Flow-injection interface for on-line coupling solid-phase extraction and X-ray fluorescence measurements

Analytica Chimica Acta 594 (2007) 69–74

Flow-injection interface for on-line coupling solid-phase extraction and X-ray fluorescence measurements J.A. P´erez-Serradilla, M.D. Luque de Castro ∗ Department of Analytical Chemistry, Annex Marie Curie Building, Campus of Rabanales, University of C´ordoba, E-14071 C´ordoba, Spain Received 5 March 2007; received in revised form 10 May 2007; accepted 11 May 2007 Available online 21 May 2007

Abstract A flow-injection (FI) manifold including a solid-phase extraction (SPE) mini-column has been coupled on-line to an energy-dispersive X-ray fluorescence detector (EDXRFD) by locating a lab-modified 18-␮L flow-cell, 10-mm path length, connected to the SPE column by PTFE tubing of 0.5 mm i.d., in the X-ray spectrometer chamber. The optical window of the flow-cell was adjusted and fixed to the X-ray irradiation zone of the spectrometer. Two PTFE tubes connected the flow-cell to the FI device and were introduced into the spectrometer chamber by a small orifice without distortion nor modification of the instrument. The SPE–EDXRFD coupling was tested for Pb and Cd aqueous solutions using Dowex 50 cation-exchange resin as sorbent, and flushing the eluate through the flow-cell for monitoring. The LODs and LOQs thus obtained were 1 and 3.2 ␮g for Pb and 1.8 and 4.8 ␮g for Cd, respectively; values which allow using the approach for the analysis of waste water by injecting 20 mL of sample into the FI manifold. The linear dynamic ranges are a function of the sample volume circulated through the mini-column. For a sample volume of 20 mL the ranges are between 1 and 4000 ␮g for Pb and between 1.8 and 2000 ␮g for Cd. The method was validated by the standard addition method using ground-water samples. The SPE–EDXRFD coupling enables to carry out the study of those variables influencing the SPE process – namely, the effect of the sample volume flushed through the column, concentration of analytes in the sample, amount of resin packed, breakthrough volume of the resin, elution profiles, sample pH and retention and elution flow-rates – in an automatic, cheap, fast and precise way. © 2007 Elsevier B.V. All rights reserved. Keywords: Solid-phase extraction; Flow-injection; Energy-dispersive X-ray fluorescence spectrometry; Cadmium; Lead; On-line coupling; Ion-exchange; Dowex

1. Introduction X-ray fluorescence spectrometry (XRFS) is a multielement and nondestructive analytical technique applicable to both liquid and solid samples. The main disadvantage of this technique is poor sensitivity, highly dependent, however, on the type of instrument used [1]. The major drawback in analysing liquids by XRFS is the high background and scattering of the radiation, which make the determination of low concentrated elements difficult. In general, the analysis of non-preconcentrated aqueous samples results in limits of detection in the high ppm range, which is a very limiting factor for the analysis of most natural waters. In order to overcome this drawback, many preconcentration methods for the analysis of metals in aqueous solutions by XRFS have been

∗

Corresponding author. Tel.: +34 957 218615; fax: +34 957 218615. E-mail address: [email protected] (M.D. Luque de Castro).

0003-2670/$ – see front matter © 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.aca.2007.05.018

reported [2]. This step can be carried out in different ways, such as evaporation of the solution on a solid substrate of paper filter [3–5], by precipitation or coprecipitation [6–8] or by preconcentration in ion-exchange resins or sorbents [9–11]; being the last strategy – solid-phase extraction (SPE) – the most widely applied in the last years [12]. In this context, resins such as Chelex-100, Dowex 50, or Amberlite IR-120 are the most frequent sorbents to concentrate cations from natural waters. The necessity for preconcentration in routine analysis [13–15] has led to on-line connection of this step with detection, thus minimizing handling processing [16]; connection which has found a proper interface in flow-injection (FI) manifolds [17]. Flame AA, ICP Emission Spectroscopy and Graphite Furnace AAS are the detection techniques more widely coupled to SPE [17]. Optimization and/or understanding of how the sorption– desorption process takes place involves the study of several influential factors such as volume of sample flushed through the column, analyte concentration in the sample, amount of

