Brian M. De Borbaa and Edward Todd Urbansky*b
aMetrohm-Peak, Inc., 12521 Gulf Freeway, Houston, TX 77034, USA. E-mail: brian.deborba@ metrohm-peak.com; Tel: +1 281 484 5000; Fax: +1 281 484 5001
bUnited States Environmental Protection Agency, Office of Research and Development, National Risk Management Research Laboratory, Water Supply and Water Resources Division, 26 West Martin Luther King Drive, Cincinnati, OH 45268, USA. E-mail: urbansky.edward@epa.gov; Tel: +1 513 569 7655; Fax: +1 513 569 7658
First published on 12th December 2001
Interest in possible sources of perchlorate (ClO4−) that could lead to environmental release has been heightened since the Environmental Protection Agency placed this anion on its Contaminant Candidate List for drinking water. Although recent investigations have suggested that fertilizers are minor contributors to environmental perchlorate contamination overall, there is still interest in screening commercial products for possible contamination and quantitating perchlorate when it is found. Ion chromatography (IC) has been used for this application owing to its speed, low detection limits, widespread availability, and moderate ruggedness relative to other techniques. However, fertilizer matrixes complicate the IC analysis relative to potable water matrixes. In this study, the performance of poly(vinyl alcohol) gel resin IC columns (100 mm and 150 mm) was evaluated for fertilizer matrixes using method EPA/600/R-01/026. The NaOH eluent included an organic salt, sodium 4-cyanophenoxide. Detection was by suppressed conductivity. A set of 55 different field samples representing 48 products and previously used by the EPA to assess occurrence of perchlorate in fertilizers (EPA/600/R-01/049) was reanalyzed on the 150 mm column. The 100 mm column was used to further investigate the positive hits. Both columns gave satisfactory performance in fertilizer matrixes, with spike recoveries (±15%), assured reporting levels (0.5–225 µg g−1 except for one at 1000 µg g−1), accuracy (relative error < 30% always and most < 15%), and precision [injection-to-injection reproducibility < 3% relative standard deviation (RSD)] comparable to those reported in other studies. Performance did not vary substantially between column lengths. Lastly, the results of this investigation provided further evidence in support of the conclusions that had been reached previously by the EPA on the occurrence of perchlorate in fertilizers.
Ion chromatography (IC) has been the principal technique employed to test environmental samples for perchlorate,20 and it is the basis of methods for drinking waters21 and fertilizers.22 Previous applications of IC to fertilizers have relied on columns that could be used to effectively separate perchlorate without an organic additive, such as 4-cyanophenoxide, which is a polarizable anion of low charge density and can therefore displace perchlorate from the resin. In fact, in a recent comparison of laboratory performance on fertilizer analysis conducted by the EPA, all participating laboratories indicated they had used such columns.23 The high ionic strength typical of fertilizer matrixes complicates the analyses and raises IC detection limits relative to drinking water or standards made in deionized water.23 In this report, we examine the performance of poly(vinyl alcohol) (PVA) gel columns on fertilizer matrixes and compare the results to those obtained previously.23 Unlike previous fertilizer work, the eluent here contained 4-cyanophenoxide, which is prepared in situ by the addition of 4-cyanophenol to a solution of sodium hydroxide. In part, our objective was to determine if such an approach, which was equivalent to initial approaches to drinking water, could be readily applied to the fertilizer matrices. PVA columns have been used to analyze potable, waste, and ground water samples with performance comparable to that of other columns on the market,24 but have not been tested on the kinds of solutions that typically result from dissolving or leaching fertilizers. Such solutions have dissolved matter concentrations and ionic strengths that are characteristically much higher than raw or finished drinking water. Although all IC columns suffer from reduced performance as ionic strength is increased, this effect was pronounced in a study which relied on 4-cyanophenoxide anion to displace sorbed perchlorate ion.25 Consequently, matrix interference could potentially limit the applicability of columns where organic modifiers are required for analysis of fertilizers.
