Evaluation of ultrafiltration cartridges for a water sampling apparatus


Patricia M. Holowecky, Battelle, 505 King Avenue, Columbus, OH 43201, USA. E-mail: holoweckyp@battelle.org


Aims:  To determine the efficiency of various ultrafiltration cartridges (UFC) in concentrating test micro-organisms from drinking water.

Methods and Results:  Replicate drinking water samples from three potable water supplies were dosed with Bacillus anthracis Sterne, Francisella tularensis LVS, Yersinia pestis CO92, bacteriophages MS2 and phi-X174, and Cryptosporidium parvum. The test micro-organisms were dosed together in 100 l of water, which was then recirculated through one of five different UFC until the retentate volume was reduced to c. 500 ml. The micro-organisms were assayed before and after ultrafiltration concentration and per cent recoveries were calculated. There were nine statistically significant differences among pairs of filters out of a possible 180 different combinations of UFC, test micro-organisms, and water types.

Conclusions:  No filter consistently performed better or worse than the others for each test micro-organism in all water samples tested.

Significance and Impact of the Study:  This study provides performance data on the ability of several different UFC to concentrate a panel of test micro-organisms from three sources of potable water. Water utilities and first responders may use these data when selecting UFC for use in emergency response protocols. This study also provides additional data as to the efficacy of ultrafiltration for recovering bacteria, virus-like particles, and protozoan oocysts from water samples.


Bacteria, viruses, and parasites can pose significant health risks to humans through exposure via drinking water, thus these micro-organisms could be used to intentionally contaminate a water supply. Current technologies for detecting micro-organisms in drinking water, such as immunological methods or polymerase chain reaction assays, have detection limits that are at or above microbial concentrations at which health effects are likely to be observed (Buehler et al. 2004; James et al. 2004). Therefore, negative results from sampling of a water supply using these methods would not ensure that the concentration of pathogenic micro-organisms in the water was below the level that could cause disease in humans. In order to determine if pathogenic micro-organisms are present at concentrations of concern, large volumes of water (100–1000 l) need to be concentrated and tested for the presence of harmful micro-organisms (Lopez-Pila et al. 1996; Rochelle et al. 1999).

Ultrafiltration uses size-exclusion as the mechanism of concentration, where molecules that are smaller than the filter pore size pass through the membrane and out of the concentration apparatus while larger particles are concentrated in the retentate and continue to recirculate within the concentration apparatus. The pore size of ultrafiltration devices is between 0·001 and 0·05 μm, which is smaller than the majority of human pathogenic bacteria and viruses, and even some large protein toxin molecules. Re-circulation of the retentate reduces fouling of the membrane and makes it possible to filter large volumes of water while keeping the targeted micro-organisms in suspension (Juliano and Sobsey 1998; Kuhn and Oshima 2001; Oshima et al. 2001; Simmons et al. 2001). Ultrafiltration has been shown to be an approach to concentrate biological contaminants in drinking water samples so that currently available technologies can detect them at or below concentrations expected to cause health effects (Oshima et al. 2001; Hill et al. 2005, 2007; Powell and Timperman 2005).

The purpose of this study was to compare the performance of several different ultrafiltration cartridges (UFC) to simultaneously recover spores of Bacillus anthracis; vegetative cells of Yersinia pestis, and Francisella tularensis; bacteriophages, MS2 and phi-X174 (PHIX); and oocysts of Cryptosporidium parvum from drinking water. Water was collected from three potable water sources. Per test, 100 l of each water type was dosed with the test micro-organisms at known concentrations and then continuously re-circulated through a water concentration apparatus as described in Lindquist et al. (2007), which incorporates the selected UFC. This technique allows the filtered water to permeate through while the retained water becomes increasingly more concentrated with micro-organisms that are larger than the pore size of the filter. The concentrated water sample (retentate) was then assayed to determine the concentration of each target micro-organism. Differences in the efficiency of recovery of the various micro-organisms were noted between the UFC and water types.

Materials and methods

Ultrafiltration cartridges

The UFC that were evaluated included the Asahi Kasei (AK) APS, the Baxter Corporation (BC) Exeltra Plus 210, Fresenius Medical Co. (FMC) F200NR, Minntech Corporation (MC) HPH 1400, and AK Rexeed (Table 1).

Table 1.   Ultrafiltration cartridges
VendorModelMolecular weight cut-off (kilodaltons)Membrane compositeSurface area (m2)Hollow-fibre diameter (μm)Cartridge hold up volume (ml)
  1. ND, not determined; no data was provided for this parameter by the vendor, and the value was not determined in the laboratory.

