To evaluate the effect of chlorination on the microbial populations present in DWDS, experimental areas with reduced chlorine concentration were defined in two cities in France and submitted either to normal chlorination or to reduced chlorination (Figure 1). The drinking water treatment plant (DWTP) supplying City#1 uses influenced groundwater as its source and includes the following treatment steps: direct filtration, ozonation, ultrafiltration, and chlorination, with a production capacity of 40,000 m³ per day for 155,000 inhabitants. The DWTP supplying City#2 is only composed of groundwater pumping and chlorination, with a mean production of 81,000 m³ per day for 220,000 inhabitants. The DWDS are composed of 96% cast iron, 2.5% iron, 1.5% other materials in City#1 and 74% cast iron, 16% polyethylene, 5% iron, 3% asbestos cement, and 2% other materials for City#2.

For each city, a limited experimental area with minimized residual chlorine concentration was defined and isolated from the rest of the network where chlorine residual around 0.2 ppm was maintained (Figure 1, dashed line). Practically, the entry of each zone consists of a drinking water tank (points C and G, respectively) where chlorine injection was turned off. The experimental zone of City#1 represents a network length of 63 km (total network length: 340 km) and served around 19,500 inhabitants while the experimental zone of City#2 represents a network length of 146 km (total network length: 1450 km) and served around 24,000 inhabitants.

Experimental periods with minimized residual chlorine were between July 21st, 2014, and January 27th, 2015, and between December 16th, 2014, and June 30th, 2015, for City#1 and City#2, respectively. After at least 1-month wash out period to allow reduction of free chlorine in the experimental area, eight in-line biofilm incubators using borosilicate glass beads (Loret et al., 2005) were installed for 2 months at each sampling point (for City#1 between November 25th, 2014, and January 19th, 2015, and City#2 between January 29th and April 15th, 2015). Each incubator was composed of a PVC compartment containing 700 autoclaved glass beads of 5 mm diameter (representing a contact surface with the water of 550 cm²) and was fed with a constant water flow of 30 L/h during the sampling period. For each sampling point, physico-chemical parameters of water quality such as temperature and free chlorine were measured on grab water samples using a thermocouple thermometer (HI8757 Hanna instruments) and ISO 7393-2, respectively. Periodically, heterotrophic plate counts (HPC) at 22°C and 36°C (ISO 6222:1999) as well as fecal contamination indicators (E. coli by ISO 9308-1:2000 standard method and enterococci by ISO 7899-2:2000 standard method) were also monitored at the sampling points.

In three biofilm points (C, D, and G), a new on-line bacteria sensor (GRUNDFOS BACMON) was used to measure in a continuous way the concentrations of total bacteria and abiotic particles in bulk water feeding the biofilm incubators. With a 10-min resolution, the system is capable of counting particles in water and classifying them as either bacteria or abiotic particles, based on shape and pattern of light diffraction (Højris et al., 2016). In sampling point C, the system was placed in the period before and during the minimizing chlorine experiment in order to follow the impact of free chlorine concentration on the bacterial biomass.

At the end of the biofilm formation period, the glass beads were collected with 500 mL of drinking water sampled at the same sampling point. After shipping within 24 h in cooling conditions, all the water volume and glass beads were sonicated in water bath (Branson 130 W) for 2 min to recover the biofilm formed at their surface. Three hundred milliliter of this suspension were sent in cooling conditions to the Institute of Microbiology (Lausanne, Switzerland), where samples were stored at 4°C and processed within 3 days.

A 0.22-μm filter was used to filter 150 ml from the water sample with as little non-soluble particle material as possible. After removal, the filter was scraped in 1 ml PBS to resuspend bacteria. DNA was extracted from 500 μl of the previous suspension with the Wizard SV Genomic DNA Purification Sytem (Promega, ref. A2361), protocol for tissue, and eluted in 50 μl TE buffer. The samples were then prepared according to the “16S Metagenomic Sequencing Library Preparation” (Part. # 15044223 Rev. B) from Illumina (San Diego, United States) using the KAPA HiFi HotStart Ready Mix (KAPA Biosystems, United States). Samples were run on gel to verify the amplification of 16S DNA. The PCR used in this study amplifies the region V3 and V4 of the 16S rRNA gene using the best primer pair proposed by Klindworth et al. (2013) and amplifies most known bacterial species: S-D-Bact-0341-b-S-17 CCTACGGGNGGCWGCAG and S-D-Bact-0785-a-A-21 GACTACHVGGGTATCTAATCC. A negative control was performed using the same protocol, starting with the filtering of DNA-free water. No amplification of 16S rRNA was obtained by PCR, and the negative control was thus not processed further. PCR products were purified and prepared for sequencing using 16S Metagenomics Sequencing Library protocol (Illumina, standard protocol). For each city, the four samples were indexed and pooled in a single run of MiSeq 2 × 300 bp for sequencing.

