Background

The antimicrobial resistance (AMR) phenomenon has spread rapidly over the course of the past decades to establish itself as a major global public health threat [1] in spite of the strides made by modern medicine to apply the use of antibiotics to ensure safe surgical procedures and improve the quality of medical care that have greatly reduced morbidity and mortality in public health [2, 3]. The bacterial infections that have been able to exhibit AMR have become fatal and given birth to a possible post-antibiotic era which initially was thought to be an apocalyptic fantasy before the 21st century [2, 4]. It goes without saying that the role played by humans in exacerbating the rate at which these infectious agents have developed resistance to antibiotics has been at the forefront of accelerating this phenomenon mainly through the rampant and inappropriate use of antibiotics [4, 5]. A previous study predicted an alarming 10 million deaths per annum with $100 trillion dollars’ worth of efforts trying to combat AMR by 2050 if this is not tackled [6]. Some regions in the world especially the African continent characterized by high infectious disease burdens and limited healthcare infrastructure have the least accurate and reliable statistical data on the epidemiology and impact of AMR on the public health sectors [7,8,9]. The rapid recent evolution of genomics-based technologies applied in the diagnosis and surveillance of the epidemiology of drug-resistant bacteria has led to the generation of large amounts of genomic data that have given deeper insights into the nature and changes of AMR determinants using modern bioinformatics analysis pipelines [10]. Application of next-generation sequencing technologies (NGS) alongside conventional microbiology procedures and antimicrobial susceptibility testing (AST) may be the key to understanding AMR [11], accelerating knowledge generation, and deploying interventions tailored towards optimization of antimicrobial use in public health [12].

Therefore, this study explored and sought to understand the epidemiology, and factors driving AMR in hospital settings in Uganda and Tanzania through the use of whole-genome sequencing (WGS) data combined with the socio-demographic metadata provided by the mother study.

Materials and methods

Study design and settings

This study utilized a laboratory-based and longitudinal study design approach. The laboratory-based design was used to undertake WGS to determine the AMR elements from the bacterial isolates.

Study sites

This study was carried out at the orthopedic units of Mulago National Referral Hospital, Kampala, Uganda with geographical coordinates (0°20’16.0"N, 32°34’32.0"E) and Bugando Medical Centre (BMC), Mwanza, Tanzania with geographical coordinates (2°31’41.0"S, 32°54’27.0"E).

Study population

The study population constituted of whole-genome sequence data obtained from a total of 142 multi-drug resistant E. coli bacterial isolates provided by the mother study titled, “Understanding Transmission Dynamics and Acquisition of Antimicrobial Resistance at Referral Hospitals and Community Settings in East Africa using Conventional Microbiology and Whole-Genome Sequencing”. The multi-drug isolates were collected from both study sites in Uganda (n = 57) and Tanzania (n = 85) from patients, the immediate non-medical caretakers of these patients, the immediate health workers attending to these patients and the patients’ environment as previously discussed in detail by the recent publications from Uganda [1] and Tanzania [13].

Data collection and analysis tools

The WGS data used in this study was provided by a bigger mother study titled, “Understanding Transmission Dynamics and Acquisition of Antimicrobial Resistance at Referral Hospitals and Community settings in East Africa using Conventional Microbiology and Whole-Genome Sequencing (Grant number GCA/AMR/rnd2/058)”. The bacterial isolates were shipped and sequenced by the Earlham Institute, Norwich, located in the United Kingdom following the Low Input, Transposase Enabled (LITE) Illumina protocol using the Illumina NovaSeq 6000 System.

The analysis of WGS data was done using our previously published Linux command line-based bioinformatics workflow called “rMAP”, the Rapid Microbial Analysis pipeline [10]. Briefly, the whole-genome raw sequences together with the GenBank Escherichia coli str. K-12 substr. MG1655 with Accession NC_000913 reference was fed into the rMAP pipeline. Sequences in the format fastq.gz were used as the input for the pipeline. All sequences were inspected for quality in the rMAP pipeline [10] before any subsequent processes using the embedded FastQC [14] to generate individual sample reports and MultiQC [15] for aggregating all the multiple reports into one report. Adapters were trimmed off the sequences using Trimmomatic [16] with the selected parameters including minimum length and phred score set to 200 and 32 respectively.

