The use of reverse osmosis (RO) membranes has been expanding not only to medical applications but also to water supply and reclaimed water applications due to its strong ability to remove a wide range of contaminants. Many researchers reported RO performance as a barrier against waterborne viruses; however, there are limited reports on its ability to remove bacteria from water. This investigation evaluated the removal performances of several spiral-wound RO modules and a hollow fiber ultrafiltration (UF) module in two different ways: dosing tests in batch-wise mode operation and in continuous-mode operation. The dosing tests of Escherichia coli using RO modules confirmed that E. coli could leak from the feed-side into the permeate. The log removal values (LRVs) (4.21- to >7.44-log10) by the RO modules from different production lots were found to vary greatly. In continuous-mode operation of the RO module, the LRVs for indigenous heterotrophic bacteria decreased over the operation period. These results clearly illustrate that bacteria, which originate on the feed-side, can leak into the permeate-side and then begin to proliferate in the permeate. On the other hand, using a UF module, E. coli was not detected in the permeate regardless of the operation mode.

  • The LRVs of bacteria by spiral-wound RO modules and a hollow fiber UF module were determined by E. coli dosing tests.

  • The LRVs by RO modules in different production lots varied in the range of 4.21- to >7.44-log10.

  • In continuous-mode operation of the RO module, the LRVs of indigenous HPC decreased over a period.

  • The UF module maintained extremely high LRVs for E. coli and indigenous HPC regardless of operational modes.

Graphical Abstract

Graphical Abstract
Graphical Abstract

The use of reverse osmosis (RO) membranes has been expanding to medical, desalination, water supply, and reclaimed water use applications due to its strong ability to remove a wide range of contaminants, micropollutant chemicals, odor compounds, and microorganisms. Since the 1970s, RO membranes have been used in the production processes of dialysis water and water for injection, because they were believed to reject not only soluble substances but also microbial substances and to be able to serve as a physical barrier. In the late 1970s, when a new dialysis treatment called online hemofiltration was introduced, researchers found that microorganisms were present in the dialysis water (Favero et al. 1975; Henderson & Beans 1978). Meanwhile, in the 1980s, the Japanese Pharmacopoeia and United States Pharmacopeia approved the production of water for injection using RO membranes. A high level of microbiological safety is required for the dialysate and water for injection, as they are directly injected into the patient's blood in large amounts, and especially dialysis water accounts for over 90% of the dialysate. However, as investigations of the microorganisms in the RO permeate have progressed, it has become widely known that bacteria are commonly present in the RO permeate (März et al. 1990), even though the RO membrane used for rejecting ions in water has a pore size smaller than the bacterial cell size. Based on these results, ultrafiltration (UF) membranes have been used in the final purification process to produce dialysate and water for injection with high microbiological quality (Ledebo 2002).

In addition to medical uses, the RO membrane has been applied to drinking water production and wastewater reclamation processes, because RO, including nanofiltration (NF), membranes could remove disinfection byproduct precursors and pathogens simultaneously (Kitis et al. 2003). Although lower ratios of water circulation were set in these systems compared with those in medical water production, bacteria have been still detected in the RO/NF permeate (Park & Hu 2010; Ohkouchi et al. 2014). The Surface Water Treatment Rule of 1989 in the USA requires microbial control at the level of 3-log10 and 4-log10 removal/inactivation for Giardia cysts and viruses, respectively. The adoption of these regulations stimulated research on the microbial removal performance of RO/NF membranes; however, only viruses, which are the smallest waterborne microorganisms, have been the focus of most of the research. Sorber et al. (1972) determined the virus removal performance for flat sheet-type RO membranes made of cellulose acetate using bacteriophage T2 and found that the removal rate for this virus was in the range from 99.2 to 99.999%.

Other researchers also confirmed that viruses could not be removed completely by spiral-wound RO modules due to defects in the module, such as pinholes in the membrane, or defects such as glue lines or cracks in the O-rings. For example, Adham et al. (1998a) evaluated the integrity of three different types of RO modules using both a vacuum test and a challenge test against viruses using bacteriophage MS-2 as an indicator. The log removal value (LRV) of MS-2 varied depending on the manufacturer of the RO modules in the range of 3.1- to over 5.9-log10, and these values were consistent with the results found by the direct method without any concentration. Adham et al. (1998b) also examined five different types of RO modules disassembled into flat sheet membranes to eliminate the effect of defects in the glue line or O-ring in the modules and found that the defects existed on the RO membrane itself. As it has become publicly known that RO modules are imperfect in their removal of viruses, various integrity tests for RO membranes have been proposed (Lozier et al. 2004). In those studies, a wide range of LRVs were also reported based on challenge tests using several combinations of surrogate indicators, such as MS-2, microspheres of 0.02 μm, and R-WT (rhodamine) with a molecular weight of 480. Integrity tests for RO and NF were also conducted using various configurations of membranes and scales, such as flat sheet membranes 2.5- or 4-inch modules at the bench scale and the pilot scale. LRVs for MS-2 of intact RO modules of 5- or 6-log10 were reported in these studies. In contrast, DeCarolis et al. (2005) found that the LRVs differed by manufacturer. The variation of LRVs also suggests that spiral-wound RO and NF modules might potentially have defects in the membrane, glue line, or O-ring. Kumar et al. (2007) also concluded that defects in the interconnector, O-ring, and glue line would be more predominant than defects in the membrane. Thus, there has been an ongoing discussion regarding which parts are responsible for the leaking of viruses. Such possible defects could lead to the leakage not only of viruses but also of larger microorganisms including bacteria.

