Studied waters exhibit low microbial activity

A summary of parameters in samples collected in the Metro Vancouver and Halifax Water systems is presented in Table 2:

Table 2 Characteristics of water samples collected from Metro Vancouver and Halifax Water transmission/distribution systems. Median results are shown, and the interquartile range (IQR) is represented in parenthesis (no value indicates an IQR of 0). Turbidity results from Halifax Water are not available

Treated water samples are comparable between the two sites and characterized by low microbial activity. Median HPC are below detection limits for both systems, with 78% of Metro Vancouver treated water samples (n = 25 out of 32) and 67% of Halifax Water samples (n = 190 out of 283) being non-detectable for HPC. Maximum value of HPC is 14 CFU/mL for Metro Vancouver, while 2% (n = 7) of Halifax Water samples exceed the relatively low upper limit of 250 CFU/mL. In comparison, studies showing high HPC-ATP correlation such as Duda et al. [10] reported HPC up to 105 CFU/mL, which is far greater than even Metro Vancouver raw inlet results (maximum HPC of 400 CFU/mL).

Median cATP in treated water samples for both study sites is 0.20 pg/mL, with a maximum of 2.5 pg/mL for Metro Vancouver and 7.9 pg/mL for Halifax Water. All treated water samples are therefore below the recommended limit of 10 pg/mL for corrective action [23]. cATP demonstrates a higher sensitivity than HPC, as expected: only 12% of Halifax Water samples (n = 34) are non-detectable for cATP, and none of the Metro Vancouver samples are non-detectable.

Poor correlation of cATP and HPC in waters with low microbial activity

HPC and cATP results for the Halifax Water and Metro Vancouver studies are presented in Figs. 1 and 2, respectively.

Fig. 1

A HPC versus cATP for Halifax Water system samples collected from January to November 2021. B zooms into data near the origin for better visibility. Solid circles = samples with HPC values between lower and upper limit (n = 86); hollow circles = samples with HPC values below/above the lower/upper limits (n = 197); solid line = regression line of all samples (R = 0.31, p << 0.001, n = 283); dashed line = regression line of samples between HPC lower and upper limits (R = 0.18, p = 0.09, n = 86)

Fig. 2

A HPC versus cATP for samples collected from Metro Vancouver inlet water and along the transmission system from June to July 2022. B zooms into data near the origin for better visibility. Solid circle = samples with HPC above detection limit (n = 15). Hollow circle = samples with HPC below the detection limit (n = 25). Solid line = least-squares regression line of all samples (R = 0.61, p << 0.001, n = 40). Dashed line = least-squares regression of samples above detection limits only (R = 0.46, p = 0.08, n = 15)

While Pearson R coefficients are shown to facilitate comparison with other studies, the data do not, in fact, satisfy the assumptions for simple linear regression. Namely, both Halifax Water and Metro Vancouver datasets yielded significant non-normal distributions of residuals, skewed left (p << 0.001; residual plots, histograms, and Q-Q plots are shown in Figure S1 and Figure S2). One main contributor to the asymmetrical distribution appears to be the disproportionate number of low HPC samples (< 2 CFU/mL), combined with the fact that these low-range HPC samples correspond to a wide range of cATP results (0 to 3 pg/mL). This results in a significant number of negative residuals in the low HPC/cATP range (i.e., left-skewed), which is not improved even after excluding samples below and above the HPC detection (p << 0.001).

Spearman Rs coefficient is therefore considered a more appropriate descriptor for these datasets, summarized in Table 3 for all studied parameters (equivalent correlation matrix for Pearson R coefficients provided in Table S2, for reference):

Table 3 Spearman Rs calculated for (A) Halifax Water (n = 283) and (B) Metro Vancouver (n = 40). Insignificant correlations (p > 0.05) are shown in grey. Correlations involving free chlorine for Metro Vancouver are limited to n = 32 treated water samples as free chlorine was not analyzed for n = 8 raw inlet samples

No significant relationship between HPC and cATP is observed for Halifax Water (p > 0.05). In contrast, a significant relationship of moderate strength is observed for Metro Vancouver (Rs = 0.55, p < 0.001). This aligns with studies testing HPC waters of a similar range: Nescerecka et al. [27] found Rs = 0.33 when testing chlorinated water distribution samples with HPC up to 220 CFU/mL. The likely reason for the discrepancy between Halifax Water and Metro Vancouver is the inclusion of raw source water in the Metro Vancouver dataset: the relatively higher HPC and cATP values of the source water samples allow for a relationship to be determined between these two parameters. Excluding these raw water samples (n = 8) eliminates any significant relationship between HPC and cATP (p = 0.9, n = 32; detailed results provided in Table S3).

