A syringe filter experiment was performed following a previously reported protocol (Wang et al., 2007; Figure 1). Commercially available (pre-sterilized) syringe filters with varied pore sizes were used: polyethersulfone filters (PES 0.2 and 0.45 μm) and polyvinylidene filters (PVDF 0.1 μm) after a preliminary optimization for testing different materials of syringe filters with the least background contaminant (Supplementary Figure S1; Supplementary Table S1). Evian bottled water served as a reference/control. Total and intact bacterial cell concentrations were measured in triplicates with flow cytometry-based staining using SYBR Green I and propidium iodide as previously described (Prest et al., 2013; Farhat et al., 2018). In short, 700 μL of water sample were mixed with 7 μL SYBR Green I for the total cell, or SYBR Green I (1:100 dilution in deionized water) and PI (4 μM final concentration) for the intact cell measurement. Stained samples were incubated for 10 min at 35°C in the dark and then measured with Accuri C6 Plus FCM (BD Biosciences, Accuri).

Designated premised taps (Water Desalination and Reuse Center, King Abdullah University of Science and Technology, Kingdom of Saudi Arabia) were used for sampling RO tap water throughout this study (Supplementary Figure S2). The RO tap water is produced by the KAUST seawater reverse osmosis plant with details on the process being described elsewhere (Farhat et al., 2020, 2022). The RO tap water is a mixture of RO permeate that is dosed with chlorine, CO₂, and lime and then stored in a storage tank before being distributed throughout the KAUST distribution network. We sampled the final distributed RO drinking water from the premised taps that are part of the distribution network. A one-liter sampling volume was used for biomass filtration and subsequent DNA extraction and collected using an autoclaved narrow-mouth polypropylene bottles (Nalgene, NG-2203-0010) for each sample groups (with replicates). Sample groups analyzed were: control, pre-flush, post-flush, incubated, and Evian bottled water. Pre-flush samples were taken from the first liter of water with exact stagnation durations unknown. Post-flush samples were taken after 5 mins of flushing (network quality water). Incubated samples were post-flush water, dechlorinated, and incubated to reach 10⁴–10⁵ cells/mL with naturally assimilable organic carbon contents (no nutrient spiking) for 2–3 days at 30°C (temperature for bottled water heterotrophic bacteria). Evian bottled water was used as a reference with an average cell density of 10⁴–10⁵ cells/mL. All samples were filtered for biomass and subsequently, DNA extraction, with a minimum of two biological replicates processed (the number of replicates ranged from two to eight), except for the control (blank) group which consisted of a virgin sterile membrane, a membrane filtering 1 L of Milli-Q water, and a membrane filtering 1 L of sterilized (autoclaved), incubated RO tap water for each different membrane types. A total of 1 L × 63 samples were processed for DNA extraction.

Five types of filter membranes were used for biomass filtration: Mixed ester cellulose (MEC, 0.2 μm, Millipore, GSWP04700), Polycarbonate (PC, 0.2 μm, Isopore, GTTP04700), Polyethersulfone (PES, 0.1 μm, Sartorius, 2250943), PES Sterivex (0.2 μm, Merck, Sterivex, SVGP01050), and Polyvinylidene difluoride (PVDF, 0.1 μm, Durapore, VVLP04700). Except for PES Sterivex 0.2 μm which is a pressure sterile filter unit, each membrane had a 47 mm diameter and was sterilized (autoclaved) before use. Membrane (biomass) filtration was done using a peristaltic pump Masterflex I/P (Model 77602-30, Cole-Parmer, United States) connected with a Masterflex tubing (Model 6437-73) and sterilized (autoclaved and rinsed thoroughly with MilliQ) Swinnex Millipore Unit (47 mm diameter), except for the PES Sterivex filter. The tubing and Swinnex filtration head were fastened with an adjustable plastic clamp and Teflon tape to prevent leaking. The filtration flow rate was approximately 0.04 L/min. For the PES Sterivex 0.2 μm filter unit, a portable peristaltic pump (Vampire Sampler, Buerkle) was used at an approximate flow rate of 30–50 mL/min.