70

J.A. P´erez-Serradilla, M.D. Luque de Castro / Analytica Chimica Acta 594 (2007) 69–74

resin packed, breakthrough volume of the resin, elution profiles, sample pH and retention and elution flow rates [18,19–21]. As far as the authors know, XRFS has never been coupled to dynamic preconcentration devices. In this context, the aim of the present work was to develop a continuous FI system for the automatic concentration of metals in water before energy-dispersive X-ray fluorescence spectrometry (EDXRFS) detection in order to overcome the lack of sensitivity of this atomic technique, particularly when low-price instruments are used. The concentration step and subsequent EDXRFS determination can be on-line connected by a conventional flow-cell located in the Xray measurement chamber, thus enabling the overall procedure to take place in an automatic fashion. This on-line approach would enable to carry out, for the first time, the on-line coupling of XRFS and SPE. In addition, the SPE–EDXRFS continuous approach would allow on-line monitoring of the preconcentration kinetics, and an easy way to study the variables influencing the preconcentration process on a given sorbent. Finally, monitoring by using a flow-cell should avoid problems arising for direct XRFS analysis of liquids, such as those related to the presence of bubbles owing to inadequate cuvette filling, and the difficulty to maintain a constant distance between the surface of the liquid and the X-ray tube anode [22]. 2. Experimental 2.1. Instruments and apparatus An energy-dispersive Fischerscope X-ray XAN spectrometer (Helmut Fischer GmbH + Co. KG, Sindelfingen, Germany), designed for coating thickness measurement and materials testing, was used in this study. It consists of an X-ray tube with a tungsten anode (50 kV/0.8 mA) and a silicon semiconductor detector (Peltier cooling at −30 ◦ C; energy resolution 200 eV). The chamber dimensions are H × W × D = 90 mm × 320 mm × 460 mm. A color video microscope allows selection and view of the irradiated area with up to 25 × magnification. The spectrometer is equipped with primary filters of aluminium and nickel. Four collimators with different diameters are also available in the spectrometer. Helmut Fischer software (WinFTM XAN, version 6.05-PDM) was used for instrument control. The samples were analysed using an algorithm based on fundamental parameters [1]. The spectrometer meets all safety and health requirements of the relevant EU and USA guidelines [23]. EDXRFS spectrum acquisition and treatment were carried out by the fundamental parameters method, an atomic physics algorithm that theoretically describes the interaction between the instrument and the analytes by interpreting the processed X-ray spectra into elemental composition, taking into account parameters such as the generation of primary X-rays from the X-ray tube, secondary fluorescent X-ray production from the sample, inter-element matrix effects and detection of the emitted X-rays [24]. A Minipuls-3 low-pressure peristaltic pump from Gilson (Worthington, OH, USA), two 5041 low-pressure selection valves from Rheodyne (Cotati, CA, USA) and PTFE tub-

ing of 0.5 mm i.d. were used to construct the flow manifold. A microcomputer Compaq Contura 3/20 equipped with a laboratory-made iterative program was used for pump control. X-ray measurements were carried out in an 18-␮L flow-cell of 10-mm path-length fabricated in quartz provided by Hellma (Jamaica, NY, USA). This material produces X-ray radiation scattering effects which result in a decrease in sensitivity as compared with measurements carried out using Mylar film as sample support. In order to minimize scattering the flow-cell was modified: an 1-mm orifice was made on the flow-cell optical window to remove the quartz, which was replaced by a Mylar film. Using the video microscope of the spectrometer and choosing the 1 mm diameter collimator, the Mylar window of the modified flow-cell was adjusted with the X-ray irradiation zone and then firmly fixed to the spectrometer chamber with adhesive tape. The two PTFE tubes used to connect the flow-cell to the rest of the continuous manifold were introduced into the spectrometer chamber by a small gap in between the cover and the measuring chamber. It is worth emphasizing that none of the spectrometer parts was distorted nor modified. 2.2. Materials and reagents The strong acidic cation exchanger resin Dowex-50W (crosslinkage 2%, dry mesh 100–200) was from Sigma–Aldrich (Steinheim, Germany). 18 M cm deionized water from a Milli-Q water purification system from Millipore (Molsheim, France) was used to prepare the standards. Stock solutions (1 g L−1 ) of Pb(NO3 )2 (Sigma–Aldrich) and Cd(NO3 )2 ·4H2 O, by Merck (Darmstadt, Germany), were prepared in 1% (v/v) analytical grade nitric acid from Scharlab (Barcelona, Spain); working solutions were daily prepared by appropriate dilution of the stock solution. Acetic acid and sodium hydroxide, used to prepare the buffer solution of pH 6, were provided by Panreac (Barcelona, Spain). Hydrochloric acid, used for column conditioning and the elution step, was from Scharlab. The preconcentration mini-columns were prepared by PTFE tubing (ca. 17 mm length) of 3 mm i.d. with their ends capped by fitting PTFE tubing of 0.5 mm i.d., and packing them with 0.12 g of dry ion exchanger resin Dowex-50W, from Sigma (St. Louis, MO, USA). Before inserting in the dynamic manifold, each column was sealed at both ends with small plugs of glass wool to prevent sorbent losses. Natural water samples were collected in the Guadalquivir river, stored in polyethylene bottles at 4 ◦ C, and spiked with Pb and Cd before SPE–EDXRFS analysis. 2.3. Proposed SPE–EDXRFS continuous procedure Fig. 1 depicts the approach used. The preconcentration procedure was preceded by conditioning the resin in order to remove any metal impurities [25–27,20]. To this end, the resin beads were flushed with 20 mL of deionized water, then with 8 mL of a 1.5 mol L−1 HCl solution and, finally, with 5 mL deionized water to wash off HCl in excess. The flow-rate was 1.5 mL min−1 in all cases.

