Detection limits can influence the interpretation of pesticide monitoring data in Canadian surface waters

Chemosphere 86 (2012) 565–571

Contents lists available at SciVerse ScienceDirect

Chemosphere journal homepage: www.elsevier.com/locate/chemosphere

Detection limits can influence the interpretation of pesticide monitoring data in Canadian surface waters Shane R. de Solla a,⇑, John Struger b, Tana V. McDaniel b a b

Wildlife and Landscape Science Directorate, Environment Canada, Canada Centre for Inland Waters, 867 Lakeshore Road, Burlington, Ontario, Canada L7R 4A6 Water Science and Technology Directorate, Environment Canada, Canada Centre for Inland Waters, 867 Lakeshore Road, Burlington, Ontario, Canada L7R 4A6

a r t i c l e

i n f o

Article history: Received 11 May 2011 Received in revised form 16 September 2011 Accepted 20 September 2011 Available online 2 December 2011 Keywords: Detection limits Pesticides Detection probability Surface waters Monitoring

a b s t r a c t Water quality monitoring programs rely on residue data that are frequently left censored, due to some observations occurring below the Method Detection Limit (MDL). Our objective was to determine the influence the MDL has on the interpretation of pesticide residues in surface waters. Water samples from tributaries in southern and central Ontario were collected by Environment Canada from 2003 to 2008 and were analyzed for 27 pesticides, with MDLs that averaged 7.02 ng1 L (range 0.39–25.1 ng1 L). We then simulated MDLs ranging from 25 to 1700 ng1 L, to determine the impact this would have on the reporting of pesticide concentrations and detections. The mean number of pesticides detected simultaneously declined with increasing, i.e. less sensitive MDLs, from 5.02 pesticides (native MDL) to 0.08 pesticides detected (MDL < 1700 ng1 L). We compared the proportion of sites where pesticides were detected in surface waters under five MDL scenarios for 13 selected pesticides. The proportions decreased sharply with increasing MDLs. We calculated detection probabilities in an effort to compensate for higher MDLs using maximum likelihood; while adjusting for detection probabilities generally improved estimates of the presence of pesticides, as the MDLs increased the ability to compensate for detection probabilities deteriorated and became unviable at high MDLs. Depending on the method of substitution for observations below MDL (replacement with ½  or 0  MDL), the mean and median pesticide residues became increasingly over- and underestimated, respectively, at higher MDLs. Although monitoring programs that are focused on exceedences of water quality guidelines may not require low MDLs, the achievable goals of monitoring programs oriented towards other ecological and toxicological objectives may be limited by higher MDLs. Crown Copyright Ó 2011 Published by Elsevier Ltd. All rights reserved.

1. Introduction Method Detection Limits (MDLs) are often seen as a tangential issue in studies involving pesticides in surface waters. The nature of MDLs is generally treated as if it was primarily a numerical issue, often with little regard to the implications the MDLs can have on the hypotheses or questions that monitoring programs can address. The nature of MDLs, however, can have a dramatic influence on the interpretation of data and can affect the objectives that any given monitoring program is capable of achieving. A number of methods have been developed to produce unbiased means and variances in the presence of observations below the MDL (Helsel, 2005), as well as generating values to replace observations (data imputation) below the MDL prior to doing statistical analyses (Succop et al., 2004; de Solla et al., 2007; Antweiler and Taylor, 2008). Naïve or overly simplistic methods of handling observations below the MDL has even been construed as ‘‘data ⇑ Corresponding author. Tel.: +1 905 336 4686; fax: +1 905 336 6434. E-mail address: [email protected] (S.R. de Solla).