DNA was extracted using the DNeasy PowerWater Kit (Qiagen, Germany) following (Brandt and Albertsen, 2018) recommendation. In-house modification of the optimized DNeasy PowerWater extraction method (Vosloo et al., 2019) for low-biomass water (10³–10⁴) cells/mL was adopted throughout this study that involved mechanical, enzymatic, and chemical lysis (Vosloo et al., 2019). The optimized protocol enhances the bead-beating cell disruption method from the DNeasy PowerWater kit by involving physical, chemical, and enzymatic lysis. In this study, a different bead-beater machine was used in comparison to the original protocol (Vosloo et al., 2019), i.e., MBB-24 Mini beadbeater (Biospec, United States), and thus 2 mins cell disruption time with 3,400 rpm was used based on the manufacturer’s recommendation to maximize cell lysis. The modified DNEasy PowerWater protocol was able to increase the DNA yield of Evian water (10⁴–10⁵ cells/mL) 6-fold (Supplementary Table S3A) and 9.6-fold compared to the DNA PowerSoil (Supplementary Table S3B), ensuring high DNA yield for the subsequent process (e.g., DNA purification) for amplicon library preparation. Furthermore, the PowerWater-modified has a less preferential bias in regards to low abundance OTUs detected and best captures the overall bacterial diversity compared to DNeasy PowerSoil kit based on our preliminary analysis (Supplementary Figure S6). Multiple blanks (as mentioned in section 2.2) and negative control (no template DNA/reagent) were processed in parallel for DNA extraction. It is also beneficial to spike the sample with known bacterial cell culture to test for DNA extraction efficiency (Brumfield et al., 2020; Knupfer et al., 2020). However, the effectiveness of the current DNA extraction method without such internal bacterial culture is also reflected in the resulting bacterial diversity (compared to DNAEasy PowerSoil; Supplementary Figure S6), indicating the current extraction protocol’s ability to open and recover genetic information from bacteria of varying hardness (La Duc et al., 2008).

Extracted DNA concentrations were measured with a Qubit High Sensitivity assay kit (Invitrogen, Thermo Fisher Scientific, Germany) using a Qubit 4.0 Fluorometer (Thermo Fisher Scientific). Briefly, a volume of 190 μL Qubit working solution was mixed with 10 μL extracted DNA. The detection range of DNA concentration with Qubit HS assay is between 0.01 to 100 ng/μL. Total DNA yield was calculated based on the measured Qubit DNA concentration times the volume of the final elution buffer used (50 μL). All of the DNA extracts were sent to the DNASense Laboratory (Denmark) for sequencing and the size and purity of the sequencing libraries were checked with Tapestation 2,200 and D1000/High sensitivity D1000 screenTape assay (Agilent, United States) that produce the gel electrophoresis image of the DNA analysis. Samples with DNA yield below the detection limit were not processed for further analysis.

Two dedicated taps were chosen for this experiment: tap 1 and tap 2 (Supplementary Figure S2) which were subjected to two different stagnation periods: 14 and 30 days. After the intended stagnation periods were reached, samples were taken for pre-flush, post-flush, and incubated samples (see 2.2. DNA extraction with various membrane filters section). Samples were processed for multiple parameter analysis consisting of abundance and activity measures: total and intact cell concentration (TCC/ICC), Adenosine triphosphate (ATP), and 16S gene copy number.

TCC and ICC were done with flow cytometry-based quantification using SYBR Green I and propidium iodide (PI) staining as described above (see section 2.1). Adenosine triphosphate (ATP) analysis was measured using a luminometer according to the manufacturer’s protocol (Celsis Advance, Charles River Laboratories, Inc., United States). TCC and ICC measurements were done in technical triplicate and ATP was done in technical duplicate. qPCR analysis of the 16S rRNA gene was done by DNASense Lab (Denmark). Primer pairs, amplification conditions, and DNA standard used are described (see Supplementary Method S1).

