Dissemination of antibiotic-resistant enterobacteriaceae from hospital wastewater in Gabon: implications for the one health strategy

Abstract

Introduction

The proliferation of Enterobacteriaceae exhibiting a multidrug-resistant phenotype in aquatic ecosystems is a major concern for global public health. In Gabon, the lack of adequate pre-treatment systems means that hospital wastewater acts as a critical reservoir associated with the spread of these pathogens into surface waters. This study aimed to characterise the bacterial load and phenotypic resistance profiles of Enterobacteriaceae isolated from the effluents of eleven Gabonese hospitals, whilst assessing the environmental factors associated with their presence.

Methodology

Between 2023 and 2024, 98 samples were collected (52 from hospital effluents and 46 from environmental waters). Physico-chemical parameters (pH, temperature) and visual indicators (color, debris, foam) were recorded. Following serial dilutions and cultivation on selective agar media, 113 Enterobacteriaceae isolates were identified using API 20E and VITEK^® 2. Susceptibility to 19 antibiotics was determined using the standardized phenotypic disk diffusion method (Kirby-Bauer).

Results

The bacterial load was significantly higher in hospital wastewater (up to 1.68 × 10⁸ CFU/mL) than in surface water. Univariate analysis revealed that the presence of debris (p = 0.003), foam (p = 0.016) and a greyish or brownish colouration of the water (p = 0.028) were visual indicators significantly markers of environmental pollution rather than confirmed causal factors. All confirmed Enterobacteriaceae isolates (n = 113) exhibited a multidrug-resistant phenotype, with rates of resistance to beta-lactams exceeding 95%. The genus Escherichia (n = 45) was predominant.

Conclusion

This observational study demonstrates that untreated hospital wastewater in Gabon is a significant source of microbiological pollution. Environmental visual indicators are useful for identifying areas with high bacterial loads. It is necessary to develop pre-treatment infrastructure to mitigate the associated environmental and health risks.

Introduction

Aquatic ecosystems, subject to significant anthropogenic pressures, are dynamic environments conducive to the evolution and diversification of microbial communities [1]. Nevertheless, these aquatic ecosystems are increasingly exposed to anthropogenic sources of pollution, creating selective pressures that drive the emergence and dissemination of antimicrobial-resistant pathogens [2]. This pollution mainly results from wastewater, which originates from multiple sources of pollution such as domestic, industrial, agricultural and, more particularly, hospital activities [3]. The latter are characterised by high water consumption and intensive use of antimicrobials. In this context, hospitals appear as significant sources of pharmaceutical residue discharges, particularly antibiotics, which are among the most commonly used chemical substances in healthcare settings [4]. The extensive and diversified use of these compounds therefore promotes the formation of major reservoirs of pathogenic bacteria and antibiotic resistance genes [5]. Therefore, hospital wastewater should be adequately treated, and antibiotic use should be carefully monitored [6]. Indeed, hospital wastewater can widely spread two types of emerging pollutants, namely antibiotic residues and antibiotic-resistant bacteria [7]. In addition to high bacterial loads and a nutrient-rich environment, hospital effluents may contain sub-therapeutic concentrations of antibiotic agents, which can facilitate the emergence and dissemination of resistance genes from one bacterium to another [8]. Hospital effluents may pose a threat to public health if they are inadequately treated and discharged [9].

In several African countries, hospital effluents are discharged into the environment without treatment due to the absence of dedicated pretreatment facilities. In countries where such infrastructures do exist, clear regulatory frameworks to ensure the quality of treated wastewater are often lacking [10]. A study conducted in Mbarara, Uganda, found that effluent from a lagooning treatment plant still released a high proportion of antibiotic-resistant bacteria, indicating that the treatment was defective [10]. This in turn facilitates the spread of antibiotic-resistant microorganisms and their corresponding resistance genes in soil, groundwater and aquatic environments [11].

Untreated hospital effluent discharged into sewerage systems or watercourses carries a high bacterial load [12]. In the absence of molecular characterisation via sequencing or plasmid analysis, the simultaneous presence of high bacterial concentrations and antibiotic residues suggests an environmental context conducive to the persistence and co-selection of resistant strains [13, 14–15]. The co-occurrence of multiple classes of antibiotics further exacerbates this phenomenon, promoting ecological conditions that may favour the persistence and co-selection of resistant bacterial populations in the environment [16, 17–18].

Like many African countries, Gabon lacks pre-treatment stations for hospital wastewater, and effluents from healthcare facilities are discharged directly into the environment. In this context, it appears essential to evaluate and characterize antimicrobial resistance in hospital effluents from Gabonese healthcare facilities in order to estimate the risk of spread of multidrug-resistant bacteria. This study aims to identify bacteria of interest present in hospital wastewater, analyze the factors associated with their presence in these environments, and characterize their phenotypic antibiotic resistance profiles. It also aims to assess the potential risks of contamination of aquatic environments within a conceptual One Health perspective linking healthcare facilities and aquatic ecosystems, highlighting how environmental co-exposure may contribute to maintaining and spreading phenotypic resistance.

We hypothesized that untreated hospital wastewater in Gabon constitutes a significant reservoir of multidrug-resistant Enterobacteriaceae and contributes to the microbiological contamination of nearby aquatic environments.

Materials and methods

Management and analysis of hospital effluents in Gabon

The healthcare facilities (n = 11) included in this study are located across five of the country’s nine provinces. In several university hospital centers (CHUs), including Libreville University Hospital (CHUL) (Libreville, Estuaire), the Mother and Child University Hospital Center Jeanne Ebori Foundation (CHUJE) (Libreville, Estuaire), and the Amissa BONGO University Hospital Center (CHUAB) (Franceville, Haut-Ogooué), wastewater undergoes a dual filtration process before being discharged into receiving environments, such as surface water bodies or the groundwater table. In contrast, wastewater generated by other healthcare facilities namely the Sino-Gabonese Friendship Hospitals (HCSG in Libreville and HASG in Franceville), Regional Hospital Centers (George RAWIRI Regional Hospital Center (CHRGR) in Lambaréné, Tchibanga Regional Hospital Center (CHRT) in Tchibanga, and Port-Gentil Regional Hospital Center (CHRPG) in Port-Gentil), as well as several medical centers (Nzeng-Ayong Health Center (CSNA) in Libreville, Franceville Urban Health Center (CSUF) in Franceville, and Moanda Medical Center (CMM) in Moanda) receives no mechanical or biological treatment. (Fig. 1).

