots of the data revealed that levels on non-spray days had a fairly consistent pattern during the spraying season [either gradually increasing (e.g. 2,4-D) or increasing initially then decreasing (e.g. atrazine)]. In order to adjust for these temporal patterns, restricted cubic splines (cubic curves between knots, smooth at the knots, and linear at the ends) were used to examine the effect of time (i.e. number of days since the start of the study) for each dependent variable as described by Harrell (Harrell et al., 1988; Harrell, 2001).
Once the optimal covariance structure and time effect had been determined, a backward selection procedure using restricted maximum likelihood estimation was performed on a model containing the covariates spraying the target herbicide on days 0, –1 and –2, kg of target herbicide sprayed on days 0, –1 and –2, acres sprayed with target herbicide on days 0, –1 and –2, and duration (min) of spraying target herbicide on day 0 only (duration data were not available for days –1 and –2). These covariates were examined together because they were possibly related predictors of biomarker levels. A 0.05 level of significance was used to remove terms from the models. All other covariates and possible confounders were tested one at a time after the best base model had been determined. Continuous covariates included age, years worked as an applicator, years worked for the current company, duration of work day, number of spray jobs during the day (all herbicides combined), number of nozzle changes during the day (all herbicides combined), and number of spray jobs with a cab window open (all herbicides combined). Categorical covariates included certified applicator status (yes/no), smoking tobacco (yes/no), chewing tobacco (yes/no), glove use (yes/no), charcoal filter (yes/no), dust filter (yes/no), air conditioning in cab (yes/no) and boom location (front/back).
Total variance was computed by fitting a model containing subject only (Burstyn et al., 2000). Within- and between-worker variance components were estimated from the random effects portion of the models. In order to obtain within- and between worker variance estimates under a first-order autoregressive or a spatial power structure, both the REPEATED and RANDOM statements in PROC MIXED were used (Littell et al., 2000).
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONBACKGROUNDMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
Applicators
All 15 applicators provided a pre-season urine sample. Alachlor mercapturate eq. and metolachlor mercapturate eq. were detected in more than half (60% each) of these samples (Table 2). Atrazine eq. and 2,4-D were detected to a lesser extent (13 and 20%, respectively). Cyanazine eq., deethylatrazine and deisopropylatrazine were not found in pre-season samples.
|
During the 6 week study season, 15 applicators from seven field stations provided urine samples for 5–7 days each, for a total of 89 applicator-days and 408 urine samples (Table 3). Applicator demographics and herbicide use are reported in Hines et al. (2001). On two applicator-days, a day 1 first-morning urine sample was not provided. These two applicator-days were excluded from analyses that involved 24 h totals. For nine of the remaining 87 applicator-days, the 24 h urine volume was <600 ml (range: 353–597 ml), and on one of these applicator-days, 24 h creatinine was <0.5 g/day (0.429 g/day). The normal range for 24 h urine volume and 24 h creatinine in men is 600–2500 ml and 0.5–3.5 g/day, respectively (Que Hee, 1993). Since we could not be certain that lower than normal 24 h urine volume or creatinine actually indicated incomplete urine collection, these data were left in the analysis. Insufficient hydration may explain low urinary volumes.
|
Biomarker results were reported on all 408 urine samples, except for 2,4-D and the dealkylated atrazine metabolites where the number of samples with results was 406 and 405, respectively, due to insufficient sample volumes (Table 4). For cyanazine eq., deethylatrazine and deisopropylatrazine, where >50% of the samples were below the LOD, only the range of concentrations above the LOD is reported (Table 4). Parent atrazine was found in only two of 405 (0.5%) spray season samples at levels of 53 and 29 nmol/l. Parent cyanazine, simazine and propazine were not detected in any urine samples. Total biomarker excreted in 24 h adjusted and unadjusted for body weight, biomarker concentration in the composite 24 h urine sample and biomarker excretion rates by collection time period were also computed for alachlor mercapturate eq., atrazine eq., 2,4-D and metolachlor mercapturate eq. (Tables 5 and 6). Excretion rates peaked in the afternoon and were lowest in the morning and overnight samples. The sample size was not sufficient to determine the effect of spraying on excretion rates by time period.