Therefore, the present study supports the observation that low HPC waters, in the ranges typically found in chlorinated potable distribution systems, are unlikely to produce strong correlations between HPC and ATP. This also can be due to the fact that ATP is not sufficiently accurate in aquatic environments with low microbial concentrations [16]. In comparison, studies finding a high correlation, such as R = 0.90 by Duda et al. [10] involved chlorinated waters collected at point-of-use with a mean HPC of 537 CFU/mL (reported as log10), which are orders of magnitude higher than the present study median HPC that is below detection limits (< 1–2 CFU/mL).

A limitation of ATP analysis, which can contribute to discrepancies with HPC, is that not all cells may be successfully lysed during sample preparation. Interestingly, while Gram-positive bacteria have a thicker peptidoglycan layer that provides structural support and resistance to chemical agents, Turner et al. [36] determined that it is in fact Gram-negative bacteria which experience incomplete cell lysis, thus resulting in underestimation of ATP levels. Additionally, the presence of an outer membrane in Gram-negative bacteria contributes to their lower surface charge, ranging from -80 to -140 mC m−2, compared to the higher surface charge of -350 to -450 mC m−2 in Gram-positive bacteria [7]. This difference in bioelectrical environments affects the efficiency of ATP extraction and measurement. The outer membrane's complex structure in Gram-negative bacteria hinders processes like electron transfer, which are crucial for accurate ATP measurement, leading to potential inaccuracies [28]. Conversely, ATP measurement in Gram-positive bacteria is typically more straightforward, with a more linear relationship between microbial suspension density and ATP concentration [33]. Therefore, the differences in cell wall structure, bioelectrical properties, and the methodologies used significantly influence the accuracy of ATP measurements across these bacterial groups. Gram-negative bacteria can comprise a significant portion of the microbial population in surface waters, making up as much as 90% of isolates studied in Lake Ontario and Lake Superior [3]. Moreover, the percentage of Gram-negative bacteria can also shift significantly from source to point-of-use: Pepper et al. [30] found Gram-negative bacteria comprised 76% of the studied source river water but only 12% in the chlorinated distribution system and 0.2% at the tap. This may explain the poor correlation between HPC and ATP in water samples, even in waters with higher biological activity.

Additionally, some of the concerning pathogenic heterotrophic bacteria that are Gram-negative are total coliforms, E. coli, Shigella, Pseudomonas aeruginosa, Legionella pneumophila, Aeromonas hydrophila, Klebsiella pneumoniae, Salmonella enterica [6]. Therefore, considering the prevalence, variability, and pathogenic concern of Gram-negative bacteria, it is advisable to complement ATP testing with species-specific tests, particularly for systems that utilize source waters of pathogenic concern. Alternatively, ATP detection limits for the mentioned pathogenic species can also be established in pure culture form. For instance, Turner et al. [36] reported that the detection limit of E. coli and Staphylococcus aureus was 104 and 102 organisms, respectively.

Interpretation of HPC and cATP results can yield consistent conclusions

Despite weak correlation between cATP and HPC, it is worth noting that using guideline ranges to interpret cATP and HPC data can lead to highly consistent conclusions. To illustrate, HPC results ≥ 100 CFU/mL can be considered indicative of a need for investigation, based on the drinking water standard limit of 100 CFU/mL employed in jurisdictions like Germany, The Netherlands, and Japan [32] (in comparison, the US EPA guideline for HPC of 500 CFU/mL [18] is too high to be relevant for the studied waters, none of which exceeded this value). Similarly, cATP results ≥ 10 pg/mL has been suggested by the test kit manufacturer as being indicative of a need for corrective action [23]. Application of this pass/fail criteria to the data set can be visualized as follows:

For the Halifax Water study, 96% of the samples (n = 273) are in agreement between cATP and HPC, in that they pass both cATP criterion (< 10 pg/mL) and HPC criterion (< 100 CFU/mL; Fig. 3A, bottom shaded region). The remaining 4% (n = 10) pass cATP (< 10 pg/mL) but fail HPC (≥ 100 CFU/mL; Fig. 3A, top unshaded region); these can be described as false negatives if cATP is treated as the predicting parameter, i.e., cATP “wrongly” identifies a sample as exhibiting low microbial activity when HPC is in fact elevated. None of the samples from Halifax Water exceed the cATP fail criterion of 10 pg/mL.

