w a t e r r e s e a r c h 4 6 ( 2 0 1 2 ) 3 9 3 4 e3 9 4 2

Available online at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Comparison of drinking water treatment process streams for optimal bacteriological water quality
Lionel Ho*, Kalan Braun, Rolando Fabris, Daniel Hoefel, Jim Morran, Paul Monis, Mary Drikas
Australian Water Quality Centre, SA Water Corporation, 250 Victoria Square, Adelaide, SA 5000, Australia

article info
Article history: Received 19 December 2011 Received in revised form 23 April 2012 Accepted 25 April 2012 Available online 4 May 2012 Keywords: Denaturing gradient gel electrophoresis (DGGE) Flow cytometry Heterotrophic plate count (HPC) Magnetic ion exchange (MIEX) Photometric dispersion analyser (PDA) Water treatment

abstract
Four pilot-scale treatment process streams (Stream 1 e Conventional treatment (coagulation/occulation/dual media ltration); Stream 2 e Magnetic ion exchange (MIEX)/ Conventional treatment; Stream 3 e MIEX/Conventional treatment/granular activated carbon (GAC) ltration; Stream 4 e Microltration/nanoltration) were commissioned to compare their effectiveness in producing high quality potable water prior to disinfection. Despite receiving highly variable source water quality throughout the investigation, each stream consistently reduced colour and turbidity to below Australian Drinking Water Guideline levels, with the exception of Stream 1 which was difcult to manage due to the reactive nature of coagulation control. Of particular interest was the bacteriological quality of the treated waters where ow cytometry was shown to be the superior monitoring tool in comparison to the traditional heterotrophic plate count method. Based on removal of total and active bacteria, the treatment process streams were ranked in the order: Stream 4 (average log removal of 2.7) > Stream 2 (average log removal of 2.3) > Stream 3 (average log removal of 1.5) > Stream 1 (average log removal of 1.0). The lower removals in Stream 3 were attributed to bacteria detaching from the GAC lter. Bacterial community analysis revealed that the treatments affected the bacteria present, with the communities in streams incorporating conventional treatment clustering with each other, while the community composition of Stream 4 was very different to those of Streams 1, 2 and 3. MIEX treatment was shown to enhance removal of bacteria due to more efcient occulation which was validated through the novel application of the photometric dispersion analyser. 2012 Elsevier Ltd. All rights reserved.

1.

Introduction

The primary goal of water utilities is to safeguard drinking water for consumers. Consequently, drinking water must be of a standard or quality that aligns with many water safety plans. This involves removing contaminants of concern, whether they be biological or chemical, and a range of water treatment methods have been developed over the past century to ensure that these contaminants are removed or

minimised in drinking water distribution systems. The efcacy of these treatment methods is governed by routine monitoring of specic indicators, including the removal of pathogenic organisms and chemicals of concern (e.g. disinfection by-products, algal toxins, etc.). Surrogate parameters are generally used to assess the efcacy of treatment processes. For example, monitoring of natural organic material (NOM), in particular, dissolved organic carbon (DOC), colour and UV absorbance, can be used

* Corresponding author. Tel.: 61 8 7424 2119; fax: 61 8 7003 2119. E-mail address: lionel.ho@sawater.com.au (L. Ho). 0043-1354/$ e see front matter 2012 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2012.04.041

w a t e r r e s e a r c h 4 6 ( 2 0 1 2 ) 3 9 3 4 e3 9 4 2

3935

to assess the precursors of disinfection by-products. Likewise, the general bacteriological quality of drinking water can be monitored using heterotrophic plate counts (HPC), a method which has been in use for over a century (Bartram et al., 2003; Allen et al., 2004; Berney et al., 2008). In recent times, new detection methods have emerged to evaluate bacteriological quality in water, including measuring adenosine triphosphate (ATP) and ow cytometry (FCM) in conjunction with uorescence staining methods (Hoefel et al., 2003; Hammes et al., 2008; Siebel et al., 2008). These detection methods offer numerous advantages over the HPC method as they are not only rapid, accurate and enable high throughput, but they can also detect bacteria which are non-culturable under the conditions of the HPC method. While studies have utilised FCM with uorescence stains to characterise bacterial removal through conventional water treatment and distribution systems (Lebaron et al., 1998; Rinta-Kanto et al., 2004; Hoefel et al., 2005; Hammes et al., 2008), few studies, to date, have utilised such an approach to compare various treatment processes in parallel to assess their ability to remove bacteria. With many water utilities commissioning water treatment plants (WTP) that employ new technologies such as membrane ltration and/or ion exchange resins (in addition to utilities retrotting or upgrading their existing plants), there is a requirement to validate specic treatments for their bacterial removal capacity. This can ensure that they adopt the multi-barrier treatment approach to comply with water safety plan guidelines and water quality targets. Such validation studies will facilitate the design of specic treatment processes for utilities, in addition to optimisation and best management practices of these processes. The aim of this study was to evaluate and compare the quality of potable water produced from four different water treatment processes in parallel, prior to nal disinfection. Moreover, a major emphasis of this study was to characterise the bacteriological quality of the product waters from the various treatments as this can play an important role in distribution systems including the formation of biolms within such systems.