Sample | Descriptionb | Samplec | Description | Sample | Description |
---|---|---|---|---|---|
a Detailed manufacturer/supplier and site data are presented in ref. 23.b The designation “ag” refers to an agricultural grade commodity chemical; these grades often contain impurities or additives not found in reagent grade chemicals. The designation “ACSr” refers to an American Chemical Society reagent grade chemical. SPF = water-soluble plant food; TRF = timed-release fertilizer (e.g., a lawn winterizer). The hyphenated numbers are the guaranteed analyses (N∶P∶K ratios), where the macronutrients are expressed as follows: nitrogen as N, phosphorus as P2O5, and potassium as K2O.c For samples 31, 33, 34, 48, and 55, the letter notation (a or b) identifies blind duplicate solid samples from which distinct test solutions were prepared and analyzed by IC.d These samples are derived from caliche; therefore, perchlorate is expected as a naturally occurring impurity. Typical perchlorate concentrations encountered in recent manufacturing lots have varied from 1 to 2 mg g−1, but steps have been taken to reduce the concentrations by 90–95%. For further detail, see refs. 1 and 13.e Granular triple superphosphate (GTSP) is a somewhat ill-defined mixture of hydrated calcium phosphates. It contains primarily calcium dihydrogenphosphate monohydrate, Ca(H2PO4)2·H2O, but species with other degrees of protonation and hydration may also be found as part of the admixture. In pure form, GTSP should normally have a guaranteed analysis of 0-46-0.f These items were supplied as quality control standards. They were prepared and evaluated by the EPA. The cation listed in parentheses indicates the perchlorate salt used to fortify the material, if any. The native materials were verified to be perchlorate-free within the limits of experimental error by the EPA prior to fortification. No analyte was added to samples 53, 54, or 55. However, sample 55 is a caliche-derived material and contains perchlorate naturally at a concentration of 1.7 mg g−1. | |||||
1 | TRF 22-3-14 | 21 | ag K2SO4 | 38 | ag GTSPe |
2 | ag (NH4)2HPO4 | 22 | ag (NH4)SO4 | 39 | ag (NH4)H2PO4 |
3 | ag urea | 23 | ag (NH4)H2PO4 | 40 | ag (NH4)2HPO4 |
4 | ag KCl | 24 | ag iron oxide | 41 | ag limestone |
5 | ag iron oxide | 25 | ag limestone | 42 | ag KNO3d |
6 | ag limestone | 26 | ag urea | 43 | TRF 10-10-10 |
7 | ag K2Mg2(SO4)3 | 27 | ag clay | 44 | Clay |
8 | ag KCl | 28 | ag KCl | 45 | ag K2Mg2(SO4)3 |
9 | TRF 18-6-12 | 29 | ag urea | 46 | ag KNO3 |
10 | TRF 36-6-6 | 30 | ag NH4NO3 | 47 | ag (NH4)2HPO4 |
11 | SPF 20-20-20 | 31a | ag (NH4)2HPO4 | 48a | ag GTSPe |
12 | ag K2Mg2(SO4)3 | 31b | ag (NH4)2HPO4 | 48b | ag GTSPe |
13 | ag K2Mg2(SO4)3 | 32 | ag (NH4)H2PO4 | 49 | ACSr KCl + 6.8 mg g−1 ClO4− (K+)f |
14 | ag K2Mg2(SO4)3 | 33a | ag KNO3 + NaNO3d | 50 | SPF + 6.2 mg g−1 ClO4− (Na+)f |
15 | ag limestone | 33b | ag KNO3 + NaNO3d | 51 | ag GTSP + 2.7 mg g−1 ClO4− (Na+)f |
16 | ag (NH4)SO4 | 34a | ag NH4NO3 | 52 | ag urea + 1.8 mg g−1 ClO4− (NH4+)f |
17 | ag urea | 34b | ag NH4NO3 | 53 | ag KClf |
18 | ag (NH4)SO4 | 35 | ag KNO3d | 54 | ag NH4NO3f |
19 | ag (NH4)H2PO4 | 36 | ag NaNO3d | 55a | ag NaNO3d,f |
20 | ag K2Mg2(SO4)3 | 37 | ag NH4NO3 | 55b | ag NaNO3d,f |
Metrosep A Supp 5 (4 × 100 mm and 4 × 150 mm) columns (Metrohm Ltd.) with an average resin particle diameter of 5 µm were used throughout this study. All samples were subjected to analysis on the 150 mm column, while the 100 mm analytical column was reserved for further exploration and comparison of those samples that were already found to contain perchlorate using the 150 mm column. Sample loop sizes were 500 µL for the 100 mm column, and 1000 µL for the150 mm column. The eluent was ∼17–20 mM NaOH with sodium 4-cyanophenoxide (prepared in situ) used as a modifier at a concentration of 1 mM for the 100 mm column or 2.5 mM for the 150 mm column. The exact NaOH concentration was unimportant because the perchlorate peak was well resolved from other peaks and any shifts in retention time were discernible based on the behavior of an analyte-fortified sample. The eluent flow rate was 0.70 mL min−1 regardless of column length. Column lifetime is about 9–12 months with heavy use, that is, several days a week of nonstop running during which strong base is in constant contact with the resin.