Fig. 3

HPC and cATP data for Halifax Water (A) and Metro Vancouver (B), previously presented in Figs. 1 and 2 respectively, shown again here with pass/fail criteria of HPC = 100 CFU/mL and cATP = 10 pg/mL. Solid circles = HPC within detection limit, hollow circles = HPC outside of detection limit

For the Metro Vancouver study, 95% of the samples (n = 38) agree between HPC and cATP; samples either pass both criteria (n = 34, Fig. 3B, bottom left shaded region), or fail both criteria (n = 4, Fig. 3B, top right shaded). One false negative of low cATP and high HPC is observed (Fig. 3B, top left unshaded), along with one false positive of high cATP and low HPC (Fig. 3B, bottom right unshaded). These results are summarized in Table 4:

Table 4 Analysis of error probability, based on criteria of cATP ≥ 10 pg/mL and HPC ≥ 100 CFU/mL as indicative of the need for corrective action. False positive: cATP indicates corrective action (≥ 10 pg/mL), but HPC does not (< 100 CFU/mL). False negative: cATP does not indicate corrective action (< 10 pg/mL) but HPC does (≥ 100 CFU/mL). Correct: cATP and HPC results are consistent (both above or both below 10 pg/mL and 100 CFU/mL, respectively)

To be clear, this error analysis is only provided for illustrative purposes to demonstrate how cATP data may be used by utility managers to replace HPC. In practice, system-specific baselines and trends should be established for cATP, HPC and any other bacteriological monitoring technique (e.g., cell count using flowcytometry, gene copies count using q-PCR, bacterial relative and absolute abundance using microbiome analysis, etc.) over a sufficient test period, which is consistent with the recommended practice for how these bacteriological indicators are best used for utility monitoring [1, 17, 20].

Correlating free chlorine and turbidity to HPC and cATP

Correlations between free chlorine, turbidity, HPC, and cATP for Metro Vancouver are summarized in Table 3. The Pearson coefficients for the same data are provided in Table S2. A weak but statistically significant negative correlation between free chlorine and cATP is observed for Halifax Water (Rs = -0.13, p = 0.04); the negative relationship is consistent with the disinfecting action of free chlorine. On the other hand, no significant relationship is observed between free chlorine and HPC for Halifax Water (p > 0.05). This is in agreement with Kennedy et al. [19] who reported a better correlation of cATP with free chlorine than HPC.

For Metro Vancouver, no correlation could be observed between free chlorine and cATP, nor between free chlorine and HPC. This disagreement with Halifax Water data could be attributed to the smaller sample size of the Metro Vancouver study, and it is possible that more samples collected over a longer period of time to capture a wider range of system conditions would have resulted in a significant correlation between free chlorine and cATP. It is also worth noting that raw inlet water samples (n = 8) were not analyzed for free chlorine and have been excluded from this analysis. If inlet samples are included with an assumed free chlorine of 0 mg/L, a significant correlation between cATP and free chlorine does emerge (Rs = -0.39, p = 0.01), though the same is also true for HPC and free chlorine (Rs = -0.44, p = 0.005). Notwithstanding, the present data only permits the tentative conclusion that cATP may correlate better than HPC to free chlorine, on the basis that one study determined significant correlation between cATP and free chlorine, whereas neither study found a correlation between HPC and free chlorine.

Turbidity data is only available from Metro Vancouver. Similar correlations between cATP and turbidity versus HPC and turbidity are observed (Rs = 0.33 and 0.46, respectively), i.e., neither appear to be obviously advantageous over the other. Furthermore, these correlations can be attributed to the inclusion of raw source water samples, without which no significant relationships (p > 0.05) are observed between turbidity and either HPC or cATP (Table S3). Therefore, the study results do not necessarily indicate that turbidity can be practically correlated to either parameter when sampling from chlorinated transmission/distribution systems.

Sodium thiosulfate does not significantly affect cATP

As regulated microbial tests like for E. coli require samples to be chlorine-quenched, the impact of chlorine quenching on cATP assay was studied to assess whether thiosulfate-pre-charged sampling bottles can be used (i.e., to unify sample handling). As discussed in Methods Section "Impact of quenching agent", n = 7 pairs of samples with and without thiosulfate addition were collected and extracted at three time points, resulting in up to n = 6 pairs of comparative cATP results at each time point (not all samples could be extracted at all three time points) for a total of n = 15 data pairs (Table 5).

Table 5 cATP results for samples with and without thiosulfate addition, reported as median with interquartile range represented in parentheses. Percent difference (calculated as “no thiosulfate” minus “with thiosulfate”) reported as mean ± one standard deviation

Median cATP results for samples with and without thiosulfate are comparable across the extraction times, with a mean percent difference of up to 36% (non-thiosulfate cATP higher than thiosulfate). Paired t-test and Wilcoxon signed rank test (the latter for non-normally distributed sample sets) conducted for thiosulfate versus non-thiosulfate results, at each extraction time all indicate a lack of significant difference (p > 0.4, detailed results in Table S4), suggesting that thiosulfate addition does not significantly impact cATP results at the conditions tested.

