This study was approved by the Research Ethic Committee, Naval Medical Department, Royal Thai Navy and Walter Reed Army Institute of Research (WRAIR), Silver Spring, MD, USA. WRAIR Human Subjects Protection Branch determined that this was non-human subjects research (NHSR) and all samples and data associated with clinical isolates was de-identified prior to transfer to AFRIMS.

We carried out surveillance of hospital wastewater effluent from Queen Sirikit Naval Hospital in Chonburi, Thailand for a ten month period from March to December 2024. An overview of the monthly testing process is shown in Supplementary Figure S1.

Water samples comprising of untreated hospital wastewater, treated hospital wastewater, and community water (local public water reservoir) were collected once a month (at the first week of each month) from March to December 2024, spanning both the dry and wet seasons. In each collection, water was manually sampled with volumes of 500 ml for untreated water and 5 L for treated water and community water as single-time point grab sample. These volumes were selected based on preliminary data on the bacterial burden in untreated wastewater, treated wastewater, and community water at these sites. Upon collection, the water samples were refrigerated during transportation to the laboratory to ensure sample integrity. Water samples were stored at 4 °C and processed within 24 h after collection.

At the laboratory, water samples were filtered through 5 μM membrane (Merck Millipore, Burlington, MA) to remove the large particles. Afterwards, the filtrate was filtered through 0.45 μM membranes (Merck Millipore, Burlington, MA) and thoroughly rinsed the filtered membranes with 5 ml Phosphate Buffered Saline (PBS; Sigma-Aldrich, Burlington, MA). To isolate bacteria, the suspension was performed 3-fold dilution in PBS then 200 μl of the diluted suspension was spreaded onto the large petri-dish (diameter 150 mm) containing Brucella agar (Becton Dickinson, Franklin Lakes, NJ) with 5% sheep blood, CHROMagar^TM MRSA, mSuperCARBA, extended-spectrum beta-lactamase (ESBL), and VRE plates (CHROMagar™, Paris, France) and Eosin-methylene blue (EMB) agar plates (Becton Dickinson, Franklin Lakes, NJ) supplemented with 4 μg/ml colistin (Sigma-Aldrich, Burlington, MA). Brucella Blood agar plate, representing non-selective media, was used to identify total number of bacteria, while antibiotic plates were used to identify antibiotic-resistant bacteria. The plates were incubated under aerobic conditions at 37 °C for 24–48 h. Colony forming units (CFUs) were enumerated and converted into CFU/ml. In order to provide a relative measure for the degree of antimicrobial resistance in a given sample, a Resistance Index was calculated based on the ratio of the number of colonies found on a selective plate with the number of colonies found on a non-selective blood agar plate.

Bacterial isolates on each antibiotic agar plates were further identified using MALDI-TOF (Bruker MALDI Biotyper system, Bremen, Germany) (Rahman et al., 2023). The lyophilized Bruker HCCA (4-Hydroxy-α-cyanocinnamic acid) matrix 2.5 mg was dissolved in 250 μl of standard solvent (50% acetonitrile, 47.5% water, and 2.5% trifluoroacetic acid). Mass spectrometry was performed and the bacterial isolates were analyzed and identified by Bruker MALDI Biotyper system and database (MALDI Biotyper 4.1.60 Software, Bremen, Germany). MALDI scoring method was used for interpretation. If the obtained score was ≥2.00, the genus and species identification were acceptable with a high confidence. If the score was 1.70–1.99, only the genus identification was acceptable. If the score was below the cut-off (1.70), the sample was recorded as no identification possible (Buchan et al., 2012).

We carried out a two-step process for AST. As an initial screen, ESKAPEE isolates identified through mass spectrometry were subjected to AST by disk diffusion to identify isolates that show an MDR phenotype by testing against the following antibiotics: amikacin, gentamicin, trimethoprim/sulfamethoxazole, ciprofloxacin, cefepime, cefotaxime, ceftazidime, ceftriaxone, imipenem, meropenem, tetracycline, ampicillin, penicillin, and vancomycin. The subset of isolates that showed an MDR phenotype in the initial screen were then analyzed using the BD Phoenix^TM M50 system (Becton Dickinson, Franklin Lakes, NJ), using the NMIC/ID-504 panel for gram-negative ESKAPEE pathogens and PMIC/ID-95 panel for gram-positive ESKAPEE pathogens, according to the manufacturer's instructions (BD Diagnostics, Sparks, MD). Interpretations of resistance phenotypes from the disk diffusion method and the BD Phoenix^TM M50 system were made following CLSI guidelines (CLSI., 2024), using the guidelines for clinical isolates so that direct comparisons could be made between clinical and wastewater isolates.