Fig. 1 e Schematic of the four treatment streams: S1 e Conventional treatment (coagulation/occulation/dual media ltration); S2 e MIEX/Conventional treatment; S3 e MIEX/Conventional treatment/GAC; S4 e Microltration/ nanoltration.

2.
2.1.

Experimental procedures
Treatment processes

Four different water treatment process streams (full- and pilot-scale) were designed and/or adapted to generate waters of varied water quality from the same source water (see Fig. 1). The feed water for the streams was supplied from the inlet to the Mt. Pleasant WTP in South Australia. This water is sourced from the River Murray via the Mannum to Adelaide pipeline. The treatment streams were evaluated from June 2010 to June 2011. Each stream was designed to generate a product ow rate of 250 L h1. Details of the treatment streams are described below:

ltration, utilising an upow clarier and a gravity fed perspex lter column. The coagulant employed was aluminium sulphate (alum) as Al2(SO4)3.18H2O. The alum dose ranged from 20 to 160 mg L1. Coagulation pH of between 6.0 and 6.5 was maintained through addition of sodium hydroxide or sodium bicarbonate buffering, depending on source water alkalinity. In addition, a coagulant aid, either anionic polyacrylamide (LT20, BASF Chemicals, Australia) or high molecular weight poly-DADMAC (LT425, BASF Chemicals, Australia) was also dosed downstream of the coagulant. This process was selected as a baseline/control as it represents the most widely applied drinking water treatment process employed in Australia.

2.1.2.

Stream 2 e MIEX/conventional treatment

2.1.1.

Stream 1 e conventional treatment

This pilot-scale conventional treatment stream comprised of coagulation/occulation/dual media (sand/anthracite)

Treated water from this process stream was sourced directly from the full-scale WTP at Mt. Pleasant. Full details of this WTP have been described previously (Drikas et al., 2011). Briey, treatment comprised of high rate magnetic ion exchange contact (MIEX DOC process, Orica, Australia) for DOC removal coupled with coagulation/occulation/dual media (sand/anthracite) ltration. The average MIEX resin

3936

w a t e r r e s e a r c h 4 6 ( 2 0 1 2 ) 3 9 3 4 e3 9 4 2

dose applied during the study was 15 L kL1. A continuous stirred-tank reactor with a cone settler operating at 10% resin regeneration was employed for the MIEX DOC process. The primary coagulant used was alum as Al2(SO4)3.18H2O; however, additional coagulant aids, LT22 and LT425 (BASF Chemicals, Australia) were also dosed periodically during coagulation as required. Due to the ability of the MIEX DOC process to efciently remove absorbable organic materials, the subsequent coagulation treatment is primarily a clarication step following the main organic carbon removal by the MIEX resin. As such, the coagulant demand is reduced leading to a lower and less variable alum dose range (10e80 mg L1).

cooled 15 mW argon ion laser, emitting at a xed wavelength of 488 nm. Fluorescent lters and detectors were all standard with green uorescence collected in the FL1 channel (530 30 nm), orange uorescence collected in the FL2 channel (585 42 nm) and red uorescence collected in the FL3 channel (>670 nm). Data were analysed using CellQuest software (Becton Dickinson, USA). Total numbers of bacteria were enumerated following staining of the bacteria with SYTO-9 and the BacLight bacterial viability kit (Molecular Probes, USA) as described previously (Hoefel et al., 2003). Results for FCM were presented as cells mL1.

2.4. 2.1.3. Stream 3 e MIEX/conventional treatment/GAC

The third treatment stream was comprised of the product water from Stream 2 (described above) with the addition of two parallel pilot-scale granular activated carbon (GAC) lters utilising F400 GAC (Calgon Carbon Corporation, USA). F400 is a bituminous coal-based GAC with effective granule size 0.55e0.75 mm which is commonly applied in water and wastewater applications for organic contaminant removal. Filtration was achieved using packed bed columns with gravity fed empty bed contact times (EBCT) of approximately 14 min at 125 L h1 for each column.

Bacterial community analysis

2.1.4.