The trimmed reads were loaded into the Shovill [17] de-novo pipeline using the Skesa as the assembler of choice. K-mer sizes of 31, 55, 79, 103 and 127 were used to determine the optimum genome assembly. Pilon was used for checking assembly errors, correcting ambiguous gaps, insertions, deletions and finally polishing the genomes [18]. Genome annotations were performed using the Prokka tool [19].

Single nucleotide polymorphism (SNP) variant calling was performed using SAMtools, Burrows-Wheeler Aligner (BWA), SAMclip, Freebayes and SnpEff [20,21,22]. The trimmed reads were aligned against an indexed reference in fasta format (GenBank reference Escherichia coli strain K-12 sub strain MG1655 Accession: NC_000913.3) using Burrows-Wheeler aligner [21] to produce Sequence Alignment Map (SAM) files. Soft and hard clipped alignments were removed from the SAM files using SAMclip(https://github.com/tseemann/samclip). SAMtools [20] then sorted, marked duplicates and indexed the resultant Binary Alignment Map (BAM) files. Freebayes [23]was used to call variants using Bayesian models to produce variant call format (VCF) files containing SNP information which was filtered using BCFtools(https://github.com/samtools/bcftools) and normalized of biallelic regions using Vt [24]. The filtered VCF files were annotated using snpEff [22]. Missense variants that were associated with resistance were identified from the VCF files according to their respective sites. Only true SNPs were considered for the downstream analysis; insertions, deletions, and complex SNPs were filtered out from the resistance-associated SNPs.

Phylogenetic inference by maximum likelihood was performed using MAFFT, IQtree, Vcf2phylip, and BMGE [25,26,27,28]. The rMAP pipeline collated all the individual VCF files into a single VCF containing all the samples and their SNPs before being transposed by vcf2phylip [26] into a multi-alignment fasta file. MAFFT software package was used to perform multiple sequence alignment [27]; removal of ambiguously aligned reads as well as extraction of informative sites was performed to infer phylogeny using BMGE [25]. IQtree [28]was then used to test various substitution models and construct trees from the alignments using the maximum-likelihood method in 1,000 bootstraps. The resulting trees were visualized in the form of rectangular phylograms.

Mass screening for AMR genes against CARD [29], ARG-ANNOT [30], NCBI, ResFinder, and MEGARES [31] databases was performed for each of the study isolates using the Abricate tool (https://github.com/tseemann/abricate). For consistency purposes, we compared results from the two most commonly used well-annotated AMR databases across the E. coli isolates (CARD and ResFinder) from both study sites. From our findings, we found more AMR genes conforming to > 90% cut-off for both coverage and identity being detected from the ResFinder database which were then presented in form frequencies and heatmaps in the results section.

Multi-locus sequence typing was performed using MLST (https://github.com/tseemann/mlst) from the E. coli assembled contigs. The E. coli isolates were determined against a set of seven (7) E. coli housekeeping genes (adk4, fum26, gyrB2, icd25, mdh5, purA5, and recA19).

Virulence factors were determined against the virulence factor database (VFDB) [32] using the Abricate tool (https://github.com/tseemann/abricate) to identify elements that conformed to > 90% cut-off for both coverage and identity.

Results

Pre-sequencing quality control

A total of 142 ESBL multi-drug resistant E. coli bacterial isolates from both Tanzania and Uganda were sequenced. Only 36/57 (63.1%) of the ESBL-producing E. coli from Uganda and 67/85 (78.8%) of the ESBL-producing E. coli isolates from Tanzania passed the initial pre-sequencing Transposase Enabled (LITE) Illumina protocol and were selected for the downstream bioinformatics analysis as shown in Fig. 1.

Fig. 1
figure 1

Proportions of the ESBL-producing E. coli from the study sites that conformed to initial pre-sequencing (LITE) Illumina protocol

Full size image

Raw sequence quality assessment, adapter, and poor read sequence trimming

The average read length of the sequences was 150 bp with an average Q-score of 32 as discussed in the methods section.

