Filtration Artifacts Caused by Overloading Membrane Filters

Environmental Protection Agency's STAR program through grant R82-5395, it ... (8) Benoit, G.; Oktay-Marshall, S. D.; Cantu, A., II; Hood, E. M.;. Cole...
0 downloads 0 Views 176KB Size
Download PDF
Environ. Sci. Technol. 2001, 35, 3774-3779

Filtration Artifacts Caused by Overloading Membrane Filters MATTHEW A. MORRISON AND GABOURY BENOIT* Yale School of Forestry and Environmental Studies, 370 Prospect Street, New Haven, Connecticut 06511

The conventional practice of using 0.45 or 0.40 µm membranes to distinguish between the particulate and dissolved phases in natural waters neglects the importance of colloids. Many of the colloids in natural waters pass through 0.45 or 0.40 µm membranes, but a significant fraction at the upper end of the colloidal particle size range is retained. Membrane clogging during filtration decreases the effective pore size and can cause the retention of increasing amounts of colloids. This filtration artifact can cause serious errors in sampling and in assigning trace metals to various particle size classes. We evaluated the effect of membrane loading for two common membrane types (0.45 µm Millipore Durapore and 0.40 µm Nuclepore) on the retention of colloidal Fe, Al, Mn, and OM in three Connecticut rivers. In addition, we used a 1.0 µm Nuclepore membrane to estimate the amount of colloids in the 0.40-1.0 µm size fraction that are retained by membranes during conventional filtration. All samples were collected with clean techniques, and all filtrations were carried out in a class 100 clean room. A peristaltic pump, set at an initial flow rate of 120 mL/min, was used to pump samples through 47 mm diameter inline Teflon filter holders. Back pressure and flow rate were monitored during filtration, and both are good indicators for the onset of membrane clogging. The results show a consistent correlation between increasing back pressure and decreasing concentration of colloidal Fe and sometimes Al, Mn, and OM in the filtrate for all membrane types. Although the shape of the loadingretention curves varied dramatically by site and by membrane type, the essential relationship between back pressure, flow rate, and filtration artifacts during membrane clogging remained the same.

Introduction Colloidal particles range from 1 nm to 1 µm in size and are increasingly recognized for their role in the transport of trace metals, nutrients, and organic contaminants in natural waters (1-5). The lower limit of the colloidal size range can be defined by microparticles that provide a physicochemical environment distinct from the bulk aqueous phase and into or onto which truly dissolved species can partition (3). The upper end of the colloidal size range is defined by particles whose movement begins to be significantly affected by gravitational settling and can extend under some circumstances to include particles up to 10 µm in diameter (3). Typical inorganic colloids are iron oxyhydroxides, manganese oxides, and clays. Typical organic colloids include polysaccharides, biological debris, and humic acids. Due to their complexity and their propensity to undergo rapid transformations once they are removed from natural systems, the 3774

