Introduction

Despite the presence of disinfectant residuals in drinking water distribution systems (DWDSs) and premise plumbing (PP) systems, these systems can be breeding grounds for (opportunistic) pathogens1. Pathogen persistence is particularly relevant in countries with aging water infrastructure, including the U.S., with over 5.6 million kilometers of distribution pipes2, most of which have not been replaced or cleaned since their construction nearly a century ago3. Europe has a similar challenge, with over 7 million kilometers of distribution pipes4. This is also a prevalent problem with intermittently operated DWDSs that are common in many parts of the world5. Intermittent release of microbes from DWDSs and particularly PP biofilms6 can present risks to consumers7 that is challenging to monitor. Pathogen release from consumer taps presents a serious public health issue. About half of the hospitalization and death incidents resulting from drinking water exposure are related to pathogens released from distribution system biofilms8. Of 187 outbreaks of drinking water-related illnesses in 28 U.S. states between 2015 and 2020, 184 were caused by Legionella and two by nontuberculous Mycobacterium9. Continuous microbial monitoring is thus key to ensuring tap water safety. However, obtaining representative samples from the both DWDSs and PP distribution systems for such analyses remains challenging10.

Drinking water microbiological investigations rely on collecting water samples (often > 8 L) through terminal faucets in homes or buildings11,12, or by sampling biofilms from often difficult to access parts of DWDSs13. Intermittent biofilm sampling methods cannot capture microbial population dynamics relevant to exposure, including potential pathogen detachment from PP biofilms during mid- to long-term operation of faucets or other outlets. While regular terminal faucet sampling can provide accurate test results of what microbes discharge from the faucet at the time of sampling, this instantaneous sampling approach may miss dynamic changes including occasional pathogen occurrence. Long-term and repeated sampling from faucets can be a robust technique for reliable monitoring of drinking water; however, this approach is labor intensive, intrusive, and impractical in many communities. Sampling from a section of the pipe or from end-of-life water meters is challenging because of the heterogeneous distribution of biofilms within the PPs/DWDSs and the long lifetimes of water meters (10 to 20 years)14 and other infrastructure components.

In recent years, the expanding adoption of at-home point-of-use (POU) devices, particularly activated carbon and reverse osmosis (RO) systems, has surged with a market value exceeding 13.5 billion USD per year. Further growth is expected at a projected cumulative average growth rate of 19%15. Approximately, 77% of US respondents to a recent survey indicated that their water undergoes some in-home filtration16. POU filters are designed for removal of pathogens and chemical contaminants and are effective to various degrees. The most common POU filters rely on activated carbon cartridges to remove chemical contaminants but are not complete barriers to microbial pathogens17. RO membranes, increasingly popular in both institutional and residential settings, excel at removing microbes and contaminants, making these suitable for untreated water sources and municipal supplies17. Periodic replacement, every 6 to 24 months per manufacturer recommendations, is vital for optimal performance of both POU carbon filters and RO membranes18.

This study considered common RO POU systems, specifically those incorporating a carbon prefilter. Carbon filters are by themselves, or combined with other filters, commonly used in pitchers, faucets, bottle stations, and refrigerator filters. The tested systems included RO membranes in addition to the carbon filter. RO POU systems typically comprise pretreatment units (including carbon filters), an RO membrane unit, post-treatment mineral cartridge, and a storage tank. Pretreatment involves removing particles, chlorine, and odorous compounds. The RO membrane efficiently removes microorganisms and most dissolved substances. Post-treatment reintroduces minerals for improving taste of the purified water. In this work, we propose a scalable method for monitoring PPs by repurposing discarded POU filter materials from homes and institutional settings. Home-use POU RO units were installed in four US research laboratories at the University of Texas Austin (UTA), Rice University, University of Tennessee at Knoxville (UTK), and Oak Ridge National Laboratory (ORNL) and operated to simulate typical household usage. We show that carbon and RO filters can accumulate PP microbiomes reaching a faucet, potentially offering more comprehensive microbial exposure data than traditional instantaneous tap water sampling. Additionally, our approach demonstrates surveillance feasibility for detecting opportunistic pathogens, such as Legionella and Mycobacteria. Collecting and analyzing the used POU filters holds promise for user-oriented DWDS/PP surveillance for mid- to long-term monitoring with minimal intrusion for monitoring drinking water safety.

