Freshwater snail faecal metagenomes reveal environmental reservoirs of antimicrobial resistance genes across two continents

Abstract

The transfer of antimicrobial resistance genes (ARGs) from environmental microbes to pathogens is a critical but underexplored One Health driver of antimicrobial resistance. Here, we evaluate freshwater snails, which are geographically widespread aquatic invertebrates, as environmental reservoirs of ARGs. We collected faeces from eight gastropod genera at 15 freshwater locations across Malawi, Uganda, Zanzibar and the UK and conducted the first freshwater snail faecal metagenomic study. We detected putative ARGs predicted to confer resistance to 13 antibiotic classes, including carbapenems. All ARGs that could be assembled into metagenome-assembled genomes (MAGs) were found within Proteobacteria, which dominated the faecal microbiomes and were strongly associated with the total ARG load. In Malawi, we linked bla_OXA-181 (bla_OXA-48-like), a previously mobilized carbapenemase gene, to Shewanella xiamenensis, the gene’s known progenitor. We detected another bla_OXA-48-like gene by read mapping from a sample in the UK. We identified mobile colistin resistance (mcr)-like genes at 11 of 15 locations, with two mcr-7-like genes being found in an Aeromonas jandaei MAG in Uganda. Our findings highlight freshwater snail faeces as a One Health-relevant environmental reservoir of clinically important ARGs.

Impact Statement

This study presents the first metagenomic analysis of freshwater snail faeces across two continents, uncovering the role of aquatic invertebrates as reservoirs of antimicrobial resistance (AMR). Within environmental bacteria, we identified AMR genes that are predicted to confer resistance to last-resort antibiotics, such as carbapenems and colistin. These findings reinforce the One Health significance of AMR by exposing a previously overlooked environmental AMR reservoir. This reservoir poses a potential public health threat, particularly in regions where human, animal and environmental health intersect within freshwater bodies and where reliance on last-resort antibiotics may increase due to rising resistance to other available treatments.

Data Summary

Short reads from all samples sequenced in this study have been deposited in the Sequence Read Archive (SRA) under BioProject PRJNA1211045, with accession numbers SRX27371064–SRX27371078. Data and code used to carry out analyses in R are available at https://github.com/amoreo71/freshwater_snail_faecal.

Introduction

Antimicrobial resistance (AMR) is a growing global health emergency driven by industrial-scale use of antimicrobials in medicine, veterinary practice and agriculture since their discovery in the twentieth century [1]. The impact of AMR on human health is largely driven by the acquisition of antimicrobial resistance genes (ARGs) in pathogenic bacteria [2]. Consequently, most data used to investigate AMR originate from disease-causing organisms, yet there is increasing recognition that One Health frameworks accounting for the interdependence of human, animal and environmental health systems need consideration to better understand the origins, drivers and spread of AMR [3]. This approach is essential for assessing how AMR may be linked to crossing planetary boundaries, such as those related to freshwater use, biodiversity loss and climate change [4,5].

Outside the context of AMR, freshwater gastropod snails are already of interest in One Health research. Snails are globally distributed in aquatic environments, often playing critical roles as obligate intermediate hosts in the transmission of neglected tropical diseases when freshwater sources are contaminated with human and animal excreta. For instance, snails of the Bulinus, Biomphalaria and Oncomelania genera are key in the transmission of schistosomiasis [6], while Lymnaea snails transmit fascioliasis [7]. Recent efforts to characterize the microbiomes of medically important freshwater snails are generally motivated by the growing recognition that host-associated micro-organisms can significantly influence host–parasite interactions [8]. The microbiomes of Biomphalaria snails have been shown by 16S taxonomic profiling to be diverse and distinct from those of their freshwater environments [9], yet bacteriome transplant experiments have demonstrated the potential for these microbiomes to change depending on the surrounding environmental bacteria [10]. Freshwater snail populations are highly mobile [11], capable of translocating due to changes in ecosystems caused by flooding or the formation of artificial water bodies [12,13], and are distributed across a range of human-influenced areas [14,15]. Considering that the environments where freshwater snails graze – typically on detritus – can be contaminated with AMR enteric bacteria at sites of interaction with humans and animals, we wanted to explore the potential role of snails as widespread environmental reservoirs of ARGs [16]. This may be particularly relevant in low- and middle-income countries, where AMR is disproportionately linked to increased mortality [17] and where poor environmental health infrastructure is linked to AMR transmission across various One Health compartments [18,19].

The analysis of shotgun metagenomic sequence data is increasingly being used in the study of complex microbial samples and AMR, made possible by advances in next-generation sequencing and computational science technologies [20]. Metagenomic studies are not reliant on culture steps that alter microbial communities and are therefore able to yield non-selective data on sample composition [21]. Metagenomic analyses generally utilize short-read sequence data, which when coupled with the complexity of microbiome samples poses challenges in the assembly of metagenome-assembled genomes (MAGs) for low-abundance organisms without ultra-deep sequencing [22,23]. Approaches that map reads directly to reference databases are therefore useful for taxonomic profiling and the detection of low-abundance target genes as they do not rely on metagenome assembly and have proved to be more sensitive than assembly-based approaches for the detection of ARGs in complex samples such as faeces [24]. When combining read-based profiling with assembly-based approaches that uncover the genomic context of AMR, allowing the linking of ARGs to specific hosts and mobile genetic elements (MGEs), metagenomics can enable a comprehensive understanding of the resistome and its potential impacts on microbial ecology and global health.

