Trace-level analysis of organic contaminants in drinking waters and groundwaters

Transcription

1 Trends Trends in Analytical Chemistry, Vol. 30, No. 4, 2011 Trace-level analysis of organic contaminants in drinking waters and groundwaters M.J. Capdeville, H. Budzinski We expect drinking water and groundwater samples to be contaminated very little, so they are subject to trace-level analysis. Due to the very low levels of contamination, this sort of analysis requires not only powerful analytical technologies to reach limits around the ng/l level, but also quality-control parameters (e.g., blank and spike samples) to monitor potential contamination or losses during sample treatment. Based on a literature review and laboratory experience, we discuss the problems linked to the difficulties of calculating limits of detection, distinguishing instrumental from methodological limits and preventing false-positive results in cases of sample contamination, or false-negative results in cases of compound losses. When possible, we suggest solutions to compensate for, or to prevent, these problems. ª 2011 Elsevier Ltd. All rights reserved. Keywords: Blank sample; Compound loss; Contamination; Drinking water; False negative; False positive; Groundwater; Limit of detection; Spike sample; Trace-level analysis M.J. Capdeville, H. Budzinski* ISM-LPTC, UMR 5255, CNRS- Université Bordeaux 1, 351 cours de la Libération, Talence cedex, France * Corresponding author. Tel.: +33 (0) ; Fax: +33 (0) ; h.budzinski@ism. u-bordeaux1.fr 1. Introduction Contamination of environmental waters (e.g., surface waters and groundwaters) and of waters intended for human consumption by trace levels of organic substances is a subject of increasing concern in western countries. European regulations [1] on the quality of water intended for human consumption have fixed quality limits of concentration for some substances, including polycyclic aromatic hydrocarbons (PAHs) [e.g., benzo[a]pyrene (0.01 lg/l)], pesticides [e.g., aldrin (0.03 lg/l)] or residues of food-contact materials [e.g., vinyl chloride (0.50 lg/l)]. Nevertheless, occurrence in groundwaters and drinking waters of some other unregulated substances have also been reported in the literature, including pharmaceuticals and personal-care products (PPCPs) (e.g., carbamazepine, caffeine and sulfamethoxazole), endocrine-disrupting compounds (EDCs) {e.g., nonylphenol and flame retardants} [e.g., TCEP (tris(2-chloroethyl) phosphate)] (Table 1). All these substances were detected in the range low lg/l ng/l. The evolution of the analytical devices allows us to improve instrumental sensitivity, pushing scientists to publish protocols referring to lower limits of detection (LODs) and lower limits of quantification (LOQs). Unfortunately, in the field of ultratrace analysis of organic contaminants, other factors, apart from the instrumental performance, have to be taken into account in determination of those limits. Indeed, the contamination of the water sample during its manipulation by some substances present in the working environment (e.g., naphthalene) or carried by the analyst (e.g., caffeine) can produce false-positive results. By contrast, the difficulties of extracting small amounts from water, combined to the breakthrough-volume problems (e.g., paracetamol), can result in false-negative results. It is then necessary to take some precautions when manipulating such samples and it is also important to distinguish the instrumental LODs (IDLs) or instrumental LOQs (IQLs) from the real limits by considering the whole protocol, from sampling to pure analysis. This is even more the case for drinking waters and groundwaters that are very clean matrices requiring very low LODs and LOQs. This article deals with the complexity of reaching a reliable measure to qualify the contamination of a sample at the ultra /$ - see front matter ª 2011 Elsevier Ltd. All rights reserved. doi: /j.trac

2 587 Table 1. Overview of organic compounds and their concentrations (ng/l) found in groundwater and drinking water worldwide. Only positive quantitative data, above the limits of quantification, are reported. Negative data (compounds not found, less than the quantification and detection limits) are not reported Ref. USA [2] Drinking water [3] Drinking water [4] Drinking water Sample PPCPs 2 EDCs (hormones, AkPs, Pesticides 4 Flame type 1 BPA, phthalates) 3 retardants 5 AHTN 490 BPA 420 DEET 66 TBEP 350 Anthraquinone 72 HHCB 82 Prometon 96 TCEP 99 Benzophenone 130 Caffeine 119 TDIP 250 Bromoform 21,000 Carbamazepine 258 TBP 100 Tetrachloroethylene 100 Cotinine 25 Dehydronifedipine 4 Triethyl citrate 62 Erythromycin 4.9 Tylosin 4.2 Roxithromycin 1.4 Oxolinic acid Flumequine Sulfamethoxazole AHTN BPA DEET 70 TDIP 100 Tetrachloroethylene Camphor NP 2 TBP 200 Triethyl citrate 100 OP2EO 100 TCEP 80 Carbamazepine Cotinine 1 20 Caffeine 90 Paracetamol Dehydronifedipine 2 4 Cimetidine 10 Codeine 30 Diltiazem 5 Diphenydramine 4 Sulfathiazole 5 [5] Groundwater 1,4-dichlorobenzene 1170 BPA 2550 DEET 13,500 TCEP 737 Naphthalene 1510 Lincomycin 320 Cholesterol methyl-1Hbenzotriazole 2080 Sulfamethazine 360 Coprostanol 1290 Acetophenone 2670 Sulfamethoxazole 1110 Ethanol, butoxyphosphate Dehydronifedipine 22 Diltiazem 28 Fluoxetine 56 1,7-dimethylxanthine 57 Paracetamol 380 Caffeine 130 Ibuprofen 3110 Line missing Others Trends in Analytical Chemistry, Vol. 30, No. 4, 2011 Trends