Recently, several studies have evaluated the removal performance of pilot-scale RO or NF systems using online bacterial monitoring techniques (Buysschaert et al. 2018; Fujioka et al. 2018; Fujioka & Boivin 2020; Miller et al. 2020), and the LRVs of 1.67- to approximately 4-log10 have been observed. The LRVs, however, might not be ‘actual’ due to several reasons as follows: (1) the pretreated feed water using the microfiltration (MF) or UF membrane was used in these investigations assuming wastewater reuse in order to reduce microbial load against RO membranes, (2) in a full-scale RO treatment process, bacterial regrowth might have occurred in the RO permeate. Unlike viruses, which cannot multiply outside their hosts, bacteria are able to proliferate in water. Once bacterial cells leak into the permeate-side, the microbiological water quality could deteriorate rapidly. Bacteria, not only pathogens but also nonpathogenic bacteria, in the RO permeate can also pose various adverse health risks, because it is widely known that regrown bacteria can support the proliferation of other microorganisms such as free-living amoebae (Thomas & Ashbolt 2011). Information on the removal performances of RO or NF would be helpful for estimating the health risks caused by bacteria in the permeate, but unfortunately the information available is still insufficient. Information on the origins of bacteria detected in the RO permeate is also important when considering the measures to be taken, which can change depending on the origin of the bacteria.

In this investigation, the removal performances of several spiral-wound RO modules and a hollow fiber UF module were evaluated in two different ways: by dosing Escherichia coli as a bacterial indicator in a batch-wise mode operation and by monitoring indigenous heterotrophic bacteria in a continuous-mode operation. The effect of inlet pressure on the removal performances of RO modules was also evaluated. Finally, the fact that bacteria are detected in the RO permeate was discussed.

Equipment and tested membrane modules

Equipment consisting of a prefilter, an activated carbon fiber cartridge filter, a booster pump (SM21-51T: Sanwa Hydrotech Co., Osaka, Japan), a high pressure pump (CO 1604 XE: Nippon Oil Pump Co., Ltd, Saitama, Japan), a 30 L polyethylene tank (hereafter a tank), and an RO module or UF module (Figure 1) was used for evaluation. A cartridge filter (RDCPP-010-250: Central Filter Mfg. Co., Ltd, Tokyo, Japan) with a nominal pore diameter of 10 μm as a prefilter and an activated carbon fiber cartridge filter (DAC-250: Central Filter Mfg Co., Ltd, Tokyo, Japan) for quenching chlorine residual were used. The spiral-wound RO modules (SV021GV-DRA9810: Daicen Membrane Systems Ltd, Tokyo, Japan) had a membrane area of 0.6 m2 and showed 98% salt rejection. The hollow fiber-type UF module (FB02-FC-FUS1582: Daicen Membrane Systems Ltd, Tokyo, Japan) had a membrane area of 0.13 m2, and its molecular weight cut-off was 150,000 Da. These RO and UF membrane modules are miniaturized modules for testing to consider the installation of membrane equipment. There is no difference beside their size between miniaturized modules and commercial modules.

Figure 1

Flow diagrams applied to E. coli dosing tests in batch-wise mode and continuous-mode operation using RO modules (a, c) and a UF module (b, d), respectively.

Figure 1

Flow diagrams applied to E. coli dosing tests in batch-wise mode and continuous-mode operation using RO modules (a, c) and a UF module (b, d), respectively.

Close modal

The major properties of the tested modules are presented in Table 1.

Table 1

Properties of tested RO/UF modules

Module typeRO moduleUF module
SV021GV-DRA9810FB02-FC-FUS1582EXP
Module size φ77 mm × L445.6 mm φ60 mm × L356 mm 
Configuration Spiral-wound Hollow fiber 
Manufacture Daicen Membrane Systems 
Membrane materials Polyamide (composite) Polyethersulfone (asymmetric) 
Membrane area (m20.6 0.13 
Molecular weight cut-off (Da) – 150,000 
Salt rejection (%) 98.0 – 
Flux L/(m2·h) 18.0–25.0 67.7–112.3 
Module typeRO moduleUF module
SV021GV-DRA9810FB02-FC-FUS1582EXP
Module size φ77 mm × L445.6 mm φ60 mm × L356 mm 
Configuration Spiral-wound Hollow fiber 
Manufacture Daicen Membrane Systems 
Membrane materials Polyamide (composite) Polyethersulfone (asymmetric) 
Membrane area (m20.6 0.13 
Molecular weight cut-off (Da) – 150,000 
Salt rejection (%) 98.0 – 
Flux L/(m2·h) 18.0–25.0 67.7–112.3 

The equipment was set up in a laboratory at Azabu University. A high pressure pump was used only for the test using the RO modules, because the UF module could be operated at low pressure. Prior to the test, all of the equipment including the sampling tube was disinfected with a commercial disinfectant (Minncare Cold Sterilant: Mar Cor Purification, USA) composed of 4.5% peracetic acid and 22% hydrogen peroxide by the following procedure. Approximately 15 L of peracetic acid-based disinfectant prepared at 450 ppm was circulated by pump for three cycles, and the equipment, filled with the disinfectant, was left for 15–16 h. The next day, the disinfectant was circulated for one further cycle. After all disinfection procedures were completed, the remaining disinfectant in the equipment was thoroughly rinsed off with tap water. The concentration of hydrogen peroxide in the remaining disinfectant was checked using a pack test (WAK-H2O2: Kyoritsu Chemical-Check Lab., Corp., Tokyo, Japan). When the hydrogen peroxide concentration fell below 0.05 ppm, the rinse was considered completed.