fabrication’’ (Helsel, 2006), where inappropriate values are substituted for values below the MDL, causing artifacts that are interpreted as real relationships. Regardless, these methods can be seen as a poor remedy for ‘‘real data’’, and should only be used as a last resort; ‘‘From a statistical perspective, the use of these limits is an abomination,’’ (Oehlert, 2006). Nevertheless, censored data are unavoidable as there are lower limits of residues that analytical techniques, however sophisticated, can detect. Logistical issues, including financial cost, may limit the lower boundary for MDLs for monitoring programs, due to large numbers of samples to be analyzed. Choosing laboratories or methods that are cheaper and/ or faster may result in higher MDLs, along with all the difficulties associated with high MDLs. The question remains, how much loss in detection sensitivity, or how high can MDLs be, before there are serious impediments to the types of questions that can be answered in monitoring programs? A personal experience by the first author was the impetus for this study; in an unrelated study (de Solla and Martin, 2011), representatives (with chemistry backgrounds) from a funding committee urged the first author to switch laboratories to one with

0045-6535/$ - see front matter Crown Copyright Ó 2011 Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.chemosphere.2011.09.026

566

S.R. de Solla et al. / Chemosphere 86 (2012) 565–571

lower costs but higher MDLs, without recognizing the statistical implications. After much discussion, the first author declined, largely due to fears that the higher MDLs would compromise the objectives of the study. This experience begged the question; do MDLs influence or limit the interpretation of contaminants data? Our current objective was to determine the influence the MDL has on the interpretation of pesticide residues in surface waters. We used three measures to accomplish this: comparison of means and medians, detection probabilities of individual pesticides, and number of simultaneously detected pesticides. We used the ‘‘native’’ MDL for pesticides analyzed from an ongoing pesticide monitoring program in surface waters in Ontario; we also modeled different scenarios with a wide range of MDLs, from 25 to 1700 ng1 L. 2. Methods 2.1. Pesticide analysis Water samples (n = 651) from tributaries in southern and central Ontario that were collected by Environment Canada from 2003 to 2008 were analyzed for pesticide residues (see Kurt-Karakus et al., 2008 for site locations). Water samples were collected year round. Pesticide residue scans were analyzed by Environment Canada’s NLET laboratory and by Axys Analytical Services. For a detailed description of analytical methods see Environment Canada (1997) and Axys Analytical Services (Woudneh et al., 2009; Glozier et al., in press). Pesticides were extracted with dichloromethane (100 mL per liter of sample). Although the total list of pesticides monitored is longer, in this study we report data for the following 27 pesticides: 2,4,5-T, 2,4-D, 2,4-DB, 2,4-DP, 2,3,6-trichlorobenzoic acid, atrazine, bromoxynil, chlorpyrifos, clopyralid, diazinon, dicamba, dimethoate, ethion, fonofos, malathion, MCPA, MCPB, metolachlor, metribuzin, parathion, phorate, picloram, silvex, simazine, terbufos, triallate, and trifluralin. All fractions were analyzed by gas chromatography/mass spectrometry (GC–MS). Detection limits for the pesticides ranged from 0.37 to 25.1 ng1 L. The MDLs, mean, maximum, and minimum concentrations, and frequency of detections are included in Supplementary Table S1. 2.2. Statistics and data analysis We estimated detection probabilities for selected pesticides in surface waters of southern Ontario under five method detection limit scenarios (native, >25, >50, >100, >500 ng1 L). By using maximum likelihood models, we estimated: wi, probability the pesticide was present at site i; and pit, detection probabilities at site i at time t, which varied among sampling periods (MacKenzie et al., 2002). Native is the actual MDL for each pesticide used in this study. We used the program PRESENCE (Proteus Inc., MacKenzie, 2005) to calculate these detection probabilities. We selected a subset of pesticides to include those with a high and low frequencies of detection (e.g. atrazine, 2,4-D, metolachlor, dicamba, vs. simazine, 2,4,5-T), pesticides whose registered use expired in Ontario in the 1980s (2,4,5-T, silvex), as well as some pesticides that had high toxicity (diazinon, chlropyrifos). Detection probabilities were estimated using presence/absence of pesticides in surface waters collected from 69 sites and represents the probability of detecting the compound in a given year at a given site. The proportion of sites where the pesticides occurred in surface waters was also estimated by PRESENCE using these detection probabilities, for each of the five method detection limit scenarios. In order to highlight the effect of increasing detection limits on detection probabilities during different times of the year we also estimated detection probabilities bi-monthly from late March to early November for both dicamba and atrazine.