In addition to directly sampling end of study carbon filters and RO membranes we installed a passive RO membrane based sampling system that operated in parallel with the POU unit at the Austin location. This passive sampler, known as a Membrane Fouling Simulator (MFS), contains parallel RO membrane coupons over which the feedwater flows and has been used to study temporal evolution of fouling propensity in RO systems. MFSs are simple flow-through devices that are operated under low pressure but these mimic the hydrodynamics of the RO membrane modules and allow easy periodic collection of membrane biofilm samples19. We used an MFS system containing 5 to 6 membrane leaves to provide access to RO biofilm samples in a time-resolved manner, and to enable studies of temporal evolution of microbial communities in the parallel POU RO systems.

The study was conducted through two sampling campaigns. The first sampling campaign involved sample collection from prefilters and RO membranes after 6 months of operation. Additionally, two MFSs containing 12 pieces of membranes (total) were sampled in parallel with the RO membrane stages at the Austin site to assess temporal variations of microbial communities in POU RO membranes during the 6-month sampling period. The RO units operated approximately twice a day, producing 5 to 10 L each time on weekdays - these conditions simulated water use patterns in residential settings. The second sampling campaign took place over a period of one month at the Austin location and included sampling from the prefilter, RO membrane, feed tap water, and a section of pipe at the end of the distribution system (preceding the POU RO system).

Results and discussion

Spatial microbial diversity in POU systems

The spatial diversity of the microbial community structure was investigated by analyzing carbon filters and RO membrane samples collected from all four locations to obtain initial information about the microbial communities (Fig. 1a). The dominant taxa recovered through these analyses showed significant differences based on the location. For example, amplicon sequences representing Afipia and Sphingomonas had a higher relative abundance in the carbon filters and RO membranes collected at the Knoxville, TN and Oak Ridge, TN locations compared to the other locations. The TN locations are approximately 40 kilometers apart from each other, and receive drinking water from different DWDSs. The genera Afipia and Sphingomonas were found in lower relative abundance in the two other locations (i.e., in Houston, TX and Austin, TX). However, the distinct water sources for the Knoxville and Oak Ridge samples, the Tennessee River for Knoxville and the Clinch River for Oak Ridge, resulted in some differences. For example, sequences representing Reyranella were abundant in the Knoxville sample whereas Methyloversatillis sequences dominated the Oak Ridge sample. The sequences with the highest relative abundances recovered from the POU RO systems operated at the two Texas locations also differed. Cyanobium sequences dominated the carbon filter from the Austin location, but sequences representing this genus were present at much lower abundances in the other sampling locations, while Rhodocyclaceae and Meytholomonas were only dominant in Houston samples. The results show the ability of biofilms collected from these POU device components to reflect differences in PP systems.

Fig. 1: Bacterial community composition differed for different locations, and POU devices could capture differences in community structure, with carbon filters and RO membranes demonstrating similar results but with important differences.
figure 1

a Bubble plot presenting the relative taxonomic abundance of the most abundant bacterial taxa across samples collected from four different locations. The size (area) of the circle is scaled to the relative abundance of that taxon. For each sample type, we used an average of the abundances detected across all samples/replicates from that sample point and the top 30 genera are shown. Carbon filters and RO membranes were sampled after running for 180 days, while the MFS samples were sampled 5 times during the 180-day sampling process. b Bray-Curtis distance analysis of microbial communities across samples visualized using a two-coordinate principal coordinate analysis (PCoA) indicates that RO samples from different locations are distinct from each other but are more similar to the carbon cartridge biofilm samples from the same location. Each symbol represents one sample. The samples are color-coded for type. The symbol shapes indicate the location of samples collected. The numbers on the axes indicate the percent of variation explained by each coordinate. The symbol shapes with days labeled are samples of MFSs from Austin collected at different time. The ellipses are hand drawn and demarcate samples from the same site.