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 35, NO. 18, 2001

physical and chemical properties of colloids are poorly understood and remain an area of active research (2, 3, 6-12). Many methods exist for isolating suspended particles from natural waters, but filtration is the most common because of its field-portability, simplicity, and low cost. Conventional filtration using a 0.45 or 0.40 µm membrane offers a few distinct advantages (13) which keep it in widespread use. First, it removes suspended sediment (i.e. silt, sand and large clay particles) and large natural organic matter that would otherwise settle out of water samples during transport and storage; settling of larger particles can alter the colloidal characteristics through differential sedimentation (1). Second, conventional filtration aids in the preservation of samples by removing some bacteria and other microorganisms that may cause biological transformations in the water sample during storage. Third, conventional filtration is the technique that is most widely used by researchers in aquatic chemistry and thus provides a standard for the intercomparison of results from different studies. However, conventional filtration has been widely criticized because (1) it is defined operationally and may not cause an unambiguous separation of particle size classes and (2) filtration artifacts call into question the comparability of results from different studies (14). Since the early 1970s, researchers have demonstrated that filtration can cause significant errors in the measurement of trace metals in filtered water samples (13-20). These artifacts, which include contamination, adsorption of solutes and colloids, membrane clogging, and concentration polarization, are thoroughly reviewed in the book Environmental Particles (13). This paper focuses on membrane clogging, which is accompanied by a decrease in the effective pore size of the filter membrane over the course of the filtration. Membrane clogging traps increasingly smaller particles as the filtration progresses and can reduce the concentration of some trace elements in the filtrate (notably Fe) to less than 5% of their original concentration. Some researchers attribute this artifact to the trapping of larger (e.g., 0.1-1.0 µm) inorganic colloids (15-17), while other researchers suggest that the entire range of colloidal particles may be affected by membrane clogging (21, 22). Two types of membranes are routinely used for conventional filtration and are typically assumed to yield comparable filtrate fractions. Research typically demonstrates that tortuous path (TP) or “depth” membranes made of mixed cellulose acetate and cellulose nitrate (e.g., Millipore and Sartorious membranes) retain a larger spectrum of particle sizes, including particles smaller than the rated pore size, while passing some that are larger (8, 12), but that these membranes do not clog as easily as sieve-type (S) membranes (e.g., Nuclepore). The advantages of TP membrane behavior are that a larger volume of water containing a constant concentration of trace elements may be filtered through these membranes and that larger colloids such as clays are excluded from the filtrate. Researchers typically recommend polycarbonate S membranes for size fractionation (8, 13, 18) because, although they retain some particles that are smaller than the rated pore size, they do not pass any that are larger. However, S membranes clog much more easily than TP membranes, which can make them impractical for studies that require large volumes of filtered water. This research compares the behavior of two common filtration membranes, 0.45 µm Millipore Durapore (TP) and 0.40 µm Nuclepore (S), during membrane clogging and correlates the changes in trace element concentrations in the filtrate to the easily measured parameters of flow rate 10.1021/es010670k CCC: $20.00

 2001 American Chemical Society Published on Web 08/16/2001

and back pressure. In addition, a 1.0 µm Nuclepore (S) membrane is used to estimate the amount of colloids in the 0.40-1.0 µm size class that contributes to and is subject to filtration artifacts. The results demonstrate that artifacts occur even in natural waters containing low concentrations of suspended particulate matter (SPM) and that these effects are variable depending on the physical and chemical properties of the water.

Experimental Section Sampling Sites. The Hammonasset River (sampled at 41°19′39"N, 72°36′43"W on 9/12/98) was chosen as the initial site for this study because of recent work which demonstrates the importance of colloids to the particle concentration effect in this and other Connecticut rivers (9, 23). The Hammonasset is characterized by a low level of watershed development, low concentrations of suspended particulate matter (SPM ) 4.1 mg/L when sampled, range of 0.6-4.5 for 24 bimonthly samples, (24)), and high concentrations of natural organic matter (449 ( 13 µM when sampled and 160-500 µM for 24 bimonthly samples (24)). The organic matter in the Hammonasset River is mostly composed of tannic and fulvic acids (25), which color the river a pale reddish orange during the summer and early fall months when these samples were collected. The Quinnipiac River (sampled at 41°25′58"N, 72°51′05"W on 9/18/98) has a highly industrialized watershed with significant point and nonpoint sources of pollution. The sampling site for this study is 12.7 km below the outfall of a sewage treatment plant (and downstream from a total of 3 such plants), which may explain the high concentration of OC (898 ( 89 µM when sampled, range of 230-1010 µM for 24 bimonthly samples (24)). The banks and flood plains of the Quinnipiac River are highly erodable and generate substantial amounts of SPM during storm events, but the baseflow SPM for this sample was 8.8 mg/L (range 0.8-18.2 for 24 bimonthly samples (24)). The final site, Wintergreen Brook (sampled at 41°20′03"N, 72°57′23"W on 10/2/98 and 10/9/98), is a small, undeveloped tributary to the West River in New Haven, CT. It was sampled twice, once for baseflow and once for stormflow, but only the stormflow data are presented in detail, since the results were similar on the two dates. SPM in this brook was 13.9 when sampled (range of 0.8-8.2 mg/L and OC 180 to 480 µM in 13 monthly baseflow samples (26)). Sampling and Analysis. All samples were collected using clean techniques (27, 28) and kept on ice for transport and storage prior to laboratory filtration. Whole water samples were collected in FEP bottles, and all laboratory filtrations were completed within 5 h of collection. Major metal analyses were performed on a Perkin-Elmer Optima 3000 ICP-AES without sample preconcentration, which provides a detection limit of 5-10 ppb for Al, Fe, and Mn and analytical precision for triplicate measurements of better than 5%. Trace metal analyses were conducted on a Perkin-Elmer model 3300 atomic absorption spectrometer (AAS) with HGA-600 graphite furnace following evaporative preconcentration (20 times) in PFA beakers. Detection limits are 0.70 µg/L for Zn (unconcentrated), 36 ng/L for Cu, 6 ng/L for Cd, and 24 ng/L for Pb based on the standard deviation of replicate measurements for the method blank; analytical precision is better than 10% for replicate measurements. A Shimadzu TOC 5000-C analyzer was used for determination of total and dissolved organic carbon (OC). Evaluation of Colloidal Stability over Time. Triplicate filtrations with 0.45 µm Durapore membranes were conducted at the Hammonasset River site. In addition, 6 L of unfiltered water were collected in FEP bottles, stored on ice, and returned to the lab. Immediately upon return (1 h after collection), and every hour for 8 h, triplicate 0.45 µm Durapore filtrates were collected for analysis of Fe concentrations.

