Legionella Diversity and Spatiotemporal Variation in the Occurrence of Opportunistic Pathogens within a Large Building Water System

Understanding Legionella survival mechanisms within building water systems (BWSs) is challenging due to varying engineering, operational, and water quality characteristics unique to each system. This study aimed to evaluate Legionella, mycobacteria, and free-living amoebae occurrence within a BWS over 18–28 months at six locations differing in plumbing material and potable water age, quality, and usage. A total of 114 bulk water and 57 biofilm samples were analyzed. Legionella culturability fluctuated seasonally with most culture-positive samples being collected during the winter compared to the spring, summer, and fall months. Positive and negative correlations between Legionella and L. pneumophila occurrence and other physiochemical and microbial water quality parameters varied between location and sample types. Whole genome sequencing of 19 presumptive Legionella isolates, from four locations across three time points, identified nine isolates as L. pneumophila serogroup (sg) 1 sequence-type (ST) 1; three as L. pneumophila sg5 ST1950 and ST2037; six as L. feeleii; and one as Ochrobactrum. Results showed the presence of a diverse Legionella population with consistent and sporadic occurrence at four and two locations, respectively. Viewed collectively with similar studies, this information will enable a better understanding of the engineering, operational, and water quality parameters supporting Legionella growth within BWSs.


Introduction
Various Legionella species, such as L. anisa, L. feeleii, L. longbeachae, L. pneumophila, and L. micdadei, can cause legionellosis, which are bacterial infections resulting in either a mild flu-like illness (Pontiac Fever) or a potentially fatal form of pneumonia (Legionnaires' Disease (LD)) [1]. These infections are primarily caused by the inhalation of Legionella-contaminated aerosols generated from engineered water systems [2,3]. Of the 74 drinking water-associated outbreaks reported in the US between 2011 and 2014, Legionella was responsible for 61% of those outbreaks causing 17% (241/1437) of the illness cases, 88% (200/226) of hospitalizations, and 100% of the outbreak deaths (27/27) [4,5]. Legionella presence in the building water systems (BWSs) was cited as the main deficiency leading to those outbreaks underscoring the need to control and prevent Legionella growth within these BWSs. sampled locations (Table 1) with previous detection of Legionella in the cold water (data not shown). This building was also chosen because it contains both office and production facilities where large volumes of water are used in the latter portions (e.g., for production processes and cooling) and smaller volumes in the office spaces for employees. Three locations had polyvinyl chloride (PVC) schedule 80 valves that supplied monochloramine-treated water (PVC-MA) or chlorine-treated water with and without passage through the building's chiller/refrigeration system (PVC-R and PVC-FC, respectively); and the other three sampling locations supplied chlorine-treated water from a cast brass spigot (Spigot), a chrome-plated, forged brass faucet (Faucet), and a drinking water fountain with a stainless steel and ethylene propylene diene bubbler head (Fountain) ( Figure S1, Table 1).
The water usage ranged from 5 L to 1.4 million L per season with an average of 278,030 L ± a standard deviation (SD) of 598,785 L per season for each sampling location. The PVC-R location is a high water usage site due to the presence of multiple online and remote water quality sensors that require constant water flow; thus, when this site is excluded, the water usage of the other five locations had an average of 35,194 ± 76,876 L per season. The pipe material from the outlet to the supply feed were either all copper (Spigot, Faucet, and Fountain), all PVC (PVC-MA), a mixture of both (PVC-FC), or a mixture of fiberglass ductile iron and PVC reinforced plastic tubing (PVC-R).
In this study, a total of 114 bulk water and 57 biofilm samples were collected across the six locations. Bulk water samples were analyzed for pH, turbidity, temperature, and free and total chlorine ( Table 2). There were no statistical differences between the pH of the first and second draw samples for each location and sampling time points, but there were differences in pH between the PVC-MA samples and the other five locations (P < 0.001). There were no statistical differences between the turbidity of all water samples at each location most likely due to the large range in Nephelometric Turbidity Units (NTUs) observed for each sample (Table 2, Figure S2).
Temperatures between first and second draw samples within each location and time point were also not significant, except for Spigot-first draw versus Spigot-second draw samples (P < 0.05). Free and total chlorine levels for first and second draw samples between, and within, each location were not significant, except for Faucet-first draw versus Faucet-second draw and PVC-R-first vs Faucet-first draw samples (P < 0.01, Table 2, Figure S3). There were no statistical differences between monochloramine and total chlorine levels in the bulk water from location PVC-MA (P > 0.05). There were statistical differences between the heterotrophic plate count (HPC) levels of the first vs second draw samples at only the PVC-R, Spigot, and Faucet locations (P < 0.01, Table 2). For biofilm samples, the average HPC level observed at the Faucet location was the highest compared to other locations but was statistically different only from the PVC-R biofilm samples (P < 0.01, Table 2).