3 588 Table 1. (continued) Ref. Canada [6] Drinking water NP1EO NP2EO NP1EC 10 20,000 NP2EC 10 15,000 [7] Drinking water Clofibric acid NP Carbamazepine NP1EO Caffeine NP2EO Cotinine NP3EO Ofloxacin NP4EO Sulfamethoxazole NP5EO NP6EO NP7EO NP8EO NP9EO BPA Di(2-ethylhexyl)phthalate Di-n-butylphthalate [8] Tap water Carbamazepine 5.6 Atrazine Deethylatrazine Deisopropylatrazine 8 11 Cyanazine Simazine 7 16 Europe [9] Tap water Iopamidol NP1EO Diatrizoate NP1EC Diatrizoate 45 Stigmasterol Diatrizoate 18 b-sitosterol Iopromide 29 Iomeprol 12 France [10] Drinking water before chlorination process Line missing Sample PPCPs 2 EDCs (hormones, AkPs, Pesticides 4 Flame type 1 BPA, phthalates) 3 retardants 5 Amitryptiline 1.4 Caffeine 22.9 Carbamazepine 43.2 Diclofenac 2.5 Ibuprofen 0.6 Ketoprofen 3 Naproxen 0.2 Paracetamol 210 Others tr- Trends Trends in Analytical Chemistry, Vol. 30, No. 4, 2011

4 589 [11] Drinking water Atenolol Androstenedione Bezafibrate Androsterone Carbamazepine Oestrone 0.3 Diclofenac Levonorgestrel Fenofibric acid Norethindrone Ibuprofen 1.3 Progesterone Ketoprofen Testosterone Metoprolol Naproxen 0.5 Oxazepam Paracetamol Pravastatin 0.2 Roxithromycin 18.1 Salicylic acid Sulfamethoxazole 0.8 Trimethoprime 1 Germany [12] Groundwater Sotalol 560 Phenazone 25 Diclofenac 590 Iopamidol 300 Amidotrizoic acid 1100 Carbamazepine 900 Anhydro-erythromycin 49 Sulfamethoxazole 410 [13] Drinking water Phenazone 400 Propiphenazone 120 AMDOPH 900 [14] Drinking water Clofibric acid N-(phenylsulfonyl)-sarcosine Propiphenazone 80 [15] Drinking water Phenazone 300 Propiphenazone 90 DP 1000 PDP 150 AMDOPH 550 AMPH 120 Spain [16] Drinking water Simazine 43 Atrazine Desisopropylatrazine Desethylatrazine 33 Line missing Metolachlor 40 Trends in Analytical Chemistry, Vol. 30, No. 4, 2011 Trends

5 590 Table 1. (continued) Ref. Sample PPCPs 2 EDCs (hormones, AkPs, Pesticides 4 Flame type 1 BPA, phthalates) 3 retardants 5 [17] Tap water 4-NP 24 BPA 6 25 Dimethylphthalate 3 4 Diethylphthalate Di-n-butylphthalate Butylbenzylphthalate Di(2-ethylhexyl)phthalate 331 Bottled water BPA 7 (PET) Di-n-butylphthalate 59 Bottled water BPA 2 (PE) Diethylphthalate Bottled water 4-NP 78 (glass) [18] Drinking water NP 85 NP1EC 12 NP2EC 10 [19] Drinking water BPA 5 Simazine 5 32 Atrazine 1 18 Israel [20] Groundwater Sulfamethoxazole 20 1 Drinking water = water sampling at the end of the drinking water treatment before the distribution system; Tap water = water collected from residential tap water [8] or public fountain [17]. 2 PPCPs, Pharmaceuticals and personal care products; AHTN, Acetyl hexamethyl tetrahydro naphthalene; HHCB, Hexahydrohexamethyl cyclopentabenzopyran; DP, 1,5-dimethyl-1,2-dehydro- 3-pyrazolone; PDP, 4-(2-methylethyl)-1,5-dimethyl-1,2-dehydro-3-pyrazolone; AMPH, 1-acetyl-1-methyl-2-phenylhydrazide; AMDOPH, 1-acetyl-1-methyl-2-dimethyloxamoyl-2-phenylhydrazide. 3 EDCs, Endocrine-disrupting compounds; AkPs, Alkylphenols; NP, Nonylphenol; NPEO, Nonylphenol polyethoxylate; NPEC, Nonylphenol carboxylate; BPA, Bisphenol A. 4 DEET, N,N-diethyltoluamide. 5 TBEP, Tris(2-butoxyethyl) phosphate; TBP, Tributyl phosphate; TCEP, Tris(2-chloroethyl) phosphate; TDIP, Tris(dichloroisopropyl) phosphate. Others Trends Trends in Analytical Chemistry, Vol. 30, No. 4, 2011