To detect the significance of differences in each of the measured parameters (DNA yield, and additional parameters: TCC, ICC, ATP, 16S gene copy number) among different sample groups, a two-way analysis of variance (ANOVA) was carried out among the following fixed factors for each experiment: (i) DNA extraction with various membrane filters: “Group type/cell density” (five factors: control, pre-flush, post-flush, incubated, Evian), “Filter membrane type” (five factors: MEC 0.2 μm, PC 0.2 μm, PES Sterivex 0.2 μm, PES 0.1 μm, PVDF 0.1 μm); (ii) controlled stagnation experiment: “Tap location” (two factors: Tap 1 and Tap 2), “Group type/cell density” (four factors: control, pre-flush, post-flush, incubated). The data were previously checked for normality and homogeneity of variance using a visual check for homogeneity of variances (residuals vs. the fitted values) and the normal probability plot of residuals (Mangiafico, 2015; Holbert, 2022). All statistical analyses were done using R within the R studio environment (version 4.3.1). For post-hoc multiple comparisons, the Tukey Honest Significant Differences (TukeyHSD) test was used (Mangiafico, 2015).

All samples from the controlled stagnation experiment were processed for 16S rRNA gene-based high-throughput sequencing on a MiSeq platform (2 × 300 bp; Illumina, United States). Library preparation and sequencing were carried out by DNASense (Denmark) following their Illumina custom protocol (Illumina, 2013, 2021). The primer pairs covering V3-V4 (bV34A) regions of the bacterial 16S rRNA gene were used: [341F] CCTACGGGNGGCWGCAG and [806RB] GGACTACNVGGGTWTCTAAT (Klindworth et al., 2013). Up to 10 ng of DNA extract was used for PCR amplification (first PCR), and Illumina Nextera adaptors were added at the second PCR step which is necessary for sequencing. After quality control and bioinformatics processing, reads between 8,284 and 47,917 were considered satisfactory (parameter: filtReads <8,000). Rarefied counts of each OTU were used in all subsequent data cleaning and statistical analyses.

For the controlled stagnation experiment, a total of 22 samples were successfully sequenced and processed for multivariate analysis to determine the impact of the following descriptors: “taps location’, “group type/cell density” and their interactive effect on the bacterial community composition data using a two-way permutational analysis of variance (PERMANOVA). The statistical test was done with the Bray-Curtis distance matrix. The pairwise-adonis test was then performed for multiple comparisons to determine which group type (cell density) had differing bacterial communities. A differential abundance test using the Differential gene expression analysis based on the negative binomial distribution (DESeq2; Love et al., 2014; Weiss et al., 2017) was done to identify which taxa were significantly (the least p-value) enriched post-flush and incubated samples vs. controls. This statistical test yields information on the artifact taxa and shift of bacterial community following cell density enrichment.

16S rRNA amplicon sequencing data can be accessed via the Sequence Read Archive (SRA) of the National Center for Biotechnology Information (NCBI): Accession number PRJNA1043459.

Measuring the percentage of intact filterable bacteria from RO tap water through various syringe filter pore sizes serves as a proxy test of biomass filtration efficiency of a particular filter membrane pore size. RO tap water samples had a high percentage (66.2%) of bacterial filterability through 0.2 μm pore size filters (Figure 2). This filterable fraction of intact bacteria was further reduced (18.1%) with lower pore size filters (0.1 μm; Figure 2). This fraction of bacteria is likely to reflect the population that survives chlorination in the network.

The influence of different syringe filter materials/manufacturers was tested (Supplementary Figure S1; Supplementary Table S1), showing that a certain type of filter material, i.e., cellulose acetate overestimated the actual bacterial count in the filtrate. These findings highlight the importance of the pre-filtration step before using a syringe filter in any subsequent analysis, e.g., microbial growth potential assay, to reduce the potential source of allochthonous carbon. Pre-washing a syringe filter with at least 200 mL of Milli-Q can remove contamination and reduce measurement error. Based on the results shown in Figure 1, it is reasonable to expect that DNA yield will be highly improved when using lower pore-size filters.

We investigated the technical factors influencing DNA extraction and yield enhancement from low-biomass RO tap water. First, to identify the optimal filter membrane type (material and pore size), a one-liter sample from each sample group (varied cell density) was filtered for DNA extraction. The sample groups included: pre-flush, post-flush, and incubated RO-produced tap water, Evian, and multiple controls. DNA concentrations were consistently low for post-flush and control samples (BDL < 0.01 ng/μL), irrespective of the filter membrane type used. Meanwhile, for higher cell density samples: RO-produced tap water and Evian, quantifiable DNA yields were obtained for both groups, 3.96–49.01 and 0.63–140.5 ng, respectively, across all filter membranes tested (Figure 3). The pore size effect, which is dependent on filter membrane material, showed no linear correlation with DNA yield, i.e., smaller pore size filters did not equate to higher DNA yields. Among different filter membrane types, significant differences in DNA yield were found (p < 0.05, Two-way ANOVA) and the polycarbonate (PC) filter membrane with 0.22 μm pore size substantially yielded more DNA compared to the other filter membranes tested (p < 0.05, post-hoc TukeyHSD test).