|
|
Exposure modeling
Regression model parameters for atrazine eq. and 2,4-D are given in Tables 7 and 8. Two models are presented for atrazine eq.: model A, which includes kg of atrazine sprayed (all other terms dropped out), and model B, in which kg of atrazine sprayed was excluded a priori. The latter model was constructed because information on the amount of herbicide active ingredient sprayed may not be available to the investigator. A compound symmetric covariance structure was used for atrazine eq. and a spatial power covariance structure for 2,4-D. Time (i.e. days since the study started) was significant in the atrazine eq. model and best modeled as a three-knot restricted cubic spline function. The percentage of applicator-days where all samples comprising the summed 24 h biomarker total were below the LOD for atrazine eq. and 2,4-D was 10 and 19, respectively. For 2,4-D, the percentage of applicator-days where all samples in the 24 h total were below the minimum instrumental response was 3.
|
|
In both atrazine eq. models, either kg of atrazine sprayed or spraying atrazine (yes/no) on day –1 had a greater impact on the amount of atrazine eq. excreted in 24 h compared with day 0 (factors of 2.32 versus 1.75 and 2.86 versus 1.95, respectively). Day –2 was not significant in either model A or B. Spraying 2,4-D one and two days before urine collection was highly predictive of the amount of 2,4-D excreted in the 24 h urine sample (factors of 4.25 and 3.90, respectively), while spraying 2,4-D on day 0 was not significant. The kg of 2,4-D sprayed on any of the 3 days did not remain in the final model. Spraying the target herbicide 3 days before urine collection (day –3) and all other covariates were not significant (P > 0.05) in any of the models. Using computed biomarker concentration in 24 h or the amount of biomarker excreted in 24 h adjusted for body weight as dependent variables had negligible effect on the atrazine eq. and 2,4-D models.
Variance components were computed (Table 9) for the atrazine eq. and 2,4-D models. The within-worker variance component, expressed as a geometric standard deviation (WGSD range: 2.5–2.9) was substantially larger than the between-worker component (BGSD range: 1.3–1.5) for both analytes. The within-worker component explained 47–62% of the total variance. Fixed effects explained 28–49% of the total variance.
|
Alachlor mercapturate eq. and metolachlor mercapturate eq. were not modeled because the number of spray days on day 0, day –1 and day –2 were not sufficient (range 1–2 days for alachlor mercapturate eq.; 6–9 days for metolachlor mercapturate eq.). In addition, the metolachlor mercapturate immunoassay moderately cross-reacted with acetochlor mercapturate (Table 1). Acetochlor was commonly substituted for alachlor during the study. The atrazine antibody also cross-reacted moderately with simazine (Table 1); however, simazine was not detected in any urine samples.
Forty-six men were recruited as a reference population. Their median age (40 yr; range 19–63 yr) compared favorably with the applicator population (median age 40 yr; range 23–58 yr). The frequency of detection of the seven herbicide biomarkers in the reference population was low (range 0% for deisopropylatrazine to 10.9% for 2,4-D) (Table 2).
Quality control
Mean recoveries in laboratory QC samples ranged from 77 to 122%, while mean percent relative standard deviations for laboratory QC samples and for field duplicates ranged from 10 to 27 and from 12 to 24, respectively (Table 10). Field samples have not been corrected for recovery.
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONBACKGROUNDMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
Custom applicators in this study were exposed, on average, to 13 (± 2.8, range 9–17) different herbicides. The group was exposed to >30 different herbicides. All monitored biomarkers were detected to varying degrees of frequency and intensity in the urine of participating applicators. In pre-season samples, alachlor mercapturate eq. and metolachlor mercapturate eq. were detected at substantially higher frequencies (60% each) than other biomarkers (range 0–20%). The concentration range for these two biomarkers was low compared with samples collected during the spray season. Because applicators had not sprayed the target herbicides before the pre-season samples were collected, the high frequency of detection suggests applicator exposure to unidentified herbicide sources prior to sample collection.