Between 1.6 and 3.5 million reads were obtained for each sample. The quality of raw reads was controlled with FastQC (Andrews, 2010). For each sample, pairs of reads were assembled using PandaSeq (Masella et al., 2012) (parameters: -p CCTACGGGNGGCWGCAG -q GACTACHVGGGTATCTAATCC -A simple_bayesian -l 390 -L 450 -B -N), and the resulting sequences were dereplicated using vsearch (Rognes et al., 2016) (vsearch –derep_fulllength). Sequences for all the samples were then pooled and dereplicated again filtering out singletons (vsearch –derep_fulllength –sizeout –minuniquesize 2), before clustering (vsearch –cluster_size –sizein –id 0.97), removal of chimera (vsearch –uchime_denovo –abskew 2 –sizein), dereplication of nonchimeric reads and relabeling (vsearch –derep_fulllength –sizein –relabel OTU_ –xsize). OTUs were assigned to a taxonomical unit with QIIME (Caporaso et al., 2010) using the RDP classifier (Wang et al., 2007) and the EzBioCloud database (Yoon et al., 2017) for the V3–V4 region. For each sample, the occurrence of OTUs was calculated starting from the dereplicated sequences using the global alignment search of vsearch (Rognes et al., 2016) (vsearch –usearch_global –id 0.97 –strand plus). Results were further processed using R 3.3.3 (Hornik, 2010) with packages dplyr, tidyr, dendextend, ggplot2, RColorBrewer, Vegan, and lazyeval. Distance among samples was calculated using Bray–Curtis model on the presence/absence of each OTU, and the hierarchical clustering was inferred using the Ward.d2 method. Non-metric multidimensional scaling was used to represent differences in microbial composition in two dimensions based on the presence-absence and quantification of OTUs.

A subsampling of read counts to 1 Mio was performed to normalize across samples. The number of different OTUs identified, Shannon or Simpson indexes were not correlated to the number of assembled paired-reads obtained in each sample (correlation -0.04, p-value 0.91; 0.56, p-value 0.15; and -0.66, 0.07, respectively), before or after rarefaction. Raw read counts for each OTU are available as Supplementary Table S1.

Raw sequencing reads were deposited in the European Nucleotide Archive (ENA) under the project number PRJEB27988.

The average and maximum values for free chlorine, temperature, and heterotrophic plate counts (HPC) at 22 and 36°C as monitored during the biofilm formation are shown in Table 1. Average temperatures varied from 9°C to 13°C in the DWDS of City#1. The temperature was very constant around 10°C in City#2, if we except sample H located in a public building that reached an average temperature of 18°C. In the main distribution network, the average free chlorine concentration varied between 0.11 and 0.24 mg/L in both cities. In the experimental area with minimized residual chlorine, the average concentration was reduced between 0.06 and 0.08 mg/L in DW Tank outlets, thus representing a three- to fourfold reduction compared to DW Tank outlets in the rest of the network. Furthermore, chlorine concentration was reduced to 0 and 0.03 mg/L in samples further down in the distribution network. As can be expected, HPC at 22°C values are negatively correlated to chlorine concentration (Figure 2A), but remained much lower than the binding limit value 200 cfu/ml at consumer’s tap applied in Denmark where drinking water is based solely on groundwater and distributed without disinfection residual (Miljøstyrelsen, 2017). However, HPC tests also recover microorganisms belonging to the natural and typically non-hazardous microbiota of water. Further analyses of coliforms, including E. coli and enterococci, measured in various point of the experimental area during chlorine minimization period did not show any non-conform measures of water quality.

As shown in Figure 2B, during the extended experiment period, residual chlorine values at sampling point C varied from 0 to 0.32 mg/L mainly due to a constant dose together with a varying flow at the waterwork. Free chlorine level remained in a relatively stable operation in June and July, before a decrease in August. Responses to changes in the system are slow. Bacterial counts by the BACMON show a steady but slow increase over the 1 month period with low chlorine concentration. However, an increase of chlorine concentration to higher levels (>0.2 mg/L) during only approximately 1 week was sufficient to get the system back at normal operation, with a large decrease in bacterial counts per milliliter. When comparing bacterial concentration in the distribution system at different level of residual chlorine, a statistical difference (p-value <2.2e-16) was observed between mean total bacterial counts at low (<0.1 mg/L), intermediate (0.1–0.2 mg/L), and high (>0.2 mg/L) free chlorine concentration (Figure 2C). Real-time monitoring performed under minimized residual chlorine conditions at the two other experimental points (D and G) showed stable bacteria counts along the whole period, excepted punctual variations due to hydraulic events (results not shown).