Full size image

Figure 1b illustrates beta (taxonomic) diversity between different RO and carbon filter samples at all locations with a principal co-ordinate analysis (PCoA) plot (Fig. 1b). The Bray-Curtis distance metric was used to characterize beta diversity. The distances between samples collected from the carbon filter and the RO membrane located at Knoxville and Oak Ridge are small, indicating similarity in the microbial community composition of the source water. The underlying reason for such similarity may reflect geographical proximity; however, the two locations are on distinct DWDSs. Overall, the microbial community makeup of the carbon filters and RO membranes for each location are observed to be closer to each other when compared to the respective sample from another location. This suggests that the microbial community composition of that in the carbon filter and that corresponding to the RO membranes are similar. This is likely because they share the same water source, even though some microorganisms could permeate through the carbon filters, which have a nominal pore size of 5–10 microns. These results are consistent with our previous report of a full scale RO desalination plant, where we observed passage of bacteria through similar micron-sized prefilters and subsequent accumulation of bacteria on the RO membranes20. We propose that the microbial community makeup for both the carbon filter and the RO should be considered simultaneously to obtain a more representative microbiome of a particular PP system.

The time-resolved MFS samples obtained from Austin show a similar microbial composition to each other, as evident from the clustering of the data points for the MFS samples in the PCoA plot (Fig. 1b) and the lack of statistically significant differences in the alpha diversity values accounting for individual sample diversity (Supplementary Fig. 2). This indicates that the local feed water quality did not significantly change over the 6-month sampling period. Similarly, the MFS samples and the RO membranes in Austin exhibit similar microbial community composition. This indicated that the microbiome associated with the MFS membranes closely resembled the RO membrane samples collected at the 6-month time-point. The carbon filter microbial composition at the Austin site was similar to that of the MFS samples as well. This was expected as carbon filter microbial composition is similar to RO membrane samples and the MFS system was designed to collect microbial samples that would represent the RO microbial compositions.

POU system capture species that would likely reach the tap

Figure 2 presents data from the second sampling campaign, where a newly installed PVC-lined pipe (leading to a POU RO system) was sampled after operating for one month, with the tap water exiting the system sampled eight times during this month. Figure 2a illustrates that the dominant bacterial genera on carbon prefilters and RO membranes that remained abundant throughout the monitoring period. We propose that the microbial distribution seen on these filters could reflect the persistent dominant bacterial genera, including potential opportunistic pathogens, in tap water that would exit if such POU devices were not present. The microbial community in the carbon prefilters is similar to that of the tap water (labelled as “water” in the figure) when compared to the RO membranes and pipe samples (Fig. 2b). Mycobacterium, a genus that includes potential pathogens, is occasionally found in DWDSs worldwide21,22,23,24. This genus contributed 14% of the tap water and 12% of the carbon prefilter amplicon sequences, dominating the microbial distribution in both samples. Some Mycobacterium species, e.g., M. avium25, can cause lung disease in immunocompromised individuals26,27. Thus, regular testing for opportunistic Mycobacterium species in distribution systems, particularly PP systems, is important. Additionally, various Alpha- and Gamma-proteobacteria have been detected in both tap water and carbon filter samples. The presence of certain bacteria, such as those from the SAR11 clade and Sediminibacterium, in the RO samples can be attributed to their small size (length of < 1 μm and diameter < 0.2 μm), allowing easy passage through the carbon filter to reach the RO membrane (Fig. 2a). These bacteria have a lower relative abundance in the carbon filter but represented a higher fraction of the microbial community on the RO membrane. We propose that such small microorganisms may pass through commonly used faucet-mounted filters (which are typically carbon based) or ion exchange media filters28. Surface properties of microorganisms may also play a role in transport across media filters which are akin to depth filters29,30,31. Overall, all of the top 30 genera found in tap water are also detected in the carbon filter and/or the RO membrane. Thus, using POU RO units containing carbon filters should provide actionable information on the dominant bacteria reaching the tap within the same distribution systems with similar premise plumbing materials.