Asahi KaseiAPS series
Baxter Corp.Exeltra Plus 21070Cellulose triacetate2·1NDND
Fresenius Medical Co.F200NR30Polysulfone2·0200112
Minntech Corp.HPH 140065Polysulfone1·320095
Asahi KaseiRexeed20Polysulfone2·5185128

Test micro-organisms, growth and enumeration

Bacillus anthracis Sterne

Bacillus anthracis is the causative agent of anthracis, which can cause inhalational, gastrointestinal, or dermal illnesses. The B. anthracis Sterne strain was obtained by Battelle from Colorado Serum Co. (Denver, CO), which manufactures a US Department of Agriculture (USDA)-approved veterinary vaccine for anthrax using this strain. A B. anthracis spore bank was produced from the vaccine stock by growing the organism to sporulation (c. 48 h) in modified G broth (modG; Kim and Goepfert 1974). The sporulated culture was harvested by centrifugation at 12 000 g at 4°C for 15 min, washed twice with sterile distilled water, and stored at 2–8°C in sterile distilled water. The spore bank was shown to be pure based on colony and spore morphology, and, based on microscopic observations, the suspension consisted of >95% refractile spores with minimal cell debris. On the day of each test, 100 l of test water was seeded with a known quantity of B. anthracis spores targeting a concentration of 1 × 106 per 100 l, mixed by swirling, and then run through the ultrafiltration procedure. Following each test, the concentration of B. anthracis in the final water concentrate (retentate) was determined using standard microbiology spread plate methods (Bacteriological Analytical Manual, 8th edn) using blood agar and phosphate-buffered saline (PBS) at pH 7·2. Enumeration plates for B. anthracis were incubated at 35–37°C until the colonies reached adequate sizes for counting, but were still distinguishable from one another (14–24 h).

Yersinia pestis Colorado 92 (CO92)

Yersinia pestis is the causative agent of bubonic and pneumonic plague. The Y. pestis CO92 strain is attenuated and devoid of the 75 kb low-calcium response (Lcr) virulence plasmid. The Y. pestis CO92 used in this study was obtained from the T. Quan, Centers for Disease Control (Ft. Collins, Colorado). A working cell bank (WCB) was prepared from a culture grown in heart infusion broth (HIB; Difco Laboratories, Franklin Lakes, NJ) supplemented with 0·2% xylose for 24 h at 37°C. Sterile glycerol was added to the culture to achieve a final concentration of 15% and then 1-ml aliquots were frozen at < −70°C. Prior to each test, a vial of the WCB was used to inoculate blood agar and incubated for 48 h at 35–37°C. On the day of testing, a seed stock was prepared in PBS and added to the water sample at an appropriate volume to achieve a target concentration of 107 CFU per 100 l. The seed stock and the final retentate from the ultrafiltration test were enumerated using standard spread plate methods using blood agar and PBS. The enumeration plates were incubated at 35–37°C for c. 48 h or until the colonies reached a distinguishable size.

Francisella tularensis live vaccine strain (LVS)

Francisella tularensis is the causative agent of tularemia, also known as rabbit fever or new world plague. Francisella tularensis LVS is registered as an investigation new drug with the Food and Drug Administration (FDA) as a candidate vaccine against tularemia. The F. tularensis LVS used in this study was obtained from the American Type Culture Collection (Manassas, VA; ATCC 29684). An WCB was prepared by cultivating F. tularensis in peptone cysteine broth, adding glycerol to a final concentration of 15%, and then freezing 1-ml aliquots at < −70°C. The F. tularensis WCB was checked for purity by cultivation on Thayer-Martin agar (TMA; Cat no. 01884, Remel, Lenexa, KS) and chocolate agar. This cell bank was then used to prepare a culture grown onto chocolate agar for 48 h, which, in turn, was used to prepare a concentrated cell suspension in PBS. The concentration of the cell suspension was estimated based on the optical density at 600 nm. The actual concentration (CFU ml−1) was then determined using standard microbiology spread plate methods using chocolate agar and PBS. The dosing concentration of F. tularensis LVS was 107 CFU per 100 l. The final concentration was determined by counting the CFU from triplicate 100-μl aliquot spread plates on TMA. These plates were incubated at 37°C and checked every 48–96 h.

Bacteriophages MS2 and PHIX

MS2 and phi-X174 are bacteriophages that infect bacteria, specifically only certain strains of Escherichia coli. In this study, MS2 was propagated in E. coli ATCC 15597, and PHIX was propagated in E. coli ATCC 13706. MS2 and PHIX are found regularly in the environment and wastewater treatment facilities. The bacteriophage and E. coli hosts used in this study were obtained from the ATCC. The purchased bacteriophage was resuspended in PBS, enumerated, stored refrigerated, and then used as seed stocks to dose the water for each ultrafiltration test. The seeding concentration of MS2 and PHIX was 106 and 105, plaque-forming units per 100 l, respectively.