Stream 4 e microltration/nanoltration

Dual pilot-scale membrane ltration consisted of microltration (MF) pre-treatment for particulate removal using a single submerged hollow bre module (Memcor CMF-S system, USA) followed by a single FILMTEC NF 270-4040 spiral wound nanoltration (NF) membrane (DOW Chemical Company, USA). The MF system was operated at 1000 L h1 with 75% permeate recovery. The NF system operated in cross-ow conguration at 43% permeate recovery, producing 325 L h1. Nominal pore size for the MF is reported as 0.2 mm with the molecular weight cut-off for the NF being 270 Da.

2.2.

Analyses

Colour measurements (at 456 nm) were made through a 5 cm quartz cell using an Evolution 60 Spectrophotometer (Thermo Scientic, USA) according to a published method (Bennett and Drikas, 1993). Results were presented in Hazen units (HU). Turbidity measurements were conducted on a 2100AN Laboratory Turbidimeter (Hach, USA) with results expressed in nephelometric turbidity units (NTU).

The effects of the different treatment processes on the bacteria in the raw water was assessed by proling the bacterial community composition of the raw water and product waters using denaturing gradient gel electrophoresis (DGGE) analyses. Water samples were analysed by FCM and bacterial numbers adjusted to 2.0 106 cells mL1, with the exception of treated water from Stream 4, which could only be concentrated to 5.0 105 cells mL1. Duplicate 1 mL samples from each water type were concentrated by centrifugation, resuspended in 5 mM TriseHCl pH 7.5 and subjected to three cycles of freeze-thawing (liquid N2 and 100  C). The resultant DNA was used as a template for universal 16S rDNA genedirected nested PCR using the primer sets 27F/1492R and 357F-GC/518R, and the products of the reaction analysed by DGGE (D-GENE Gel Electrophoresis System, Bio-Rad, USA) as reported previously (Hoefel et al., 2005). Positive and negative controls used in DGGE were as described by Hoefel et al. (2005). The resulting DGGE proles were analysed using Phoretix 1D version 11.2 (TotalLab, Newcastle upon Tyne, UK) with the following settings: lanes were identied automatically (lanes controlling controls were excluded from the analysis), background subtraction used the rolling ball method with a radius of 200, bands were manually called, Gaussian peaks were tted to bands using the advanced tting option with manual adjustment as required, bands were aligned using a synthetic reference generated by the software and similarity of proles was assessed using the UPGMA option.

2.5.

Flocculation index determination

2.3.

Bacterial enumeration

Bacterial enumeration was conducted using HPCs and FCM. HPCs were performed in accordance with the Australian Standard AS/NZS 4276.3.1 (Australian Standard, 1995) using R2A solid media (Oxoid, Australia). Dilutions, when necessary, were performed in maximum recovery buffer (0.1% (w/v) neutralised bacteriological peptone, 0.85% (w/v) NaCl, pH 7.0). Incubation was performed using standard conditions of 20  C for 72 h. Results for HPC were presented as colony forming units per mL (CFU mL1). FCM analyses were conducted using a FACSCalibur ow cytometer (Becton Dickinson, USA) equipped with an air-

The photometric dispersion analyser (PDA 2000, Rank Bros Ltd., Cambridge, UK), is a laboratory instrument used for analysis of owing suspensions (Gregory and Nelson, 1984, 1986). The method employed was similar to Staaks et al. (2011) with slight modications. Briey, the PDA was connected, via exible tubing, to one jar during jar testing. A peristaltic pump circulated the sample water at 21.6 mL min1. The pump was located after the PDA to avoid deterioration of the ocs. A volume of 1 mm3 of the owing suspension is illuminated by a narrow beam of light from a high intensity light emitting diode at 850 nm wavelength (Yukselen and Gregory, 2004). The intensity of transmitted light uctuates concurrently with the number of particles and is detected by a sensitive photodiode. The optical signal is converted to a voltage recorded by a computer equipped with a data logging system. The resultant PDA output is a graph of

w a t e r r e s e a r c h 4 6 ( 2 0 1 2 ) 3 9 3 4 e3 9 4 2

3937

the occulation index (FI) as a function of time. The FI is a relative value generated from a ratio of the root mean square (RMS) and direct current (DC) signals and has been used to compare and characterise occulation processes (Gregory and Nelson, 1984, 1986; Yukselen and Gregory, 2004; Staaks et al., 2011). In our study, three key parameters were extracted from the FI graphs: the initial oc aggregation (IFA), the relative settling factor (RSF) and the variance. The derivation of these parameters has been documented previously (Hopkins and Ducosto, 2003; Staaks et al., 2011). The relevance of these parameters will be discussed in the following sections.

3.

