At the DKU Environmental Research Lab, all samples were stored in a −80°C freezer prior to processing. For each sample, the filter membrane was removed from the sealed syringe filter unit and cut into eight strips using a sterile razor blade on a disposable petri dish. DNA extraction was then conducted using the Qiagen DNeasy Mini Kit, following the manufacturer’s instructions, with an additional bead-beating step to enhance extraction efficiency. In brief, the filter strips were placed into a 2 mL microcentrifuge tube, where 0.2 g of 0.1 mm Zr beads and 400 microliters of lysis buffer AP-1 were added. The mixture was then beaten at 2,000 rpm for 5 min.

For PCR amplification, the V4 region of the 16S rRNA gene was amplified by the universal primer pairs 515F (5′-GTGCCAGCMGCCGCGGTAA-3′) and 805R (5′-GACTACNVGGGTATCTAAT-3′) with dual barcode index and heterogenous spacers (Lin et al., 2019; Kozich et al., 2013). KAPA HiFi PCR Kit and the manufacturer’s protocol were adopted. All PCR reactions were performed in triplicates with 25 μL of each reaction mixture. Agarose gel electrophoresis was then performed to visualize amplicon fragments, and PCR products were purified using the QIAquick PCR Purification Kit (Qiagen) following the manufacturer’s instructions. The purified PCR amplicons from each sample were dissolved in 30 μL of elution buffer and stored in a −80°C freezer.

Finally, the concentration of the purified PCR amplicons was measured using a Qubit™ 4 Fluorometer. Equimolar amounts of purified amplicons were pooled together and sent to Genewiz in Suzhou for an Illumina Miseq (250 PE) sequencing run.

Out of the 50 drinking water samples, 40 passed our quality control and they were collected from 27 administrative regions across China (Table 1). Ten samples were excluded from the study because of either lab processing errors or improper handling during shipping and/or storage (e.g., all ice packs melt), which may have compromised DNA quality. Among the 40 valid samples, 29 samples showed positive PCR signals which were collected from 16 cities in 10 provinces, 4 municipalities, and Macau (Figure 1). Notably, all three tap water samples collected in Zhejiang Province, including the post-typhoon Hangzhou sample showed no PCR signal, suggesting that microbial contamination was below the detection limit of the PCR approach for the 1-liter water samples. This result may reflect the high quality of source water in Zhejiang Province (Han et al., 2020) and effective drinking water treatment in developed cities such as Hangzhou and Ningbo. Alternatively, factors such as new/clean plumbing systems, suitable plumbing materials that do not support the growth of microorganisms, and shorter water stagnation time in the plumbing could also account for the absence of detectable microbial DNA (Ley et al., 2020; Ji et al., 2015; Ling et al., 2018).

This study is among the few investigations on the tap water microbiome conducted through citizen science, which can serve as a proof of concept for national-scale microbiological monitoring of tap water using citizen science and show its competitive advantages compared to non-citizen science sampling methods.

Firstly, the study demonstrates the cost-effectiveness and extensive coverage of the approach across China. With decentralized volunteer participation and affordable sampling kits, the sampling sites spanned 31 regions and various seasons, ensuring spatial and temporal diversity. It is worth mentioning that the use of Corning^® syringe filters (~$2.5–4/unit) (Corning, 2023) reduced costs by 1 to 1.5 times compared to the traditional MilliporeSigma^® Sterivex filters (~$8–13/unit) commonly used for aquatic microbiome sampling (Millipore Sigma, 2023). Also, our DNA extraction protocol was specifically optimized for this low-cost filter. Additionally, this citizen science sampling approach significantly reduces travel and personnel expenses compared to traditional fieldwork methods. While the lack of strict oversight during the volunteers’ sample collection is a limitation, the high validity, with 40 out of 50 drinking water samples passing quality control, and the consistency in results discussed in the following sections support the approach’s applicability and reliability.

Furthermore, due to the flexibility of this sampling approach, we conducted timely opportunistic sampling of extreme weather events—specifically, Typhoon In-Fa and 2021 Henan Floods—by leveraging local volunteers in Hangzhou, Changzhou, and Zhengzhou. This citizen science approach proved to be particularly fast-responding and effective given the limited funding and the urgent nature of these events. Additionally, the sample storage tests (Supplementary Text 1.1) confirmed that maintaining samples for 7 days at −18°C did not generate false positives in the absence of bacteria during collection, ensuring accurate pathogen identification and reliability of using household freezer for storage, thereby enhancing the flexibility and feasibility of the citizen science sampling approach.

