Biofilms are aggregates of microbial cells, usually accumulated at a solid–liquid interface and encased in a matrix of highly hydrated EPS (Flemming and Wingender, 2010). Biofilms are initiated by adsorption of bacterial cells to a surface, followed by production of EPS by attached cells, and bacterial proliferation within the formed biofilm (Figure 3). Biofilm cell density in drinking water distribution systems can vary significantly with cell numbers in the range of 10⁴ to 10⁸ cells/cm², and with the amount of active biomass, as measured with ATP concentrations in the range of 10² to 10⁴ pg ATP/cm² (Boe-Hansen et al., 2002b; Wingender and Flemming, 2004; Långmark et al., 2005; Liu et al., 2014). The EPS structure offers a protective environment against disinfectant residuals and against grazing organisms, and binds organic and inorganic compounds (LeChevallier et al., 1988; Flemming and Wingender, 2010). Due to extracellular enzymes in the EPS, biofilm bacteria can utilize complex organic substrates, such as humic acids that are not easily biodegradable and usually not used by bulk water bacteria (Camper, 2004; Flemming and Wingender, 2010). Availability of additional nutrients creates new ecological niches and thus enables growth of different microorganisms than present in bulk water. Liu et al. (2014) reported that about 12% of the total bacteria (OTUs) found in biofilms were not shared with the bulk water. Shift in bacterial community between suspended and attached phases are furthermore influenced by specific pipe materials applied in drinking water distribution networks (see details in section “Biological Stability in Drinking Water: Implications for Treatment and Distribution”). Studies have shown that young biofilms display similar characteristics to the bulk water bacterial communities, while mature biofilms displayed lower growth rates and lower community richness, indicating a different bacterial community (Boe-Hansen et al., 2002b; Martiny et al., 2003).

Sediment formation is the result of particle deposition under favorable hydraulic conditions (Vreeburg et al., 2008) (Figure 3), and distribution networks can contains as much as 3000 mg/m loose deposits (Barbeau et al., 2005; Vreeburg et al., 2008). Deposited particles offer a favorable environment for bacterial growth, as (i) they provide a large surface area, (ii) are usually composed of organic compounds and also (iii) contain inorganic substrates (e.g., Ca, Fe, Mn; Gauthier et al., 1999; Zacheus et al., 2001). Liu et al. (2014) found that sediments may favor the growth of specific bacterial species, for example bacteria involved in iron and arsenic cycling (e.g., Rhodoferax sp. and Geobacter sp.). Particle sedimentation combined with biofilm formation and EPS production consolidates the sediment structure, which expands as long as sediments are not re-suspended during high hydraulic peaks (cf. details in section “Biological Stability in Drinking Water: Implications for Treatment and Distribution”). During sediment expansion, anoxic or anaerobic conditions are likely to be created, providing a selective environment for the growth of bacteria not found in the bulk water phase (e.g., Rhodoferax sp. and Geobacter sp., or bacteria from actinomycetes group; Zacheus et al., 2001; Liu et al., 2014). Sediments were shown to contain large amounts of biomass, in the range of 700 to 4000 ng ATP/g loose deposit (Liu et al., 2014) and up to 10¹¹ cells/g (Barbeau et al., 2005) and to harbor the largest bacterial diversity, compared to bulk water and biofilm phases, with 29% of the total bacteria which were not shared with the bulk water (Liu et al., 2014). Sediments can be the source of hygienic and operational problems, as they offer a protective environment for bacteria to grow, particularly for undesirable organisms (Gauthier et al., 1999), and iron-oxidizing bacteria shown to increase corrosion processes (Sun et al., 2014). Furthermore, sediments can be the source of colored water when re-suspended into the bulk water (Vreeburg and Boxall, 2007) and host invertebrates (Christensen et al., 2011).

Mechanisms of interactions between biofilm, sediment, and bulk water bacteria have been investigated to estimate to which extent biofilms and sediments affect the bacterial community in bulk water in terms of abundance and community composition and structure, thus how these interactions might affect biological stability. Biofilms have long been considered as containing the largest fraction (up to 95%) of bacterial cells in drinking water distribution systems (Flemming et al., 2002). However, the sediment phase has been largely overlooked due to sampling difficulties (Liu et al., 2013a). A recent study has shown that 98% of bacterial cells were situated in both biofilms and sediments, of which 60 to 90% were actually situated in the sediment phase (Liu et al., 2014). It was first assumed that the majority of bulk water bacteria originate from biofilm detachment (LeChevallier et al., 1987; van der Wende et al., 1989), rather than bacterial growth in the bulk water phase. However, this hypothesis was challenged by a study by Boe-Hansen et al. (2002a), which demonstrated higher bacterial activity and growth rates in bulk water than in biofilm (0.30 day^-1 in bulk water compared to 0.048 day^-1 in biofilm). The study showed that bacterial production in the bulk water constituted 37% of the total bacterial production in a model drinking water system. Recently, Liu et al. (2014) detected different bacterial community compositions in bulk water and biofilm, and Henne et al. (2012) found significantly different core bacterial communities in both water and biofilm sampled from full-scale and long used distribution systems. From these observations, it was suggested that bulk water bacteria function as a seed bank for biofilms and sediments (Henne et al., 2012; Liu et al., 2014), each phase thereafter developing its own bacterial community, with different predominant species depending on specific environmental conditions. However, interactions between bacteria in water, sediment, and biofilm phases are still unclear. Detachment of biofilms and re-suspension of sediments would arguably contribute to bacterial cell concentrations and community composition in the bulk water (cf. details on hydraulic conditions in section “Biological Stability in Drinking Water: Implications for Treatment and Distribution”). Moreover, bacterial cells in biofilm and sediments would compete with bulk water bacteria for available nutrients. As the largest proportion of bacteria is present in the two phases, substantially fewer nutrients would remain available for bulk water bacteria. Consequently, the role and interactions between bulk water, biofilms, and sediments should be taken into account when considering biological stability in drinking water distribution systems.