Procedures for Laboratory Filtrations. Laboratory filtrations were carried out in a Class 100 clean room with a Masterflex peristaltic pump, PTFE tubing, and 47 mm PFA inline filter holders (Savillex). The Masterflex silicone tubing (peroxide-cured) that was used in the pump head was first cleaned by recirculating 10% trace metal grade HCl and then 1% trace metal grade HNO3. Whole water samples were stirred during filtration, and the flow was set to an initial rate of 120 mL/min. Back pressure was monitored via an Ashcroft glycerin-filled pressure gauge, which was mounted between the peristaltic pump and the filter. The pressure gauge was separated from the sample stream by a PFA gauge guard to avoid contamination. Flow rate was calculated from time measurements taken at the beginning and end of each aliquot during the filtration and thus represents an average value for each aliquot. All membrane filtrations were conducted in triplicate, but replicates for a specific membrane were not carried out in direct succession so that systematic errors were minimized. For each membrane the total volume of the filtrate aliquots that were collected depended on the total volume of filtrate that would pass through the membrane prior to the onset of membrane clogging. Typically 8-12 aliquots were collected for each membrane; filtrate aliquots were collected in 55 mL acid-cleaned centrifuge tubes (Falcon, polypropylene). TOC analysis was performed immediately following filtration, and aliquots for trace metal analysis were acidified to 2% HNO3 (trace metal grade) for preservation prior to analysis. An Amicon hollow fiber ultrafiltration cartridge (H1P3-20) with a 3000 MW cutoff was used to distinguish between the colloidal and truly dissolved fractions for all water samples (29).

Results and Discussion Sample Stability during Transport and Laboratory Storage. To determine if the transport and storage methods employed in this study would cause a change in the particle size distribution during the course of the experiments, we compared field filtration of Hammonasset River water to laboratory filtration (only for the 0.45 µm Durapore membranes). Triplicate 0.45 µm filtrates, ended prior to any observed decrease in filtrate flow rate, collected in the field were compared to triplicate filtrates taken at hourly intervals for up to 8 h after returning to the laboratory (Figure 1). The mean concentration for Fe in these filtrates (n ) 27) was 2.034 ( 0.070 mg/L. The data showed no decrease in the filtrate Fe concentration over time that would indicate aggregation of colloids in the samples. This initial test was supported by the excellent reproducibility (typically < 10% RSD, error bars Figure 1) between replicate filtrates during the artifact experiments. Hammonasset River. Figure 2 shows the changes in Fe, Mn, Al, and organic carbon (OC) concentrations and back pressure during membrane clogging for the three filter types. This site is different from the other two chosen for this study in that the Fe was mostly colloidal (the concentration of Fe in the 1.0 µm filtrate was 89% of the total Fe, while Fe in the ultrafiltrate was below the detection limit); the other two sites had a substantial amount of macroparticulate Fe. A significant fraction of the Mn (40%) was also colloidal, and, in contrast to the other sites, no Mn was found in the ultrafiltrate. All of the Al and OC at this site was less than 1.0 µm in size (72% of Al and 77% of OC were colloidal), but in contrast to Fe and Mn, 28% of Al and 23% of OC were present in the truly dissolved fraction. It is notable that both Fe and Al declined to approximately the concentration found in ultrafiltrate. In other words, the clogged filters acted approximately like ultrafilters for these elements. In contrast, in filtrate from overloaded membranes Mn and especially OC remained well above ultrafiltrate concentrations. These results suggest that the colloidal size distributions are VOL. 35, NO. 18, 2001 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3775