Here, we present a dataset characterizing the first freshwater snail faecal metagenomes from 15 locations within Malawi, Uganda, Zanzibar (a self-governing region within the Republic of Tanzania) and the UK. Acknowledging that freshwater snail microbiomes vary by organ [25], and to mirror the natural mechanisms of faecal bacterial transfer from gastropods to the environment, we sampled faeces excreted by snails immediately after they were collected from their habitats, rather than extracting DNA directly from snail organs or rearing snails under laboratory conditions prior to DNA extraction. We use whole-genome metagenomics to investigate the role of freshwater gastropods as reservoirs, and potentially vectors, of AMR. Our goal is to identify clinically relevant ARGs within these common aquatic invertebrates and to examine their genomic context, providing insights and targets for future studies on ARG transfer to human and animal pathogens. We utilize direct read mapping for taxonomic profiling and ARG identification, followed by less sensitive but highly insightful assembly-based methods to explore the genomic context of ARGs.

Methods

Sampling, processing and sequencing

In 2023 and 2024, we collected freshwater snails at 15 freshwater locations in 4 administrative regions across 2 continents, targeting snails of a single genus at each location (Table S1, available in the online Supplementary Material). In Africa, collections were made from six locations in Malawi (five Bulinus and one Cleopatra), three locations in Uganda (one each of Biomphalaria, Bulinus and Physella) and four locations on Unguja Island, Zanzibar (one each of Bulinus, Cleopatra, Lanistes and Pila). In Europe, two locations in the UK were sampled (one Anisus and one Lymnaea). Sampling was opportunistic in nature and was carried out in conjunction with ongoing parasitological surveys that were not originally designed for balanced ecological comparisons between snail genera. As such, the presence of snail genera across the 15 sites reflects natural distributions at the time of sampling, rather than a standardized genus-level survey.

At each location, groups of 50 snails from the identified genus were gathered, rinsed and temporarily pooled in sterile water. After 24 h, faecal material deposited by each pooled snail group was collected to obtain adequate biomass for DNA extraction and sequencing. DNA was extracted from PBS-washed faecal samples with the QIAamp PowerFecal Pro DNA Kit (QIAGEN, Germany), as per the manufacturer’s instructions. Concentration, purity and integrity of extracted DNA were assessed using a Qubit 4 fluorometer (Thermo Fisher Scientific, USA), NanoDrop One spectrophotometry (Thermo Fisher Scientific) and the TapeStation System (Agilent Technologies, USA). Purified DNA extracts were shipped to Genewiz/AZENTA (UK) for library preparation, validation and shotgun metagenomic sequencing. Sequencing was carried out with the Illumina NovaSeq instrument using a 2×150 bp paired-end configuration. The number of 150 bp Illumina paired-end reads per sample ranged from 9,340,457 to 35,991,622 (Table S2). The resulting 15 metagenomes proceeded to downstream analyses, each representing the faeces of 50 snails from a specific genus at a specific location.

Bioinformatics

Read processing

Raw FASTQ sequence files were quality assessed using FASTQC (https://github.com/s-andrews/FastQC) (Version 0.11.9). Trimming with Trimmomatic [26] (Version 0.39) was performed to remove adaptor sequences and low-quality bases with a sliding window quality cutoff of Q20 and a minimum read length of 50 bp. Standard practice in metagenomic studies is to filter out host DNA by precisely aligning all reads to host reference genomes. As our study involved the analysis of pooled DNA samples from hosts whose genomes are not all available in databases, we applied the Kraken Software Suite protocol outlined by Lu et al. [27] to identify and extract contaminant and host DNA in a multi-step approach. Trimmed reads were classified using Kraken 2 [28] (Version 2.1.3). Kraken 2 works by matching read k-mers against a database of k-mers from existing genomes, before assigning each read to the lowest common ancestor of all genomes that share those k-mers. We initially classified reads using the Kraken 2 standard database, which contains RefSeq archaea, bacteria, viruses, plasmid complete genomes, UniVec Core and the human reference genome, GRCh38. We then filtered out reads identified as eukaryotic (mapping to the human reference genome, either due to low-level human DNA contamination or because they were eukaryote-generic reads originating from snails) using the KrakenTools ‘extract_kraken_reads.py’ script (https://github.com/jenniferlu717/KrakenTools). Retained reads were considered to have passed quality control.

We also built a custom Kraken 2 Gastropoda database containing the available GenBank snail genomes for all species within the genera represented in our samples: Anisus vortex (GCA_949126835.1), Anisus vorticulus (GCA_964264155.1), Biomphalaria glabrata (GCA_947242115.1), Biomphalaria pfeifferi (GCA_030265305.1), Biomphalaria straminea (GCA_021533235.1), Biomphalaria sudanica (GCA_036873155.1), Bulinus truncatus (GCA_021962125.1), Lanistes nyassanus (GCA_004794575.1), Lymnaea stagnalis (GCA_964033795.1) and Physella acuta (GCA_028476545.3). Genomes from the Cleopatra and Pila genera were not included, as they were not available within GenBank. Separately from classifications using the Kraken 2 standard database, we mapped all trimmed reads passing quality control to the custom-built Gastropoda database. We then extracted reads identified to be of gastropod origin that had not already been discarded after being labelled as eukaryotic by the Kraken 2 standard database. This conservatively extracted subset of remaining Gastropoda reads that passed quality control (0.09–0.44% per sample) also underwent ARG detection steps to provide us with reassurance that our inability to precisely align each sample’s reads to a specific host reference genome did not bias ARG profiles.

Taxonomic profiling

Kraken 2 reports containing bacterial read taxonomic assignments were used to run species abundance estimation with Bracken [29] (Version 2.9), with a minimum number of reads required for a classification at the specified rank set at 10, to reduce noise from potential low-abundance contaminant species, as per default recommendations.

ARG database and clustering

We used the ResFinder database [30] (Version 2.3.2) to screen for ARGs. This database was selected because it focuses on acquired ARGs with known clinical relevance and mobile potential, thereby aligning with our aim to identify resistance mechanisms that may contribute to the spread of AMR from environmental reservoirs to clinical settings. Sequences in the ResFinder database were downloaded and then clustered using CD-HIT [31] (Version 4.8.1) to identify ARG clusters containing sequences with a minimum nt identity of 90%, referred to as Cluster90s [24].