Water treatment strategies are adapted to the characteristics of raw water, which can be very diverse. Deep ground water typically contains very low bacterial cell concentrations (10³–10⁴ cells/mL), are often anaerobic and contain low organic nutrients (e.g., AOC below 10 μg ac-C/L) but potentially high methane (e.g., 0.01 to 9 mg/L) and ammonium concentrations (e.g., 0.2 to 5 mg/L; van der Kooij et al., 1982; de Vet et al., 2009; Hedegaard and Albrechtsen, 2014). Surface waters, on the other hand, typically contain high bacterial cell numbers (10⁵–10⁶ cells/mL) and relatively high concentrations of organics (e.g., AOC in the range of 5–150 μg Ac-C/L; van der Kooij, 1990; Hammes et al., 2010a; van der Wielen and van der Kooij, 2010), while numerous in-between situations can be found with, e.g., infiltrated water or phreatic aerobic ground water (van der Kooij et al., 1982; van der Kooij and Hijnen, 1985b). Treatment strategies typically aim to inactivate hygienically relevant organisms, remove micro-pollutants, improve aesthetic aspects (turbidity, taste, and odor), and prevent bacterial growth during water distribution. Therefore, combinations of different treatment types are applied. In several European countries, stable water is produced from surface water using extensive multistep treatments, without the use of residual disinfectant. As an example, treatment trains applied in Zurich and Amsterdam include one or several disinfection steps (e.g., ozonation) and a combination of biological filtration processes (e.g., rapid sand filter, slow sand filter, activated carbon filter, and/or dune infiltration; Hammes et al., 2010a; Vital et al., 2012a). The choice of treatment strategies and combinations are crucial in the production of biological stable water, as it determines (i) composition and concentration of individual organic and inorganic nutrients, and (ii) bacterial community characteristics (abundance, activity, community composition).

Oxidation processes, such as chlorination and ozonation or UV/H₂O₂ advanced oxidation, are commonly applied as primary disinfection strategies for bacterial inactivation (Bernarde et al., 1967; Hoefel et al., 2005; Ramseier et al., 2011a). Oxidative treatments often result in modification of the substrate composition (van der Kooij et al., 1989; Schmidt et al., 1998; Okuda et al., 2009; Sarathy and Mohseni, 2009; Ramseier et al., 2011b). As an example, Vital et al. (2012a) reported an AOC increase from 20 to 120 μg ac-C/L after ozonation, while DOC concentration did not change, indicating a clear change in the composition of organic compounds in water. In this regard, Hammes et al. (2006) showed that 60–90% of the AOC formed after ozonation of Lake water was composed of organic acids. Primary disinfection processes with high ozone or free chlorine dosage also typically result in inactivation of the entire bacterial community (Hammes et al., 2008, 2010a), thus leaving room for new microorganisms to colonize the treated water and consume the altered nutrient pool. Overall, oxidative processes create highly unstable water due to combined effects of (i) increased nutrients availability and (ii) absence and/or inactivation of bacterial cells, thus creating a new niche for bacterial growth.