Parametera | ppb calibration | ppm calibration | ||
---|---|---|---|---|
100 mm | 150 mm | 100 mm | 150 mm | |
a Key: m = slope, εm = standard error in slope, εm/m = relative error in slope; b = y-intercept, εb = standard error in y-intercept, εb/b = relative error in y-intercept; A = a peak area for the concentration indicated by the subscript.b The notation (ppb) or (ppm) refers to the columns of data to which the parameter applies, columns 2 and 3 or columns 4 and 5, respectively.c [ClO4−]calc is the back-calculated value of the analyte concentration assuming a peak area equal to either the y-intercept ([ClO4−]calc = b/m) or the sum of the y-intercept and its standard error {[ClO4−]calc = (|b| + εb)/m] as indicated in the entry. This is an artefactual detection limit based on the calibration standards; see text for further explanation. It is negative when the y-intercept is negative, and its absolute value represents a measure of an artefactual method detection limit resulting from the choice of calibration standards. Units for this parameter are (ng mL−1) for columns 2 and 3 and (µg mL−1) for columns 4 amd 5.d Units for the slope are reciprocal concentration units, i.e., (mL ng−1) for columns 2 and 3 and (mL µg−1) for columns 4 and 5. | ||||
m ± εmd | 44.3 ± 0.07 | 52.8 ± 0.2 | 252 ± 7 | 848 ± 27 |
εm/m (%) | 0.15 | 0.34 | 2.81 | 3.18 |
b ± εb (unitless) | −38 ± 18 | −18 ± 48 | −256 ± 400 | −565 ± 1525 |
εb/b (%) | −46 | −268 | −156 | −270 |
R2 | 0.99997 | 0.9998 | 0.992 | 0.990 |
1 − |b/εb| | −1.18 | — | — | — |
|b|/A5ppb (ppb)b or |b|/A0.5ppm (ppm) (%) | 20 | 7.3 | 74 | 18 |
|b|/A10ppb (ppb)b or |b|/A1ppm (ppm) (%) | 9.5 | 3.6 | 30 | 8.0 |
εb/A10ppb (ppb)b or εb/A1ppm (ppm) (%) | 4.4 | 9.7 | 47 | 12 |
εb/A100ppb (ppb)b or εb/A10ppm (ppm) (%) | 0.42 | 0.94 | 3.4 | 1.0 |
[ClO4−]calc for bc | −0.86 | −0.33 | −1.01 | −0.67 |
[ClO4−]calc for |b| + εbc | 1.26 | 1.23 | 2.61 | 2.47 |
There is attenuation of the response in moving from the ppb calibration to the ppm calibration. Specifically, we would expect the slopes for the ppm calibrations to be about 1000-fold the slopes of the ppb calibrations based simply on the analyte concentrations. However, the ratio of the slopes of the 150 mm column is 16, and the ratio of the slopes of the 100 mm column is 5.6. While all of the regression coefficients are greater than 0.99, there is clearly nonlinear variation in sensitivity, as indicated by the change in slope; therefore, careful calibration is required over the concentration range of interest. When the calibration region is large enough that deviations from linearity become apparent, as is the case here, an unweighted least squares regression sacrifices the quality of the fit for data nearer to the origin (where the ordinate values are smaller in magnitude) and necessitates multiple calibration intervals. This is most conspicuous when examining the ppm calibration. The data for the low-concentration standards essentially become irrelevant in an unweighted fit. The y-intercepts for both columns are statistically indistinct from zero, i.e., their uncertainties are larger than their magnitudes. Nevertheless, that alone is not sufficient proof of the goodness of fit, for the magnitude of both the y-intercept (which incidentally is negative) and its standard error are such that no concentration below 0.7 ppm can be quantitated using the regression line