Read-based ARG detection

To identify ARGs within each sample, reads were aligned to reference sequences in the ResFinder database using KMA [32] (Version 1.4.14), a tool designed to map raw reads directly against redundant databases such as ARG databases. Alignments were then filtered to retain only those with a minimum 90% query identity and 60% template coverage – a common approach in the analysis of complex environmental metagenomes that balances accuracy and sensitivity by favouring reliable ARG identification while allowing for partial matches, thereby reducing the risk of missing low-abundance or divergent ARGs [33]. For all reference sequences with alignment hits after filtering, the corresponding Cluster90 was identified and annotated manually. We calculated the normalized abundance of each ARG from read mapping data, expressed as reads per kilobase per million (RPKM) mapped bacterial reads, and then summed the RPKM of all ARGs to calculate total ARG load per sample. We also performed read-based ARG detection on random subsets of reads (5 million, 10 million, 15 million and 20 million) from samples with >20 million read pairs following quality control (Uk1, Z1, Z3 and Z4). The ARG detection data generated from these read subsets were used to plot rarefaction curves and assess diversity saturation.

Metagenome assembly, binning and classification

Reads identified as being bacterial by Kraken 2 were assembled with metaSPAdes [34] (Version 3.11.1), using the default k-mer lengths of 21, 33 and 55. Assembly statistics were calculated with QUAST [35] (Version 5.0.2) using contiguous sequences (contigs) ≥500 bp, to ignore low-quality sequences and ensure compatibility with ARG-containing regions (Table S2). Contigs were indexed using the Burrows Wheeler Aligner (BWA) (Version 0.7.17), before bacterial reads were aligned to the indexed contigs with ‘bwa-mem’ [36]. The contigs and resulting sorted BAM files were parsed to the ‘jgi_summarize_bam_contig_depth’ script from MetaBAT2 [37] (Version 2.17), before the resulting depth files were used by MetaBAT2 to bin assembled contigs of ≥2,500 bp to putative genomes (i.e. MAGs). CheckM [38] (Version 1.1.2) was used to assess the completeness and contamination of each MAG. CheckM outputs were compared to standards set by the genome standard consortium to classify MAGs as high-quality (>90% completeness and <5% contamination), medium-quality MAGs (>50% completeness and <10% contamination) or low-quality (<50% completeness and/or >10% contamination) [39]. MAGs deemed as medium- or high-quality proceeded to further analysis. MAGs underwent taxonomic classification using GTDB-Tk [40] (Version 2.1.1) against the Genome Taxonomy Database (Version 09-RS220).

Assembly-based profiling of ARGs and MGEs

Assemblies were screened against the ResFinder [30] and ISfinder [41] databases using ABRicate (https://github.com/tseemann/abricate) (Version 1.0.1) at default settings (minimum 80% identity and 80% query cover) to identify contigs containing ARGs and insertion sequence (IS) elements. Compared to our read mapping approach, in which we used a higher identity threshold to identify hits (90%) in combination with a lower minimum query cover (60%) due to the fragmented nature of using raw sequences to identify low-abundance genes, the assembly-based approach allowed for the detection of ARGs and MGEs that could then be characterized further within their genomic context following annotation. Annotations were performed using Bakta [42] (Version 1.9.3; full database Version 5.1). To elucidate the potential origins and dissemination pathways of ARGs, selected contigs were subjected to a sequence similarity search using the Nucleotide Basic Local Alignment Search Tool (blast) webtool (megablast algorithm) against the GenBank Nucleotide database. This step aimed to identify existing homologous sequences and/or gene clusters. The Clinker webtool (https://cagecat.bioinformatics.nl/tools/clinker) [43] was used to visualize gene cluster comparisons.

Statistical analysis and visualization

Species abundance and read-based ARG assay data were combined with hierarchical/classification data and sample metadata to generate a TreeSummarizedExperiment data container [44]. Data associated with MAGs were handled separately. Within the R environment, downstream data wrangling, analysis and visualization took place predominantly using the R/Bioconductor mia and miaViz (Version 1.12.0) packages and the R/CRAN ggplot2 (Version 3.5.1) and pheatmap (Version 1.0.12) packages.

We calculated alpha diversity statistics at the species level, leveraging the high taxonomic resolution of shotgun metagenomic data. A Bray–Curtis dissimilarity matrix containing pairwise dissimilarities among microbiome samples was generated at the species level using relative abundance data to investigate beta diversity and to perform principal coordinate analysis (PCoA). Ahead of hierarchical clustering of samples and taxa at the order level, centred log-ratio (CLR) transformations were applied to microbial composition data [45] to normalize for compositional effects, with a pseudocount added to manage zero values using the mia package. Hierarchical clustering of both taxa and samples was applied to standardized CLR-transformed (CLR-z) data using Euclidean distance and complete linkage, as implemented by default in the pheatmap package. Before plotting ARG Cluster90 heatmaps, a log-transformation was applied to ARG load to enable visualization across a wide range of abundances, with a pseudocount added to manage zero values using the mia package.

To assess monotonic relationships between numerical variables, such as the relative abundance of Proteobacteria and ARG load, we used Spearman’s rank correlation coefficient testing. This non-parametric test was chosen due to the small number of locations sampled (n=15), its suitability for non-normally distributed data and its robustness to outliers. Significance was determined with a threshold of P<0.05. These analyses were followed by analyses of ARGs and MGEs found in MAGs, which supplement read-based profiles with high-resolution, species-specific context. Here, we used Wilcoxon rank-sum tests (non-parametric) to assess differences in the counts of ARGs and IS elements detected per complete MAG (calculated using raw ARG and IS element counts alongside genome completeness values per MAG) between taxa. Again, the threshold for statistical significance was set at P<0.05.