Biological filtration processes are applied worldwide and are usually implemented after primary disinfection processes (van der Kooij, 1992; Prévost et al., 1998; Gauthier et al., 1999; Hammes et al., 2010a; Pinto et al., 2012). Biological filtration processes are believed to be an essential step for production of biological stable water (Rittmann and Snoeyink, 1984; van der Kooij, 1990; Smeets et al., 2009). Different types of filtration technologies, applied for diverse purposes, can act as biological filtration. These include active carbon, rapid or slow sand filtrations, or soil infiltration. When water flows through biological filters, bacterial cells attach to filter particles (Servais et al., 1994; Velten et al., 2007) and consume substrates provided by the water, resulting in formation of a diverse bacterial community. Pinto et al. (2012) recorded the presence of 14 different bacterial phyla in the effluent of a rapid dual media filter. Bacterial community abundance, activity and composition in filters depends on nutrient composition of water flowing through the filter, resulting in the presence of different organisms being able to metabolize different types of substrate, and on environmental parameters, such as temperature (cf. section “A Deeper Look into Microbial Dynamics in Drinking Water”; Fonseca et al., 2001). Bacterial cells regularly detach from the filters, and bacteria are found in water after biological filtration typically in bacterial cell concentrations ranging between 10⁴ and 10⁵ cells/mL (Servais et al., 1994; Hammes et al., 2010a; Lautenschlager et al., 2014). Studies have shown that biological filtration processes shape the bacterial community composition in treated water (Pinto et al., 2012; Lin et al., 2014), which is only slightly modified during water distribution (Henne et al., 2012). Successive biological filtration steps of different types are often applied in order to consume fast degradable and more complex organic compounds. As an example, Lautenschlager et al. (2014) showed in a multi-step treatment plant in Zürich that low molecular weight humic substances were mainly removed during rapid sand filtration, while polysaccharides were degraded in subsequent slow sand filtration. This can be attributed to residence time of water in the slow sand filter, which can be up to 50 times longer than in rapid sand filter (Huisman and Wood, 1974). Application of successive biological filtration steps thus gradually shapes both the nutrient pool and bacterial community characteristics in treated water (Lautenschlager et al., 2014). Changing raw water quality or temperature variations potentially affect bacterial community characteristics in treated water in time, including bacterial community composition (Pinto et al., 2014) and/or activity and/or abundance (van der Wielen and van der Kooij, 2010). Application of biological filtration processes has two main advantages: (i) growth-supporting organic and inorganic nutrients in treated water are considerably reduced (up to 80–90% AOC removal; van der Kooij, 1987; Hijnen and van der Kooij, 1993; Servais et al., 1994) and (ii) a very diverse autochthonous bacterial community is released in treated water, therefore covering a large substrate utilization spectrum and contributing to biological stability of water (cf. section “A Deeper Look into Microbial Dynamics in Drinking Water”).

In recent years, application of membrane filtration methods rather than chemical additions to water, have been proposed as an alternative primary or secondary disinfection approach (Sibille et al., 1997; Sibille, 1998; Vrouwenvelder et al., 2004). Several membrane types can be distinguished based on their removal capacity, and include MF, UF, NF, and RO (Mallevialle et al., 1996). The advantage of this approach is that it lowers significantly bacterial cell concentration in treated water, by removing over 99.5% of bacterial cells (Liikanen et al., 2003), without producing any by-products that have potential adverse health effects, or contribute to available nutrients for bacterial growth. Moreover, membrane filtration considerably reduces the amount of suspended particles entering the distribution system, thus lowering the potential for sediment deposition (Vreeburg et al., 2008) and associated problems (cf. sections “Location of Bacterial Cells: In Water, Sediment, and Biofilm” and “Water Temperature”).

The use of membrane-based filtration methods, however, potentially leads to production of unstable water (Okabe et al., 2002). While bacterial cells are physically removed from the water, a significant fraction of nutrients passes through the membranes, whatever the cut-off value (Liu et al., 2013b). While over 90% of BDOC is retained with NF membranes (Escobar and Randall, 2001; Vrouwenvelder et al., 2004), a large proportion of AOC remain in the filtered water, ranging between 5 and 90%, depending on the cut-off value and the influent water quality (Liikanen et al., 2003; Meylan et al., 2007). Even RO membranes, which enable an efficient bacterial cell and organic compound removal (over 99.5% rejection of bacterial cells, and over 90% of TOC rejection), do not retain all AOC, with about 80% AOC rejection efficiency (Park and Hu, 2010). As no system remains sterile (Liikanen et al., 2003; Liu et al., 2013b), these nutrients are fully available for new bacterial growth in the distribution system. In such case, water retains a certain level of growth potential, as highlighted by Liikanen et al. (2003) who observed an HPC increase of over a log unit during incubation of the permeate of nanofilters for 20 days. Similarly, significant regrowth occurred in a model distribution system fed with RO permeate, both in the bulk water (from 50 bacterial cells/mL in the RO permeate up to about 10³ cells/mL in the bulk water after 20 days residence time)and in the form of a biofilm (containing 8 × 10³ cells/cm²) developed on PVC coupons in a biofilm reactor fed with the RO-treated water during 20 days. Bacterial cell numbers remained, however, extremely low in comparison with a similar system fed with the same tap water without prior RO treatment (2 × 10⁵ cells/mL in bulk water, and 7 × 10⁵ cells/cm² in biofilm), showing a reduced growth potential of the water after RO filtration. In summary, membrane filtered water remains unstable due to the permeability of certain compounds through the membrane, the lowest growth potential being obtained by membranes retaining the highest proportion of nutrients (e.g., RO).

