Environmental Consequences of Anthropogenic Pollution: Non-antibiotic-Drug-Driven Antibiotic Resistance Selection in a Model Aquatic Ecosystem

Abstract

Non-antibiotic drugs (NADs) used in human therapy may induce antibiotic resistance selection and dissemination in vitro. However, the potential risks of antibiotic resistance emergence associated with environmental NAD pollution have not been addressed. Here, we conducted a multidisciplinary study on river water microcosms using growth kinetics, qPCR, metagenomics, 16S rRNA sequencing, and liquid chromatography–tandem mass spectrometry (LC-MS/MS) to determine whether NADs alter river bacterial ecology and select for antibiotic resistance genes (ARGs). Four NADs with different mechanisms of action were included at a high (mg/L) and low (μg/L) dose to establish dose–response relationships: chlorpromazine (antipsychotic), diclofenac (anti-inflammatory), diphenhydramine (antihistamine), and fluoxetine (antidepressant). Although the community response to NAD pollution was compound-specific and dose-dependent, all NADs and doses were stable in the environment, altered the composition and activity of bacterial communities, and selected for several ARGs, mostly β-lactamases and aminoglycoside resistance genes, some of which were associated with horizontal gene transfer genes. Pseudomonas (including some ARG-harboring subpopulations) was identified as a key player in the response to NAD pollution. Here, we demonstrate NAD-driven antibiotic resistance selection in complex aquatic communities, raising concerns about the collateral effects on human and environmental health due to the extensive anthropocentric use of NADs.

1. Introduction

Antibiotic resistance is a leading cause of mortality worldwide, contributing to 9% of global deaths. Due to the insufficient commercialization rate of new antibiotics, therapies combining antibiotics with non-antibiotic drugs (NADs) have been used as an alternative to fight the emergence of antibiotic resistance. This repositioning of several non-antibiotic drugs that show a synergistic effect with antibiotics might address infections caused by high-priority human pathogens. − However, in addition to (and maybe as a consequence of) their antibacterial effects on the gut microbiome, various NADs at clinical doses have also been associated with the development of antibiotic resistance in human-associated bacteria. , Several intra- and intergenera conjugation studies show that multiple NADs, including β-blockers, antiepileptics, antidepressants, and nonsteroidal anti-inflammatory drugs (NSAIDs) − increase the transfer of plasmids containing antibiotic resistance genes (ARGs). These studies pointed to an increased reactive oxygen species (ROS) production, the activation of the SOS response and efflux-pump expression, and increases in cell permeability, pilus generation, and cell-to-cell contact. Several NSAIDs, a lipid-lowering drug, and a β-blocker have also been associated with increased bacterial transformation. Thus, the use of drugs designed to target eukaryotic cells for their original application and as adjuvants for antibiotic therapy could have collateral effects on the human microbiome, increasing both ARG dissemination through horizontal gene transfer (HGT) and phenotypic resistance to antibiotics. These NADs might also affect the environment’s microbial communities.

The environment is a major source of antibiotic resistance that requires global attention in order to tackle the antibiotic resistance crisis. The aquatic environment is a particular concern because it can transport pollutants and bacteria over long distances, facilitating the spread of antibiotic resistance to the human microbiome via drinking water, recreational bathing, and crop irrigation. Thus, anthropogenic pollutants reaching surface waters at residual concentrations may pose a risk to human health. NADs are constantly introduced in aquatic ecosystems through sewage and industrial waste and are found at relatively low concentrations in water bodies worldwide. In vitro evidence has shown that environmentally relevant doses of multiple NADs may promote the development of antibiotic resistance through the increase of mutation rates and the dissemination of ARGs to single strains and complex bacterial communities through ROS-mediated plasmid conjugation (similar to what was observed at clinical doses). Environmental doses of NADs might also contribute to the development and spread of antibiotic resistance in the environment. However, there is a lack of studies addressing the impact of residual doses of NADs on antibiotic resistance selection and dissemination in aquatic ecosystems. To determine the risk to human health associated with NAD aquatic pollution, the potential effect of low doses of NADs on antibiotic resistance in aquatic ecosystems and the dose–response relationships and interactions between NADs and bacteria in complex environmental communities need to be determined.

The objective of this study was to assess this knowledge gap by determining the impact of NADs on antibiotic resistance selection using river water microcosms as a model aquatic ecosystem. To this end, four NADs belonging to different classes were selected to account for the diversity of mechanisms of action, chemical structures, and potential interactions with the aquatic ecosystem (both inert and living particles) encompassed by NADs. Diclofenac, a NSAID that inhibits inflammation through the inhibition of prostaglandin production, and fluoxetine, an antidepressant that blocks the reuptake of serotonin in the central nervous system (CNS), were selected because of their frequent detection in water bodies, their documented effects on antibiotic resistance dissemination at high doses in vitro and their identification as high-risk aquatic pollutants that deserve urgent attention. On the other hand, two drugs with contradictory or nonreported effects on antibiotic resistance dissemination were selected to provide new insights into the complexity of NAD-induced selective processes. Chlorpromazine, an antipsychotic that blocks dopamine, serotonin, and histamine receptors in the CNS, reduced plasmid conjugation in an in vitro study. Diphenhydramine, an antihistamine that acts as a competitive antagonist of histamine in the receptors of the CNS and is found in multiple surface water streams, has not been included in published research addressing antibiotic resistance selection mediated by NADs. Two different doses were used for each NAD, and their influence on antibiotic resistance selection and the response of the bacterial community were determined over 8 days using a combination of growth kinetics, qPCR, metagenomics, and 16S rRNA gene sequencing. In addition, NAD concentrations were determined in all samples by liquid chromatography–tandem mass spectrometry (LC-MS/MS) to improve our understanding of NAD fate and behavior in water bodies and to establish dose–response relationships for each NAD.