In addition to these parameters, assimilable organic carbon (AOC) determination was performed on six samples during the experimental period. AOC is usually determined to characterize the ability of water to support bacterial growth. Nitisoravut et al. (1997) proposed a threshold value of 30 μgC/l, considering the analytical sensitivity and variability of AOC determination, to ensure a limited bacteria growth in a non-chlorinated system. AOC was determined according to van der Kooij (1992). The results showed low average AOC concentrations in treated water from DWTP in City #1 and #2, respectively, around 30 and <10 μgC/L.

16S amplicon sequencing was performed to investigate biofilm composition in the DWDS with varying chlorine concentration (Table 2). Whereas sequence assignment to genus level was successful for over 95% of the dataset, species-level assignment was more challenging, due to the presence of sequences with lower than 97% identity to known bacterial species or sequences equally distant to known bacterial species. Using hierarchical clustering on the presence or absence of OTUs, we observed two major groups corresponding to the two cities (Figure 3A), indicating that each city has a signature composition of bacteria in its DWDS. This signature could also be affected by the sampling at different time of the year (Dec–Jan and Feb–April). Samples with similar concentrations of chlorine formed secondary groups in City#1 only. The large number of bacterial phyla identified in the samples suggests this hidden diversity of lowly abundant species that contribute to forming the city signature (Figure 3C).

A similar clustering taking into account the frequency of each OTU, rather grouped samples with high chlorine (A, F, B, E, and to a lesser extent G) and those with lower chlorine (C, D, H), notably reflecting the large predominance of Pseudomonas spp. in the samples with high chlorine (Figure 3D). The non-metric multidimensional scaling highlights the particular bacterial communities found in sample H (Figure 3B). Sample H from the distribution network of City#2 exhibits a large number of OTUs present in very small proportion in the sample forming a biofilm with very particular bacterial composition that could be linked to the sampling point (end of distribution network).

Samples with low chlorine harbored significantly higher bacterial diversity calculated with Shannon (p-value = 0.009502) or Simpson (p-value = 0.01435) index that account for both abundance and evenness of OTUs in the samples (Table 2). Pearson’s correlation coefficient between chlorine concentration and Shannon index was -0.65 but did not reach statistical significance due to the low number of samples considered here (p-value = 0.07987). A larger sample size would have been required to increase the power of the study and thus the significance of the observed correlation and differences. The higher bacterial diversity in samples with lower chlorine concentration comprised several clades of poorly described proteobacteria from the order Sphingomonadales. The many known and unknown species recovered in the various samples were particularly prevalent in those with lower chlorine concentration, although they were also present in small proportion in other samples.

A closer analysis of bacterial phyla identified in these water samples showed the presence of several known colonizers and potential pathogens of water environments (Table 3). Obligate intracellular bacteria belonging to the order Chlamydiales were observed in six samples mostly in very small amounts only retrieved due to the high number of reads per sample. However, sample H showed a large diversity of OTUs classified as belonging to the Chlamydiales order. Members of the family Parachlamydiaceae were the most common (35 OTUs), followed by Rhabdochlamydiaceae (17 OTUs). One OTU of an unknown Parachlamydiaceae was particularly prevalent with over 2% of the sample H reads. These so-called Chlamydia-related bacteria are known to grow steadily within amoebae, and in particular Acanthamoeba spp., that are found ubiquitely in water environments (Greub and Raoult, 2004; Kebbi-Beghdadi and Greub, 2014). The pathogenic potential of some of these organisms is still debated, but many strains are proposed to be rather environmental isolates with no, or little, pathogenic potential (Horn, 2008; Greub, 2009; Taylor-Brown et al., 2015). Known pathogens such as Legionella pneumophila or Legionella lytica were only observed as traces once (0.001% of reads), although other known and unknown Legionella species most likely also growing in amoebae were recovered at very low concentration (up to 0.3% of reads) in five samples. Their occurrence in potable water plumbing systems has been widely reported, and legionella are known to thrive in warmer and stagnant water (Loret and Greub, 2010; Jjemba et al., 2015). The higher temperature observed in sampling point H, as well as its particular situation in the DWDS could explain the higher prevalence and diversity of legionella and chlamydia observed (Table 3). Unknown Mycobacterium spp. were recovered in six different samples comprising only up to 0.1% of the sample reads. Finally, reads classified as Pseudomonas aeruginosa were present in all samples of City#1 and sample G of City#2 in very small proportion (between 0.0001 and 0.01%).