Could the load of carbapenemase genes in hospital wastewater be a proxy for emerging resistance to carbapenems in humans?

Abstract

Background

Antimicrobial resistance (AMR) poses a growing threat to global public health and is a key concern for infection control teams in hospitals. However, AMR surveillance is time-consuming and limited in most countries, resulting in incomplete findings. In high-income countries, infection control teams ensure the contact tracing of every patient carrying an emerging extensively drug-resistant bacterium which is very time-consuming. Wastewater surveillance (WWS) has been proposed as an alternative approach for the surveillance of infectious diseases. This study aims to test the feasibility of AMR WWS under real-world conditions in hospital. It investigates the dynamics of endemic (bla_CTX-M) and emerging AMR genes (bla_OXA-48, bla_NDM, bla_KPC and vanA) in wastewater from two hospital buildings where patients with contrasting risk for carrying resistant bacteria were cared for and compares results with clinical data.

Methods

The sampling programmes were adapted according to the sampling sites and patient flow for each hospital building. Genes were quantified in the effluent using qPCR and dPCR. Cultivable carbapenemase-producing Gram-negative bacteria were characterised using MALDI-TOF MS and PCR.

Results

The feasibility of AMR monitoring in wastewater in real hospital conditions was demonstrated by dPCR and qPCR, which produced correlated results. The presence of peaks and the low load of the vanA and bla_NDM genes in wastewater (compared to bla_CTX-M) were consistent with their known emerging status, as indicated by national and local clinical data. However, the high concentration of bla_OXA-48 and bla_KPC in wastewater was unexpected because it did not reflect the known clinical involvement of these emerging resistances, particularly in the case of bla_KPC. Bacterial culture also revealed discrepancies between the species isolated in wastewater and those isolated in patients in the hospital, with a majority of Citrobacter spp. carrying bla_KPC and bla_OXA-48 in wastewater, whereas Escherichia coli and bla_OXA-48 dominated in patients. Quantifying carbapenemase genes in wastewater was able to differentiate between buildings housing patients contrasting risks of emerging AMR.

Conclusion

This study shows the WWS feasibility in real hospital conditions and preliminary findings regarding patient populations but identified obstacles that need to be overcome prior to use WWS for routine surveillance in an infection control hospital context.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13756-026-01697-9.

Background

Wastewater is a complex matrix that transports a wide range of chemical and biological molecules. The analysis of drug residues in wastewater to monitor public health coined the concept of wastewater-based epidemiology (WBE), also called wastewater surveillance (WWS) [1]. After several attempts at WWS for viral infections [2], the monitoring of SARS-CoV-2 in wastewater as a proxy for human epidemiology highlighted the utility of WWS for emerging infectious diseases [3]. WWS is simpler and less costly than individual testing of a human population, particularly for infectious agents that are often associated with asymptomatic carriage, such as SARS-CoV-2 [3, 4] and multi- or extensively drug-resistant bacteria [5, 6]. In routine wastewater surveillance, quantitative PCR (qPCR) and digital PCR (dPCR) are the most widely used methods for quantifying nucleic acids in wastewater [4]. The use of dPCR is increasing due to its ability to provide absolute quantification without the need for a calibration curve, as well as its reported greater sensitivity, tolerance to inhibition, precision and reproducibility than qPCR [5]. However, both approaches still need to be comparatively evaluated for use in WWS, particularly for bacterial genes [5].

Antimicrobial resistance (AMR) is a growing threat to global public health because it can lead to increased morbidity and mortality from treatment failures, and can also slow progress in critical care or surgery [7]. The emergence, spread and increasing prevalence of carbapenemase-producing Enterobacterales (CPE) is of particular concern because carbapenemases inactivate carbapenems, which are considered the antibiotics of last resort for serious gram-negative infections. As a result, CPE has been identified by the World Health Organization (WHO) as one of the top three critical priorities [8]. CPE was spreading in an epidemic mode, while other multi-drug-resistant bacteria also identified by the WHO as critical priorities, such as extended-spectrum beta-lactamase (ESBL)-producing bacteria (mainly CTX-M-producing bacteria), are now spreading in an endemic mode [9]. In France, the most common gene associated with CPE is bla_OXA-48, followed by bla_NDM and less common genes (bla_KPC, bla_VIM, bla_IMI, bla_GES…) [10]. In addition to CPE, vancomycin-resistant enterococci (VRE) are emerging as a major cause of healthcare-associated infections and are among the highest global public health priorities [8]. VRE express the vanA or vanB genes, which cause cell wall modifications and confer resistance to vancomycin and other glycopeptides that are antibiotics of last resort.

In most European countries, surveillance for CPE and VRE is based on voluntary reporting by medical laboratories or hospital infection control teams (ICTs). In the hospital, detection of asymptomatic carriage is achieved by individual rectal screening, either systematically for patients in intensive care units or targeted only at patients at high risk of carriage, such as those hospitalised in high endemic areas or who have been in contact with a CPE or VRE-carrying patient. Contact tracing for each CPE or VRE case, as well as individual screening of high-risk patients, is poorly received by patients. It is also burdensome and time-consuming for healthcare workers, and it only provides patchy epidemiological data. There is clearly a need for more efficient tools to facilitate patient follow-up and enable more comprehensive surveillance of critical emerging resistant pathogens.

Given the faecal shedding of CPE and VRE, WWS of emerging AMR may be a reasonable option. The post-COVID-19 surge in WWS indications, combined with strong political will for the development of environmental AMR surveillance, could rapidly lead to the establishment of regulatory obligations for AMR WWS [11]. However, the correlation between WWS and clinical surveillance, i.e. between the antimicrobial resistance gene (ARG) signal in wastewater and the frequency of isolation of a particular resistant bacterium in medical laboratories, remains hardly evaluated [11–13].