J.A. P´erez-Serradilla, M.D. Luque de Castro / Analytica Chimica Acta 594 (2007) 69–74

71

Fig. 1. Experimental set-up used for the proposed SPE–EDXRF continuous method. PC, personal computer; FC, flow-cell; W, waste; PP, peristaltic pump; SV, selection valve; S, sample; CS, conditioning solvent, ES, elution solvent.

In the preconcentration step, 20 mL of sample buffered to pH 6 was flushed at 1.5 mL min−1 through the mini-column. The elution step was carried out by passing through the column 3 mol L−1 HCl solution (a very expeditive desorption reagent thanks to the ability of chloride ions to form metal complexes with the target analytes [28]), at 0.5 mL min−1 ; then, the eluate circulated through the flow-cell and was led to waste. The column was regenerated by passing 8 mL of 1.5 mol L−1 HCl solution and 5 mL deionized water to wash off the excess HCl; 1.5 mL min−1 flow-rate was applied in the regeneration step. The column showed no loss of efficiency for at least 100 experiments. Measurements were carried out during the circulation of the eluate through the flow-cell; each X-ray spectrum was the average of those collected during 30 s using the aluminium primary filter. No air bubbles in the flow-cell were observed by the microscope. Spectra treatment was as follows: a blank-spectrum was subtracted from each sample-spectrum – average of those obtained for 30 s – , and the peak areas were obtained by integration of the resulting spectrum, named as analytical spectrum, as shown in Fig. 2 for both analytes. In this case the response area for Pb was the sum of the two peak areas it provides within 10–13 keV, and that appearing at 23 keV for Cd. Ten analytical spectra were recorded from each sample during the 5-min elution time, interval within which the elution is supposed to be complete; the Pb and Cd response areas obtained from each analytical spectrum were added and the total response areas thus obtained for each analyte were taken as experimental response for quantification purposes.

for subsequent measurements as it yielded the best results in terms of signal-to-noise ratio (SNR). As the diameter of the flow-cell window is approximately 1 mm, a collimator with the same diameter was used. The stop-flow mode was tested in order to make measurements in the portion of eluate with the highest concentration of analytes. Despite this modality provided better SNRs, it was rejected as the reproducibility was poor and no information of the extraction kinetics was obtained. Thus, continuous monitoring was adopted. Long counting times (viz. time interval during which spectra are collected for subsequent average to obtain each spectrum) have a positive influence on the instrumental response in conventional X-ray analysis [1], as the measurements are carried out in static solutions and thus with constant concentration during the monitoring time. In our case, measurements are carried out in dynamic solutions where the concentration of the analytes varies continuously; this is the reason why the counting time has a key influence on the instrumental response. Short counting times (in

3. Results and discussion 3.1. Optimization study The influence of the two available primary filters (aluminium and nickel) was investigated and that of aluminium was chosen

Fig. 2. An analytical EDXRF spectrum collected during the first 30 s of an elution step; 20 mL of sample containing 20 ␮g mL−1 of each analyte, Pb and Cd, was flushed through the SPE column.

