Investigation of Conventional Membrane and Tangential Flow

Nov 16, 2004 - MATTHEW A. MORRISON* AND. GABOURY BENOIT. Yale School of the Environment, Yale University,. New Haven, Connecticut 06520...
0 downloads 0 Views 347KB Size
Environ. Sci. Technol. 2004, 38, 6817-6823

Investigation of Conventional Membrane and Tangential Flow Ultrafiltration Artifacts and Their Application to the Characterization of Freshwater Colloids MATTHEW A. MORRISON* AND GABOURY BENOIT Yale School of the Environment, Yale University, New Haven, Connecticut 06520

Artifacts associated with the fractionation of colloids in a freshwater sample were investigated for conventional membrane filtration (0.45 µm cutoff), and two tangential flow ultrafiltration cartridges (0.1 µm cutoff and 3000 MW cutoff). Membrane clogging during conventional filtration removed some colloids smaller than 0.1 µm in diameter, much smaller than the nominal size of the filter. For certain constituents (e.g., Fe), filter clogging had a significant effect on filtrate concentrations, while artifacts associated with tangential flow ultrafiltration using a 0.1 µm cutoff were minimal. Artifacts occurred during tangential flow ultrafiltration with a 3000 MW cutoff, but did not deviate from predicted changes in concentration based on a standard permeation model. Comparison of filtrate concentrations for membrane filtration (at 1.0 and 0.45 µm) and tangential flow ultrafiltration (at 0.1 µm) for a large number of samples from Connecticut rivers shows that significant and consistent differences exist between their separation characteristics. Results for organic carbon, Fe, Mn, Al, Cu, and Pb demonstrate the magnitude of the effects of the fractionation technique on filtrate element concentration, show variability by element and flow condition, and highlight the importance of larger colloids to freshwater metal speciation. One implication of the research is that tangential flow ultrafiltration with large size cutoff membranes (e.g., ∼0.1 µm) may be superior to conventional filtration with filters in the same size discrimination range, and potentially more appropriate for the fractionation of natural water samples.

Introduction Current research demonstrates the importance of colloidal constituents to the speciation of trace metals and organic carbon in natural waters (1-5). No direct comparisons exist, however, between the widely used membrane filtration and tangential flow filtration techniques employed to separate colloids from suspended sediments in natural waters. Conventional membrane filtration (typically at 0.4 or 0.45 µm pore size) is known to be subject to significant artifacts (see, e.g., refs 6-8), and does not separate particles at the * Corresponding author present address: National Risk Management Research Laboratory, U.S. Environmental Protection Agency, Cincinnati, OH 45268; phone: (513)569-7441; fax: (513)569-7105; e-mail: [email protected]. 10.1021/es049710l CCC: $27.50 Published on Web 11/16/2004

 2004 American Chemical Society

commonly agreed upon cutoff between colloids and suspended sediments of 1.0 µm (9). During membrane clogging, increasingly smaller colloids are retained, but no studies have quantified the extent of pore size reduction due to membrane clogging. Tangential flow ultrafiltration (TFU) is the most widely used technique for the fractionation of colloids (see, e.g., refs 3, 10, and 11-15). The main advantage of TFU is a potential reduction of filtration artifacts in comparison to those of membrane filtration. However, many fundamental questions remain unanswered concerning the separation of colloids by TFU, such as the exact size discrimination characteristics for a given filter type and the artifacts associated with routine application of TFU. Much of the research on TFU artifacts focuses on membrane integrity, and utilizes model compounds (e.g., various MW dextrans), rather than natural water samples, to test that integrity and to evaluate the retention characteristics of specific TFU cartridges (16-18). The study of breakthrough and retention for model compounds is reviewed and expanded by Dai et al. (19) and Guo et al. (20), and will not be discussed here in detail. The research presented in this paper demonstrates the use of a combined membrane filtration (0.45 µm Durapore) and TFU (0.1 µm and 3000 MW hollow fiber cartridges) artifact study for the investigation of colloidal characteristics in a riverine sample. Membrane filtration artifacts for 0.45 µm membrane filters in combination with 0.1 µm and 3000 MW TFU cartridge results show that, as membrane filters clog, some colloids smaller than 0.1 µm can become trapped, but other colloids and dissolved species may pass freely through clogged membrane filters. A larger study investigated the existence of systematic differences between filtrate concentrations for three separation techniques with similar size cutoffs. This study compares filtrate concentrations from 1.0 µm polycarbonate (Nuclepore) and 0.45 µm mixed cellulose (Durapore) membrane filters, and 0.1 µm TFU cartridge (Spectrum) separation. The results demonstrate that systematic relationships exist between the filtrate element concentrations for a large number of sites over a range of flow conditions.