Here, we used qPCR and dPCR to quantify the bla_CTX-M, bla_OXA-48, bla_NDM, bla_KPC and vanA genes in hospital wastewater. The study aimed to demonstrate the feasibility of AMR WWS in real hospital conditions and to describe the dynamics of endemic and emerging ARGs in wastewater from two hospital buildings housing patients with contrasting risks for AMR. Ultimately, the study highlights the value of AMR WWS, as well as some limitations and confounding factors that must be considered before it can be implemented in routine infection control procedures.

Methods

Hospital wastewater sampling

The University Hospital of Montpellier, in the south of France, is a 2600-bed tertiary care teaching hospital. Hospital wastewater sampling was carried out using automatic samplers supplied by the IAGE Company (Montpellier, France) in distinct buildings with independent wastewater collection pipes and tanks. The sampling and sub-samples storage were done from 12 to 15 °C along the sampling period. The GPS coordinates of the sampling sites in the Montpellier University Hospital were as follows: 43.630501792764484, 3.850851011755367 (SRR site) and 43.63036309242587, 3.864716021935508 (HRR site). Samples were conducted between 05–11-2022 and 06–02-2022 in SSR building (9 samples), and between 03–07-2022 and 05–23-2022 in HRR building (12 samples).

The sampling programmes were designed to obtain the theoretical volume of 1 L. In the SRR building, 24-h composite samples were collected in a wastewater retention tank at a sampling frequency of 20 s per hour. Sampling was continuous throughout the day with a break between 11.00 am and 01.00 pm. In the HRR building, 1 L of wastewater was collected weekly in the main building drain, where the wastewater does not stagnate due to rapid flow with each use of the toilet, sink or shower. The frequency and volume of sampling was optimised according to the flow in the collector during the hours of maximum use of the toilets and bathrooms by patients: 1 min of sampling every 2.5 min between 7.30 am and 11 am, then between 4.30 pm and 8 pm. Samples were transported in coolers, carefully homogenised and separated into two subsamples, one for qPCR and the other for dPCR.

Protocols for qPCR and dPCR quantifications

For the qPCR analyses, extraction was performed on the pellet obtained from 3 mL of wastewater (3 min centrifugation at 12,000 g). The MasterPure Gram-Positive DNA Purification Kit (Lucigen, LGC Biosearch Technologies) was used in accordance with the manufacturer’s instructions for DNA extraction in triplicate. This kit enabled the lysis of both gram-negative and gram-positive bacteria. Quantities and qualities of DNA extracts were measured using a NanoDrop spectrophotometer (ThermoFisher Scientific). DNA extracts were stored at  − 20 °C until use. qPCR reactions were performed using SYBRgreen chemistry on a LightCycler Nano (Roche) in eight-well strips for the quantification of six target genes: 16S rRNA gene [14], bla_CTX-M [15, 16], bla_OXA-48, bla_NDM, bla_KPC [17, 18] and vanA [19]. Each reaction contained SensiFAST SYBR No-ROX mix 1X (Bioline Meridian Bioscience), 0.4 µM of forward and reverse primers (Supplementary Table S1), 1 µL of tenfold diluted DNA (10 and 1000-fold diluted DNA for 16S rRNA gene amplification) and sterile water to a final volume of 10 µL. qPCR cycling conditions started with polymerase activation at 95 °C for 3 min, followed by 40 cycles composed by 10 s at 95 °C, 10 s at 60 °C (16S rRNA gene, bla_NDM, bla_OXA-48), at 62 °C (bla_CTX-M and bla_KPC) or at 58 °C (vanA), and 10 s at 72 °C. Finally, amplification products were gradually heated from 60 to 95 °C (0.1 °C/s) in order to obtain the melting temperatures and to check specificity for each sample. Standard curves were obtained by quantification of tenfold serial dilution of known quantities of linearized plasmids containing specific amplicons. DNAs extracted from characterized natural isolates with native plasmids carrying bla_OXA-48 (Klebsiella pneumoniae ARS619), bla_NDM (Escherichia coli B26), bla_KPC (E. coli ARS224), bla_CTX-M (K. pneumoniae 2M2E5) and vanA (Enterococcus faecium strain T301-1) genes were used as positive controls, and sterile water as negative control. For each sample, quantification was performed in technical duplicate. For inhibition assays, i) the same protocol was used to quantify the bla_NDM gene in a reaction containing a known amount of standard plasmid in addition to 1 µL of tenfold diluted DNA extracted from a wastewater sample known to be negative for this gene and ii) quantifications of 16S rRNA genes from 10 and 1000-fold diluted DNA extracts were compared. qPCR quantification data were only considered if the amplification had a specific Tm, if the quantification was above the limit of quantification (Supplementary Table S2 and if it was i) present in both technical replicates and ii) present in at least two of the three DNA extraction replicates. If this was not the case, the gene was considered to be detected but not quantifiable. Data were analysed using the LightCycler Nano Software suite (v1.0).