Genomic DNA from bacterial isolates were extracted using the DNeasy Blood and Tissue Kit (QIAGEN, Hilden, Germany). The concentration and purity of the extracted DNA was using a Nanodrop 2000c Spectrophotometer (Thermo Scientific, Waltham, MA). The DNA, at a concentration between 1 and 10 ng/μl, was used as the template for PCR and real-time PCR and was stored at −20 °C until use. Real-time PCR assays were performed as described previously to determine the presence of carbapenemase antimicrobial resistance genes (bla_NDM and bla_KPC) (Milillo et al., 2013) and methicillin resistance gene (mecA) (Mc Gann et al., 2013) using SensiFAST Probe No-ROX Mix (Bioline, London, UK) on CFX96 Touch Deep Well™ Real-Time PCR Detection System (Bio-Rad, Hercules, CA). Analysis was performed using Bio-Rad CFX Manager software version 3.1 (Bio-Rad, Hercules, CA). PCR assays were performed as described previously to determine the presence of ESBL genes (Dallenne et al., 2010), carbapenemase genes (bla_{OXA − 23}, bla_{OXA − 24}, bla_{OXA − 51}, bla_{OXA − 58}) (Feldgarden et al., 2019), colistin resistance genes (mcr-1 to mcr-9) (Rebelo et al., 2018; Borowiak et al., 2020), and vancomycin resistance genes (vanA, vanB, vanC, vanD, vanE, and vanG) (Pourcel et al., 2017) using the AmpliTaq^® Gold DNA polymerase (Thermo Fisher Scientific, Waltham, MA) on Mastercycler^® nexus (Eppendorf, Hamburg, Germany). Primers for mcr-10 were designed based on full-length mcr-10 gene (GenBank accession number MT468575) using Primer3Plus (https://www.primer3plus.com/index.html). The sequences are as follows: MCR-10_133F (5′-GCAATAACCCGACGCTGAAC-3′) and MCR-10_133R (5′-GTAACGCGCCTTGCATCATC-3′). These primers were incorporated into mcr-6 to mcr-9 multiplex (Borowiak et al., 2020).

We obtained 743 clinical isolates from the Microbiology Laboratory at Queen Sirikit Naval Hospital (QSH) as part of on-going AMR surveillance effort between AFRIMS and QSH between March and December of 2024. Bacteria isolated from clinical samples (e.g., pus, urine, rectal swabs, sputum, and blood) that were collected as part of routine medical care in both inpatient and outpatient settings were tested at QSH using the VITEK2 system (bioMérieux, Durham, NC). Isolates that were determined to be ESKAPEE organisms and classified as MDR based on non-susceptibility to at least one drug in three or more antimicrobial drug classes (Magiorakos et al., 2012) were selected and transferred to AFRIMS for confirmation of both speciation and AST phenotypes by the BD Phoenix^TM M50 system. The pure isolates were grown on tryptic soy agar plates and Brucella Blood agar plates in preparation for analyses.

Wastewater isolates (n = 52) were selected for whole genome sequencing (WGS) based on whether they had a similar AST phenotype profile as clinical isolates collected from the same hospital. This was done to maximize the use of limited WGS resource toward identifying wastewater isolates of high clinical relevance. As part of on-going clinical surveillance at QSH, 127 clinical MDR ESKAPEE isolates were also selected for WGS. DNA was extracted using the DNeasy UltraClean Microbial Kit (QIAGEN, Hilden, Germany). Library construction was performed with KAPA Library Quantification Kit (Roche Dianostics Corporation, Indianapolis, IN) for sequencing on an Illumina MiSeq Benchtop sequencer (Illumina, San Diego, CA) as previously described (Galac et al., 2020). Minimum thresholds for contig size and coverage were set at 200 bp and 49.5+, respectively. Assembled sequences were annotated using Prokka v1.14.6 (Seemann, 2014).