Results

The freshwater snail faecal microbiome is diverse, varied, yet consistently rich in Proteobacteria

Species abundance estimations showed that Proteobacteria was the most abundant phylum observed, with a mean relative abundance of 66.27% (Fig. 1a). Enterobacterales, an order within the Proteobacteria containing multiple species of clinical concern due to their ability to acquire and exchange ARGs, were well represented, with a mean relative abundance of 4.79%. The Uk1 sample was a notable outlier in which the relative abundance of Enterobacterales was estimated to be 41.70%. When excluding Uk1, the mean relative abundance of Enterobacterales dropped to 2.15%. Four orders were more abundant than Enterobacterales (Burkholderiales, Rhizobiales, Aeromonadales and Pseudomonadales), all of which had a mean relative abundance of greater than 5% (Fig. 1b). All samples exhibited high community (alpha) diversity. Shannon’s diversity index ranged from 3.881 (M5) to 7.372 (Z2) (mean=6.474, median=6.888). Simpson’s diversity index, in which scores from 0 to 1 are possible (1=maximum possible diversity), ranged from 0.793 (M5) to 0.999 (Z2) (mean=0.972, median=0.994). We observed that alpha diversity (Shannon’s diversity index) was negatively correlated with the relative abundance of Proteobacteria (Spearman’s rank correlation coefficient=−0.636; P=0.0129).

Fig. 1. Exploration and comparison of bacterial taxa within freshwater snail faeces: (a) relative abundance of the top phyla, (b) relative abundance of the top orders and (c) clustered heatmap of standardized order abundances (CLR-z) across snail hosts and geographic locations. Column and row trees cluster samples and taxa, respectively, based on a hierarchical clustering approach. Taxa with a mean relative abundance <1% are classified as ‘Other’ in parts (a) and (b).

Relative abundance data were aggregated at the order level, subjected to CLR transformation and subsequently standardized (CLR-z) to facilitate visual comparisons between taxa with varying absolute abundances (Fig. 1c). From the 160 bacterial orders identified across all samples, the 50 most abundant orders, accounting for 97.92% of the total microbiome, were selected for hierarchical clustering and visualization. Samples from the same administrative region clustered closely together, with the exception of Malawian samples, which were split into two distinct groups. This split corresponded to the two districts where the samples were collected: samples M1, M2, M3 and M4, collected in Mangochi District on the southern shorelines of Lake Malawi, clustered together, while samples M5 and M6, collected in Nsanje District, over 275 km from Mangochi District sampling sites, formed a separate cluster. Although they clustered together, northern hemisphere samples (UK) did not form a highly divergent cluster despite likely differences in climate-related and environmental conditions. Microbiomes from Bulinus snails (the most-represented genus, obtained from seven locations) showed notable divergence: the sample collected in Zanzibar clustered at one extreme of the dendrogram alongside samples from other snail genera, distinct from Bulinus samples collected in Uganda and the two Malawian districts.

When investigating diversity between samples at the species level, a Bray–Curtis dissimilarity matrix (Fig. S1a) showed a range of divergence between Malawian Bulinus snails. We observed moderate divergence (0.455) between two samples (M2 and M4) collected a little over 5 km from each other, both from stagnant water bodies close to the Lake Malawi shoreline, suggesting that local factors play a key role in influencing the microbiome. Despite these local variations, PCoA performed using Bray–Curtis distances at the species level (Fig. S1b) revealed similar locational trends to our order-level hierarchical clustering visualization, in particular showing that the four samples from Mangochi District, Malawi (M1–M4), as well as the four from Unguja Island, Zanzibar (Z1–Z4), clustered closely together within their respective groups.

ARGs conferring resistance to first- and last-line antibiotics detected by read mapping

Using read-based k-mer alignment to screen for ARGs against the ResFinder database [30,32], reads mapped to 54 unique ARGs in total (minimum 90% query identity and 60% template coverage), belonging to 41 ARG Cluster90s (in which reference genes share a minimum 90% nt identity) [24]. ARG richness (the number of unique ARGs identified by read mapping in any given sample) ranged from 1 to 16 (mean=7.07, median=6). ARG load ranged from 3.16 to 357.04 RPKM (mean=87.99, median=59.62). Although the depth of sequencing and number of unique ARGs detected per sample were not correlated (Spearman’s rank correlation coefficient=0.283; P=0.3065) (Fig. S2), we used random subsets of 5 million, 10 million, 15 million and 20 million read pairs from samples with more than 20 million pairs passing quality control (Uk1, Z1, Z3 and Z4) to further explore ARG saturation. Rarefaction curves showed that while ARG saturation can be reached by 10 million reads (e.g. Uk1), other samples (Z1, Z3 and Z4) required up to 20 million reads to achieve saturation (Fig. S3). The ARGs that were not detected in smaller read subsets were only detected at very low abundance in full read sets, therefore having a minimal effect on ARG load.

When we aggregated ARG relative abundance by class (i.e. the antibiotic that the ARGs confer resistance to), beta-lactamase genes (excluding carbapenemases) were the most abundant, contributing to 53.58% of the total ARG load across samples, followed by genes conferring resistance to colistin (14.38%), carbapenems (10.27%), sulphonamides (5.32%), tetracyclines (5.12%), aminoglycosides (2.77%), fosfomycin (2.15%), macrolides (2.14%), trimethoprim (1.90%), lincosamides (0.51%), chloramphenicol (0.47%), fusidic acid (0.19%) and quinolones (0.03%) (Fig. S4). Efflux pump genes conferring multidrug resistance contributed 1.16% of the total ARG load. Beta-lactamase genes from the OXA-12 Cluster90 were the most prevalent in our samples and were detected in 14 of the 15 metagenomes (Fig. 2). We also detected ARGs from the CphA Cluster90, a family of carbapenem-hydrolysing metallo-beta-lactamase genes, in 10 of the 15 samples. Three OXA-carbapenemase genes were identified, each from a different country. These included two ARGs from the OXA-48 Cluster90 (bla_OXA-181 and bla_OXA-547, in Malawi and the UK, respectively), of concern globally due to plasmid-borne dissemination among Enterobacterales [46], and bla_OXA-212 (Uganda) from the OXA-211 Cluster90 of carbapenemase genes that naturally occur in Acinetobacter johnsonii [47]. We also detected putative colistin resistance genes, predominantly variants of the mcr-7.1 gene, in 11 metagenomes.