The DNA of microorganisms in 29 household drinking water samples were successfully extracted, PCR amplified with 16S primers, and submitted for Illumina sequencing. Following the demultiplexing process, five samples did not yield significant reads (all < 50 per library) and were excluded from the dataset. In addition, two samples collected after the 2021 Henan Floods were excluded after quality filtering and chimera checking due to having relatively low reads (<3,000 per library). Analysis of the sequencing results indicated that samples with low read counts were primarily associated with specific primer sets, suggesting that the reduced output was likely due to primer efficiency or compatibility issues rather than issues with DNA quality. As a result, the 16S rRNA gene amplicon dataset in this study comprised a total of 22 samples.

In DADA2 quality filtering, the proportions of output and input read for 13 (59.1%) samples were lower than 50%, indicating low-quality raw DNA reads. This may be due to (1) potential degradation of DNA samples during sampling, shipping, or DNA processing; and (2) residual DNA from dead bacteria cells, which are harmless to humans. After quality filtering and chimera removal, the sequencing reads per sample ranged from 9,794 to 150,656, with an average of 60,884. From the resulting dataset, 7,635 ASVs were identified (the most abundant ASVs are provided in Supplementary Text 1.2 and Supplementary Table 2.1).

Fifty two microbial phyla including 46 bacterial phyla and 6 archaeal phyla were detected in the household drinking water samples (n = 22). In all samples, 9 bacterial phyla and 1 archaeal phylum accounted for over 90.7% of total taxonomically assigned reads at the phylum level (Table 2). All 10 phyla are both abundant and prevalent (i.e., occur in 91–100% of all samples), indicating a relatively even microbial distribution pattern at the phylum level in the samples. The five most dominant bacterial phyla were Proteobacteria (mean percentage ± SD: 55.0% ± 19.8%), Planctomycetota (10.5% ± 7.9%), Acidobacteriota (7.0% ± 6.0%), Actinobacteria (5.9% ± 6.6%), and Cyanobacteria (4.7% ± 3.5%), comparable to a previous study in China (Han et al., 2020). Four of them were reported to be tolerant to water treatment and distribution processes, except for Acidobacteriota (Han et al., 2020; Godinho et al., 2024). Interestingly, this bacterial composition differs from the primary bacterial assemblages revealed by several highly cited studies conducted in other countries, including the United States and Portugal, which predominantly feature Proteobacteria, Actinobacteria, and Bacteroidetes (Hull et al., 2017; Pinto et al., 2012; Ivone et al., 2013). This may imply a unique microbiome intrinsic to China’s drinking water systems.

The phylum-level taxonomic composition for each sample is detailed in Figure 4 and Supplementary Table 2.2 (the most abundant classes are summarized in Table 3). In this study, Proteobacteria (classes α-and γ-Proteobacteria) was the most predominant phylum in 20 samples and the second most prevalent in others. It accounts for 29.9–99.0% of the reads in each sample, with the highest relative abundance (RA) detected in Huizhou (Huizhou_0219). Among all the samples, Sphingomonas was the most abundant genus of Proteobacteria. This result was consistent with a previous study (Han et al., 2020) which found Proteobacteria to be dominant in tap water collected mainly from central and eastern China, and the genus Sphingomonas grew during chlorination (Jia et al., 2015) or monochloramine treatment (Chiao et al., 2014).

However, two samples, Changzhou_0220 (Changzhou, Feb. 20) and Nanchang_0210 (Nanchang, Feb. 10), were dominated by Crenarchaeota, a common archaeal phylum. Specifically, Crenarchaeota accounted for 33.3 and 35.0%, with the genus Candidatus Nitrosotenuis (32.0%) and Candidatus Nitrosotalea (34.9%) being most abundant in samples Changzhou_0220 and Nanchang_0210, respectively. This indicates that, in addition to a variety of bacteria, archaea can also grow in tap water, which is supported by studies that have detected the archaeal phylum Crenarchaeota in drinking water distribution systems and drinking water-related environments (Roeselers et al., 2015; Inkinen et al., 2021; Dai et al., 2020; Franca et al., 2015; Bautista-de los Santos et al., 2016). In particular, Inkinen et al. (2021) found a high abundance of archaeal reads from the genus Candidatus Nitrosotenuis and Candidatus Nitrosotalea in drinking water distribution systems supplying non-disinfected waters. This suggests that the disinfection processes of samples Changzhou_0220 and Nanchang_0210 may be less effective compared to others.