Results and discussion

3.1. Comparison of treatment streams for colour and turbidity reduction

During this study (June 2010eJune 2011) the inlet (raw) water to Mt. Pleasant WTP and subsequently the pilot-scale processes were challenged with water which was out of its usual specication; a consequence of two major water quality events brought about by large inows into the MurrayeDarling Basin from eastern Australia. These ood waters resulted in large spikes in turbidity with a maximum of approximately 190 NTU, followed by periods of high colour with values in excess of 100 HU. These events followed a period of extended drought where river inows were minimal and source water quality was relatively stable. From an operational standpoint, the monitoring of these two water quality parameters (turbidity and colour) are generally indicative of how well the treatment processes are performing; in addition to conforming to appropriate water quality standards and/or guideline levels. For example, turbidity has been used as a surrogate for parasites such as Cryptosporidium and Giardia, while colour is generally regarded

as an aesthetic parameter which can also be used as a surrogate for organic matter. To put things into perspective, the Australian Drinking Water Guideline levels for turbidity and colour are 0.5 NTU and 15 HU, respectively. Despite these signicant water quality challenges, the pilot-scale treatment processes were generally efcient in reducing both the colour and turbidity as shown in Figs. 2 and 3, respectively. For example, colour reduction was consistently high, especially for the advanced multi-stage processes (Streams 3 and 4) which averaged greater than 98% reduction over the period. Some difculty was encountered in maintaining optimum coagulation conditions throughout the changing water quality periods, especially when rapid changes occurred, and this is reected in the poorer removals in colour and turbidity by conventional treatment (Stream 1). This was in part due to the reactive nature of coagulation control where decline of treated water quality dictated the operational changes. During these periods, additional chemicals (including the coagulant aids) were dosed to maintain target pH and oc settleability for acceptable lter run times but only after water quality showed deterioration, resulting in the largest span between maximum and minimum reduction percentages of all the treatments.

3.2. Comparison of treatment streams for removal of bacteria

The bacteriological quality of the four treated waters was evaluated using both HPCs and FCM. Results for HPC showed no clear trends between each of the treatment streams, suggesting that each of the treatment processes were equally effective in removing bacteria (Fig. 4). Furthermore, large uctuations in bacterial numbers in the treated waters were evident with a numbers ranging from 2 CFU mL1 up to w7 103 CFU mL1. In contrast to the HPC data, FCM analyses of the treated waters showed more denitive and stable trends between

Fig. 2 e Colour measurements before (raw) and after the four treatment processes. Dashed line represents Australian Drinking Water Guideline level of 15 HU.

3938

w a t e r r e s e a r c h 4 6 ( 2 0 1 2 ) 3 9 3 4 e3 9 4 2

Fig. 3 e Turbidity measurements before (raw) and after the four treatment processes. Dashed line represents Australian Drinking Water Guideline level of 0.5 NTU.

each of the treatment processes, as shown in Fig. 5. This highlights the shortcomings of utilising HPCs for monitoring bacteriological quality, a nding supported by others (Hammes and Egli, 2005; Berney et al., 2008; Hammes et al., 2008; Siebel et al., 2008). Many of the authors ascribe the deciency of HPCs to human error. For example, the statistical accuracy of the plating method is dependent upon colonies being counted between 30 and 300 per plate, and this is dependent upon the appropriate dilution factor. Hammes et al. (2008) documented that the standard error of HPC results was >30% compared with FCM results which were <5%. Another deciency and perhaps the biggest drawback of the HPC method is its selectivity as it is unable to enumerate viable, non-culturable bacteria, which explains why HPC

results are on average two orders of magnitude lower than bacterial enumeration by FCM (Siebel et al., 2008). This in part is due to the nutrient concentrations on conventional HPC agar plates which can be between 800 and 1000 times higher than the concentrations detected in drinking water (Berney et al., 2008; Hammes et al., 2008). The large discrepancy between HPC and FCM results has led some to suggest for a reconsideration of existing drinking water guidelines and legislation (Berney et al., 2008). The raw water total bacterial count averaged 1.8 107 cells mL1 (minimum 8.5 106 cells mL1, maximum 3.2 107 cells mL1) during the study period, of which 55% were shown to be active, as determined by FCM. This number is relatively high in comparison to other water

Fig. 4 e Heterotrophic plate counts (HPC) before (raw) and after the four treatment processes.

w a t e r r e s e a r c h 4 6 ( 2 0 1 2 ) 3 9 3 4 e3 9 4 2

3939

Fig. 5 e Bacterial enumeration by ow cytometry (FCM) before (raw) and after the four treatment processes.