Fig. 2. Heatmap of normalized ARG Cluster90 abundances across samples, expressed as log(RPKM). Cluster90s are colour-coded by the antibiotic class that they are predicted to confer resistance against.

We observed a positive correlation between the relative abundance of Proteobacteria (the predominant bacterial order) and ARG load (Spearman’s rank correlation coefficient=0.846; P<0.0001) (Fig. 3a) . A significant association between Proteobacteria relative abundance and ARG richness was not detected (Spearman’s rank correlation coefficient=0.294; P=0.2877). Among the 20 orders of bacteria with a mean relative abundance above 1% (Table S3), only Pseudomonadales had a significant positive correlation with ARG load (Spearman’s rank correlation coefficient=0.604; P=0.0195), although this was not a clear linear trend (Fig. 3b) . Meanwhile, the relative abundance of Aeromonadales was particularly high in two samples (M5 and Ug3), within which ARG load was also markedly higher (>300 RPKM) than all other samples (<100 RPKM). However, the overall correlation between samples was deemed non-significant (Spearman’s rank correlation coefficient=0.357; P=0.1916). These results indicate that different orders within the Proteobacteria are involved in AMR to varying extents in different snail microbiomes and that assembly-based methods that elucidate the genomic context of AMR are needed to explore this with greater resolution.

Fig. 3. Associations between Proteobacteria and ARG load: (a) relative abundance of Proteobacteria versus ARG load and (b) relative abundance of the five most abundant orders within the Proteobacteria versus ARG load. Within-plot labels in part (a) are snail genera. Spearman’s rank correlation coefficients and associated P-values are presented for each taxonomic group for relative abundance versus ARG load.

Assembly-based methods link ARGs and MGEs to various bacterial genera and identify known progenitors of mobile carbapenemase and colistin resistance

We detected fewer ARGs by screening metagenome assemblies against the ResFinder database using ABRicate (mean=2.53, median=2) than we did by read mapping (Fig. S5). However, we were able to acquire information on the carriage of ARGs by specific taxonomic groups following assembly and binning of contigs. Across all samples, 10 of the 38 ARGs (26.3%) assembled on contigs could be binned to high-quality (>90% completeness and <5% contamination) or medium-quality (>50% completeness and <10% contamination) MAGs. These ARGs were all binned to MAGs classified as Proteobacteria (genera: Aeromonas, n=7; Shewanella, n=1; and Stenotrophomonas, n=2) (Fig. 4) . Specifically, these MAGs all belonged to the Gammaproteobacteria class, from which ARGs were identified at a rate of 0.62 per complete genome. Even within our small dataset of 29 medium- and high-quality MAGs, the difference in ARG detection rates in Gammaproteobacteria, which made up 85.7% (18/21) of Proteobacteria MAGs and 62.1% (18/29) of all MAGs, was significantly higher than in all other bacterial classes (P=0.0391). We detected between 8 and 97 putative IS elements per sample (mean=37.8, median=29). IS elements were more commonly identified in Gammaproteobacteria MAGs (1.03 per complete genome) than in other MAGs (0.57 per complete genome). Although the difference observed did not meet the threshold for statistical significance within this relatively small dataset (P=0.1223), the trend likely reflects underlying biological differences in MGE content between taxonomic groups.

Fig. 4. Summary of medium- and high-quality MAGs: (a) genera, (b) AMR determinants (annotated as ‘-like’ if nt identity is <100% when compared to reference) and (c) IS families (assigned based on the ISfinder annotation of the corresponding hit).

We identified two contigs containing both IS elements and ARGs, neither of which could be binned to a MAG. A 4,042 bp contig (Ug2 NODE_1302: Bulinus spp. faeces, Lake Victoria, Uganda) contained sul1 (sulphonamide-resistant dihydropteroate synthase) and bla_OXA-10 (beta-lactamase) genes, on which both ARGs were located upstream of IS6100. The gene cluster also contained qacEdelta (a gene conferring resistance to antiseptics) and yhbS (putative N-acetyltransferase). The contig displays 100% query cover with one complete genome in the GenBank Nucleotide database – Aeromonas hydrophila (accession no. AP025277) – with 99.97% identity, but also maintains high query cover (>95%) and identity (>99%) with chromosomal and plasmid sequences from various other bacteria (Pseudomonas aeruginosa, Vibrio cholerae, Morganella morganii, Salmonella enterica and Klebsiella pneumoniae). A second, shorter, 3,502 bp contig (M2 NODE_687: Bulinus spp. faeces, Mangochi District, Malawi) aligned with 99.94% nt identity to Ug NODE_1302 over the region containing qacEdelta, sul1 and yhbS upstream of IS6100 (Fig. 5), with a single nucleotide polymorphism (SNP) in the sul1 gene. The similarity observed between the two contigs from Ug2 and M2, and with sequences from a range of bacterial species in GenBank, demonstrates the widespread nature of certain ARGs associated with MGEs, which we show can be found in freshwater snail faeces. However, we could only link MGEs directly to ARGs on contigs in the above two examples. Searches for ARGs associated with IS elements in transposons should ideally capture 24.34 kbp of DNA (based on the length of composite and unit transposons in The Transposon Registry) [48]. The median length of contigs containing ARGs in our dataset was 2.10 kbp, demonstrating the limitations in assembling large contigs from short reads in diverse metagenomic samples [23] and the associated inability to capture all co-localized ARGs and MGEs.