Fig. 3.

Frequency distribution of the five dominant bacterial genera.The bar chart displays the total number of isolates for each of the most prevalent genera recovered from wastewater samples in Gabon. Escherichia emerges as the most frequently identified genus, followed by Enterobacter and Klebsiella. The total number of isolates included in this analysis is 113

Fig. 1.

Geospatial distribution of Enterobacteriaceae sampling sites across Gabon. Map detailing the geographical locations of the surveyed healthcare facilities. Each sampling site is represented by a colored circle, where the marker area is directly proportional to the total number of samples collected (n = 98). The colors correspond to the categorical palette established in the multivariate analyses, and the five sampled provinces are highlighted in green

Site sampling and methodology

Hospital wastewater was sampled at the point of generation, directly from the hospital sewer sumps (before reaching the main sewage network). Wastewater samples were collected directly from the manholes and sumps of each healthcare unit, as well as prior to their discharge into the municipal sanitation network. Sampling continued in the streams located near the healthcare facilities. The samples were collected over a six (6)-month period, divided into three sampling campaigns, regardless of the research seasons. Three sampling campaigns were conducted between 2023 and 2024 to cover different hospital settings in several regions of Gabon. The first campaign took place from July 31 to September 30, 2023, in the Haut-Ogooué province (Franceville and Moanda) at the CHUAB, HASG, CSUF, and CMM. The second campaign was carried out from January 12 to March 12, 2024, in the Estuaire province (Libreville), including the CHUL, CHUJE, HCSG, and CSNA. Finally, the third campaign was held from June 22 to August 25, 2024, in the Moyen-Ogooué, Nyanga, and Ogooué-Maritime provinces (Lambaréné, Tchibanga, and Port-Gentil), covering the CHRGR, CHRT, and CHRP, respectively.

The samples were collected in sterile glass containers (1,000 mL) at a depth of approximately 30 cm below the water surface. Sampling was performed at the same time (between 8 am and 10 am) of day, as effluent discharge peaks could not be assessed continuously due to the lack of 24-hour automatic samplers at the various sites, but did not account for variations in hospital effluent flow rates. A total of 98 samples were collected during the analysis period: 46 environmental water samples (40 from watercourses and 6 form rivers) and 52 hospital department wastewater samples (14 from maternity, 8 from pediatrics, 10 from internal medicine department, 2 from hemodialysis, 8 from surgery and operating rooms, and 2 from intensive care and emergency units) (Table 1). The samples were transported to the laboratory in insulated containers at a temperature of 4 °C and analyzed within two hours at partner laboratories (CIRMF, IRET, and healthcare facilities).

Table 1.

Sampling strategy in different healthcare facilities

Provinces	Institutions (n°)	Wastewater sample from healthcare facilities (n°)	Environmental water samples (n°)	Treated / untreated wastewater
Etuaire	CHUL (11)	Maternity (2), pediatric (1), Internal Medicine (1), surgery (1), operating theatre (1), resuscitation (1)	watercourse (4)	Treated
	CHUJE (9)	Maternity (1), pediatric (1), surgery (1), surgery (1), operating theatre (1)	watercourse (4)	Treated
	HCSG (8)	Maternity (1), pediatric (1), Internal Medicine (1), operating theatre (1)	watercourse (3); river (1)	Untreated
	CSNA (7)	Maternity (1), physician (1), surgery (1)	watercourse (2); river (2)	Untreated
H-Ogooué	CHUAB (13)	Maternity (1), pediatric (1), Internal Medicine (1), haemodialysis (2), surgery (1), operating theatre (1), resuscitation (1), emergency (1)	watercourse (5)	Treated
	HASG (11)	Maternity (1), pediatric (1), Internal Medicine (1), surgery (1), operating (1), operating theatre (1)	watercourse (4); river (1)	Untreated
	CMM (6)	Maternity (1), pediatric (1), Internal Medicine (1)	watercourse (3)	Untreated
	CSUF (8)	Maternity (1), pediatric (1), Internal Medicine (1), surgery (1)	watercourse (4)	Untreated
M-Ogooué	CHRGR (7)	Maternity (1), Internal Medicine (1), operating theatre (1)	watercourse (3); river (1)	Untreated
Nyanga	CHRT (7)	Maternity (1), Internal Medicine (1), operating theatre (1)	watercourse (4)	Untreated
O-Maritime	CHRP (11)	Maternity (1), pediatric (1), Internal Medicine (1), surgery (1), operating theatre (1), emergency (1)	watercourse (4); river (1)	Untreated

(n°) number of samples collected from various departments

Determination of physicochemical parameters

The physicochemical parameters evaluated for all environmental and wastewater samples included temperature (°C) and pH. The pH measurement was performed using pH paper strips (filtraTECH, Saint Jean de Braye, France), providing an accuracy of 1.0 pH units. The temperature was determined using a standard THIES thermometer (BLET Measurement Group, Saint-Germain-en-Laye, France), ensuring a reliable and precise measurement. These analyses allowed for a rigorous characterization of the physicochemical conditions of the studied samples.

To document real-time environmental variations at the time of collection, continuous monitoring of flow rate, temperature, pH, and the presence of algae, insects, and waste was observed and measured directly in the field. Measurements were taken at four (4) minute intervals for the flow rate, and one (1) minute intervals for temperature, thereby ensuring a detailed and dynamic description of the environmental conditions influencing the samples’ characteristics.