sources and may be attributed to the water being sourced from the River Murray via the Mannum to Adelaide pipeline. The residence time in this non-disinfected pipeline is between 2 and 3 d prior to the Mt. Pleasant WTP, which subjects the pipeline to sloughing of biolm and consequently higher numbers of bacteria entering the WTP. The order of effectiveness of the processes based on removal of total and active bacteria followed the trend: Stream 4 > Stream 2 > Stream 3 > Stream 1 (see Table 1). As expected, the advanced multi-stage process of MF/NF was the superior treatment stream due to its size-exclusion nature (2.7-log removal of both total and active bacteria). However, an cells mL1 average number of 4.5 104 3 (minimum 8.3 10 cells mL1, 5 1 maximum 2.0 10 cells mL ) was still detected in the NF treated water, even though the nominal molecular weight cutoff of the membrane is quoted as 270 Da; approximately 100e10,000 times smaller than bacterial cells (between 0.5 and 10 mm in size). The limit of detection of the FCM method in this study is 5.0 103 cells mL1 (unpublished work), suggesting that either some bacteria were breaking through the membrane or that there was possibly some form of

contamination or re-growth after the membrane during sampling. The latter is possible since the sampling point for the NF efuent is located on a stainless steel pipe approximately 2 m after the NF module. Comparison of the bacterial diversity in the raw and treated waters by DGGE (Fig. 6) showed that Streams 1, 2 and 3 had similar proles to the raw water, with some minor band differences between these samples and a dominant band apparent in the treated samples and not detected in the raw water. However, the prole from Stream 4 was noticeably different to the raw water or the other treated waters, with only a few bands in common (Fig. 6A), suggesting that this community is different to the communities in the other samples. This result suggests that either Stream 4 treatment was allowing particular bacterial species to breakthrough (that are not dominant in the raw water and consequently not detected), or that bacteria colonised the pipe post-NF and these were being detected in the Stream 4 sample. Considering that there were a few bands in common between Stream 4 and the raw water or other streams, a combination of some breakthrough of bacteria from the raw water and biolm from the post-NF pipe would also be consistent with this result.

Table 1 e Average bacterial numbers (total and active) in the efuent of the treatment processes and log removal values of bacteria (total and active) by each of the treatment processes from July 2010 to June 2011. Treatment process
Stream 1 conventional treatment Stream 2 MIEX/conventional treatment Stream 3 MIEX/conventional treatment/GAC Stream 4 MF/NF

Average total numbers in efuent (cells mL1)

2.7 106 1.5 105 6.1 105 4.5 104

Average active numbers in efuent (cells mL1)

1.6 106 7.7 104 5.1 105 2.5 104

Log removal (total)

1.0 0.3 2.3 0.4 1.5 0.2 2.7 0.4

Log removal (active)

0.9 0.3 2.3 0.3 1.3 0.2 2.7 0.3

3940

w a t e r r e s e a r c h 4 6 ( 2 0 1 2 ) 3 9 3 4 e3 9 4 2

Fig. 6 e Analysis of bacterial communities present in raw and treated waters by denaturing gradient gel electrophoresis of the V3 region of 16S rDNA. (A) Raw results: ntc [ no template PCR control, showing contribution of background bacterial DNA in reagents; pos [ positive PCR control using genomic DNA from Escherichia coli, Aeromonas hydrophila and Staphylococcus epidermidis; Raw [ raw water sample; S1eS4 [ Streams 1e4. (B) Dendrogram showing similarity of the bacterial communities inferred by analysis of the banding patterns using Phoretix 1D software.

The relationships of the communities were determined using Phoretix 1D software, which incorporates the presence/ absence of bands and also the relative band intensity to calculate the relative similarity for each pair-wise combination of samples (where 1 indicates the samples are 100% identical, 0.5 indicates 50% similarity, etc). Cluster analysis of the sample similarity matrix resulted in the dendrogram shown in Fig. 6B. The clustering pattern suggests that the communities in Streams 2 and 3 were approximately 70% similar to each other (with the same 4 dominant bands present), with the Stream 1 community approximately 60% similar to these (with the same 4 dominant bands as well as some additional dominant bands present). The raw water community was approximately 50% similar to the communities in Streams 1, 2 and 3, with most of the difference attributable to the presence/absence of minor bands. If the dominant bands were considered in isolation, these samples would be 70e80% similar. Stream 4 community was only 35% similar to the rest of the samples, with only 2 of 6 dominant bands in common with any of the other samples. These results support the qualitative observation that Stream 4 microbial community is substantially different to the communities in the raw water and other treatment streams. Furthermore, the pattern of clustering of the treated water communities correlated with the treatments. Both Streams 2 and 3 incorporated MIEX treatment (with the addition of GAC for Stream 3) and these were the most similar communities.