Ion exchange has also been proposed as an additional step in full-scale conventional drinking water treatment system to improve color of final water and biological stability (e.g., Heijman et al., 1999). IEX is efficient in removing low molecular weight organics (Bolto et al., 2002), and has been shown to significantly reduce both DOC (e.g., by 50%) and AOC (e.g., by 60%) of final water (Grefte et al., 2011). Biofilm formation rate of produced water was subsequently reduced by 70%. However, water treated by IEX retains a certain degree of bacterial growth potential: in a study by Liu et al. (2013b), bacterial growth was detected in a water treated with IEX and incubated for 24 h, with an increase in ATP of 23% and in total bacterial cell concentration of 41%, showing that the IEX treatment did not fully prevent bacterial growth.

In many cases, a secondary disinfection is applied to the produced drinking water before distribution, typically with free chlorine, chlorine dioxide, or monochloramine (van der Kooij, 1992; LeChevallier et al., 1996; Prévost et al., 1998; Batté et al., 2006; Pinto et al., 2012). The aim of secondary disinfection is either to prevent bacterial growth in drinking water distribution systems, or to reduce HPC values below the guideline value. In both cases, application of a disinfectant alters significantly the bacterial community composition and structure, usually resulting in a lower diversity and richness, as well as the nutrient pool (cf. details in section “A Deeper Look into Microbial Dynamics in Drinking Water”). Consequently, the growth potential of the water is affected with new niches available for bacterial growth once disinfectant residual is depleted (cf. section “A Deeper Look into Microbial Dynamics in Drinking Water”).

Pipe material composition influences biofilm development on pipe surfaces (Figure 3). A single distribution system typically includes diverse materials such as metal pipes (e.g., cast iron, stainless steel), cement, and/or synthetic polymers (e.g., PVC), on which widely different growth rates, bacterial densities, and community compositions were measured (Niquette et al., 2000; Yu et al., 2010; Wang et al., 2014). Several studies have demonstrated that iron pipes allow the highest bacterial densities, with up to 45 times higher bacterial biomass fixed on iron coupons than on plastic materials (Norton and LeChevallier, 2000; Niquette et al., 2000). Corrosion of iron pipes leads to the release of particles and deposit formation (Figure 3), on which organic and inorganic compounds adsorb, and which act as attachment sites, where bacteria are protected from disinfectant residuals (Camper, 2004; Morton et al., 2005). Synthetic polymeric pipe materials, such as cross-linked poly-ethylene (PEX), polybutylene (PB), or PVC, were shown to release biodegradable organic substances (Figure 3), modifying the available nutrient source for bacteria to grow (Skjevrak et al., 2003; Bucheli-Witschel et al., 2012). Even if stable water is produced at the treatment plant, release of additional nutrients into water can cause biological instability. As an example, van der Kooij and Veenendaal (2001) reported an increase by up to 200% in ATP concentration after incubation of water with plastic materials (e.g., plasticized PVC, PVCp) compared to the incubation of the same water alone. Pipe materials influence the bacterial community predominantly during its first stage of development (Martiny et al., 2003; Henne et al., 2012). Consequently, construction of new distribution systems or replacement of pipe segments in old distribution systems profoundly affects the biological stability of drinking water for a period of time that can reach up to several years before a stable system is reached again (Martiny et al., 2003).

Hydraulic changes in distribution systems are frequent and play a major role in interactions between bulk water, sediment, and biofilm phases (Figure 3). Low water consumption periods result in low flow velocities or even water stagnation in reservoirs and parts of distribution systems, enabling particle deposition, increased residence time, and offering favorable conditions for bacterial growth to occur (Gauthier et al., 1999; Zacheus et al., 2001; Liu et al., 2013a,b). On the other hand, hydraulic peaks caused by high consumption periods, fire-fighting actions, pipe flushing or system malfunctioning such as pipe breaks, unavoidably result in increased biofilm detachment and possible sediment re-suspension (Lehtola et al., 2006; Vreeburg and Boxall, 2007). This increases bacterial dispersal in the network water and will modify bacterial abundance and community composition in the bulk water, thus affecting biological stability (Figure 3). Hydraulic peaks can cause release of hygienically relevant organisms (Torvinen et al., 2004; Wingender and Flemming, 2011), and serious aesthetic issues such as presence of invertebrates in the bulk water (van Lieverloo et al., 2002a), or color deterioration of water at the point of consumption (Vreeburg et al., 2004). To avoid such problems, Vreeburg et al. (2009) have proposed to adapt distribution system design, to control hydrodynamic conditions in distribution systems and limit particle deposition, sediment formation, and avoid dramatic hydraulic peaks.