Fig. 2: Comparative sampling of distribution system pipe, tap water, carbon filter and RO membrane microbiomes indicate that carbon filter and RO samples together capture most of the dominant species reaching the tap (represented by water in the figure) over a month (8 samples).
figure 2

a Bubble plot presenting the relative taxonomic abundance of taxa across samples collected from Austin. b Beta diversity analysis of microbial communities across samples. Visualization of the degree of similarity between samples using a two-dimensional projection of their composition metric.

Full size image

Samples collected form the PVC pipe exhibited distinct microbial composition when compared to the tap water and prefilters. During the one-month experiment, Mycobacterium sequences dominated approximately 80% of the amplicon sequencing data in PVC pipe biofilm. The finding aligns with previous reports from China and Finland32,33 and from showerhead studies conducted in Austin34. Additionally, Sphingomonadaceae, typically represented by the genus Sphingomonas, are commonly found in DWDSs11,35 and were also detected. The significantly higher abundance of Mycobacterium and Sphingomonadaceae in the pipe samples indicates that direct sampling from pipe biofilms may not represent the species composition observed in tap water. This finding highlights the shortcomings of the currently used approaches as the two common modes of sampling are either tap water samples or biofilm samples from the distribution systems such as pipe or meter samples or from passive samplers that are installed as coupons within the distribution system. Furthermore, certain sequences representing Cyanobium and algal chloroplasts are detected in both the tap water and carbon prefilter, but are not found in the top 30 genera found in pipe biofilm samples. The Cyanobium sequences likely represent non-photosynthetic class of Cyanobacteria (mostly classified to “Candidatus” Melanibacteria), that have also been reported in DWDSs13.

Detection of legionella and mycobacterium

Legionella and/or Mycobacterium may persist in DWDSs32 and are of particular concern in premise plumbing systems because of their proximity to users. These were found at varying abundance levels as revealed by 16S rRNA gene amplicon analyses of both tap water and POU carbon filter samples (Fig. 2a). A follow-up analysis with targeted qPCR analyses was conducted on the samples from the second campaign to quantify gene copy levels of these potential pathogens (Fig. 3), to provide quantitative abundance data.

Fig. 3: Quantitative gene abundance analysis indicated that carbon filter samples are most similar to the tap water samples.
figure 3

Average Mycobacterium gene copies are normalized to water quantity processed and area sampled from analysis of samples collected from the Austin site. Tap water samples are the average of 8 tap water samples taken from one month sampling time. The other three points are the samples taken from the same sampling period. The ellipses are hand drawn and are solely used for illustrating similar samples. Error bars represent the standard deviation.

Full size image

Detection of Legionella in the carbon filter as well as in the monthly MFS samples from the first campaign suggest that their occurrence is frequent. The detection of Legionella on the MFS suggests that certain species can pass through the carbon filter and reach the MFS RO membrane, which was sampled monthly. However, for all the samples collected during the one-month second campaign, where we used Legionella-specific 16S rRNA gene-targeted qPCR assays, we did not detect 16S rRNA gene amplicons of Legionella above the detection limit in the samples.

We also conducted Mycobacterium 16S rRNA gene-targeted qPCR analyses of pipe, carbon filter, tap water, and RO membrane in the second sampling campaign. The gene copy numbers obtained were normalized to overall volume filtered/contacted for each of the samples and sampling area from which the biofilm was obtained or through which the water was filtered (Fig. 3). The pipe sample exhibited a significantly higher abundance of Mycobacterium as seen by the higher 16S rRNA gene copy numbers, surpassing the other samples by at least an order of magnitude. These results are in line with preferential colonization of PVC pipes by Mycobacterium32,33. Consistent with previous findings, tap water, carbon filter, and RO samples show similar abundances (within the same order of magnitude) of Mycobacterium. Although tap water samples were highly variable, qPCR analysis of carbon filter and RO membrane samples yielded Mycobacterium gene copy numbers similar to the average values for tap water samples.