The dPCR experiments were performed by the company IAGE according to a patented method for pathogen detection in liquid matrix (n° WO2022144432A1). For dPCR, 30 mL of the homogenised wastewater were vortexed 3 times for 10 s each. The sample was cooled on ice for 15 s between each vortex burst. An Amicon Ultra-15 centrifugal filter unit (Merck Millipore, cut-off: 10 kDa) was hydrated with 2.5 mL of ultrapure water and spun at 3234 g for 10 min at 4 °C. Fifteen mL of the vortexed samples were transferred in the Amicon Ultra-15 Centrifugal Filter Unit and were then concentrated by centrifugation at 3234 g for 35 min at 4 °C (ultrafiltration). The extraction of total DNA was performed using the IndiMag Pathogen Kit (Indical Bioscience) on the total volume of the concentrated sample. DNA extracts were then stored at  − 20 °C until use. dPCR reactions were conducted using a 5-plex assay developed by IAGE on a QIAcuity Eight Plateform System, using the QIAcuity Probe PCR Kit and QIAcuity Nanoplates 26 K 24-wells (Qiagen, Germany). The same primers as for qPCR assays were used, supplemented with hydrolysis probe (Supplementary Table S1). The dPCR reaction mixtures were prepared in a standard PCR plate as follows: each reaction contained 10 µL of 4X Probe PCR Master Mix, 5 μL of the primers/probe mix (2.25 µM of both primers, 0.625 µM of probe), 4 µL of DNA extract and nuclease free water to a final volume of 40 µL. The reaction mixtures were transferred into a QIAcuity Nanoplate and the plate was loaded onto the QIAcuity Eight automated system. The workflow included (i) a priming and rolling step to generate and isolate the chamber partitions (26,000 partitions), (ii) an amplification step with the following cycling protocol: 95 °C for 2 min for enzyme activation, 95 °C for 5 s for denaturation, and 58 °C for 60 s for annealing/extension for 40 cycles; and (iii) imaging step by reading fluorescence emission after excitation of the probe at the appropriate wavelength. The gene bla_KPC was not quantified by dPCR but by qPCR only. Data were analyzed using the QIAcuity Software suite v3.1.0.0.

qPCR and dPCR results obtained in gene copy number per µL of DNA were both transformed in copy number mL⁻¹ of wastewater. Normalised genes abundances were calculated as the ratio between resistance genes and the 16S rRNA gene copy numbers. These data were plotted on a logarithmic scale (log₁₀).

To study the correlations between dPCR and qPCR data, and between absolute and normalised quantifications, Pearson’s or Spearman’s correlation coefficients (r or rs) were calculated depending on whether the data followed a normal distribution (Shapiro–Wilk test), and the nullity of these coefficients were tested. Box-plots were performed to study the distribution of the data of the different sites of sampling and analysis protocols. Comparison between sites and protocols were made using a Mann–Whitney test for those that did not follow a normal distribution (GraphPad Prism software 8.0.1). Rejection of the H₀ hypothesis was considered significant when the p value ≤ 0.05.

CPE cultures and species identification

Fifty µL of pure, tenfold and 100-fold diluted samples were plated on CHROMagar mSuperCARBA (Graso Biotech) medium using an easySpiral automatic plater (Interscience) in exponential and constant modes. After overnight incubation at 37 °C, colonies were counted and each morphotype (form size and colour) were selected for subculture on the same medium. Each selected strain was frozen at − 80 °C in trypticase soy broth (DIFCO) supplemented with 20% glycerol, which allowed the constitution of a library of 145 bacterial strains. Bacterial identification was performed on overnight cultures on trypticase soy agar using matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF MS) (Brucker). The threshold for species affiliation was 2.0 according to the in vitro diagnosis protocols of the supplier (Brucker). Fisher’s exact test was used to compare the proportion of CPE and carbapenemase producing bacteria between the two buildings (GraphPad Prism software 8.0.1). Simpson’s Diversity Index (D) was calculated for the two buildings using the formula where (S) is the number of species and (a) is the proportional abundance of the nth species. Mann–Whitney test was conducted to compare the diversity indices between the two buildings (GraphPad Prism software 8.0.1). Rejection of the H₀ hypothesis was considered significant when the p value ≤ 0.05.

End-point PCR of fimH, bla_NDM, bla_OXA-48, bla_KPC and bla_CTX-M genes

Endpoint PCR amplifications for resistance genes were performed with the primer indicated in Supplementary Table S1. The fimH gene was amplified according to Johnson et al. [20], (Supplementary Table S1). PCR amplifications were done on a Hybaid® thermocycler, starting with an initial denaturation for 2 min at 95 °C, followed by 35 cycles consisting in 30 s (45 s for fimH) at 95 °C, 30 s (45 s for fimH) at 60 °C (bla_NDM, bla_OXA-48), 62 °C (bla_CTX-M), 55 °C (bla_KPC), or 58 °C (fimH), 30 s (60 s for fimH) at 72 °C and a final extension for 7 min at 72 °C. Reactions contained GoTaq Flexi Buffer 1X and 0.625U of GoTaq DNA Polymerase (Promega), 1.75 mM of MgCl₂, 0.2 µM of each primer (Supplementary Table S1), 25 ng of DNA from wastewater (for fimH) or 1 µL of boiling/freezing crude extract from bacterial strains for the other genes. Sterile water was used as negative control. Positive controls were DNAs extracted from a laboratory collection of bacterial strains: E. coli T352 for fimH gene and K. pneumoniae 21DU, ARS619, Fonc147, 2FA2E1 for bla_NDM, bla_OXA-48, bla_KPC and bla_CTX-M genes, respectively. Endpoint PCR results were visualised on 1.5% agarose gel with ethidium bromide after electrophoresis with TriDye 1 kb DNA Ladder (New England Biolabs).

Clinical AMR data

The Infection Control Team (ICT) data used in this paper are Excel files that collate the species and resistotypes of all bacteria isolated from hospital patients. These data are generated by the medical microbiology laboratory and exported to the ICT surveillance file. Regional and national data about CPE and VRE were obtained from the reference national centre for antimicrobial resistance [10] and from the European Centre for Disease Prevention and Control [21].