Interestingly, compared to the winter Changzhou sample Changzhou_0220, the pre-typhoon Changzhou_0712 and post-typhoon samples Changzhou_0728 collected from the same urban household in July were much more similar in overall composition at the phylum level, indicating a seasonal effect. However, the post-typhoon sample Changzhou_0728 exhibited higher levels of Actinobacteria (increased from 2.2 to 7.1%) and Cyanobacteria (increased from 5.2 to 14.9%), which are phyla containing potential waterborne pathogens and the species that produce cyanotoxins, respectively. Elevated levels of the pathogen Mycobacterium spp. (more details in Section 3.5) as well as toxin-producing Cyanobacteria spp. were observed in post-typhoon sample Changzhou_0728. Specifically for cyanobacteria, Microcystis spp. had a higher RA, while Cylindrospermopsis sp. and Dolichospermum sp. appeared after the typhoon event. Other toxic species of Cyanobacteria, including Aphanizomenon sp. and Anabaena sp., were detected in normal samples collected from Shanghai, Lanzhou, Xi’an, and other locations. Many Cyanobacteria spp. from those genera can produce a variety of cyanotoxins such as Microcystins and Cylindrospermopsin, which can cause liver and kidney damage and have potential carcinogenicity (EPA, 2022). Similarly, a substantial rise in the RA of Cyanobacteria was reported in treated water samples collected from a drinking water treatment plant in Jiangsu Province after Typhoon Lekima in August 2019 (p < 0.05) (Tang et al., 2021). In this study, although the RA of Cyanobacteria in the post-typhoon sample Changzhou_0728 remained detectable, it decreased to the pre-typhoon level by the third day after the typhoon event.

In total, six bacteria genera containing pathogenic species and three pathogenic species were detected in all the PCR positive samples (n = 22) (Table 4). Five genera and one species that occurred in more than 30% of the samples (7 samples) were categorized as common pathogens, while the rest were grouped into rare pathogens (Supplementary Table 2.3). It is worth noting that the overall pattern of pathogens in the normal samples remains unchanged when the two post-weather samples are included. This finding suggests that pathogen contamination in tap water could be a widespread phenomenon (Ley et al., 2020; Ma et al., 2022).

The distribution of these potential pathogens within each water sample is detailed in Figure 5, and the BLAST+ results of the potential pathogenic ASVs are provided in Supplementary Table 2.4. The mean RA of common pathogens in all tap water samples ranged from 0.02% to 2.67% (normal samples: 0.02%–1.95%), while that of rare pathogens was extremely low. Mycobacterium spp. (mean RA 2.67%, including rainfall events), Acinetobacter spp. (1.43%), and Legionella spp. (0.46%) occurred in all the samples while Leptospira spp. (0.02%) were found in half of the samples. Notably, Escherichia coli ASVs (0.06%) (E-value = 6e-132, Percent identity = 100%) were detected in 36.4% of the samples. In addition, two Brevundimonas species, B. vesicularis and B. diminuta, which are particularly recognized as emerging global opportunistic pathogens (Ryan and Pembroke, 2018), were detected in the majority of samples (90.9%). Compared to normal samples, three more pathogenic bacteria species were detected in the post-weather samples, including Salmonella enterica, Brevundimonas diminuta, and Aeromonas hydrophila.

E. coli, a common fecal indicator bacteria (WHO, 2017), was widely detected in this study. The tap water samples with the highest RA of E. coli were Changzhou_0728 (RA: 0.61%), Beijing_0620 (0.51%), Zhengzhou_0802 (0.05%), Changzhou_0712 (0.03%), and Jinan_0219 (0.02%). The elevated RA of E. coli in the two post-weather samples suggests that the contamination might be related to the extreme rainfall events. Notably, the proportion of E. coli in Beijing_0620 was dramatically higher than that of other normal samples. It is worth mentioning that traditional E. coli tests can be a generally reliable indicator of enteropathogenic serotypes in drinking water; however, potentially viable but non-culturable E. coli cells could result in underestimations of actual water contamination (Liu et al., 2008). Therefore, it is recommended to use PCR or quantitative PCR (qPCR) methods for the monitoring of E. coli (Wolf-Baca and Siedlecka, 2019).