Preparation of bacterial suspension for dosing tests

E. coli NBRC 3301 was purchased from the Biological Resource Center at the National Institute of Technology and Evaluation. E. coli was cultured in 2 L LB broth (Lennox, Thermo Fisher Scientific K. K., Tokyo, Japan) for 24 h at 37 °C and then harvested by centrifugation at 8,000 rpm for 10 min (Himac CR-21N, Koki Holdings Co., Ltd, Tokyo, Japan). The harvested bacterial pellets were rinsed twice with sterilized Milli-Q water and were resuspended in 200 mL Milli-Q sterilized water. After being well dispersed, the prepared E. coli suspension solution was used for the dosing tests as soon as possible.

Dosing tests of E. coli in batch-wise mode

The flow diagram for batch-wise mode experiments using RO and UF modules is shown in Figure 1(a) and 1(b), respectively. Four different RO modules and one UF module were evaluated. Twenty liters of Milli-Q water or tap water freshly collected in a tank were used as feed water. During the batch-wise mode experiments, all of the permeate was returned back to the tank through a soft tube with a diameter of 2 mm. Prior to the addition of the bacterial suspension, the tank water was circulated in the closed system for 30 min, and a blank test was carried out for each dosing test to confirm whether indigenous E. coli was undetected. The samples of feed water and permeate for E. coli enumeration were discharged via the soft tubes and directly collected into sterile bottles.

Dosing tests in batch-wise mode were designed according to ‘JIS K 3823: 2012 Test methods for determining bacterial rejection of ultrafiltration modules’ (2012). The dosing tests in batch-wise mode using four different RO modules or a UF module were repeated one to three times with intervals of 20 days or more. The fluxes for the RO modules and a UF module varied seasonally and were in the range of 18.0–25.0 and 67.7–112.3 L/(m2·h), respectively. The inlet pressure of the RO modules was set at 0.75 MPa, while that of the UF module was 0.08 MPa or less. Each dosing test was conducted at two different bacterial loadings to evaluate the influence of the concentration of E. coli in the feed water on LRV. At each load, two samplings were conducted with an interval of 30 min. First, 1/10 volumes of the freshly prepared E. coli suspension were added to Milli-Q water or tap water in the tank for the low loading test. After the two samplings were finished, the remaining prepared E. coli suspension was added to the tank for the higher loading test. The expected E. coli concentrations in feed water were around 106 and 107 CFU/mL for the lower and higher loading tests, respectively.

Evaluation of LRV in continuous-mode operation

Between the second and third batch-wise mode tests under different RO inlet pressure conditions, the RO module was provided for continuous-mode tests for 44 days to monitor the time-dependent changes in bacterial LRV of the modules using indigenous heterotrophic bacteria as a surrogate indicator. The flow diagram applied in this test is presented in Figure 1(c). In this test, tap water was supplied continuously as feed at an inlet pressure of 0.55 MPa. The heterotrophic plate count (HPC) in the feed water and permeate was monitored regularly during operation to calculate the LRV for the RO module. During the 30-day interval of the batch-wise mode operation of the UF module, the UF module was also tested in continuous-mode operation (Figure 1(d)). The HPC in the feed water and permeate of the UF module was also monitored in the same manner.

Bacterial enumeration and LRV calculation

The bacteria, E. coli or heterotrophic bacteria, in the feed or permeate water were enumerated using culture methods. In batch-wise mode tests, the serially diluted feed water samples were incubated using CHROMagar ECC medium (Kanto Chemical Co., Inc., Tokyo, Japan) for E. coli detection at 37 °C for 24 h, then the colonies that appeared blue were counted as E. coli. If necessary, the E. coli in the permeate samples (10–1,000 mL) were trapped on a sterilized 0.2 μm membrane filter (Toyo Roshi Kaisha Ltd, Tokyo, Japan) by vacuum filtration, and the filters placed onto the medium were incubated under the same conditions. In continuous-mode tests, heterotrophic bacteria in the feed water and permeate were incubated using R2A agar (Nihon Pharmaceutical Co., Ltd, Tokyo, Japan) at 20 °C for 7 days after dilution or concentration in the same manner described above, if necessary.

The LRVs of E. coli or heterotrophic bacteria by the RO or UF modules were calculated using the following equation:
where Cp and Cf are the bacterial concentrations (CFU/mL) in the permeate and feed water, respectively.