Results

Hospital buildings and patients’ populations in relation to CPE and VRE

The first hospital building has been considered as a standard resistance risk building (SRR) because the incidence of multi-drug-resistant bacteria was 0.44/1000 hospitalisation days in 2022. The patients cared for daily in SRR for short-stays mainly for orthopaedic diseases (median length of stay about 4.5 days in 450 beds). During the sampling period of 4 weeks, approximately 7200 patients contributed to the wastewater of this building. Most of the SRR patients were not screened for resistant bacteria unless they had a known individual risk, such as a recent hospitalisation in a high-risk country or a history of contact with a patient carrying CPE or VRE.

On the opposite, the second building hosted patients with high risk of resistance (HRR) due to the general use of broad-spectrum antibiotics for proliferative blood disorders associated to immunosuppression (median length of stay about 10 days in 60 beds). The incidence of multi-drug-resistant bacteria in HRR was 9.39/1000 hospitalisation days in 2022. HRR patients were screened weekly for resistant bacteria. To ensure that most of the patient population changed between each sample, the wastewater from the HRR was analysed once a week for 11 weeks (1120 patients, from 07 March to 23 May 2022). No CPE or VRE outbreak was declared in SRR or HRR during the study period. The healthcare workers and visitors also contributed to SRR and HRR wastewater. The difference between the two buildings for patients AMR risks was confirmed by the difference in presence of patients who carried CPE or VRE during the study period (bars in Fig. 1).

Fig. 1.

Normalised quantifications of resistance genes in 21 wastewater samples from two hospital buildings. The genes bla_CTX-M, bla_OXA-48, bla_KPC, bla_NDM and vanA were quantified using qPCR in wastewater of A standard resistance risk (SRR) and B high resistance risk (HRR) buildings, and using dPCR in wastewater of C SRR and of D HRR buildings. Quantifications are the mean of DNA extraction triplicates. An upward-pointing triangle on the x-axis indicates that the gene was detected, but its quantity was below the quantification limit in qPCR. The dPCR quantification of was not performed. The number of patients detected as carriers of CPE and VRE during the sampling period are indicated by coloured bars in secondary y axis

Comparison of two protocols for ARG quantifications in hospital wastewater

The in-house qPCR protocol involved an enzymatic DNA extraction method on a raw pellet obtained by centrifugation of wastewater. The IAGE company carried out the dPCR experiments according to their patented protocol, which involved a bead-beating based extraction on a wastewater concentrate obtained by ultrafiltration.

The two sample concentration methods prior to DNA extraction (centrifugation vs. ultrafiltration) were compared on the same sample HRR-1403 (Supplementary Table S3). The performance of DNA enzymatic extraction from a raw pellet and concentrated wastewater was compared. Based on the 260/280 and 260/230 optical density ratios, the quality of the extracted DNA was similar, as well as was the quantity, for both methods (Supplementary Table S3). The sample concentration by centrifugation (raw pellet) was chosen for the in-house because it is simple and inexpensive, making it suitable for regular WWS monitoring. The company IAGE used ultrafiltration in its commercialised analyses because more suitable for high-throughput processes. This study did not compare the performance of enzymatic and bead-beating based extractions, but the cell lysis in the enzymatic extraction kit is probably better suited to extracting DNA from gram-positive bacteria such as Enterococcus spp.

As qPCR is described as being more susceptible to inhibition than dPCR, a comparative quantification of bla_NDM in a reaction containing a known amount of standard (bla_NDM gene on native plasmid) with or without diluted DNA extracted from wastewater negative for bla_NDM was performed. Similar concentration (440 ± 98.12 copies mL⁻¹ and 487 ± 13.49 copies mL⁻¹, respectively) was observed suggesting the absence of inhibition when using the tenfold diluted DNA samples. The absence of qPCR inhibition was also confirmed for each sample by the results of 10- and 1000-fold diluted DNA 16S rRNA gene quantification (Supplementary Fig. S1). Indeed, the number of copies of 16S rRNA gene normally decreases by approximately 100-fold between samples diluted tenfold and those diluted 1000-fold. Moreover, when the results are expressed in copies per millilitre, a decline in the number of copies is observed when using the 1000-fold dilution in comparison to tenfold (Supplementary Fig. S1).

In the absence of a gold standard, the performance of each protocol is compared with that of the other. The Fig. 1 shows the qPCR and dPCR normalised quantifications for the 21 wastewater samples. Most of the 21 samples from the 2 hospital buildings gave signals with both dPCR and qPCR protocols. When the two methods are compared, the percentage of gene detection varies according to the gene. Both methods are highly sensitive for the detection of the genes bla_OXA-48, bla_KPC and bla_CTX-M, with a positive detection rate of 90.5% to 100%. However, dPCR appeared less sensitive than qPCR for the detection of bla_NDM, with a positive detection rate of 57% in qPCR versus 38% in dPCR. However, four of the negative samples for bla_NDM in dPCR were detected but were unquantifiable in qPCR, suggesting non-specific signals in qPCR. Conversely, qPCR-positive vanA tests that were negative in dPCR gave quantifiable, specific signals. The qPCR protocol’s superior sensitivity for vanA detection (95% vs. 57% for dPCR) could be due to the use of a more efficient extraction method for gram-positive bacteria than that used in the dPCR protocol.

To enable a more global comparison of the qPCR and dPCR results, the distribution of normalised ARG concentrations was compared between the two protocols. Figure 2 shows that there were no significant differences, despite the larger distribution of dPCR quantification for the vanA gene.