A repeated measures ANOVA was used to compare the number of pesticides detected simultaneously for each water sample. An assumption is the repeated measures factors are independent or orthogonal with each other (i.e. assumption of sphericity); if this assumption was violated we used the Greenhouse–Geiser adjustment. The various detection scenarios (native, > 25, > 50, > 80, > 100, > 250, >500, and > 1700 ng1 L) were the repeated measures. Extreme value distributions were used to create a line of best fit, as this distribution had a better fit compared to normal or lognormal distributions; the curves were primarily for illustrative purposes. We also used atrazine as an example, to examine how changes in the MDL affected detection, particularly in reference to the Canadian Council of Ministers of the Environment (CCME) water quality guideline for the protection of aquatic life (CCME, 1999). We also used both repeated measures ANOVA and nonparametric Friedman ANOVA repeated measures tests to determine if the mean and median concentrations of pesticides, respectively, varied among MDLs. Furthermore, we attempted to elucidate the effect that the interaction between MDL and the method of substituting replacement values has on the interpretation of pesticide residues. Two common methods of replacing observations below MDL, representing the extreme options typically used, are to replace values with ½ the MDL, or with zero. Other methods are also available, such as using maximum likelihood (Helsel, 1990; de Solla et al., 2007), regression on order (Helsel, 2005), and calculating unbiased means and variance without substitution (Perkins et al., 1990), but comparison of these methods have been dwelt with elsewhere (Helsel, 2005) and thus will not be evaluated here. 3. Results We compared the proportion of sites where pesticides were detected in surface waters under five method detection limit scenarios, for the 13 selected pesticides. The proportion of sites where pesticides were detected decreased sharply with increasing MDLs (Fig 1a). Commonly detected pesticides which tend to be found at higher concentrations such as atrazine and metolachlor, did not initially drop as precipitously with declining MDLs, while others, such as dicamba, MPCA, 2,4-DP, bromoxynil, 2,4,5-T and silvex (2-(2,4,5-trichlorophenoxy)propionic acid) dropped sharply as soon as the MDLs exceeded the native detection limit. The proportion of sites for relatively commonly detected compounds such as MPCA and 2,4-DP, fell from 0.7 at the native MDLs (0.58 and 0.42 ng1 L) to 0.21 and 0.12 respectively, even with a MDL of 25 ng1 L. This trend was reflected in the detection probabilities for each compound under the five MDL scenarios (Fig. 1b). Some less common pesticides, such as 2,4,5-T and silvex, were only detected when the most sensitive (native) detection limit (0.39 ng1 L) was used. We used the detection probabilities estimated in PRESENCE to calculate the estimated proportion of sites where each pesticide occurred (Fig. 1c). This model should compensate to some degree for decreases in detection probability. Note that the amplitude in decline in the estimated proportion of sites detected is reduced at lower MDLs (Fig. 1c). However, at higher detection limits, in cases where the detection probability is below 0.1, the model is unable to produce reasonable estimates of the proportion of sites occupied, as for MPCA, bromoxynil, diazinon, and chlorpyrifos. We also examined the impact of increasing MDLs of two commonly detected pesticides in surface waters, dicamba and atrazine, on the seasonal ability to detect them (Fig. 2). The probability of detecting dicamba outside of the growing season was remote at all detection limits above native. Although it was still possible to detect atrazine throughout the year, the detection probability outside of the growing season (May–September) declined steadily as the detection limit increased.