found from these standards. This is evident from the back-calculated analyte concentrations (identified as [ClO4−]calc in Table 2) found using either the y-intercept or its error. We emphasize that this behavior is a result of the concentrations chosen for the calibration standards and therefore is related to these calibration curves rather than the columns specifically. It does not mean that the columns and instrument are incapable of measuring concentrations below these values, but that the specific calibration lines under discussion cannot be used. These artefactual limits of detection can be improved by increasing the number of points chosen at low concentration, by weighting those points, by restricting the concentration domain, and/or by fitting the data to a nonlinear function. At some point, the systematic variation in sensitivity (i.e., net instrument/detector nonlinearity) is no longer a governing factor, and the indeterminate error in the measurement controls the value obtained for the detection limit.
Similarly, we can compare the peak areas of the lowest standards with the magnitudes of the y-intercepts and their standard errors. This comparison is especially telling for the ppb calibrations. For the 100 mm column, the y-intercept is equal to 20% of the area of a 5.00 ng mL−1 standard. On the other hand, for the 150 mm column, the y-intercept is equal to only 7.3% of the area of a 5.00 ng mL−1 standard. At first glance, this suggests that the detection limits should be lower on the 150 mm column. Nevertheless, we must also consider the standard errors of the y-intercepts. The trend in ratios (εb/A5ppb) of the standard errors of the y-intercepts to the areas of the 5 ppb standards is reversed; that is, the 100 mm column has a ratio of 0.42 (better), while the 150 mm column has a ratio of 0.94 (worse). The larger this ratio, the closer the standard error is in magnitude to the area of the lowest standard. Other similar ratios are also provided in Table 2 for comparison. Furthermore, the back-calculated concentrations, [ClO4−]calc, allow the contrast to be seen straightforwardly. In this fashion, performance near the detection limit is concluded to be more-or-less the same for both column lengths.
As explained above, performance exceeding the artefactual detection limits induced by the calibration curves is attainable. For example, if a 3.00 ng mL−1 standard is injected eight times, a method detection limit (MDL) may be calculated as described.13 At the 99% confidence level, it is determined as follows: MDL = t0.01,7 × σn − 1 × [3.00 ng mL−1]/Aav, where t0.01,7 = 2.998 (the value of Student's t for seven degrees of freedom at the 99% confidence level), σn − 1 = the standard deviation of the eight replicate peak areas, and Aav = the average of the eight replicate peak areas. For the 100 mm column, the MDL was 0.48 ng mL−1 (Aav = 106.8, σn − 1 = 5.8 ng mL−1, RSD = 5.4%); for the 150 mm column, the MDL was 0.35 ng mL−1 (Aav = 149.3, σn − 1 = 5.9 ng mL−1, RSD = 4.0%). If the MDL is used as the criterion of performance, the 150 mm column outperformed the 100 mm column just slightly since the 100 mm column has an MDL 38% larger than that of the 150 mm column. Ten consecutive injections of a 5.00 ng mL−1 standard gave an MDL of 0.16 ng mL−1; when repeated the next day, an MDL of 0.18 ng mL−1 was obtained (150 mm column; t0.01,9 = 1.383).