6 Trends in Analytical Chemistry, Vol. 30, No. 4, 2011 Trends ace level, in the field of very pure matrices (e.g., drinking waters and groundwaters). We review the major analytical difficulties at ultra-trace level (e.g., LODs, falsepositive and false-negative results), including literature advice and feedbacks from experience of our laboratory and know-how acquired in this field. 2. General analytical validation parameters Before going in depth into the specific problems of the trace analysis, it is important to provide a brief reminder of the criteria essential to validation of a method, whatever the level of contamination studied: (1) Use of high-purity reference compounds (99%) [21]. If the purity is lower than 96%, the exact degree of purity of the product must be known to correct the concentrations [22]. The stock and working solutions must be prepared in solvents that are not too volatile; otherwise their evaporation can involve changes in concentrations [22]. (2) Identification of the compounds during the analysis by at least three characteristic parameters {e.g., retention time or the transition of quantification, as recommended by the European Commission (EC) [23,24]}. In mass spectrometry, it is better to have four identification points to ensure the identity of the compound: retention time, at least two ionic signals [one signal for quantification and one for confirmation, two ions in single-ion monitoring (SIM) mode in mass spectrometry (MS) or two transitions in multi-reaction monitoring (MRM) mode in tandem MS (MS 2 )] and the ratio between the two ionic signals [21]. (3) Evaluation of the linearity, the precision (repeatability and reproducibility) and the accuracy of the analytical answer [8,12,18,24 27]. For most authors, the precision of a method was determined by repeated intra-day and inter-day analysis (successive injection of a standard solution in one day and in several successive days, respectively) and was expressed as the relative standard deviation of these replicate measurements [8,24,25]. The EC defines accuracy as the closeness of agreement between a test result and the accepted reference value, and accuracy is calculated by determining trueness and precision, with trueness being assessed through recovery of certified reference material or spiked samples [23]. (4) If possible, the addition of internal standards to evaluate the method performances [28] 3. Limits of detection and quantification As underlined by Glaser et al. [29], LODs are the most important criteria to evaluate the performance of a method. The LOD corresponds to a minimal quantity or concentration, different from zero, which can be reliably detected with a certain degree of confidence [29]. There are various ways of calculating the LOD [30]: (1) use of the signal-to-noise ratio (S/N); (2) statistical calculations based on the variation of the analytical response; (3) via the calibration curve obtained by spiking either pure water or the samples to take into account the matrix effect. When the LOD is defined from the signal-to-noise ratio, the response of the analytes is compared to that of the background noise. Usually, the LOD corresponds to the concentration that gives an S/N ratio equal to 3 and the LOQ to a concentration that gives an S/N ratio equal to 10 [3,17,18,25,27,31,32]. In other cases, the LOD corresponds to the variation (standard deviation) of the response of the successive injection of 7 10 replicates of a sample spiked at a very low concentration, multiplied by a factor of statistical confidence. This method is used by the US EPA and USGS [5,21,28,33]. A third technique, which can be used for both LOD and LOQ, involves using a calibration curve. Wenzel et al. [9] determined their LOQ for iodinated contrast media, alkylphenol polyethoxylates (APEOs) and alkylphenol carboxylic acids (APECs) according to the DIN method, whereby the LOQ is calculated according to the confidence interval at 99% of a calibration curve, obtained with a groundwater spiked with the compounds of interest. Because there is no universal standard method for the calculation of the LOD, the values of LOD can differ depending on the method used. This issue is illustrated for benzo[a]pyrene in meat [34]. For trace analyses, one sample may be considered as contaminated, or not, depending on the way that the LOD is calculated. In order to ensure reliable results, several authors modify their LODs with corrective factors. These modified LODs are more or less equivalent to LOQs. Barnes et al. [5] and Focazio et al. [28] would rather use the term reporting level, which corresponds to five times the methodological LOD. These reported limits correspond to the value beyond which they can quantify the contamination and give the values with confidence. In order to reduce the risks of a false positive, Furlong et al. [21], use interim reporting levels (IRLs) which correspond to twice the methodological LODs. Lastly, Garcia-ac et al. [8] used the term Method Detection Limits (MDLs) when the LODs are established on the transition of quantification, and the term Method Confirmation Limits (MCLs), when the LODs are established on the transition of confirmation. However, the EC [23] and the most recent guidelines on analysis by chromatography and MS recommend use of both transitions (quantitative and confirmatory) to establish the LODs and LOQs

7 Trends Trends in Analytical Chemistry, Vol. 30, No. 4, 2011 In addition, once the way of calculating is defined for LOD and LOQ, it is necessary to understand the difference between the instrumental limits (IDLs or IQLs) and the methodological limits (MDLs or MQLs). The IDLs and IQLs are defined from the injection of standard solutions and are directly related to the sensitivity and to the performances of the analytical instruments. For example, the IDLs of ofloxacin were 4759 pg, 214 pg and 2 pg injected, when it was analyzed, respectively, by: LC/UV (HPLC thermo pump P1500 spectrasystem, injector AS100 spectraseries, detector DAD thermo UV6000; mobile phases: water % TFA/acetonitrile % TFA); LC/MS (HPLC Agilent 1100 series, MS Agilent 1100 series single quadrupole; mobile phases: water + 0.3% HCOOH/acetonitrile + 0.3% HCOOH); and, LC/MS 2 (Rapid Resolution Liquid phase Chromatography Agilent 1200 series, MS/MS Agilent 6410 QQQ; mobile phases: water + 0.3% HCOOH/acetonitrile + 0.3% HCOOH). Globally, for various pharmaceuticals analyzed with these three instruments, a factor of 10 in sensitivity is obtained between LC/UV and LC/MS and another factor of 10 between LC/MS and LC/MS 2. Petrovic et al. [18] also highlight the instrumental dependence of the IDL by comparing values obtained by LC/MS (liquid phase chromatography mass spectrometry) to those obtained in LC/MS 2. The LODs of LC/MS 2 are roughly three times lower than the LODs of LC/MS. Since these LODs are instrument-dependent, it was necessary to achieve technological improvements in order to be able to perform trace analyses, especially for polar and hydrophilic contaminants. The MDLs take into account both the instrumental sensitivity and the impacts of the sample-preparation protocols. The first way of calculating them involves extrapolating the IDLs by taking into account the sample volume and the factor of reconcentration (theoretical IDLs). But this calculation supposes the absence of methodological error (i.e., no loss and no amplification phenomena during sample preparation and no matrix effect during the analysis). Another possibility involves applying the whole protocol (extraction, reconcentration and analysis) to pure water, spiked with very small quantities of compounds. These MDLs are then measured by taking into account all the possible impacts of the protocol but do not take into account matrix interferences during the extraction and the analysis. MDLs obtained by extrapolation of IDLs and considering a factor of reconcentration of 20,000 (extrapolated LODs) and MDLs obtained by the extraction of 1 L of drinking water spiked with targeted compounds (measured LODs) were compared for some pharmaceuticals in the laboratory. In drinking water, the matrix effects are low and the two MDLs have an equal order of magnitude. For example, for diclofenac, extrapolated and measured LODs were 0.5 ng/l and 0.9 ng/l, respectively. However, there are some differences, depending on the compounds: for carbamazepine, sample pre-treatment makes it possible to improve the measured MDLs (0.8 ng/l) compared to those extrapolated from the IDLs (2.5 ng/l); by contrast, when the handling of the sample causes compound losses, which is typically the case for caffeine, the measured MDLs (1.5 ng/l) are more important than those extrapolated from the IDLs (0.7 ng/l). Thus, although the extrapolated LODs are in most cases good estimators of MDLs, the last two examples justify interest in measuring them. Finally, there is a third kind of MDL, corresponding to the LODs measured from the analysis of various types of waters spiked with the targeted compounds. In this case, the potential problems brought during the samplepreparation step, and those related to the nature of the samples themselves, are both taken into account. The effects that the matrix can cause on the extraction recoveries or on the analytical sensitivity are integrated into those calculations. They are the most accurate MDLs, but, in the case of drinking water analyses, the disturbances brought by the matrix are unlikely. Table 2 illustrates the matrix impact on the MDLs for four types of water (i.e., raw tap water, surface water, marine water and wastewater-treatment-plant effluent). All were analyzed according to the same protocol: extraction of the compounds by solid-phase extraction (SPE) on Oasis MCX cartridges with ethyl acetate, ethyl acetate/acetone and ethyl acetate/acetone/ ammonium hydroxide as elution solvents; and, analysis with GC/MS with an HP5/MS capillary column and ultrapure helium as carrier gas [10]. It clearly highlights that the MDLs increase with increasing content of organic matter in water. Ternes [49] showed that the MQLs are instrument dependent and methodology dependent. The LOQs are 10 times lower for the antibiotic analysis when he used SPE rather than freeze-drying as extraction and reconcentration technique. He also highlighted the matrixdependent character of this parameter, since, from drinking water to groundwater, the estrogen LOQ increases by a factor of 2. Although in drinking-water analyses, the LODs are unlikely to be modified due to the matrix effect or signalextinction problems, they can be modified if contamination occurs. Indeed, if blanks are contaminated and the problem cannot be solved, the blank values must be taken into account to define the LOQs. A contaminationbackground noise for bisphenol A of about 4 ng/l was noticed in the laboratory, by compiling a great number of experimental data. Consequently, any sample that potentially contains less than 4 ng/l BPA cannot be analyzed with confidence, because the potential contamination of the sample is masked by the background 592