FIGURE 1. Variation of 0.45 µm filtrate Fe with holding time of sample for the Hammonasset River, CT. There was no variation, within experimental error, for a period up to 8 h, which was much longer than other filtration experiments in this study.

FIGURE 3. Filtration data for the Quinnipiac River, CT. Of the parameters measured, only Fe declined substantially, though other constituents showed steady but very small declines. Unreported Al data also showed large decreases, but results were too close to detection limits for statistical significance. Note that in the upper panel, Fe data have a different scale from the others.

FIGURE 2. Variation in concentrations of filtrate forms of several compounds as a function of volume filtered for the Hammonasset River, CT. Filter loading was monitored via back pressure and flow rate. On each concentration axis of the top panel (0.45 µm Durapore), arrows indicate total (unfiltered) and ultrafiltrate levels of constituents at this site. The difference between the upper arrow and the initial filtrate concentration is the amount of the macroparticulate form of the given constituent. The difference between initial filtrate and the lower arrow is the colloidal forms, and the difference between the lower arrow and baseline is truly dissolved forms. All figures are plotted to the same scale to facilitate comparison among filter types. At this site, all constituents eventually showed significant declines as a consequence of filter overloading. Clogging occurred in the sequence: 0.40 µm Nuclepore > 0.45 µm Durapore > 1.0 µm Nuclepore. dissimilar for different compounds and that only a subset of (perhaps larger) colloidal particles are affected by membrane clogging. Comparison of the 0.40 µm Nuclepore and 0.45 µm Durapore membranes for the Hammonasset shows behavior that is similar to other studies of membrane clogging (15, 17). The 0.40 µm Nuclepore membrane clogged rapidly, possibly due to the adsorption of natural organic matter (19), and quickly had lower concentrations of Fe and Al in the 3776

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 35, NO. 18, 2001

filtrate. The behavior of the 1.0 µm Nuclepore membrane was markedly different, however, and allowed roughly four times more artifact free water to pass than the 0.45 µm Durapore. This behavior is attributed to the relative absence of particles larger than 1.0 µm that might clog the larger pore-size filter. Despite these membrane differences, Fe, Mn, Al, and OC exhibited roughly similar behavior during membrane clogging, both within and between membranes. Changes in back pressure and flow rate during filtration closely corresponded to the measured changes in trace element concentrations. At this site, both 0.45 µm Durapore and 1.0 µm Nuclepore filters gave similar results until they clogged, though the latter filter yielded much greater amounts of artifact-free water. The 0.40 µm Nuclepore membranes began to clog immediately and would be expected to give very inconsistent results if trying to discriminate between macroparticulate and colloidal and dissolved forms. However, as the next two sites demonstrate, there is substantial site to site variability of filtration artifacts. Quinnipiac River. In the Quinnipiac River (Figure 3), Fe and Mn size distributions were more typical of conditions found in some other studies (13-15, 21, 22) in that a large proportion of Fe was macroparticulate (retained by the membranes even in the initial aliquot), and 52% of the Mn was in the truly dissolved phase (only 12% colloidal). The lower Fe levels in the initial aliquots for the 0.40 µm compared to the 1.0 µm Nuclepore membranes indicate that a significant portion of the Fe may be present as colloids in the size range between 0.40 and 1.0 µm. Also, Fe in the initial aliquot for the 0.45 µm Durapore membrane (9 µg/L) was much lower than for the other two filters (40 and 61 µg/L), suggesting that some colloids smaller than 0.40 µm may be retained by the Durapore membrane at this site even before the membranes begin to clog.

