Jorien Favere, Benjamin Buysschaert, Nico Boon, Bart De Gusseme
Abstract
Microbial regrowth during drinking water distribution can result in a variety of problems such as a deviating taste and odor, and may even pose a risk to public health. Frequent monitoring is essential to anticipate events of biological instability, and relevant microbial parameters for operational control of biostability of drinking water should be developed. Here, online flow cytometry and derived biological metrics were used to assess the biological stability of a full-scale drinking water tower during normal and disturbed flow regime. Pronounced operational events, such as switching from drinking water source,and seasonal changes, were detected in the total cell counts, and regrowth was observed despite the short hydraulic residence time of 6e8 h. Based on the flow cytometric fingerprints, the Bray-Curtis dissimilarity was calculated and was developed as unambiguous parameter to indicate or warn for changing microbial drinking water quality during operational events. In the studied water tower, drastic microbial water quality changes were reflected in the Bray-Curtis dissimilarity, which demonstrates its use as an indicator to follow-up and detect microbial quality changes in practice. Hence, the Bray Curtis dissimilarity can be used in an online setup as a straightforward parameter during full-scale operation of drinking water distribution, and combined with the cell concentration, it serves as an early-warning system for biological instability.
1. Introduction
Excessive microbial regrowth may result in a variety of problems during drinking water distribution. Microbial growth and activity can entail aesthetical issues, such as deviating taste and odor, and may eventually form a threat to public health (Skjevrak et al., 2004; Wingender and Flemming, 2011). Biofilms adhered to the pipe walls may cause operational problems, such as loss of pressure, and can accelerate pipe corrosion (Liu et al., 2016). Therefore, it is important to produce drinking water that is biologically stable, which implies sensu stricto that the microbial community composition and its abundance remain unchanged during distribution (Lautenschlager et al., 2013; Rittman and Snoeyinck, 1984; WHO, 2011).
“Biostability” is a theoretical concept that is now becoming an operational goal (Prest et al., 2016a). Biostability is most often enforced through the addition and maintenance of various disinfectants in the distribution network to eliminate the presence of viable bacteria (Zhang et al., 2002). However, disinfectants, such as chlorine, only have a temporary effect, and may furthermore result in the availability of dead organic matter that can be used during necrotrophic and heterotrophic growth, resulting in a biologically unstable system (Temmerman et al., 2006). Carcinogenic disinfection by-products are known to be formed when reacting with organic matter (Li and Mitch, 2018). This has encouraged countries, such as Switzerland and The Netherlands, to focus on achieving biostability without maintaining disinfection residuals in the network, but by limiting nutrients and carbon availability (van der Kooij et al., 2013).
To evaluate biostability of drinking water, heterotrophic plate counts (HPC) are commonly used, as this technique is the legal standard (European Communities, 1998, US Congress, 1996, WHO, 2011). However, samples are not taken frequently enough (e.g., weekly) and the time-to-result (2e7 days) is too long to anticipate events of biological instability. To facilitate decision-making when detecting events in the abundance or changes in the composition of the microbial community, implementation of online monitoring techniques is necessary.
Though they are only sporadically implemented at full-scale, a variety of techniques for online microbial drinking water quality monitoring are available. Online ATP measurements allow fast (i.e., <10 min) quantification of microbial activity based on luciferin/ luciferase enzymatic reactions (de Vera and Wert, 2019; Vang et al., 2014). Though cellular ATP is an indicator for bacterial activity, ATP quantification is prone to interferences of extracellular ATP, and is not always correlated to regrowth (Nescerecka et al., 2016; Vang et al., 2014; Vital et al., 2012). Other enzymatic techniques have been developed for near-real-time selective detection of Escherichia coli presence, targeting the specific enzyme b-glucuronidase (Hesari et al., 2016; Koschelnik et al., 2015). Results can be acquired within 1-2 h, with good accuracy. Although this selective method provides information about the hygienic quality of the water, it does not indicate events of general biological instability. Alternatively, recently developed online particle counters and online flow cytometry (FCM) allow the direct quantification of microbial abundance in water using optical technology within 10 min (Besmer et al., 2016; Hammes et al., 2012; Højris et al. 2016, 2018). Flow cytometry is the most established research technique, and when combined with advanced data analysis, it allows assessment of the microbial community characteristics, such as viability and phenotypic diversity (Gillespie et al., 2014; Props et al., 2016). Established ecological biodiversity metrics such as the alpha diversity or the within-sample diversity, and the beta diversity or the community turnover can be calculated from the flow cytometry data based on the phenotypic traits of a microbial community (Props et al., 2016; Whittaker, 1972). Correspondence between the phenotypic and genotypic diversity was shown by Props et al. (2016). The multidimensional flow cytometric data, summarized in the so-called cytometric fingerprint, reacts quickly to changing water quality and conditions (Besmer et al., 2017). The advanced community data analysis is more valuable than the cell counts for decision-making, as it takes into account both changes in the composition and abundance of the microbial community. Yet, as previously stated, it is not often used for microbial monitoring, because translation and interpretation of the results into warnings or concrete actions is currently lacking. This was also one of the major conclusions of an extended review by Safford and Bischel (2019) concerning the implementation of FCM as microbial monitoring tool in drinking water. A key point is to obtain better understanding of the microbial dynamics in full-scale drinking water networks to being able to generate fundamental knowledge about the system and to define the “degree of acceptable change” during events of biological instability (Prest et al., 2016a).