Fig. 2.

Distribution of concentrations of relative quantification of ARG in both clinical settings’ wastewater. The results of qPCR and dPCR on environmental DNA are expressed on a logarithmic scale. Comparison of quantification between quantification methods was performed by the Mann–Whitney test, and p values are presented above the box plots (ns: non-significant). Median and IQR are presented on each boxplot. The dPCR quantification of bla_KPC was not performed

Morever, the results of Spearman’s correlation tests showed that there was a significant positive correlation between dPCR and qPCR normalised quantifications for bla_CTX-M, bla_OXA-48, bla_NDM and vanA genes (rs = 0.5 to 0.8, p values ≤ 0.01 to ≤ 0.001). In conclusion, both methods are suitable for use in hospital settings for WWS, with no significant discrepancies.

Finally, the normalised abundances of ARG (ratio on 16S rRNA copies mL⁻¹) were positively correlated with the absolute abundances of ARG by both qPCR and dPCR (rs_qPCR = 0.5 to 0.9, p value ≤ 0.01to ≤ 0.0001; rs_dPCR = 0.33 to 0.91, p value ≤ 0.01 to ≤ 0.0001, except for bla_OXA-48 (p value = 0.2)) (data not shown).

Concentration of resistance genes in hospital wastewater compared to patients carrying these genes and AMR epidemiology

Table 1 presents nationwide epidemiological data on CPE and VRE, alongside their associated resistance genes. The low percentage of resistant isolates (< or = 1%) confirms the emerging nature of CPE and VRE in France.

Table 1.

Epidemiological data on the resistance markers targeted in this study (France, 2022)

	CPE			ESBL-E	VRE		Data sources
Percent of resistant isolates among total tested isolates, nationwide	0.1–1*%			8–28*%	0.7%		[21]
Percent of gene types, nationwide (percent of gene types, in the hospital)	blaOXA-48 60,6% (74%)	blaKPC 2,1% (0%)	blaNDM 25,7% (26%)	blaCTX-M-15 74–92*% (na)	vanA 82,6% (100%)	vanB 14,8% (0%)	[10]; ICT unpublished data

Values in parentheses corresponds to the hospital data provided by ICT for this study. na for not available

*Depending on the species. CPE for Carbapenemase-Producing Enterobacteriales; ESBL-E for Extended-Spectrum Beta-lactamase-producing enterobacteria, VRE for Vancomycine Resistant Enterocci

By contrast, ESBL-E are multi-drug-resistant bacteria representing 8 to 25% of the enterobacterial isolates, depending on the species. This high prevalence rate suggests endemic diffusion in France. In this study, the beta-lactamase gene the most frequently associated with ESBL-E, bla_CTX-M-15, has been used as an indicator of the endemic diffusion of antimicrobial resistance.

As expected, the ESBL marker bla_CTX-M was detected in all wastewater samples in both SRR and HRR buildings (Fig. 1). The overall kinetics of bla_CTX-M and bla_OXA-48 were very similar in SRR whereas bla_OXA-48 is quantified at a higher level than bla_CTX-M in HRR (Fig. 1). As bla_OXA-48 is still emerging in France and in the hospital, the constant detection of bla_OXA-48 in both SRR and HRR, sometimes at higher levels than bla_CTX-M was unexpected. Only 36 patients carrying OXA-48 CPE were identified in the entire hospital in 2022 (ICT surveillance data). During the study period, only two outpatients carrying OXA-48 K. pneumoniae stayed in HRR for several very short stays from 22 February to 15 April 2022 (Fig. 1B, D). The dynamics of bla_OXA-48 in the HRR effluent could not be fully explained by the presence of these two outpatients, as the signal remained at the same level after their discharge from 15 April until the end of the study (23 May). The persistent and high load of bla_KPC in both SRR and HRR (Fig. 1) is even more surprising given the low rate of KPC CPE in France and the absence of KPC CPE detected in our hospital in 2022 (Table 1). The bla_NDM gene was detected less frequently. It was quantified or detected below the limit of quantification in 14 of the 21 samples. In SRR, bla_NDM was quantified at low level in only two samples (Fig. 1A, C) whereas in HRR we observed a sudden increase between 22 March and 18 April (Fig. 1B, D). After this peak, the bla_NDM signal rapidly became negative or at the limit of detection until the end of the study. These dynamics reflect the emerging status of the NDM-carbapenemase in patient epidemiology in France and in the hospital (Table 1). The sudden increase in HRR could suggested that a patient or a group of patients carrying NDM bacteria were cared for in HRR wards. However, no patient carrying NDM CPE was detected by individual biological diagnosis during this period (ICT surveillance data).

The dynamics of vanA gene during the study is shown in Fig. 1. In qPCR, it was detected in all but one sample (24 April in HRR) in both buildings. The vanA gene load is lower (1 to 3 log) than that of bla_CTX-M, considered as the standard for endemic gene, in SSR (Fig. 1A, C). This was also the case in HRR, but for this ward vanA exceeded bla_CTX-M in few samples (21 March, 28 March and 16 May) (Fig. 1B, D) while vanA resistance was still considered emerging in France and no outbreak was detected in HRR. During the study period, a maximum of 2 HRR patients per day were known as carriers of E. faecium vanA (Fig. 1B, D). From 7 March to 11 April, the effluent signal appeared to be partially linked to the number of known vanA carriers. After 12 April, similar vanA signals were observed, despite only one outpatient being admitted twice for one day in the ward (ICT surveillance data). In conclusion, vanA levels in hospital wastewater were partially correlated with the presence of individual patients who carried the gene. The vanA dynamics in wastewater of the two buildings suggested a low-level of endemicity for vanA resistance in hospital patients.