To enumerate bacteriophage in the stock solutions (before spiking), 100-μl aliquots were combined with 200 μl of an E. coli suspension (c. 1 × 109 CFU ml−1) onto an empty petri dish, and then overlaid with 20 ml of molten 0·7% Lennox agar and mixed by swirling. Each sample was plated in triplicate. The plaques were counted after 14–20 h of incubation at 35–37°C. For enumerating the retentate samples also containing the other micro-organisms, 5–10-ml aliquots of the retentate samples were filtered through a 0·45-μm porosity polyvinylidene fluoride (PVDF) filter (Millex HV, Millipore, Billerica, MA, USA) to remove cells, oocysts, and large debris. Filtered aliquots of 1 ml were then enumerated as described before.

Cryptosporidium parvum

Cryptosporidium parvum is the causative agent of cryptosporidiosis. Cryptosporidosis is often manifested by gastroenteritis of long duration, which may be fatal in people with compromised immune systems. There are a number of other symptoms associated with cryptosporidosis and the micro-organism may cause systemic infection in individuals with immune deficiencies. Cryptosporidium parvum was purchased from Waterborne, Inc. (New Orleans, LA, USA) and stored at 4°C. The sample of C. parvum used for the evaluation was received in May 2006 and all of the testing was completed by the end of October, therefore, use of the oocysts was completed within five months of receipt. The oocyst spiking suspensions, prepared daily for each filtration experiment, were enumerated in order to determine the spiking volume. This was accomplished by spotting triplicate 25-μl aliquots onto slides with an etched grid allowing the sample to air dry. These slides were heat fixed in a flame for 1–2 s. They were then stained with a Crypt-a-Glo fluorescent antibody staining kit (Waterborne Co.) following the manufacturer’s instructions. The oocysts were counted using a 100 ×  epifluorescence microscope. The seeding concentration of C. parvum was 103 oocysts per 100 l. This microscopic analysis also allowed examination of the oocysts to ensure that clumping did not occur.

After enumerating the bacteria and bacteriophage, the remainder of the concentrated water sample (between 250 and 500 ml), was centrifuged at 15 530 g for c. 15 min. The supernatant was drawn off and the pellet was resuspended in 10 ml of PBS (pH 7·2). One millilitre of the resuspended pellet was centrifuged again for c. 1 min; the supernatant was then discarded and the resultant pellet was resuspended to 100 μl with PBS. Triplicate 25-μl aliquots of this concentrated solution containing oocyst, spores, cells, and phage were dispensed onto a glass slide etched with a counting grid. The slides were air dried, heat fixed and stained, and counted.

Water sample concentration apparatus and UFC

Five different hollow-fibre UFC (Table 1) containing of polysulfone or cellulose triacetate filtration media were evaluated on their ability to concentrate test micro-organisms from 100-l water samples. The filter membrane’s molecular cut-off weights reported by the manufacturers ranged from 20 to 70 kDa and the surface areas of the filtration media ranged from 1·3 to 2·5 m2. During the evaluation, these UFC were inserted into a water sample concentration apparatus (Fig. 1) (Lindquist et al. 2007). Each component within the apparatus was connected with Tygon tubing (R-3603 formulation; St. Gobain, France) and the water was re-circulated by a peristaltic pump (Masterflex, Cole Parmer, Vernon Hills, IL, USA).

Figure 1.

 Schematic diagram of the water sample concentration apparatus.

Drinking water samples

This evaluation included the use of three different finished drinking water sources. Approximately 7500 l was collected at one time from a faucet located at Battelle in Columbus, OH (referred to in this report as ‘Battelle’ water). Battelle’s water supply is served by the City of Columbus Division of Water, specifically from a plant whose source is the Scioto River. Upon intake into the treatment plant, the water is coagulated with chemical flocculants, settled, filtered, chlorinated, and then pumped into the City of Columbus, OH distribution system. Because this sample need not have to be transported from an off-site location, a larger number of sample replicates were conducted with this water sample to increase the overall statistical power of the analysis. Five replicate concentration procedures were performed using this sample while three replicates were performed for the other two water samples. The second drinking water sample (c. 2000 l) was collected from a fire hydrant at the City of Columbus, OH, Parsons Avenue Water Treatment Plant (referred to in this report as ‘Columbus’ water). The source of water for this treatment plant is an underground aquifer. Upon intake into the treatment plant, the water is filtered, chlorinated, and then pumped into the City of Columbus, OH water distribution system. The third water sample (c. 2300 l) was collected from a fire hydrant in the Bronx borough of New York, New York (referred to in this report as ‘NYC’ water). NYC water was unique in this study because the majority of NYC water is not filtered during the treatment process. NYC water treatment relies predominantly on a system of large reservoirs that collect water from a protected watershed. The water is also disinfected and chlorine residual is added.