Fig. 5. Gene cluster comparison between Ug2 NODE_1302 and M2 NODE_687 containing bla_OXA-10 (Ug2 only), qacEdelta1, sul1 and yhbS upstream of IS6100. Regions coding for hypothetical proteins are not displayed.

One of the two bla_OXA-48-like carbapenemase genes identified by read mapping was also detected in metagenome assemblies. NODE_173 from the M5 metagenome (Cleopatra spp. faeces, Nsanje District, Malawi) contained bla_OXA-181 on a 20,996 bp contig with 17 open reading frames. The contig was binned to a high-quality MAG classified as Shewanella xiamenensis (Table 1). When the bla_OXA-181-containing contig was queried against the GenBank Nucleotide database, the closest alignment was to chromosomal DNA from an S. xiamenensis isolate (accession no. AP026732) obtained from urban drain water from Vietnam (100% query cover and 98.51% identity) [49]. With only one SNP in a non-coding region, the contig also aligns with 99.92% identity over a 1,309 bp region (8,073–9,381) to various plasmids from the Enterobacterales containing the internationally disseminated bla_OXA-181 gene and a truncated lysR (ΔlysR) gene, including the KP3-A plasmid from which bla_OXA-181 was first characterized in a clinical K. pneumoniae isolate (accession no. JN205800) [50]. The region containing bla_OXA-181 and ΔlysR within the KP3-A (and other) plasmid(s) has previously been traced back to its origin, S. xiamenensis, where bla_OXA-181 is found adjacent to a complete lysR gene [51], as is also seen in our annotated sequences (Fig. S6). The ability to detect, assemble and bin this critical priority ARG is aided by the high relative abundance of Shewanella in the M5 sample (5.07%). We also observed a strong Shewanella signal in the Uk2 sample (relative abundance=1.33 %), from which another bla_OXA-48-like variant (bla_OXA-547) was detected by read mapping, when compared to the remaining 13 samples in which no bla_OXA-48-like genes were detectable (mean Shewanella relative abundance=0.11 %).

Table 1. Summary of MAGs recovered from freshwater snail faecal microbiomes with GenBank accession numbers for high-quality MAGs.

Phylum	Species	Sample	Quality (completeness, contamination)	ARGs	IS families	GenBank accession
Bacteroidota	Chryseobacterium aquaticum	Ug3	Medium (51.7%, 0.3%)	–	–	–
	Cloacibacterium spp.	Z3	Medium (52.3%, 0.0%)	–	–	–
	Cloacibacterium spp.	Z4	Medium (63.8%, 0.0%)	–	–	–
Cyanobacteria	Trichormus spp.	M6	High (97.6%, 0.2%)	–	IS200/IS605, IS200/IS605, IS200/IS605, IS200/IS605, ISKra4	JBODRP000000000
Firmicutes	Exiguobacterium indicum	Z1	High (90.4%, 2.0%)	–	–	JBODRM000000000
	Exiguobacterium profundum	Z1	High (90.7%, 0.7%)	–	–	JBODRL000000000
	Virgibacillus spp.	M5	Medium (55.1%, 0.6%)	–	–	–
Mycoplasmatota	Mycoplasma spp.	Ug1	High (90.2%, 0.4%)	–	–	JBODRO000000000
Proteobacteria	Acinetobacter bohemicus	Uk1	Medium (51.7%, 3.4%)	–	–	–
	Acinetobacter tandoii	M5	High (97.9%, 1.2%)	–	–	JBODRR000000000
	Aeromonas jandaei	Ug3	High (95.3%, 0.6%)	blaCphA-like, blaOXA-12-like, mcr-7-like, mcr-7-like	IS630, IS1595	JBODRN000000000
	Aeromonas jandaei	Z1	Medium (85.1%, 1.2%)	blaOXA-12-like	–	–
	Aeromonas sobria	Uk2	Medium (82.1%, 0.6%)	blaOXA-12-like	–	–
	Aeromonas spp.	M5	High (93.3%, 0.6%)	blaOXA-12-like	IS4, IS4, IS4, IS4	JBODRQ000000000
	Aeromonas spp.	Z4	Medium (74.8%, 1.2%)	–	IS4	–
	Aeromonas veronii	Ug2	Medium (76.8%, 0.8%)	–	IS3, IS481	–
	Aliarcobacter spp.	M5	Medium (89.0%, 2.4%)	–	IS1182	–
	Brevundimonas nasdae	Ug2	Medium (73.4%, 1.5%)	–	–	–
	Curvibacter spp.	M1	Medium (70.1%, 0.5%)	–	–	–
	Duganella spp.	M1	Medium (76.6%, 0.5%)	–	–	–
	Erythrobacter spp.	M6	Medium (57.8%, 0.0%)	–	–	–
	Limnohabitans spp.	M2	Medium (56.5%, 0.2%)	–	IS630	–
	Massilia spp.	Ug2	Medium (68.9%, 0.7%)	–	IS256	–
	Pseudomonas sivasensis	Uk1	Medium (51.8%, 1.8%)	–	–	–
	Serratia fonticola	Uk1	Medium (71.9%, 0.0%)	–	IS21, IS110	–
	Shewanella xiamenensis	M5	High (96.0%, 0.8%)	bla OXA-181	IS630	JBODRS000000000
	Stenotrophomonas bentonitica	Ug1	Medium (55.2%, 0.0%)	–	–	–
	Stenotrophomonas maltophilia	Ug1	Medium (84.1%, 1.7%)	aph(3′)-like, blaL1-like	IS110	–
	Vogesella spp.	Z3	Medium (65.5%, 0.0%)	–	–	–

ARGs are annotated as ‘-like’ if nt identity is <100% when compared to reference. Each IS family was assigned based on the ISfinder annotation of the corresponding hit.

'–', indicates none found (for ARGs and IS elements) or no GenBank accession (for medium-quality MAGs).