Legionella Culture Results
Bulk water and biofilm samples were processed and enumerated for Legionella colony forming units (CFU) as described in Sections 4.2 and 4.4. No culturable Legionella was detected in samples from the PVC-MA, Spigot, or Fountain location; from the second draw and biofilm samples from PVC-R; and biofilm samples from PVC-FC. However, culturable Legionella was detected at various time points for the first and second draw samples from PVC-FC; all sample types from the Faucet; and only the W2018 first draw sample from PVC-R, which contained Legionella non-pneumophila bacteria as confirmed by colony lysate PCR and latex agglutination (Table 3).  For the PVC-FC bulk water samples, Legionella non-pneumophila colonies were detected in both the first and second draw samples during W2018 (4.5 and 2.6 log 10 CFU 100 mL −1 , respectively) and W2019 (2.1 and 2.6 log 10 CFU 100 mL −1 , respectively) and only the second draw (3.0 log 10 CFU 100 mL −1 ) during F2018 via colony confirmation PCR (Table 3). The F2018 isolate was identified as L. feeleii via indirect immunofluorescent antibody assay as described in Section 4.4. L. pneumophila and non-pneumophila colonies were detected in the first and second draw samples (3.8 and 2.8 log 10 CFU 100 mL −1 , respectively) at this location during Su2018 via colony confirmation PCR. The L. pneumophila PCR-positive Su2018 colonies either gave an inconclusive latex agglutination result or were identified as belonging to serogroup (sg) 2-14. Three Su2018 isolates (one from the first draw and two from the second draw samples), one F2018 isolate (from the second draw sample), and four W2019 isolates (two from the first and two from the second draw samples) were processed for whole genome sequencing. One of the two Su2018 second draw isolates was identified as L. pneumophila sg5 via indirect immunofluorescent antibody assay.
Throughout the study period, only one second draw (W2019) and one biofilm (F2016) sample, at the Faucet location had culturable Legionella bacteria at concentrations of 2.1 log 10 CFU 100 mL −1 and 1.9 log 10 CFU cm −2 , respectively ( Table 3). The Faucet second draw sample was confirmed as L. pneumophila sg1, and the biofilm sample contained a mixture of L. pneumophila sg1 and 2-14 colonies as confirmed by PCR and latex agglutination. Fifty percent of the Faucet first draw samples were Legionella culture positive: Sp2017, F2017, W2018, Su2018, and W2019 with an average ± SD concentration of 2.0 ± 0.8 log 10 CFU 100 mL −1 (Table 3). Legionella identified within these Faucet first draw samples were diverse with only L. pneumophila sg1 being identified in the Su2018 and W2019 samples; only L. pneumophila sg2-14 in the Sp2017 sample; a mixture of L. pneumophila sg1 and 2-14 in the W2018 sample; and Legionella non-pneumophila identified in the F2017 sample. Five isolates obtained from the Su2018 and one from the W2019 Faucet first draw samples were processed for whole genome sequencing.
For only the Su2018, F2018, and W2019 time points, L. pneumophila most probable number (MPN) in bulk water and biofilm samples were enumerated using Legiolert ® as described in Section 4.4. All samples at each location had no detectable L. pneumophila MPN except for the Faucet first draw (Su2018 1.8 log 10 MPN 100 mL −1 ; W2019 3.8 log 10 MPN 100 mL −1 ) and second draw (Su2018 1.4 log 10 MPN 100 mL −1 ; W2019 2.5 log 10 MPN 100 mL −1 ) samples. Four wells from the first draw and three wells from the second draw Legiolert ® tray, read as L. pneumophila positive, were sampled and pure colonies were obtained as described in Section 4.4. Isolates were Legionella and L. pneumophila PCR-positive but gave an inconclusive latex agglutination result. One of the W2019 Faucet second draw Legiolert ® isolates was processed for whole genome sequencing and identified as sg5 via indirect immunofluorescent antibody assay. In addition to culture methods, bulk water and biofilm samples were also analyzed for the presence of Legionella spp. and L. pneumophila via qPCR as described in Section 4.6. In agreement with the specificity and sensitivity of the Legionella genus and L. pneumophila species-specific qPCR assays used in this study, the total Legionella levels detected in the bulk water and biofilm samples were higher than the L. pneumophila levels observed in the corresponding samples (Figure 1a-f and g-l). The presence of Legionella and L. pneumophila was detected at all sampling locations and sample types and within each location, detection generally occurred more frequently, and at higher levels, in the bulk water compared to the biofilm samples ( Figure 1, first draw, circles, and second draw, squares, compared to biofilm, triangles).  (a-f) and L. pneumophila (g-l) 16S rRNA qPCR as described in Section 4.6 from each of the six locations listed on the left. Each data point is the mean of duplicate wells with standard deviation. No sampling occurred in F2016-Sp2017 for location PVC-FC. The limit of detection for bulk water and biofilm samples were 1.6 log 10 genomic copies (GC) L −1 , and 1.3 log 10 GC cm −2 , respectively.
Biofilm samples from all locations had sporadic detection of Legionella and L. pneumophila except for the Faucet and Fountain locations (Figure 1, triangles). At the Faucet location, both were detected in all biofilm samples except for the Sp2017 and F2018 time points for L. pneumophila (Figure 1e

Free-living Amoebae and Mycobacterium intracellulare Occurrence
To determine if Legionella occurrence correlated with the presence of their eukaryotic hosts and another water-based human opportunistic pathogen, bulk water and biofilm samples at each location were analyzed for two different free-living amoebae, Vermamoeba vermiformis and Acanthamoeba spp., and Mycobacterium intracellulare ( At location PVC-MA, V. vermiformis was detected in 80% of the bulk water samples (9/10 first draw; 7/10 second draw) and 40% of the biofilm samples (4/10). V. vermiformis was detected in 43% of the bulk water samples at the PVC-FC (1/7 first draw; 5/7 second draw); 30% at the Spigot (6/10 first draw; 0/10 second draw); and 45% at the Faucet (7/10 first draw; 2/10 second draw) locations (

