Assessing the contribution of biofilm to bacterial growth during stagnation in shower hoses

Stagnation is an inevitable outcome in the building water supply system or premises plumbing. For example, the stagnation time at shower heads, which are one of the distal ends of building water supplies, is 23.8 h/day (Proctor et al. 2018). Stagnation can lead to an increase in water temperature, the depletion of disinfectant residuals, and the leaching of organic compounds and metals from pipes. In addition, stagnation further encourages the exchange of nutrients and cells between pipe biofilm and the bulk water phase (Proctor & Hammes 2014; Zhang et al. 2015; Liu et al. 2016; Zlatanovic et al. 2017). Thus, the stagnant water becomes an ideal growth environment for microbes including opportunistic pathogens such as non-tuberculous mycobacteria (Feazel et al. 2009), Legionella pneumophila (Lau & Ashbolt 2009; Dai et al. 2018) and so on. Increment of the number of microbes in water is the most obvious outcome of stagnation.

The duration of stagnation can vary from a few hours to several days, and has different effects on bacterial growth. 12 h stagnation can cause a dramatic increase in the total cell number and drastic shift in the bacterial communities in water, as reported consistently by many studies (Boe-Hansen et al. 2002; Lehtola et al. 2007; Lautenschlager et al. 2010; Ji et al. 2015; Liu et al. 2016). However, longer stagnation time does not contribute to more increase in total cell counts (TCC) due to the exhausting of carbon and nutrients (Boe-Hansen et al. 2002; Lautenschlager et al. 2010; Liu et al. 2015, 2016; Bedard et al. 2018). Many researchers have studied the reasons for bacterial growth during stagnation. In general, the increase of microbial cells is due to the growth of microbes present initially in the planktonic phase or detached microbes from the biofilm phase (Lautenschlager et al. 2010; Ling et al. 2018). Only few studies have quantitatively investigated the contribution of pipe biofilms to bacterial growth in stagnant water. Given that biofilms represent more than 95% of the biomass in drinking water distribution systems (DWDS) (Flemming et al. 2002), most of the bacterial growth is assumed to originate from biofilm (Van der Wende et al. 1989; Laurent et al. 1993; Lehtola et al. 2007). Other studies suggested that microbial cells in the bulk water can account for a significant part of total bacterial growth in DWDS after 12 h stagnation using the total cell number count and leucine incorporation method (Boe-Hansen et al. 2002). Thus, it is still not clear to what extent biofilms and bulk-phase planktonic cells contribute to bacterial growth during stagnation, and better quantitative methods are needed. This is the first question needing to be answered for studying the mechanism of bacterial growth during stagnation.

Microbial source tracking (MST) was used to understand the origin of fecal pollution in aquatic systems in early studies (Wheeler et al. 2002; Griffith et al. 2003; Harwood et al. 2014). Traditional MST methods track the sources of fecal pollution by using fecal indicators and their biomarkers (Scott et al. 2002), which can be problematic when the markers are not entirely sourcing specific or when multiple sources within a system have similar marker concentrations (McCarthy et al. 2017). The development of sequencing-based bioinformatics tools, such as Ichnaea (Casanovas-Massana et al. 2015) and Sourcetracker2 (Knights et al. 2011), can estimate the origins of microbes more efficiently and accurately. Sourcetracker2 is a Bayesian method that can estimate the proportion of a community that comes from a set of source environments using the microbial community composition, namely the ‘microbial community fingerprint’. Sourcetracker2 was also used to estimate the proportion of each source contributing to a designated sink sample based on public dataset rather than on representative dataset (Knights et al. 2011; Henry et al. 2016; McCarthy et al. 2017; Liu et al. 2018). Thus, Sourcetracker2 can be used as an effective method to study the mechanism of bacterial growth during stagnation.

In this study, the shower hoses from a hotel were used as a model to study the contribution of biofilm to bacterial growth in stagnant water. Biomass from the stagnant water and on the pipe wall was first collected after subjecting shower hoses to different durations of stagnation (6 h, 12 and 24 h). New shower hoses were used as the control treatment. Next, genomic DNA of individual water and biofilm samples taken before and after stagnation was extracted and analyzed using 16S rRNA-based amplicon sequencing. Finally, the contributions of biofilm and fresh water were determined by Sourcetracker2, and the changes of microbial composition were analyzed based on diversity indices. This study aimed to address: ‘What is the difference in microbial composition in stagnant water after different stagnation time?’, and ‘To what extent do the biofilm and fresh water microbes contribute to the increase in cell counts in stagnant water?’ The present study explored the effect of stagnation time on microbial composition at the distal end of the building water supplies, and provided insights into the mechanism of bacterial growth during stagnation, which is expected to develop a framework for the development of strategies for managing biological water quality.