2. Materials and Methods

2.1. River Water Microcosm Setup and DNA/RNA Coextraction

Surface river water was sampled from the Rhône river in Lyon (45°45′08.1″N 4°50′11.4″E) on November 20th, 2023. Microcosms with 450 mL of water were polluted with four different NADs: the antipsychotic chlorpromazine (CMZ), the NSAID diclofenac (DFC), the antihistamine diphenhydramine (DPH), and the antidepressant fluoxetine (FXT). NAD analytical standards (Sigma-Aldrich) were added at two different doses: a low dose on the μg/L scale, aiming to reflect an environmental pollution scenario with NAD residues, and a high dose on the mg/L scale, aiming to reflect a clinical scenario. Doses added for each compound are detailed in SI Table S1. Microcosms were incubated at room temperature for 8 days in a vertical rotator in a room with some exposure to daylight. Triplicates were made for each of the 8 conditions (4 different NADs, low and high doses), and three nonpolluted controls were incubated in parallel. Each condition was named after the added NAD (CMZ, DFC, DPH, or FXT) and dose (“l” for low dose and “h” for high dose). After 0, 2, and 8 exposure days, 100 mL of river water was filtered from each sample and DNA/RNA were coextracted from the filters using the phenol/chloroform method described by Griffiths et al. Nucleic acids were eluted in 60 μL of nuclease-free water: 40 μL was conserved at −40 °C to analyze DNA, and 20 μL were treated with RQ1 RNase-Free DNase (Promega) and conserved at −80 °C to analyze RNA.

2.2. Quantification of NAD Concentration in Water Samples

50 mL of water from each sample and sampling time was conserved at −40 °C until the determination of NAD concentration using LC-MS/MS. The analysis was performed using an LC-MS/MS system, which consists of an ExionLC system (AB Sciex) with a QTOF X500R equipped with an electrospray ionization (ESI) Turbo V ion spray (AB Sciex). Details about the followed protocol are described in SI Description S1.

2.3. Metagenomic Sequencing and Metagenome-Based Community, Resistome and Mobilome Analyses

Metagenomes were sequenced from DNA samples extracted from river water. Metagenomic libraries were prepared from ≤1 ng of DNA using the Nextera XT Library Prep Kit and Indexes (Illumina) as detailed in Illumina’s “Nextera XT DNA Library Prep Kit’’ reference guide. DNA sequencing was performed using the NextSeq 1000 System and the NextSeq 1000/2000 P2 XLEAP-SBS Reagent Kit 600 cycles (Illumina). Sequencing depths after quality control are shown in SI Table S2. An assembly-based approach was used to reconstruct metagenome-assembled genomes (MAGs). Antibiotic resistance genes (ARGs), genes related to mobile genetic elements (MGE), and virulence genes were annotated in these MAGs, and the abundance and co-occurrence of ARGs and MGE in river water microcosms were determined using a contig-based analysis. Four independent assemblies were run in parallel: each assembly contained the samples polluted by one NAD and nonpolluted controls. The number of contigs, N50, and the number of MAGs obtained from each assembly are shown in SI Table S3. A detailed description of the protocol used for sequence treatment and analysis can be found in SI Description S2.

2.4. 16S rRNA Sequencing and Identification of Active Bacteria

The V4 hypervariable region of the 16S rRNA gene was amplified from retrotranscribed cDNA. Retrotranscription, cDNA amplification, sequencing, and sequence treatment were done as described by Joannard and Sanchez-Cid. Sequencing depths obtained after sequence treatment are shown in SI Table S2. All samples reached the plateau of amplicon sequence variant (ASV) richness discovery, as shown in SI Figure S1. Total inferred ASV abundances were obtained by dividing the ASV count by sequencing depth and multiplying by the number of 16S rRNA copies per L of water (quantified by qPCR following the protocol described in SI Description S3).

2.5. Statistical Analyses

Statistical tests were used on GraphPad Prism 9 to determine the impact of NADs on bacterial growth in vitro, overall bacterial abundance and activity in microcosms, and the abundance of MAGs, ARGs, and MGE-related genes. Normality was tested by using the Shapiro-Wilk test. When variables showed a normal distribution, two-way ANOVA tests followed by Tukey-Kramer post-hoc comparisons were used to determine statistical differences between controls and polluted samples at each sampling time. If at least one of the conditions did not follow a normal distribution, Kruskal–Wallis tests followed by Dunn’s post-hoc comparisons were used. p-Values were corrected to account for multiple comparisons (α < 0.05). Only pairwise comparisons with a p-value <0.05 are shown.

In addition, the influence of NADs on bacterial community composition and activity at an overall scale was determined by performing nonmetric multidimensional scaling (NMDS) analyses. These analyses were done on the Bray–Curtis dissimilarity calculated from the relative abundance of the MAGs and ASVs. NMDS analyses were done using the “vegan” package in R. PERMANOVA tests with 999 permutations were applied to each NMDS to determine the significance of NAD and dose influence on bacterial community composition using the “adonis” package in R and establish relationships between NAD dose and the response observed in the community. Samples at day 0 were excluded from the analyses, since the time-lapse to observe a response after NAD addition was too short.