8 Trends in Analytical Chemistry, Vol. 30, No. 4, 2011 Trends Table 2. Influence of sample composition on the method limit of detection (MDL) (1L extracted for raw tap, surface and marine waters and 500 ml for wastewater effluent) of pharmaceutical compounds analyzed by GC/MS 1 [10] Measured MDLs (ng/l) Raw tap water Surface water Marine water Wastewater effluent Amitryptiline Carbamazepine Diazepam Doxepine Imipramine Nordiazepam Aspirin Diclofenac Ibuprofen Ketoprofen Naproxen Clenbuterol Caffeine Gemfibrozil GC Agilent HP6890, MS Agilent 5973 single quadrupole; ionization by electronic impact at 70 ev; carrier gas ultrapure He; capillary column HP5/MS 5% diphenyl/95% dimethylsiloxane, 30 m 0.25 mm 0.25 lm film thickness. noise, in spite of an IDL extrapolated for the BPA at 1 ng/ L. In this case, the real LOD of the BPA should be at least 8 ng/l (i.e. twice the value of the background noise) to be able to be sure of getting rid of experimental pollution. In other cases, an analytical interference can mask the targeted compound. For example, the ketoprofen signal was shown to be interfered by a molecule that had the same retention time and the same quantification transition. The quantification of the signal in the blank gives a concentration equivalent to 2.5 ng/l. Then, in spite of an extrapolated LOD around 1 ng/l, any sample contamination lower than 3 ng/l will be hidden by the interference. Consequently, the LODs of ketoprofen should be raised to at least 5 ng/l. The last important point to define LOD, is the signal considered. Few authors mention on which signal the LODs are established [3]. Generally, to ensure compound identification in MS, one must check its retention time, one ion or one transition of quantification and another one of confirmation, and the ratio between these two signals. The reliable identification of a compound is achieved when all these criteria match simultaneously. However, the signal of confirmation is often less intense than that of quantification. In the absence of this second signal, the presence of a compound in a sample cannot be certified. Thus, the LOD should be related to the response of the second signal. However, the potential presence of a compound in a sample cannot be completely ignored if the other criteria are fulfilled, especially when the ratio between the two signals is significant. In this case, the LOD is defined according to the quantification signal. Then, to determine whether or not the sample is contaminated by the compound, it is necessary to take into account other parameters {e.g., the context (conditions in which the sample was obtained, preserved and analyzed), the type of sample (e.g., surface water, or bottled water) and other indications [e.g., published results with similar samples or laboratory internal results (frequency and certainty of detection of this compound in this type of sample)]}. Finally, the sample cannot be certified as 100% contaminated or not contaminated. To sum up, although the LOD is the most important criterion for the trace analysis, it is the most variable parameter from one methodology to another. It depends on the compound, the technology, and the matrix. The ideal way of working for trace analysis is: to define the MDL by taking into account the analytical instrument used, the whole protocol (extraction, reconcentration and analysis), the matrix effects; and, to specify the method of calculation adopted. Consequently, an accurate estimation of the LOD requires spiking at low level of one or several replicates of samples (final concentration within the range of 10 ng/l) and making them undergo the whole protocol, whatever the mode of calculation chosen. 4. False positives Good sensitivity is an essential, but not exclusive, criterion to meet during the development of a methodology for trace-level analysis of organic compounds in water samples. Determining whether the sample contamination by targeted compounds originates from the sample or relates to its handling is another very important point. Indeed, although there is no natural background for human-made organic molecules, as there is for metals, they enter the composition of materials (e.g., plastics) or 593