Dosing tests of E. coli in batch-wise mode

The dosing tests of E. coli at an RO inlet pressure of 0.75 MPa were performed using four RO modules from different production lots. All of the RO modules were brand new, and the experimental equipment including the RO modules was disinfected thoroughly just before each dosing test. Figure 2 shows the E. coli concentrations of the feed water and permeate measured in each test. With a few exceptions, E. coli was detected at low concentrations in the permeate samples, though the E. coli concentration was below the detection limit (<10−3 CFU/mL) before the addition of the E. coli suspension. This result proves that the E. coli cells in those permeates leaked from the feed-side. In the low loading tests (Figure 2(a)), the E. coli concentration of the feed water was in the range of 3.7 × 105 to 6.4 × 106 CFU/mL, and the E. coli concentration in the permeate was in the range of 1.5 × 10−2 to 1.4 × 102 CFU/mL. On the other hand, in the high loading test (Figure 2(b)), the E. coli concentration of the feed water and permeate was in the range of 1.3 × 107 to 8.2 × 107 CFU/mL and <0.5 CFU/mL (not detected) to 1.9 × 103 CFU/mL, respectively.

Figure 2

E. coli concentration in feed water and RO permeate in dosing tests at RO inlet pressure of 0.75 MPa. The dosing tests were conducted using four RO modules from different production lots. Two different E. coli loadings (low (a); high (b)) were applied. The E. coli data represent the means with ranges of two dosing tests. During the intervals between the dosing tests, the RO modules were provided for continuous-mode operation through the feeding of tap water.

Figure 2

E. coli concentration in feed water and RO permeate in dosing tests at RO inlet pressure of 0.75 MPa. The dosing tests were conducted using four RO modules from different production lots. Two different E. coli loadings (low (a); high (b)) were applied. The E. coli data represent the means with ranges of two dosing tests. During the intervals between the dosing tests, the RO modules were provided for continuous-mode operation through the feeding of tap water.

Close modal

The calculated LRVs for the removal of E. coli by the RO modules are presented in Table 2. The initial LRVs for the modules under the new conditions at an RO inlet pressure of 0.75 MPa were 4.25-, >6.71-, and 5.73-log10 on average for RO module-2, module-3, and module-4, respectively. A recent study evaluated the bacterial LRVs for a spiral-wound RO system by real-time online bacterial monitoring, which is a detection method for viable bacterial cells (Fujioka et al. 2018). This study was conducted using a pilot-scale treatment system assuming a scene of water reclamation. Although the pretreated RO feed by MF/UF to reduce the bacterial load against RO membranes was applied, viable bacterial cells were still detected in the RO permeate at cell counts of 4–342, and the viable bacterial LRVs in the 68 h after starting the operation were in the range of 98.32–99.98% (i.e. 1.77- to 3.70-log10). Fujioka & Boivin (2020) have tested the passage of intact and damaged bacterial cells in RO feed to permeate, and LRVs slightly smaller than 2-log10 were confirmed. On the other hand, they have reported much greater LRVs (6.8- to 6.9-log10) using intact RO membranes obtained before assembly and bacterial-sized fluorescence particles at 3 × 107 in RO feed water as a surrogate in another study (Fujioka & Boivin 2019). Such a large variance in the LRVs obtained in several investigations suggests that the LRVs were greatly affected by both the bacterial load in RO feed and the lower detection limit of the applied surrogate assay in each investigation. In our dosing tests, relatively high bacterial loads of 106–108 CFU/mL in RO feed were applied as shown in Figure 1. Also, E. coli in the RO permeates was enumerated after enrichment using a membrane filtration method, if necessary. Thus, the difference in the bacterial load in RO feed or in the bacterial enumeration methods could have resulted in the greater LRVs obtained in our present work.

Table 2

The calculated initial LRVs for new RO modules determined by E. coli dosing tests

RO module no.Log removal values (log10)
Low E. coli loading 1st test/2nd test (average)High E. coli loading 1st test/2nd test (average)Overall average
3.98/4.44 (4.21) 4.29/4.29 (4.29) 4.25 
6.37/6.27 (6.32) >7.39/7.48 (>7.44) >6.71 
Not tested 5.59/5.86 (5.73) 5.73 
RO module no.Log removal values (log10)
Low E. coli loading 1st test/2nd test (average)High E. coli loading 1st test/2nd test (average)Overall average
3.98/4.44 (4.21) 4.29/4.29 (4.29) 4.25 
6.37/6.27 (6.32) >7.39/7.48 (>7.44) >6.71 
Not tested 5.59/5.86 (5.73) 5.73 

The modules were operated at inlet pressure of 0.75 MPa.

Fujioka & Boivin (2019) have also reported a wide variation of 1.6- to 6.1-log10 in the LRVs of pilot-scale RO filtration systems provided by different manufacturers using fluorescence particle (FL) as a surrogate, whereas their conductivity removal was stable at approximately 2-log10. Our results highlight the fact that the initial LRVs for the brand-new RO modules varied greatly even between the production lots, though there was no difference in the salt rejection performance of those modules, and all modules passed an integrity test by the manufacturer at the end of the operation. It has been considered that the rejection efficiency of soluble substances in water by RO/NF membranes is affected by various solute parameters such as molecular weight, molecular size, acid disassociation constant, hydrophobicity/hydrophilicity, diffusion coefficient, and others (Bellona et al. 2004). These parameters can contribute to the rejection mechanisms of steric hindrance or electrostatic exclusion. In our experiments, the observed variation of LRVs between different production lots might be mainly caused by a steric hindrance mechanism, because the chemistry of the solution and the membrane surface were found to be relatively consistent in a series of dosing tests. Unfortunately, the kinds of defects in the RO module that caused the variation, such as pinholes in the membrane, defects of the glue lines, or cracks in the O-ring sealings, were not identified in this investigation. A recent bench-scale study on bacterial infiltration through a disassembled RO membrane, however, provides us useful information on the passage routes of bacteria. It proved that FLs of 0.5 μm could pass through the membrane (Fujioka & Boivin 2020) using a unique approach. The authors also implied that the high permeability of FL through the disassembled RO membrane can occur as a result of damage generated by contacting the RO skin layer and feed spacer. The proposed concept can reasonably explain the variation of LRVs for each RO membrane module. Further investigation on the causes of the variation in the LRVs for RO modules is needed.