72

J.A. P´erez-Serradilla, M.D. Luque de Castro / Analytica Chimica Acta 594 (2007) 69–74

the few seconds interval) allow a better knowledge of the elution kinetics, as it provides information on the evolution of the concentration with time, but to the detriment of the SNR. On the other hand, large counting times (in the minutes interval) provide less information on the elution kinetics but improvement of the SNR is not always warranted as the counting time can cover intervals during which the concentration of the analytes is negligible. 30 s counting time was selected for subsequent experiments as a compromise between kinetics information and SNR. 3.2. Effect of the sample volume Large volumes of sample are usually preconcentrated in order to achieve low detection limits [29]. For this reason, 5, 10, 25 and 50 mL samples containing 10 ␮g mL−1 of each cation were examined flushing them through 0.12 g of resin. Linear changes according to the total amount of the target analytes were obtained within the range of volumes and concentration studied. Thus, the analysis of low concentrated solutions can be carried out by injecting large sample volumes, but to the detriment of the analysis frequency. 3.3. Effect of analyte concentration Four different volumes of water (namely, 5, 10, 25 and 50 mL) were spiked with the same amount of both cations: 50 ␮g. No significant differences were observed between the responses obtained after flushing them through 0.12 g of resin. Thus, it can be concluded that the behaviour of the resin is independent of the metal concentration within the range studied, namely: 1–10 ␮g mL−1 . 3.4. Effect of the resin amount The amount of sorbent packed in the mini-column is a crucial variable to achieve quantitative retention of the analytes. Ten different amounts of dry resin (in the range 0.01–0.2 g) were tested, which were flushed with 20 mL of 100 ␮g mL−1 of each analyte. An increase in the resin amount leads to increased retention efficiency traduced in higher recoveries, up to an amount of resin which provides quantitative recovery for each metal and levelled off for higher amounts. Fig. 3 shows the behaviour for both cations: cadmium requires smaller amount of resin to be quantitatively retained, as compared with lead. A resin amount equal to (or higher than) 0.12 g is necessary to guarantee quantitative retention of 2 mg of both analytes. 3.5. Breakthrough volume The behaviour of each analyte was studied separately by using different volumes of 300 ␮g mL−1 of each analyte subjected to the proposed procedure. The maximum retention capability of the resin thus calculated was 100 mL of Pb or 50 mL of Cd solutions flushed through 0.12 g of the resin – higher volumes did not increase the response. Thus, the ion

Fig. 3. Effect of the resin amount on the recoveries achieved of each analyte. 20-mL volumes of a solution containing 100 ␮g mL−1 of each, Pb and Cd, were flushed through the SPE column. () Cd; () Pb.

exchange capability of the resin is estimated to be approximately 2.4 mequiv/g dry resin. 3.6. Effect of sample pH It has been established that Cd and Pb are quantitatively retained from solutions buffered at pH ranging from 4.0 to 6.5 [18]; thus, the samples were buffered to pH 6. In preliminary studies, solutions covering the full pH range were examined in order to verify the foreseeable behaviour. When the pH was out of the range 4.0–6.5 the retention efficiency decreases; this effect being more significant as the pH is farther from the optimum range, ascribed to competition with the proton for the sorption sites and precipitation of the cations hydroxides (pH lower and higher than the optimum range, respectively. 3.7. Effect of flow-rate Retention flow-rates higher than 1.5 mL min−1 decreased metal retention owing to shorter residence time of the sample in contact with the sorbent and also to growing overpressure [16]. A flow rate of 1.5 mL min−1 was considered the best for quantitative retention, also obtained at lower flow-rates, but to the detriment of the sampling frequency. This flow-rate was also used for conditioning and regeneration. A flow-rate of 0.5 mL min−1 was selected for elution as it proved the more suitable to monitor the elution kinetics of 20 mL analytes solutions in the range of 0.05–50 ␮g mL−1 . Higher flow-rates decreased the analysis time but to the detriment of precision and definition of the elution profiles, especially for diluted samples; lower flow-rates enlarged the analysis time. 3.8. Elution profiles As one analytical spectrum is collected each 30 s, the proposed methodology enables monitoring the elution kinetics. Fig. 4 shows the elution profiles obtained by analysing 20 mLsamples spiked with three different amounts of Pb and Cd. As expected, complete elution takes place in a shorter time when the amount of metals retained is smaller. By comparing the elution profiles of Cd and Pb, it can be observed that the latter elutes faster.