A subsequent analysis using qPCR was conducted to evaluate the total bacterial abundance (gene copy number/μL) in each sample group. All samples from the following: multiple controls, post-flush, and pre-flush, irrespective of filter membrane types had gene copy numbers below the detection limit (<1,000 copies/μL) and therefore not shown. Figure 4 shows the samples that had detectable gene copy numbers. MEC 0.2 μm and PES 0.1 μm filter membranes at control conditions had considerable background contamination in terms of 16S rRNA gene copy numbers (1,430 and 2,950 copies/μL, respectively, Figure 4). Meanwhile, PC 0.2 μm filters gave BDL at control conditions, corroborating the previous findings in DNA yield, and substantially resulted in higher gene copy numbers (Figure 4; p < 0.05, post hoc TukeyHSD test) for incubated samples and Evian water, both have higher cell density. From incubated samples, PC filter membranes enhanced the positive detection rate for total bacterial copy numbers with quantification values 158 times higher than controls. Overall, PC 0.2 μm pore size filters outperformed the other membranes in terms of the highest DNA yield and lowest background of 16S rRNA gene copy numbers.

Following the optimization of filter membrane type for enhancing the DNA yield, we designed the following controlled stagnation experiment to investigate the impact of increasing cell density on bacterial community composition, and the extent of its contribution to the set of parameters (TCC/ICC, ATP, 16S gene copies and DNA yield). DNA extraction was performed with PC 0.2 μm filter membranes only.

As shown in Figure 5A, stagnation had a similar effect on the TCC and ICC from the two tap locations: the maximum bacterial cell concentrations reached were 5-log [cells/mL; see Supplementary Table S1 for pre-flush and incubated samples while post-flush cell concentration is always consistent between 2- to 3-log (cells/mL)]. Consequently, these differing cell densities had substantially different DNA yields (Figure 5B), bacterial gene copy numbers, and ATP concentration (all p-values < 0.05, Two-Way ANOVA). Stagnation periods had no significant impact on the DNA yield obtained (p = 0.956, ANOVA). Pre-flush and incubated samples from tap 2 had overall lower DNA yields (6.85–9.35 ng) compared to tap 1 (29.5–60.5 ng) even though bacterial cell concentrations were within the same range with tap 1 [~4-log unit (cells/mL)] which is probably due to differences in DNA extraction efficiency and/or sample variations. However, even though differences in DNA yields were observed, the total bacterial gene copy numbers were similar for both taps [4.5-log–5.5-log unit (copies/μL)].

Our results emphasize that the lower bacterial cell concentrations in post-flush samples had significantly low DNA yield, and 16S rRNA gene copies (most are below the detection limit of 1,000 copies/μL) compared to the pre-flush and incubated samples that had at least 10⁴ cells/mL (all p values were <0.05, Post hoc TukeyHSD test). The consequence of these findings is that low-biomass RO tap water with cell density below 1,000 cells/mL will yield low DNA concentration, even after using a PC 0.2 μm filter that enhances DNA yield with minimum background bacterial copy number. Nevertheless, useful sequencing libraries were produced, and a straightforward incubation procedure evidently increased cell density, which is proportional to the higher DNA yield and higher total bacterial gene copy number.

To elucidate if increasing cell density could be a feasible alternative to improve DNA yield before performing downstream molecular analysis of low-biomass RO tap water, bacterial community profile from a controlled stagnation experiment was examined. One contaminant genus was detected to be highly abundant, i.e., Allorhizobium, Neorhizobium, Pararhizobium, and Rhizobium from Proteobacteria phylum (66%–94% of the relative abundances in controls; Figure 6A), most likely cross-contamination during the workflow or from the sequencing platform. Six more OTUs present in high abundance in the controls (Figure 6A; also confirmed by the least p-values based on DESeq analysis, data not shown) but not in samples were removed to create the second heatmap (Figure 6B) and ordination plot (Figure 7). The ordination plot clearly shows distinct bacterial communities of incubated samples compared to the pre-, and post-flush while the heatmap reveals the top 20 most abundant genera present in RO tap water from each sample group after removal of taxa that are highly enriched in the controls (Figure 6B).