The calculated LRVs, except the initial LRVs by module-3, fluctuated slightly, but there were no significant differences within each test, despite the changing of the bacterial load in the feed water. If E. coli in the feed water was adsorbed or deposited in pores of the RO membrane within a few hours, the LRVs should rise consistently under high loading conditions. The unchanged LRVs observed in many modules at different bacterial loadings showed that there was no deposition of E. coli cells in pores within each dosing test. Module-3 under the new conditions revealed the highest E. coli removal performance among the tested modules as described above. This finding suggests that RO module-3 may have a tighter pore size distribution or smaller defects than the other modules. Only in module-3, the initial LRVs observed in the high loading test were clearly at least 1-log10 greater than those in the low loading test. The bacterial cells tended to aggregate under high bacterial loadings, even though the E. coli suspension was well dispersed by vigorous shaking. Therefore, the effect of aggregated E. coli cells in the feed water on LRVs might have appeared more clearly in module-3, which has a pore size distribution biased toward the smaller sizes or smaller defects, and then might have led to higher LRVs under only the high loading condition.

After each dosing test using E. coli as a surrogate, the equipment was operated continuously through the feeding of tap water during intervals between the dosing tests. The changes in the LRV by each RO module under different loadings of E. coli across the intervals are presented in Figure 3. The initial LRV by RO module-1 was not determined. The average LRVs by the other three modules tended to rise during the operation periods. Although the rates or patterns of the increase of LRVs were different between module-3 and module-4, both LRVs rose approximately 0.7-log10, on average of both bacterial loadings, after 56 and 68 days of operation, respectively. The increase over time in the LRVs obtained in the dosing tests performed after continuous-mode operation can be attributed to the deposition of rejected substances in the feed water onto the RO membrane, because chemical disinfection, but not cleaning, with peracetic acid was conducted prior to every dosing test in a series of tests using the same modules. No chemical cleaning of the modules with alkaline or acid reagent was conducted; therefore, it is conceivable that the deposited substances could not be removed by peracetic acid disinfection, while bacteria could be easily inactivated under the same conditions. Such a deposition of rejected substances in the membrane pores, pinholes, and defects in the glue lines might give a positive effect on the bacterial removal performance. Consequently, the above-mentioned difference in LRVs between different bacterial loadings in module-3 on Day 0 was offset by the newly generated effect of deposited substances after 20 days of operation.

Figure 3

Calculated LRVs in E. coli dosing tests in batch-wise mode. After three different RO modules were operated for various periods, the dosing tests were conducted using module-2 (at low loading (LL (open circle), at high loading (HL (closed circle)), module-3 (LL (open triangle), HL (closed triangle)), and module-4 (LL (open square), HL (closed square)). The upward arrow indicates that the LRV is greater than that value. Only module-4 was operated at an RO inlet pressure of 0.40 MPa for the first 21 days, and then at 0.55 MPa for the remaining period. All other modules were operated at an RO inlet pressure of 0.75 MPa. The LRV data represent the means with ranges of two dosing tests.

Figure 3

Calculated LRVs in E. coli dosing tests in batch-wise mode. After three different RO modules were operated for various periods, the dosing tests were conducted using module-2 (at low loading (LL (open circle), at high loading (HL (closed circle)), module-3 (LL (open triangle), HL (closed triangle)), and module-4 (LL (open square), HL (closed square)). The upward arrow indicates that the LRV is greater than that value. Only module-4 was operated at an RO inlet pressure of 0.40 MPa for the first 21 days, and then at 0.55 MPa for the remaining period. All other modules were operated at an RO inlet pressure of 0.75 MPa. The LRV data represent the means with ranges of two dosing tests.

Close modal

In order to compare the removal efficiency of the hollow fiber UF module with that of the spiral-wound RO module, dosing tests were conducted twice using one UF module. The tests were carried out with an interval of 30 days between them, using operation in continuous mode with the feeding of tap water. As shown in Table 3, the E. coli concentration in the feed water was in the range of 8.5 × 105–1.9 × 106 CFU/mL and 3.9 × 106–2.1 × 107 CFU/mL at low and high bacterial loadings, respectively. Unlike with the RO modules, E. coli was not detected (<1 CFU/2 L) in the UF permeate either before or after continuous operation. The calculated LRVs for the UF module were >9.42- (Day 0) and >9.31-log10 (Day 30) under lower bacteria loading, whereas those under higher bacterial loading were >10.5- (Day 0) and >10.1-log10 (Day 30). At the high bacterial loading, the higher E. coli concentration in the feed water may have made it possible to obtain the higher LRVs. Thus, the hollow fiber UF module, which does not have glue lines like spiral-wound RO modules in its structure, showed perfect rejection of E. coli in our tested range.