The material used in the present study were water and shower hoses. Tap water was used, which is also referred to as fresh water, and the water source of the studied drinking water distribution system is Qiantang River, Zhejiang, China. Shower hoses were plastic hoses covered by a metal shell. The new shower hoses were purchased from a supplier. The used shower hoses (150 cm long and 55 mL per hose) were collected from a local hotel in Hangzhou, China. The used shower hoses were in use for about 7 years in the hotel before the sample. During shower hose sampling, the used hoses were covered by plastic stoppers after disassembly, and transported to the laboratory in the College of Civil Engineering and Architecture at Zhejiang University at 4 °C within 2 h. Stagnation experiments were carried out in the laboratory. The geographical relationship between the water treatment plant, the local hotel and the laboratory is shown in Figure 1.

Figure 1 indicates that nine used shower hoses and three new shower hoses were used and divided into three groups. Before the experiment, new hoses were sterilized with 70% ethanol in an ultrasonic cleaner (DS-5510DTH, SXSONIC, Shanghai, China) for 45 min and then with sterilized water for 15 min. Used shower hoses were conditioned with fresh water at low velocity (about 20 mL/min per hose) for about 15 h. After closing valves at both ends, the three groups of hoses were subjected to a stagnation time of 6 h, 12 and 24 h, respectively, during experiments. After stagnation, part of the stagnant water from the hoses was collected and analyzed for heterotroph plate counts (HPC) and adenosine triphosphate (ATP) concentration, and the remainder was filtered through 0.22 μm membranes to collect microbial biomass. To prepare the next stagnation experiment, a 20-minute gentle flushing (about 20 L) was conducted. Stagnation for each experimental setting was repeated four times. Additionally, 6 L of fresh water was collected before the first stagnation for detecting water quality. As shown in Table S1 (Supporting Information), the fresh water, in which the residual free chlorine concentration (0.01 Cl⁻ mg/L) was almost zero, was used as the feed water for all the stagnation tests. At the end of the stagnation experiment, shower hoses were dissected. Biofilm on the entire inner plastic surface was sampled by sterile swab, re-suspended in sterile water and collected by filtering through 0.22 μm membranes, and then the membranes with cells were stored at −20 °C.

Different physico-chemical parameters of fresh water were analyzed. Dissolved oxygen, pH and conductivity were measured using HACH meter (HQ40d, HACH, USA) and specific probes. Free chlorine was measured by the N,N-diethyl-p-phenylenediamine (DPD) method with a HACH Test Kit (HACH, USA). Turbidity was measured by portable turbidimeter (2100Q, HACH, USA). Total phosphorus, total nitrogen and nitrate nitrogen were determined by UV/Vis spectrophotometer (DR6000, HACH, USA). Total organic carbon (TOC) was determined by thermal oxidation to CO₂ with TOC-VCPH analyzer (Shimadzu, Japan). Iron and manganese were measured by atomic absorption spectrometer (ICETM 3500, Thermo Scientific, USA).

HPC of water and biofilm samples were conducted using the spread plate method in triplicates at 25 °C for 7 days before colony enumeration (Thayanukul et al. 2013). ATP content in stagnant water was measured by BacTiter-Glo^™ Microbial Cell Viability kit (Promega Corporation, USA). A water sample (300 μL) and BacTiter-Glo^™ reagent (50 μL) were incubated at 38 °C for 1 min simultaneously in separate sterile Eppendorf tubes. The samples and reagent were mixed and then the relative light units (RLU) were measured in a luminometer (Berthold Technologies, Centro XS3 LB 960 Microplate Luminometer) after 20s (Lee et al. 2016). ATP was measured in duplicate for individual samples and converted to an ATP concentration using a calibration curve constructed with a pure ATP standard (Promega Corporation, USA).