Stream 1 only included conventional treatment, and possessed a community that was the most similar to the communities in Streams 2 and 3. The treatment for Stream 4 relied solely on membrane ltration (MF/NF), and this stream had the lowest numbers of bacteria in the product water and also the most different community as assessed by DGGE analysis of the V3 region of 16S rDNA. While Lovins et al. (2002) demonstrated excellent rejection of organisms (including bacteria) using three different NF membranes (with molecular weight cut-off values from 100 to 300 Da), the authors still found that organisms did pass through the membranes, supporting the contention of bacterial breakthrough. Furthermore, Liikanen et al. (2003) and Park and Hu (2010) observed growth of bacteria in NF and reverse osmosis (RO) permeates, with bacterial numbers of between 1.2 103 and 2.1 105 cells mL1 detected, similar to the numbers in our study. This is thought to be due to the RO permeate creating more conducive conditions for bacterial growth, where more assimilable low molecular weight organics would pass through the membrane (Drewes et al., 2003; Park and Hu, 2010). It is worth bearing in mind that these are not sterile closed systems, so even in the absence of bacteria breaking through the membrane, any bacteria present in the post treatment pipes could colonise the system provided sufcient nutrients were present. The relatively higher numbers of culturable bacteria (as determined by HPC, see Fig. 4) in the NF permeate supports this contention; a consequence of the lower community diversity in the NF permeate (Park and Hu, 2010). Stream 3 was designed to be the second most effective advanced multi-stage process, based on the addition of a GAC lter. However, this did not translate to the second best treatment option in terms of the bacterial removal where an average number of 6.1 105 cells mL1 was detected in the efuent, approximately 4 times higher than Stream 2. This strongly suggests that the GAC lter contributed to the higher numbers. Stewart et al. (1990) documented that carbon particles (nes) could be detected in the efuent of GAC lters and that these nes were colonised with large numbers of bacterial cells (several thousand), lending support to this contention. Similarly, Velten et al. (2011) documented detachment of high numbers of bacteria from a GAC lter with numbers of w2.5 105 cells mL1 detected in the efuent, the same order of magnitude as in our study. The authors determined that the bacteria in the efuent represented 84% of the total bacteria colonised in the GAC lter during steady state. This is not surprising as GAC biolms are considered nutrient poor environments (Velten et al., 2011) and such conditions have been documented to decrease bacterial adhesion to porous media due to the greater production of extracellular polymeric substances (EPS) which causes the cells to be more hydrophobic (Haznedaroglu et al., 2008). Interestingly, the community in Stream 3 was very similar to the community in Stream 2, suggesting that any bacteria detaching from the GAC must be similar to the bacteria in the raw water, or the bacteria detaching are not very diverse because there are few unique bands present in the DGGE analysis that are only associated with Stream 3. The difference between Stream 2 and Stream 1 was the addition of MIEX pre-treatment prior to coagulation in Stream

w a t e r r e s e a r c h 4 6 ( 2 0 1 2 ) 3 9 3 4 e3 9 4 2

3941

Table 2 e The initial oc aggregation (IFA), relative settling factor (RSF) and variance values derived from the photometric dispersion analyser (PDA) from laboratory coagulation experiments simulating Streams 1 and 2. Treatment process
Stream 1 conventional treatment Stream 2 MIEX/conventional treatment

IFA
0.28 0.07

RSF
0.58 0.07

Variance
0.002 0.001

0.37 0.07

0.71 0.06

0.021 0.007

2. The addition of MIEX treatment not only enhances removal of organics (see colour removal in Fig. 2 and also previous studies by Singer and Bilyk (2002), Jarvis et al. (2008) and Drikas et al. (2011)), but it also appeared to achieve better removal of bacteria, where the log removal values for Stream 2 were more than double than that of Stream 1 (see Fig. 5 and Table 1). The mechanisms of organics removal by MIEX treatment are well documented by the aforementioned studies; however, its ability to enhance removal of bacteria has yet to be established and/or published. It is hypothesised that MIEX treatment in Stream 2 may have resulted in more efcient coagulation through more compact ocs which were able to entrap bacteria within their structure, resulting in more efcient removals of bacteria than the processes employed in Stream 1. In order to validate this hypothesis, additional experiments were conducted using a PDA instrument to characterise the ocs generated through laboratory simulation of Streams 1 and 2. Three key parameters were derived from these experiments, the IFA, RSF and variance. The IFA has been used to describe the growth rate of the ocs; the RSF has been used to represent the settling of the ocs; while the variance has been used to assess oc structural differences (both size and distribution) (Hopkins and Ducosto, 2003; Staaks et al., 2011). Table 2 shows results from the simulation of Streams 1 and 2. Stream 2 had higher values for each of the parameters compared with Stream 1. The higher IFA for Stream 2 indicates that the rate of oc formation is greater than for Stream