During the spray season, 2,4-D was found in applicator urine samples at GM levels 3–4 times higher than the other biomarkers (Table 4). Surprisingly, alachlor mercapturate eq. was detected in nearly 83% of the spray season urine samples and at GM levels comparable to those found for metabolites of two heavily used herbicides, atrazine and metolachlor, even though most farmers switched from alachlor to acetochlor during the study year. Alachlor was sprayed on only two of the 87 applicator-days. Since alachlor mercapturate levels were generally higher in the spray season than in the pre-season, and very little alachlor spraying was done, this increase might be due to alachlor residues on equipment from previous seasons or to antibody cross-reactivity with an unidentified compound. The high percentage of hand wash samples with detectable levels of alachlor (~70%) suggests the former may be the case (Hines et al., 2001). The presence of alachlor mercapturate in a subset of the urine samples has been confirmed by GC/MS/MS (unpublished data) The excretion rate peak seen in the afternoon for all analytes, may reflect that samples came from an occupationally exposed population where exposure typically starts early in the morning and ends late in the day.
Atrazine eq. measured by ELISA was detected more frequently in spray season samples than either of the two mono-dealkylated atrazine metabolites measured by GC. These findings are consistent with literature reports that atrazine mercapturate, which cross-reacts with the atrazine ELISA, is the major human metabolite of atrazine. Forty-six percent of the samples tested with the cyanazine immunoassay gave a response above the assay’s LOD. Since parent cyanazine was not found in any sample analyzed by GC, we hypothesize that a putative cyanazine metabolite or congener was possibly present in the urine which reacted with the cyanazine antibody.
The monitored biomarkers were found infrequently in the urine of the reference population and, when detected, levels were low compared with exposed applicators. 2,4-D was found most often in the reference population (10.9%), even though these participants did not report application of 2,4-D to their property, either by themselves or by others. 2,4-D is a widely used herbicide and participants were possibly exposed to 2,4-D at or away from home without their knowledge. Hill et al. (1995) reported a similar frequency of 2,4-D detection (12%) in a much larger set of 983 urine samples collected from adults participating in the National Health and Nutrition Examination Survey III (NHANES III). MacIntosh et al. (1999) found atrazine mercapturate in only 0.3% of 348 urine samples collected from 80 individuals living in five Maryland counties. This percentage is slightly less than the 2.2% reported here for atrazine eq., where the major metabolite measured was most likely atrazine mercapturate. General population reference data are not available for the other biomarkers monitored. Other published applicator studies have measured exposure only on spray days, whereas our study included both spray and non-spray days.
Regression modeling was used to determine significant predictors of the amount of atrazine eq. and 2,4-D excreted in the 24-h sample. For both analytes, spraying information on days prior to urine collection (day –1 for atrazine eq., and days –1 and –2 for 2,4-D) was significant (Table 7). Unlike atrazine, spraying 2,-4 D (yes/no) on day 0 was not an important predictor of total 2,4-D excreted in the 24 h sample (Table 7). Also unlike atrazine, the amount of 2,4-D sprayed in kg was less significant than whether 2,4-D was sprayed or not and did not remain in the final model. The reason for the difference in the importance of the kg covariate between the atrazine eq. and 2,4-D models is not readily apparent. The application method was the same for both herbicides; however, the range of 2,4-D EH sprayed on days 0, –1 and –2 (3–24, 2–84 and 0.3–87 kg, respectively) was much narrower than for atrazine on the corresponding days (9–464, 5–236 and 10–402 kg, respectively). The 2,4-D findings are consistent with urinary half-life estimates for 2,4-D in the range of 13 to >40 h (Sauerhoff et al., 1977; Nash et al., 1982; Harris and Solomon, 1992). The small sample size and adjustment for more than one day of spraying in the models probably reduced the power to detect other significant covariates. This contrasts with the analysis of hand and dermal exposures for this population where a significant glove effect was found for alachlor, atrazine, 2,4-D EH, and metolachlor (Hines et al., 2001).
The time or number of days into the study when the sample was collected was a significant predictor of the amount of atrazine eq. excreted independent of the amount of target herbicide sprayed (Table 7). Therefore, in order to clearly test the effect of spraying the target herbicides on the amount of biomarker excreted, it was necessary to adjust for time in the atrazine regression models. Our approach was to use the restricted cubic spline function that most closely modeled the observed atrazine time profile and gave the best model fit. Time trends, such as that found here, may be present in other longitudinal exposure data, and should be examined and adjusted for, when necessary, as a part of exposure determinant modeling.