Samplea | Grav concn/µg g−1b | 100 mm PVA column concn/µg g−1c | 150 mm PVA column concn/µg g−1c | Relative difference in concn for 100 mm vs. 150 mm columnd (%) | EPA study average concn/µg g−1c,e | Relative difference in concn for EPA study vs. 150 mmf (%) |
---|---|---|---|---|---|---|
a The letter notation (a or b) is used to delineate duplicate solid samples. These were not identified to the laboratory as the same exact material. Distinct test solutions were prepared and injected into the IC.b The gravimetric value is calculated from the known formulation; known masses of a perchlorate salt and a fertilizer (which essentially serves as a diluent) were combined in the EPA laboratory as reported in ref. 23.c Reported uncertainty is the estimated standard deviation for three replicate injections of the same test solution.d Relative difference = (C150mm − C100mm)/C150mm = 100%.e Values were taken from Table 4 in ref. 23. A duplicate of sample 55 was not included in the previous EPA study.f Relative difference is computed for the concentrations obtained using the 150 mm Metrosep A Supp 5 column versus the average from the EPA study; relative difference = (CEPA − C150mm)/CEPA × 100%. | ||||||
33a | — | 4209 ± 2 | 4113 ± 4 | −2.33% | 3970 ± 190 | −3.6 |
33b | — | 4130 ± 7 | 4116 ± 13 | −0.34% | 3980 ± 230 | −3.5 |
35 | — | 2288 ± 19 | 2277 ± 11 | −0.48% | 2330 ± 200 | +2.1 |
36 | — | 1946 ± 15 | 1937 ± 6 | −0.46% | 1920 ± 100 | −1.0 |
42 | — | 2408 ± 5 | 2438 ± 4 | 1.23% | 2480 ± 250 | +1.5 |
49 | 6800 | 5802 ± 15 | 6138 ± 12 | 5.47% | 6080 ± 250 | −0.96 |
50 | 6200 | 5575 ± 15 | 5881 ± 12 | 5.20% | 5440 ± 300 | −8.2 |
51 | 2700 | 2119 ± 61 | 2388 ± 25 | 11.26% | 2400 ± 240 | +0.51 |
52 | 1800 | 1671 ± 4 | 1600 ± 5 | −4.44% | 1620 ± 170 | +1.0 |
55a | 1700 | 1676 ± 1 | 1587 ± 4 | −5.61% | 1590 ± 100 | +0.19 |
55b | 1700 | 1712 ± 3 | 1657 ± 1 | −3.32% | — | — |
Interrun (day-to-day) variation on the 150 mm column was larger than intrarun variation. The average peak area based on 10 injections fell by 5.6% for a 5.00 ng mL−1 standard and 7.6% for a 10.0 ng mL−1 standard from one day to the next. Similarly, the average peak area based on 10 injections of the solution derived from sample 33 fell by 4.7% from one day to the next. Of course, this level of variation is within the ±20% limit imposed by EPA/600/R-01/026, but can be eliminated by recalibration.
There was a statistically significant difference in the concentrations measured using the 100 mm column as opposed to the 150 mm column. No clear bias was observable when comparing the data obtained using the different columns; sometimes the concentrations were higher on one column and sometimes on the other. The relative differences between the concentrations measured on the two columns (value from 150 mm column as compared with that from 100 mm) varied from a minimum (in magnitude) of −0.34% (sample 33b) to a maximum of 11.3% (sample 51). In comparing the relative differences on the blind duplicates, there is satisfactory precision. For sample 33, it was −2.3% for sample 33a versus −0.34% for sample 33b. For sample 55, it was −5.6% for sample 55a versus −3.3% for sample 55b. Overall, this suggests that performance is consistent.