Experimental Section Sampling Location. The Hammonasset River (sampled at 41°19′39′′ N, 72°36′43′′ W on May 17, 2001) was chosen for this study because of recent work (7, 21) which demonstrates the importance of NOM colloids to filtration artifacts and to the speciation of trace and heavy metals at this site. Those references describe the characteristics of the Hammonasset River in more detail. The concentration of organic carbon (OC) when sampled for this study was 3.23 ( 0.02 mg/L, the pH was 6.7, the water temperature was 12.3 °C, and the specific conductance was 72.8 µS/cm. Filtration Strategy. Larger colloids (ca. 0.1-1.0 µm) were chosen for detailed evaluation because of evidence suggesting that conventional methods (i.e., 0.4 or 0.45 µm membrane filtration) were not adequate for accurate and meaningful separation of particles in this size range. For the comparison study we chose 1.0 µm Nuclepore polycarbonate and 0.45 µm Millipore Durapore membranes, and a Spectrum 0.1 µm tangential flow ultrafiltration cartridge. The reasoning is as follows: (a) Filtration at 1.0 µm separates colloids from suspended sediments at the most commonly agreed upon upper particle diameter cutoff for colloids in freshwater systems. (b) The 0.45 µm pore size cutoff is, perhaps, the most commonly used one in environmental research and VOL. 38, NO. 24, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

6817

regulation. (c) We selected a lower cutoff of 0.1 µm because there is evidence that a particle size slightly smaller than the standard 0.45 µm (Millipore) or 0.4 µm (Nuclepore) is important to the speciation of elements such as Fe and Al, and should not be included in the dissolved or colloidal organic matter size fraction. We selected TFU (rather than conventional filtration) for this size cutoff because standard 0.1 µm membrane filters clog almost instantaneously (7). The TFU cutoff chosen to separate organic colloids from the dissolved phase of 3000 MW is based on previous research in our laboratory, and is only slightly larger than the 1000 MW cutoff commonly used by marine researchers. Filtration Techniques. All samples were collected using clean techniques (10), and kept on ice for transport and storage prior to laboratory processing. Field filtrates were collected in precleaned LDPE bottles, and whole water samples for laboratory filtration were collected in PFA bottles. All laboratory filtrations (0.45 µm membrane artifact filtration, 0.1 µm and 3000 MW tangential flow ultrafiltrations) were completed within 12 h of collection. Laboratory filtrations were carried out in a class 100 clean room, using methods (and subsequent analytical techniques) described previously (7). In addition to previously cited detection limits, the detection limit for organic carbon (as nonpurgeable organic carbon via Shimadzu TOC 5000A) was 0.15 mg of C/L on the basis of the standard deviation of method blanks. Method blanks for the filtration techniques were not significantly different from Barnstead Nanopure water blanks for all elemental analyses, and were run routinely as part of the QA/QC protocols. All membrane filters used in this study were of 47 mm diameter. For the evaluation of conventional membrane filtration artifacts (0.45 µm Durapore), only backpressure change during filtration was used to indicate membrane clogging, and only five filtrate fractions, with no replication, were collected. Filtrate fraction volumes were approximately 200 mL for a 1.0 L natural water sample, and corresponded to 25% increases in measured back-pressure, with the final fraction collected at a maximum back-pressure of 16.5 psig. For routine application of both membrane and tangential flow techniques, samples were typically prefiltered through 20 µm nylon membrane filters (Spectrum, 47 mm) to remove larger suspended sediment particles. Two TFU cartridge filters were employed: (i) a 0.10 µm MiniKros hollow fiber cartridge (mixed cellulose ester; Spectrum, M11M 260 01N) and (ii) a 3000 MW Amicon hollow fiber cartridge (polysulfone; H1P3-20 (Millipore, production discontinued)). Both cartridges had the same active surface area (600 cm2). A transmembrane pressure of 5 psig was used for the Amicon 3000 MW ultrafilter to increase the permeate flow rate, but this was not necessary for the MiniKros 0.10 µm cartridge. A Masterflex peristaltic pump (peroxide-cured silicone tubing in the pump head) and PTFE tubing (outside the pump head) were used for the recirculation of the cleaning and sample solution and for the collection of permeate. New TFU cartridges were cleaned initially by recirculation, with permeation, of 0.1% Micro detergent (0.5 L, 30 min) followed by no less than 10 L of Nanopure water. Prior to each filtration, the TFU cartridge was cleaned according to the following procedure: (i) recirculation, with permeation, of 0.5 L of NaOH (0.01 M, Alfa Aesar, 99.996% metals basis) for 30 min; (ii) straight-through flushing of 1.0 L of Nanopure water through the interior of the cartridge (no recirculation); (iii) filtration of 1.0 L of Nanopure water, with recirculation, through the membrane. The pH was checked at the end of this step, and this step was repeated if the pH values of the retentate and permeate were greater than 7.0. Steps 1-3 were then repeated with the substitution of 0.1 M HCl (Fisher, trace metal grade) as the cleaning solution. The pH at the end of the acid cleaning step was around 5.0, and cartridges 6818