De-novo genome assembly

Considering the main genome assembly parameters, the samples from the two study sites had an average of 132 bp (base pairs) for the mean read length and 525 bp for the contigs. An estimated average genome length of 5,138,608 bp, average GC-content of 51%, and average sequencing depth of 9.34X were reported for the isolates. A summary of the average genome assembly statistics across all analyzed samples is shown in Table 1.

Table 1 De novo genome assembly metrics and statistics for the study isolates
Full size table

Single nucleotide polymorphism (SNP) variant calling

The annotated VCF files were interrogated for SNPs that were “missense” meaning those that altered the function of the protein and had a keyword “resistance”. Generally, the most clinically significant predominant resistance-associated SNPs and respective genes affected included: peroxide stress resistance protein (yaaA), chloramphenicol resistance permease (RarD), and quinolone resistance protein (ybgl). A detailed depiction of the distribution of these SNPs between the two study sites is shown in Table 2.

Table 2 Frequency of resistance-associated SNPs identified from Ugandan and Tanzanian isolates
Full size table

Antimicrobial resistance gene determinants from Uganda and Tanzania

From the Ugandan site, 18/36 (50%) isolates were found to have AMR genes conforming to the set > 90% cut-off for both coverage and identity in the ResFinder database. The most predominant genes identified among the isolates comprised of blaCTX−M−15 with 11/18(61.1%) which confers resistance to fluoroquinolones, and third-generation cephalosporins, mdfA 14/18(77.7%) which confers resistance to chloramphenicol through its efflux pump action, tet(B) 9/18 (50%) conferring resistance to tetracyclines and sul1 8/18(44.4%) conferring resistance to sulphonamides. The heatmap in Fig. 2 shows the distribution of the respective genes across the isolates. The green color indicates the presence of a gene while the grey shows the absence of that gene per isolate.

Fig. 2
figure 2

Heat map showing Ugandan E. coli predominant genes from ResFinder database

Full size image

From the Tanzanian site, 23/67 (34.3%) isolates were found to have AMR genes conforming to the set > 90% cut-off for both coverage and identity in the ResFinder database. The most predominant genes identified among the isolates comprised of blaCTX−M−27 with 12/23 (52.1%) conferring resistance to fluoroquinolones and third-generation cephalosporins, and blaTEM−1B with 13/23 (56.5%) which confers resistance to third-generation cephalosporins, mdfA(A) 21/23 (91.3%) which confers resistance to chloramphenicol through its efflux pump action, tet(A) 12/23 (52.1%) conferring resistance to tetracyclines and sul2 15/23 (65.2%) conferring resistance to sulphonamides. The heatmap in Fig. 3 shows the distribution of the respective genes across the isolates. The green color indicates the presence of a gene while the grey shows the absence of that gene per isolate.

Fig. 3
figure 3

Heat map showing Tanzanian E. coli predominant genes from ResFinder database

Full size image

Multi-locus sequence typing (MLST)

A total of 34/36 (94.4%) isolates for which MLST was performed conformed to the set > 90% cut-off for both coverage and identity for the sequence types (STs). The most predominant E. coli from Uganda comprised of the following: ST656 accounting for 9/34 (26.4%) within the isolates, ST206, ST448, ST1193, and ST1284 each accounting for 2/34 (5.8%) within the isolates. The majority of the other STs were; ST131, ST10, ST6438, ST998, and ST38 among others each accounted for 1/34 (2.9%) within the isolates.

In comparison with the Tanzanian site, 59/67 (88%) isolates conformed to the set > 90% cut-off for both coverage and identity for the sequence types (STs). The most predominant E. coli STs from Tanzania comprised of the following: ST131 accounting for 5/59 (8.4%) within the isolates, ST10, ST2852, and ST167 each accounting for 2/59 (3.3%) within the isolates. Majority of the other STs were: ST3580, ST695, ST542, ST48, and ST38 among others accounted for 1/59 (1.6%) within the isolates. Fig. 4 below shows how the isolates between the two countries cluster in terms of AMR genes and STs.