S.R. de Solla et al. / Chemosphere 86 (2012) 565–571

a

567

We examined the interaction between the MDL and the method of substituting replacement values on the interpretation of pesticide residues, using atrazine as a test case. The mean concentration of atrazine declined with increasing MDLs when the replacement value was zero (F[1.93.1255.8] = 173.1, P < 0.0001 Fig. 4a), but mean concentrations of atrazine increased with increasing MDLs when the replacement value was ½ the MDL (F[1.08,697.0] = 146.6, P < 0.0001; Fig. 4b). Similarly, the median concentration of atrazine declined with increasing MDLs when the replacement value was zero (Friedman ANOVA; v2½8 ¼ 3116:8, P < 0.0001 Fig. 5a), but median concentrations of atrazine increased with increasing MDLs when the replacement value was ½ the MDL (Friedman ANOVA; v2½8 ¼ 2753:5, P < 0.0001; Fig. 5b).

b 4. Discussion

c

Fig. 1. (a) Naïve estimates of the proportion of sites in which selected pesticides were detected in surface water samples on a yearly basis from Ontario, 2003–2008 using different Method Detection Limits (MDLs) (ng1 L). (b) Detection probabilities for each pesticide in surface waters at these sites, on a yearly basis, with different MDLs (ng1 L). (c) The estimated proportion of sites where these pesticides occur, as adjusted by the detection probability estimates. The MDLs (ng1 L) for the selected pesticides are as follows: Atrazine: 5.76; 2,4-D: 0.47; Dicamba: 0.73; Metolachlor: 23.7; MCPA: 0.58; 2,4-DP: 0.42; Bromoxynil 0.99; Simazine: 16.4; Diazinon: 14.9 and 15.5; 2,4,5-T:0.39; Chlorpyrifos; 2.2 and 14.8; Silvex: 0.4). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

We compared the number of pesticides detected simultaneously among the different detection scenarios. The assumption of sphericity was violated for the number of detections and the MDL scenario interaction (W = 0.00004, v235 ¼ 6567:0, P < 0.0001), so the Greenhouse–Geiser adjustment was used. The average number of pesticides detected simultaneously declined with increasing MDLs (F[2.0,1356.1] = 1714.5, P < 0.0001; Fig. 3), and the mean declined from 5.02 pesticides (Native MDL) to 0.08 pesticides detected (>1700 ng1 L MDL). Even having a detection limit as low as 25 ng1 L reduced the number of pesticides detected by about half (mean = 2.45 pesticides detected). The most frequently encountered scenario (e.g. mode) using an MDL of 50 ng1 L, was zero pesticides detected simultaneously (26.1% of water samples), whereas the most common scenario using the native MDL was seven pesticides detected simultaneously (14.4% of water samples), and pesticides were not detected in only 1.7% of water samples.

The different measures used – means, medians, frequency of detections, detection probabilities and number of simultaneously detected pesticides, were all affected by the changes in MDLs. Depending on the method that one uses for replacement of observations below detection limits, means or medians will be either underestimated (0 in lieu of the MDL) or overestimated (½ MDL). Although initially there were not many changes in mean or median concentrations of atrazine when the MDL shifted from the native MDL (5.76 ng1 L) to 25 or 50 ng1 L, subsequent increases in MDLs increasingly biased both means and medians; medians were no longer viable at an MDL of 80 ng1 L, as the median was ½ the MDL or zero. Changing the MDL from native to 25 ng1 L resulted in a 51.2% decrease in the mean number of pesticides detected simultaneously; even small increases in the MDL affected the estimate; an MDL of 50 ng1 L reduced the number of detected pesticides by 27.3% from an MDL of 25 ng1 L. A non-exhaustive search of the scientific literature found that the MDLs reported for atrazine spanned three orders of magnitude, from 1.3 ng1 L to 1300 ng1 L (Table 1); presumably there is a similar range of MDLs for other common pesticides. The proportion of sites where pesticides were detected decreased as the MDL increased, as did detection probabilities for each pesticide. Adjusting for MDLs using maximum likelihood (program PRESENCE) generally improved estimates of the presence of pesticides. However, as the MDLs increased the ability to compensate for detection probabilities deteriorated. When detection probabilities were less than 5% the estimates of the proportion of sites with detections reached 95–100%. However, as MDLs declined, the proportion of sites that have detectable pesticides declined as well, despite the attempt to compensate using by adjusting estimates with their respective detection probabilities. Although using maximum likelihood to model detection probabilities improved estimates when detection probabilities were greater than 5% compared to naïve estimates, as MDLs increased estimates inevitably worsened. We speculate that the model breaks down due to its inability to reliably estimate the parameters at higher MDLs, as almost all sites have no detections, with all the detections confined to a small number of sites. The model then incorrectly estimates that the detection probability is high. Analytical costs can be a major constraint on water quality monitoring studies that seek to quantify the spatial and temporal variability of pesticide concentrations in the environment (Byer et al., 2008), which can affect both the study design as well as the frequency or extent of sampling. Thus, there is substantial incentive to minimize the analytical costs to allow more extensive and intensive sampling for pesticide monitoring. Less expensive methods for measuring pesticides in surface waters, such as ELISA techniques, are useful for broad screening or when looking for exceedences of guidelines, but often have higher MDLs. However,