Prior to use in this evaluation, each sample was stored in plastic tanks (c. 2000 l). Upon the collection of each sample, the water parameters were measured including pH, using EPA Method 150·1 (US EPA 1983a), total dissolved solids using EPA Method 160·3 (US EPA 1983b), and turbidity using EPA Method 180·1 (US EPA 1983c) (Table 2). To monitor for microbial overgrowth, 100 ml of the water samples was collected weekly and 1-ml aliquots were plated on tryptic soy agar for heterotrophic plate counts. Upon inspection of the plates, if growth was observed, chlorine was added to c. 200 parts per million (ppm) to prevent further growth.

Table 2.   Water quality data
ParameterMethodBattelleNew York City, NYColumbus, OH
  1. NTU, nephelometric turbidity units.

  2. *US EPA (1983a).

  3. US EPA (1983b).

  4. US EPA (1983b).

pHEPA 150·1*7·27·37·3
Total suspended solids (mg l−1)EPA 160·3†<5·0<5·0<5·0
Turbidity (NTU)EPA 180·1‡0·11·70·9

For each concentration procedure performed, 100-l aliquots of one water sample were collected in 100-l carboys and transported to the laboratory. On any given day, only water from a single location was tested. Chlorine was added to the carboys for a holding period of 30–60 min to a concentration of 200 ppm. Overnight, the chlorine in each sample was neutralized by adding 50 ml, 0·1 mol l−1 sodium thiosulfate to the carboy. An N,N-diethyl-p-phenylenediamine test kit was used to show that chlorine was not present in a concentration greater than the minimum detection limit (0·1 mg l−1).

Filtration procedures

A mixture of the six test micro-organisms was generated that was sufficient for all sample runs for one day. While this mixture was made to an approximate concentration based on previous enumeration of these micro-organisms, a small aliquot of the mixture was reserved and used to confirm the initial concentration of each micro-organism. The confirmed concentration was used in calculations of recovery efficiency. After dechlorination of the 100-l water sample aliquot, an appropriate amount of the micro-organism mixture was added to each 100-l sample aliquot and the samples were mixed by agitation. One litre of 0·5% bovine calf serum albumin was re-circulated within the filtration system prior to filtration to attempt to reduce nonspecific protein binding of the plastics in the system and increase recovery. One litre of distilled water was then flushed through the system to decrease the formation of bubbles while re-circulating the 100-l water sample.

The 100-l water sample, now dechlorinated and spiked with micro-organisms, was then pumped through the system at an approximate flow rate of 2 l min−1 with a transmembrane pressure of c. 20–30 psi. The transmembrane pressure was controlled by adjusting the pump speed and by strategically placed hose clamps. The system is designed to allow for re-circulation for a proportion of the concentrated sample through the filter by mixing with the sample. Filtration was terminated when the 100-l sample was depleted, and the retentate bottle contained c. 250 ml. The system was then back-flushed with 150 ml of 0·001% Tween-80 at a flow rate of c. 250 ml min−1 to elute the collected micro-organisms from the filter. Throughout the evaluation, three to four UFC were evaluated simultaneously on each testing day. The UFC to be used on a given day were chosen in random order while the water source remained consistent on testing days. The random order was generated through use of a random numbers table.

Data analysis

As described previously, each UFC was used to concentrate Battelle, NYC, and Columbus water each contaminated with all six test micro-organisms simultaneously. The per cent recovery (%R) of each test micro-organism was calculated using the following equation:


where N1 is the initial number per 100 l of each test micro-organism added to the 100-l water sample and N2 is the final number per 100 l calculated for each specific test micro-organism as recovered from the sample after processing. N1 was determined from the enumeration of the mixed culture of test micro-organisms added to the initial water sample prior to concentration. N2 was determined from the enumeration of the retentate after completion of the concentration procedure and corrected for the proportion of the sample analysed. Outliers were determined by Dixon’s Q-test and were not used in the calculation of mean %R.