Additional Sampling Sites
During the W2019 time point, 1 L of first draw samples were collected at two additional locations within the large BWS: an outlet off of the incoming main water line located in the boiler room (BWS Supply Line) and a recirculating pipe loop (PVC-Loop) previously described [32]. The bulk water sample from the BWS Supply Line had a turbidity of 0.34 NTU; 0.00/0.05 mg L −1 free/total chlorine; pH of 8.53; temperature of 39.5 • C; and HPC levels of 1.6 log 10 CFU 100 mL −1 . For the PVC-Loop, the bulk water sample had a turbidity of 0.21 NTU; 0.01/0.01 mg L −1 free/total chlorine; pH of 8.31; temperature of 20.6 • C; and HPC levels of 4.7 log 10 CFU 100 mL −1 .
The BWS Supply Line contained 3.0 log 10 CFU and MPN 100 mL −1 of L. pneumophila sg1 and 2-14 as confirmed by colony lysate PCR and latex agglutination. Two colonies were Legionella and L. pneumophila PCR positive, identified as sg1, and processed for whole genome sequencing. The BWS Supply Line sample was negative for Acanthamoeba spp. and M. intracellulare but contained 3.9 log 10 CE 100 mL −1 of V. vermiformis, 7.3 log 10 GC 100 mL −1 of Legionella spp., and 6.9 log 10 GC 100 mL −1 of L. pneumophila. The PVC-Loop bulk water sample was negative for V. vermiformis and M. intracellulare but contained 3.0 log 10 CE 100 mL −1 of Acanthamoeba spp., 7.4 log 10 GC 100 mL −1 of Legionella spp., and 7.3 log 10 GC 100 mL −1 of L. pneumophila.
The PVC-Loop had a high level of non-Legionella background that negatively impacted the enumeration of presumptive Legionella colonies. Although Legionella CFU could not be determined, 3.4 log 10 MPN 100 mL −1 of L. pneumophila was detected using Legiolert ® . Four wells of the PVC-Loop Legiolert ® tray, read as L. pneumophila positive following manufacturer's protocols, were sampled, and pure colonies were obtained as described in Section 4.4. Legiolert ® isolates were analyzed by PCR and latex agglutination. One of the four Legiolert ® isolates was Legionella and L. pneumophila PCR negative and processed for whole genome sequencing. Three of the four Legiolert ® isolates were Legionella and L. pneumophila PCR positive with 2/3 giving an inconclusive agglutination result and 1/3 identified as sg5. This PCR positive, L. pneumophila sg5 isolate was processed for whole genome sequencing.

Statistical Correlations between Water Quality Characteristics
Correlation analysis was performed, as described in Section 4.8, to determine if there were negative or positive associations between the observed water quality characteristics within each sampling location. As expected, there was a strong positive correlation between free and total chlorine in both the first and second draw bulk water samples at all locations supplied with chlorinated water, (r = 0.7-1.0, P < 0.05). For the other pairwise comparisons within each location, only the statistically significant (P < 0.05) correlations between water quality characteristics are shown in Figure 3.
Legionella and/or L. pneumophila occurrence was negatively correlated with disinfectant residual at three of the six sampling locations. Legionella spp. was negatively correlated with NH 2 Cl (r = −0.8) and TCl (r = −0.9) in the first draw, but only with TCl in the second draw samples (r = −0.9), at the PVC-MA location. At the PVC-FC location, Legionella spp. and L. pneumophila were negatively correlated with TCl and free chlorine (FCl), respectively, only in the second draw samples (r = −0.8, P < 0.05). Similarly, L. pneumophila was negatively correlated to both FCl and TCl in the Spigot second draw samples (r = −0.7). Notably, for the Faucet location that had consistent levels of Legionella spp. and L. pneumophila (Figure 1e and k), as well as culturable Legionella (Table 3), no statistical correlations were made between Legionella and disinfectant residual (Figure 3, Faucet). Moreover, occurrence of V. vermiformis was also negatively associated with FCl and TC (r = −0.8) at only the Faucet location. Legionella and/or L. pneumophila occurrence was negatively correlated with disinfectant residual at three of the six sampling locations. Legionella spp. was negatively correlated with NH2Cl (r = −0.8) and TCl (r = −0.9) in the first draw, but only with TCl in the second draw samples (r = −0.9), at the PVC-MA location. At the PVC-FC location, Legionella spp. and L. pneumophila were negatively correlated with TCl and free chlorine (FCl), respectively, only in the second draw samples (r = −0.8, P < 0.05). Similarly, L. pneumophila was negatively correlated to both FCl and TCl in the Spigot second draw samples (r = −0.7). Notably, for the Faucet location that had consistent levels of Legionella spp. and L. pneumophila (Figure 1e and k), as well as culturable Legionella (Table 3), no statistical correlations were made between Legionella and disinfectant residual (Figure 3, Faucet). Moreover, occurrence of V. vermiformis was also negatively associated with FCl and TC (r = −0.8) at only the Faucet location.
At all sampling locations except for PVC-MA, Legionella spp. was positively correlated with L. pneumophila in the first draw, second draw, and biofilm samples depending on the location (Fountain, r = 0.6; PVC-R, PVC-FC, and Faucet, r = 0.8; and Spigot, r = 1.0). Additionally, at the Faucet location, culturable Legionella was positively correlated with L. pneumophila detection in the first draw water samples (r = 0.7). For L. pneumophila and V. vermiformis, a positive correlation was found at locations PVC-FC (biofilm, r = 1.0) and Faucet (first draw, r = 0.6). Both positive and negative correlations between HPCs and chlorine residuals were observed at three of the six locations. There were positive correlations between HPC and FCl in the first draw (PVC-FC, r = 0.9) and second draw (PVC-R, r = 0.7) samples and negative correlations between HPC and both FCl and TCl in the first draw samples at the Faucet location (r = −0.7). At the Faucet location, HPCs were negatively correlated with pH (r = −0.7) in the first draw samples. HPCs were also negatively associated with culturable Legionella in the There were strong correlations between Legionella detection and various physiochemical water quality parameters. Legionella spp. was negatively correlated to temperature at only one location (Spigot, second draw, r = −0.7). Turbidity (NTU) was positively correlated with L. pneumophila detection in the Faucet second draw samples (r = 0.8); however, both positive and negative correlations were observed between NTU and Legionella spp. detection at two separate locations PVC-MA (first draw, r = 0.8) and Spigot (first draw, r = −0.7), respectively. Only a negative correlation was found between NTU and V. vermiformis detection in the first draw samples at two locations, Spigot (r = −0.8) and Fountain (r = −0.7).
Correlations between the physiochemical water quality parameters included a positive association between temperature and both NH 2 Cl and FCl at the PVC-MA (first draw, r = 0.7) and between temperature and FCl at the Spigot (second draw, r = 0.7) locations, respectively. There was a negative correlation between NTU and pH in the second draw samples at the PVC-MA (r = −0.6) and PVC-FC (r = −0.9) locations; and negative correlation between NTU and NH 2 Cl (r = −0.9) and TCl (−1.0) in the first draw samples at the PVC-MA location.