Genomic DNA was extracted using the FastDNA^® SPIN Kit for Soil (MP Biomedicals, USA) from the membranes with cells. The bacterial 16S rRNA gene was amplified with a set of primers targeting the V3-V4 hypervariable regions (338F: ACTCCTACGGGAGGCAGCA, 806R: GGACTACHVGGGTWTCTAAT). Each polymerase chain reaction (PCR) mixture (25 μL in volume) contained 5 μL of 5× reaction buffer, 5 μL of 5 × GC buffer, 2 μL of dNTP (2.5 mM), 1 μL of forward primer (10 μM), 1 μL of reverse primer (10 μM), 2 μL of DNA template, 8.75 μL of ddH₂O and 0.25 μL of DNA polymerase (New England Biolabs Inc., USA). The PCR procedure included initial denaturation at 98 °C for 2 min, followed by 30 cycles of 98 °C for 15s, annealing at 55 °C for 30s and extension at 72 °C for 30s, and a final extension at 72 °C for 5 min. Paired-end sequencing of the amplicons (2 × 300 bp) was conducted using the MiSeq Reagent Kit V3 on the Illumina MiSeq platform (Illumina, Inc., USA) at the Personal Biotechnology Co., Ltd (Shanghai, China).

In total, 4,007,908 16S rRNA gene sequences were obtained from 12 biofilm samples and 49 water samples. The demultiplexed FASTQ files were imported into QIIME2 version 2017.12 (Caporaso et al. 2010). DADA2 pipeline (Callahan et al. 2016) was used for quality control, chimeras removal and sequence variants detection. The resulting sequences were de novo clustered into operational taxonomic units (OTU) with 97% identity using the vsearch plugin in QIIME2. Alpha-diversity and beta-diversity indices were calculated based on the rarefied OTU table at a depth of 14,473 sequences per sample (Shannon indices and weighted UniFrac distance). Permutational multivariate analysis of variance (PERMANOVA) was performed to compare differences between water group and biofilm group based on weighted UniFrac distance in QIIME2. The August 2013 97% identity Greengene Database (McDonald et al. 2012) was trained for taxonomic assignments of 16S rRNA sequences. Raw sequencing data were deposited in the NCBI Sequence Read Archive (SRA) with the accession number PRJNA529279.

The OTU tables derived from quality control and OTU picking were used as input files for Sourcetracker2 (Knights et al. 2011). For the used hoses, the communities in stagnant water were considered as ‘sink’, while the communities in fresh water and biofilm were considered as ‘source’. The sourcetracker2 analysis was performed using settings with source rarefaction depth of 14,200, sink rarefaction depth of 14,200, burn-in 100 and restart 10; default alpha1 (0.001), alpha2 (0.1) and beta (10) parameter were applied (Liu et al. 2018). The analysis was performed three times, and the average was reported. Significance analysis was conducted with Student's t-test using R software.

For biofilm samples, used hoses have higher HPC than new hoses (p < 0.01). The difference between HPC in new and used hoses is at least 3 orders of magnitude. No significant difference was found among biofilm samples with different stagnation time (Figure 2). In water samples, HPC and ATP concentrations were low in new hoses under stagnation and were similar to that in fresh water, suggesting that without the present of mature biofilms on the pipe surface, bacterial growth potential was low during stagnation. In contrast, HPC and ATP concentration were high in used hoses' stagnant water. Average HPC in 6 h, 12 and 24 h stagnant water was 7.2 × 10⁴ CFU/mL, 3.2 × 10⁵ CFU/mL and 2.7 × 10⁵ CFU/mL, respectively. After 6 h stagnation, water contained significantly lower HPC than 12 and 24 h stagnation (p < 0.01, respectively). No significant difference in HPC was detected in water samples under 12 and 24 h stagnation. The HPC in the stagnant water increased after 6 h stagnation, and longer stagnation did not lead to a further increase in HPC but reached a plateau, as reported previously (Boe-Hansen et al. 2002; Lautenschlager et al. 2010). Lautenschlager et al. (2010) found that there was about 0.44 × 10⁴ intact cells per milliliter in fresh water and about 4.90 × 10⁴ cell/mL after 12 h stagnation. Cell concentration increased at a rate of 0.37 × 10⁴ h⁻¹ until 12 h of stagnation. In the present study, HPC was 0.85 × 10³ CFU/mL in fresh water, 1.41 × 10⁵ CFU/mL after 6 h stagnation in used hoses and 5.13 × 10⁵ CFU/mL after 12 h stagnation in used hoses. HPC increased at a rate of 2.33 × 10⁴ h⁻¹ until 6 h of stagnation, and at a rate of 6.20 × 10⁴ h⁻¹ until 12 h of stagnation. These results suggested that HPC increased faster than total cell counts during stagnation.