Per cent recoveries for each UFC

Statistical analyses were performed on the %R determined for each UFC, micro-organism, and water sample concentration (independent of all others) to evaluate their performance in concentrating all six test micro-organisms in the three water samples. A one-way analysis of variance (anova) was performed on the data across all six micro-organisms within each water sample. The results of this analysis provided the probability of differences in %R existing between micro-organisms. However, the anova does not provide information about which pairs of micro-organisms may be different, so Student’s t-tests were performed on the %R between each test micro-organism regardless of the significance of the anova to determine the probability that differences exist between the recoveries of each micro-organism. The probabilities determined during these analyses were adjusted to account for the propagation of error owing to multiple individual comparisons (i.e. all the possible combinations of micro-organisms and UFC) using the Benjamini and Hochberg (1995) false discovery rate. This P-value adjustment is a slightly less conservative adjustment than the commonly used Bonferroni adjustment.


Water quality

Table 2 shows the pH, total suspended solids, and turbidity for each of the three water samples. There were blank runs for each water sample and each filter, where the sample was run without the addition of any test micro-organisms. These blanks were tested for all six of the test organisms. All of these results were negative as expected.

Large volume water samples were collected and held for analysis in an effort to keep the water samples as consistent as possible. This resulted, ultimately in overgrowth of heterotrophic bacteria. These bacterial populations were controlled by chlorination of the water sample during storage of the bulk sample. Even given these measures to control the normally occurring flora, there were seven occasions when culture plates were uncountable owing to overgrowth of nontarget organisms. It should be noted that the water was dechlorinated prior to the addition of the test organisms.


Table 3 provides the %R data, including the average and SD for each water sample and test micro-organism that was concentrated using the AK APS UFC from each sample. There was no significant difference for recovery of the test micro-organisms in Battelle (= 0·22) or NYC (P = 0·65) water. However, statistically significant differences were seen between the test micro-organisms in the Columbus water (P = 0·00035). Two-sample t-tests performed on all the possible combinations of test micro-organisms showed no significant differences between the test micro-organisms in Battelle or NYC water. These tests showed that, in the Columbus sample, the recovery of Y. pestis was significantly higher than F. tularensis, MS2, and C. parvum; the recovery of F. tularensis was significantly lower than B. anthracis and C. parvum; and the recoveries of B. anthracis and C. parvum were significantly higher than MS2 (See Table 3). The differences between test micro-organisms in Columbus water were because of high recovery efficiency of B. anthracis and Y. pestis, and low recoveries of F. tularensis and MS2 as compared with the other water samples. Similar t-tests were performed to compare recoveries of the micro-organisms between the water samples as well; however, there were no significant differences. The average amount of time required to concentrate the 100-l water samples was 2 h (0·8 l min−1).

Table 3.   Recoveries of different filters, by water source and micro-organism
FilterWater source (n)Percent recovery for each micro-organism (SD)Overall
Bacillus anthracisYersinia pestisFrancisella tularensisMS2PHIXCryptosporidium parvum
  1. AK APS, Asahi Kasei APS; BC, Baxter Corp. Exeltra Plus 210; FMC, Fresenius Medical Co. F200NR; MC, Minntech Corp. HPH 1400; AK Rexeed, Asahi Kasei Rexeed; ND, not determined.

  2. *Contains one data point where fewer than 25 counts were used to calculate recovery.

  3. †Contains two data points where fewer than 25 counts were used to calculate recovery.

  4. ‡Contains three data points where fewer than 25 counts were used to calculate recovery.

  5. §Contains one data point where the data was not used owing to overgrowth of nontarget bacteria.

  6. **Contains one data point that was removed as an outlier after a Q-test.

  7. ††Contains one data point that was removed as an outlier.

AK APSBattelle (5)48 (34)*83 (28)32 (17)*77 (61)*57 (30)53 (18)58 (36)
NYC (3)88 (38)72 (8)§39 (21)††75 (53)36 (11)ND63 (35)
Columbus (3)118 (41)137 (30)10 (14)†8 (9)*78 (37)64 (16)69 (55)
Overall (11)78 (45)97 (36)27 (19)57 (55)57 (30)57 (17)62 (42)
BCBattelle (5)94 (51)§98 (72)*33 (22)*85 (47)48 (25)35 (21)64 (49)
NYC (3)83 (32)33 (19)16 (19)9 (4)**55 (43)ND41 (36)
Columbus (3)50 (13)69 (61)23 (32)54 (49)51 (47)81 (21)61 (35)
Overall (11)77 (39)72 (61)28 (22)60 (50)55 (30)52 (31)58 (43)
FMCBattelle (5)53 (16)*63 (29)51 (20)*58 (20)**52 (20)28 (16)50 (23)
NYC (3)101 (12)92 (55)103 (112)74 (47)60 (68)ND86 (59)
Columbus (3)114 (55)112 (16)18 (1624 (22)61 (20)54 (5)67 (44)
Overall (11)80 (44)84 (38)62 (62)52 (34)57 (34)38 (18)64 (42)
MCBattelle (5)67 (6)83 (30)67 (16)70 (28)73 (44)24 (14)64 (30)
NYC (3)49 (17)36 (32)38 (32)142 (68)95 (78)ND72 (61)
Columbus (3)72 (8)118 (46)38 (29)†65 (55)*69 (3)50 (18)69 (38)
Overall (11)63 (13)80 (45)51 (26)88 (55)78 (46)33 (20)67 (41)
AK RexeedBattelle (5)60 (44)†61 (5)‡**17 (10)*89 (32)83 (34)36 (27)58 (37)
NYC (3)77 (28)40 (39)*56 (84)28 (2)**73 (101)§ND66 (57)
Columbus (3)57 (11)81 (13)6 (5)†40 (47)104 (6)81 (34)65 (37)
Overall (11)64 (32)61 (26)30 (46)62 (42)95 (45)53 (36)62 (41)
All filtersBattelle (25)62 (36)78 (39)40 (24)76 (39)63 (32)35 (21)59 (36)
NCY (15)79 (29)53 (39)55 (66)73 (62)69 (60)ND66 (52)
Columbus (15)82 (40)103 (41)24 (22)38 (40)78 (24)66 (22)66 (42)
Overall (55)72 (36)79 (43)40 (39)64 (48)68 (39)47 (25)63 (42)