Further analysis concerning the difference clustering of the bacterial communities (Figure 7) observed was done with multiple pairwise comparisons, revealing significant differences for all sample group pairs, except for pre- vs. post-flush samples. The tap location was found to have a significant effect on the bacterial community composition (PERMANOVA, p = 0.01) which can be attributable to the compositional variance in the bacterial community between both taps: samples from tap 1 have more consistent communities compared to tap 2 (Betadisper, p = 0.25, Supplementary Figure S4). Nevertheless, the distinction in the bacterial community between three sample groups (incubated, pre- and post-flush) can further be explained by 17 OTUs (out of 139 total OTUs identified from all samples) that made up the core members (cut-off relative abundance of at least 0.0001; and present in at least 50% of samples; Supplementary Figure S3). Notably, two genera were detected in all samples (100% of samples), namely Porphyrobacter, the most abundant genus, and Blastomonas (Figure 7B; Supplementary Figure S3). Both bacterial genera belong to the family Sphingomonadaceae which could serve as a strong potential indicator bacteria for RO tap water microbial water quality. Collectively, these findings highlight that increasing the number of samples (i.e., one-liter volume from multiple taps) allows identification of the resident bacterial community in RO tap bulk water, together with using the correct filter (PC 0.2 μm) and optimized DNA extraction protocol as well as incorporating multiple controls. Additionally, increasing cell density via incubation can be an easy and effective strategy to enhance DNA yield, allowing accurate identification of the core microbiome of low-biomass RO tap water. Incubation of RO tap water implies the thriving bacterial groups in RO water able to consume (low-amount) naturally assimilable organic carbon present.

We hypothesized that biomass filtration efficiency will correlate with recovered DNA extracts; the higher efficiency will result in higher DNA yields. However, our results showed no linear correlation between DNA yield and pore size filter (biomass retention efficiency; Figure 3). A smaller pore size filter did not equate to higher DNA yield which can be explained by premature clogging, facilitating DNA degradation (Abreu-Silva et al., 2023). A previous study has shown that a larger pore size filter favored long (environmental) DNA fragment retention compared to short DNA (i.e., less degraded DNA; Jo et al., 2020) which likely contributes to high yields of DNA recovery.

This study identifies that a PC filter membrane with 0.2 μm pore size enables high recovery of DNA extract from low-biomass RO tap water. This filter remarkably outperforms the other four filter membranes tested, in agreement with similar previous studies that reported high DNA yield and 16S gene abundance of low microbial load water samples extracted with a PC filter (Minamoto et al., 2016; Abreu-Silva et al., 2023). PC filter membrane has a straight-through pore-channel structure, less surface roughness, least hydrophilic (higher surface hydrophobicity), and highest wettability (Dizge et al., 2011; Abreu-Silva et al., 2023) that may explain higher DNA yields and low background contaminant (16S rRNA copy number) obtained. PC filter with 0.2 μm is known to retain less protein and carbohydrate (Dizge et al., 2011), which can interfere with DNA extraction and severely affect DNA purity (Abdel-Latif and Osman, 2017). Aggregates of the solubilized microbial products, including its DNA, have been reported to be greater for PC membranes than other membranes with interconnected pore geometry (e.g., PES and MEC; Dizge et al., 2011).

However, it should be noted that intra- and intervariations between filter material and the nested pore size exist. Filter pore sizes are known to be different in their nominal value described by the manufacturer, the size of any pore can widely vary (Deiner et al., 2018). That being said, the pore size of a filter membrane is rather a probability distribution and not an exact (homogenous). Investigation on the micrograph of the pore structures of filter membranes used was out of the scope of the current study and the pore size value itself was not determined analytically, but for PC filter specifically, the scanning electron microscopy images are available elsewhere (Tung and Chuang, 2002; Paoli et al., 2020). Our results show that the PC filter gives the best DNA yield and low background 16S rRNA gene copies but the manufacturer has stated that the water volume filter should be a maximum of two liters. A composite sampling, i.e., splitting biomass filtration into multiple filters could be considered if larger sampling volumes are needed and how it affects the DNA concentration and microbial community remains to be investigated in future studies. Future works should also explore how different pore sizes that are dependent on filter materials interact with (extracellular) DNA could explain different DNA yields. Nevertheless, our results emphasize that using a PC filter with 0.2 μm pore size and multiple smaller volumes is reasonable and provides a good representation (i.e., rarefaction curves show that most bacterial taxa were captured, Supplementary Figure S5) of the bacterial community composition in low-biomass RO tap water.