The ATP concentration in water is used to reflect TCC (Magic-Knezev & Van der Kooij 2004; Velten et al. 2007; Hammes et al. 2010). In new hoses, fresh water and stagnant water exhibited a low concentration of ATP, and the concentration did not increase with stagnation time (p > 0.05). In used hoses, except U#3 at 6 h, other stagnant water samples taken at 6 h contained significantly lower ATP concentrations than samples taken at 12 and 24 h (p < 0.05, respectively), which is consistent with the results of HPC.

The changes in the microbial communities in stagnant water and biofilm were analyzed. As shown in Figure 3, the beta-diversity using weighted UniFrac distances-based principal coordinate analysis (PCoA) was used to compare the microbial community structure between fresh water, biofilm phase, and water phase after 6 h, 12 and 24 h stagnation in used hoses. Biofilm samples were observed to be significantly different from stagnant water samples after 6, 12 and 24 h stagnation (p < 0.01), and the microbial compositions of biofilm phase and water phase showed more similarity along with stagnation time. Based on the analysis of alpha-diversity, the Shannon indices of biofilm were observed to be higher in new hoses than in used hoses (Figure S1, Supporting Information). In used hoses, the biofilm phase has a lower Shannon value than the water phase. No significant difference in the water and biofilm phase was found for different stagnant times. The differences in Shannon index between the water and biofilm phase were mainly reflected by richness, but not evenness (p > 0.05, Figure S2 and Figure S3). Also stagnation has no effect on the evenness of the water and biofilm phase (Figure S2). In used hoses, no significant difference of observed OTU in the water and biofilm phase was found among different stagnation times, but more OTUs were observed in the water phase than the biofilm phase (Figure S3).

Figure 4 further illustrates the changes in the microbial community after stagnation at the phylum level (and class level for Proteobacteria). The stagnant water samples were dominated by Proteobacteria, especially Alphaproteobacteria. Compared with the fresh water sample, stagnant water harbored more Alphaproteobacteria and fewer rare phyla (i.e. those with an abundance of less than 1%). Notably, 6 h biofilm contained a high abundance of Gammaproteobacteria, which decreased significantly from 30.7 to 14.6% after 12 h stagnation (p = 0.008) and further decreased to 8.2% after 24 h stagnation. On the contrary, the relative abundance of Alphaproteobacteria increased significantly from 37.6 to 51.4% after 12 h stagnation (p = 0.029) and maintained 51.8% after 24 h stagnation. In addition, biofilm samples had fewer phyla than stagnant water samples, and the number of phyla in new hoses was higher than that in used hoses (Figure 4). After 6 h stagnation, the average number of phyla in water increased from 16 to 27, but decreased to 22 and 24 phyla after 12 and 24 h stagnation.

Sourcetracker2 was used to analyze the contribution of the core OTUs, defined as the OTUs with a cut-off of relative abundance >0.5% within at least one sample, including fresh water, stagnant water and biofilm samples. Figure 5(a) shows that the contribution of biofilm to core communities in stagnant water increased with stagnation time. At 6 h stagnation, an average of 5.6 and 11.2% of core OTUs were predicted to originate from fresh water and biofilm, respectively. At 12 h stagnation, 9.1 and 32.1% of core OTUs were tracked from fresh water and biofilm. At 24 h stagnation, the contributions were 11.3 and 53.5% for fresh water and biofilm, respectively. The contribution of fresh water did not show obvious change along with stagnation time, while biofilm had more influence on stagnant water after 24 h stagnation than 6 h stagnation (p = 0.023). The contribution was significantly different between fresh water and biofilm (p = 0.029) after 24 h stagnation.

When total OTUs were used to track sources (Figure 5(b)), the average contribution of fresh water dropped from 25.5% at 6 h stagnation to 17.6% at 12 h stagnation and remained at 17.7% at 24 h stagnation. The average contribution of biofilm increased from 20.1% at 6 h stagnation to 50.7% at 12 h stagnation, and further increased to 61.5% at 24 h stagnation. The contribution of fresh water did not show obvious change along stagnation time. However, the contributions of biofilm after 12 and 24 h were significantly higher than that after 6 h (p = 0.034 and p = 0.025, respectively). After 12 h stagnation, the contribution had significant difference between fresh water and biofilm (p = 0.031), and became more significant after 24 h stagnation (p = 0.021). Overall, the contribution from fresh water did not change significantly with stagnation time, whereas the contribution from biofilm increased significantly for the changes in microbial community after 24 h stagnation compared with 6 h stagnation.