Does resistance gene monitoring in hospital wastewater differenciate patients populations?

The two hospital buildings examined in this study serve different patient populations in terms of their risk of carrying CPE and VRE: high risk for HRR patients and standard risk for SRR patients. The difference in normalised quantification of bla_CTX-M was highly significant between SRR and HRR, but unexpectedly bla_CTX-M was more prevalent in SRR (Fig. 3A). One explanatory hypothesis is antimicrobial stewardship in HRR ward, where any sign of systemic infection leads to the immediate prescription of broad-spectrum antibiotics that cover at least ESBL-E. This could limit the load of bla_CTX-M in patient microbiota and in HRR building effluent. The loads of carbapenemase genes, bla_OXA-48 and bla_KPC, were significantly lower in SRR wastewater than in HRR (Fig. 3A, B). This was also the case for bla_NDM, even if the high rate of negative signals prevented statistical analysis (Fig. 1). While these results are preliminary, they suggest a potential link between AMR in patients and WWS. In fact, they suggest that the concentration of bla_CTX-M and carbapenemase genes in the effluents from the HRR and SRR could reflect the respective patient populations of these hospital buildings. However, the load of vanA was not significantly different between SRR than HRR (Fig. 3A, B).

Fig. 3.

Distribution of the normalised loads of bla_CTX-M, bla_OXA-48, bla_KPC and vanA in wastewaters. (HRR), high risk resistance building and (SRR), standard risk resistance building. A quantification using qPCR; B quantification using dPCR (B). The results are expressed in an inverse logarithmic scale. Comparison of ARGs quantification between hospital buildings was performed by the Mann–Whitney test, and p values are presented above the box plots (ns: non-significant; *: 0.01 < p-value ≤ 0.05; ***: 0.0001 < p value ≤ 0.001; ****: p value ≤ 0.0001). Median and IQR are presented on each scatter plot. The dPCR quantification of bla_KPC was not performed

Does the resistant culturable gammaproteobacterial community in wastewater reflected gene load in wastewater and local epidemiology ?

The numbers of gamma-proteobacteria growing onto media selective for carbapenem resistance are showed in Fig. 4. The load of gamma-proteobacteria on CHROMagar mSuperCARBA in the effluent varied from 10⁴ to 2.10⁸ CFU mL⁻¹ in SRR and from 3.10⁵ to 5.10⁸ CFU mL⁻¹ in HRR. The results of the culture tests should be treated with caution because most culture media are not validated for use with wastewater. Nevertheless, the higher concentration of carbapenem-resistant bacteria in the HRR follows the trends observed for ARG load in the two buildings. The dynamics of the resistant bacteria showed peaks that were not related to the sum of the normalised loadings of the three carbapenemase genes (Fig. 4), suggesting that the mechanism conferring resistance to carbapenems at the concentrations present in the CHROMagar™ mSuperCARBA medium does not involve carbapenemase genes detected in this study.

Fig. 4.

Load of carbapenemase resistant bacteria versus summation of normalised loads of carbapenemase genes. A standard resistance risk (SRR) building; B high resistance risk (HRR) building. Bacteria are grown onto mSuperCARBA™ medium. Carbapenemase genes are quantified by dPCR and qPCR

More than half of the selected bacteria (76/145 = 52%) resisted to carbapenems by carrying the carbapenemase encoding genes bla_OXA-48, bla_KPC or bla_NDM. Fourteen different species belonging to 7 genera of carbapenem-producing gamma-proteobacteria were identified including 4 genera of Enterobacterales: Citrobacter, Enterobacter, Klebsiella and Kluyvera (Fig. 5) but also members of the genera Aeromonas, Pseudomonas and Shewanella.

Fig. 5.

Identification and resistotype of carbapenemase-producing species. Bacteria isolated from SRR (A, n_tot = 81) and HRR (B, n_tot = 64) wastewaters onto mSuperCARBA™ medium

The most common carbapenem resistotypes in Enterobacterales were OXA-48-producing Citrobacter freundii (n = 20 in HRR and n = 1 in SRR), KPC-producing C. freundii (n = 10 in each building) and KPC-producing Citrobacter braakii (n = 9 in HRR and n = 2 in SRR). Six strains carried both bla_OXA-48 and bla_KPC gene (4 C. freundii, 1 C. braakii and 1 Klebsiella oxytoca), all of them were isolated from HRR wastewater (Fig. 5). Finally, 49 of the 53 Citrobacter spp. strains isolated in this study encoded bla_OXA-48, bla_KPC or bla_NDM. The rate of viable/cultivable OXA-48-, NDM- and KPC-CPE among strains isolated on selective medium was significantly higher in HRR (65%) than in SRR (41%) (Fisher’s exact test, p value = 0.0042). Furthermore, the majority of bacteria growing on the carbapenem resistance selective medium resisted to carbapenem via OXA-48 and KPC-carbapenemase production in HRR, but not in SRR where KPC CPE largely dominated (Fig. 5).

The pathogens E. coli and Pseudomonas aeruginosa were not isolated in this study. Concerning E. coli, the specific fimH gene was searched in wastewater and was detected in every wastewater sample (data not shown). This suggests that although E. coli DNA is present in wastewater, carbapenem-resistant E. coli, if present, are no longer viable or cultivable in the samples tested.