Water temperature is an essential factor influencing bacterial growth kinetics and competition processes. Drinking water temperatures typically range between 3 and 25°C in European countries, (Kerneis et al., 1995; Niquette et al., 2001; Uhl and Schaule, 2004), and fluctuate seasonally within this temperature range even within a single drinking water distribution system. Elevated water temperatures have often been associated with increased bacterial abundance in drinking water distribution systems (Servais et al., 1992; Kerneis et al., 1995; Francisque et al., 2009; Liu et al., 2013a), and with higher numbers in indicator organisms such as coliforms or Aeromonas (Burke et al., 1984; Volk and Joret, 1994; LeChevallier et al., 1996). Francisque et al. (2009) recorded five times more occurrences of HPC concentrations above 100 CFU/mL at temperatures above 18°C than at lower water temperatures. In addition, water temperature can also affect bacterial community composition, by providing competitive advantages to specific bacterial species in defined temperature ranges, including pathogenic species (Vital et al., 2007, 2012b). For example, Vital et al. (2012b) showed that the maximum growth rate and competitive fitness of E. coli grown with an indigenous drinking water community increased with temperature in the range of 12–30°C. There is therefore increased chance for problems associated with bacterial growth in summer periods (with higher water temperatures), such as hygienic risks, deterioration of aesthetic aspects of water, malfunctioning of water installations, exceeding of legal guidelines, for e.g., heterotrophic plate counts (cf. section “Problems Associated with Bacterial Growth in Drinking Water Distribution Systems”). In this regard, specific attention is given to the influence of anticipated global warming on drinking water quality, as average water temperatures are expected to increase, concomitantly with longer periods at higher temperatures (Levin et al., 2002). Moreover, significant seasonal shifts in bacterial community composition have been reported in effluents of treatment utilities (Pinto et al., 2014). Though the cause of such variations is not clear at this stage, seasonal variations in water temperature could well be involved in bacterial community characteristics of water entering the distribution system. Further research would be needed to determine to which extent these changes affect bacterial competition processes within the drinking water distribution system.

Residence times can reach up to a few days within a distribution system (Kerneis et al., 1995), leaving time for bacterial growth to occur. Residence times depend on (i) distance from treatment plant, up to 100 km or more in the case of extended cities or remote villages (Cook et al., 2014), (ii) pipe diameters, varying from a few meters in water mains down to a few millimeters in service pipes, and (iii) water flow velocity caused by water consumption. The latter also influences additional residence time of water within reservoirs at the treatment outlet and/or within distribution systems. In general, higher bacterial abundances were observed at higher water residence times in the network (Maul et al., 1985; Kerneis et al., 1995; Uhl and Schaule, 2004; Nescerecka et al., 2014). In the case of chlorinated water, increased bacterial abundance at long residence time is often congruent with decay of disinfectant residual (Servais et al., 1995). However, in systems distributing water without residual disinfectant, long residence times do not systematically lead to an increase, but occasionally to a decrease in bacterial abundances and/or activity (van der Wielen and van der Kooij, 2010), possibly be due to substrate limitations. In summary, the effect of residence time on microbial dynamics is not clear and is often dependent on other factors such as residual disinfectant decay and/or substrate availability.

While biological stability could theoretically be achieved by maintenance of sufficient disinfectant residuals at all points of a distribution system, this is challenging to achieve. Many of the factors mentioned in this section contribute to disinfectant decay within distribution systems. For example, dissolved nutrients in bulk water, as well as EPS and organic and inorganic nutrients adsorbed on biofilms and sediments can react with the disinfectant (Gauthier et al., 1999; Campos and Harmant, 2002), resulting in lowered concentrations or even absence of residual disinfectant at long residence times (Maul et al., 1985; Kerneis et al., 1995). This phenomenon is increased in pipe sections with small diameter, in which the surface to volume ratio is higher, thus increasing contact of water with pipe materials and/or biofilms and sediments. Prévost et al. (1998) have observed a faster decay in 150 mm diameter pipes than in larger water mains. Chemical reactions are further affected by water temperature, also modifying the disinfection capacity in time (Bernarde et al., 1967; Urano et al., 1983). Loss of disinfection residuals undoubtedly results in biological instability and bacterial regrowth (Servais et al., 1992; LeChevallier et al., 1996; Nescerecka et al., 2014).

Good practices for construction, operation, and maintenance of water distribution systems are also essential to maintain biological stability in distribution systems (Smeets et al., 2009). A code for hygiene during pipe installation was developed by water utilities in the Netherlands (van Lieverloo et al., 2002b). Besides, maintenance of sufficient pressure in the system for protection against intrusion of external water, and good maintenance of pipelines to maintain physical integrity and limit leakages, are essential to prevent external contamination. In this way, intrusion of both external organisms and organic and inorganic nutrients is avoided, that would profoundly affect the nutrient pool and bacterial community composition in drinking water.

A range of methods have been developed to assess growth promoting properties (growth potential) of drinking water and are traditional indicators for biological stability (van der Kooij et al., 1982; Servais et al., 1989; Hu et al., 1999; Ross et al., 2013). In essence, these methods are predictive, as the water is analyzed before distribution, and the tests are used to predict the extent of growth that could potentially occur during water distribution.