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 38, NO. 24, 2004

TABLE 1. Recovery Data for 0.1 µm TFUa percent of total recovery

permeate retentate acid rinse total

OC

Fe

Al

Mn

Cu

98 3 2

58 11 2

86 10 8

77 5 3

91 1 16

103

71

104

85

108

a

Permeate recoveries for base cations were (Na) 110%, (Mg) 99%, (K) 101%, and (Ca) 99%.

TABLE 2. Recovery Data for 3000 MW TFUa percent of total recovery

OC

Fe

Al

Mn

Cu

permeate retentate acid rinse

32 54 9

0 84 10

14 52 9

44 30 18

47 50 29

total

95

94

75

92

126

a

Permeate and retentate recoveries for base cations were (Na) 104% (7% retentate), (Mg) 81% (12%), (K) 94% (6%), and (Ca) 81% (13%).

were stored in this condition (with the cartridge filled with Nanopure water) between filtrations. For long-term storage, when the cartridges were not to be used for more than 7-14 days, they were filled with water and stored in clean Ziploc bags in the refrigerator (4 °C, dark). The cartridges were preconditioned with sample or standard solution (125 mL) prior to the filtration of standard solutions and natural water samples. Preconditioning ensures that the permeate is not diluted by entrained water, and may also fill surface adsorption sites (19) and exchange acidic protons (from the HCl cleaning solution) on the hollow fiber surface (13). During filtration, permeate fractions were collected in LDPE bottles (60 mL, acid-cleaned). Only the final retentate fraction was collected for analysis. Following sample filtration, the cartridge was drained and a solution of 0.01 M HNO3 (Seastar) was recirculated, with permeation, to recover adsorbed constituents from the hollow fiber surfaces. This treatment allows for a semiquantitative mass balance for each sample, but does not remove irreversibly adsorbed, base-soluble, or physically trapped constituents from the cartridge.

Results and Discussion Recovery Data. TFU recovery data are summarized in Tables 1 and 2; recovery data were calculated only for the TFU methods, because a retentate is not collected during conventional dead-end filtration. For the TFU methods, differences in recovery data should be governed by differences in membrane materials and pore size (ca. 1.5 nm for the 3000 MW cartridge vs 100 nm), since the active surface area is the same for each cartridge. Research suggests that polysulfone (3000 MW) adsorbs OC more strongly than mixed cellulose ester (0.1 µm) (19), but this effect has not been tested for the specific membranes used in this study or for natural freshwater samples. As shown by the recovery data in Tables 1 and 2, adsorption of OC during TFU does not appear to be significant for Hammonasset River water. Recoveries of OC (permeate + retentate) for the two TFU cartridges were 101% and 86% for the 0.1 µm and 3000 MW cartridges, respectively. The lower recovery for the 3000 MW cartridge may be due to stronger adsorption, but then one would expect low initial permeate concentrations of OC, which were not observed (see Figure 1). Recovery data for the trace metals Fe, Mn, and Al provide further clues as to the cause of incomplete recoveries. The