Finally, statistical differences in total inferred ASV abundance between polluted samples and nonpolluted controls at days 2 and 8 were determined using the DESeq2 package in R. Log2FoldChange values were adjusted using the Approximate Posterior Estimation for generalized linear model in the “apeglm” package in R. Differences were defined based on a p-value <0.001 and a log2FoldChange ≥5. In addition, Spearman correlations were established using GraphPad Prism 9 between NAD doses and MAG, ARG, MGE, and 16S rRNA gene ASV relative abundance to determine links between NAD pollution levels and changes in bacterial communities, the resistome and the mobilome. One correlation analysis was done per coassembly (i.e., individual correlation analyses were done for each subset of samples polluted by a single NAD and the corresponding controls). Two variables were considered to be correlated at R ± 0.75 and p < 0.001.

3. Results and Discussion

3.1. Fate of Non-antibiotic Drugs in River Water Microcosms

The concentrations of the added pollutants were measured in river water microcosms using LC-MS/MS to determine the fate of these NADs in aquatic ecosystems over 8 days (Figure ). Slight increases in NAD concentration were observed from day 0 to day 2 in chlorpromazine at high doses (Figure A), diclofenac at high and low doses (Figure B), and diphenhydramine and fluoxetine at high doses (Figure C,D). Considering the relatively high hydrophobicity of these compounds (log K _ow CMZ = 5.41, log K _ow DFC = 4.51, log K _ow DPH = 3.27, log K _ow FXT = 4.05), these increases could be related to an initial adsorption of a small fraction of the added pollutants onto river water particles and their subsequent desorption over time. The high sorption of these four NADs has been widely reported. − Measured diclofenac doses were slightly higher than those of the other three NADs. Regarding background pollution levels, chlorpromazine and fluoxetine were not detected in nonpolluted controls, whereas some background residual pollution was measured for diclofenac and diphenhydramine. Diclofenac background pollution was no longer detected on day 2, whereas diphenhydramine was the only NAD still detected in controls on days 2 and 8. These initial background levels are consistent with previous findings; diclofenac and diphenhydramine have been found at much higher doses than fluoxetine in several urban streams at concentrations similar to those detected here. On the other hand, chlorpromazine is rarely measured in wastewater and river water. Overall, all added compounds and doses showed high stability in the microcosms, and final concentrations were similar to initial concentrations for all four NADs. Previous studies have found a high stability (no degradation over 3 years) of fluoxetine and diphenhydramine in outdoor soil-biosolid mesocosms, whereas diclofenac and chlorpromazine are mainly photodegraded in water, showing no chemical or biological degradation. ,, The high stability of these compounds in aquatic ecosystems, as well as their high sorption capacity onto particles present in river waters, may impose a long-term pressure on aquatic bacteria that might lead to the development of adaptative mechanisms with detrimental consequences for animal (including human) and environmental health.

1.

Concentrations in μg/L of (A): Chlorpromazine (CMZ), (B): diclofenac (DFC), (C): Diphenhydramine (DPH), and (D): Fluoxetine (FXT) measured in polluted river water microcosms and nonpolluted controls at days 0, 2, and 8 at low (l) and high (h) doses. CMZ and FXT limit of detection (LOD): 0.015 μg/L; limit of quantification (LOQ): 0.05 μg/L. DFC and DPH LOD: 0.0015 μg/L, LOQ: 0.005 μg/L. Values under the LOD were represented as 1/2 LOD for visualization purposes on the log scale. LOD lines are shown to facilitate visual interpretation of the data: a discontinuous line indicates the LOD in each graph. n = 3.

3.2. Impact of NAD Pollution on the Composition and Activity of Bacterial Communities in Microcosms

First, the effect of the four tested compounds on the Rhône river bacterial growth and activity was measured in vitro (growth assays) and in microcosms (qPCR) as detailed in SI Description S3. Overall, all NADs at high doses and all NADs at low doses except fluoxetine showed a reduction of growth in vitro during the exponential phase. However, only high doses of chlorpromazine and diphenhydramine showed an inhibitory effect after 48 h (SI Figures S2 and S3). This inhibitory effect in vitro was not reflected overall in river water microcosms, where only a transiently lower abundance was detected for high doses of chlorpromazine at day 2. After 8 days, only high doses of fluoxetine led to an increased overall bacterial abundance (SI Figure S3). All NAD doses used in this study were overall subinhibitory to river water bacteria in microcosms, although they all influenced bacterial growth during the exponential phase in vitro. The inhibitory effects of NADs on human microbiome bacteria have been extensively reported at clinical doses in vitro. , However, this study provides novel data on overall bacterial growth, abundance, and activity of NAD-exposed aquatic bacteria in vitro and in microcosms at both high and low doses.