Serial dilution of samples from study sites

Dilutions were performed as described: in a cryotube containing 9 ml of sterile water, 1 ml of the sample was added. The mixture was thoroughly mixed. This represented the 10⁻¹ dilution. From the 10⁻¹ dilution, 1 ml was added to 9 ml of sterile water and carefully mixed. This represented the 10⁻² dilution. By repeating the same procedure, the dilution was carried out up to 10⁻⁵. This procedure was performed in a laminar flow cabinet. Subsequently, 100 µL of the 10⁻¹, 10⁻², 10⁻³, 10⁻⁴ et 10⁻⁵ dilutions were spread onto MacConkey agar (BioMérieux, Marcy-l’Étoile, France). The inoculated Petri dishes were incubated for 24 h at 37 °C. The resulting colonies were counted and expressed as colony-forming units (CFU) per 1 mL of wastewater, then recorded in a database. The enumerated isolates were stored in a mixture containing 30% glycerol and 70% PBS at − 20 °C until further use.

Isolation and Identification of Enterobacteriaceae from hospital wastewater samples

A total of 98 samples were inoculated using the spread-plate method. This technique consists of depositing a defined volume (0.1 mL) of the sample onto the surface of a solidified agar medium (MacConkey), then spreading it uniformly using a sterile spreader to ensure a homogenous distribution of bacterial cells across the agar surface. The inoculated plates were then incubated under aerobic conditions for 24 h at 37 °C. For plates displaying multiple colony morphologies, a second inoculation was performed to obtain isolated colonies, which were incubated at 37 °C for 24 h. The resulting colonies were initially identified (biochemical testing) via Gram staining, catalase test, and oxidase test. In total, 158 tested strains were provisionally classified as Gram-negative bacteria. For optimal preservation, representative isolates were stored in duplicate for subsequent analyses: one set for identification testing and the other at − 80 °C, both in a solution composed of 70% phosphate-buffered saline (PBS) and 30% glycerol.

After incubation, distinct colonies after inoculation and biochemical testing, were collected and re-inoculated onto fresh plates following the same procedure, then incubated under identical conditions to obtain pure cultures. The purified colonies were subsequently subjected to comprehensive biochemical identification. Colonies exhibiting characteristics of Enterobacteriaceae were identified using the API 20E system (BioMérieux, Marcy-l’Etoile, France) and the automated VITEK^® 2 Compact 15 system (BioMérieux, Marcy-l’Etoile, France) (it was used solely to resolve ambiguities in the identification of certain species), allowing precise and rapid species-level identification as decribed by [19].

Evaluation of antibiotic resistance in isolated enterobacteriaceae

All Enterobacteriaceae isolates were evaluated for their resistance to various antibiotic agents using the disk diffusion method (Kirby-Bauer) on Mueller-Hinton (MH) agar (BioMérieux, France), in accordance with the guidelines of the Clinical and Laboratory Standards Institute (CLSI) [20]. A panel of 19 antibiotics belonging to eleven distinct classes was used. The antibiotics tested included amoxicillin-clavulanic acid (AML, 30 µg), ampicillin (AMP, 10 µg), aztreonam (ATM, 30 µg), amoxicillin (AMC), tetracycline (TE, 30 µg), amikacin (AK, 30 µg), gentamicin (GEN, 10 µg), ceftazidime (CAZ, 30 µg), cefepime (FEP, 30 µg), cefoxitin (FOX, 30 µg), cefotaxime (CTX, 30 µg), imipenem (IMI, 10 µg), ertapenem (ETP, 10 µg), meropenem (MRP, 10 µg), ofloxacin (OFX, 5 µg), nalidixic acid (NA, 30 µg), fosfomycin (FOS, 200 µg), ticarcillin/clavulanic acid (TTC, 85 µg), and nitrofurantoin (F, 300 µg). In brief, MH agar plates were inoculated with a standardized bacterial suspension (0.5 McFarland) prepared from 24-hour primary cultures of each isolate. The agar plates were then incubated at 37 °C for 24 h. The diameters of inhibition zones were measured in accordance with CLSI, 2023 guidelines after 24 h of incubation at 37 °C. All tested isolates were classified as either “S” (susceptible) or “R” (resistant) and recorded in a database for further analyses. Isolates classified as ‘I’ (intermediate) have been included in “R”, as required by CLSI.

Data analysis

All data preprocessing, statistical analyses, and visualizations were performed using the Python programming language (v3.12.2) [21] within a Jupyter Notebook environment [22] Key libraries used include pandas (v2.2.3) [23] for data manipulation, GeoPandas (v0.14.4) [24] for geospatial analysis, SciPy (v1.13.1) [25] for statistical computations, and Matplotlib (v3.9.0) [26] with Seaborn (v0.13.2) [27] for figure generation. Additional specialized libraries included scikit-bio for multivariate distance-based statistics, prince and scikit-learn for ordination techniques, and statsmodels for multiple testing corrections. The antibiotic susceptibility test results were analyzed using Microsoft Excel 2010.

To ensure mathematical stability across all multivariate models, the dataset was strictly harmonized at the individual isolate level (n = 113), excluding any profiles with missing resistance values. The Cumulative Resistance Score (CRS) was calculated for each isolate, and global differences across multi-level spatial or taxonomic factors were assessed using the non-parametric Kruskal-Wallis test.

Methodological Rationale for Fig. 8: in order to assess the phenotypic association dynamics between the various antibiotic molecules within our harmonised environmental cohort (n = 113 isolates), a cross-correlation matrix was generated.

Fig. 8.