1. Furthermore, the higher RSF in Stream 2 is indicative of better oc settling performance. Finally, and perhaps most relevant is the variance, where a higher value indicates not only larger ocs (with a larger range of oc sizes) but also a stronger more compact oc (Hopkins and Ducosto, 2003; Staaks et al., 2011). These results corroborate the previous contention of more efcient coagulation and enhanced bacterial removal with the incorporation of MIEX treatment. Additional supporting evidence is displayed in Fig. 7, where samples taken after the laboratory simulation of Streams 1 and 2 show lower bacterial numbers (total and active) after Stream 2 treatment. Each of the treatment processes generally removed active bacterial cells equally to that of the total bacterial cells with log removals ranging from 0.9 0.3 (Stream 1) to 2.7 0.4 (Stream 4) (Table 1). Interestingly, the percentage of the active bacterial numbers was between 50 and 60% of the total bacterial numbers in the efuents of the treatment processes, except for Stream 3 where the percentage was considerably higher at 84%. This supports the previous contention of bacterial detachment from the GAC lter and that a majority were active due to their ability to produce EPS, a physiological mechanism thought to resist stressful (oligotrophic) conditions (Haznedaroglu et al., 2008). Such EPS-producing bacteria could potentially result in greater biolm formation in downstream distribution systems.

4.

Summary and conclusions

Despite signicant water quality challenges, the four pilotscale treatment process streams employed were able to effectively reduce colour and turbidity to below ADWG levels, with the exception of Stream 1 which periodically struggled to comply with the turbidity target; a consequence of the reactive nature of coagulation control where the decline of treated water quality dictated the operational changes. In terms of the bacterial enumeration, FCM was shown to be a better monitoring tool than HPCs, which allowed for more denitive comparisons to be made between each of the treatment streams. This suggests that FCM can be used to monitor water quality during treatment and distribution and could be useful in facilitating the design and optimisation of specic treatment processes. Based on removal of total and active bacteria, the treatment process streams were ranked in the order: Stream 4 (MF/ NF) > Stream 2 (MIEX/Conventional treatment) > Stream 3 (MIEX/Conventional treatment/GAC) > Stream 1 (Conventional treatment). Some of the interesting observations included:  detection of bacteria in NF efuent with an average number of 4.5 104 cells mL1;  demonstration that the bacterial community in the NF efuent (Stream 4) was very different to the communities present in the other treated water streams;  detachment of bacteria from GAC with an average total number of 6.1 105 cells mL1 detected in the lter efuent, of which 84% were shown to be active (in comparison with the other processes which ranged between 50 and 60%);

Fig. 7 e Numbers of bacteria (total and active, as determined by ow cytometry) sampled after laboratory simulation of Streams 1 and 2 (using photometric dispersion analyser).

3942

w a t e r r e s e a r c h 4 6 ( 2 0 1 2 ) 3 9 3 4 e3 9 4 2

 DGGE analysis identied only a few novel amplicons (representing 1 or 2 bacterial species) associated with the GAC treated stream;  overall, the community analysis suggested that the treatments affected the bacteria present, with the communities in streams incorporating conventional treatment clustering with each other, and the streams with MIEX being the most similar of the communities compared;  verication that MIEX treatment enhanced removal of bacteria through more efcient coagulation by the novel application of the PDA (eg. greater rate of oc formation, better oc settling performance and larger, more compact and stronger ocs). Negligible differences were observed between the removal of active bacteria cells compared with total bacteria cells by the treatment processes with log removals ranging from 0.9 0.3 to 2.7 0.4.

Acknowledgements
This project was supported by Water Quality Research Australia, South Australian Water Corporation, United Water International, Grampians Wimmera Mallee Water, Water Corporation, Delft University of Technology, DCM Process Control and Orica Watercare. The assistance of Jasper Verberk, Paul Colby, Renae Phillips and Nic Reid are duly acknowledged.