In natural ecosystems, drug effects on growth may differ from in vitro, and overall subinhibitory concentrations may inhibit the growth of certain members of the community and benefit others. Thus, changes in bacterial dynamics at the community level induced by each of the NADs were determined at the metagenome-assembled genome (MAG) and amplicon sequence variant (ASV) levels for bacterial community composition and activity using NMDS and PERMANOVA analyses. Overall, high doses of all four NADs showed apparent effects on both bacterial community composition and activity from day 2, whereas low doses formed distinct clusters only at day 8 (Figure A–H). For all tested NADs, low and high doses clustered apart from each other and, in most cases, lower doses were closer to controls than high doses. All NADs showed significant (but low) correlations (p < 0.001) between dose and the composition of bacterial communities (PERMANOVA chlorpromazine R ² = 0.35, F = 13; diclofenac R ² = 0.2, F = 6; diphenhydramine R ² = 0.29, F = 10; fluoxetine R ² = 0.26, F = 12) and active bacterial communities (PERMANOVA chlorpromazine R ² = 0.28, F = 8; diclofenac R ² = 0.23, F = 7; diphenhydramine R ² = 0.36, F = 12; fluoxetine R ² = 0.31, F = 11). However, the NAD influence on bacterial community composition and activity was compound-, time-, and dose-dependent. Chlorpromazine had the highest correlation between the dose and community composition, whereas diphenhydramine dose had the highest correlation with active community composition. On the other hand, although diclofenac affected bacterial community composition and activity both at high and low doses, dose played a more limited role: correlations between diclofenac dose and bacterial community composition and activity were the lowest of all tested NADs. Previous research using artificial wastewater revealed that diclofenac-induced shifts in the community were more important when it was used as the sole carbon source: the use of diclofenac as a carbon source in these microcosms could partially explain the stronger divergence between low doses and controls observed for DFC than for the other NADs. All of these compounds have been associated with shifts in bacterial community composition in the human gut microbiome (chlorpromazine), in that of other mammals (fluoxetine), in activated sludge (diclofenac), and in stream biofilms (diphenhydramine). Here, we demonstrated the alteration of the composition and activity of bacterial communities in river water following NAD pollution at both high and low doses: even residual doses of NADs had an impact on the aquatic microbiome. This is consistent with previous research identifying medication as the factor explaining the largest variance in the composition of the human microbiome.

2.

Impact of the four tested NADs on the composition (A, C, E, G) and activity (B, D, F, H) of bacterial communities in river water microcosms. The dotted color represents the different NADs: (A, B): Chlorpromazine (CMZ) in green; (C, D): Diclofenac (DFC) in orange; (E, F): Diphenhydramine (DPH) in pink; (G, H): Fluoxetine (FXT) in blue. Lighter colors represent low doses (l), and darker colors represent high doses (h). The dot size represents sampling time (small = day 0, medium = day 2, and high = day 8). Bacterial community composition (A–G) was evaluated using Bray–Curtis dissimilarity calculated from the percentage of recruitment of MAGs across samples (percentage of reads from a sample recruited into a MAG). Active bacterial community composition (B–H) was evaluated using Bray–Curtis dissimilarity calculated from the total inferred abundance of 16S rRNA ASVs (number of 16S rRNA ASVs divided by sequencing depth and multiplied by the number of 16S rRNA copies per L). n = 3.

3.3. Antibiotic Resistance Selection under NAD Pollution

Since all NADs and doses showed an impact on the overall composition of the aquatic microbiome, MAGs were analyzed individually to identify bacterial populations selected by NAD exposure. All NADs except diclofenac were selected for MAGs that contained specific ARGs (i.e., all antibiotic resistance genes except those coding for efflux pumps) at high doses (Table ). Diclofenac might induce changes in the resistome mainly by acting as a carbon source that selects for endogenous bacteria, some of which carry efflux-pump genes (SI Table S4). Chlorpromazine was selected only for a Pseudomonas MAG at day 2 (19-fold increase relative to controls) and day 8 (506-fold increase) that contained an aphc­(9′)-Ic aminoglycoside resistance gene as well as 3 genes coding for efflux pumps (Table ). Two MAGs containing specific ARGs had an increased relative abundance in samples exposed to high doses of diphenhydramine compared to the nonpolluted controls. A Pseudomonas MAG selected both after day 2 (31-fold-increase) and day 8 (190-fold-increase) had a vatF streptogramin resistance gene and 17 genes coding for efflux pumps. A complete genome associated with Perlucidibaca selected by diphenhydramine after day 8 contained an aadA2 aminoglycoside resistance gene, a pmrE polymyxin resistance gene, and 3 genes involved in antibiotic efflux. Finally, fluoxetine was also selected for two MAGs containing specific ARGs. Sphingobium, selected at both day 2 and day 8, had an sgm-1 β-lactamase and an efflux-pump coding gene. Pseudoxanthomonas, selected only on day 8, contained a pjm-1 β-lactamase and 4 efflux-pump genes. All of these MAGs contained genes related to MGE and virulence genes (SI Table S5). In addition, all NADs except chlorpromazine (fluoxetine at low doses, diphenhydramine at high doses, and diclofenac at both low and high doses) increased the abundance of MAGs containing efflux pumps (SI Table S4). Here, we demonstrate the compound- and dose-specific and consistent selection of taxa harboring ARGs by NADs of different nature in aquatic ecosystems. Interestingly, two different NADs, chlorpromazine and diphenhydramine, were selected for two Pseudomonas MAGs that contain ARGs and are likely two different subpopulations according to their ARG, MGE, and virulence gene content. This genus, which is frequently detected in Rhône river water, , includes some opportunistic pathogen species. , Although the resolution of these analyses is insufficient to determine whether the subpopulations selected by NADs in this study are pathogens, both Pseudomonas MAGs harbor several virulence genes that might contribute to their pathogenicity to human hosts (as do the other selected MAGs). Further research should identify antibiotic-resistant aquatic subpopulations selected by NAD pollution to the species level in order to quantify the contribution of NAD aquatic pollution to the selection of antibiotic-resistant pathogens.