Different from single marker MST, Sourcetracker2 can characterize thousands of markers simultaneously. The results of Sourcetracker2 (Figure 5) show that the contribution of biofilm to stagnant water whether using core OTUs or total OTUs increased with stagnation time, and was more than that of fresh water after 24-h stagnation. By comparing the results of source tracking based on core OTUs and total OTUs, we can analyze the mechanism of bacterial growth during stagnation. On the one hand, the contribution of fresh water to the microbial community after stagnation did not change significantly with stagnation time when core OTUs and total OTUs were characterized. However, fresh water has more influence on total OTUs than on core OTUs only under 6 h stagnation (p= 0.023). It is suggested that many rare OTUs originated mainly from fresh water in 6-h stagnant water; whereas after further stagnation some of the rare species would become detectable and even remarkable in water, which makes no difference in the effect of fresh water on core OTUs and total OTUs. On the other hand, when using core OTUs to track sources, the contribution of biofilm improved only after 24 h stagnation. However, when total OTUs were used, its contribution increased significantly after 12 h stagnation. It can be inferred that some rare OTUs in water detached from the biofilm after 12 h stagnation. Overall, according to the results of Sourcetracker2, in the early stage of stagnation, such as 6 h stagnation, rare OTUs in water are mainly from fresh water; during 6 to 12 h stagnation, core OTUs and rare OTUs are both mainly affected by biofilm and the contribution of biofilm remains high during further stagnation. As negative control, bacterial counts in stagnant water from new hoses were much lower than those from used hoses under three stagnations. This phenomenon emphasizes the important role of the presence of mature biofilms in bacterial growth during stagnation, which also responds to the results of SourceTracker2.

The situation of water stagnation in shower hoses is similar to the situation of water transportation in DWDS to consumers, both of which have long contact time between water and biofilm phase, causing the deterioration of water quality. Liu et al. (2018) used SourceTracker2 and found that 8% of core communities in tap water came from treated water, while 25% were from biofilm and loose deposits in DWDS, which also confirmed that the attached phase, including biofilm and loose deposits, has more effect on the core communities in stagnant (or transported long-distance) water than fresh water.

Residual chlorine is a key factor to control microbial growth in water. In the present study, the chlorine residual in fresh water was almost zero (<0.05 Cl⁻ mg/L, see Table S1), which could not effectively suppress bacterial growth during stagnation. In the present study, HPC in stagnant water from new hoses did not have significant difference among three stagnation times (p > 0.05), but HPC in stagnant water from used hoses did show significant increase along with stagnation time. Van der Wielen et al. (2016) found that HPC, Aeromonas, Mycobacterium and fungi were more reliable indicators than ATP and cell number to monitor bacterial growth in unchlorinated drinking water. However, in the present study, stagnation did not have similar effect on the relative abundance of Mycobacterium. The relative abundance of Mycobacterium was averagely 0.74, 1.32 and 0.49% for 6 h, 12 and 24 h stagnant water, respectively. Compared with 12 h stagnation, its relative abundance reduced significantly after 24 h stagnation (p < 0.05). Different from HPC results, the relative abundance of Mycobacterium did not increase with stagnation time but decreased after 24 h stagnation. So, Mycobacterium could not be an indicator of bacterial growth under stagnation. However, Mycobacterium was highlighted in shower facilities biofilm (i.e. the shower head and shower hose), as well as Legionella, Pseudomonas, Nocardia and Burkholderia (Feazel et al. 2009; Proctor et al. 2016). These genera include opportunistic pathogens which can cause respiratory or wound infection. The relative abundance of Mycobacterium in fresh water was 1.01%, stagnation did not increase its relative abundance. The relative abundance of Legionella was 0.00, 0.04, 0.01 and 0.43% for fresh water, 6 h, 12 and 24 h stagnant water, respectively. The relative abundance of Pseudomonas was 0.10, 0.79, 0.26 and 0.35% for fresh water, 6 h, 12 h and 24 h stagnant water, respectively. The relative abundance of Nocardia was 0.00, 0.02, 0.00 and 0.00% for fresh water, 6 h, 12 h and 24 h stagnant water, respectively. The relative abundance of Burkholderia was 0.00, 0.02, 0.00 and 0.02% for fresh water, 6 h, 12 h and 24 h stagnant water, respectively. The results show that stagnation time has no significant effect on four genera's relative abundance and gene copy number (data not shown). Previous studies showed different conclusion. Wang et al. (2014) found that water age (1 d–5.7 d) has a significant positive effect on the gene copy number of Mycobacterium and Pseudomonas aeruginosa in drinking water. This result may be caused by shorter stagnation, so that the genera have not begun to show a trend of increasing abundance. But compared to fresh water, the relative abundance of Pseudomonas increases two to seven times, and Burkholderia and Nocardia became detectable after stagnation.