568

S.R. de Solla et al. / Chemosphere 86 (2012) 565–571

Dicamba

1 0.8 0.6 0.4 0.2

50

S ep 8O ct 22 -O ct 5N ov

-S ep

10

24 -

ug -A ug

-A

100

27

-J ul

30

16

25

13

l Ju

Native

-J ul

un

2-

n

ay

Ju

-J 18

4-

21

7-

-M

pr M ay

r

-A

Ap

23

26 -

9-

M

ar

0

500

Atrazine

1 0.8 0.6 0.4 0.2

Native

25

50

100

ov

ct

5N

ct

O

8O

22 -

4Ju n 18 -J un 2Ju l 16 -J ul 30 -J u 13 l -A ug 27 -A ug 10 -S ep 24 -S ep

ay -M

21

M ay

7-

Ap r

Ap r

23 -

9-

26 -M

ar

0

500

Fig. 2. Seasonal changes in the detection probability of dicamba and atrazine in surface water samples from Ontario, 2003–2008, under different method detection limits (ng1 L). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

300

Native (>0.39 to < 25.1) 50 100

250

Frequency

200

150

100

50

0 0

1

2

3

4

5

6

7

8

9

10

11

12

13

14

Number of pesticides detected Fig. 3. Number of pesticides detected simultaneously in individual water samples, with MDLs corresponding to the native, 50, or 100 ng1 L MDLs; curves were generated using the extreme value distribution. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

a few pesticides have CCME guidelines that are below the higher MDLs (Table 2); there are six pesticides with CCME guidelines which are at or below 50 ng1 L (Table 2). High detection limits may restrict the types of inferences one may make about pesticide residues in surface waters. For example,

analytical methods with high detection limits are insufficient to make claims about the absence of pesticides. Given the native MDL of 5.76 ng1 L (i.e. assuming that atrazine is not present at concentrations lower than this), methods having MDLs of 50 ng1 L would miss about 41% of the atrazine detections. It is also possible

569

S.R. de Solla et al. / Chemosphere 86 (2012) 565–571

a

350

Concentration of atrazine (ng/L)

Mean concentration atrazine (ng/L)

a

300 250 200 150 100 50 0

500

400

300

200

100

0

5.76

25

50

80

100

250

500

Median 25%-75% 10%-90%

1500 1700

5.76

Concentration of atrazine (ng/L)

Mean concentration atrazine (ng/L)

b

1000 900 800 700 600 500 400 300 200 100

25

50

80

100

250

700

250

500

1500 1700

Median 25%-75% 10%-90%

500 400 300 200 100 5.76

25

50

80

100

250

500

1500 1700

Method Detection Limit (ng/L)