First, in terms of absolute abundance (i.e., number of different OTUs), the low number of OTUs (139 OTUs) detected reflects the low bacterial cell numbers in the bulk water, as found elsewhere (Kantor et al., 2019). Increasing sampling volume does not fundamentally address the problem (Stamps et al., 2018; Putri et al., 2021) due to the sparse distribution of bacterial cells in RO bulk water (i.e., low cell density, low ATP, and 16S gene copy number, this study). The tap water incubation procedure increased bacterial cell density which correlated with an increase in DNA yield and gene copy number (Figure 5B). However, the gene copy number had less variability compared to the DNA yield (Figures 5A,B), and thus more indicative of the high cell density. Adding 16S rRNA gene copies parameter allows for reducing false negative of low DNA yield samples before proceeding with amplicon sequencing. Second, in terms of beta diversity analysis we observed differences in bacterial communities in terms of overall richness (i.e., number of OTUs; Figure 6B) after deducting from the contaminant detectable in the controls (Figure 6B). The contaminant genus is known to be isolated from wastewater samples (Bigott et al., 2022) and appeared exceptionally high (~66%–94%) in the controls, likely to be cross-contamination from the sequencer (and/or sequencing provider). The present study also rules out more contaminant genera: Ralstonia, Sphingomonas, Corynebacterium_1, Methylobacterium, Idiomarina, and Rhodococcus that were present in high abundant and high coverages in the control and samples irrespective of cell density regimes. All these genera have been previously reported as common contaminant taxa from low-biomass environments (Salter et al., 2014).

Post-flush network samples had the highest diversity, i.e., more appearance of low abundance OTUs compared to incubated and pre-flush samples (Figure 6B). This difference could reflect the actual environmental heterogeneity between those samples, such as temperature, sample confinement (e.g., materials) during incubation, and hydrodynamics conditions (Neu et al., 2018; Proctor et al., 2018; Dai et al., 2020). Nonetheless, the core bacterial community (Supplementary Figure S3) between post-flush (low DNA yield) and incubated tap water samples (high DNA yield) from the same environment converged, showing that the incubation strategy works in reflecting the true diversity of low-biomass chlorinated RO drinking water samples. It is well acknowledged that amplicon sequencing cannot exhaustively reveal the total microbial content in a given sample and can only serve as a tool to characterize the dominant taxa in a sample and for comparative assessment between different samples or grouping factors. However, it remains a faster and more cost-effective method for identifying microbial taxa compared to shotgun metagenomics. Performing quantitative PCR targeting a specific bacterium (other than the total 16S rRNA gene copies) that may be predominant in the sample microbiome composition can also be done to ensure that a low DNA yield sample has higher biological signals than the control (Zhang et al., 2015).

Based on the core microbiome (i.e., shared bacterial taxa) analysis using cutoff: taxon presence in at least 50% of samples (occupancy-based) and a total abundance higher than 0.001% across all samples (Custer Gordon et al., 2023), we identified two genera that were found consistently throughout all cell density regimes, either pre-flush (in-situ stagnation), post-flush or incubated samples: Porphyrobacter and Blastomonas, highlighting their importance in the (chlorinated) RO-produced drinking water biological stability. Porphyrobacter has been reported to proliferate after surviving chlorine dioxide disinfection (Ma et al., 2020) while Blastomonas is a dominant bacterial genus found in tap water resistant to chlorine dioxide (Paduano et al., 2020; Sala-Comorera et al., 2020) and biofilm in a chlorinated distribution system (Buffet-Bataillon et al., 2010). These two genera are very likely to survive the residual chlorine present in the RO bulk water, explaining their high abundance. The overall bacterial community profile from this study could serve as baseline data for chlorinated RO-produced drinking water. The genus level information provided also highlights potential ecological attributes of those two genera that are often overlooked when using a higher level of taxonomical information (e.g., phylum, class; Hermans et al., 2017). We did the comparison of bacterial community from the bulk water previously at the upstream part of KAUST RO tap water before and after chlorination point (Farhat et al., 2022) and found a high abundant of Blastomonas genus in concurrence with the current study, showing that this bacterium is the actual resident within chlorinated RO tap network that is resistant to chlorine. Further studies could be done to investigate temporal (and spatial) variations (e.g., physicochemical conditions, sampling from different buildings) from chlorinated RO tap water and how these two genera correlates with those factors.