Species identification was determined using Kraken2 (v2.0.8-β) and in silico ST detection was identified for all isolates using species specific schemes for E. coli, Enterobacter spp., A. baumannii, K. pneumoniae, P. aeruginosa, and S. aureus (www.pubmlst.org). Antimicrobial resistance genetic determinants were annotated using AMRFinderPlus (Feldgarden et al., 2019) and ARIBA (Hunt et al., 2017). Minimum spanning trees (MST) were generated using core genome MLST (cgMLST) with species specific cgMLST schemes (https://www.cgmlst.org/ncs). Isolates with 0-10 allelic differences are considered highly genetically related based on prior studies on epidemiological outbreak detection using cgMLST (Coles, 1988; Leeper et al., 2023). This genome-based genetic relatedness analysis was performed on the 52 wastewater isolates and 127 clinical isolates collected during the study period in addition to >1,600 clinical isolates collected from the same hospital and sequenced since September 2022.

Descriptive statistics was used to determine significant differences between bacterial burden from different water sources. Phenotypic analysis of AST profiles was analyzed using the phenotype classification as determined by the BD Phoenix^TM M50 system. We applied a clustering approach for identifying groups of isolates with similar AST phenotypes that uses a pairwise distances between AST phenotype profiles of each isolate identify groups of isolates with similar AST profiles based on a method developed by Sanderson et al. (2019). First each isolate, its AST phenotype profile was represented as a character string with the number of digits and position of each character corresponding to a drug in the AST panel, with the value of the character being “1” if the phenotype for that isolate for that drug was R, and “0” if the phenotype was S or I. AST profiles of gram-negative bacteria were represented by an 18-digit string while profiles for gram-positive bacteria were represented by a 19-digit string, corresponding to the number of antibiotics in the BD Phoenix^TM M50 testing panels. A distance between the AST phenotype profiles of two isolates was calculated as the Hamming Distance (Hamming, 1950) between their AST phenotype character strings, defined as the sum of mismatches between the two strings. For each organism, a pairwise distance matrix for all isolates was generated and used for hierarchical clustering using the hclust algorithm in the R statistical software platform. AST phenotype clusters were defined using the cutree function with the height parameter set to 1. Supplementary Figure S2 illustrates numbers of ESKAPEE in untreated wastewater characterized by MALDI-TOF through whole genome sequencing analysis.

Water samples were collected monthly from the hospital, filtered, cultured on selective and non-selective media and assessed by Resistance Index as a relative measure of drug resistance in the bacterial biomass (Figure 1A). We found that the Resistance Index varied significantly from month to month. For carbapenems, we found, in general, that untreated and treated water samples tended to have higher resistance than community water. For vancomycin, interestingly, we found that community water tended to have higher resistance. For ESBL resistance, we did find that untreated water had significantly higher resistance than treated water. For colistin, the peaked resistance was found in particular month such as September, October, and June for untreated water, treated water, and community water, respectively. Overall, we found that, in general, untreated water had higher cumulative CFU/ml than treated water and community water for all the selective plates (Figure 1B, Supplementary Table S1), showing that untreated hospital water was enriched for antibiotic resistance bacteria.

We selected drug resistant isolates from selective plates for speciation using MALDI-TOF mass spectrometry. Figure 2 shows the profile of both ESKAPEE and non-ESKAPEE pathogens in untreated wastewater, treated wastewater, and community water from the hospital. Overall, we found that a majority of drug resistant isolates from untreated hospital wastewater (1,550 of 2,735, 56.7%) were ESKAPEE pathogens, with K. pneumoniae, E. coli, and Enterobacter spp. being the most common, accounting for 48.9% of drug resistant isolates. By contrast, treated wastewater had significantly fewer ESKAPEE isolates (14.6%, 52 of 356 isolates), predominantly A. baumannii, and Enterobacter spp. Likewise, we found that ESKAPEE isolates represented a very small fraction (13 of 544 isolates, 2.4%) of drug resistant isolates in community water. Non-ESKAPEE isolates varied based water source with Achromobacter isolates being the most common in untreated hospital wastewater, Stenotrophomone isolates being the most common in treated hospital wastewater and Sphingobacter isolates being the most common in community water, with Pseudomonas spp. (non-aeruginosa) being the second most common in all three sources (Figure 2).