![]() | ||
Fig. 1 Ion chromatograms of Chilean sodium nitrate fertilizer (sample 55a) prepared at 2.0 mg mL−1 (1/100 dilution of 10 g dL−1) on Metrosep A Supp 5 columns; (a and b) on 100 mm column using 500 µL loop: (a) unspiked, showing native perchlorate peak, (b) the same solution as in (a) plus a 20% spike; (c and d) on 150 mm column using 1000 µL loop: (c) unspiked, showing native perchlorate peak, (d) the same solution as in (c) plus a 20% spike. |
![]() | ||
Fig. 2 Ion chromatograms of 22-3-14 lawn fertilizer (sample 1) prepared at 5.0 mg mL−1 (1/20 dilution of of 10 g dL−1) on a 150 mm Metrosep A Supp 5 column: (a) no perchlorate peak is discernible in the original solution, (b) solution in (a) fortified with 10 ng mL−1 of the analyte. |
100 mm column (%) | 150 mm column (%) | |||
---|---|---|---|---|
Sample | 20% spike | 50% spike | 20% spike | 50% spike |
a The term 20% spike refers to a fortification equal to 20% of the analyte concentration found in the injected solution so that the post-fortification analyte concentration is 1.2 times the original concentration measured in the solution. Likewise, the term 50% spike refers to a fortification equal to 50% of the analyte concentration found in an injected solution so that the post-fortification analyte concentration is 1.5 times the original concentration measured in the solution. | ||||
33a | 102.6 | 105.2 | 101.8 | 102.8 |
33b | 102.5 | 104.0 | 102.3 | 103.3 |
35 | 100.7 | 103.0 | 101.5 | 100.8 |
36 | 100.9 | 102.6 | 101.8 | 101.8 |
42 | 100.7 | 103.5 | 100.7 | 102.0 |
49 | 102.5 | 105.1 | 100.8 | 103.4 |
50 | 101.6 | 100.8 | 97.9 | 102.1 |
51 | 97.9 | 100.1 | 101.7 | 104.4 |
52 | 101.5 | 101.9 | 100.3 | 99.3 |
55a | 101.1 | 102.9 | 98.6 | 102.1 |
55b | 102.7 | 105.3 | 100.7 | 102.9 |
Sample | Dilution factora | Recovery (%) | pARL/µg g−1 | Sample | Dilution factor | Recovery (%) | pARL/µg g−1 |
---|---|---|---|---|---|---|---|
a Dilution factor refers to the subsequent volumetric dilution of 1 part of a 10 g dL−1 leachate or solution to the number of parts listed in the table prior to injection on the IC. | |||||||
1 | 20 | 100.4 | 2 | 26 | 20 | 98.9 | 2 |
2 | 500 | 103.7 | 50 | 27 | 10 | 101.4 | 1 |
3 | 20 | 96.1 | 2 | 28 | 50 | 99.0 | 5 |
4 | 50 | 94.5 | 5 | 29 | 10 | 95.2 | 1 |
5 | 100 | 103.7 | 10 | 30 | 50 | 99.3 | 5 |
6 | 5 | 93.9 | 0.5 | 31a | 100 | 94.3 | 10 |
7 | 20 | 99.7 | 2 | 31b | 450 | 100.4 | 45 |
8 | 50 | 95.2 | 5 | 32 | 470 | 92.6 | 47 |
9 | 470 | 103.3 | 47 | 34a | 50 | 96.7 | 5 |
10 | 470 | 103.8 | 47 | 34b | 50 | 96.9 | 5 |
11 | 970 | 94.1 | 97 | 37 | 50 | 99.5 | 5 |
12 | 20 | 103.1 | 2 | 38 | 1000 | 94.5 | 100 |
13 | 20 | 91.8 | 2 | 39 | 450 | 96.5 | 45 |
14 | 20 | 98.0 | 2 | 40 | 100 | 97.1 | 10 |
15 | 10 | 98.5 | 1 | 41 | 5 | 97.3 | 0.5 |
16 | 50 | 96.0 | 5 | 43 | 450 | 101.0 | 45 |
17 | 10 | 98.1 | 1 | 44 | 10 | 107.8 | 1 |
18 | 100 | 95.6 | 10 | 45 | 20 | 100.9 | 2 |
19 | 20 | 93.8 | 2 | 46 | 50 | 99.7 | 5 |
20 | 20 | 97.5 | 2 | 47 | 420 | 95.9 | 42 |
21 | 2250 | 95.3 | 225 | 48a | 10000 | 84.6 | 1000 |
22 | 50 | 95.8 | 5 | 48b | 1000 | 93.4 | 100 |
23 | 450 | 101.8 | 45 | 53 | 50 | 97.0 | 5 |
24 | 100 | 102.7 | 10 | 54 | 50 | 103.7 | 5 |
25 | 10 | 98.0 | 1 |
Footnote |
† This is the work of a US Government employee engaged in his official duties. As such it is in the public domain and exempt from copyright. ©US Government. |
This journal is © The Royal Society of Chemistry 2002 |