J.A. P´erez-Serradilla, M.D. Luque de Castro / Analytica Chimica Acta 594 (2007) 69–74

73

Table 1 Recoveries, expressed as percentage, obtained for the SPE–EDXRFS determination of Pb and Cd in 20-mL ground-water samples spiked with different metal amounts (n = 5 replicates) Analyte Metal amount added (␮g) 0 Pb Cd

6

n.d. 104 (7.05) n.d. 99 (8.95)

10

20

50

100

102 (7.35) 96 (9.22)

98 (6.73) 96 (8.64)

100 (5.58) 98 (7.85)

102 (4.81) 100 (7.07)

n.d.: not detected; errors, in brackets, expressed as percentage.

Two measurements per solution and day were performed on 7 days. The repeatability and within-laboratory reproducibility, both expressed as relative standard deviation, were 4.65 and 10.23% for Pb and 6.76 and 13% for Cd, respectively. 3.10. Application of the proposed method to spiked water samples

Fig. 4. Elution profiles obtained after charging the resin with 20 mL of 10 (A), 50 (B) and 100 ␮g mL−1 (C) of each analyte. Percent elution is expressed as the amount of analyte eluted each 30 s with respect to the total amount of analyte eluted from the resin. EDXRF counting time is 30 s. Cd, black bars.

The LODs achieved with the proposed approach are in the low ␮g range – not bad, taking into account the characteristics of the EDXRFS detector at our disposal – , so the proposed method can be used for the analysis of waste waters without introducing large amounts of sample, as most of the European environmental regulations establish the maximum concentration of Pb and Cd allowed in waste waters within 0.1–0.5 ␮g mL−1 [31]. The need for high sample volumes makes expensive the use of certified water reference materials to validate the proposed method; so, spiked ground-water samples were analysed in order to demonstrate its applicability. Thus, 20-mL ground-water samples buffered to pH 6 were spiked with five different amounts of Pb and Cd (between 6 and 100 ␮g) and analysed using the standard addition method (none of the analytes had been detected in the samples). The results, summarized in Table 1, show that nearly quantitative recoveries were obtained in the range studied. As expected, poorer precision values were obtained at the lowest concentrations.

3.9. Characterization of the method 4. Conclusions and future outlook Calibration curves were obtained by plotting the experimental response of Pb and Cd as a function of the amount of analyte added to 20 mL buffer. The optimal working conditions described in Section 2 were used in all instances. The regression coefficients were 0.9995 and 0.9939 for Pb and Cd, respectively. The limit of detection (LOD) and that of quantification (LOQ) for each analyte were expressed as the mass of analyte which gives a signal that is 3σ and 10σ, respectively, above the mean blank signal (where σ is the standard deviation of the blank signal). The LOD and LOQ values were 1 and 3.2 ␮g for Pb, and 1.8 and 4.8 ␮g for Cd, respectively. The linear dynamic ranges are between 3.2 and 4000 ␮g for Pb and between 4.8 and 2000 ␮g for Cd when the sample volume was 20 mL. The precision of the proposed method in terms of withinlaboratory reproducibility and repeatability was assessed by using a single experimental setup with duplicates [30]. Tests involved the analysis under the optimal working conditions of 20 mL of water solutions spiked with 200 ␮g of each analyte.

Coupling SPE and EDXRFS using a FI manifold as interface has for the first time been carried out. The LODs obtained after on-line preconcentration of 20-mL sample prove its usefulness for waste water analysis. Lower LODs can be achieved by using higher sample volumes, but to the detriment of the analysis frequency. A promising door is here open to future couplings using more sensitive X-ray fluorescence detectors which may lower the LODs at the ng mL−1 without the use of large sample volumes, thus enabling automatic and fast analysis of drinking waters. As compared with similar studies in the literature using discontinuous batch approaches [18–20], the SPE–EDXRFS coupling enables to carry out a complete study of influential factors on SPE processes (viz. the effect of the volume of sample flushed through the column, that of analyte concentration in the sample, sorbent amount packed, breakthrough volume of the sorbent, elution profiles, sample pH and retention and elu-