1. Relative Abundance (Average and Standard Deviation) and Abundance Fold-Increase in Polluted Samples Compared to Controls of the Metagenome-Assembled Genomes (MAGs) Containing Specific Antibiotic Resistance Genes Selected by NADs after 2 and/or 8 Exposure Days .

NAD	MAG	day	dose	relative abundance	fold-increase to controls
CMZ	taxonomy: Pseudomonas	day 0	controls	0.02 ± 3.9 × 10–3
			low	7.2 × 10–3 ± 1.8 × 10–3	0.47 (ns)
			high	1.3 ± 1.3	85 (ns)
	completion: 71.8% redundancy: 7%	day 2	controls	0.4 ± 0.6
			low	0.02 ± 5.6 × 10–3	0.05 (ns)
			high	6.6 ± 2.9	18.6 (**)
	ARGs: aph-(9′)-Ic , mexF, muxC, opmH	day 8	controls	0.01 ± 0.01
			low	0.02 ± 2.2 × 10–3	1.47 (ns)
			ligh	7 ± 4.2	505.63 (***)
DPH	taxonomy: Perlucidibaca	day 0	controls	1.6 × 10–3 ± 3.3 × 10–4
			low	1.9 × 10–3 ± 5.8 × 10–4	1.14 (ns)
			high	0.06 ± 0.01	35.43 (ns)
	completion: 100% redundancy: 0%	day 2	controls	0.02 ± 0.03
			low	0.01 ± 0.01	0.69 (ns)
			high	0.5 ± 0.29	24.45 (ns)
	ARGs: aadA2, pmrE, smeR, acrB,ugd, kdpE	day 8	controls	3 × 10–3 ± 1.5 × 10–3
			low	0.01 ± 5.9 × 10–3	3.87 (ns)
			high	2.06 ± 1.01	690.96 (****)
	taxonomy: Pseudomonas	day 0	Controls	0.01 ± 3 × 10–3
			low	0.01 ± 4.3 × 10–3	0.97 (ns)
			high	0.48 ± 0.17	43.13 (ns)
	completion: 62% redundancy: 2.8%	day 2	controls	0.32 ± 0.53
			low	0.25 ± 5.2 × 10–2	0.79 (ns)
			high	9.93 ± 5.17	31.05 (****)
	ARGs:  vatF, 17 efflux-pump genes	day 8	controls	0.02 ± 0.02
			low	0.11 ± 0.03	4.91 (ns)
			high	4.28 ± 1.59	190.19 (*)
FXT	taxonomy: Sphingobium	day 0	Controls	0.14 ± 0.02
			low	0.16 ± 0.04	1.12 (ns)
			high	0.37 ± 0.09	2.62 (ns)
	completion: 69% redundancy: 5.6%	day 2	controls	0.28 ± 0.25
			low	0.24 ± 0.22	0.87 (ns)
			high	2.68 ± 0.50	9.66 (****)
	ARGs: sgm-1, acrB	day 8	controls	0.07 ± 0.03
			low	0.09 ± 0.03	1.31 (ns)
			high	3.92 ± 0.63	58.74 (****)
	taxonomy: Pseudoxanthomonas	day 0	controls	0.01 ± 4.1 × 10–3
			low	0.02 ± 3.1 × 10–3	1.16 (ns)
			high	0.01 ± 4.6 × 10–3	1.39 (ns)
	completion: 77.5% redundancy: 2.8%	day 2	Controls	0.04 ± 0.03
			low	0.01 ± 2.2 × 10–3	0.31 (ns)
			high	0.04 ± 0.01	1.06 (ns)
	ARGs: sgm-1, acrB	day 8	controls	0.02 ± 8 × 10–3
			low	0.02 ± 2.3 × 10–3	0.91 (ns)
			high	1.1 ± 0.74	66.23 (****)

^aMAG relative abundance is expressed as the percentage of recruitment (percentage of sequences from a sample that are recruited into a MAG). Specific ARGs are underlined. ns = non-significant. **p < 0.01, ***p < 0.001, and ****p < 0.0001. n = 3.

Despite their overall subinhibitory levels, all NADs inhibited some members of the communities, as shown by the negative correlations between NAD dose and the relative abundance of several MAGs (6/29 for CMZ, 2/34 for DFC, 1/29 for DPH, and 5/26 for fluoxetine; SI Tables S6–S9). This is consistent with a targeted inhibition of sensitive members at overall subinhibitory levels and a selection of nonsensitive members, some of which carry antibiotic resistance traits. CMZ was positively correlated only to Pseudomonas, DPH to Pseudomonas, Perlucidibaca, Acinetobacter, and Ferrovibrio, and no positive correlations were found between DFC and FXT dose and the relative abundance of any MAG. The selection of ARG-harboring taxa could be direct (i.e., the selected taxa contain specific mechanisms to survive NAD pressure) or indirect (i.e., selected taxa are benefiting from the inhibition of other members of the community and the lower competition for resources). Mechanisms underlying this selection could include NAD metabolism/biodegradation by bacterial communities, the activation of stress responses and the acquisition of new ARGs by the selected bacterial hosts, and coresistance mechanisms (coselection and/or cross-resistance). This study highlights the urgent need for further extensive research to elucidate the mechanisms, which are likely to be compound- and dose-specific, involved in NAD-driven selection in complex environmental communities using multiomics approaches.