Table 3

Determined LRVs for the UF module by E. coli dosing tests

E. coli loading (elapsed time)E. coli concentration (CFU/mL)
LRVs (log10) Average
Feed waterUF permeate
Low loading (Day 0) 1.92 × 106 N.D. > 9.42 
9.15 × 105 N.D. 
Low loading (Day 30) 1.22 × 106 N.D. > 9.31 
8.50 × 105 N.D. 
High loading (Day 0) 2.05 × 107 N.D. > 10.5 
1.03 × 107 N.D. 
High loading (Day 30) 3.90 × 106 N.D. > 10.1 
1.04 × 107 N.D. 
E. coli loading (elapsed time)E. coli concentration (CFU/mL)
LRVs (log10) Average
Feed waterUF permeate
Low loading (Day 0) 1.92 × 106 N.D. > 9.42 
9.15 × 105 N.D. 
Low loading (Day 30) 1.22 × 106 N.D. > 9.31 
8.50 × 105 N.D. 
High loading (Day 0) 2.05 × 107 N.D. > 10.5 
1.03 × 107 N.D. 
High loading (Day 30) 3.90 × 106 N.D. > 10.1 
1.04 × 107 N.D. 

N.D. means that E. coli concentrations were below detection limit (<5 × 10−4 (CFU/mL)).

Influence of RO inlet pressure on E. coli removal

The influence of the RO inlet pressure on the LRV of E. coli was also investigated using RO module-4. The calculated LRVs at different inlet pressures are compared in Figure 4. At high E. coli loadings, the averaged LRVs at 0.30 and 0.75 MPa in the first dosing test were 5.86- and 5.73-log10, respectively, while those at 0.55 and 0.75 MPa in the second test were 5.99- and 5.91-log10, respectively. These results indicate that there was no difference between the LRVs at different RO inlet pressures on the same day, and, therefore, the RO inlet pressure was considered to not affect the LRVs by the RO module in the tested pressure range. Also, there was no clear difference between the LRVs at low and high bacterial loadings.

Figure 4

Effect of RO inlet pressure on the LRVs by RO module-4 in E. coli dosing tests. The loadings of E. coli in feed water were set at low (open bar) and high (closed bar). The data represent the means with the ranges of two dosing tests.

Figure 4

Effect of RO inlet pressure on the LRVs by RO module-4 in E. coli dosing tests. The loadings of E. coli in feed water were set at low (open bar) and high (closed bar). The data represent the means with the ranges of two dosing tests.

Close modal

On the other hand, a trend of increasing LRVs obtained at an RO inlet pressure of 0.75 MPa was observed as the operating period of RO module-4 was prolonged. For example, the average LRVs at high bacterial loadings at 0.75 MPa shifted from 5.73-log10 to 6.51-log10 during the operation. Of course, the shift of LRVs includes the effect of the interval operation at a different inlet pressure, but the increase in LRVs can be mainly attributed to a positive effect of deposited substances in the feed water as described in the previous section.

Evaluation of bacterial removal by RO/UF modules in continuous-mode operation

During the intervals between the dosing tests, the time-dependent changes in the removal efficiencies of the RO or UF modules were evaluated. The time-dependent change in HPC in the feed water and RO permeate during continuous-mode operation using RO module-4 at an inlet pressure of 0.55 MPa is shown in Figure 5. The HPC in the feed water drastically increased during the first 7 days, and then it decreased gradually over a month. The HPC in the RO permeate also increased within 7 days, and then it stabilized around 5 CFU/mL. The LRVs for the RO module-4 are also presented in Figure 5. The initial LRV could not be determined due to the HPC being too low in the permeate, but it can be estimated as >6-log10 from the results of the E. coli dosing tests. After the operation was started, the LRVs decreased gradually from 3.72-log10 on Day 7 to 2.86-log10 at the end of the operation. Miller et al. (2020) studied the LRVs of the NF/RO process in a full-scale treatment plant for wastewater reclamation and reported LRVs of total ATP (2.71- ±0.39-log10) that were greater than those of the total cell counts (1.67- ±0.40-log10) determined by flow cytometry in the same samples. The reported LRVs of ATP were very close to our value obtained at the end of the operation in the continuous mode. One of the reasons for the decrease in the LRV over time might be that the continuous-mode operation allows heterotrophic bacteria in the feed water and permeate to be able to proliferate. Our previous research shows that the growth rate of heterotrophic bacteria on the permeate-side is 4-fold greater than that on the feed-side when the RO system is operated in continuous mode. The difference in the growth rate can lead to a decrease of LRV (Ase & Ohkouchi 2020). The discrepancy in the LRV between the batch-wise mode test and the continuous-mode test suggests that the LRV of the HPC or ATP obtained by continuous-mode operation might indicate just an ‘apparent LRV,’ which must be vastly undervalued due to the presence of bacteria that can multiply on the permeate-side, whereas the LRV greater than 6-log10 obtained in the batch-wise mode dosing tests revealed a true physical barrier of this RO module. Thus, the bacterial regrowth is not negligible in continuous-mode operation.