Method Detection Limit (ng/L)

that higher MDLs may also miss biologically relevant concentrations in surface waters. For example the CCME guideline for atrazine is 1800 ng1 L, Hayes et al. (2002) argued that concentrations as low as 100 ng1 L were sufficient to cause alterations in the gondal development of leopard frogs (Rana pipiens). Some methods for measuring atrazine with MDLs > 100 ng1 L would miss the presence of this biologically relevant concentration (Table 2). One of the studies testing for the effects of atrazine on Xenopus laevis had a MDL marginally lower than the 100 ng1 L hypothesized to cause gonadal abnormalities in ranid frogs, but the Limit of Quantification (LOQ) was 222 ng1 L, which was above the hypothesized effect level. Similarly, a Lowest Observed Effect Level (LOEL) of 10 ng1 L was found for endosulfan for rainbow trout (Oncorhynchus mykiss) (Arnold et al., 1996), whereas the CCME guideline is 20 ng1 L (Table 2). Valuable information could be lost if MDLs are set for environmental compliance benchmarks only. For the pesticides that are most toxic to aquatic organisms (e.g. endosulfan, azinphos-methyl), monitoring programs with high MDLs may be inadequate to be protective, as the MDLs may exceed toxicity benchmarks even if they do not exceed guideline concentrations. There are also concerns regarding philosophical differences between monitoring for environmental compliance versus general surveillance monitoring. An example of information that could be lost by increasing MDLs is the ability to detect banned pesticides. Silvex and 2,4,5-T were banned for use in North America in 1985

100

600

500 1500 1700

Fig. 4. Mean concentration (SD) of atrazine in surface waters using different method detection limits, with observations below MDL replaced by (a) zero, (b) ½ the MDL.

80

800

0

5.76

50

Method Detection Limit (ng/L)

Method Detection Limit (ng/L)

b

25

Fig. 5. Median (75% and 90% percentiles) concentrations of atrazine in surface waters using different method detection limits, with (a) observations below MDL replaced by zero, and (b) observations below MDL replaced with ½ the MDL.

Table 1 Detection limits reported from various studies for measuring atrazine in water samples. MDL (ng1 L)

References

1.3 1.7 4 14 30 74 155 240 601 1300 5.76 1800

Herranz et al., 2008 Ma et al., 2003 Maqbool et al., 2008 Beale et al., 2009 Hackett et al., 2005 Hecker et al., 2004 Rodriguez-Mozaz et al., 2004 Katsumata et al., 2006 Zhou et al., 2008 Beale et al., 2009 This study CCME guideline for atrazine

due to concerns regarding dioxin contamination during the manufacturing process (Ontario Ministry of Natural Resources, 2011). In this study, the ability to detect silvex and 2,4,5-T was dependent upon using the most sensitive MDL (Fig. 1). Also the ability to characterize seasonal variability of some pesticides such as dicamba was lost (Fig. 2a). High detection limits could also reduce the efficacy of surveillance monitoring for restricted pesticides such as the recent ban on cosmetic pesticides in Canadian municipalities and provinces (Glozier et al., in press).

570

S.R. de Solla et al. / Chemosphere 86 (2012) 565–571

Table 2 Canadian water quality guidelines for the protection of aquatic life (CCME, 1999). Pesticide

Guideline (ng1 L)

Atrazine Bromoxynil Captan Carbaryl Carbofuran Chlorothalonil Chlorpyrifos Cyanazine Deltamethrin Dicamba Diclofop-methyl Dimethoate Dinoseb Endosulfan Glyphosate Imidacloprid Linuron MCPA Metolachlor Metribuzin Tributyltin Permethrin Picloram Simazine Tebuthiuron Triallate Trifluralin

1800 5000 1300 200 1800 180 3.5 2000 0.4 10 000 6100 6200 50 20 65 000 230 7000 2600 7800 1000 8 4 29 000 10 000 1600 240 200

5. Summary We have demonstrated that increasing MDLs (e.g. reduced sensitivity) worsens summary statistics, such as means and medians, and decreases the probability of detecting pesticides in surface waters. It also makes it more difficult to compensate for changes in the detection probability, even when using statistical techniques designed for such situations. At higher MDLs, one cannot adequately characterize seasonality in pesticide concentrations in surface waters, which is critical for doing risk assessments. Increasing MDLs reduces the ability to do investigative or retrospective work such as environmental forensics. Analytical techniques that have high MDLs may be sufficient for compliance monitoring (i.e. guidelines) however they will surely miss ecologically and toxicologically relevant concentrations. Earlier studies have shown that the statistical methods used to deal with values below MDLs can affect the interpretation of the data (e.g. Helsel, 2005, 2006). Despite the importance of using optimal statistical models for compensating for values below MDLs, our data suggests that choosing optimal MDLs is more important than the statistical methods used in response to MDLs.