Although we cannot conclude that the detected Legionella and Mycobacterium taxa are pathogenic, species in these genera are known pathogens (e.g., Legionella pneumophila36 and Mycobacteroides abscessus37). Thus, our occasional detection of Legionella and Mycobacterium sequences highlights the need for species-level monitoring of microbial communities in DWDSs, and in particular in PP systems, to ensure drinking water safety. Infections caused by opportunistic pathogens such as L. pneumophila and M. avium can lead to severe disease outcomes. In the United States, Legionella infections occur at an average rate of 10 cases per million people annually38,39, while Mycobacterium infections occur at a rate of 13.2 cases per million people, annually40. Mycobacteria are particularly challenging to deal with because of their high resistance to disinfectants, attributed primarily to its highly hydrophobic surface41. The occurrence of Legionella has been positively correlated with occurrence of Mycobacterium, and both have been detected in various DWDSs42,43,44. Controlling the concentration of residual chlorine or chloramine is effective in limiting the occurrence of these bacteria in DWDSs45. However, changes to disinfectant residuals should be balanced against the risk from the formation of disinfection byproducts. Overall, it is necessary to establish a robust monitoring system at the terminus of the distribution system, specifically targeting (opportunistic) pathogens. The proposed monitoring approach may ensure early detection and prevention of potential infections, thereby safeguarding the overall water quality and protecting public health.

Repurposing end-of-life POU filters as passive samplers

This study demonstrates that simple carbon filters, such as cartridge filters, are effective in capturing DWDS bacteria but are not complete barriers. The similarity in the microbiome of the carbon filters and tap water is more pronounced than that between tap water and pipe biofilm samples. Additionally, carbon filters and RO membranes can effectively trap a significant proportion of microorganisms that may be missed during instantaneous tap water sampling. Tap water analysis is a snapshot in time and detection may be limited because of the low abundance of target microbes (i.e., target organisms may be below the detection limit). Also, biofilms may slough off occasionally and such microbes may not be captured by occasional or even daily tap water sampling. As a result, tap water sampling does not provide an accurate indication of the microbial quality of water leaving the tap in a consistent manner. Alternatively, filters installed on systems sharing water with taps can serve as a collection system, capturing all microorganisms encountered during each filter replacement period. Although microbial analyses from POU filters per se cannot represent the precise structure of PP biofilms because of population shifts associated with differential growth after attachment, collecting and analyzing these filters could help screen for opportunistic pathogens exposure from PP. Consequently, collection of used carbon or RO membrane filters from individual households/institutions can serve as a resource for monitoring microorganisms within municipal DWDSs. Compared to the traditional instantaneous tap water sampling method, this proposed approach is non-intrusive, can be easily applied in households and institutional buildings, and provides an expanded platform for pathogen detection. This approach also holds great potential for expanding our understanding of microbial population dynamics in tap water over a medium term (6–24 months) and enhancing water borne disease surveillance efforts.

Analyzing end-of-life filters presents a valuable opportunity to gain insights into the microbial dynamics within DWDSs and premise plumbing, ultimately leading to improved drinking water safety. By collecting and analyzing information about microorganisms over temporal scales, drinking water facilities can proactively mitigate potential pathogen exposure risks through protective measures such as installation of shower filters or adjusting hot water temperatures46. Furthermore, system-wide (for e.g. city wide) monitoring of taxa can serve as an indicator for potential systemic challenges, such as inadequate disinfectant residual levels or the need for pipe replacements. While not addressed here, several POU systems are also designed to trap contaminants such as lead, arsenic, and disinfection byproducts (DBPs)17. Thus, it may be possible to extend this methodology to encompass a wider range of public health related contaminants.