This study also allowed an empirical evaluation of the statistical impact of using observed estimates for 2,4-D below the LOD compared with imputing a constant value. Not surprisingly, use of a single imputed value in regression models reduced total variability (Table 9) and decreased standard errors (Tables 7 and 8). When an imputed value was used (compared with using all observed estimates), reduction in the within-worker variability was more pronounced than in the between-worker variability (Table 9). The effect of using an imputed value versus observed estimates is seen even more clearly in estimates of the geometric mean (GM) and standard deviation (GSD) for 2,4-D levels in individual samples (Table 4) where estimates using a single imputed value resulted in ~2-fold higher GMs and ~2-fold lower GSDs as compared to using observed estimates.
Traditionally, chemists have been largely concerned with the reporting of individual measurements rather than the use of data to make statistical inferences and estimates (Lambert et al., 1991; Porter et al., 1988). The Analytical Methods Committee of the Royal Society of Chemistry (1986) noted, however, that ‘At low levels of analyte, many observations of concentration will fall below the cL [detection limit] given an unbiased analytical method. Such observations are not devoid of meaning: they are estimates of true concentrations’ and ‘...the preferred method of reporting such values depends on the intended use of the data [emphasis added]’, Moreover, when a value is part of a data set subjected to statistical analysis ‘various substitute values ranging from 0 to cL have been suggested but are not recommended’ and ‘Generally the observed value should be used.’ The use of observed estimates below the LOD, although controversial, preserves information, and in particular, approximates the true variability in the data set better than using substitute values. Clearly, the level of censoring in the data should still be considered before statistical analyses are performed, regardless of the method used for observations below the detection limit; however, when statistical analyses are performed more consideration should be given to using all observed data to ensure more reliable estimates and inferences.
Few studies have reported within- and between-worker variability for pesticide biomarkers in urine. The high within-worker variability observed in this study (WGSD range 2.5–2.9) indicates that repeated measurements on workers in this job are necessary in order to estimate individual biomarker levels more accurately, if a similar sampling frequency (i.e. ~1–2 times per week) is used. Within-worker variability of dimethylphosphate, a metabolite of azinphosmethyl, was also greater than between-worker variability (WGSD range 2.2–2.6; BGSD range 1.1–1.4) in spot urine samples collected one to several days apart from apples thinners (Simcox et al., 1999). On the other hand, between-worker variability of creatinine-corrected urinary 3,5,6-trichloro-2-pyridinol was greater than within-worker variability (WGSD = 1.4, BGSD = 2.4) when spot urine samples were collected every day for a week from termiticide applicators using chlorpyrifos (Hines and Deddens, 2001). Variability in the environmental exposure, the elimination half-life of a substance, and the frequency of sample collection are all factors that may influence within- and between-worker variability.
Some study limitations should be noted. The concurrent collection of dermal and urine samples, and any incomplete urine collection may have produced underestimates of biomarker levels. Herbicide captured or removed by dermal sampling would not have been available for uptake. In the 2,4-D analysis, samples did not undergo acid hydrolysis and any 2,4-D present as a conjugate would not have been measured. The study year was unusually wet and applicators sprayed on only one-third of the applicator-days compared with two-thirds of the days in a more typical year. Therefore, biomarker levels in this study may not be representative of levels found in years with more spraying.
|
CONCLUSION |
|---|
|
TOPABSTRACTINTRODUCTIONBACKGROUNDMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
The specific days of herbicide spraying, measured as either a dichotomous variable (yes/no) or as kg of herbicide sprayed, that were associated with an increased amount of atrazine eq. and 2,4-D excreted in the 24-h urine samples, varied for the two biomarkers and included information from one or more days prior to the urine sample collection. This inclusion of day(s) prior to urine collection in the models confirms the importance of collecting covariate information on the day(s) most relevant to the biomarker of interest. In addition, time trends in longitudinal data that may be independent of the effects of interest should be evaluated and adjusted for in statistical models. Larger sample sizes may be needed to evaluate exposure determinants for biological measures of exposure compared with non-biological measures if the biomarker of interest is not excreted within 24 h and exposures on days prior to biomarker collection are significant.
Acknowledgements—We would like to acknowledge the contributions of Donald Booher, Laura Denovan, Dominic Giambri, Gail Gislason, Cathy Hall, Matthew Jackson, Barbara Jenkins, Steve Lenhart, Wayne Sanderson, Lloyd Stettler, Alexander Teass, Elizabeth Ward and Thurman Wenzl to this study. We would also like to thank Randall Smith for statistical advice, and Jeffrey McLaurin, Dawn Ramsey and Derek Stinson for sample analyses. Most importantly, we would like to thank the custom applicators for their participation.