Organic matter in the Quinnipiac was found entirely in the colloidal (52%) and dissolved (48%) fractions yet was unaffected by membrane clogging for any membrane type. Again there was a substantial difference in the way that these membranes, when clogged, handle Fe and OC colloids. The former were retained while the latter were passed, and this was true across membrane types. It seems as if there are two distinct pools of colloids: large colloids composed mainly of Fe oxides and smaller ones containing substantial amounts of OC. This distribution matches results from detailed studies combining field flow fractionation and ICP-MS detection (30). Aluminum at this site was low (56 µg/L) and almost entirely macroparticulate (none detected in any of the filtrates for this site). The results for Fe (Figure 3) illustrate an example where filtration artifacts are pronounced with Nuclepore membranes. Neither the 0.40 nor the 1.0 µm Nuclepore membrane exhibited a region of constant Fe concentration prior to the onset of membrane clogging, and this occurred much more quickly for these membranes than for the 0.45 µm Durapore membrane. (However, the initial filtrate Fe was much lower for the Durapore membrane than for the Nuclepore membranes.) Mn was not affected substantially by membrane clogging during filtration of the Quinnipiac River water for any of the membranes tested. This can be explained by the small fraction of colloidal Mn in this sample. Mn was, however, affected by another artifact in the initial aliquot of the 0.45 µm Durapore membrane, as shown by the low value for the first 50 mL aliquot. We observed this artifact in a number of locations and ascribe it to either adsorption or retention of a limited quantity of small colloids in the interior of the Durapore membrane. The results for the Quinnipiac River support the idea that different pools of colloidal particles are present in natural waters and are affected differently by filtration artifacts. Wintergreen Brook. Wintergreen Brook (Figure 4) had a relatively large proportion of truly dissolved OC (88% of the filtrate fraction), with very little in the colloidal size fraction (12%). Organic carbon concentrations were not affected by membrane clogging during filtration of water from this site. Much of the Fe in this brook was macroparticulate. Fortyfour percent of the total Fe initially passed through the 1.0 µm Nuclepore membrane compared to 31% passing the 0.40 µm Nuclepore membrane and just 24% passing through the 0.45 µm Durapore membrane. The behavior of Fe during filtration of water from this site was similar to the Quinnipiac River but more pronounced. As in the Quinnipiac River samples, the concentration of Fe in the Nuclepore filtrates declined rapidly as the filters became clogged, while the somewhat smaller concentration of Fe in the Durapore filtrates remained relatively constant (94 ( 7% of the initial value) until the onset of membrane clogging (sixth aliquot). Manganese in Wintergreen Brook was present mainly in the colloidal (53%) and dissolved (25%) phases and was not affected by membrane clogging. The data from this site on both dates, however, shows a more pronounced undermeasurement of Mn in the first four aliquots through the 0.45 µm Durapore membrane that preclude any attribution of this artifact to dilution effects from entrained water. Changes in back pressure and flow rate were similar to data from the Quinnipiac River and show that decreases in Fe concentration in the filtrate are closely correlated to increases in back pressure and decreases in flow rate. Again Al was mainly macroparticulate, with no significant amount detected in the initial aliquot through either the 0.45 or 0.40 µm membranes. A substantial amount was measured in the 1.0 µm filtrate (46% of total), indicating that a large amount of the Al was just smaller than this size. Conditions varied from site to site and also with time. A baseflow collection from Wintergreen Brook a week earlier