Two mcr-7-like lipid A phosphoethanolamine transferase genes were detected in a high-quality Aeromonas jandaei MAG (Table 1) recovered from the Ug3 metagenome (Physella spp. faeces, Kimi Island, Uganda) on separate contigs (NODE_28 and NODE_32 : 86.88% and 81.93% nt identity to mcr-7.1, respectively). Lipid A phosphoethanolamine transferase enzymes cause a reduction in the overall net-negative charge of the outer membrane, conferring resistance to colistin among some gram-negative bacteria, and are a major public health concern due to their propensity to be mobilized from progenitor species chromosomes to plasmids, limiting last-resort antibiotic treatment options for infections caused by multi-drug-resistant pathogens [52]. Aeromonas spp. are the known progenitors of mcr-3 and mcr-7 [53]. From our 15 samples, reads mapped to the MCR-3 Cluster90 in one and the MCR-7 Cluster90 in 10, underscoring the widespread prevalence of Aeromonas-associated mcr-like genes in the sampled environment.

Discussion

Our findings provide novel insights into the role that freshwater snails play as AMR reservoirs, revealing a potential widespread environmental source of clinically relevant ARGs. Through analysis of snail faecal metagenomes from diverse geographic locations, we demonstrate that their microbiomes harbour a wide array of bacterial species and ARGs, including those conferring resistance to both first- and last-line antibiotics. Identification of environmental progenitors of potentially mobile ARGs conferring resistance to carbapenems and colistin is particularly noteworthy in sub-Saharan Africa, where such antibiotics are not commonly available or prescribed for patients with severe infections caused by Enterobacterales, but where resistance among Enterobacterales to other treatment options such as penicillins, third-generation cephalosporins and gentamicin has increased [54]. A likely increase in carbapenem prescribing as a result of this may introduce additional selective pressures to aquatic habitats, driving selection and transfer of ARGs from environmental reservoirs to clinically relevant bacteria [55]. Our results emphasize the potential for snails to contribute to this dissemination in freshwater ecosystems, particularly in those modified by anthropogenic change with deficits in local WASH (water, sanitation and hygiene) infrastructures [3,19], although we are not able to confirm ARG mobility from these sequencing data alone.

Proteobacteria dominated the snail faecal microbiomes, reinforcing results from previous studies that found Proteobacteria to be highly abundant in other freshwater snail microbiomes such as the gut and haemolymph [9,56,58]. The significant correlations observed between Proteobacteria abundance and both alpha diversity and ARG load suggest that an abundance of Proteobacteria reduces overall microbiome diversity while increasing the relative abundance of ARGs within the microbial community. We binned genes from the three most prevalent ARG Cluster90s (CphA, MCR-7 and OXA-12) to Aeromonas spp. MAGs, highlighting this group of bacteria as geographically widespread AMR reservoirs. The Proteobacteria order and Aeromonas genus have both recently been shown to be core taxa in gut samples extracted from laboratory-reared Biomphalaria snails [58]; as well as signalling that snail faecal microbiomes are a reflection of the intestinal tract, our results indicate that these core features are preserved across various snail genera and within their natural habitats. Meanwhile, our hierarchical clustering and PCoA visualizations demonstrate the variation in numerous other bacterial taxa between snails of the same genus (Bulinus) observed between countries and between districts in the case of Malawi. The moderate diversity observed between proximal samples from Malawian Bulinus snails at the species level highlights the potential role of local environmental conditions, such as water pollution, climate and antibiotic exposure, in addition to host-specific characteristics, in shaping faecal microbiomes beyond the core taxa.

We recovered 8 high-quality MAGs and 21 medium-quality MAGs from 20 bacterial genera in total, yet 1,246 unique genera were detected in total by Kraken 2 read profiling, demonstrating the limitations of assembly-based approaches in complex samples such as these. It is therefore expected that fewer ARGs were detectable in MAGs, which are generally recovered from high-abundance organisms, than by read mapping or by screening all contigs, including those that could not be binned to MAGs. The inability to recover medium- to high-quality MAGs (with >50% completeness) in 1,226 of the 1,246 genera detected by read profiling (98.39%) suggests that some ARGs from low-abundance species are likely to have been undetectable even by our read-based approach, particularly in samples with lower sequencing depths. We specified that 60% template coverage was needed to call a hit to reduce the chance of false positives, and hence, the ARG profiles that we present here are likely biased towards those found in high-abundance genera such as Aeromonas. This idea is supported by our rarefaction analysis, in which varying sequencing depths were required to achieve ARG saturation. Nonetheless, ARGs associated with abundant bacterial taxa were commonly detected across the four samples with fewer than 10 million reads.

The detection of ARGs from the CphA, MCR-7 and OXA-12 Cluster90s in 66.7%, 66.7% and 93.3% of samples, respectively, all of which originate from Aeromonas species [59,60], is a reflection of the skew towards the detection of ARGs associated with abundant taxa. The mcr-7.1 colistin resistance gene was first identified on plasmids in K. pneumoniae isolates from chickens in China [61], but chromosome-borne variants of the gene have more recently been identified in Aeromonas isolated from poultry in the USA [62] and in global environmental sources (water) [63]. The suggestion that Aeromonas species are the natural reservoirs of mcr-3 and mcr-7 has since been proven by robust evolutionary models [53], and we were able to assemble and bin mcr-7-like variants to a putative Aeromonas jandaei MAG. Our results signpost to the potential for the emergence of new variants of these ARGs from freshwater gastropod faeces and, specifically, the Aeromonas reservoirs within.