Methods

POU RO systems set up and sampling

To investigate the potential of utilizing POU RO systems for sampling and identifying the presence of consumer-relevant microorganisms in PPs/DWDSs, four model number ROES-50 POU units were purchased from APEC Water (City of Industry, CA) and installed at different locations including the University of Texas at Austin (Austin, TX), Rice University (Houston, TX), the University of Tennessee Knoxville (Knoxville, TN), and Oak Ridge National Laboratory (Oak Ridge, TN). The units were reconfigured to simplify the overall design (Supplementary Fig. 1). The systems were comprised of an activated carbon prefilter stage (carbon filter) and an RO membrane stage with no post-treatment stage. The MFS membranes coupons and the tap water RO elements that were placed in the ROES-50 units were generously provided by Dupont Water (Edina, MN).

The first of the two sampling campaigns was conducted between April 2021 and October 2021, which involved sample collection from prefilters and RO membranes after six months of operation. Additionally, MFSs were installed and sampled in parallel with the RO membrane stages at the Austin site to assess temporal variations during the six-month sampling period. MFSs are simple flow-through devices that are operated under low pressure but these mimic the hydrodynamics of the RO membrane modules and allow easy periodic collection of membrane biofilm samples19. MFS operation provided a mean to evaluate temporal evolution of microbial communities in POU RO membranes. The RO units operated approximately twice a day, producing 5–10 L each time during the week days; these conditions simulated water use patterns in residential settings. The MFSs operated at a flow of 300 mL/min for 2 h every week day to simulate the same inflow with the POU RO membranes.

The second sampling campaign was conducted from June 2022 to July 2022 at the Austin site, which included sampling from the prefilter, RO membranes, feed tap water, and a section of a pipe at the end of the distribution system (preceding the RO system). Ten liters of tap water (preceding the POU filter) were sampled by filtering through Sterivex™ Filter units (EMD Millipore, Burlington, MA) twice a week (8 samples total) over the testing period and the filters were stored at −80 °C. A new section of the PVC pipe evaluated was installed at the beginning of the campaign, upstream of the POU unit. One-inch square areas of membrane biofilm and pipe biofilm samples were collected at the end of the test for processing. A comprehensive sample list is presented in Supplementary Information.

Sample analysis

DNA extraction, 16S rRNA gene amplification (V4 variable region), and sequencing followed established procedures as described elsewhere20,47. Obtained amplicon sequence variants (ASVs) were classified to different microbial taxonomic levels and represented the relative distribution of microbial genera, including pathogenic signatures (e.g., for that of Legionella and Mycobacterium). Quantitative Polymerase Chain Reaction (qPCR) assays targeting Legionella and Mycobacterium were performed in the second campaign. The qPCR assays targeting Legionella and Mycobacterium 16S rRNA genes used primers, probes, and conditions previously reported48,49. All assays were validated with synthetic linear DNA fragments (LDF) of Legionella pneumophila and Mycobacterium avium obtained from Integrated DNA Technologies (Coralville, IA). The performance criteria of all assays met the MIQE guidelines50 for reporting qPCR results (e.g., amplification efficiency [E] between 89% to 96%; standard curve slopes ranging from − 3.5 to − 3.6; R2 > 0.998; y-intercepts ranging from 36–41). The limit of detection (LOD) and limit of quantification (LOQ) were determined based on standard curves, prepared using a 10-fold serial dilution of the LDF at a concentration range of 100 to 109 per μL for both Mycobacterium and Legionella LDFs. The highest cycle threshold (Ct) value observed for a truly positive sample is assumed to be the LOD. The highest DNA dilution that results in an amplification signal within the linear range of the standard curve for each assay is considered the LOQ. The linear range for the Mycobacterium and Legionella assays is 10E + 2 to 10E + 7 and 10E + 2 to 10E + 6 respectively. The LOD and LOQ for the Mycobacterium and Legionella assays are 10E + 1 for the LOD and 10E + 2 for the LOQ51,52. The obtained DNA sequencing data were processed with Qiime 250. Data were first filtered through demultiplexing and quality control, and then used for phylogenetic and taxonomic analyses. Details are described in the Supplementary Information.