Furthermore, contig-based analyses revealed that all four NADs induced the selection of antibiotic resistance genes at high doses and that the resistome responded to NAD pollution in a compound-specific manner (Figure ). Chlorpromazine increased the relative abundance of β-lactamase genes and efflux-pump coding genes (Figure A). CMZ at high doses increased the abundance (p < 0.05) of 10 β-lactamases and the aminoglycoside resistance gene identified in the selected Pseudomonas MAG, at day 2. At day 8, CMZ selected 4 β-lactamases, 2 aminoglycoside resistance genes, and a fluoroquinolone resistance gene (Figure B). The strong selection for β-lactamase resistance genes induced by CMZ could be associated with its described synergistic effects with β-lactams and CMZ-induced bacterial cell wall alterations. The impact of CMZ on cell wall integrity could also be related to CMZ being the NAD with the most negative correlation between dose and MAG abundance described above. CMZ has been identified as an efflux-pump inhibitor , and an inductor of the heat shock response in Escherichia coli. Furthermore, despite previous research showing a CMZ-induced reduction of plasmid transmission, it selected for integration/excision and transfer genes (Figure A) and for a plasmid-encoded qnrB22 fluoroquinolone resistance gene in the microcosms at high doses. This drug, with strong selective effects and potential consequences for the emergence and dissemination of antibiotic resistance reported here, is nevertheless seldom monitored in river waters. Our results highlight the need to include CMZ in routine measurements to improve our understanding of its occurrence, fate, and effects on aquatic ecosystems.

3.

Changes in the resistome and mobilome induced by NAD pollution. (A): Selection of ARGs and MGE-associated genes induced by chlorpromazine (CMZ), diclofenac (DFC), diphenhydramine (DPH), and fluoxetine (FXT) at low (l) and high (h) doses. Red squares represent selection (increased abundance of contigs containing ARGs or MGE-related genes, p-value <0.05). D2: Day 2. D8: Day 8. (B): Relative abundance (percentage of recruitment) of ARG-containing contigs in samples polluted by chlorpromazine (CMZ). (C): Relative abundance of individual ARG-containing contigs in samples polluted by diclofenac (DFC). (D): Relative abundance of individual ARG-containing contigs in samples polluted by diphenhydramine (DPH). (E): Relative abundance of individual ARGs in samples polluted by fluoxetine (FXT). Blue: β-lactam resistance genes. Orange: aminoglycoside resistance genes. Green: fluoroquinolone resistance genes. Pink: streptogramin resistance genes. Gold: rifamycin resistance genes. Purple: phenicol resistance genes. Brown: macrolide resistance genes. *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001. n = 3.

Diclofenac had a limited impact on the resistome, increasing the abundance of aminoglycoside resistance genes only at day 2 (Figure A) and of efflux pumps at days 2 and 8 (Figure A). High doses were selected only for 2 β-lactamase genes at day 8 (Figure C). Despite being one of the most prevalent NADs detected in river waters worldwide, diclofenac has only been associated with antibiotic resistance by SOS-mediated processes: mutagenesis, transformation, and conjugation, consistent with the selection of replication/recombination/repair and transfer genes at high doses of diclofenac (Figure A). The SOS response has been extensively characterized in human microbiome bacteria, particularly in E. coli, which is a model widely used in these studies describing the NAD-induced SOS response. However, the SOS response is far from being universal, as illustrated by the variability in the lexA sequences and the composition of its regulon, the absence of SOS response genes in some bacteria and the lexA-independent response mediated by RecA in Acinetobacter baumanii, which is an opportunistic pathogen ubiquitous in the environment. The implication of this phenomenon in the response to SOS-inductors such as diclofenac in aquatic ecosystems might thus be overestimated, and further research should quantify the SOS response in environmental scenarios to determine its importance relative to clinical settings.

Diphenhydramine (DPH) increased the abundance of β-lactamase and efflux-pump coding genes at day 2 and day 8, of aminoglycoside resistance genes at day 8, and of other ARGs at day 2 (Figure A). Two β-lactamases had increased abundance after a 2-day exposure to high doses of DPH, and at day 8, the genes selected by high DPH doses included 5 β-lactamases, 1 aminoglycoside resistance gene, and a streptogramin resistance gene identified in the Pseudomonas MAG selected by DPH (Figure D). Moreover, MGE-related genes (except those related to phages) were selected under DPH exposure (Figure A). Despite the frequent detection and high stability of diphenhydramine in environmental settings, there is a lack of studies assessing its selective potential on antibiotic resistance. This study is the first to show DPH-driven antibiotic resistance selection (both at the ARG and at the MAG level) and the increased abundance of genes related to mobility under DPH exposure, highlighting the need for incorporating it in future studies.

Fluoxetine (FXT) had the strongest effect on ARG selection: it increased the abundance of β-lactamase genes, aminoglycoside resistance genes, efflux-pump genes, and other ARGs both at day 2 and day 8 (Figure A). FXT had the most consistent selection pattern between days 2 and 8:7 β-lactamases, 2 aminoglycoside resistance genes, and 2 phenicol resistance genes were selected by high doses of FXT at both day 2 and day 8 (Figure E). Some other genes were selected only on day 2 (2 β-lactamases and 1 aminoglycoside resistance gene) or on day 8 (the aph­(3′)-Ib aminoglycoside resistance gene). FXT was also selected for genes related to integration/excision, stability, and transfer (Figure A). These strong selective effects are consistent with previous research pointing to fluoxetine as a strong antibacterial (even stronger than some antibiotics) and its role as a stimulator of horizontal gene transfer.