Whole Genome Sequencing of Drinking Water Isolates
Nineteen representative L. pneumophila and non-pneumophila isolates from the Su2018, F2018, and W2019 time points and PVC-FC, Faucet, BWS Supply Line, and PVC-Loop locations were submitted for whole genome sequencing as described in Section 4.7 (Table 4). All seven Faucet isolates from the Su2018 (Faucet 1-5) and W2019 (Faucet Legiolert ® 1 and Faucet 2) time points, and the W2019 BWS Supply Line 1 and 2 isolates, were identified as L. pneumophila Sequence Type (ST) 1 with an average genome size of 3.6 million base pairs (Mbp), a guanine-cytosine (G + C) content of 38%, and approximately 3200 predicted genes. The average nucleotide identity (ANI) between these seven Faucet and two BWS Supply Line isolates was between 99.97% and 100% indicating that these isolates were the same L. pneumophila strain (Figure 4). These isolates were also identified as sg1 via latex agglutination as described in Section 2.2 and 2.4.
Between the L. pneumophila sg5 ST1950 isolate and sg5 ST2037 isolates, the ANI was 92%, suggesting that they belong to the same species, but are different strains ( Figure 4). Notably, the nine L. pneumophila sg1 ST1 isolates had a higher ANI of 96% with the L. pneumophila sg5 ST1950 isolate compared to the 92% ANI with the L. pneumophila sg5 ST2037 isolates. This higher nucleotide similarity between the L. pneumophila sg1 ST1 and L. pneumophila sg5 ST1950 can be visualized with their clustering in the phylogenetic tree ( Figure 5). The L. pneumophila sg1 ST1 strains (W2019 BWS Supply Line, Su2018 and W2019 Faucet) were in the same branch as the L. pneumophila sg5 ST1950 strain, while the L. pneumophila sg5 ST2037 strain (Su2018 PVC-FC) was in a separate branch. The PVC-FC isolates were identified as L. feeleii via indirect immunofluorescent antibody assay and sequence similarities to two L. feeleii reference genomes (strain WO-44C and NCTC11978) which displayed >99.9% similarity based on 16S rRNA and ANI of 98.5%. Figure 5 shows the L. feeleii isolates in their own cluster with the reference genomes of L. massiliensis and L. nautarum in the next closest branch. The two L. feeleii reference genomes were not included in the sequence database at the time of this analysis and thus are not represented in the phylogenetic tree ( Figure 5).

Discussion
The main objective of this study was to gain a better understanding of water quality parameters that correlated with Legionella occurrence within a large complex building water system (BWS) over an extended period of time. Cold bulk water and biofilm samples were collected to monitor the occurrence and distribution of Legionella, other water-based pathogens, and eukaryotic hosts, while considering seasonal fluctuations, and varying engineering, operational, and water quality characteristics. Cold water was analyzed in this study given (1) the high prevalence of Legionella contamination previously identified in cold water samples within public building, private residences, healthcare facilities, and water storage tanks; (2) Legionella transmission and infections epidemiologically linked to cold water exposure; and (3) the current recommendations to monitor both hot and cold water in BWS to control for Legionella [33]. In this study, both culture and molecular methods were used for Legionella detection and culture isolates were submitted for whole genome sequencing for further genetic characterization.
Two percent of biofilm samples (1/57) and 12% of bulk water samples (14/114) were culturepositive for Legionella of which 57% (8/14) were isolated during the winter; 21% (3/14) during the summer; 14% (2/14) during the fall; and 7% (1/14) during the spring. Legionnaires' disease (LD) cases typically peak during the summer and fall seasons [34,35]. However, other epidemiological studies reported a winter peak for non-travel related cases [36] and no monthly or seasonal correlations for nosocomial-acquired [37] and community-acquired Legionella pneumonia cases [38]. Higher incidences of legionellosis have been associated with various meteorological factors (e.g., humidity, temperature, rainfall, atmospheric pressure); geographic location; and properties of the local watershed and source water (e.g., proximity and water temperature, levels, and flow) [39][40][41][42]. Thus, the exact environmental mechanisms and triggers of Legionella transmission and subsequent disease incidences are yet to be clearly defined.
Culture and qPCR are frequently used methods for Legionella environmental detection [43][44][45]. However, due to discrepancies between and within each of these methods, interpretations of Legionella occurrence across various studies and correlations to legionellosis health risks within BWSs are challenging [46]. Moreover, false positivity rates for the Legiolert ® method have been reported to be between 0% and 3.3% for potable water and 4.9% and 11% for non-potable water [47][48][49]. Such as