Role of the funding source This study was funded by Environment Canada’s Environmental Sustainability Indicators and Pesticide Science Fund Initiative. The managers of the source of funding did not participate in the design of the study, nor the interpretation or writing of the manuscript. All such decisions were made by S.R. de Solla, J. Struger, and T.V. McDaniel.

Acknowledgements The authors would like to thank the sampling efforts of Environment Canada field staff, Jake Kraft and Jeffrey Hanna for sample collection, and Logan Heslip for compiling the data. We would also like to thank Axys Analytical and the National Laboratory for Envi-

ronmental Testing (NLET) for analytical support. We gratefully acknowledge the helpful comments from Sarah Gewurtz.

Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.chemosphere.2011.09.026. References Antweiler, R.C., Taylor, H.E., 2008. Evaluation of statistical treatments of leftcensored environmental data using coincident uncensored data sets: I. Summary statistics. Environ. Sci. Technol. 42, 3732–3738. Arnold, H., Pluta, H.J., Braunbeck, T., 1996. Cytological alterations in the liver of rainbow trout Oncorhynchus mykiss after prolonged exposure to low concentrations of waterborne endosulfan. Dis. Aquat. Org. 25, 39–52. Beale, D.J., Porter, N.A., Roddick, F.A., 2009. A fast screening method for the presence of atrazine and other triazines in water using flow injection with chemiluminescent detection. Talanta. 78, 342–347. Byer, J.D., Struger, J., Klawunn, P., Todd, A., Sverko, E.D., 2008. Low cost monitoring of glyphosate in surface waters using the ELISA method: an evaluation. Environ. Sci. Technol. 42, 6052–6057. CCME (Canadian Council of Ministers of the Environment), 1999. Canadian water quality guidelines for the protection of aquatic life. In: Canadian Environmental Quality Guidelines, 1999, Canadian Council of Ministers of the Environment, Winnipeg. de Solla, S.R., Fernie, K.J., Letcher, R.J., Chu, S.G., Drouillard, K.G., Shahmiri, S., 2007. Snapping turtles (Chelydra serpentina) as bioindicators in Canadian areas of concern in the Great Lakes Basin. I. Spatial distribution of PBDEs, PCBs, and organochlorine pesticides in eggs (Chelydra serpentina). Environ. Sci. Technol. 41, 7252–7259. de Solla, S.R., Martin, P.A., 2011. Accumulation of pesticides in snapping turtle eggs (Chelydra serpentina) from treated soil. Chemosphere 85, 820–825. Environment Canada, 1997. Manual of analytical methods, vol. 3, Organics. The National Laboratory for Environmental Testing (NLET), vol. 42. Burlington, ON, Canada. Environ. Sci. Technol. pp. 8452–8458. Glozier, N.E., Struger, J., Cessna, A.J., Gledhill, M., Rondeau, M., Ernst, W.R., Sekela, M.A., Cagampan, S., Sverko, E., Murphy, C., Murray J., Donald. D.B., in press. Occurrence of glyphosate and acidic herbicides in select urban rivers and streams in Canada, 2007. Environ. Sci. Pollut. Res. doi:10.1007/s11356-0110600-7. Hackett, A.G., Gustafson, D.I., Moran, S.J., Hendley, P., Van Wesenbeeck, I., Simmons, N.D., Klein, A.J., Kronenberg, J.M., Fuhrman, J.D., Honegger, J.L., Hanzas, J., Healy, D., Stone, C.T., 2005. The acetochlor registration partnership surface water monitoring program for four corn herbicides. J. Environ. Qual. 34, 