Furthermore, although NAD-driven antibiotic resistance selection was dose-dependent, all NADs were selected for at least one ARG at low doses. CMZ, DFC, and DPH were selected for 2 β-lactamases and 1 aminoglycoside resistance gene (Figure B–D), whereas FXT was selected for 1 β-lactamase, 1 aminoglycoside, and 1 rifamycin resistance gene (Figure E). Thus, even at low environmental doses, the four NADs evaluated here are capable of inducing antibiotic resistance selection and thus impose a potential risk to human health.

Finally, contigs containing ARGs were inspected to determine whether these genes co-occurred with MGE-related and virulence genes. The CRP β-lactamase genes selected by all tested NADs at high doses and all tested NADs except for fluoxetine at low doses had a frequent co-occurrence (21/22 contigs) with the virulence factor vfr associated with the type IV pili involved in plasmid conjugation (SI Table S10). In addition, the aadA2 gene identified in the complete Perlucidibaca MAG selected by high doses of DPH was found in the same contig as a class 2 intI2 integrase gene. Finally, the aph­(3′)-Ib gene identified in contigs from the fluoxetine coassembly was associated with an integration/excision sequence. Therefore, some of the ARGs selected by NAD pollution seem to have the potential to be transferred between bacteria.

3.4. Identification of Bacteria Responding to NAD Pollution

Finally, active bacteria targeted by 16S rRNA gene sequencing were analyzed to identify members actively responding to (and potentially benefiting from) NAD pollution. Overall, high doses of CMZ, DPH, and FXT decrease the activity of higher proportions of the community than low doses (SI Table S11), whereas diclofenac (DFC) showed similar numbers of inhibited bacteria at low and high doses and a higher proportion of active bacteria at high doses (SI Table S11). This increase in the number of active ASVs is also observed at high doses of DPH and FXT but not CMZ. CMZ could have a dose-dependent inhibitory effect on several subpopulations in the Rhône river, while the trends observed for DFC could support its use as a carbon source by the community. High doses of DPH and FXT inhibited and activated more bacteria than did low doses. ASVs changing in response to NAD pollution varied as a function of time, compound, and dose, and were overall more diverse in DFC-polluted samples and less in CMZ- and DPH-polluted samples (SI Table S12). All compounds, particularly at high doses, increased the activity of several ASVs associated with the genus Pseudomonas. Thus, the changes in abundance of Pseudomonas ASVs from nonpolluted controls to polluted samples were analyzed at a closer level to identify patterns in the response of Pseudomonas subpopulations (Figure ). Overall, clear relationships were observable between the NAD dose and Pseudomonas response. In addition, diphenhydramine (DPH) at high doses induced the greatest increase in Pseudomonas activity, particularly on day 2, and diclofenac (DFC) induced the least increase in Pseudomonas activity for all tested MAGs. Of the 45 ASVs associated with Pseudomonas, only one (ASV 11) did not show increased activity under NAD exposure. Some ASVs responded to high doses of all tested NADs (ASVs number 3, 15 at day 2, and ASV 9 at day 8), but in general, Pseudomonas subpopulations responding to NAD pollution were compound-dependent (SI Tables S6–9). Pseudomonas has been identified in previous research in the Rhône river as a key player in the response to other aquatic pollutants, such as antibiotics and metals. , In addition, Pseudomonas strains were the most resistant phenotypes to CMZ across 10 human pathogen strains, and Pseudomonas was selected during diclofenac exposure in activated sludge. In this study, we demonstrated that this genus responded to NADs in an aquatic ecosystem with an increase of 44 ASVs (as well as two Pseudomonas MAGs selected by CMZ and DPH) under NAD pollution conditions. The variability in the response of Pseudomonas subpopulations as a function of the compound and dose also highlights the complex interactions and selective processes that may be involved. Further research targeting this genus in aquatic ecosystems should be carried out to understand subpopulation dynamics and the mechanisms that drive its pivotal role in polluted waters and quantify the risk that the pollutant-driven selection of Pseudomonas poses to human health.

4.

Log2 Fold increase in polluted samples relative to nonpolluted controls in total inferred 16S rRNA gene abundance of ASVs associated with the genus Pseudomonas. CMZ: chlorpromazine. DFC: diclofenac. DPH: diphenhydramine. FXT: fluoxetine. D2: day 2, D8: day 8, l: low dose, h: high dose. The y-axis represents the specific Pseudomonas ASV selected by each NAD and dose. Only ASVs that showed >5 log₂ Fold increase relative to controls at a p-value <0.001 are shown. Total inferred abundance (ASVs per L) was obtained by dividing ASV counts by sequencing depth and multiplying by 16S rRNA numbers determined by qPCR. n = 3.