BC Exeltra Plus 210

Table 3 provides the %R data for each water sample and test micro-organism concentrated using the BC UFC. Per cent recoveries for each micro-organism were not significantly different from one another when the BC UFC was used with any of the three water samples (Battelle, P = 0·08; Columbus, P = 0·63; NYC, P = 0·37). Two-sample t-testing showed no significant differences between the micro-organisms or water samples when using the BC UFC (Table 4). The average amount of time required to concentrate the 100-l water samples was 2 h (0·8 l min−1).

Table 4.   UFC pairs with significantly different %R (average %R)
Micro-organismBattelleColumbusNYCAmong sources
  1. AK APS, Asahi Kasei APS; PHIX, phi-X174; NA, not applicable.

Bacillus anthracisNo significant differencesNo significant differencesFresenius (101) > Minntech (49)No significant differences
Yersinia pestisNo significant differencesAK APS (137) > AK Rexeed (81)AK APS (88) > Baxter (33)AK APS > AK Rexeed
Francisella tularensisMinntech (67) > AK APS (33);
Minntech > Baxter (33);
Fresenius (51) > AK Rexeed (17);
Minntech > AK Rexeed
No significant differencesNo significant differencesNo significant differences
MS2No significant differencesNo significant differencesNo significant differencesNo significant differences
PHIXNo significant differencesAK Rexeed (104) > Fresenius (61);
AK Rexeed > Minntech (69)
No significant differencesNo significant differences
Cryptosporidium parvumNo significant differencesNo significant differencesNANo significant differences


Table 3 provides the %R data for each water sample and test micro-organism that was concentrated using the FMC UFC. The per cent recoveries of the micro-organisms were not significantly different from one another when the FMC UFC was used with Battelle (P = 0·12) or NYC (P = 0·92) water (Table 4). However, there were differences between the test micro-organisms in the Columbus water sample (P = 0·0031). While two-sample t-tests showed no significant differences between the test micro-organisms in Battelle or NYC water, the recovery of Y. pestis was significantly higher than F. tularensis, C. parvum, and MS2 in Columbus water (see Table 3). The differences between test micro-organism recoveries in Columbus water were because of the high recoveries of B. anthracis and Y. pestis and low recoveries of F. tularensis and MS2 compared with the other samples. Comparisons of the recoveries of the micro-organisms between the water samples showed a significantly higher recovery of B. anthracis in the NYC water than the Battelle water (P = 0·03). The average time to concentrate 100 l of water was 2·5 h (0·7 l min−1).

MC HPH 1400

Table 3 provides the %R data for each water sample and test micro-organism that was concentrated using the MC UFC. There were no significant differences in %R between water samples using the MC UFC (Table 4). While anova showed that there were no significant differences in recoveries between the test micro-organisms in NYC water (P = 0·11) or Columbus water (P = 0·13), the anova results suggested recovery differences between micro-organisms in the Battelle water (P = 0·025) (Table 4). Specifically, the recoveries of B. anthracis, Y. pestis, and F. tularensis were significantly higher than C. parvum. The recoveries of B. anthracis, Y. pestis, and F. tularensis were high, though not significantly so, relative to the recoveries in other samples, and C. parvum recoveries were low relative to the other samples. Similar t-tests were performed to compare the recoveries of the micro-organisms between the water samples as well. There were no significant differences in the recovery of the micro-organisms between the water samples. The average time to concentrate 100 l of water was 3 h (0·6 l min−1).