FIGURE 1. Changes in OC, Fe, Mn, Al, and Cu concentrations during conventional membrane filtration (0.45 µm) and TFU using 0.1 µm and 3000 MW cutoff hollow fiber cartridges. The figure also shows the model fit for 3000 MW permeate data. Data shown are for Hammonasset River water, collected May 17, 2001. pores of hollow fiber membranes are not perfect cylinders, but have been shown by the manufacturers to contain pores with a range of diameters, twists and turns, and dead-end paths. If chemical adsorption is minimal, then it is likely that unrecovered elements are present in the source water as colloids close in size to the rated pore size of the membrane. Incomplete recovery of Fe and Mn during 0.1 µm TFU (69% and 82%, respectively) may be due to the trapping of colloids, and the differences observed in the recovery of Al (96% and 66% for 0.1 µm and 3000 MW, respectively) indicate differences in the chemical and physical speciation of these elements. Incomplete recoveries may also be influenced by the retentate to permeate flow ratio, which was not measured as part of the research protocols (18). Comparison of Filtration Artifacts for Hammonasset River Water. Conventional membrane and tangential flow ultrafiltration processes are subject to many of the same artifacts. The artifacts associated with conventional membrane filtration increase with loading, and can be indicated by changes in flow rate and back-pressure (6, 7). The artifacts associated with TFU also increase with colloid loading in the retentate, a property quantified by the concentration factor (CF), which is a measure of the completeness of the filtration. Changes in back-pressure and concentration factor during filtration are presented in graphical and tabular form in the Supporting Information. It should be noted that other studies have used much higher CF values for TFU separations (see, e.g., ref 20), but this difference is likely to be less significant to the broader study of filtration artifacts than site-specific differences in source water characteristics (e.g., coarse and

fine suspended sediment, colloidal and dissolved organic matter concentration). The three filtration artifact data setss0.45 µm membrane filtration, and 0.1 µm and 3000 MW TFUsfor the Hammonasset River are presented together in Figures 1 and 2. Figure 1 shows the change in filtrate/permeate concentration vs cumulative filtrate/permeate volume for OC, Fe, Mn, Al, and Cu concentrations in Hammonasset River water. The total (unfiltered) concentration for each element is shown on each graph as a dashed horizontal line. For OC and Cu the total, 0.45 µm filtrate and 0.1 µm TFU permeate concentrations are the same, and do not change significantly during filtration; this is also true for Na, Mg, K, and Ca (Figure 2; total concentration lines and conventional filtration data were omitted from Figure 2 for clarity). Conventional filtration artifacts for other elements are minimal, but telling. As observed previously (7), the filtrate concentrations of Fe, Mn, and Al decrease with increased membrane loading. The concentration of Al in the 0.45 µm filtrate appears to stabilize for the last two fractions and does not fall below the average concentration of the 0.1 µm TFU permeate. The concentrations of Fe and Mn, however, continue to decrease below the concentration of these elements in the 0.1 µm TFU permeate. The filtration artifact data in Figures 1 and 2 support the following two observations: (i) The fact that Fe and Mn concentrations continue to decrease below the level of the 0.1 µm TFU permeate during membrane clogging-while the concentration of Al does not-suggests that Fe and Mn colloids approximately 0.1 µm in diameter are present in significant quantity. The presence of colloidal Fe and Mn in VOL. 38, NO. 24, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

6819

FIGURE 2. Changes in Na, Mg, K, and Ca concentrations during TFU using 0.1 µm and 3000 MW cutoff hollow fiber cartridges. The figure shows the model fit for 3000 MW permeate data. Data shown are for Hammonasset River water, collected May 17, 2001. the upper range of the 0.1 µm to 3000 MW colloidal fraction cannot be shown definitively, because the size range between these two ultrafilters is too large, but the conclusion is supported by existing research (2, 22-24). (ii) The data confirm the supposition from previous research (7) that some colloidal fractions are unaffected by membrane clogging during conventional filtration. Specifically, these are substances that exist mainly as very fine colloids and truly dissolved forms (e.g., OC, Cu, and base cations). Differences in the behavior of various elements during membrane clogging underscore the potential need for dissimilar sampling strategies for different substances. Ultrafiltration Data. The model results for the permeate concentration (Cp; line only in Figures 1 and 2) were calculated using a standard permeation model equation (15, 19, 20). Data and calculations for the 0.1 µm and 3000 MW TFU studies for the Hammonasset River are presented in full as Supporting Information. None of the constituents analyzed exhibited a significant change in the 0.1 µm TFU permeate concentration with increasing CF, despite (i) the high CF achieved during ultrafiltration (21.5) with the 0.1 µm pore size cartridge, (ii) incomplete recoveries (permeable + retained species) for Fe and Mn (69% and 82%, respectively), and (iii) substantial increases in the retentate concentrations of OC, Fe, Al, and Mn (1.6, 3.0, 3.0, and 1.7 times, respectively, compared to the initial feed concentration). Application of the permeation equation to the 0.1 µm TFU data in Figures 2 and 3 provides permeation coefficients between 0.94 and 0.98 for all elements except Cu (not applicable due to data scatter). For the most part, the model fits (not shown) appear to be valid, but may not be particularly useful since the permeate concentration changes are largely insignificant. The lack of artifacts for the 6820