Similar reservoirs of ARGs in low-abundance species may be more widespread than our data indicate. We demonstrate the presence of bla_OXA-181 and bla_OXA-547 in samples from Malawi and the UK, respectively, both of which are bla_OXA-48-like genes. In the case of bla_OXA-181, we could bin the >20 kbp contig containing the gene to a high-quality MAG of S. xiamenensis, the species known to be the gene’s progenitor [51]. Shewanella species have been identified as the progenitors of numerous bla_OXA-48 variants [64], and the enzymes coded for by these genes hydrolyse carbapenems and confer resistance to most beta-lactam, beta-lactamase inhibitor combinations. They emerged and proliferated over the last two decades to become the most prevalent carbapenemases among Enterobacterales in many parts of Europe, Northern Africa and the Middle East [46]. Consequently, they are a major public health concern, with carbapenemase-producing Enterobacterales (CPE) being listed by the World Health Organization as critical priority pathogens for which new antimicrobial agents are urgently needed [65]. While we cannot infer mobilization of bla_OXA-48- or mcr-like genes from environmental species to Enterobacterales in our study, the rapid dissemination of ARGs co-localized with IS elements among various species within the Enterobacterales often occurs as a result of one initial mobilization event, after which transposable units can move between replicons and cells [64,66].

To our knowledge, this is the first report of bla_OXA-181 in Malawi, a country where CPE are uncommon in the clinical setting, but extended-spectrum beta-lactamase (ESBL)-producing Enterobacterales resistant to a range of other beta-lactam antibiotics contribute to increased mortality in hospital patients [67,68]. ESBL-producing Enterobacterales have also been widely detected in water systems as well as human and animal faeces in Malawi [68,69]. Efforts to characterize the selective pressures behind the spread of ESBL-producing Enterobacterales in Malawian wastewater and freshwater are ongoing, but low environmental antimicrobial concentrations are known to drive the transfer and persistence of ARGs in water systems [55]. This selective pressure may also facilitate the emergence of bla_OXA-48-like genes from environmental reservoirs where environmental and human health intersect. Meanwhile, such reservoirs of carbapenem resistance genes could allow bacteria to react rapidly if carbapenems are introduced more routinely into clinical practice, limiting their efficacy. The emergence and spread of ARGs is often driven by horizontal gene transfer (HGT) associated with MGEs, such as conjugative plasmids, which regularly carry multiple IS elements [3,70]. We show that IS elements are abundant within snail faecal microbiomes. However, the suggestion of HGT is purely hypothetical in our context, without capturing widespread co-localization of ARGs on MGEs and performing functional assays. We could only demonstrate co-localization of ARGs and IS elements in two instances where sul1-containing gene clusters were identified upstream of IS6100 in samples from Malawi and Uganda. This is a common issue in short-read metagenomics due to the fragmented nature of assemblies [71].

Our study does have several limitations. For example, the use of pooled snail faeces for metagenomic analysis, while necessary to obtain sufficient biomass for DNA extraction and sequencing, limits the ability to assess individual snail variability and the potential heterogeneity in microbiome composition and ARG content in snails from the same location. Meanwhile, the use of short-read sequencing, although effective for taxonomic profiling and ARG identification, poses challenges in assembling complete genes and genomes, particularly for low-abundance species, potentially leading to an underestimation of the full richness of ARGs. This limitation is compounded by the inherent bias toward detecting ARGs in more abundant taxa which may skew the representation of ARGs in less abundant microbes. Ultra-deep sequencing may be required to detect such ARGs, even by read mapping methods that do not rely on metagenome assembly. Additionally, while read-based ARG detection approaches are widely used in the analysis of complex metagenomes to enable the detection of low-abundance ARGs in the absence of assembly [24,33], partial hits may represent truncated or non-functional genes. Inherently, short-read metagenome assemblies from diverse samples are highly fragmented. This limited our ability to explore co-localization of ARGs with MGEs, such as plasmids and composite transposons. The accuracy of, and access to, long-read metagenomic sequencing technologies is rapidly improving [71], and these technologies could be used in future metagenomic studies to address the issues associated with fragmented assemblies. Proximity ligation methods such as Hi-C could also assist by linking MGEs to host genomes [72], though this was not a feasible addition to our study due to limited sample biomass and the need for dedicated extraction protocols. Furthermore, while this study highlights the potential role of freshwater snails as AMR reservoirs, the cross-sectional nature of the sampling limits our ability to infer causality or the dynamics of ARG transmission over time. Longitudinal studies are needed to better understand temporal changes in ARG prevalence and the environmental factors influencing these patterns. Finally, while the study identifies ARGs predicted or known to be of clinical relevance, it does not explore the functional impact of these ARGs, nor does it confirm their potential for HGT to human pathogens, a critical future step to quantify the risks associated with our findings.

In conclusion, our study represents the first metagenomic analysis of AMR in freshwater gastropod faeces, providing important insights into the potential roles of these invertebrates as reservoirs of clinically relevant ARGs. Our findings have significant implications for AMR surveillance within One Health frameworks, particularly in areas where agricultural animals, wild animals and humans utilize the same open water sources. This reservoir of ARGs poses a threat to public health in an era of industrial-scale antibiotic use, particularly where access to clean water and sanitation is limited. Future research should build on these results through longitudinal sampling and by investigating how environmental factors, such as the presence and concentration of antibiotics, influence ARG richness and abundance in snail microbiomes. Meanwhile, incorporating long-read sequencing, proximity ligation methods and functional assays in future work, where feasible, will improve our understanding of ARG mobility. Additional sampling in nearby human and animal populations to map potential transmission of ARGs from snail-associated bacteria to human pathogens will be essential for designing targeted interventions to curb AMR spread from and into shared aquatic environments.

Abbreviations

AMR

antimicrobial resistance

ARG

antimicrobial resistance gene

CLR

centred log-ratio

CPE

carbapenemase-producing Enterobacterales

ESBL

extended-spectrum beta-lactamase

HGT

horizontal gene transfer

IS

insertion sequence

MAG

metagenome-assembled genome

MGE

mobile genetic element

PCoA

principal coordinate analysis

RPKM

reads per kilobase per million

SNP

single nucleotide polymorphism