Hierarchical clustering heatmap of phenotypic antibiotic susceptibility profiles among the 113 Enterobacteriaceae isolates. Dendrogram structures are based on a Jaccard dissimilarity matrix calculated from binary susceptibility data (Red = Resistant/Intermediate; Light yellow = Susceptible). The vertical left bar identifies the facility of origin, and boxes outline the phenotypic clades A, B, and C

As the input phenotypic data are strictly dichotomous (1 = Resistant, 0 = Susceptible), application of the correlation equation inherently calculates the Phi coefficient (ɸ). This non-parametric metric is the mathematical standard for measuring the strength of the association between two binary variables. The scale was strictly limited to 0 to 1 (or -1 to 1, including negative correlations) to prevent any visual overinterpretation. According to statistical standards, a value of ɸ ≥ 0.7 indicates a strong association, whilst a value between 0.4 and 0.7 indicates a moderate association.

Terminology and standard definitions

For the purpose of scientific coherence throughout this manuscript, the following standardized terms are defined and applied:

1. Hospital wastewater (or hospital effluents): Refers strictly to liquid waste generated by healthcare facilities, collected at the point of generation or from sewer sumps prior to any environmental discharge.

2. Enterobacteriaceae: Refers to the specific bacterial family confirmed via Gram staining, biochemical testing (catalase, oxidase), API 20E, and VITEK^® 2 automated systems.

3. Phenotypic resistance: Refers to the antibiotic resistance profiles determined exclusively by structural growth inhibition zone diameters using the standardized disk diffusion method, without extrapolation to underlying genetic mechanisms.

4. Cumulative Resistance Score (CRS): Defined as the absolute number of distinct antimicrobial classes to which a single bacterial isolate exhibits phenotypic non-susceptibility (resistant or intermediate categories).

5. Multidrug-resistance (MDR) phenotype: Defined strictly in accordance with international standards (Magiorakos et al., 2012) as phenotypic non-susceptibility to at least one agent in three or more antimicrobial chemical classes.

Results

Environmental factors and bacterial load

Univariate analyses correlating field-measured environmental variables with the total bacterial load (log₁₀ CFU/mL) identified environmental variables markers of environmental pollution rather than confirmed causal factors within the effluents (Fig. 2).

Fig. 2.

Distribution of total bacterial load (log₁₀ CFU/mL) stratified by environmental parameters for the dominant enteric genera (n–e = 113). Panels (a) represent binary field observations, and panels (f–h) display categorical environmental characteristics. P-values are derived from non-parametric Mann-Whitney U tests or Kruskal-Wallis tests

Statistical analysis shows that the presence of anthropogenic solid waste (p = 0.003, Fig. 2b) and foam (p = 0.016, Fig. 2e), as well as the greyish colour of the water (p = 0.028, Fig. 2g), are visual indicators significantly associated with variations in bacterial load, acting as markers of environmental pollution rather than confirmed causal factors of bacterial proliferation. Conversely, the presence of algae was significantly associated with a reduction in the bacterial load (p = 0.046, Fig. 2d), potentially due to biological competition or algal antimicrobial exudates. Total rainfall within the last 24 h (p = 0.475, Fig. 2a) and effluent pH (p = 0.735, Fig. 2h) did not significantly alter the recovered bacterial loads.

Isolation and identification of presumptive Enterobacteriaceae, enumeration of isolates

The concentrations of Escherichia coli, Klebsiella spp., Salmonella spp., Serratia spp., and Enterobacter spp. in hospital effluent samples ranged from 1.6 × 10³ to 1.68 × 10⁸ CFU/mL, whereas those observed in environmental water samples varied between 1.3 × 10⁴ and 1.55 × 10⁶ CFU/mL. Overall, bacterial loads were markedly higher in hospital effluents than in surface water samples. Of the 158 presumptive Gram-negative bacterial isolates recovered from hospital effluents and surface waters, 113 isolates (71.52%) were confirmed as members of the Enterobacteriaceae family and 45 isolates (28.48%) were identified as non-Enterobacteriaceae. Figure 3 presents the distribution of these isolates using a bar chart, with bars representing the number of isolates per genus and arranged in decreasing order to highlight the most prevalent genera. The results indicate that Escherichia was the most frequently isolated genus, accounting for approximately 45 isolates more than twice the number observed for any other genus. Klebsiella and Enterobacter showed comparable frequencies, with 20 and 19 isolates, respectively, while Serratia accounted for about 14 isolates. Salmonella was the least represented genus, with 15 isolates.

Distribution of cumulative resistance scores (CRS) among bacterial genera isolated from hospital effluents

Geospatial analysis of the 113 environmental isolates from the five dominant bacterial genera reveals a high degree of heterogeneity in phenotypic resistance burden across Gabon (Fig. 4).

Fig. 4.

Spatial distribution of the mean Cumulative Resistance Score (CRS) among Enterobacteriaceae isolates across the sampled locations in Gabon. Marker size indicates the number of isolates recovered at each healthcare facility, and marker color corresponds to the calculated mean CRS value for each site

Sampling disparity: The Franceville site (Haut-Ogooué) has the highest density of isolates (close to 50), followed by Libreville (Estuaire) and the CHRP (Ogooué-Maritime). The sites in Moyen-Ogooué (CHRGR) and Nyanga (CHRT) show lower representativeness. Resistance hotspot in Franceville: The mean cumulative resistance score (Mean CRS) peaks notably in Franceville, characterised by the highest score in the study (Mean CRS > 5.75, dark red zone). Intermediate and moderate profiles: Intermediate levels of resistance (Mean CRS 4.50–5.00) are observed at the CHRP and CHRT. Conversely, the most moderate levels (Mean CRS ~ 4.00) are concentrated in Libreville and at the CHRGR.

Cumulative resistance scores (CRS) per genera

The Cumulative Resistance Score (CRS), measuring the absolute number of resistant classes per isolate, was systematically evaluated across multiple epidemiological strata (Figs. 5 and 6). The global median CRS for the entire Gabonese cohort was 5 resistant classes (Fig. 6).

Fig. 5.