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 38, NO. 24, 2004

0.1 µm TFU permeate suggests that tangential flow ultrafiltration techniques with large pore sizes (0.1-1.0 µm cutoff) may be the appropriate solution to the long-standing problem with membrane filtration artifacts. TFU systems can be scaled to filter large volumes in short periods of time, especially for larger pore sizes, and can be reused if proper cleaning protocols are followed. TFU artifacts for the 3000 MW cartridge, as shown in Figure 1, are significant for OC, Al, and Cu. Calculated permeation coefficients (PC) are 0.40 for Cu, 0.44 for OC, and 0.54 for Al. PC values in this range indicate significant interaction with the cartridge pores during permeation, and imply that a significant portion of the permeate fraction for these elements is macromolecular. Both OC and Cu are expected to be present as, or complexed to, macromolecular natural organic matter (NOM), but the result for Al demonstrates that it may also be present complexed to macromolecular NOM. Fulvic acid (MW ≈ 1000), which is known to form strong complexes with Cu (25) and Al (26, 27), is present in the NOM of the Hammonasset River. Mn data show no significant change in the 3000 MW permeate during the filtration, and Mn has a calculated Pc of 0.90. Fe was below the method detection limits for all permeate fractions, and was recovered nearly quantitatively in the retentate solution (84%). Similar results for Fe and Mn have been observed by other researchers (2) TFU artifacts for Ca, Mg, K, and Na in the 3000 MW permeate are shown in Figure 2. Conventional 0.45 µm filtration data are omitted for clarity, and because there was no change in the concentration of any of the four base cations during membrane filtration. Systematic changes in permeation concentration during tangential flow ultrafiltration for these elements have also been observed by Guo et al. (28). The results show that Ca and Mg are retained to a greater

FIGURE 3. Comparison of 0.45 µm Durapore membrane filtration data with 1.0 µm Nuclepore membrane filtration and 0.1 µm tangential flow filtration data for OC, Cu, and Mn in a series of Connecticut rivers. Slopes are shown for regression lines (with 95% confidence intervals). extent by the 3000 MW cartridge than K and Na, and that the differences are divided according to charge. The permeation coefficients for Ca and Mg are both 0.73, while that for K is 0.80 and that for Na is 0.90 (note however the significant scatter in the Na data points, which is due to analytical problems during analysis). The two possible reasons for the observed artifacts are (i) the association of base cations, by complexation or ionic interactions, with NOM colloids and (ii) interaction of aqueous cations with the membrane pores, enhanced by the negative surface charge of polysulfone (28). The argument for charge-based interactions between major ions and the tangential flow ultrafiltration membrane is made by Guo et al. (28), who observed permeation coefficients of 0.86 and 0.87 for Ca and Mg during TFU of Trinity River water using a 1000 MW cartridge. The similarity in the permeation behavior of singly and doubly charged species supports the argument that membrane surface charge rather than complexation might be the dominant cause of the observed artifacts. Complexation of Ca by fulvic acid has, however, been studied (25), and complexation of Mg by NOM, although less studied, may be stronger (according to the Irving-Williams series of complexing strength based on a divalent metal’s ionic radius; 26); both Ca and Mg would interact more strongly with colloidal NOM than Na and K. These two mechanisms cannot be distinguished on the basis of the available data. Comparison of Methods and Cutoffs for Larger Colloids. One of the critical questions concerning filtration of natural waters is the degree to which a choice of technique or size cutoff will affect filtrate concentrations for different elements. This question was investigated by comparing conventional 0.45 µm membrane filtrate concentrations with 1.0 µm Nuclepore membrane and 0.1 µm TFU cartridge filtrate concentrations for a number of Connecticut rivers and streams, under both baseflow and stormflow conditions. A Nuclepore membrane with 1.0 µm cutoff was chosen because 1 µm is the accepted upper size range cutoff for colloids, and