Discussion
The main objective of this study was to gain a better understanding of water quality parameters that correlated with Legionella occurrence within a large complex building water system (BWS) over an extended period of time. Cold bulk water and biofilm samples were collected to monitor the occurrence and distribution of Legionella, other water-based pathogens, and eukaryotic hosts, while considering seasonal fluctuations, and varying engineering, operational, and water quality characteristics. Cold water was analyzed in this study given (1) the high prevalence of Legionella contamination previously identified in cold water samples within public building, private residences, healthcare facilities, and water storage tanks; (2) Legionella transmission and infections epidemiologically linked to cold water exposure; and (3) the current recommendations to monitor both hot and cold water in BWS to control for Legionella [33]. In this study, both culture and molecular methods were used for Legionella detection and culture isolates were submitted for whole genome sequencing for further genetic characterization.
Two percent of biofilm samples (1/57) and 12% of bulk water samples (14/114) were culture-positive for Legionella of which 57% (8/14) were isolated during the winter; 21% (3/14) during the summer; 14% (2/14) during the fall; and 7% (1/14) during the spring. Legionnaires' disease (LD) cases typically peak during the summer and fall seasons [34,35]. However, other epidemiological studies reported a winter peak for non-travel related cases [36] and no monthly or seasonal correlations for nosocomial-acquired [37] and community-acquired Legionella pneumonia cases [38]. Higher incidences of legionellosis have been associated with various meteorological factors (e.g., humidity, temperature, rainfall, atmospheric pressure); geographic location; and properties of the local watershed and source water (e.g., proximity and water temperature, levels, and flow) [39][40][41][42]. Thus, the exact environmental mechanisms and triggers of Legionella transmission and subsequent disease incidences are yet to be clearly defined.
Culture and qPCR are frequently used methods for Legionella environmental detection [43][44][45]. However, due to discrepancies between and within each of these methods, interpretations of Legionella occurrence across various studies and correlations to legionellosis health risks within BWSs are challenging [46]. Moreover, false positivity rates for the Legiolert ® method have been reported to be between 0% and 3.3% for potable water and 4.9% and 11% for non-potable water [47][48][49]. Such as with Ochrobactrum identified in the PVC-Loop location, stable colonization of a Legiolert ® false positive causing strain at an environmental monitoring site may continually overestimate L. pneumophila levels and confound interpretations of their occurrence at those sites. Culture-based methods are a reliable indicator of pathogen viability and potential health risks associated with Legionella detection; however, periodic validation of culture-based results can be performed using molecular methods.
In this study, the PVC-R W2018 first draw sample was Legionella culture positive, but Legionella qPCR negative (Table 3, Figure 1b), while the PVC-FC Su2018 second draw sample was Legionella and L. pneumophila culture positive, but only L. pneumophila qPCR negative (Table 3, Figure 1c and i). Culture-positivity/qPCR-negativity has been observed previously and was associated more with drinking water compared to cooling tower water samples [50][51][52]. Furthermore, L. pneumophila qPCR levels were higher than those of Legionella spp. in four first draw samples (Spigot F2017, W2019 and Fountain F2017, Su2018) (Figure 1d, j and f, l) and L. pneumophila was also observed in hot water, but not cooling tower water samples [53]. These discrepancies have been attributed to the presence of PCR inhibitors and competing non-target organisms, and varying culture conditions (e.g., growth temperature, agar type) and sample processing steps that may select for, or inhibit growth of, different strains of Legionella [46].
Differences in physiochemical and microbial water quality parameters between first and second draw samples were previously reported with microbial loads generally higher in the first draw depending on the sampling location and volume collected [56,57]. As stated in Section 2.1, there were no statistical differences in pH, temperature, free chlorine, monochloramine, and total chlorine between the first and second draw samples at all locations with the exceptions of temperature for the Spigot samples and free and total chlorine for the Faucet samples. Three of the six locations had statistical differences between the first and second draw samples for HPCs: PVC-R, Spigot, and Faucet. For Legionella levels detected by qPCR, there were strong positive correlations between the first and second draw samples from all locations: PVC-MA (r = 0.9, P < 0.001), PVC-R (r = 0.6, P < 0.05), PVC-FC (r = 0.9, P < 0.01), Faucet (r = 0.8, P < 0.01), and Fountain (r = 0.8, P < 0.01), except for the Spigot location (r = 0.2, P = 0.575). For L. pneumophila levels as detected by qPCR, there were only strong positive correlations between the first and second draw samples at three of the six locations: PVC-MA (r = 1.0, P < 0.001), PVC-FC (r = 0.9, P < 0.01), and Faucet (r = 0.7, P < 0.05).
The results between the bulk water samples suggested that an analysis of either the first or second draw samples was able to indicate the presence of Legionella. Moreover, it is unclear why only the Faucet location showed statistical differences of both free and total chlorine between the first and second draw samples as water usage at this site was higher than two other locations and the distance of the Faucet outlet was greater than and almost equal to three other sampling locations (Table 1). The only difference between the Faucet location and all others was the ability to draw hot water from this outlet, but only cold water was analyzed in this study; thus, hot water may be contributing to bacterial contamination in the Faucet.
After conversion to monochloramine disinfection within BWSs, there were reductions in the number of distal sites testing positive for Legionella (39-100% to 0-18% positivity); however, there was a large range in the log reduction of Legionella levels (0.2 to 3 log 10 CFU L −1 ) with one study reporting no changes in levels post-conversion during the one-to three-year monitoring period [58][59][60]. Control of biofilm-associated Legionella was also observed [58,61], most likely due to the better penetration of monochloramine into biofilms compared to chlorine [62]; however, control of biofilm-associated L. pneumophila (Lp) was previously reported to be pipe material specific during chlorine and monochloramine treatment [63]. The PVC-MA location used in this study has been operating with a monochloramine residual for approximately 10 years with a two-month chlorine conversion from December 2013 to February 2014. During the sampling period, no culturable Legionella was detected, but molecular analyses indicated a consistent and high level of Legionella and sporadic detection of L. pneumophila in the bulk water and biofilm samples. Specifically, 100% (20/20) of the bulk water samples and 10% (1/10) of biofilm samples had detectable Legionella; while L. pneumophila was detected in 30% (6/20) of bulk water samples (three first draw and three second draw samples) and 10% (1/10) of biofilm samples (Figure 1a and g). Using E. coli, monochloramine was shown to disrupt protein-mediated metabolic processes with no damage to the cell envelope or nucleic acids [64]. This finding is supported by other studies demonstrating no significant impacts on Legionella 16S rRNA gene transcript levels during monochloramine versus chlorine treatment [63] and Pseudomonas aeruginosa extracellular polymeric substances material limiting and delaying monochloramine access to the cell surface [65]. Thus, it is unclear what the exact mechanisms are for monochloramine control of Legionella bacteria and whether different surface properties (e.g., lipopolysaccharides used for L. pneumophila serogroup identification) would result in varying degrees of inactivation with intermediate stages allowing for Legionella regrowth.
Numerous studies have reported more frequent detection of V. vermiformis within BWSs compared to Acanthamoeba spp. [56,66,67]. Concordantly, Acanthamoeba spp. were undetectable in the bulk water and biofilm samples during F2016, thus the following five sampling time points were excluded; however, the analysis was resumed in Sp2018 to confirm the low frequency and/or undetectable observation for Acanthamoeba spp. in BWS samples (Figure 2g-l). Acanthamoeba spp. was detected in only one second draw sample at the PVC-MA location (Figure 2g). Similarly, M. intracellulare was infrequently detected in this study (Figure 2n and p) as previously observed for water samples from chlorinated BWSs [23,68]. V. vermiformis was detected at all locations except PVC-R (Figure 2a-f) with positive correlations between L. pneumophila and V. vermiformis in PVC-FC biofilms and Faucet first draw samples. Given that FLA detection has been correlated to Legionella and Mycobacterium in drinking water systems [69], there is utility in monitoring for FLA to better understand the conditions and FLA members contributing to Legionella and mycobacterial survival within BWSs.
Although diverse populations of Legionella bacteria have been described in drinking water [70,71], environmental monitoring of BWSs focuses primarily on L. pneumophila detection since this species makes up the majority of clinical isolates, with > 80% of those isolates belonging to serogroup (sg) 1 [72,73]. Serotyping allows for the differentiation of L. pneumophila isolates based on their reactivity to the Dresden Panel of antibodies that recognize distinct structures on the bacterial lipopolysaccharide molecule [74,75]. L. pneumophila contains 17 serogroups and 10 subgroups within sg1 [76,77]. The latex agglutination serotyping method used in this study has been shown to produce false negatives as Legionella-like colonies isolated from drinking water samples, confirmed as L. pneumophila via 16S rRNA sequence analyses, were agglutination negative (this study; [78]). Thus, molecular-based methods, such as sequence-based typing (SBT) and whole genome sequencing (WGS), are increasingly being used due to their reliability, better resolution, and discriminatory power for describing genetic diversity, environmental distribution, evolution, population structure, clonal expansion, and virulence properties of Legionella isolates [77,79].
In this study, L. feeleii, L. pneumophila sg1 sequence-type (ST) 1, and L. pneumophila sg5 ST 1950 and ST2037 were identified via 16S rRNA-based analyses, serotyping (latex agglutination and indirect immunofluorescent antibody assays), WGS, and SBT analyses. These unique Legionella strains were isolated from the Faucet, PVC-FC, PVC-Loop, and BWS Supply Line locations during the Su2018, F2018, and W2019 time points (Table 4 andFigure 4). ST1 is the most commonly identified and globally distributed strain isolated from both environmental and clinical samples ( [79,80]; this study). Identification of previously unknown STs (e.g., ST1950 and ST2037 from this study) supports the continued SBT of clinical and environmental isolates. This will help elucidate which STs may be more globally distributed or geographically confined, and which are more associated with disease cases, such that when those are environmentally identified, preventative measures can be implemented to limit public health and exposure risks to these pathogens. WGS analyses provide more detailed genetic information about the Legionella strain, beyond those obtained from SBT alone, enabling potential subspecies identification, refined taxonomic classification, and genetic profiling for virulence properties [77]. Legionella diversity and distribution data may also reveal environmental parameters that influence Legionella occurrence and survival within specific environments such as BWSs.
Due to lengthy incubation periods required for Legionella culture, the tendency for qPCR to overestimate their levels, and the discrepancies associated with these methods, as described above, use of other microbial and/or physiochemical water quality parameters as potential indicators for Legionella presence in BWSs has been investigated [70]. As described in Section 2.5, there were strong statistical correlations observed between various water quality parameters and Legionella occurrence such as V. vermiformis, HPC, chlorine residual, temperature, and turbidity ( Figure 3). However, these correlations were location and sample type specific with conflicting positive and negative correlations for turbidity and HPC. Conflicting correlations between the latter and Legionella levels have been reported previously with either strong correlations [81], no correlations [82][83][84][85], or possible seasonal dependencies for these correlations [7]. Other conflicting correlations were reported between Legionella occurrence and pH, temperature, various minerals and metals (Ca, Cu, Fe, Mg, Mn, and Zn), total organic carbon, conductivity, and free chlorine at the sampling site and building supply feed [85][86][87].
Thus, further analyses of existing data and correlations to Legionella occurrence, as well as more in-depth studies on identifying these correlations, need to be performed. This will enable determinations as to whether correlations are dependent on complex factors, either individually or in concert, such as disinfectant type; source water quality fluctuations; genetic background of detected Legionella populations; presence of other drinking water microorganisms; or certain engineering and operational water system aspects specific to sampled locations. The notable observations from this study were (1) detection and culture of Legionella from outlets that neither supply, nor are connected to plumbing for, hot water; (2) isolation of diverse L. pneumophila and non-pneumophila strains from different locations; (3) utility of whole genome sequencing and sequence-based typing for enhanced isolate description and characterization of their distribution; (4) Legionella levels detected during monitoring can significantly differ between the first and second draw sample; and (5) negative and positive correlations between Legionella and various water quality parameters were location and sample type specific.
Confoundingly, occurrence alone is not the most important factor for legionellosis risk. Environmental, bacterial, and host specific factors such as aerosolization into respirable droplets and their potential for human exposure; virulence of the environmental Legionella strain; and host immune status and susceptibility to infection, collectively play an important role in exposure risks and disease outcome. The main objective of this study was to gain a better understanding of Legionella occurrence and water quality parameters supporting their growth within a large, complex building water system. Information from this, and future studies, will help elucidate ways to effectively manage the risks associated with Legionella exposure within these drinking water distribution systems.