Initial focus of these methods was on biodegradable organic carbon and included the assessment of AOC, initially proposed by van der Kooij et al. (1982), van der Kooij and Hijnen (1985b), and BDOC method, proposed by Servais et al. (1987, 1989). Both methods have been the subject to numerous adaptations for improving the tests representativeness, ease of handling and time (Werner, 1984; Joret and Levi, 1986; Joret et al., 1988; Lucena et al., 1990; Ribas et al., 1991; Hambsch et al., 1992; LeChevallier et al., 1993; Miettinen et al., 1997; Haddix et al., 2004; Hammes and Egli, 2005; Sack et al., 2009; Ross et al., 2013). While the AOC assays by definition focus on easily available substrates for planktonic growth, BDOC assays enable the assessment of the refractory fraction of biodegradable organic carbon, which can be used by biofilm-bacteria in distribution systems (Camper, 2004; Flemming and Wingender, 2010). In general, higher BDOC values than AOC values are measured in drinking water. For example, Volk and LeChevallier (1999) reported BDOC values in the range of 0.15–0.75 mg/L and AOC values in the range of 0.10–0.33 mg/L in treated surface waters. Systematic application of the methods for full-scale studies showed that little AOC uptake occurred and HPC values remained below guideline values (100 cfu/mL) during distribution of non-chlorinated waters with an AOC level below 10 μg Ac-C/L (van der Kooij, 1992). In chlorinated water, statistically fewer occurrences of coliforms were observed at AOC concentrations below 100 μg Ac-C/L than in waters with higher AOC concentrations (LeChevallier et al., 1996). No decrease in BDOC was observed during distribution of waters with a BDOC levels below 150 μg C/L in chlorinated distribution systems (Servais et al., 1992; Volk and Joret, 1994). These different AOC and BDOC values have been associated with no/limited bacterial growth and have been subsequently used as guidelines for biological stable water (van der Kooij, 1987; LeChevallier et al., 1992).

Both AOC and BDOC methods focus primarily on organic carbon as the only growth-limiting substrate in drinking water. However, other compounds were identified as possible microbial growth controlling substances (organic carbon, ammonium, manganese, iron; Rittmann and Snoeyink, 1984; States et al., 1985; van der Kooij and Hijnen, 1985b; Miettinen et al., 1997), and might be in some cases more critical to describe and understand microbial dynamics in full-scale drinking water systems. The bacterial growth that water can support as such, independently of its growth-limiting element, can be assessed by incubation of a water samples without pre-treatment in controlled conditions (Gillespie et al., 2014). The same approach can be extended for the assessment of growth-limiting compounds in water samples by step-wise addition of single substrates or combinations of substrates (Miettinen et al., 1999; Ihssen and Egli, 2004), and subsequent measurement of the bacterial growth potential.

A large number of devices have been used to study the property of water to support development of biofilms in drinking water distribution systems. These include annular reactors (e.g., Volk and LeChevallier, 1999), Propella^® reactors (e.g., Lehtola et al., 2006), flow cell systems (e.g., Manuel et al., 2010), and Robbins devices (e.g., Kalmbach et al., 1997). An extensive overview of available systems has been provided by Gomes et al. (2014). Many of these devices comprise coupons of defined materials such as copper, PVC or cement, which can have an influence on growth of biofilm. Alternatively, van der Kooij et al. (1995) proposed to quantify the potential of water to form biofilms on inert materials. Subsequently, additional guideline values for biological stable water have been introduced, such as BFR which should be kept smaller than 10 pg ATP/cm².d in produced drinking water (van der Kooij et al., 1995).

Materials in contact with drinking water can profoundly influence growth potential. A number of assays have been developed to assess this effect, and the approach varies from country to country. For example, a standardized method in the United Kingdom measures consumed dissolved oxygen as an indicator for bacterial growth on 150 cm² of material in contact with water during 7 weeks (Colbourne and Brown, 1979; BSI, 2000). In Germany, formation of biomass on 800 cm² material surface is measured as the volume of slime produced during 12 weeks (ÖNorm B 5018-12, 2002) and in the Netherlands, BFP is measured on 50 cm² material (van der Kooij and Veenendaal, 1994) with ATP analysis. The latter test was further adapted to estimate the potential of tested material to promote bacterial growth both on its surface (biofilm) and in the bulk water in contact with the material (BPP; van der Kooij and Veenendaal, 2001). More recently, Bucheli-Witschel et al. (2012) have developed a new test package which requires 14 days and that measures the BPP (quantified with FCM) as well as the AOC fraction in water after a series of high-temperature standardized migration assays. Though assessment of growth-promoting properties of materials in contact with water is a useful decision-making tool for selection of appropriate materials, it is challenging to assess growth-promoting properties of materials already installed in networks. The latter can be largely affected by long-term aging in specific conditions in distribution systems, including continuous flow and presence of biofilms, sediments and/or specific degrading organisms.