This study aims at demonstrating that advanced data analysis of flow cytometric data can be used for the development of a straightforward parameter to define the microbial water quality in a more holistic way, by not only taking into account the microbial abundance but also the stability of the microbial community composition. Therefore, a water tower with incoming drinking water produced from two different sources was chosen as full-scale model setup during regular and disturbed flow regime. Microbial regrowth in the water tower was evaluated through total cell counts. Also, biological metrics calculated from the flow cytometric fingerprint, such as the Bray-Curtis dissimilarity with a value between 0 and 1 that expresses how different two cytometric fingerprints are, was used for the detection of operational events that resulted in a drastic microbial water quality change (Greenacre and Primicerio, 2014).
2. Material and Methods
2.1. Water Tower Characteristic's
The microbial dynamics of a water tower in Flanders, Belgium were monitored in spring 2018 (April 18th - 27th) and summer 2018 (August 6th - 27th). Before entering the distribution network, the water is treated and disinfected, after which the drinking water is transported to the water tower through a high-pressure piping, and from which it flows to the customers through a low-pressure network (Fig. 1). This water tower was selected specifically because two drinking water feed streams produced from different source waters enter the reservoir. The first feed is produced from surface water (length feeder to tower = 47.58 km, diameter = 0.7-1.0 m) and the second feed stream is produced from groundwater (length feeder to tower = 24.44 km, diameter = 1.0 m). This way, the most important types of drinking water, i.e., produced from either surface water, groundwater or a mixture of both, are included in the monitoring setup. The physicochemical characteristics of the incoming and exiting streams of the water tower are summarized in Table 1. The water tower complex has a total buffering capacity of 6,500 m3. Based on the average daily intake, 72.48 ± 19.14% of the drinking water entering the tower is produced from surface water, with an average residence time of 6-8 h.
During the monitoring period in April, the water composition was altered in two different ways by closing one of the two feeders. During these events, the incoming water consisted out of drinking water either mostly produced from surface water or completely produced from groundwater (Fig. 1). During the first event on April 24th (07h34 - 23h19), the feed originated for almost 100% from surface water, which is referred to as “event 1”. On April 26th (09h06 - 15h29) and also on April 27th (09h13 - 12h30), the feed was 100% produced from groundwater, further referred to as “event 2” and “event 3” (Fig. 3). During event 2 and event 3, drinkingwater produced from groundwater was entering through both feeders (Fig. 1). As seasonal differences are an important factor known to affect the microbial drinking water quality during distribution systems (Nescerecka et al., 2018; Pinto et al., 2014; Prest et al., 2016b), monitoring was repeated in August 2018 without disturbances in the flow, to evaluate the seasonal impact on the biostability of drinking water.
2.2. Flow cytometric measurements
Dynamics of the microbial community of the incoming streams and exiting stream were monitored with online flow cytometry. Samples from all streams were taken automatically every 40 min, were stained and incubated for 20 ± 2 min at 37 °C prior to measurement
to allow for stable staining of all cell types. No pretreatment nor dilution was performed. During the monitoring period in April, 933 flow cytometric samples were taken (311 samples per stream), and during the monitoring period in August 2049 samples were taken (683 samples per stream). For automation of these measurements, an onCyt© (onCyt Microbiology AG, Switzerland) robotwas coupled to an Accuri™C6 flowcytometer or an Accuri™ C6 Plus flow cytometer (BD Biosciences, Belgium). The flow cytometers are equipped with a blue (20 mW, 488 nm) and a red laser (12.5 mW, 640 nm), two scatter detectors configured on the blue laser, and four fluorescence detectors with bandpass filters. Three of the bandpass filters are for the blue laser emission (FL1: 533/30 nm, FL2: 585/40 nm, and FL3: 670 LP) and one is for the red laser emission (FL4: 675/25 nm). The lower detection limit of flow cytometry in this type of application is in the order of 102 cells/mL and the instrument precision error is below 5%, independent of the cell concentration (Hammes et al. 2008, 2010). The cell concentrations in this study were at least 10 times higher than the lower detection limit, ensuring accurate and precise
quantification of cell concentrations in all streams. MilliQ (Merck, Belgium) was used as sheath fluid. A bacteriostatic concentrate solution containing 4.65% EDTA and 0.82% sodium fluoride, (BD Biosciences, Belgium) was added (0.79 vol% final concentration) to prevent bacterial growth in the sheath fluid during the course of the experiments. Staining was performed using a 1000x dilution of SYBR® Green I concentrate (Invitrogen, Belgium) in TRIS buffer with pH 8.2, with a 10 vol% final concentration. Flow cytometry samples were run in fixed volume mode (50 mL) at a flow rate of 66 mL/min. A Cavro® XCalibur Pump (Tecan Trading AG, Switzerland) with 12 channels connects the necessary fluidics, air and waste with the three chambers in the onCyt© robot. The pump regime is adapted to the desired sampling frequency (40 min interval) and size (800 mL total volume) using the onCyt© software.