Distribution of Cumulative Resistance Scores (CRS) across dominant bacterial genera. Boxplots illustrate the distribution of the CRS(absolute number of antibiotic classes for which an isolate exhibits phenotypic resistance) among confirmed Enterobacteriaceae (n = 113). The genera are ordered on the y-axis according to their median CRS, with the associated sample size (n). The overall variance was assessed using a non-parametric Kruskal-Wallis test (p = 0.837)

Fig. 6.

Distribution of Cumulative Resistance Scores (CRS) across epidemiological and taxonomic strata (n = 113). Boxplots illustrate the distribution of CRS values according to (a) environmental source, (b) province, (c) healthcare facility, (d) hospital ward, and (e) bacterial genus. The dashed horizontal line represents the overall cohort mean CRS

Kruskal-Wallis tests demonstrated no statistically significant differences in CRS when stratified by: Bacterial Genus (p = 0.837, Fig. 5/6e), with Escherichia, Klebsiella, Enterobacter, Salmonella, and Serratia all displaying nearly identical median CRS values ranging from 5 to 6. Environment vs. Hospital setting (p = 0.475, Fig. 6a). Province (p = 0.174, Fig. 6b). Specific Hospital (p = 0.125, Fig. 6c). Hospital Service/Ward (p = 0.646, Fig. 8d). This lack of statistical significance indicates a homogeneous distribution of cumulative resistance scores across the sampled institutional and taxonomic categories.

Antibiotic resistance profile by institution and geographical area

The harmonized cohort analysis underscores an high baseline prevalence of antimicrobial resistance (AMR) across Gabon (Fig. 7). At the provincial level, the Estuaire and Haut-Ogooué provinces exhibit the highest cumulative resistance rates.

Fig. 7.

Spatial distribution of phenotypic antibiotic resistance prevalence across healthcare facilities and environmental reservoirs. This heatmap displays the percentage of phenotypic resistance for each tested antibiotic across the sampled geographical locations, derived from the cohort of dominant bacterial isolates (n = 113). Healthcare facilities yielding only a solitary isolate (CHRM, n = 1) are excluded from this spatial comparison. Sampling locations (y-axis) are grouped by their respective provinces, with sample sizes (n) indicated in parentheses. The aggregated non-hospital environmental baseline (Communal) and the overall national average (Global Gabon) are positioned at the base of the matrix. The color gradient reflects resistance prevalence ranging from 0% (white) to 100% (dark red). Antibiotics (x-axis) are color-coded by therapeutic class

Individually, healthcare facilities such as CHUAB (n = 23), HASG (n = 16), and CHUJE (n = 10) exhibit high resistance rates, particularly against first-line bêta-lactams (AMP and AML reaching up to 100% resistance in multiple centers). Phenotypic resistance to third- and fourth-generation cephalosporins (CTX, CAZ, FEP) routinely exceeds 50% across most facilities. Furthermore, carbapenem resistance (specifically ETP) is already well-established, reaching 91% at CHUAB and 94% at HASG, pointing toward a reduced therapeutic options.

Hierarchical classification and limits of the MDR standard

The unsupervised topological analysis (Fig. 8) initially identified three distinct epidemiological sub-populations (Clades A, B, and C) sharing similar resistance profiles, independent of any a priori geographical or taxonomic considerations.

Clade A represents an extended multidrug-resistance phénotypic profile, characterized by non-susceptibility to a high number of tested antibiotic classes, including last-resort carbapenems (ETP, IMI, and MRP) and advanced cephalosporins. Clade B displays a restricted phenotypic resistance profile, maintaining susceptibility to aminoglycosides and critical beta-lactams. Clade C exhibits intermediate-to-high multi-drug resistance (MDR) phenotypes, with marked resistance toward fosfomycin (FOS) and amikacin (AK). Crucially, the dendrogram alignment demonstrates that these clades do not segregate by healthcare facility or geography; isolates from CHUAB, HASG, or CMM are interspersed across all three clades, highlighting a systemic dissemination of uniform MDR patterns within the regional sanitation infrastructure.

To clinically characterize these clades, we applied the international standard for multidrug resistance (MDR, defined as non-susceptibility to at least three antimicrobial classes according to Magiorakos et al., 2012). This approach reveals a systemic ceiling effect within this environmental cohort: the baseline burden is extreme, with over 90% of the isolates from the average cohort classified as MDR.

Consequently, although the three clades identified in Fig. 8 display an MDR prevalence of 100%, Monte Carlo simulations indicate that this rate no longer differs significantly from the background leve (Clade A: p = 0.102; Clade B: p = 0.626; Clade C: p = 0.348).

Multiple correspondence analysis (MCA) of phenotypic resistance profiles

The Multiple Correspondence Analysis (MCA) successfully mapped the global phenotypic resistance space of the bacterial isolates, with Dimension 1 accounting for the vast majority of the total variance (81.6%) and Dimension 2 explaining 18.4% (Fig. 9a and b). The spatial distribution of isolates on Dimension 1 was primarily driven by resistance to fosfomycin (FOS), ertapenem (ETP), and aztreonam (ATM) (Fig. 9c), whereas Dimension 2 was predominantly influenced by amoxicillin/clavulanic acid (AMC), fosfomycin (FOS), and cefepime (FEP) (Fig. 9d).

Fig. 9.

Multiple Correspondence Analysis (MCA) of phenotypic antibiotic resistance profiles (n = 113). Panels (a) and (b) project individual isolates and group centroids categorized by healthcare facility and bacterial genus, respectively. Panels (c) and (d) display the individual coordinates and contributions of each antibiotic resistance trait to Dimensions 1 and 2

Permutational Multivariate Analysis of Variance (PERMANOVA) revealed highly significant differences in resistance profiles across different healthcare facilities (p = 0.001, Fig. 9a). However, a significant PERMDISP result (p = 0.005) indicates substantial variations in multivariate dispersion among hospitals, reflecting highly heterogeneous institutional resistance pressures. Conversely, when looking at taxonomy, no significant clustering was observed among the major bacterial genera (PERMANOVA: p = 0.124; PERMDISP: p = 0.953, Fig. 9b), suggesting that phenotypic resistance profiles are shaped by environmental and institutional selective pressures rather than taxonomic boundaries.