AK Rexeed

Table 3 provides the %R data for each water sample and test micro-organism that was concentrated using the AK Rexeed UFC. The %R of the test micro-organisms were not significantly different from one another when the AK Rexeed UFC was used with NYC water (P = 0·91). However, there were statistically significant differences between the test micro-organisms in both Battelle (P = 0·022) and Columbus (= 0·0054) water. In the Battelle water, the recovery of F. tularensis was significantly lower than Y. pestis, MS2, and PHIX; in the Columbus water, the recovery of F. tularensis was significantly lower than B. anthracis, Y. pestis, and PHIX and the recovery of PHIX was significantly higher than B. anthracis (Table 3). The differences between test micro-organisms in the Battelle water were because of the consistently low recovery of F. tularensis in combination with high MS2 and PHIX recoveries, compared with the other samples. Similarly, in the Columbus water, the differences were also because of low recovery of F. tularensis and the high recovery of PHIX. There were no significant differences in recoveries of the test micro-organisms between the water samples (Table 3). The average time to concentrate 100 l of water was 1·5 h (1·1 l min−1).

Comparison between UFC

Table 3 shows the overall recovery of each micro-organism for each filter. For each test micro-organism, a two-way anova model was fitted to the data, including all five different UFC and three water samples as fixed effects. The fitted model was used to determine whether there were statistically significant differences between the five UFC or the three water samples. Paired t-test comparisons of the five UFC (a total of 10 possible pairs of UFC) were performed for each water sample to identify pairs of UFC or water samples with significantly different recoveries. Table 4 summarizes these results. The UFC with the higher recovery is listed first and is followed by a ‘greater than’ sign. Also, the average %R is given in parentheses for each UFC. Although there were some differences in UFC performance within some water samples, there was only one significant difference between UFC for one test micro-organism across all water samples. In this instance, the AK APS had a statistically significantly greater %R of Y. pestis than the AK Rexeed across all samples.

Another important parameter to consider is the potential variability in recovery from filter to filter. The average relative SD of the recoveries in Battelle water were compared across all six target micro-organisms in order to evaluate whether any of the UFC generated more repeatable recovery results. The average SD were: AK APS (54% ± 19%), BC (60% ± 8%), FMC (41% ± 10%), MC (38% ± 20%), and AK Rexeed (49% ± 26%). The MC and FMC results were more reproducible, but the uncertainties associated with those averages overlapped with the other UFC making it impossible to determine if any of the UFC generated significantly more repeatable results than the others. A similar comparison was performed across the test micro-organisms to determine if any of the test micro-organisms had inherently greater or lesser variability in recovery. The average variability across water samples and UFC was: B. anthracis (48% ± 28%), Y. pestis (40% ± 24%), F. tularensis (48% ± 17%), MS2 (49% ± 19%), PHIX (49% ± 9%), and C. parvum (57% ± 15%). There was no statistically significant difference in the variability of recovery of the different micro-organisms.


None of the UFC consistently performed better or worse than the others for all micro-organisms or water samples. Given the number of UFC, test micro-organisms, and water samples there were 180 different possible combinations to compare UFC. Of these, there were nine instances where pairs of UFC performed significantly different from one another. Of these significantly different pairs, the MC UFC had five occurrences (three times with higher recoveries and twice with lower recoveries); AK Rexeed had five occurrences (three times with lower recoveries and twice with higher recoveries); AK APS and FMC three; and BC two, respectively.

Any study of this type suffers from the inability to compare every potential UFC, water sample, and micro-organism against every other one. Water samples change over time, throughout the year, and even daily. Micro-organisms vary from organism to organism, strain to strain, and may vary according to culture conditions. Ultrafilters may also vary from lot to lot, manufacturer to manufacturer, and so on. It is impossible to conduct a single comprehensive study of each possible value for all of these potential variables. The conditions under which the samples were collected and stored until use, and the use of chlorination to control bacterial overgrowth may mean that, to some extent, the samples are not representative of water freshly drawn from a distribution system. As the purpose of this project was to compare filters, this method was chosen to attempt to reduce the variability within each water sample collected.