2.3. Data analysis
Flow cytometric datawas analyzed in R (v3.4.4). Flow Cytometry Standard (.fcs) files were imported using the flowCore package (v1.44.2) (Hahne et al., 2009). The FlowAI package (v1.14.0) was used to check the flowcytometry data quality and to remove events having anomalous values in terms of flow rate stability, signal acquisition and the dynamic range, caused by technical issues (Monaco et al., 2016). Background noise caused by abiotic interferences was removed by manually drawing a gate based on the FL1 and FL3 fluorescence data, as the combination of these two fluorescence parameters results in the most optimal signal and noise separation for drinking water samples (Hammes and Egli 2005, 2010). Further data processing was done using the Phenoflow package (v1.1.1) based on the signal height values, with rescaling based on the maximum FL1 fluorescence signal height (FL1-H) after gating of the bacterial population as described by Props et al. (2016). To detect the impact of the disturbed flow regimes on the microbiological quality of water, phenotypic community analysis was performed on the FCM data by the use of advanced fingerprinting (Props et al., 2016; Rogers and Holyst, 2009). In this analysis, the flow cytometry data of every sample is transformed, discretized and concatenated into a one-dimensional vector that serves as basis for further phenotypic community analysis. Fingerprinting was performed using probability binning approach of the package flowFP (v1.40.1), constructing a model grid with 5 recursions (25 bins) (Rogers and Holyst, 2009). The model was built based on the fluorescence signal height values FL1-H and FL3-H of all samples. From these fingerprints, beta diversity analysis and Bray-Curtis dissimilarity calculations were performed using the vegan package (v2.5.4) (Oksanen et al., 2019). Resampling to the lowest sample size (n . 124 cells) was performed prior to beta diversity analysis to account for size-dependent differences. The Bray-Curtis dissimilarity was chosen above other biological metrics derived from the cytometric fingerprints (e.g., phenotypic typing (Fig. S1) and diversity analysis) to evaluate differences between the microbial community of samples, as, with regard to the industry’s needs, this is an easily interpretable, unequivocal parameter quantifying the difference between two cytometric
fingerprints. Furthermore, this is a frequently used quantitative dissimilarity index with ecological value (Bray and Curtis, 1957; Legendre and Legendre, 2012). Two identical fingerprints will have a Bray-Curtis dissimilarity of 0, whereas two fingerprints that have no non-empty bins in common, will have a Bray-Curtis dissimilarity of 1 (Greenacre and Primicerio, 2014). A sufficiently large representative set of samples taken during normal operation of theater tower at the beginning of each monitoring period was taken as a baseline (n . 104 in April, n . 106 in August) for dissimilarity comparison. The Bray-Curtis dissimilarity that was assigned to a sample was calculated as the average of the Bray-Curtis dissimilarities between that sample and all of the baseline samples (Fig. 5). As a basis for decisions concerning biological instability, a threshold for event detection was set at the average Bray-Curtis dissimilarity calculated between all baseline samples plus three times the standard deviation of this distribution. This is a rather conservative threshold, that finds its origin in the normal distribution, where it will include 99.87% of the data and thus regard the remaining 0.13% as deviating events (Howell et al., 1998). This method was carried out to account for inter-sample variability, resulting in a threshold of 0.301 in the case of the monitoring period in April (Fig. 5).
2.4. Statistics
Correlation between variables was calculated using the Pearson correlation coefficient (rp) for linear correlations. Prior to calculation of correlations, the normality and homoscedasticity of the data was checked and approved (Fig. S2). All statistical analyses were performed in R using the stats package (v.3.5.2). The ANOSIM analysis in the vegan package (v2.5.4) with 100 permutations was used to determine differences between the cytometric fingerprints of different water types.
2.5. Data availability
The raw flow cytometry data sets have been deposited on FlowRepository and are publicly available under accession ID FRFCM- Z25U (April monitoring campaign) and accession ID FRFCM-Z25T (August monitoring campaign).