FIGURE 4. Filtration data for Wintergreen Brook, New Haven, CT. This site was similar to the Quinnipiac in that only Fe, of constituents measured, declined dramatically. Note the increase in Mn for the Durapore filter. showed higher total Fe (2.49 compared to 1.07 mg/L) and an even greater deficit of Fe in the initial aliquot of the 0.45 membrane when compared to the 0.40 µm membrane result (Fe0.45 ) 11% of Fe0.40) similar to what was observed in the Quinnipiac River. Other measurements were similar on the two dates, with OC and Mn exhibiting nearly constant concentrations as a function of membrane loading. (Baseflow Mn in Wintergreen Brook was entirely in the colloidal (26%) and dissolved (74%) phases.) Correlation of Fe Concentration with Back Pressure and Flow Rate. Of the elements tested, Fe most consistently was observed to decrease with increasing membrane loading. Figure 5 is a graph of the Fe concentration data from all of the sites (including a second set for Wintergreen Brook) and all of the filter types plotted as a function of back pressure. The Fe data are normalized to the concentration in the initial filtrate aliquot collected from each membrane in order to show the relative change in concentration over the course of the filtration. The back pressure data is normalized to the maximum value for each filtration because this maximum value varied (( 10%) over the course of the study. This graph clearly demonstrates a threshold effect for changes in Fe concentration during membrane clogging in relation to back pressure regardless of sample site and/or membrane type. Below 90% of the maximum back pressure value, the Fe concentration is largely unaffected by membrane clogging. The only data set that showed significant variability below 90% of the maximum back pressure was for the Quinnipiac River. This could be due to the concentrations of Fe in the Quinnipiac River, which were substantially lower than at the other two sites. This data set shows that monitoring back pressure during filtration and ending the collection of filtrate when the back pressure reaches 80-90% of its maximum value would virtually eliminate colloid-induced clogging artifacts for the sites in this study. VOL. 35, NO. 18, 2001 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3777

FIGURE 5. Fe concentrations in filtrate as a function of back pressure for all of the sites and all filter types for each site. The Fe data are normalized to the concentration in the initial filtrate aliquot collected from each membrane in order to show the relative change in concentration over the course of the filtration. The back pressure data are normalized to the maximum value for each filtration. Below 90% of the maximum back pressure value, the Fe concentration is largely unaffected by membrane clogging. FIGURE 7. Concentrations of OC, Fe, Cd, Pb, Mn, Zn, and Cu in 0.45 µm Durapore filtrate of Leadmine Brook water as a function of back pressure. Mn, Zn, and Cu are unaffected by filter overloading. OC, Cd, and Pb like Fe decline dramatically when filters become clogged and back-pressure rises.

FIGURE 6. Fe concentrations as a function of flow rate normalized to an initial value of 125 mL/min; data shown for all of the sites and all filter types for each site. Figure 6 shows a similar graph of Fe concentration vs normalized flow rate. The flow rate is normalized to 125 mL/ min, which is the value for deionized water through the membranes under the conditions employed for this study. Figure 6 shows a strong correlation between the concentration of Fe and the flow rate but, in contrast to Figure 5, does not provide a distinct threshold for the discrimination against filtration artifacts due to membrane clogging. This is because flow rate changed gradually over the course of the filtration, rather than reaching a threshold and declining rapidly. For this reason, flow rate does not provide as good an indication of when membrane clogging begins to affect the concentration of Fe in the filtrate. On the other hand, a decline in filtration rate can be perceived qualitatively without the need for additional equipment like a pressure gauge but should be used with caution because artifacts affect filtrate concentrations well before the filtrate flow is reduced to a drip (typically at 40-50% of the initial flow). To demonstrate the applicability of the results from this study in routine sampling and to test the effect of membrane clogging on heavy metals, a single undeveloped stream site was chosen. Other elements that are associated with Fe (e.g., P (31)) at this site would be expected to decline as well when membranes clog. Samples were collected from Leadmine Brook (near its mouth; 41°42′0.1′′ N, 73°03′32.2′′ W), filtered in triplicate using a 0.45 µm Durapore membrane, and measured for OC, Fe, Mn, Cu, Cd, Pb, and Zn. Three filtrate fractions were collected and analyzed: low back pressure (6-10 psig), rising back pressure (11-13 psig), and maximum 3778