9 Trends Trends in Analytical Chemistry, Vol. 30, No. 4, 2011 products for standard human consumption (e.g., cosmetics) and are present in our every-day environment. As a result, they can be transferred to the samples. The most common sources of sample contamination by organic molecules are: the atmosphere; equipment that is insufficiently cleaned or made with materials inappropriate for the targeted compounds; bad care of the samples during the sampling and the transport (adapted USGS 2006) [35]. In practice, solvents, reagents, glassware, analytical instruments, equipment and work environment are the sources of contamination [22]. They make it possible to qualify a clean sample as contaminated. In the following paragraphs, we review the definition and the necessity to perform blanks in parallel to real samples as well as the various sources of pollution and the possible solutions from literature data and/or feedback from our laboratory experience. In addition to the blanks, the extraction recoveries can also reflect pollution of the samples, when the recovery rates are much greater than 100%. For example, during one of their experiments, Furlong et al. [21] reported an extraction recovery up to 560% for caffeine, clearly proving the contamination of sample. As a result, blanks and extraction-recovery experiments must be performed in parallel Blanks To prevent false positives, many authors make blanks go through analysis in parallel with their samples. In general, they use water free from organic matter and free from the targeted analytes. Depending on the authors, various kinds of water are employed: water used for LC mobile phases (e.g., HPLC grade [33]), ultra pure water (e.g., MilliQ water [17,32]), bottled spring water [37], bottled natural mineral water [38] or homemade water {distilled water [7], laboratory grade [2,36], water filtered on activated carbon [37]}. Several blanks can be made during sample treatment. They are introduced at different moments of the sample treatment, from the sampling to the injection on the analytical instrument, in order to identify the origin of the contamination such as the environment at sampling time, the equipment used for sampling, filtration or extraction, the needle of injection or the chromatographic solvent and gases. Table 3 adapted from USGS [35] lists the principal types of blank that can be realized to validate only the sampling step. In practice, the field blank is the most complete, since it undergoes all the sampling stages (collection, transport and filtration) and makes it possible to validate the whole sampling step [28,32,36]. Berryman et al. [6] made these blanks by using field bottles containing pure water and the same preservative agent that is added to each sample (formaldehyde). These bottles were opened during the collection step in the water-treatment plant, in order to evaluate the potential contamination brought by the equipment, the ambient air or transport. Nevertheless, the field blank is not precise enough to identify the exact source of contamination. The second type of blank that must be integrated in a QC approach is known as laboratory blanks, which undergo the same treatment as the environmental samples [5,7,17,28]. They make it possible to be sure that the sample is not polluted in the laboratory. For Furlong et al. [21], these manipulation blanks correspond to water samples free from organic matter and spiked with Table 3. Common types of blank samples and the questions they address [35] Type Targeted Source(s) of Bias 1 Field blank Equipment blank Sampler blank Filter blank Ambient blank Source-solution blank Sample collection, processing and transport process Basic QC sample: Was my sample contaminated as a result of field activities and exposure? Sample collection and processing equipment system Topical QC sample: Does an initial equipment assessment 2 confirm the suitability of the equipment to provide samples within my data-quality requirements? Topical QC sample: Is my equipment-cleaning protocol adequate? Sampling device (e.g., the D-95 sampler, Fultz pump, or peristaltic-pump tubing) Topical QC sample: Is my sampling device the source of contamination? Filtration device (e.g., the capsule filter, in-line filter holder, aluminum plate filter) Topical QC sample: Is my filtration device the source of contamination? Exposure to atmospheric outfall or other conditions Topical QC sample: Was sample exposure to the atmosphere a contaminant source? The blank water used (e.g., IBW, PBW, or VPBW) Topical QC sample: Was my blank water tainted with respect to my analyte(s) of interest? [QC, Quality control; IBW, Inorganic-grade blank water; PBW, Pesticide-grade blank water; VPBW, Volatile-organic-compound and pesticidegrade blank water]. 1 The bias and variability measured include that from laboratory handling, processing, and analysis of the sample in addition to the targeted source listed. 2 An equipment blank is required for USGS investigations to determine the equipment suitability to provide the analyte data needed to meet study objectives

10 Trends in Analytical Chemistry, Vol. 30, No. 4, 2011 Trends only internal standards. They are extracted and analyzed in parallel with a series of 10 environmental samples. Those blanks really reflect what the handling can bring to a sample during processing, since they correspond to the processing of material, solvents, and gases used for the extraction or purification. They also take into account the atmosphere of the laboratory where the sample treatment is performed, because compounds present in the atmosphere and can be transferred to the solvents, can adsorb on glassware or in the phases that are used (e.g., in SPE methodologies). The last type of blank to be included in the QC approach is the injection blank [17,21,32,33]. Injection blanks make it possible to supervise a possible injectionto-injection carryover. This cross-contamination between injections can result from incomplete injection of the previous sample or insufficient washing of the injection needle. These blanks contain only solvent or water free of organic matter, plus internal and surrogate standards. They are placed directly on the rack of injection. Cahill et al. [33] inserted their injection blanks every 10 analyses, whereas Furlong et al. [21] analyzed them just after the calibration/verification sample (sample containing all the selected compounds). In Method 1694, the US EPA [37] also recommended placing an injection blank right after the analysis of samples containing the targeted analytes, in order to make sure that there is no transfer of pollutants. Even if many authors carry out analysis of blanks in parallel with their samples to ensure the validity of their environmental results, none of them exploit the information provided in the same way, when the blanks appear to be contaminated. Some authors {e.g., Casajuana and Lacorte [17], Chen et al. [7] or Garcia-Ac et al. [8]} subtracted the values of their laboratory and injection blanks from the values of their samples. By contrast, Furlong et al. [21] did not, as was also the case in US EPA methods and 527 [22,39]. One should not subtract the values of the blank from those of the samples, because the concentrations in the blanks are variable. Others {e.g., Barnes et al. [5] or Focazio et al. [28]} would rather use these values to raise their LODs. Thus, Barnes et al. [5] reported that the concentration of compounds in the environmental samples were lower than their MDLs if the values were lower than at least 10 times the contamination of the blanks. Lastly, at the National Water Quality Laboratory (NWQL, USA), the presence of a compound in a sample is validated only if the concentration in the sample is 10 times that in the blank and the concentration in the sample is higher than the LOQs [21]. Other authors have decided on an individual basis, according to the compounds and the samples. Stackelberg et al. [2] specified that the quantities of compounds found in the blanks were either much lower than those found in the associated samples, or, on the contrary, that the compounds identified in the blanks were not found in the associated samples. For example, carbamazepine was dosed at 0.3 ng/l in the field blank whereas it was dosed with a value more than 100 times greater in the associated sample, or, by contrast, the sulfamethoxazole and the trimethoprim, respectively quantified at 0.9 ng/l and 0.1 ng/l in the field blank, are not found in the associated samples. In both cases, they deduced that identification of these compounds in the field blanks or in the laboratory blanks did not compromise qualitatively and quantitatively the results of their analysis. In general, the compounds that are the most frequently detected in the blanks are caffeine, phthalates and bisphenol A (BPA). Blanks of Focazio et al. [28] revealed frequent detection of caffeine, in the range ng/l. They also underlined occasional detection of the metabolite of caffeine (1,7-dimethylxanthine), plasticizers (e.g., bisphenol A or triphenyl phosphate), or molecules that were constituents of cosmetics (e.g., 1,4- dichlorobenzene or methyl salicylate). Furlong et al. [21] found at least once in the blanks 11 of the 14 pharmaceutical substances that they were looking for. Caffeine was found five times out of 99 blanks analyzed, with an average concentration of 77.8 ng/l, and with a maximum concentration of 239 ng/l. Codeine is found three times out of 99 blanks analyzed, with an average concentration of 15 ng/l and with a maximum concentration of 33.4 ng/l. The authors concluded that the contamination of the blank by PPCPs is infrequent and that one laboratory blank every 10 environmental samples treated was enough, but yet essential, to evaluate it Contributions of the technician One of the sources of pollution during sample analysis is the technician. Some compounds [e.g., caffeine or cotinine (a nicotine metabolite), both found in standard products for human consumption like coffee or cigarettes, and acetyl hexamethyl tetrahydro naphthalene (AHTN) found in perfume, or aspirin, an anti-inflammatory drug also found in dermatological creams] can be transferred to samples via the analyst. Tests have been carried out to compare the impact a coffee drinker can have on the transfer of caffeine to the samples. The same water analyzed by a coffee drinker contains 250 ng/l of caffeine, whereas it contains no more than 30 ng/l when it is analyzed by a person who does not drink coffee. This concentration even goes down to less than 5 ng/l if the analyst washes his hands carefully and wears gloves. Other compounds are also transferable. It has been noticed that the presence of phenanthrene or pyrene in laboratory blanks was characteristic of a smoking technician, as illustrated in Table 4. Also, an experienced technician, after working for several years on this problem, pollutes samples less than a beginner, as shown 595