74

J.A. P´erez-Serradilla, M.D. Luque de Castro / Analytica Chimica Acta 594 (2007) 69–74

tion flow-rates) in an automatic, non-expensive, fast and precise way. Acknowledgements Spain’s Ministry of Science and Technology is gratefully acknowledged for financial support (Project No. CTQ200601614). J.A. P´erez-Serradilla is also grateful to Spain’s Ministry of Education for award of his research training fellowship. References [1] B. Beckhoff, B. Kanngieber, N. Langhoff, R. Wedell, H. Wolff, Practical X-Ray Fluorescence Analysis, Springer, Berlin, 2006. [2] R. Van Grieken, Anal. Chim. Acta 143 (1982) 3. [3] J. Smits, J. Nelissen, R. Van Grieken, Anal. Chim. Acta 111 (1979) 215. [4] I. Hanif, J. Hanif, S.M. Hasany, M.Z. Iqbal, X-Ray Spectrom. 24 (1995) 298. [5] A.F. Wilson, D.C. Turner, A.A. Robbins, Process Analytics, Orem, UT, 1998. [6] S. Per¨aniemi, Veps¨al¨ainen, H. Mustalahti, M. Ahlgr´en, Fresenius J. Anal. Chem. 344 (1992) 118. [7] E. Almeida, V.F. Nascimento Filho, E.P.E. Valencia, R.M. Cunha e Silva, J. Radioanal. Nucl. Chem. 252 (2002) 541. ´ ´ ´ [8] A. Montero-Alvarez, J.R. Est´evez-Alvarez, R. Padilla-Alvarez, J. Radioanal. Nucl. Chem. 245 (2000) 485. [9] D.E. Leyden, T.A. Patterson, J.J. Alberts, Anal. Chem. 47 (1975) 733. [10] V. Cojocaru, S. Spiridon, J. Radioanal. Nucl. Chem. 170 (1993) 259. [11] Z.T. Jiang, J.C. Yu, H.Y. Liu, Anal. Sci. 21 (2005) 851. [12] P.J. Potts, A.T. Ellis, P. Kregsamer, C. Streli, C. Vanhoof, M. West, P. Wobrauschek, J. Anal. At. Spectrom. 20 (2005) 1124.

[13] P. Benes, V. Majer, Trace Chemistry of Aqueous Solutions, Elsevier, Amsterdam, 1980. [14] J.W. Patterson, R. Passion, Metals: Speciation, Separation and Recovery, Lewis Publishers, Chelsea, 1990. [15] D.E. Leyden, W. Wegscheider, Anal. Chem. 53 (1981) 1059. [16] M. Valc´arcel, M.D. Luque de Castro, Flow Injection Analysis: Principles and Applications, Ellis Horwood, Chichester, 1987. [17] M. Valc´arcel, M.D. Luque de Castro, Non-chromatographic Continuous Separation Techniques, The Royal Society of Chemistry, Cambridge, 1991. [18] P. Pohl, Z. Bogdal, B. Prusisz, Microchim. Acta 150 (2005) 253. [19] E. Pehlivan, T. Altun, J. Hazard. Mater. B143 (2006) 149. [20] P. Pohl, B. Prusisz, Anal. Chim. Acta 502 (2004) 83. [21] X. Hou, H.L. Peters, Z. Yang, K.A. Wagner, J.D. Batchelor, M.M. Daniel, B.T. Jones, Appl. Spectrosc. 57 (2003) 338. [22] L.P. Eksperiandova, A.B. Blank, Y.N. Makarovskaya, X-Ray Spectrom. 31 (2002) 259. [23] Fischerscope X-ray Xan Operator’s Manual, Version 2.0, 07/02. [24] D. Kloos, Precious Met. 24 (2001) 17. [25] S. Stra␤burg, D. Wollenweber, G. W¨unsch, Fresenius J. Anal. Chem. 360 (1998) 792. [26] C. Pin, S. Joannon, Talanta 57 (2002) 393. [27] A. Krata, K. Pyrzynska, E. Bulska, Anal. Bioanal. Chem. 377 (2003) 735. [28] O. Samuelson, Ion Exchange Separation in Analytical Chemistry, John Wiley, New York, 1963. [29] B. Salbu, E. Steinnes, Trace Elements in Natural Waters, CRC Press, Boca Raton, FL, 1995. [30] D.L. Massart, B.G.M. Vanderginste, L.M.C. Buydens, S. De Jong, P.J. Lewi, J. Smeyers-Verbeke, Handbook of Chemometrics and Qualimetrics, Part A, Elsevier, Amsterdam, 1997. [31] European directives: 83/513/CEE, 86/280/CEE, 90/415/CEE, 2000/60/CE, 96/61/CE.