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 35, NO. 18, 2001

back pressure (14-15 psig). Figure 7 shows the results, with OC, Fe, Cd, and Pb all exhibiting substantial declines as back pressure increased, while Mn, Zn, and Cu remained constant within the uncertainty of the measurements. This result demonstrates that back pressure can be used to collect filtrate that is free from artifacts and indicates that a number of other freshwater constituents may commonly be subject to the same artifacts as Fe when filtration membranes become clogged. Recommendations. This research demonstrates the importance of colloids in the study of filtration artifacts. Based on the often large difference between material retained by 1.0 µm Nuclepore membranes compared to 0.40 µm Nuclepore and 0.45 µm Durapore membranes, it is evident that a significant fraction of the colloids in some natural waters exists in the 0.4-1.0 µm size range. These particles are significantly influenced by membrane clogging and may indeed be largely responsible for its occurrence. The majority of these particles is likely to be classified as macroparticulate matter when conventional filtration is used to define the upper size range for colloidal particles. Caution should be exercised in this practice, because these particles will not be removed readily by sedimentation and may be responsible for substantial transport of contaminants in natural waters. The very different filtering behavior of Fe, Mn, OC, and Al indicates that multiple size classes of colloidal particles exist in natural waters and that some of these colloids are less affected by membrane clogging. This suggests that the practice of exhaustive filtration (16, 21, 32), which relies upon membrane clogging, may not yield filtrate that is representative of the truly dissolved fraction. Even within a small geographic area, the systematics of particle size distributions, and the association of trace elements with various size classes, vary tremendously. The “best” filter to use in any given situation depends on the goal of the separation (e.g., where the cutoff needs to be delineated), the ambient particle size distribution (e.g., are colloids mainly close to the rated pore size or much smaller), the element(s) of main interest (e.g., Fe vs Mn), the volume of filtrate required, and other factors. It seems unlikely that any single protocol will yield similar results across many

sites, especially if multiple analytes are being considered. For research level measurement campaigns, it would be best to conduct studies of the sort described here, to evaluate the relative merits of different filters, and to determine the impact of filtration on the particle size distribution of the water from the study site. For routine measurements, analysts rarely have the luxury of conducting extensive preliminary studies and often must sample water from diverse freshwater sites. In this case, simple rules can be of great value. This research shows that a substantial increase in back pressure or decrease in flow rate during filtration is always a sign that membrane clogging has begun to cause a major shift in the size discrimination characteristics of the membrane filter for some elements. Other specific recommendations include the following: (1) Discard the first 25-50 mL of filtrate when using TP membranes in order to avoid undermeasurement of certain elements (e.g. Mn, see Wintergreen Brook data) that are lost to the membrane through either adsorption or entrapment. (2) TP membranes, while apparently retaining some colloidal particles less than 0.45 µm, are less affected by artifacts associated with membrane clogging across the range of freshwaters sampled for this study. (3) When using membrane filters, collect small volumes of filtrate and terminate collection well before the onset of membrane clogging (especially when using decreased flow rate as an indication of membrane clogging). This is becoming increasingly feasible with the advent of analytical techniques that do not require large sample volumes (e.g., ICP-MS) for the analysis of trace and heavy metals. If large volumes of water must be filtered for the analysis of particulate matter, use multiple filters for the collection and analysis of the filter-passing fraction (4). Great care (e.g., use of clean techniques and rapid sample processing) must be taken when collecting and processing natural water samples for the analysis of trace and heavy metals and organic matter, especially when information on particulate and colloidal forms is to be determined.

Acknowledgments This research has been supported by a grant from the U.S. Environmental Protection Agency’s science to achieve results (STAR) program. Although the research described in the article has been funded wholly or in part by the U.S. Environmental Protection Agency’s STAR program through grant R82-5395, it has not been subjected to any EPA review and therefore does not necessarily reflected the views of the agency, and no official endorsement should be inferred.

Literature Cited (1) Stumm, W.; Morgan, J. J. Aquatic Chemistry, 3rd ed.; WileyInterscience: New York, 1996; 1022 pp. (2) Buffle, J.; Leppard, G. G. Environ. Sci. Technol. 1995, 29, 21692175. (3) Gustafsson, O.; Gschwend, P. M. Limnol. Oceanogr. 1997, 42, 519-528.