Projection and Spatial Characterisation of Evolutionary Trajectories (Clades)

To link these findings, the simultaneous projection of hierarchical clades (from Fig. 5) and resistance vectors onto the MCA space (Fig. 9) visually and mathematically validates the existence of three divergent ecological trajectories. Post-hoc statistics are consistent with absolute segregation (FDR p = 0.001 for all inter-clade comparisons and against the background cohort). The mapping of these phenotypes (Fig. 6) allows for the specification of their territorial anchoring.

Clade A (High-Level Multiclass Resistance Profile): This cluster (n = 31) represents the widespread resistance profile. Projected in the lower-left quadrant of the MCA (Fig. 10), it is defined by the accumulation of carbapenems and quinolones. Although it forms a cohesive structure (intra-clade Jaccard distance = 0.365) and is statistically divergent (Separation Ratio = 1.55; Pseudo-F = 8.2), its spatial distribution is ubiquitous (Fig. 9). It is not significantly linked to any specific hospital (Fisher p > 0.05). Found simultaneously at the CHUAB (8 isolates), the CSNA (4 isolates), and in the effluents of the Communal Hospital (5 isolates), t he widespread distribution of this clade across multiple healthcare facilities and environmental sites suggests that highly multidrug-resistant Enterobacteriaceae may be circulating beyond hospital boundaries and extending into surrounding urban aquatic environments.

Fig. 10.

Integration of hierarchical phenotypic clades onto the Multiple Correspondence Analysis (MCA) multidimensional space. The biplot projects the environmental isolates (n = 113) onto a two-dimensional space defined by their phenotypic antibiotic susceptibility profiles. Benzécri’s correction was applied to the eigenvalues to adjust for variance underestimation, with the two main dimensions accounting for 100% of the corrected inertia (Dimension 1: 81.6%). Individual isolates (dots) are color-coded according to the distinct categories (Clades A, B, and C) identified via hierarchical clustering. Grey vectors display the coordinates of the resistant modality for each tested antibiotic. Spatial separation among the three clades was evaluated using PERMANOVA and PERMDISP on a Jaccard distance matrix (p = 0.001, 999 permutations), with subsequent FDR-adjusted pairwise comparisons showing statistically significant differences between individual clades (p_adj = 0.001)

Clade B (Susceptible/Alternative Escape): Restricted in size (n = 9), this clade is fundamental to understanding the dynamics of the environment. Isolated at the extreme right of Dimension 1 (Fig. 10), it stands in diametric opposition to the major resistance vectors. It represents an escape subpopulation that has maintained a largely susceptible phenotype to critical antibiotics in the face of ambient hyper-resistance. This clade, slightly more dispersed (intra-clade distance = 0.504; Ratio = 1.31; Pseudo-F = 7.2), displays a marked geographical identity on the map (Fig. 9): more than 55% of its isolates (5/9) originate from the Estuaire province (Libreville).

Clade C (FOS/AK-driven/ Localized resistance cluster): This clade (n = 20) is the major epidemiological discovery of this study. Pulled along Dimension 2 (Fig. 10) by last-resort molecules (Fosfomycin, Amikacin, Aztreonam), it displays the most extreme topological divergence (Separation Ratio = 1.68; Pseudo-F = 9.7) and remarkable phenotypic homogeneity (intra-clade distance = 0.342). Shared across several bacterial genera, this profile is not ubiquitous; rather, it is the result of a focused selection pressure. As illustrated by the spatial distribution (Fig. 10), nearly 85% of these isolates are located in the Haut-Ogooué province, and more specifically within the CHUAB of Franceville (60% of the clade). Fisher’s exact test confirms a massive and highly significant enrichment for this hospital: bacteria from this clade are 11 times more likely to originate from CHUAB effluents than from the rest of the sampled sites (Odds Ratio = 11.2, Fisher p < 0.001).

Co-resistance correlation matrix among antibiotic classes isolated from hospital effluents

The exhaustive quantitative analysis of the matrix (Fig. 11) reveals a complex structure of the phenotypic burden. Unlike clonal clinical profiles where resistance is often acquired in rigid “blocks” (generating ɸ > 0.8), the Gabonese hospital environment exhibits an extremely diversified and “chaotic” co-resistance network.

Fig. 11.

Phenotypic co-resistance matrix across dominant bacterial genera. This heatmap illustrates the pairwise associations of phenotypic resistance between tested antibiotics within the harmonized cohort (n = 113). Because the input data are strictly binary (1 = Resistant, 0 = Susceptible), the mathematical association is expressed using the Phi coefficient (ɸ), anchored to a scale from 0 (no association, white) to 1 (perfect positive association, dark red). To ensure analytical clarity and prevent visual inflation of weak signals, exact ɸ values are annotated within the lower triangular matrix. Notably, the entire network lacks strong correlations (ɸ ≥ 0.7). The strongest observed signals represent moderate cross-class associations such as the strongest observed association involved fosfomycin (FOS) and ertapenem (ETP) (ɸ = 0.48) reflecting the complex and deeply interconnected co-selection pressures within these environmental reservoirs. Antibiotic labels are color-coded by therapeutic class

Absence of strong associations and intra-class decoupling

The maximum correlation measured across the entire panel is ɸ = 0.468 (between Cefepime (FEP) and Amoxicillin/Clavulanic Acid (AMC)). Even more remarkable, intra-class associations are weak. For instance, within the carbapenem class, the co-resistance between Ertapenem (ETP) and Imipenem (IMI) is only ɸ = 0.128. This decoupling demonstrates that environmental selection pressure generates a highly heterogeneous mosaic of phenotypes rather than the simple dissemination of a uniform isolate resistant to all antibiotics tested.