Overall, 3–16 phyla appeared after stagnation in water, but no significant difference was found among different stagnation lengths. Lautenschlager et al. (2010) found that 20–100% of the microbial community changed after overnight stagnation in a non-chlorinated water system. Ji et al. (2015) also reported that 6–13 phyla appeared after 8 h stagnation compared to the microbial community in influents. Likely the increase was due to the potential seeding from biofilm or the growth of rare species that were not detected initially due to their low abundance. In this study, 25–100% of the phyla that appeared after stagnation in water were due to the growth of the rare species initially present at low abundance, suggesting that they were not detected in fresh water and biofilm samples but in stagnant water samples. The number of phyla in new hoses was higher than that in used hoses due to the microbial succession and maturation of biofilm (Martiny et al. 2003).

The present study further showed that stagnation could affect microbial diversity and led to changes in microbial community structure, in particular, due to the contribution from biofilm. Based on the average Shannon index, stagnant water samples generally contained lower microbial diversity than fresh water (Figure S1). Significant differences in microbial diversity were observed between 6 h and 24 h stagnant water (p < 0.01) as well as between 12 h and 24 h stagnant water (p < 0.05). These observations were also reported in a previous study after water was subjected to stagnation (Boe-Hansen et al. 2002). The decrease in microbial diversity with the 24 h stagnant water was due to the growth of dominant bacteria like Proteobacteria and Planctomycetes. Water samples deviated further away than biofilm samples along with longer stagnation (Figure 3), indicating that stagnation had more impact on microbial abundance and diversity in water than that in biofilm. After 12 h and 24 h stagnation, water samples and biofilm samples clustered closer than after 6 h stagnation.

Theoretically, the microbial cells present in stagnant water should come from either fresh water or biofilm. However, a significant proportion of contributions to the bacterial growth in stagnant water remain unknown. One reason is that there was only one fresh water sample used in source tracking analysis. Sequence depth can also be a limitation for an unknown source, OTUs with low abundance but that become detectable after stagnation might be identified from the unknown source. In this study, the shower hoses were transferred from a local hotel 6.0 km away from the laboratory and conditioned with fresh water at low velocity (about 20 mL/min per hose) for about 15 h before the experiment. Previous studies (Pryor et al. 2004; Stanish et al. 2016) showed that the new water environment could change the microbial community in biofilm. This could introduce some uncertainty to the present results.

The findings in this study indicate that the HPC in stagnant water increased during the 6 h and 12 h stagnation, but longer stagnation time did not lead to a further increase. The same trend was observed in ATP concentration, 6 h stagnant water contained significant lower ATP concentration than 12 h and 24 h. The microbial diversity in stagnant water significantly decreased after long-term stagnation. About 3–16 more phyla were observed in water after stagnation, 25–100% of them were due to the growth of the rare species, which were not detected in water before stagnation. The main contribution to the increase in microbes under 6 h stagnation is the growth of microbes present initially in fresh water, while biofilm contributes more after 12 h and 24 h stagnation. These findings provided insights into the mechanism of bacterial growth during stagnation, and could serve as a framework for future development of strategies to manage biological water quality at the distal end of the building water supplies. Last but not least, consumers are suggested to keep water running for a while before using shower hoses after a period of stagnation, in order to reduce the exposure to microbes.

The authors would like to acknowledge the financial support from the National Natural Science Foundation of China (51678520), the Major Science and Technology Program for Water Pollution Control and Treatment (2017ZX07201004), and the Joint UIUC-ZJU Research Program. The authors would also like to thank Limao Dong from Zhejiang University for providing assistance in sampling.

All relevant data are included in the paper or its Supplementary Information.