In addition, 17 OTUs (Supplementary Figure S3) were identified successfully as core microbiomes of RO bulk water that are highly adapted to oligotrophic (low level of AOC) conditions and chlorine disinfection specific to our drinking water network. It is worth noting, that some taxa were enriched in pre-flush but not post-flush or incubated samples (Figure 6B) such as Dechloromonas, Novosphingobium, and Piscinibacter. Dechloromonas has been reported to dominate the biofilm in disinfected drinking water, playing a role in inhibiting corrosion (Wang et al., 2017a). Novosphingobium has been known to increase its antibiotic resistance gene expression via biotransformation with antibiotic (Ciprofloxacin)-chlorination products (Wang et al., 2017b). Meanwhile, Piscinibacter has been reported to predominate cultured biofilm growing in chlorinated polyvinyl chloride that can utilize a single carbon compound (van der Kooij et al., 2017). We hypothesized that these bacteria can grow in pre-flush samples following the rise in assimilable organic carbon (AOC) due to the stagnation and depletion of residual chlorine. The range of AOC in pre-flush samples was between 0.025–0.05 μg (1 μg AOC = 10⁶ cells/mL; Hammes and Egli, 2005) AOC after stagnation.

In summary, this study shows that a strategy employing a good filter (PC 0.2 μm) membrane for (microbial) DNA extraction with low background contaminant, optimizing DNA extraction protocol, incorporating multiple controls, having multiple smaller (1 L) sampling volumes, and increasing cell density via incubation allows to profile microorganisms in low-biomass RO tap water conclusively. Future studies should investigate how different filter membrane materials and pore sizes affect different microbial endpoints (eukaryotes, fungi, viruses) in this particular type of water. Different groups of microorganisms present in low-biomass tap water will probably have different cut-offs of filter membranes (type/pore size). We also observed a shift in abundance of the core microbiome between pre-, post-flush, and incubated samples that warrant further investigation of how these communities change across seasonal and spatial variations. Amplicon sequencing analysis also does not delineate active and dead cells so further examination could be performed, such as using propidium monoazide-quantitative polymerase chain reaction (PMA-qPCR) targeting Porphyrobacter and/or Blastomonas.

There are practical implications to the findings of this study. First, circumventing the low DNA yield from low-biomass RO tap water before performing downstream DNA-based analysis can be done with multiple smaller (i.e., one-liter) sampling volumes and incubation of the bulk water to reach higher bacterial cell density (with naturally occurring AOC). These two approaches allow for conclusively profiling the bacteria present and influence the biological stability in the premise plumbing and the distribution network conditions. Our current results expand prior work of the optimized DNA extraction protocol (i.e., DNEasy Power Water kit) that was adopted in this study for low-range bacterial cell concentration (10^3–10⁵ cells/mL; Vosloo et al., 2019). Importantly, to the authors’ knowledge, our study is the first of its kind to identify the best filter membrane (i.e., PC 0.2 μm) for DNA extraction of (chlorinated) low-biomass RO-produced drinking water with the highest DNA yield and lowest background copy number of 16S rRNA gene. Second, incorporating low-biomass blanks is crucial as well as subjecting it to the same amplicon sequencing/bioinformatics pipeline for the best results’ interpretation of microbial community analysis from this type of water. Here, we added a Milli-Q and sterilized (autoclaved) version of incubated RO tap water to the blanks/controls filtered through the same filter membrane in addition to the well-established blanks (no template DNA blank, and unused virgin/sterile filter membrane blank) that are commonly used, and thus we also recommend future studies to incorporate this type of blanks.