Sampling Locations
A 40-year old, 33,000 square ft. building with an average potable water usage of 3.6 million gallons (13.6 million L) per year was used in this study. Water usage at any given location varies widely depending upon the activity (floor washing, water storage tank cleanout, etc.) being conducted and the facility cooling demand during warmer months. The building's potable water supply is derived from river water treated by coagulation, flocculation, and sedimentation; followed by sand, gravel, and granular activated carbon filtration; and then chlorination. Cold potable water samples were collected seasonally every three months from six locations throughout the building (Table 1, Figure S1). The total number of samples for each site was 30 (10 first and second draw bulk water and 10 biofilm samples) collected over a 28-month period, October 2016 to February 2019; except for site PVC-FC, where the total number of samples was 21 (seven first and second draw bulk water and biofilm samples) collected over an 18-month period, August 2017 to February 2019. Sampling time points are denoted F, for fall; W, for winter; Sp, for spring; and Su, for summer followed by the corresponding year. F, W, Sp, and Su samples were collected during the months of October, November, February, May, and August, respectively.
Within this building, a semi-closed pipe loop distribution system simulator was fed with the chlorinated municipal drinking water, described above, and amended with ammonium hydroxide and sodium hypochlorite (Sigma Aldrich, St. Louis, MO, USA) to yield a 2 mg L −1 monochloramine residual as previously described [88]. Average monochloramine and ammonia levels (± SD) during this sampling period were 1.25 ± 0.37 and 0.16 ± 0.07 ppm, respectively.