For each ESKAPEE isolate from the untreated hospital wastewater, we carried out AST using disk diffusion followed by the automated BD Phoenix^TM M50 system. Overall, 1,550 isolates were identified as ESKAPEE organisms by MALDI-TOF. Of that, 1,078 isolates were selected for AST by disk diffusion, and 367 of the isolates exhibiting MDR pattern were characterized by the BD Phoenix^TM M50 system (Figure 3). The most common drug resistant ESKAPEE isolates based on disk diffusion were Enterobacter spp., E. coli, and K. pneumoniae. Based on the BD Phoenix^TM M50 data, we classified isolates as MDR (non-susceptible to at least three classes of antibiotics), and XDR (non-susceptible to all but one or fewer classes of antibiotics). We found that most of the ESKAPEE isolates we collected (799 of 1,078, 74%) were MDR, with almost all (12 of 13, 92.3%) the P. aeruginosa isolates being XDR.

The breakdown of ESKAPEE isolates from hospital wastewater based on resistance types is shown in Supplementary Figure S3. Overall, we found that high rates of carbapenemase producers among A. baumannii, Enterobacter spp., K. pneumoniae, and P. aeruginosa isolates (60-90%). We found high rates of ESBL producers among E. coli and K. pneumoniae isolates (~70%). For colistin resistance, we observed relatively low resistance in E. coli and K. pneumoniae isolates (20–30%) but high reistance among Enterobacter spp. isolates (55%). Finally, we found that 24% of E. faecium isolates showed vancomycin resistance.

Our approach was able to capture gram-positive ESKAPEE pathogens, specifically 35 isolates of E. faecium. However, among MDR clinical isolates collected during the same period, the only gram-positive pathogen that was identified was S. aureus (17 out of 743 clinical isolates). Because of the low prevalance of gram-positive MDR pathogens among clinical isolates and the lack of overlap between gram-positive ESKAPEE pathogens in the clinical and wastewater samples (E. faecium in wastewater vs. S. aureus in clinical samples), the subsequent analyses focus on the following gram-negative ESKAPEE pathogens: A. baumannii, E. cloacae, K. pneumoniae, P. aeruginosa and E. coli.

We carried out AST using disk diffusion method followed by automated identification and testing for a subset of 416 wastewater isolates, of which 251 were confirmed as ESKAPEE pathogens. We combined this data with BD Phoenix^TM M50 data from 743 clinical ESKAPEE isolates and generated AST profiles for each isolate based on whether they were sensitive (S or I) or resistant (R) to each antibiotic in the drug panel (Sanderson et al., 2019). We then carried out hierarchical clustering to define clusters of isolates with similar AST profiles (Figures 4, 5). Overall, we found that for five ESKAPEE pathogens: K. pneumoniae, A. baumannii, E. coli, P. aeruginosa, and E. cloacae, we found many clusters where both clinical and wastewater isolates had similar or identical AST profiles.

We sought to determine the degree to which the wastewater isolates captured the diversity of clinical isolates (Table 1) by determining what percentage of clinical isolate clusters (representing unique clinical phenotypes) contained a wastewater isolate with a matching AST phenotype. For E. coli, we found that almost 50% of clinical AST phenotype clusters included a wastewater isolate with the same AST phenotype. For A. baumannii, K. pneumoniae, E. cloacae that number was 35, 28, and 27%, respectively, compared to 17% for P. aeruginosa. Overall, this shows that while some clinical AST phenotype patterns are captured in the AST phenotype profiles of wastewater isolates, many clinical phenotypes are still missing.

We also sought to determine the degree of coverage that wastewater isolates provide in capturing high-frequency clinical phenotypes (Table 2), by assessing what proportion of all clinical isolates have a wastewater isolate with the same AST phenotype. We found that for A. baumannii and P. aeruginosa, the coverage was relatively high, with 72 and 73% of clinical isolates being represented by a phenotypically matching wastewater isolate, respectively. For E. coli and K. pneumoniae, there was moderate coverage with 38 and 32% coverage of clinical isolates, while for E. cloacae there was very low coverage of 10%. When we assessed what proportion of wastewater isolates have a clinical isolate of the same AST phenotype, we found that for A. baumannii and P. aeruginosa, coverage is very high at 94 and 100%, respectively, suggesting that most or all of wastewater isolates for these two organisms may be clinically relevant. For E. coli there was more moderate coverage of 70%, while it was lower for E. cloacae and K. pneumoniae, at 38 and 17% coverage, respectively.