The anchoring of fosfomycin to critical resistance traits

Examination of the data highlights the particular behavior of Fosfomycin (FOS). Far from evolving independently from the rest of the panel, the acquisition of FOS resistance displays significant levels of association with last-resort molecules. The FOS vs. ETP pair (ɸ = 0.48) constitutes one of the highest correlations within the cohort. FOS is also moderately correlated with Aztreonam (ATM, ɸ = 0.424) and Amikacin (AK, ɸ = 0.334).

The Phi coefficient (ɸ) matrix identified several significant, positive phenotypic co-resistance correlations within the wastewater isolates (Fig. 11). The strongest positive co-resistance links were observed between ceftazidime (CAZ) and aztreonam (ATM) (ɸ = 0.46), as well as between cefepime (FEP) and amoxicillin/clavulanic acid (AMC) (ɸ = 0.45).

Crucially, fosfomycin (FOS) resistance was strongly coupled with ertapenem (ETP) resistance (ɸ = 0.42) and amikacin (AK) resistance (ɸ = 0.33). Similarly, tetracycline (TE) resistance showed a strong co-resistance pattern with ETP (ɸ = 0.46).

Discussion

Microbiological burden and the impact of the absence of hospital effluent pretreatment

Quantitative analysis of the samples in this study reveals substantial microbiological contamination of hospital effluents in Gabon, with bacterial loads reaching up to 1.68 × 10⁸ CFU/mL. This substantial concentration highlights the critical absence of functional pretreatment infrastructure within Gabonese healthcare facilities, a finding that is unfortunately recurrent across many regions of sub-Saharan Africa. Compared to environmental surface waters, the load in hospital effluents is significantly higher, confirming that healthcare centers act as important sources of bacterial pollution.

These results corroborate the observations of Chonova et al. [28] and Katale et al. [29], who demonstrated that untreated hospital wastewater releases microbiological densities far exceeding tolerable environmental thresholds, rapidly saturating the self-purification capacities of receiving water bodies. Upstream, the absence of specific physical or biological barriers within hospital sewage systems not only promotes the persistence of this biomass but also transforms these niches into high-density cellular incubators [30]. In the Gabonese context, the direct discharge of these effluents into the municipal network or adjacent riverbeds poses an potential epidemiological risk to peripheral human and animal populations reliant on these water resources, thus highlighting issues that are relevant to the broader One Health framework.

Visual and environmental determinants associated with bacterial dynamics

One of the original findings of our study lies in the identification of robust statistical correlations between certain macroscopic visual indicators (presence of solid debris, p = 0.003; presence of foam, p = 0.016; gray or brown discoloration, p = 0.028) and variations in enterobacterial load. The positive association between bacterial load and the presence of solid debris or foam may reflect environmental conditions favourable to bacterial persistence. However, the observational design of the study does not allow causal relationships to be established. These macromolecular aggregates may reflect environmental conditions associated with bacterial persistence and biofilm development. However, the observational design of this study does not allow causal relationships to be established by Stalder et al. [31]. Conversely, the lack of significant influence of pH and rainfall on the overall bacterial load in Gabon reflects the relative stability of the ecological niche of hospital effluents, where the continuous influx of anthropogenic nutrients overrides minor seasonal or physicochemical variations. This observation deviates from certain West African models where seasonality dictates bacterial dynamics [32], highlighting here that internal hospital emission pressure supplants natural dilution factors. The utilization of these visual markers by health authorities could provide a low-cost, rapid screening tool to identify high-risk epidemiological sedimentation and exposure zones. In our study, the use of pH paper was a practical constraint associated with remote sampling at 11 sites across the country.

Homogeneity of the cumulative resistance score (CRS) and environmental saturation (“Ceiling Effect”)

The phenotypic antibiotic resistance profiles of the 113 isolates in our cohort reveal an extremely severe situation: 100% of the enterobacterial isolates, dominated by the genera Escherichia (n = 45) and Klebsiella (n = 20), exhibit a multidrug-resistant (MDR) phenotype. The application of non-parametric Kruskal-Wallis tests to the Cumulative Resistance Score (CRS) revealed a total absence of statistically significant differences when data were stratified by bacterial genus (p = 0,837), province (p = 0,174), or clinical service (p = 0,646). Geospatial analysis reveals heterogeneity in mean CRS scores at the national level. The highest mean CRS value was recorded in facilities located in Franceville (Haut-Ogooué) (mean CRS > 5.75), whilst intermediate to moderate values were observed in Libreville and Lambaréné. This ubiquitous homogeneity results in an epidemiological ceiling effect. This indicates that the resistance burden has reached an widespread occurrence stage within the studied hydrosystems. The background noise of AMR in the Gabonese environment is so high that it reduces standard taxonomic or institutional fluctuations. This top-level leveling phenomenon corroborates the findings of the pan-European study by Berendonk et al. [33] and the work of Larsson and Flach (2022) [34], which postulate that when anthropogenic selection pressures exceed a critical threshold, aquatic ecosystems transform into a uniformly resistant matrix. In our study, the national median CRS stands at an elevated level of 5 antibiotic classes per isolate, reflecting the joint accumulation of resistance determinants to first-line molecules (ampicillin, amoxicillin) as well as third- and fourth-generation cephalosporins.