Disclaimer—Mention of any company or product does not constitute endorsement by the National Institute for Occupational Safety and Health.
|
|
FOOTNOTES |
|---|
* To whom correspondence should be addressed. Tel: +1-513-841-4453; fax: +1-513-841-4486; e-mail: chines{at}cdc.gov
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONBACKGROUNDMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
Analytical Methods Committee, Royal Society of Chemistry. (1987) Recommendations for the definition, estimation and use of the detection limit. Analyst; 112: 199–204.[CrossRef]
Biagini RE, Tolos W, Sanderson WT, Henningsen GM, MacKenzie B. (1995) Urinary biomonitoring for alachlor exposure in commercial pesticide applicators by immunoassay. Bull Environ Contam Toxicol; 54: 245–50.[ISI][Medline]
Bradway DE, Moseman RF. (1982) Determination of urinary residue levels of the N-dealkyl metabolites of triazine herbicides. J Agric Food Chem; 30: 244–7.
Brady JF, Turner J, Cheung MW et al. (1998) An immunochemical approach to estimating worker exposure to atrazine. In Ballantine LG, McFarland JE, Hackett DS (editors). Triazine Herbicides: Risk Assessment. ACS Symposium Series 683. Washington, DC: American Chemical Society, pp. 131–140.
Buchholz BA, Fultz E, Haack KW et al. (1999) HPLC-accelerator MS measurement of atrazine metabolites in human urine after dermal exposure. Anal Chem; 71: 3519–25.[Medline]
Burstyn I, Kromhout H, Kauppinen T, Heikkila P, Boffetta P. (2000) Statistical modelling of the determinants of historical exposure to bitumen and polycyclic aromatic hydrocarbons among paving workers. Ann Occup Hyg; 44: 43–56.
Catenacci G, Barbieri F, Bersani M, Ferioli A, Cottica D, Maroni M. (1993) Biological monitoring of human exposure to atrazine. Toxicol Lett; 69: 217–22.[Medline]
Denovan LA, Lu C, Hines CJ, Fenske RA. (2000) Saliva biomonitoring of atrazine exposure among herbicide applicators. Int Arch Occup Environ Health; 73: 457–62.[Medline]
Driskell WJ, Hill RH. (1997) Identification of a major human urinary metabolite of metolachlor by LC-MS/MS. Bull Environ Contam Toxicol; 58: 929–33.[CrossRef][ISI][Medline]
Driskell WJ, Hill RH, Shealy DB, Hull RD, Hines CJ. (1996): Identification of a major human urinary metabolite of alachlor by LC-MS/MS. Bull Environ Contam Toxicol; 56: 853–9.[CrossRef][ISI][Medline]
Feng PCC, Sharp CR, Horton SR. (1994) Quantitation of alachlor residues in monkey urine by ELISA. J Agric Food Chem; 42: 316–9.
Feldmann RJ, Maibach HI. (1974) Percutaneous pentration of some pesticides and herbicides in man. Toxicol Appl Pharmacol; 28: 126–32.[CrossRef][ISI][Medline]
Gilman SD, Gee SJ, Hammock BD et al. (1998): Analytical performance of accelerator mass spectrometry and liquid scintillation counting for detection of 14C-labeled atrazine metabolites in human urine. Anal Chem; 70: 3463–9.[Medline]
Harris SA, Solomon KR. (1992) Percutaneous penetration of 2,4-dichlorophenoxyacetic acid and 2,4-D dimethylamine salt in human volunteers. J Toxicol Environ Health; 36: 233–40.[Medline]
Harrell FE. (2001) Regression modeling strategies. New York: Springer-Verlag. pp. 20–4. ISBN 0-387-95232-2.
Harrell FE, Lee KL, Pollock BG. (1988) Regression models in clinical studies: determining relationships between predictors and response. J Natl Cancer Inst 80; 1198–202.
Hill RH, Head SL, Baker S et al. (1995) Pesticide residues in urine of adults living in the United States: reference range concentrations. Environ Res; 71: 99–108.[Medline]
Hines CJ, Deddens JA. (2001) Determinants of chlorpyrifos exposures and urinary 3,5,6-trichloro-2-pyridinol levels among termiticide applicators. Appl Occup Hyg; 45: 309–21.