The ICT data from the routine surveillance of CPE in hospital showed that the most commonly isolated CPE from patients in the Montpellier hospital, all buildings combined, in the studied period, were K. pneumoniae, E. coli, Enterobacter cloacae and C. freundii (Fig. 6). This clearly differed from wastewater, where Citrobacter dominated and E. coli was not detected. Figure 6 also showed that, with the exception of bla_OXA-48, which is common in both human and wastewater, the proportion of bacterial species resistant to carbapenems by production of NDM and KPC differed between human and wastewater. In conclusion, the taxonomic diversity and resistotypes of CPE isolated from patients in the routine practice of medical microbiology did not match those of resistant bacteria circulating in wastewater and isolated on the CHROMagar™ mSuperCARBA medium.

Fig. 6.

Distribution of resistotypes and taxa of carbapenemase-producing enterobacteria in patients and hospital wastewater

Discussion

A recent systematic review [22] showed that most published studies on ARGs in wastewater have focused on the risk of spread from hospitals or other AMR hotspots into the environment, and the subsequent risk to public health [23]. Few papers, however, have addressed the WWS of AMR. While previous publications have presented ARGs in wastewater as potential AMR indicators for surveillance purposes, the concept of WWS applied to AMR surveillance has only recently been highlighted in publications such as an editorial in Science [24] and a commentary by Larsson et al. in Nature Reviews Microbiology [25]. The scientific basis for WWS of AMR was a comprehensive set of experiments based on metagenomics. These experiments demonstrated the reliability of WWS of AMR at national [26] and global [27] levels. Furthermore, correlations with socio-economic, health, and environmental factors have been demonstrated [28]. Approaches targeting resistant bacteria or ARGs of interest have also been developed, showing the correlation between AMR in wastewater and AMR epidemiology [29–31]. If global surveillance is needed to inform national and international agencies, local or territorial surveillance, for example at the city level [32] or within hospitals [12, 33], could provide insights into the extent of antibiotic use, the local mode of bacterial resistance spread (endemic or epidemic), and thus guide effective antibiotic stewardship and infection control interventions.

A recent study established a practical approach for WWS as a potential tool for public health monitoring of AMR burden in healthcare facilities [5]. Our present study proposes to monitor emerging resistance genes in wastewater of a hospital under real-world conditions. Specifically, the study design does not involve any changes in the structure of the wastewater network or in patient monitoring. Thus, feasibility and barriers have been assessed in a real hospital context. We examined wastewater from two buildings displaying different effluent systems: a retention tank (SRR building) and a drainpipe directly connected to the patients’ sanitary system (HRR building). To limit biases linked to these structural differences, we collected pooled samples, rather than grab samples, because pooling has been shown to be the optimal method for obtaining the highest bacterial diversity in hospital drains [34]. The tank, located between the proximal SRR outfalls and the municipal collector, probably increased the homogeneity of the effluent, thereby reducing the number of sub-samples required. In HRR, the flow in the outfall pipe was intermittent, depending on the use of the sanitation systems. Given the small contributing population and the short distance between the toilet and the sampling point in HRR, it was determined that very small and frequent subsamples would be required, in line with previous studies [34, 35] while other studies underlined the advantage of passive sampling [5].

Previous studies have shown that the enrichment of resistant bacteria and the reduction of E. coli due to the biocidal selection of hospital wastewater have been observed in vitro at 20 °C [34]. The unexpected absence of culturable carbapenem-resistant E. coli despite the presence of fimH, an E. coli-specific gene, in each sample suggests a potential biocidal effect. It is noteworthy that E. coli is the most common organism among CPE isolates from patients in France and in the Montpellier hospital. A publication describing WWS in a Swedish hospital [12] reported a high prevalence of carbapenemase-producing E. coli. This suggests the possibility of a depletion of carbapenemase-producing E. coli in the Montpellier hospital effluent, as evidenced by the findings of Urase et al. [13] who detected only six E. coli among 247 carbapenem-resistant bacteria isolated from wastewater in Japan. It is also the case in Poretsky et al. where only one strain of carbapenemase-producing E. coli was detected in the wastewater of a clinic in Chicago, USA, while NDM-producing E. coli was the more frequently species and resistotype isolated in patients [5].

Comparing the load of AMR genes in wastewater with data from clinical AMR surveillance in hospitals in France produced expected and unexpected results, raising several questions. The inconstant detection of bla_NDM is consistent with the clinical data on NDM-CPE, which is emerging in France and Europe. For the vanA gene that is still emerging in France, WWS suggested rather a low endemicity in the hospital. It is noteworthy that after the study, in 2023, a large vanA outbreak occurred in several wards of the hospital (ICT, unpublished), suggesting that WWS could be predictive of a risk of local outbreak. The bla_OXA-48 gene is consistently detected, sometimes at a higher level than the bla_CTX-M gene, which is used as a reference for endemic resistance. This suggests a bias due to a resistant microbial community that inhabited pipe network or calls into question the current emerging status assigned to OXA-48 CPE in our hospital and in France. Current surveillance, which is only targeted at patients who are supposed to be at high risk of carriage, could overlook the rate of OXA-48 CPE in the general population. The dominance of bla_KPC in hospital effluents was the most surprising finding of this study, given that KPC-producing bacteria are in the minority in France [10]. Similarly, the wastewater-resistant bacterial community was clearly dominated by carbapenemase-producing Citrobacter spp., which were sporadically detected in patients in 2022 in France. At our hospital, the bla_KPC gene and carbapenemase-producing Citrobacter were almost undetectable in patients during the study period (only one patient carried a VIM-producing C. freundii). Similar results were showed in a recent study conducted in the United States finding the same dominant species in hospital wastewater as those found in this study, namely KPC-producing Aeromonas and Citrobacter, even though these bacteria are rarely found in patients, either as carriers or in infections [5]. These discrepancies could be due to a specific community of resistant bacteria residing not only in the wastewater network, but also in siphons, sinks and toilets [36]. These plumbing devices generally harbour bacterial communities in biofilms that favour the persistence of resistant bacteria and the exchange of genes via horizontal transfer between bacterial species [36]. Furthermore, plumbing components could be involved in bacterial transmission to patients and in hospital outbreaks [36–38].