Institutional versus taxonomic factors in the structuring of AMR (MCA & PERMANOVA)

The implementation of Multiple Correspondence Analysis (MCA), combined with multivariate robustness tests, provides major conceptual insights. The PERMANOVA applied to binary resistance profiles demonstrates in a highly significant manner that the structuring of AMR profiles is dictated by the specific hospital environment of origin (p = 0.001) rather than by the taxonomic identity of the bacterial genera (p = 0.124). In parallel, the significant PERMDISP test (p = 0.005) confirms a marked heterogeneity in multivariate dispersion between healthcare facilities, illustrating the existence of local selection pressures specific to each healthcare structure (for instance, resistance rates to ertapenem reaching 91% at the CHUAB of Franceville and 94% at the HASG), the absence of molecular analyses (PCR, whole-genome sequencing or plasmid profiling) means that it is not possible to confirm the plasmid-mediated nature of these resistances or the prevalence of specific carbapenemase genes. These profiles are described strictly on a phenotypic basis (Kirby-Bauer). These results demonstrate that local medical prescription practices, the management of reserve antibiotic stocks, and the internal plumbing architecture of each hospital exclusively shape the phenotypic profile of resident bacteria. This finding aligns perfectly with the ecological theories of Baquero et al. [35] and empirical studies by Karkman et al. [36], which assert that the immediate environmental compartment (the hospital sewer) acts as a directional selection filter. Although horizontal gene transfer was not directly investigated, the similarity of resistance profiles across genera may reflect the occurrence of similar phenotypic resistance patterns across bacterial genera: similar resistance phenotypes were observed across several genera between Escherichia, Klebsiella, and Enterobacter, unifying their phenotypic behaviors against antibiotics [37].

Divergent ecological trajectories (Clades A, B, C) and co-resistance mosaic

Unsupervised topological analysis enabled the mapping of the enterobacterial population into three distinct epidemiological clades (A, B, and C) projected into the MCA space. Clade A represents a nationwide widespread high-level multidrug resistance cluster (with no exclusive hospital anchoring, Fisher’s exact test p > 0.05), suggests the widespread occurrence of highly resistant enterobacterial isolates within Gabon’s global anthropogenic network. In contrast, Clade C embodies an ultra-localized resistance pattern, presenting a massive and highly significant enrichment for the CHUAB of Franceville (Odds Ratio = 11.2; p < 0.001). This specific clade is driven by co-resistances involving last-resort therapeutic molecules: fosfomycin (FOS), amikacin (AK), and aztreonam (ATM). The analysis of the Phi (ɸ) correlation matrix provides a mechanistic explanation for this structure. The entire resistance network is characterized by an absence of perfect linear associations (ɸ < 0.7), reflecting a highly diversified and mosaic accumulation of resistance determinants rather than the rigid clonal expansion of a single variant [38]. Nevertheless, moderate but highly concerning cross-linkages were mapped, particularly between fosfomycin and ertapenem (ɸ = 0.48 depending on the adjusted indexes). This close association between resistance to carbapenems and fosfomycin may suggest a phenotypic association that may reflect co-selection mechanisms, as has recently been observed in similar contexts in East Africa [39].

The data presented in Fig. 11 suggest that the determinants of fosfomycin resistance do not circulate in isolation in these effluents, but are statistically and ecologically intertwined with phenotypes that already exhibit resistance to carbapenems (ETP) and monobactams (ATM).

Although these results contribute data that may inform future One Health investigations of antimicrobial resistance in Gabon, two fundamental limitations must be borne in mind when interpreting them: genomic limitation: Phenotypic co-occurrence (e.g. FOS and ETP) may indicate shared ecological or selective pressures. However, in the absence of whole-genome sequencing (WGS), it is impossible to state definitively that these resistance mechanisms are co-localised on the same mobile element. Clinical limitation: The correlations described reflect a complex environmental ‘mixture’ (effluents). They measure population-level ecological risk and must under no circumstances be extrapolated to predict individual co-resistances for the purpose of guiding empirical treatment protocols. These phenotypic associations suggest that resistance to different antibiotic classes may occur together within the same environmental settings that the clinical use of beta-lactams or carbapenems in Gabon is sufficient to maintain, through indirect pressure, environmental resistance to fosfomycin, thereby severely compromising the efficacy of salvage therapies for severe human infections.

Conclusion

This multisite study across Gabon demonstrates the important role of untreated hospital effluents in contaminating hydrosystems with multidrug-resistant (MDR) enterobacteria. The massive microbiological loads measured (up to 1.68 × 10⁸ CFU/mL) suggest that the absence of systematic pretreatment transforms clinical drainage networks into important sources of biological contamination. Furthermore, the use of macroscopic visual indicators (solid debris and foam) proved to be a reliable and low-cost predictive tool, well-suited for field surveillance in sub-Saharan Africa. Epidemiologically, the generalized multidrug resistance (100% of isolates) combined with a ceiling effect in the Cumulative Resistance Score (CRS) reveals an endemic saturation of AMR in the environment. Multivariate analyses (MCA and PERMANOVA) highlighted that the architecture of resistance profiles is dictated by local selection pressures specific to each facility rather than by taxonomic boundaries. Finally, the identification of a divergent epidemiological clade (Clade C) at the CHUAB of Franceville, which strongly couples resistance to fosfomycin and ertapenem, these findings raise concerns about the potential co-selection of resistance mechanisms affectingof last-resort human therapies. From a One Health perspective, these findings highlight potential links between hospital activities, wastewater contamination and environmental exposure, this research underscores the inseparable interconnection between hospital prescription practices, surface water degradation, and the exposure of local populations. Facing the reduced therapeutic options, it is urgent for Gabonese authorities to enforce a binding regulatory framework compelling every healthcare center to install dedicated pretreatment stations. to halt the dissemination of these resistant bacteria into the environment.

Authors’ contributions

Conceptualization, W.B.E.C., A-C.D., and R.O.; methodology, W.B.E.C., and R.O.; validation, W.B.E.C., A-C.D., R.O., G.F., A.L.L.B.O., L.F.O., and B.N.; formal analysis, G.F. and B.N.; investigation, W.B.E.C.; resources, W.B.E.C., A-C.D. and R.O.; writing original draft preparation, W.B.E.C., and R.O.; writing review and editing, W.B.E.C., R.O., and A-C.D.; visualization, W.B.E.C., R.O., G.F., A-C.D., and B.N.; supervision, R.O., and A-C.D., project administration, W.B.E.C. and R.O. All authors have read and approved the published version of the manuscript.

Funding

This research received no external funding.

Data availability

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

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.