We carried out PCR-based testing of AMR genes in the ESKAPEE isolates from wastewater used for AST testing and compared it to AMR genes identified in clinical samples to determine the degree to which AMR genes identified in wastewater isolates recapitulated AMR genes identified in clinical isolates. A high degree of overlap would indicate that wastewater surveillance can be used to monitor the emergence of clinically relevant AMR genes in ESKAPEE pathogens. Table 3 shows the list of AMR genes tested in gram-negative ESKAPEE pathogens which shows which AMR genes were identified in at least one wastewater isolate (denoted by a “+”), and then the proportion of clinical isolates that also have that same gene. For K. pneumoniae, we found 80% sensitivity and 100% specificity in wastewater AMR genes identifying from clinical isolates, notably for bla_NDM, bla_TEM, bla_SHV, bla_OXA, and bla_{CTX − M−1} genes which showed high prevalence in clinical samples.

For P. aeruginosa and A. baumannii, we found that analysis of wastewater isolates had moderate sensitivity (~60%) and high specificity (100%) in identifying AMR genes from their respective clinical isolates, notably for high-prevalence genes bla_NDM and bla_VEB in P. aeruginosa and bla_NDM, bla_TEM, bla_{OXA − 23}, bla_{OXA − 51}, and bla_{OXA − 69} in A. baumannii. For E. coli, we found that wastewater isolates had high sensitivity (80%) but relatively low specificity (62%), with a number of AMR genes identified in wastewater isolates that were absent in clinical isolates, including bla_KPC, bla_{CTX − M−8/25}, bla_GES, bla_VEB, and bla_KPC. For these four ESKAPEE pathogens, it is notable that there were few cases of a highly prevalent AMR gene in a clinical isolate that was not found in the wastewater isolate—the bla_IMP gene in clinical P. aeruginosa isolates (45% prevalence in clinical samples) and bla_{OXA − 48} family genes in K. pneumoniae (19% prevalence in clinical samples).

For E. cloacae, we found low sensitivity (33%) and moderate specificity (57%), with numerous clinical AMR genes missing in the wastewater samples including bla_SHV, bla_OXA, bla_CMY, bla_EBC, and bla_PER. It is important to note that in most of the cases were an AMR gene from clinical isolate was absent in wastewater isolates, it had relatively low prevalence in the clinical samples, suggesting that the wastewater surveillance may have been limited by low sample size.

We selected a subset of 52 ESKAPEE isolates for WGS by selecting wastewater isolates that were found in the same AST phenotype cluster as clinical isolates including E. coli (n = 22), K. pneumoniae (n = 10), P. aeruginosa (n = 9), A. baumannii (n = 5), E. faecium (n = 3), and E. cloacae (n = 3). Using cgMLST analysis, we found that for A. baumannii (Figure 6), E. coli (Figure 7), K. pneumoniae (Figure 8), and P. aeruginosa (Figure 9), wastewater isolates have the same sequence type as clinical isolates, and were highly genetically related. For E. cloacae, the sequenced wastewater isolates were not genetically related to clinical isolates (Supplementary Figure S4) suggesting that any similarities in AST profiles in this pathogen are not reflective of relatedness.

Table 4 shows a summary of sequenced wastewater isolates that were found to be of the same sequence type (ST) as either a clinical isolate collected during the study period, or a historical clinical isolate collected earlier at the same site. For E. coli, we found 7 of 22 sequenced wastewater isolates were in the same ST as a clinical isolate, including an ST744 isolate and ST361 isolate where there were no allelic differences found between the wastewater and clinical isolates, indicating extremely high genetic relatedness. AST phenotypes for these shared strains include MDR and XDR phenotypes. For K. pneumoniae, we found 5 of 10 sequenced wastewater isolates were in the same ST as a clinical isolate, including two ST147 isolates and one ST37 isolate which also showed no allelic differences to a clinical isolate.

For P. aeruginosa, we found that 7 of 9 sequenced wastewater isolates shared the same ST type as a clinical isolate for ST233, ST235, ST654, and ST274, all of which showed less than 10 allelic differences to the nearest clinical isolate. Most of these shared strains had the XDR phenotype. Finally, for A. baumannii, we found 4 out of 5 sequenced wastewater isolates had the same ST type as clinical isolates for ST2 and ST164, in both cases with no allelic differences. This was most pronounced for ST2, where we had 3 of 5 wastewater isolates and over a 100 clinical isolates that were highly genetically related and had the XDR phenotype.