877–889. Hayes, T.B., Collins, A., Lee, M., Mendoza, M., Noriega, N., Stuart, A.A., Vonk, A., 2002. Hermaphroditic, demasculinized frogs after exposure to the herbicide atrazine at low ecologically relevant doses. Proc. Natl. Acad. Sci. USA. 99, 5476–5480. Hecker, M., Giesy, J.P., Jones, P.D., Jooste, A.M., Carr, J.A., Solomon, K.R., Smith, E.E., Van der Kraak, G., Kendall, R.J., Du Preez, L., 2004. Plasma sex steroid concentrations and gonadal aromatase activities in African clawed frogs (Xenopus laevis) from South Africa. Environ. Toxicol. Chem. 23, 1996–2007. Helsel, D.L., 2005. Nondetects and Data Analysis: Statistics for Censored Environmental Data. John Wily and Sons, New York. Helsel, D.L., 2006. Fabricating data: how substituting values for nondetects can ruin results, and what can be done about it. Chemosphere 65, 2434–2439. Helsel, D.R., 1990. Less than obvious: statistical treatment of data below the detection limit. Environ. Sci. Technol. 24, 1766–1774. ~ a, E., Marazuela, M.D., Marco, M.P., MorenoHerranz, S., Ramo´n-Azco´n, J., Benito-Pen Bondi, M.C., 2008. Preparation of antibodies and development of a sensitive immunoassay with fluorescence detection for triazine herbicides. Anal. Bioanal. Chem. 391, 1801–1812. Katsumata, H., Kaneco, S., Suzuki, T., Ohta, K., 2006. Determination of atrazine and simazine in water samples by high-performance liquid chromatography after preconcentration with heat-treated diatomaceous earth. Anal. Chim. Acta. 577, 214–219. Kurt-Karakus, P.B., Bidleman, T.F., Muir, D.C.G., Cagampan, S.J., Struger, J., Sverko, E., Small, J.M., Jantunen, L.M., 2008. Chiral current-use herbicides in Ontario streams. Environ. Sci. Technol. 42, 8452–8458. Ma, W.T., Fu, K.K., Cai, Z., Jiang, G.B., 2003. Gas chromatography/mass spectrometry applied for the analysis of triazine herbicides in environmental waters. Chemosphere 52, 1627–1632. MacKenzie, D.L., 2005. Was it there? Dealing with imperfect detection for species presence/absence data. Aust. NZ. J. Stat. 47, 65–74. MacKenzie, D.L., Nichols, J.D., Lachman, G.B., Droege, S., Royle, A., Langtimm, C.A., 2002. Estimating site occupancy rates when detection probabilities are less than one. Ecology 83, 2248–2255. Maqbool, U., Anwar-Ul-Haq, Qureshi, M.J., 2008. Comparison of in-house-developed ELISA with HPLC techniques for the analysis of atrazine residues. J. Environ. Sci. Health, Part B 43, 224–230. Oehlert, G.W., 2006. Detection Limits and Related Tragedies. Statistics 5932 course. School of Statistics, University of Minnesota.

S.R. de Solla et al. / Chemosphere 86 (2012) 565–571 Ontario Ministry of Natural Resources. 2011. 2,4,5-T Herbicide and Health Concerns (accessed 03.06.11). Perkins, J.L., Cutter, G.N., Cleveland, M.S., 1990. Estimating the mean, variance, and confidence limits from censored (
571

Succop, P.A., Clark, S., Chen, M., Galke, W., 2004. Imputation of data values that are less than a detection limit. J. Occup. Environ. Hyg. 1, 436–441. Woudneh, M.B., Ou, Z., Sekela, M., Tuominen, T., Gledhill, M., 2009. Pesticide multiresidues in waters of the lower Fraser Valley, British Columbia, Canada Part I. Surface water.. J. Environ. Qual. 38, 940–947. Zhou, Q.X., Xie, G.H., Pang, L., 2008. Rapid determination of atrazine in environmental water samples by a novel liquid phase microextraction. Chin. Chem. Lett. 19, 89–91.