Figure 5

Time-dependent changes in heterotrophic bacteria in the feed water (closed circle) and in the permeate (open circle). The calculated LRVs (gray bars) are also presented in the graph. The downward and upward arrows indicate an HPC below the detection limit (10−3 (CFU/mL)) and an LRV greater than that value, respectively.

Figure 5

Time-dependent changes in heterotrophic bacteria in the feed water (closed circle) and in the permeate (open circle). The calculated LRVs (gray bars) are also presented in the graph. The downward and upward arrows indicate an HPC below the detection limit (10−3 (CFU/mL)) and an LRV greater than that value, respectively.

Close modal

The UF module was also evaluated by measuring the HPC as a bacterial indicator in continuous-mode operation. In contrast to the results of the RO modules, heterotrophic bacteria were not detected in 10 L of the UF permeate even after 30 days operation as shown in Table 4. The much greater LRV of the hollow fiber UF obtained in a dosing test also supports this result. Our comparison using two different types of membrane modules again highlighted the superiority of the hollow fiber UF module in terms of its bacterial removal performance.

Table 4

Determined LRVs for the UF module by continuous-mode operation

Operation (days)HPC (CFU/mL)
LRVs (log10)
Feed waterUF permeate
Not tested Not tested Not tested 
10 N.D. > 4.80 
30 35 N.D. > 5.50 
Operation (days)HPC (CFU/mL)
LRVs (log10)
Feed waterUF permeate
Not tested Not tested Not tested 
10 N.D. > 4.80 
30 35 N.D. > 5.50 

N.D. means that HPCs were below detection limit (<1 × 10−4 (CFU/mL)).

As mentioned in the Introduction, the origin of the bacteria detected in the RO permeate has been the subject of an ongoing discussion. Our results of E. coli dosing tests surely proved that bacteria leaks from the feed-side into the RO permeate even though spiral-wound RO modules have LRVs of 4-log10 or greater. Also, even though under the conditions in which high LRVs were successfully maintained, such as immediately after disinfection of the RO modules, bacteria might leak from the feed-side if sufficient bacterial cells are loaded to the feed-side of the RO modules. For example, assuming a bacterial concentration of 100 CFU/mL in the feed water and a membrane flux of 20 L/(m2·h), the bacterial loading from the feed-side onto the RO membrane (0.6 m2) used in our research is estimated to be 6 × 106CFU per 5 h. By applying an LRV of 6-log10 as observed in our dosing tests, a bacteria loading of 6 CFU per 5 h could pass through the membrane probabilistically. This slight amount of leaked bacteria accumulates in the permeate lines and begins to proliferate over time, and finally, it could lead to the deterioration of the apparent LRV by the RO modules in continuous-mode operation. However, the mechanism by which the bacteria that leak through the RO module can survive and proliferate in the RO permeate under very low nutrient conditions within a short period has not been elucidated yet and should be further investigated.

This investigation evaluated the bacterial removal performances of spiral-wound RO modules to ensure control of bacterial contamination in the RO permeate. Four different new RO modules were tested by E. coli dosing tests in batch-wise mode operation, and low levels of E. coli were detected in almost all RO permeates. This result confirmed that some of the bacteria detected in permeate samples can leak from the feed-side by passing through the RO modules, but we did not identify the mechanism by which this occurs. RO modules from different production lots also exhibited a wide range of LRVs from 4.21- to >7.44-log10, though they exhibited the same salt rejection rate. These LRVs obtained by new RO modules represent a true physical barrier for bacteria. The LRVs tended to increase with the operating period of the RO module, but were not affected by RO inlet pressures in the range of 0.30–0.75 MPa. In contrast, a hollow fiber UF module exhibited extremely high LRVs of over 9-log10, and this was maintained even after 30 days of operation.

In continuous-mode operation at an RO inlet pressure of 0.55 MPa, the LRV by the module decreased over the operation period. This result strongly suggests the active proliferation of bacteria that leaked from the feed-side into the permeate-side of the RO module. On the other hand, no heterotrophic bacteria were detected in the UF permeate even after 30 days of the continuous-mode operation of the UF module, and the LRV by the module was maintained above 5.5-log10. The obtained results show that bacteria on the feed-side can leak into the permeate-side very quickly with the operation of the RO module and then begin to proliferate in the RO permeate.

The authors thank Mr Naoto Nishimura, Ms Aiko Wakimoto, Mr Kouki Kiyosawa, and Ms Miu Tanaka for their technical assistance in the E. coli dosing tests.

Data cannot be made publicly available; readers should contact the corresponding author for details.