Hines CJ, Deddens JA, Tucker SP, Hornung RW (2001) Distributions and determinants of pre-emergent herbicide exposures among custom applicators. Ann Occup Hyg; 45: 227–39.
Hornung RW, Reed L. (1990) Estimation of average concentration in the presence on non-detectable values. Appl Occup Environ Hyg; 5: 46–51.
Kohli JD, Khanna RN, Gupta BN, Dhar MM, Tandon JS, Sircar KP. (1974) Absorption and excretion of 2,4-dichlorophenoxyacetic acid in man. Xenobiotica; 4: 97–100.[ISI][Medline]
Kronenberg JM, Fuhremann TW, Johnson DE. (1988) Percutaneous absorption and excretion of alachlor in rhesus monkeys. Fundam Appl Toxicol; 10: 664–71.[Medline]
Lambert D, Peterson B, Terpenning I. (1991) Nondetects, detection limits, and the probability of detection. J Am Stat Assoc; 86: 266–77.
Littell RC, Pendergast J, Natarajan R. (2000) Tutorial in biostatistics: modelling covariance structure in the analysis of repeated measures data. Stat Med; 19: 1793–819.[CrossRef][ISI][Medline]
Lucas AD, Jones AD, Goodrow MH et al. (1993): Determination of atrazine metabolites in human urine: development of a biomarker of exposure. Chem Res Toxicol; 6: 107–16.[CrossRef][ISI][Medline]
MacIntosh DL, Needham LL, Hammerstrom KA, Ryan PB. (1999) A longitudinal investigation of selected pesticide metabolites in urine. J Exp Anal Environ Epidemiol; 9: 494–501.[CrossRef][ISI][Medline]
Menda G, Tkalcevic B, Drevendar V. (2000) Determination of chloro- and methylthiotriazine compounds in human urine: extraction with diethyl ether and C18 solid-phase extraction for gas chromatographic analysis with nitrogen-selective and electron capture detection. Anal Chim Acta; 424: 7–18.[CrossRef]
Nash RG, Kearney PC, Maitlen JC, Sell CR, Fertig SN. (1982) Agricultural applicators exposure to 2,4-dichlorophenoxyacetic acid. In Plimmer JR (editor). Pesticide Residues and Exposure. American Chemical Society Symposium Series No. 182. Washington, DC: American Chemical Society. pp. 119–32. ISBN 0-8412-0701-1.
NIOSH. (1994) Method 8315: triazine herbicides and their metabolites in urine. In Eller PM, Cassinelli ME (editors). NIOSH Manual of Analytical Methods, 4th edn, 3rd supplement. Cincinnati, OH: DHHS (NIOSH).
Porter PS, Ward RC, Bell HF. (1988) The detection limit. Environ Sci Technol; 22: 856–61.
Que Hee H. (1993) Biological monitoring: an introduction. New York: Van Nostrand Reinhold. pp. 141–2. ISBN 0-442-23677-8.
Sauerhoff MW, Braun WH, Blau GE, Gehring PJ. (1977) The fate of 2,4-dichlorophenoxyacetic acid (2,4-D) following oral administration in man. Toxicology; 8: 3–11.[CrossRef][ISI][Medline]
Simcox NJ, Camp J, Kalman D et al. (1999) Farmworker exposure to orangophosphorus pesticide residues during apple thinning in central Washington state. Am Ind Hyg Assoc J; 60: 752–61.[Medline]
Striley CAF, Biagini RE, Mastin JP, MacKenzie BA, Robertson SK. (1999) Development and validation of an ELISA for metolachlor mercapturate. Anal Chim Acta; 399: 109–14.[CrossRef]
Symanski E, Chan W, Chang C. (2001) Mixed-effects models for the evaluation of long-term trends in exposure levels with an example from the nickel industry. Ann Occup Hyg; 45: 71–81.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  B. D. CURWIN, M. J. HEIN, W. T. SANDERSON, C. STRILEY, D. HEEDERIK, H. KROMHOUT, S. J. REYNOLDS, and M. C. ALAVANJA Urinary Pesticide Concentrations Among Children, Mothers and Fathers Living in Farm and Non-Farm Households in Iowa Ann. Hyg., January 1, 2007; 51(1): 53 - 65. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||