The high prevalence of carbapenemase-producing C. freundii and C. braakii in wastewater is concerning, given that these bacteria are increasingly being reported in healthcare-associated infections [39, 40]. Moreover, evidence that Citrobacter species are emerging carriers of carbapenem-resistance genes has recently been published [41]. The role of hospital plumbing devices as reservoirs of carbapenemase-producing C. freundii in persistent outbreaks of CPE was demonstrated [40]. Since our study, the regional epidemiology changed with more and more carbapenemase-producing Citrobacter described in hospital outbreak (ICT surveillance data 2023–2024). A recent study in Germany showed that C. freundii is the most common species producing carbapenemases in humans at the national level [42].

The abundance of carbapenemase-producing Citrobacter spp., Aeromonas spp. and Pseudomonas spp. in the microbial community of effluent network suggested a possible exchange of emerging carbapenemase genes between clinically relevant and environmental bacteria. Indeed, Citrobacter spp., Aeromonas spp. and Pseudomonas spp. can also colonise the human microbiota and act as a shuttle between humans and the environment. Aeromonas spp. were recently described for the first time as carriers of bla_OXA-48 and bla_KPC [43]. This led to the hypothesis of an exchange of ARG from human bacteria to autochthonous wastewater bacterial communities, with some species and resistotypes finding favourable conditions in the wastewater network. Therefore, wastewater could be used as an indicator of the further emergence of AMR and predict human risk, even if the signals are not related to real-time clinical epidemiology.

The primary and ultimate function of WWS is to determine the prevalence of an infectious disease in a human population without testing every individual. We tested this in real hospital conditions by comparing the levels of carbapenemase genes in wastewater from two patient populations characterised by different levels of risk of carrying or being infected with CPE or VRE. Significant differences were observed between the HRR and SRR in the load of AMR genes and the rate of cultivable carbapenem-resistant bacteria. It should be noted that the contribution of healthcare workers and visitors to the carbapenemase signal in wastewater cannot be evaluated. However, it is unlikely that these two groups carried CPE or VRE at a level that would affect the corresponding gene signals in wastewater. For the majority of genes and resistotypes, a higher prevalence was observed in the HRR, corresponding to the increased risk of CPE and VRE carriage in HRR patients. Majlander et al. [33] have previously observed this between two hospitals in Finland. This suggests that, despite the lack of strict correspondence between the AMR signal in wastewater and individual patient cases on any given day, quantifying carbapenemase genes in wastewater could be used to detect differences between contrasting patient populations. Additional studies are needed to determine if, at the building scale and in an infection control purpose, the WWS of a targeted gene during nosocomial outbreaks could be a complementary tool to patients tracing. In this context, WWS of AMR could be a valuable and efficient tool for local infection control.

Conclusion and perspectives

Thanks to adjustments to sampling programmes based on patient flow and network structure, we have demonstrated that assessing emerging AMR in wastewater can be carried out in a real hospital environment. Although WWS identified two distinct patient groups within our hospital, the WWS signals were only partially consistent with the patients’ clinical data. This suggests that interpreting WWS for AMR will be more challenging than interpreting it for viruses. This is probably because bacteria can colonise wastewater networks and plumbing, multiply, and exchange genes.

The WWS of AMR in hospitals remains a rarely studied topic. The results of the available studies, including this one, have informed future works through the identification of three main objectives: i) establishing and assessing reproducible and standardised protocols; (ii) fundamental studies using metagenomics and epicPCR to investigate the molecular links between the WWS signal and clinical cases; (iii) for an applied infection control (IC) purpose, setting up continuous and prospective WWS surveillance to test its capacity to detect nosocomial outbreaks or changes in resistant bacteria within a hospital. This is necessary for modelling ARG dynamics, setting threshold alerts, linking WWS signals to infection control policies, and evaluating WWS in predictive epidemiology.

Abbreviations

AMR

Antimicrobial resistance

ARG

Antimicrobial resistance gene

CPE

Carbapenemase-producing Enterobacterales

ESBL

Extended-spectrum beta-lactamase

ESBL-E

Extended-spectrum beta-lactamase-producing enterobacteria

HRR

High resistance risk

ICT

Infection control teams

MALDI-TOF MS

Matrix-assisted laser desorption ionization time-of-flight mass spectrometry

SRR

Standard resistance risk

VRE

Vancomycin-resistant enterococci

WBE

Wastewater-based epidemiology

WHO

World Health Organization

WWS

Wastewater surveillance

Author contributions

CF: Conceptualization, resources, formal analysis, writing—original draft, writing—review and editing. MT: Conceptualization, resources, methodology, formal analysis, writing—original draft, writing—review and editing. IZ: Methodology. OC: Conceptualization, formal analysis. FD: conceptualisation. PLF: Validation, writing—review and editing, funding acquisition. EJB: Conceptualization, validation, writing—original draft, writing—review and editing, supervision, project administration, funding acquisition.

Funding

The study was funded by the PHySE-HSM project (university of Montpellier), the Mi2H platform and the Tremplin program of the Montpellier Academic Hospital.

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.