because research has shown that Nuclepore membranes provide an accurate size separation (see, e.g., ref 29) for particles. The use of 0.1 µm TFU cartridge filtration provides a size cutoff just below the commonly employed membrane filter size cutoffs of 0.45, 0.4, and 0.2 µm, and has the advantage of removing biological material and fine clays that may pass conventional 0.45 µm membranes. The results are shown in Figures 3 (OC, Cu, Mn) and 4 (Fe, Al, Pb). The graphs in Figures 3 and 4 show regression lines for each comparison (95% confidence intervals shown by dashed lines) that demonstrate deviation from a 1:1 relationship (i.e., no difference between filtrate element concentrations). The slope, intercept, and r2 values for all regressions are included in the Supporting Information. For comparisons of the 0.45 and 1.0 µm membrane filters, OC, Cu, and Mn filtrate concentrations (slope 0.98, 1.07, and 1.03, respectively) show little significant difference and little scatter in the data. The remaining elements, Fe, Al, and Pb (slope 1.07, 1.6, and 1.9 respectively), show significant differences in filtrate concentration. Although the slope of the regression line for Fe is close to 1.0, the intercept is 30, and there is significant scatter in the data. The relationship between Fe and Al concentrations in 0.45 and 1.0 µm filtrates is consistent with previous observations (7), but while the 1.0 µm filtrate concentrations are consistently higher, the results are quite variable (r2 ) 0.66 and 0.36 for Fe and Al regressions, respectively). The difference between the 0.45 and 1.0 µm filtrates appears to be due to differences in the colloidal and suspended sediment characteristics of the source water, as evidenced by the variable behavior of Fe and Al, present in freshwater systems as oxides and clay particles. The regression slopes for the relationship between the 0.45 µm membrane filtrate and the 0.1 µm tangential flow filtrate show significant differences for all elements. Organic carbon and Mn show the least significant differences in filtrate concentration (slope 0.76 and 0.74, respectively), and, along with Cu, the least scatter in the data (r2 ) 0.88 or 0.89). The VOL. 38, NO. 24, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

6821

FIGURE 4. Comparison of 0.45 µm Durapore membrane filtration data with 1.0 µm Nuclepore membrane filtration and 0.1 µm tangential flow filtration data for Fe, Al, and Pb in a series of Connecticut rivers. Slopes are shown for regression lines (with 95% confidence intervals). slope of the Cu regression (0.35) is very close to the slope of the Fe regression (0.36), but the scatter of the Fe and Al data (r2 values of 0.36 and 0.16, respectively) belie a simple relationship between the elements. Cu may be present as a trace constituent on the surface of iron (hydr)oxides, but may preferentially bind to smaller particles and/or may bind with organic matter coatings on particle surfaces. The Pb data do not follow the same patterns as the Cu data, and were included only to illustrate this point; there are not enough data for Pb to draw firm conclusions concerning filtrate concentration differences. The filtration artifact study and the data presented in Figures 3 and 4 demonstrate the importance of colloids within the size range between 0.1 and 1.0 µm to the speciation of trace metals and organic carbon, and show that, as expected, iron and aluminum (hydr)oxide colloids-and associated heavy metals-exhibit the greatest variance when filtration methods are compared. The surprising result is that OC, typically present in natural freshwaters as macromolecular colloids, and Mn, mainly thought to be present in natural freshwaters as manganese oxide solids, do not show greater artifacts when fractionated by the methods described in this research. These results emphasize the need for researchers studying trace metal speciation to choose their method of separation carefully, on the basis of the element to be studied and the goals of the research.

Supporting Information Available Full data for 0.45 µm membrane and 0.1 µm and 3000 MW tangential flow ultrafiltrations of Hammonasset River water (Tables S1-S3, respectively), including calculated values for average and modeled permeate concentrations, Figure S1 depicting the change in back-pressure for membrane filtration and the change in concentration factor (CF) during tangential flow ultrafiltrations, and Tables S4 and S5 containing slope, intercept, and r2 data for the regressions 6822

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 38, NO. 24, 2004

depicted in Figures 3 and 4. This material is available free of charge via the Internet at http://pubs.acs.org.

Literature Cited (1) Rostad, C. E.; Leenheer, J. A.; Daniel, S. R. Organic carbon and nitrogen content associated with colloids and suspended particulates from the Mississippi River and some of its tributaries. Environ. Sci. Technol. 1997, 31, 3218-3225. (2) Pham, M. K.; Garnier, J.-M. Distribution of trace elements associated with dissolved compounds (