Adham
S.
Gagliardo
P.
Smith
D.
Ross
D.
Gramith
K.
Trussell
R.
1998a
Monitoring the integrity of reverse osmosis membranes
.
Desalination
119
(
1
),
143
150
.
https://doi.org/10.1016/S0011-9164(98)00134-9
.
Adham
S. S.
Trussell
R. S.
Gagliardo
P. F.
Trussell
R. R.
1998b
Rejection of MS-2 virus by RO membranes
.
Journal of American Water Works Association
90
(
9
),
130
135
.
https://doi.org/10.1002/j.1551-8833.1998.tb08505.x
.
Ase
T.
Ohkouchi
Y.
2020
Time-dependent changes in bacterial count in permeate of spiral-wound RO modules and evaluation of factors related to their changes
.
Journal of Antibacterial and Antifungal Agents
48
(
3
),
101
110
.
Buysschaert
B.
Vermijs
L.
Naka
A.
Boon
N.
De Gusseme
B.
2018
Online flow cytometric monitoring of microbial water quality in a full-scale water treatment plant
.
Npj Clean Water
1
.
https://doi.org/10.1038/s41545-018-0017-7
DeCarolis
J.
Adham
S.
Kumar
M.
Pearce
B.
Wasserman
L.
2005
Integrity and performance evaluation of new generation desalination membranes during municipal wastewater reclamation
.
Proceedings of the Water Environment Federation
3518
3529
.
https://doi.org/10.2175/193864705783865776
.
Favero
M. S.
Petersen
N. J.
Carson
L. A.
Bond
W. W.
Hindman
S. H.
1975
Gram-negative water bacteria in hemodialysis systems
.
Health Laboratory Science
12
(
4
),
321
334
.
Fujioka
T.
Boivin
S.
2019
Assessing the passage of particles through polyamide reverse osmosis membranes
.
Separation and Purification Technology
226
,
8
12
.
https://doi.org/10.1016/j.seppur.2019.05.076
.
Fujioka
T.
Boivin
S.
2020
Assessing bacterial infiltration through reverse osmosis membrane
.
Environmental Technology & Innovation
19
,
100818
.
https://doi.org/10.1016/j.eti.2020.100818
.
Fujioka
T.
Hoang
A. T.
Aizawa
H.
Ashiba
H.
Fujimaki
M.
Leddy
M.
2018
Real-time online monitoring for assessing removal of bacteria by reverse osmosis
.
Environmental Science & Technology Letters
5
(
6
),
389
393
.
https://doi.org/10.1021/acs.estlett.8b00200
.
Henderson
L. W.
Beans
E.
1978
Successful production of sterile pyrogen-free electrolyte solution by ultrafiltration
.
Kidney International
14
(
5
),
522
525
.
https://doi.org/10.1038/ki.1978.157
.
Japanese Industrial Standards Committee
2012
JIS K 3823: 2012 Test Methods for Determining Bacterial Rejection of Ultrafiltration Modules
.
Japanese Industrial Standards Committee
,
Tokyo
.
Kitis
M.
Lozier
J. C.
Kim
J. H.
Mi
B.
Mariñas
B. J.
2003
Evaluation of biologic and non-biologic methods for assessing virus removal by and integrity of high pressure membrane systems
.
Water Supply
3
(
5–6
),
81
92
.
https://doi.org/10.2166/ws.2003.0153
.
Kumar
M.
Adham
S.
DeCarolis
J.
2007
Reverse osmosis integrity monitoring
.
Desalination
214
(
1
),
138
149
.
https://doi.org/10.1016/j.desal.2006.10.021
.
Ledebo
I.
2002
On-line preparation of solutions for dialysis: practical aspects versus safety and regulations
.
Journal of the American Society of Nephrology: JASN
13
(
Suppl. 1
),
S78
S83
.
Lozier
J.
Kitis
M.
Colvin
C.
Kim
J.
Mi
B.
2004
Microbial Removal And Integrity Monitoring Of High-Pressure Membranes
.
IWA Publishing
,
London
.
März
F.
Scheer
R.
Graf
E.
1990
Microbiological aspects in the production of water for injection by reverse osmosis
.
International Journal of Pharmaceutics
58
(
2
),
155
164
.
https://doi.org/10.1016/0378-5173(90)90253-Z
.
Miller
S. E.
Rodriguez
R. A.
Nelson
K. L.
2020
Removal and growth of microorganisms across treatment and simulated distribution at a pilot-scale direct potable reuse facility
.
Environmental Science: Water Research & Technology
.
https://doi.org/10.1039/C9EW01087D
Ohkouchi
Y.
Yata
Y.
Bun
R.
Itoh
S.
2014
Chlorine requirement for biologically stable drinking water after nanofiltration
.
Water Science & Technology: Water Supply
14
,
405
.
https://doi.org/10.2166/ws.2013.214
.
Park
S.
Hu
J. Y.
2010
Assessment of the extent of bacterial growth in reverse osmosis system for improving drinking water quality
.
Journal of Environmental Science and Health. Part A, Toxic/Hazardous Substances & Environmental Engineering
45
(
8
),
968
977
.
https://doi.org/10.1080/10934521003772386
.
Sorber
C. A.
Malina
J. F.
Sagik
B. P.
1972
Virus rejection by the reverse osmosis-ultrafiltration processes
.
Water Research
6
(
11
),
1377
1388
.
https://doi.org/10.1016/0043-1354(72)90200-X
.
Thomas
J. M.
Ashbolt
N. J.
2011
Do free-living amoebae in treated drinking water systems present an emerging health risk?
Environmental Science & Technology
45
(
3
),
860
869
.
https://doi.org/10.1021/es102876y
.

Author notes

Present address: Technology Management Division, Chuou Sekkei Engineering Co., Ltd, Yokohama Bashamichi Bldg.,4-55 Otamachi, Naka-ku, Yokohama, 231-0011 Kanagawa, Japan