Additionally, the RA of total Legionella spp. was most abundant in Zhengzhou_0802 (1.32%), Changzhou_0220 (1.25%), and Zibo_0502 (1.09%). Almost all species in the genera Legionella are thought to be potential human pathogens, but L. pneumophila (on Contaminant Candidate List 5 - CCL 5) is the leading cause of Legionnaires’ disease (pneumonia) and Pontiac fever (a milder infection) (WHO, 2017; EPA, 2023). Potential L. pneumophila ASVs were detected in 22.7% (Ramirez-Castillo et al., 2015) of the tap water samples. Among those samples, the highest L. pneumophila RA was observed in Zhengzhou_0219 (RA: 0.237%), followed by Macau_0524 (0.035%), Nanchang_0210 (0.019%), Xi’an_0217 (0.011%), and Shanghai_0411 (0.010%). Moreover, some other pathogenic species such as L. oakridgensis and L. maceachernii occurred in Tianjin_0715, Lanzhou_0822, Xi’an_0217, Lijiang_0722, and Zhengzhou_0802.

Salmonella enterica (ASV 3082, RA: 0.39%), a highly pathogenic species, was detected in the post-flood tap water microbiome from Zhengzhou (Zhengzhou_0802), despite its known susceptibility to disinfection. However, the potential health risks remain uncertain, as the severity of the disease depends on the serotype and host factors of Salmonella (WHO, 2017). Nonetheless, the presence of Salmonella enterica after the 2021 Henan Floods indicates potential contamination in the household drinking water after an extreme weather event. While this could suggest fecal contamination, alternative sources, such as residual contamination from washing raw meat in kitchen sinks, cannot be ruled out.

Spearman’s correlations between potential pathogens and total microbiome alpha diversity indexes (Chao1 and Shannon) are shown in Figure 6. When the influence of extreme rainfall events was excluded (n = 20), the RA of Brevundimonas spp. showed strong positive correlations with multiple pathogens and alpha diversity indexes (Supplementary Table 2.5), especially with Mycobacterium spp. (r = 0.84, p < 0.001) and the total RA of potential pathogens (r = 0.67, p < 0.05). These findings suggest that Brevundimonas spp. may serve as a useful ecological indicator of microbial risk in tap water systems.

Interestingly, we also observed that the presence of E. coli was significantly associated with lower alpha diversity of the microbial community (Chao1: r = −0.24, p < 0.05; Shannon: r = −0.21, p < 0.05). Reduced alpha diversity may reflect microbial imbalance or stress conditions that favor pathogen persistence. Understanding the relationship between indigenous water microbiomes and opportunistic or fecal pathogens, such as E. coli, could therefore provide insights into the ecological conditions that support pathogen survival in tap water.

While some correlations may be driven by low sample detection frequencies or shared habitat traits, the broader pattern highlights the value of co-occurrence analysis in generating hypotheses about ecological interactions or stress responses in tap water microbiomes. We acknowledge these are exploratory findings, and future studies incorporating environmental covariates (e.g., chlorine levels, pipe material, water age) would help validate these relationships.

The citizen science sampling procedure exhibits several limitations and could be improved in the future. First, to further minimize bias during the sample collection process, closer supervision of the volunteers (e.g., through cell phone video recording) and duplicate sample collection from the same location would be beneficial. Second, systematic time-series sampling from the same location is necessary to understand the temporal patterns of microbes. Additionally, to pinpoint contamination sources, future studies could sample water treatment system effluents to determine if contamination originates from treatment or pipeline issues. Moreover, collecting a comprehensive set of metadata (e.g., water temperature, nutrients, pH) associated with microbiome sampling will help reveal the environmental factors shaping the drinking water microbiome.

To improve the success rate of sample collection, it is essential to implement stricter controls on shipping temperatures (e.g., adding a reusable temperature logger to the sampling kit) and to collaborate with shipping companies. This collaboration will help reduce sample transportation costs, a critical factor in expanding the citizen science outreach beyond university affiliates. Meanwhile, developing an online platform could help disseminate the research results to the public, thereby promoting science education and citizen engagement.