(4) Douglas, G. B.; Hart, B. T.; Beckett, R.; Gray, C. M.; Oliver, R. L. Aquat. Geochem. 1999, 5, 167-194. (5) Ross, J. M.; Sherrell, R. M. Limnol. Oceanogr. 1999, 44, 10191034. (6) Buffle, J.; Leppard, G. G. Environ. Sci. Technol. 1995, 29, 21762184. (7) Perret, D.; Newman, M. E.; Ne`gre, J.; Chen, Y.; Buffle, J. Water Res. 1994, 28, 91-106. (8) Benoit, G.; Oktay-Marshall, S. D.; Cantu, A., II; Hood, E. M.; Coleman, C. H.; Corapcioglu, M. O.; Santschi, P. H. Mar. Chem. 1994, 45, 307-336. (9) Benoit, G.; Rozan, T. F. Geochim. Cosmochim. Acta 1999, 63, 113-127. (10) Martin, J.; Dai, M. Limnol. Oceanogr. 1995, 40, 119-131. (11) Buffle, J.; Wilkinson, K. J.; Stoll, S.; Filella, M.; Zhang, J. A. Environ. Sci. Technol. 1998, 32, 2887-2899. (12) Grout, H.; Wiesner, M. R.; Bottero, J. Environ. Sci. Technol. 1999, 33, 831-839. (13) Buffle, J.; Perret, D.; Newman, M. In Environmental Particles; Buffle, J., van Leeuwen, H. P., Eds; Lewis Publishers: Boca Raton, FL, 1992; Vol. 1, pp 171-230. (14) Horowitz, A. J.; Lum, K. R.; Garbarino, J. R.; Hall, G. E. M.; Lemieux, C.; Demas, C. R. Environ. Sci. Technol. 1996, 30, 954963. (15) Sheldon, R. W. Limnol. Oceanogr. 1972, 17, 494-498. (16) Kennedy, V. C.; Zellweger, G. W.; Jones, B. F. Water Resour. Res. 1974, 10, 785-790. (17) Danielsson, L. G. Water Res. 1982, 16, 179-182. (18) Laxen, D. P. H.; Chandler, I. M. Anal. Chem. 1982, 54, 13501355. (19) Horowitz, A. J.; Elrick, K. A.; Colberg, M. R. Water Res. 1992, 26, 753-763. (20) Karlsson, S.; Peterson, A.; Ha¨kansson, K.; Ledin, A. Sci. Total Environ. 1994, 149, 215-223. (21) Shiller, A. M.; Taylor, H. E. Environ. Sci. Technol. 1996, 30, 33983399. (22) Horowitz, A. J.; Lum, K. R.; Garbarino, J. R.; Hall, G. E. M.; Lemieux, C.; Demas, C. R. Environ. Sci. Technol. 1996, 30, 33993400. (23) Benoit, G. Geochim. Cosmochim. Acta 1995, 59, 2677-2687. (24) Rozan, T. F. Doctoral Dissertation, Yale University, 1998. (25) Rozan, T. F.; Benoit, G. Geochim. Cosmochim. Acta 1999, 63, 3311-3319. (26) Benoit, G.; Parrett, N. West River Water Quality Report: Year 2 (1997-1998); Connecticut DEP: Hartford CT, 2000. (27) Benoit, G. Environ. Sci. Technol. 1994, 28, 1987-1991. (28) Benoit, G.; Hunter, K. S.; Rozan, T. F. Anal. Chem. 1997, 69, 1006-1011. (29) Wang, E. X.; Benoit, G. Environ. Sci. Technol. 1996, 30, 22112219. (30) Hasselo¨v, M. Doctoral Dissertation, Go¨teborg University, 1999. (31) Fox, L. E. Geochim. Cosmochim. Acta 1989, 53, 417-428. (32) Taylor, H. E.; Shiller, A. M. Environ. Sci. Technol. 1995, 29, 13131317.

Received for review February 21, 2001. Revised manuscript received June 18, 2001. Accepted July 5, 2001. ES010670K

VOL. 35, NO. 18, 2001 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3779