An increase in bacterial abundance is a clear sign of biological instability, and can be measured as a change in specific bacteria, bacterial groups, or in the total bacterial community (Servais et al., 1992; LeChevallier et al., 1996; van der Wielen and van der Kooij, 2010; Nescerecka et al., 2014). Specific detection usually focuses on hygienically relevant organisms such as Legionella, Mycobacterium, Pseudomonas aeruginosa, total coliforms, Enterococcus and E. coli (Tallon et al., 2005). In the Netherlands, the presence of Aeromonas is also included in legislation and often tested as an indicator for biological stability (Waterleidingbesluit, 2001). Specific detection is traditionally performed by cultivation on selective media, and more recently with molecular-based techniques such as qPCR or with the use of specific antibodies (Rompré et al., 2002; Schets et al., 2002; Tallon et al., 2005; Hügler et al., 2011; Douterelo et al., 2014). Besides specific detection, cultivation is also used worldwide in drinking water monitoring for HPC methods (Allen et al., 2004; Pepper et al., 2004; Uhl and Schaule, 2004; Batté et al., 2006). There is more or less universal agreement that the fraction of bacterial cells detected by HPC methods is less than 1% of the total bacterial concentration in drinking water (Staley and Konopka, 1985; Hoefel et al., 2003; Hammes et al., 2010a; Epstein, 2013). Quantification of all bacterial cells in a water sample is achieved by cell labeling with fluorescent dyes (e.g., SYBR Green I or DAPI) and subsequent detection by either epifluorescence microscopy (Servais et al., 1992; Niquette et al., 2001) or FCM (Hoefel et al., 2003; Hammes et al., 2008). FCM enables extremely fast water analysis with limited sample handling (Hammes et al., 2008; Prest et al., 2013), as well as possibility for automation (Hammes et al., 2012; Besmer et al., 2014). The use of FCM has revealed an increase in total cell concentrations in the range of 11–23% in drinking water distributed without disinfectant residual (Hammes et al., 2010a; Prest et al., 2014) and up to 300% in household taps after stagnation (Lautenschlager et al., 2010). While all methods above are well adapted for enumeration of suspended bacterial cells, they cannot be applied directly for enumeration of bacteria attached to particles, sediments, or biofilms. Pre-treatments such as ultrasonication or physical removal (e.g., swabbing or scratching the surface) are required to detach and suspend bacteria (Boe-Hansen et al., 2002b; Magic-Knezev and van der Kooij, 2004).

A large number of viable drinking water bacteria is not detected with cultivation methods as a result of being either so-called unculturable bacteria (Kaeberlein et al., 2002; D’Onofrio et al., 2010; Epstein, 2013) or in the so-called “VBNC” state (Oliver, 2005; Hammes et al., 2011). It is particularly useful to determine cultivation-independent bacterial viability/activity when disinfection is applied. Viability of all bacterial cells can be assessed by cell labeling with fluorescent dyes targeting specific features of bacterial cells related to bacterial viability, such as cell membrane integrity (e.g., propidium iodide), membrane potential [e.g., DiBAC₄(3)], respiratory activity (e.g., CTC), and subsequent detection with epifluorescence microscopy or flow cytometry (Prévost et al., 1998; Hoefel et al., 2005; Berney et al., 2008; Ramseier et al., 2011a; Hammes et al., 2011). As an example, Berney et al. (2008) have shown that water sampled at a household tap after non-chlorinated water distribution contained 1.7 × 10⁵ cells/mL, of which about 70% had intact, polarized bacterial cell membranes, but only 20% displayed esterase activity. Noticeable biological instability has been detected in a chlorinated drinking water distribution system, in which the total bacterial cell concentration increased from 1.62 × 10⁵ to 1.07 × 10⁶ cells/mL and the percentage of cells with intact membranes increased from 3 to 59% (Nescerecka et al., 2014). Besides the above single-cell methods, ATP is a useful bulk water measurement of biological activity, as ATP is present only in living cells (Karl, 1980). Typically, total ATP concentrations in the range of 0.8 to 12 ng ATP/L are found in drinking water (Lautenschlager et al., 2010; van der Wielen and van der Kooij, 2010). For increased specificity in the method, it is possible to distinguish between intracellular and extracellular ATP and to estimate the average amount of bacterial ATP per cell when combined with FCM quantification (Hammes et al., 2010b; Vital et al., 2012a). ATP measurements have successively been applied in several case studies to assess biological stability (van der Wielen and van der Kooij, 2010; Vital et al., 2012a; Liu et al., 2014; Nescerecka et al., 2014). A few alternatives for bacterial activity measurements in drinking water have been proposed (Staley and Konopka, 1985), including the measurement of H³-Leucine or H³-thymidine incorporation (Servais et al., 1992; Boe-Hansen et al., 2002a).

A change in microbial community composition is indicative of instability (Lautenschlager et al., 2013; Pinto et al., 2014; El-Chakhtoura et al., 2015). Molecular methods for this purpose are predominantly based on DNA extraction and PCR amplification (e.g., of 16S rRNA gene) and are roughly classified into either fingerprinting methods or high-throughput sequencing methods (Read et al., 2011; Douterelo et al., 2014). Fingerprinting methods such as DGGE and T-RFLP provide insight in the bacterial community composition, without identification of specific bacterial groups or species. Fingerprinting methods are useful for a quick comparison between bacterial community in different water samples, and have been applied to study, e.g., the effect on bacterial communities of different pipe materials (Roeder et al., 2010; Yu et al., 2010) or of water stagnation and flushing of household taps (Lautenschlager et al., 2010; Manuel et al., 2010). Recently, non-molecular techniques such as flow cytometric fingerprints have been shown to be indicative of the bacterial community composition, and have been useful for detection of bacterial community shifts after changing environmental conditions or after water distribution (De Roy et al., 2012; Koch et al., 2013; Prest et al., 2014).