11 Trends Trends in Analytical Chemistry, Vol. 30, No. 4, 2011 Table 4. Influence of the technician characteristics (experienced/inexperienced, smoker or not) on the detection of PAHs in experimental blanks (experienced technician technician used for several years to work on traces level samples; inexperienced technician technician working for the first time on such samples) Quantity (ng) of PAH compounds Experienced non-smoker technician Inexperienced non-smoker technician Experienced smoker technician Naphthalene Acenaphthylene Acenaphthene Fluorene Phenanthrene Anthracene 0.09 nd 0.22 Fluoranthene Pyrene Benzo[a]anthacene nd Chrysene + triphenylene Benzofluoranthene nd nd nd Benzo[e]pyrene nd nd nd Benzo[a] pyrene nd nd nd Perylene nd 4.27 nd Indone[c d]pyrene nd nd nd Dibenzo[a,h + a,c]anthracene nd nd nd Benzo[ghi]perylene nd nd nd Sum with naphthalene and phenanthrene but also pyrene and perylene (Table 4). These two examples confirm that the technician can be the source of sample pollution. In the USGS handbook [35] about the quality of data acquisition for water analysis, chapter 4 on quality control (QC) lists the substances that people can introduce into the sample during sampling. It also applies to analysis of the sample itself. Caffeine, nicotine or N,N-diethyltoluamide (DEET, a mosquito repellent) can be transferred via dirty hands or gloves, alcohol can be brought by breath and many other substances can be transferred from clothes or hair. To reduce the risks of a technician polluting samples, Barnes et al. [5], Focazio et al. [28] and the USGS [35] recommended avoiding use of perfume or insect repellant (DEET) and not consuming products containing caffeine or tobacco when their components are also the target molecules. In conclusion, to limit the transfer of compounds from the technician to the sample, it is recommended to select the person according to their own characteristics (smoker or not, coffee drinker or not) and the targeted analytes. Wearing gloves, a laboratory coat and a mask throughout the process of sampling and analysis is recommended to reduce transfers. In addition, the experience of the analyst strongly contributes to reducing the risks of pollution during handling (Table 4) Contributions of the working environment The environment of the laboratory is also a source of contamination. In our laboratory, a room has been dedicated to experiments on samples that are supposed to be highly contaminated [e.g., waters from sewagetreatment plants (STPs)] and another room to experiments on samples that are not or only slightly, contaminated (e.g., groundwater or drinking water). Initially, the two rooms were the same, but, due to the constant treatment of contaminated samples in the first one the work environment there is more likely to convey contaminants and could consequently pollute future samples for analysis. For this reason, this room was named the contaminated room. By contrast, the environment of the second room is supposed to be cleaner, so this second room is dedicated to analyses of traces and was named the clean room. Their respective ambient atmospheres are verified daily. PAHs were analyzed in blank samples carried out in parallel with environmental samples in each room. The average quantity of PAHs found in blank samples were 17 ng (n = 7) and 8 ng (n = 11) in the contaminated and clean room, respectively. In the same way, polychlorinated biphenyls (PCBs) and polybrominated diphenyl ethers (PBDEs) were analyzed in blank samples carried out in parallel with environmental samples in each room (Fig. 1). Even if there is an overlap area between the quantities found in the blanks of these two rooms, it can noticed that, for the two classes of compounds and for a same parameter, as for PAHs, the values are systematically higher in the room dedicated to the contaminated matrices. These results perfectly reflect the pollution that the laboratory environment itself can generate. As a result, we can strongly recommend to dedicate rooms and hoods to specific applications. Moreover, samples of different levels of contamination should not be analyzed during the same extraction series or at the same time, in the same room, even if they are in different 596