Sample Collection and Processing
For each sampling location, the first draw sample was taken immediately after turning the tap on, while the second draw was collected after 10 s of flushing (approximately 4 L), except for Fountain where the second draw sample was collected after 30 s of flushing (approximately 2 L). The 10-30-second flush time was used to ensure collection of non-stagnant water that was still representative of water quality conditions within the BWS. Sampling took place early in the morning after an overnight stagnation period. Water samples were collected in sterile 1 L plastic bottles and 1 mL of 10% w/v sodium thiosulfate was added to neutralize any disinfectant residual. An additional 100 mL was also collected for water quality analysis as described below. Approximately 1 L of each bulk water sample was filtered through a 0.2 µm polyethersulfone membrane (Supor ® Membrane, PALL Life Sciences, Nassau, NY, USA). Filters were placed into 11 mL of UV-light dechlorinated, 0.22 µm filtered drinking water (dfH 2 O), and vortexed at maximum speed for 1 min to resuspend the concentrated bulk water material. For biofilm collection, a sterile polyester tipped applicator was used to swab an approximate area of 2 cm 2 inside the tap. The applicator was then placed in a 14-mL round bottom tube containing 2 mL of dfH 2 O and vortexed vigorously for 1 min to resuspend the collected biofilm material.
Approximately 1 mL of the concentrated bulk water and biofilm suspension was analyzed for CFU, as described in Section 4.4, and the remaining volume was centrifuged at high speed (13,000 rcf, room temperature, 10 min; Eppendorf, Foster City, CA, USA). Pellets were resuspended in 200 µL of dfH 2 O and placed in a Lysing Matrix A tube (MP Biomedicals, Solon, OH, USA) along with the washed filter or biofilm swab for nucleic acid extraction as described below.