From the possible 180 combinations of conditions, it would be expected that 5% of these would show statistically significant differences at a P-value of 0·05 or less even if there was no true difference in the performance of the filters. In this study, there were nine statistically significant differences in comparisons, or precisely 5% of the 180 total possible comparisons. Moreover, there was no trend in these significant differences. There was no single filter, micro-organism, or water sample that performed with consistently greater or lesser recovery than the others. There was no trend to suggest any variability in the performance of these particular filters, given the conditions of testing. The Benjamini and Hochberg (1995) false discovery rate correction was used to account for the occurrence of spurious statistical differences caused by multiple comparisons. Although a more conservative method could have been chosen to adjust for multiple comparison errors it may not be possible to entirely statistically correct for the detection of occasionally significant but unimportant differences when simultaneously testing a large number of parameters. Given the large inherent variability in this data set, a relatively nonconservative method for multiple comparison adjustment was justified to avoid missing patterns of significance that may otherwise have been obscured. For example, if a particular UFC achieved a lower %R consistently, but not always statistically significantly.

The lack of important significant differences between the various UFC may partially be because of large variation inherent in the data. The reason for this large variability is not clear. Bacteria, viruses, and to some extent oocysts are all susceptible to small but significant changes in experimental conditions. Reduction of variability may be difficult to accomplish in projects of this scale. However, the scale of variability within this experiment is comparable with the other reports of variability in recovery testing systems of recovery for micro-organisms from environmental samples (Juliano and Sobsey 1998; Morales-Morales et al. 2003; Hill et al. 2007; Lindquist et al. 2007).

Ideally, the %R for each of the test micro-organisms would be 100% in each replicate of each drinking water sample. Often recovery was less than 100%. In these cases, some test micro-organisms may be rendered nonviable during the concentration procedure owing to the drinking water matrix or the conditions. They may bind to the walls of the tubing, containers, or to the UFC, or pass through the UFC into the waste container. There were also cases of apparent over-recovery (>100%), as seen in Tables 2–4. In these cases, it is likely that recovery of micro-organisms was high, but the counting procedure underestimated the number of micro-organisms spiked into the sample, thus giving an apparent recovery greater than 100%. There are several possible causes of faulty enumeration including cell-clumping, matrix effects, sampling, and so on.

Most ultrafiltration concentration work reported in the literature involves single micro-organisms (Kfir et al. 1995; Simmons et al. 2001; Kuhn and Oshima 2002); however, there are at least four published examples (Juliano and Sobsey 1998; Morales-Morales et al. 2003; Hill et al. 2007; Lindquist et al. 2007) of work involving simultaneous concentration of test micro-organisms. The overall range of recoveries reported in this manuscript falls within the ranges of recoveries and variation found in the literature for similar organisms.

The UFC were all very similar from a functional standpoint. With the exception of the MC UFC, each of the five UFC concentrated the 100-l water samples in c. 2 h. The MC UFC took c. 3 h to complete each concentration procedure. This difference was likely because of the fact that there is c. 35% less membrane surface area in this UFC than the others tested.

Three water samples were used during this study. The Battelle and Columbus water samples were tap water samples from a filtered and chlorinated water system originating from surface and groundwater sources, respectively. In addition, the NYC water was a tap water sample from a predominantly nonfiltered chlorinated water system from a surface water source. For each UFC, five replicates of the Battelle water were used. Additional samples were used to increase the overall statistical power of the analysis. Overall, the technique worked for each sample, but difficulty arose with the NYC water. The increased particle load in the unfiltered NYC water made accurate microscopic enumeration of C. parvum oocysts impossible. Use of selective concentration of oocysts by a technique such as immunomagnetic separation may have allowed for analysis of these samples. This sample also showed increase in variability in recovery of bacteria and viruses, possibly as a result of the higher particulate load.

One of the main reasons for selecting the NYC water for this evaluation was that it is principally unfiltered; therefore, it is likely to be a relatively turbid water sample. The turbidity of the sample was 1·7 nephalometric turbidity units (NTU). While the total suspended solids measured in the initial characterization of the NYC water, was not different than the other two water samples, the turbidity was higher (Table 2). Following the concentration of the drinking water samples, the overall turbidity of the concentrated water caused difficulty with counting the fluorescently labelled C. parvum oocysts. After attempting several approaches to conclusively distinguish between the oocysts and the background particles, it was decided that the other water samples provided adequate performance information on C. parvum. Therefore, there are no C. parvum recovery data for the NYC water.

The five UFC that were evaluated performed very similar to one another under the conditions tested. This included analysis of three different water samples and six different test micro-organisms. The large variability in the results may have made small differences in the data appear to be statistically insignificant. However, there were no overall or general trends that might have suggested any important differences in the recovery performance of the various UFC.


This study was undertaken as a part of the Technology Testing and Evaluation Program (TTEP) and was funded by the US Environmental Protection Agency. The authors would like to thank Vincent Hill of the US Centers for Disease Control and Prevention and Kevin Oshima of the US EPA for their assistance during this study as well as Minntech Corporation, Fresenius Medical, and Baxter Corporation for donation of their UFC.