16S rRNA-gene based NGS methods, including 454-pyrosequencing, Illumina, or Ion-torrent, enable taxonomic and/or functional classification of organisms present in drinking water at various phylogenetic levels. These high throughput techniques evolved rapidly in the past decade and are consequently used in increasing numbers of studies on drinking water (e.g., Hwang et al., 2012; Wang et al., 2013b, 2014; Pinto et al., 2014; Lin et al., 2014; Liu et al., 2014; Roeselers et al., 2015). 16S rRNA gene based NGS techniques also enable the study of specific bacterial groups such as ammonia-oxidizing, iron-oxidizing, or sulfate-reducing bacteria (Gomez-Alvarez et al., 2013; Sun et al., 2014).

Routine monitoring of drinking water quality is essential for surveillance of water quality. A number of chemical/environmental parameters (e.g., pH, conductivity, temperature) are nowadays used for water quality monitoring in real time with on-line apparatus. However, monitoring of microbial aspects remain conservative with the use of cultivation methods for both general water quality (HPC) and hygienically relevant organisms. Disadvantages of cultivation-based methods have been discussed in Section “How is Biological Stability Assessed?”, but essentially, results are only obtained few days after sampling, which does not allow for rapid detection of system failure (e.g., pipe leakage or break) and for immediate corrective actions. Flow cytometry in this regard is an excellent candidate as rapid method for general microbial water quality monitoring, by providing results within 15 min (Hammes et al., 2008; Prest et al., 2013). Moreover, on-line flow cytometric technologies are emerging (Besmer et al., 2014) and one could envision in future real-time monitoring of total and intact bacterial cell counts in treatment effluents and water at several locations in distribution systems. Measurement of total and free ATP concentrations are also to consider for viability assessment and can easily be implemented as routine measurement, as already applied by water utilities in the Netherlands. Collection of large data on water quality in distribution systems in space and time enables description of inherent variations (e.g., daily or seasonal) to each drinking water distribution system under normal conditions and to provide a baseline for detection of abnormal changes (LeChevallier et al., 1996; Besmer et al., 2014). Application of such an approach on long-term would provide sound basis for establishing the degree of acceptable change specific to a given distribution network, or when action is required to safeguard water quality during distribution. In such a case of abnormal change, the continuous control with flow cytometry should be complimented with other methods, such as high-throughput sequencing, or screening for specific pathogenic organisms.

One key advantage of implementing monitoring of microbial parameters in water on a high-frequency basis is the sensitivity for detection of a change in drinking water quality or characteristics. Any change in chemical/environmental property of water would result in a change in the bacterial community (cf. section “A Deeper Look into Microbial Dynamics in Drinking Water”). This change could be detected sensitively by flow cytometry, ATP or 16S rRNA-gene sequencing methods, while other methods for measuring chemical properties of water would not be sufficiently sensitive. As an example, an increase in organic components in water of 1 μg C/L would result in an increase of 10⁴ cells/mL. Flow cytometry enables the counting of bacterial cells down to concentrations of 100 cells/mL. Theoretically, this implies that FCM would enable to quantify bacterial growth occurring from the consumption of 0.01 μg C/L. In comparison, detection limits of AOC methods are usually in the range of 1–10 μg C/L, while the current analytical techniques for DOC measurements have higher detection limits (from 10 μg C/L in the case of high quality apparatus).

Combined results obtained from in situ and laboratory-scale analyses provide a basis for water treatment and distribution conditions evaluation. The extent of bacterial growth that water can support in bulk water and in biofilm can be assessed using growth potential tests without sample pre-treatment (Gillespie et al., 2014) and the BFR test. Adaptation of the growth potential test is also useful for identification of growth-limiting compounds. The gathered information provide basis for choosing adequate treatment(s) to reduce growth potential of water, by decreasing concentration of growth-limiting compound. Besides, the effect of changing operation conditions of specific treatment (e.g., contact time within biofilters) or of implementation of new treatment steps can be evaluated using the same methods. However, decisions should also be grounded on basis of on-site measurements that evaluate the extent of growth actually occurring within a specific distribution system, based on flow cytometric and ATP measurements.

The large set of methods proposed can be applied to unravel key knowledge gaps highlighted in previous sections. Particularly, high-throughput 16S rRNA-gene sequencing methods combined with quantitative methods, such as FCM and ATP, offer new opportunities to investigate drinking water ecology related questions (cf. section “How is Biological Stability Assessed?”), including the occurrence of autotrophic growth or the interactions between bulk water, sediments, and biofilms. Similarly, growth potential tests would be of specific value to investigate water limitations in inorganic nutrients. Furthermore, systematic large-scale studies applying the combined methods would provide information related to the establishment of guideline values for a degree of acceptable change in drinking water distribution systems.