3.5. Implications for Human and Environmental Health Risk

This research demonstrated that four different NADs with diverse mechanisms of action in human cells induced a response and were selected for antibiotic resistance in river water bacterial communities. Although this response was compound- and dose-dependent, some common trends were identified. All NADs and doses were stable in the environment and overall subinhibitory to river water bacteria, although they were selective for specific members of the community. In addition to this alteration of the bacterial communities under NAD exposure, a selection of ARG-harboring bacteria that also contained MGE-related and virulence genes was observed. The induction of Pseudomonas activity and the selection of ARGs, mostly those involved in the resistance to two major antibiotic classes used in clinics, β-lactams and aminoglycosides, was observed under exposure to all tested NADs. Thus, pharmaceuticals reaching surface waters on a daily basis may increase the selection and dissemination of antibiotic resistance in the environment, even at low, residual doses. Two major implications of this work are the need for further understanding of the mechanisms of action of NADs on bacterial cells and the urgency to assess the risk associated with NAD environmental pollution.

NADs induced a response in bacterial communities that may be detrimental to human health, and this research raises multiple questions regarding the mechanisms undermining this response in complex environmental communities. For example, the drug with the weakest effect on antibiotic resistance selection (diclofenac) is the only drug that does not target the central nervous system. Research efforts are needed to understand whether there is a link between the mechanisms of action of NADs in human cells and their effects on the microbiome as a whole (including but not limited to the human gut). Bacterial cells may be a target for NADs due to a common evolutionary origin of bacterial and mammalian targets. Diphenhydramine has been proposed as a coadjuvant to increase antibiotic effects in respiratory infections and in chemotherapy to sensitize tumoral cells and protect normal ones. Furthermore, specific genes to resist an antidiabetic drug have been identified in the human microbiome, and therefore, bacteria could contain NAD resistance genes that have not yet been identified. Other possibilities involve NADs acting as a substrate for bacterial efflux pumps and the involvement of ARGs in the response to NADs. For example, the consistent selection of β-lactamase genes observed in this study supports previous observations showing a decreased sensitivity to β-lactams in Gram-negative bacteria after NAD exposure. This raises questions regarding the role of β-lactamases in the response to NADs. These questions could not be addressed in this study, and an important body of research is needed to elucidate the interactions between NADs and antibiotic resistance and improve our understanding of their mechanisms of action beyond the human body.

Finally, we need to understand that our anthropocentric view of NADs as drugs acting uniquely on human target cells is obsolete and that they pose a significant risk to human, animal, and ecosystem health. Besides the concerns raised by their disruptive effects on fish behavior, their toxicity to algae and their potential dissemination across ecological boundaries by insects, this study shows that NADs represent a potential environmental risk associated with antibiotic resistance selection in freshwater ecosystems. This research used only four model compounds out of the hundreds of prescribed, nonprescribed, and illicit drugs that are found in surface waters. Research efforts are needed to assess the magnitude of the risk posed by NAD residues in surface waters, improve their removal from residual waters, and design strategies to mitigate their effects.

4. Conclusions

This research demonstrates how several non-antibiotic drugs with different mechanisms of action in human cells are stable in the aquatic environment and select for antibiotic resistance genes in a model aquatic ecosystem, both at high and residual doses. Although the response of aquatic bacterial communities to NAD pollution was compound-specific and dose-dependent, Pseudomonas, a genus that can be associated with opportunistic pathogens, was identified as a key member responding to NAD pollution. In addition, some of the selected antibiotic genes were associated with horizontal transfer and virulence genes, which may increase the risk of dissemination across bacterial populations and the pathogenicity of their hosts. Thus, the presence of NADs in the aquatic environment may increase human health risk associated with the emergence of antibiotic resistance in environmental settings. Further research should focus on the mechanisms of action of NADs on bacterial cells and their interactions with antibiotic resistance, as well as on quantifying the risk associated with the presence of NAD residues in surface waters.

The data sets generated and analyzed during the current study are available at the DDBJ repository, BioProject PRJDB20794, DRA accession DRA021299. All codes used in this study are available at: https://github.com/concscid/Sanchez-Cid-et-al-2025-NADs

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/envhealth.5c00238.

• Procedure for NAD quantification in river water samples; protocol for metagenomic sequence treatment and analysis; protocol for bacterial growth and qPCR assays; rarefaction curve of ASV richness discovery in the 16S rRNA sequencing; bacterial growth in 1:10 TSB medium under exposure to chlorpromazine, diclofenac, diphenhydramine, and fluoxetine at low and high doses for 48 h; impact of the four tested NADs at low and high doses on overall bacterial growth in 1:10 TSB medium for 48 h and on overall bacterial abundance and activity in river water microcosms for 0, 2, and 8 days; doses of NADs used in this study measured at day 0 by LC-MS/MS; sequencing depths obtained from the metagenomic and 16S rRNA sequencing from each sample; number of contigs, N50 and number of MAGs obtained from each metagenomic assembly; characterization of MAGs selected by NAD pollution that contained generalist ARGs (efflux-pump genes); MGE-related genes and virulence genes in MAGs selected by NAD pollution that contained specific ARGs; significant correlations between chlorpromazine, diclofenac, diphenhydramine, and fluoxetine dose and analyzed parameters (MAGs, ARGs, MGE-related genes, Pseudomonas 16S rRNA ASVs) in samples from days 2 and 8 combined; co-occurrence of ARGs, MGE-related genes, and virulence genes in the contigs; number of ASVs that were found to be statistically higher in total inferred abundance in polluted samples and in controls; and taxonomy of the ASVs selected by NADs (PDF)

The authors declare no competing financial interest.

Published as part of Environment & Health special issue “New Pollutants: Challenges and Prospects”.