references

Allen, M.J., Edberg, S.C., Reasoner, D.J., 2004. Heterotrophic plate count bacteria: what is their signicance in drinking water? International Journal of Food Microbiology 92, 265e274. Australian Standard, 1995. Water Microbiology. Heterotrophic Colony Count Methods-Pour Plate Method Using Plate Count Agar. Standards Australia Int., Stratheld, New South Wales, Australia. AS/NZS 4276.3.1. Bartram, J., Cotruvo, J., Exner, M., Fricker, C., Glasmacher, A., 2003. Heterotrophic Plate Counts and Drinking-water Safety. IWA Publishing on behalf of the World Health Organization, London, UK. Bennett, L.E., Drikas, M., 1993. The evaluation of colour in natural waters. Water Research 27, 1209e1218. Berney, M., Vital, M., Hulshoff, I., Weilenmann, H.-U., Egli, T., Hammes, F., 2008. Rapid, cultivation-independent assessment of microbial viability in drinking water. Water Research 42, 4010e4018. Drewes, J.E., Reinhard, M., Fox, P., 2003. Comparing microltration-reverse osmosis and soil-aquifer treatment for indirect potable reuse of water. Water Research 37, 3612e3621. Drikas, M., Dixon, M., Morran, J., 2011. Long term case study of MIEX pre-treatment in drinking water; understanding NOM removal. Water Research 45, 1539e1548. Gregory, J., Nelson, D.W., 1984. A new optical method for occulation monitoring. In: SolideLiquid Separation. Ellis Horwood, Chichester, UK, pp. 172e182. Gregory, J., Nelson, D.W., 1986. Monitoring of aggregates in owing suspensions. Colloid Surface 18, 175e185.

Hammes, F.A., Egli, T., 2005. New method for assimilable organic carbon determination using ow-cytometric enumeration and a natural microbial consortium as inoculum. Environmental Science and Technology 39, 3289e3294. Hammes, F., Berney, M., Wang, Y., Vital, M., Koster, O., Egli, T., 2008. Flow-cytometric total bacterial cell counts as a descriptive microbiological parameter for drinking water treatment processes. Water Research 42, 269e277. Haznedaroglu, B.Z., Bolster, C.H., Walker, S.L., 2008. The role of starvation on Escherichia coli adhesion and transport in saturated porous media. Water Research 42, 1547e1554. Hoefel, D., Grooby, W.L., Monis, P.T., Andrews, S., Saint, C.P., 2003. Enumeration of water-borne bacteria using viability assays and ow cytometry: a comparison to culture-based techniques. Journal of Microbiological Methods 55, 585e597. Hoefel, D., Monis, P.T., Grooby, W.L., Andrews, S., Saint, C.P., 2005. Proling bacterial survival through a water treatment process and subsequent distribution system. Journal of Applied Microbiology 99, 175e186. Hopkins, D.C., Ducosto, J.J., 2003. Characterizing occulation under heterogeneous turbulence. Journal of Colloid and Interface Science 264, 184e194. Jarvis, P., Mergen, M., Banks, J., McIntosh, B., Parson, S.A., Jefferson, B., 2008. Pilot scale comparison of enhanced coagulation with magnetic resin plus coagulation systems. Environmental Science and Technology 42, 1276e1282. Liikanen, R., Miettinen, I., Laukkanen, R., 2003. Selection of NF membrane to improve quality of chemically treated surface water. Water Research 37, 864e872. Lovins, W.A., Taylor, J.S., Hong, S.K., 2002. Micro-organism rejection by membrane systems. Environmental Engineering Science 19, 453e465. Lebaron, P., Parthuisot, N., Catala, P., 1998. Comparison of blue nucleic acid dyes for ow cytometric enumeration of bacteria in aquatic systems. Applied and Environmental Microbiology 64, 1725e1730. Park, S.-K., Hu, J.Y., 2010. Assessment of the extent of bacterial growth in reverse osmosis system for improving drinking water quality. Journal of Environmental Health Part A 45, 968e977. Rinta-Kanto, J.M., Lehtola, M.J., Vartiainen, T., Martikainen, P.J., 2004. Rapid enumeration of virus-like particles in drinking water samples using SYBR green I-staining. Water Research 38, 2614e2618. Siebel, E., Wang, Y., Egli, T., Hammes, F., 2008. Correlations between total cell concentration, total adenosine triphosphate concentration and heterotrophic plate counts during microbial monitoring of drinking water. Drinking Water Engineering and Science 1, 1e6. Singer, P.C., Bilyk, K., 2002. Enhanced coagulation using a magnetic ion exchange resin. Water Research 36, 4009e4022. Staaks, C., Fabris, R., Lowe, T., Chow, C.W.K., van Leeuwen, J.A., Drikas, M., 2011. Coagulation assessment and optimisation with a photometric dispersion analyser and organic characterisation for natural organic matter removal performance. Chemical Engineering Journal 168, 629e634. Stewart, M.H., Wolfe, R.Y., Means, E.G., 1990. Assessment of the bacteriological activity associated with granular activated carbon treatment of drinking water. Applied and Environmental Microbiology 56, 3822e3829. Velten, S., Boller, M., Koster, O., Helbing, J., Weilenmann, H.-U., Hammes, F., 2011. Development of biomass in a drinking water granular active carbon (GAC) lter. Water Research 45, 6347e6354. Yukselen, M.A., Gregory, J., 2004. The reversibility of oc breakage. International Journal of Mineral Processing 73, 251e259.