12 Trends in Analytical Chemistry, Vol. 30, No. 4, 2011 Trends sum PCB+PBDE (ng) average maximum minimum median contaminated room (n=3) clean room (n=9) room type Figure 1. Impact of the room where the experiments are done on the quantity of PCBs and PBDEs detected in blank samples (Contaminated room means room where samples supposed to be highly contaminated are treated and clean room means room where only samples supposed not to be or only slightly contaminated are treated; in parallel, the ambient contamination is checked to verify levels for targeted contaminants to be analyzed). extraction series. This will avoid cross-contamination between samples. Respecting this system of specific rooms according to the level of contamination implies doubling all necessary equipment to avoid changing rooms during the experiment. Consequently, it implies a certain cost, organizational rules and strict working methods. These practices are part of demanding QC procedures and methodologies [e.g., Good Laboratory Practice (GLP) or Hazard Analysis Critical Control Point (HACCP)]. Despite these precautions, if some compounds are identified in the laboratory blanks, it is necessary to distinguish the compounds brought by the sample treatment from those present initially. One strategy involves increasing the volume of the tested sample without changing extraction time significantly. If the contamination comes from the sample, then the quantities extracted will increase in proportion to increases in the volume treated. However, if the contamination comes from the working atmosphere, then, whatever the volume treated, the quantity of extracted compounds will be the same. But increasing the sample volume is not always possible because matrix problems can increase or sample size can be limited. Caffeine was analyzed in eight raw waters (RWs) used in the production of drinking waters. With a sample volume of 100 ml, the contamination level of the laboratory blank (11.5 ng) and of the samples ( ng) were roughly the same. In this particular case, it was impossible to know whether the caffeine analyzed in the samples came from the samples themselves or was the result of contamination during sample processing. Increasing the volume of the treated sample from 100 ml to 500 ml enabled going over the background noise (e.g., RW 4). Extraction of 100 ml and 500 ml of RW 4 gave 18.8 ng and 54.8 ng of caffeine, respectively, whereas extraction of 100 ml and 500 ml of blank samples gave 11.5 ng and 20.2 ng of caffeine, respectively. We can conclude that RW 4 probably contained an average of 70 ng/l of caffeine, after subtraction of the blank. By contrast, the other samples always contained caffeine in the same proportions ( ng for 500 ml extracted water) as the laboratory blanks (or even below). We can deduce that these waters initially did not contain caffeine and that the quantities analyzed resulted only from the pollution brought during handling. The differences between samples are due to the variability in ambient contamination. Generally, the more a sample is handled, the more it could be contaminated by the working atmosphere. To reduce this risk, there are certain techniques that can be used {e.g., direct injection [40,41], solid-phase micro-extraction (SPME) [42] or online SPE [8]}; they reduce both handling costs and risks of pollution Contribution of the laboratory equipment used According to the class of targeted contaminants, pollution of the laboratory blank is more likely to be caused by the working environment (e.g., naphthalene, caffeine), the technician (e.g., pyrene, caffeine) or the equipment. Indeed, when targeted compounds are components of plastics (e.g., alkylphenols, bisphenol A or phthalates), the equipment used for the analysis should be chosen carefully. To illustrate this point, on three occasions, two different volumes of the same drinking water were extracted, one by SPE with plastics pipes and one with glass Pasteur pipettes, in order to analyze BPA and 4-NP. The quantities of 4-NP and BPA were, respectively, 3 times and 100 times lower when the analysis was realized with glass Pasteur pipettes rather than plastics pipes. The contamination of the sample by the equipment was obvious. Moreover, with the same equipment employed, the quantities found were higher when greater volumes were extracted. For example, quantities of BPA were respectively 45.5 ng and ng for 0.5 L and 1.5 L of drinking water extracted with plastics pipes, and 0.3 ng 597

13 Trends Trends in Analytical Chemistry, Vol. 30, No. 4, 2011 and 0.7 ng for 0.01 L and 0.1 L of the same drinking water extracted with glass Pasteur pipettes. The larger the volume of sample treated, the longer was the contact time and the greater was the quantity of BPA. Indeed, the equipment was not inert and the longer the sample was in contact with plastics, the more it absorbed BPA (and 4-NP), explaining why the quantities increased according to the increase in volume treated. Casajuana and Lacorte [17] observed that the contamination of their laboratory blanks, made in parallel for analysis of phthalates and other EDCs, could come from polypropylene cartridges and polyethylene frits during SPE. As the equipment used for SPE is a source of pollution, SPE does not seem to be an appropriate technique to analyze phthalates at low levels. Moreover, for the analysis of phthalates, we can notice that the less a sample is handled the less it is polluted. SPME should be preferred because it allows us to minimize the handling of aqueous samples. However, the average (n =3) chromatographic peak areas for di(2-ethylhexyl)phthalate (DEHP) in an injected blank sample were 25,048 ± 5235 and 73,791 ± 12,979 with PDMS 100 lm and PDMS DVB fibers, respectively, showing that, when DEHP was analyzed, laboratory blanks systematically contain the targeted analyte, whatever the type of fiber employed. This unavoidable background noise can be managed by making a fiber blank before the analysis of each sample. If the DEHP area in the sample is not at least twice as high as that in the previous blank, then it cannot be concluded that the sample is contaminated with DEHP. In addition, the state of the analytical system employed (more or less recently subjected to service and maintenance) and the type of fiber chosen could affect the quantities of DEHP detected in the blanks, and could vary from one day to another (Fig. 2). Fig. 2 shows the decrease in DEHP contamination observed in fiber, which can be explained by the decrease in the pollution released from the fiber or a decrease in analytical sensitivity. The impact of glassware on phthalate analysis was studied by SPME. The DEHP concentrations found were 4.2 ± 2.1 ng/l, 3.7 ± 1.2 ng/l, 3.3 ± 1.6 ng/l and 1.5 ± 0.2 ng/l in the same water analyzed with glassware used only for drinking-water analysis, new glassware, new glassware that had been baked and new glassware washed and baked, respectively. New, washed and baked glassware could reduce both the DEHP concentrations and concentration variations in the water analyzed. But, whatever the precautions taken, there was an unavoidable background noise of about 4 ng/l. This residual contamination masked all DEHP levels below 5 ng/l and imposed an LOD of at least 10 ng/l. The influence of the state of the glassware on caffeine analysis was also highlighted in the laboratory. When the whole equipment used for extraction, reconcentration and analysis of caffeine was new, the caffeine level detected in the laboratory blank was 1 ng/l. But, if the equipment used was not new, even if it had been used for drinking-water analysis only and it had been cleaned with detergents and baked at 450 C, traces of caffeine were quantified in the concentration range 3 30 ng/l. This is called the memory effect. The memory effect can be stronger if the equipment used for trace analyses is not clean enough or if it has already been used with contaminated samples. By contrast, the molecules can be lost if the equipment used is not adapted. For example, they can be adsorbed on the inside surface of the containers {e.g., some pesticides on plastic area new fibre + new elements same fibre after a break into the usage new fibre + same elements cleaned new fibre + new elements Figure 2. Evolution of blank fiber contamination depending on the analytical system and fiber states (elements = glass insert SPME, gold plated inlet seal, column cutting)