Water Quality Analysis
Bulk water samples were analyzed for pH, turbidity, temperature, disinfectant residual, and heterotrophic plate count (HPC). Free chlorine and total chlorine measurements were performed using the DPD colorimetric method (Powder Pillows; Hach USA) and monochloramine and free ammonia measurements were performed using the indophenol method (method 10200, Powder Pillows, free ammonia chlorinating solution; Hach USA). HPCs were enumerated by the spread plate method on Reasoner's 2A agar (R2A, Difco Laboratories, Detroit, MI, USA) following incubation at 28 • C for 7 d. The limit of detection (LOD) for bulk water samples was 1.0 log 10 CFU 100 mL −1 and 0.7 log 10 CFU cm −2 for biofilm samples.

Legionella Enumeration and Presumptive Colony Analysis
For colony forming unit (CFU) enumeration, undiluted and serially diluted suspensions were spread plated on buffered charcoal yeast extract (BCYE) agar plates (BD Diagnostics, Franklin Lakes, NJ, USA) and incubated for 4-6 days at 37 • C [44]. Presumptive Legionella colonies were counted; and a subset was isolated and confirmed as Legionella spp. or L. pneumophila via polymerase chain reaction (PCR) using the 16S rRNA gene assays described in Section 4.6. An aliquot of the processed bulk water and biofilm samples was also heat-treated (incubation in a 55 • C water bath for 30 min) before plating on BCYE agar plates to evaluate potential differences in Legionella recovery from this pretreatment method [44]. Although growth of non-Legionella bacteria was inhibited by heat treatment, there were no significant differences between Legionella CFU observed between unheated and heated samples (data not shown).
For most probable number (MPN) enumeration, Legiolert ® (Idexx Laboratories, Westbrook, ME, USA) was used to analyze 10 mL of the unconcentrated bulk water samples and 0.5 mL of the resuspended biofilm samples for only the Su2018, F2018, and W2019 time points. To obtain pure isolates from the Legiolert ® tray, positive wells were punctured using a 26-gauge needle and 50-1000 µL of the well contents was collected. A 20 µL aliquot of the sampled well was streaked onto a BCYE agar plate and incubated for 4-6 days at 37 • C.
To account for zero values, 1 was added to all data points before conversion to the log10 scale (e.g., log10 (CFU + 1)). Calculations from CFU and molecular analyses were adjusted and expressed as units per mL or cm 2 for bulk water samples and biofilms, respectively. The LOD for bulk water samples was 1.0 log 10 CFU 100 mL −1 and 0.7 log 10 CFU cm −2 for biofilm samples.

Isolation and Preparation of Total DNA
DNA was extracted from bacterial cells using the MasterPure™ Complete DNA purification kit (Epicentre Biotechnologies Inc., Madison, WI, USA) according to manufacturer's protocol and the Mini-Beadbeater−16 (Biospec Products, Bartlesville, OK, USA) where samples were processed twice for 30 s at 3450 oscillations min −1 . The DNA pellet was resuspended in 100 µL of molecular grade water.

Quantitative Polymerase Chain Reaction (qPCR)
Biofilm and bulk water DNA samples were analyzed in duplicate using the Applied Biosystems QuantStudio 6 Flex Fast Real-Time PCR system (ThermoFisher, Waltham, MA, USA). A 10-fold dilution of each sample was also analyzed in duplicate to test for presence of environmental qPCR inhibitors. The TaqMan qPCR assay for Legionella spp., L. pneumophila, Mycobacterium intracellulare detection, targeting the 16S rRNA gene, was performed as previously described [63,89,90]. The TaqMan qPCR assay for Acanthamoeba spp. and SYBR green qPCR assay for Vermamoeba vermiformis detection, targeting the 18S rRNA gene, was performed as previously described [91,92].
The forward and reverse primers and probe sequences (5' to 3') and cycling parameters used in this study for the Legionella spp. qPCR assay, respectively, are 16S-LegF1c: TAG TGG AAT TTC CGG determined by alignment similarity to a select subset of COG (Clusters of Orthologous Groups) domains. The phylogenetic tree is reconstructed using FastTree 2 [95] to determine maximum likelihood phylogeny. Average nucleotide identity (ANI), an index of similarity between two genomes [96], was calculated using FastANI v1.3 (https://github.com/ParBLiSS/FastANI) [97]. ANI is defined as mean nucleotide identity of orthologous gene pairs shared between two microbial genomes. No ANI output is reported for a genome pair if the ANI value is below 80%.

Statistical Analysis
For each water quality parameter, a Shapiro-Wilk normality test was conducted for each site to determine distribution of the data throughout the sampling period. A one-way analysis of variance (ANOVA) using the Tukey multiple comparisons test was conducted between each site and sample type. P-values of < 0.05 were considered statistically significant. Analyses were performed using Prism 8 (GraphPad Software, San Diego, CA, USA). The R functions cor() and cor.test() were used, with a Spearman correction, to determine the direction (positive or negative) and significance of correlation between pairs of water quality characteristics within each sample location [98]. The correlation between the pairs was denoted by the number, r, which varies between -1 and +1, with 0 meaning no correlation, +1 a complete positive correlation, and -1 a complete negative correlation.
Supplementary Materials: The following are available online at http://www.mdpi.com/2076-0817/9/7/567/s1; Figure S1: Images of outlets at each sampling location; Figure S2: Scatterplot of turbidity measurements for bulk water samples; Figure S3: Scatterplot of free (A) and total (B) chlorine measurements for bulk water samples.