To further support the analysis of these results, the baseline variance of cATP is determined from the mean coefficient of variance for n = 26 sets of triplicates (detailed results provided in Table S5). The one-way ANOVA test confirms that the coefficients of variance evaluated under the various sampling conditions (Methods section "Establishing baseline of cATP variance") are not significantly different (p = 0.4). As such, all n = 26 triplicates are used to calculate a mean coefficient of variance of 35 ± 17%, which is consistent with literature values of 32 ± 16% [29]. Percent differences evaluated between thiosulfate versus non-thiosulfate (Table 5) are not significantly different from baseline variance (p > 0.7, Welch’s t-test; detailed results in Table S6), further supporting the conclusion that thiosulfate addition does not significantly impact cATP results.

Extraction within 24 h is acceptable for cATP analysis

To assess the impact of sample extraction time on cATP results, samples were extracted at 4, 6 and 24 h after collection. Percent changes in cATP between different extraction times are presented in Fig. 4 (values reported in Table S7):

Fig. 4

Percent change in cATP values between the extraction times studied (i.e., 4, 6, 24 h) for samples with and without sodium thiosulfate addition. Values shown are mean ± one standard deviation. Note that 4 to 24 h percent change is not necessarily the cumulative result of 4 to 6 h and 6 to 24 h because only select samples could be extracted at 4 h, with the result that different sample pairs are used to calculate each percent change (detailed breakdown provided in Table S7)

Samples without thiosulfate demonstrate a consistent decrease in cATP by 30 to 40% over 24 h, with an average standard deviation of 43%. Reduction in cATP would be consistent with the absence of chlorine quench by thiosulfate; chlorine in the samples is expected to accelerate cell death and reduce cATP over time.

Samples with thiosulfate show an inconsistent trend. A significant decrease (-75%) is observed over a relatively short time period of 4 to 6 h, followed by a small increase (+ 19%) from 6 to 24 h. Overall, from 4 to 24 h, cATP decreases by 30%, which is comparable to samples without thiosulfate (-34%). Notably, a larger standard deviation (average 79% over the three studied time-pairs) is observed than that for samples without thiosulfate (average 43%). The significant decrease from 4 to 6 h (-75%, n = 5 pairs) is difficult to explain, especially as the expected effect of thiosulfate addition would be improved preservation of cATP. It is worth noting that, of the five data pairs averaged in this calculation, two of the greatest changes (most negative) are observed for samples of relatively low cATP (0.13 pg/mL). That is, while a very large percent change is observed for these two sample pairs (-150%), the actual change in cATP for these samples is only 0.12 pg/mL, which is fairly small compared to the scale of cATP alert criterion (10 pg/mL). In other words, it is worth noting that the large 75% decrease from 4 to 6 h is the result of relatively small changes to already-low cATP values.

Despite these various apparent trends, none of the percent changes, whether for samples with or without thiosulfate, are statistically different than baseline variance (35 ± 17%). This is confirmed by Welch’s t-test yielding p > 0.3 at all conditions (statistical test conducted using absolute value of percent changes; detailed results in Table S8 and Table S9). In other words, fluctuations in cATP readings over the 24-h period can still be attributed to baseline variance, thus supporting the manufacturer’s claim that analysis within 24 h of collection is acceptable. As this is observed for samples with and without thiosulfate, the results further support the conclusion in Section "Sodium thiosulfate does not significantly affect cATP" that addition of sodium thiosulfate is acceptable for cATP analysis.

Cost and other considerations

While a full cost analysis is beyond the scope of the study, the authors believe that some general remarks on cost comparison between cATP and HPC testing can be valuable to utility managers. As of the date of this writing, the cost of consumables for cATP testing is approximately $20 CAD per sample (LuminUltra® QGA-25), while the cost of a luminometer is approximately $8,000–9,000 CAD (LuminUltra® PhotonMaster). In comparison, HPC costs for utility managers can vary widely depending on the scope of their operations. For utility managers operating certified laboratories for in-house HPC testing (e.g., Metro Vancouver), one of the main consumables is agar growth media. Cost of R2A agar can range from $0.35 CAD per test if prepared from dehydrated powder, up to $17 CAD per test if using prepared plates (Thermo Scientific R2A Agar). This excludes cost of equipment such as incubators, and additional equipment required for media preparation (e.g., drying oven). For utility managers that contract third-party laboratories for HPC testing (e.g., Halifax Water), the cost will naturally vary based on external laboratory pricing. That said, cATP testing is likely to be a cost-effective alternative especially considering its simplicity of operation, with the only equipment required being the aforementioned portable luminometer.

Finally, it is worth noting that a sizeable amount of plastic waste is produced for cATP testing, including the one-time use filters, syringes, and cuvettes. Plastic waste produced in HPC testing varies based on laboratory-specific procedures, with zero waste possible through the use of autoclavable glass plates and pipettes (as is the case for Metro Vancouver). For utility managers using prepared, disposable plates, however, the amount of waste generated becomes comparable with cATP testing.