14 Trends in Analytical Chemistry, Vol. 30, No. 4, 2011 Trends [43] or the tetracyclines on glass [44]}. Consequently, it is fundamental to pay attention to the choice of the material used and its cleaning at every step of the analysis protocol. For example, for the sampling step, almost all authors recommended use of amber or inactinic glass bottles, to prevent the photodegradation of the contaminants [2,3,16,20]. Furlong et al. [21] use stainless-steel containers. For handling, many authors recommended use of laboratory-glass material to limit pollution of the samples by the material employed [21,22,37,39]. For example Montiel [45] recommended use of Pyrex bottles, and Quintana et al. [16] borosilicate Pyrex bottles. Moreover, it is necessary to avoid pollution coming from caps by using Teflon caps to close flasks and bottles [21,22]. In addition, it is very important to pay attention to the products used for cleaning, as some detergents contain compounds [e.g., alkylphenol polyethoxylates (APEOs)]. Looking at the literature, each team seems to have its own cleaning protocol. To minimize the contamination of their material by APEOs, Wenzel et al. [9] cleaned their glassware with precautions, heating it at 250 C for at least 24 hours, then rinsing it with organic solvents. They cleaned the caps of their injection vials for 24 h at 70 C under reduced pressure (50 mbar). Boyd et al. [46] washed their glassware with soap, then soaked it in a solution containing a specific detergent for industrial and medical use and then in a hydrochloric-acid solution. To finish, they baked it at 450 C. The Teflon material was washed in the same way as the glassware except for the baking step. Ye et al. [3] washed their bottles with acid. The US EPA method 1694 [37] recommended washing the glassware with solvents and a solution of detergent, right after use. The various elements have to be disassembled before washing. To optimize elimination of substances, the glassware containing the solution of detergent can be sonicated for a few seconds, and, once washed, it was immediately rinsed with several solvents: methanol, hot water, methanol, acetone, and, lastly, methylene chloride. Finally, it was baked at C. Montiel [45] forbade the use of detergents, but recommended washing with nitric acid and then rinsing the glassware with water, baking it for 30 min at 450 C, and then rinsing it again with the extraction solvent of the compounds that were looking for. Nevertheless, re-use of a bottle can lead to pollution, because of the memory effect. As a result, a bottle that has contained highly polluted samples should not be used again for the analysis of trace materials. Depending on the target compounds, one possible option involves using single-use polyethylene bottles. Watkinson et al. [32] autoclaved their glassware before successively washing it with acetone, methanol and MilliQ water. In addition to their glassware, Furlong et al. [21] also baked Pasteur pipettes and glass-fiber filters at 450 C for 4 h, in order to eliminate all organic-compound residues. Many authors recommended calcination of the glassware [2,9,21,45,46]. To sum up, except for certain particular pollutants, we advise working with glass containers and using Teflon caps. The glassware must be washed with the extraction solvents of the compounds and/or adapted detergents and then baked at 450 C for several hours. The caps must also be washed with adapted detergents. In addition to glassware, the SPE vacuum manifold used for the SPE extraction is another possible source of contamination. The same experiment was performed with and without an SPE manifold, for aspirin analysis. In both cases, SPE cartridges were conditioned with ethyl acetate, then with pure water. They were immediately eluted. In that way, the quantities detected exclusively reflected the contamination brought by the equipment and particularly by the SPE manifold, since it was the only variable element between the two experiments. Blanks contained an average basis of 13 ng of aspirin when the analysis was carried out with the SPE manifold versus 5 ng when the experiment was carried out without the manifold. This clearly showed that the SPE manifold had an impact on sample contamination. The sample-reconcentration step under nitrogen flow can also be a source of pollution. Fluoranthene was detected in all the protocol blanks of various experiments at concentration in the range ng. To identify the source of contamination, the experiment was divided into several steps and several blanks were analyzed after each step. The conclusion was that the problem came from the reconcentration stage under nitrogen flow and from the purity of the gases that were used. It is important to remember that working with analytical gases of high quality (purity superior to % for nitrogen or switch to argon) and excluding gases of industrial quality is of primary importance. Furlong et al. [21] used ultrapure quality nitrogen for the reconcentration stage. Another source of pollution identified in laboratories is linked to the purity of standard compounds, more particularly internal standards. Indeed, erythromycin was analyzed with concentrations in the range ng/l in several drinking waters during various series. In parallel, it was observed that in the same series, concentrations were almost identical, at an average level of 6 ± 0.6 ng/ L(n = 8) or 0.6 ± 0.1 ng/l (n = 3) and that there was a link between the quantity of erythromycin- 13 C 2 added to the samples and the concentration of erythromycin that was quantified. Finally, LC/MS 2 injection of a solution containing only erythromycin- 13 C 2 revealed the presence of erythromycin. Consequently, the erythromycin 599