Fig. 4
figure 4

Phylogenetic tree showing the distribution of AMR genes and STs across the two study sites

Full size image

Virulence factor detection(VF)

In addition to their AMR genes, the E. coli species from the two study sites were found to possess arsenals of virulence factors that are inter-species in nature that is, they originated from other different bacterial species, a clear depiction of horizontal transfer of genetic material between bacteria. The isolates from the two countries had most of their virulence factors identical to those found in Shigella dysenteriae Sd197, Yersinia pestis CO92, Salmonella enterica subsp. enterica serovar Typhimurium str. LT2, Escherichia coli O157:H7 str. EDL933, and Pseudomonas aeruginosa PAO1. A comprehensive table depicting the distribution of the different virulence factors of the isolates from the two sites is shown in Table 3.

Table 3 Distribution of virulence factors among the E. coli isolates from Uganda and Tanzania
Full size table

Discussion

Control of multi-drug resistant infections is fundamental in reducing the disease burden and costs incurred while treating these pathogens in tandem with the Global Action Plan set by WHO [33] to combat AMR. This study comes in at the right point in time where the scale at which global public health is threatened by the increasing infection rates. In this study, the aim was to explore the genetic determinants that confer AMR from isolates obtained at Mulago National Referral Hospital, Bugando Medical Centre, and their environmental settings. The findings from this study are provocative and inform the dire need to strengthen the existing infection-prevention controls (IPC) together with surveillance and monitoring systems.

This study was predominantly comprised of ESBL organisms; with 36/57 (63.1%) originating from Uganda and 67/85 (78.8%) originating from Tanzania. Previous findings from studies in Uganda reported ESBL E. coli prevalence rates of 5.3% conducted between 2006 and 2007 [34], 62% carried out in 2014 [35] at Mulago National Referral Hospital, and between 2015 and 2016 at Kasese Regional Referral Hospital at 62% [36]. Related meta-analysis and systematic review studies from East Africa carried out in hospitals and surrounding communities reported a similar predominance of ESBL-producing Escherichia coli and Klebsiella pneumoniae [37,38,39]. The Enterobacteriaceae family has been reported to shape the nosocomial pathogen eco-system because of the plasticity of their genome and their ability to perform inter-species and intra-species incorporation and transfer of drug resistance mediating determinants like plasmids, transposons, insertion sequences, and virulence factors via horizontal gene transfer [40,41,42].

Detection of SNP-associated mutations in the genes; RarD, yaaA, and ybgl conferring resistances to chloramphenicol, peroxidase, and quinolones from the isolates from both Uganda and Tanzania further depicts how these organisms evolve resistance towards some of the most commonly used antibiotics and antiseptic used for the day-to-day management of clinical cases. A related study highlighted the roles played by these SNPs in the evolution of antimicrobial resistance and in shaping the genome of these organisms [43].

Our results reported very high frequencies for blaCTX−M−15 accounting for 11/18(61.1%), and blaCTX−M−27 with 12/23 (52.1%), blaTEM−1B with 13/23 (56.5%) of isolates originating from Uganda and Tanzania respectively. These genes are responsible for conferring resistance to penicillin, fluroquinolones, and third-generation cephalosporins (ceftazidime and cefotaxime) which are part of routinely prescribed antibiotics used in the treatment of medical cases within the two sites similar to other previous related studies that reported phenotypic AMR profiles of the same organisms [44]. Tanzania had relatively higher tetracycline resistance gene tet(A) with 12/23 (52.1%) as compared to the Ugandan isolates with tet(A) with 7/18 (38.8%) prevalence which are in agreement with similar studies performed across six Tanzanian hospitals [45]. We also reported relatively a high prevalence of sulphonamide-resistance conferring genes sul1 8/18(44.4%) and sul2 15/23 (65.2%) from Uganda and Tanzania respectively. Chloramphenicol resistance gene mdfA(A) with 21/23 (91.3%) from Tanzanian isolates and trimethoprim resistance-conferring gene dfrA17 with 8/23 (34.7%) in Ugandan isolates were also observed within the two cohorts. The authors propose that the high resistance observed in the majority of over-the-counter antibiotics can likely be explained by their affordability in the two countries. In contrast, expensive drugs such as piperacillin-tazobactam, amikacin, and carbapenems are less accessible in most drug shops within the two study sites. Consequently, these expensive drugs are less likely to be misused, leading to lower detected resistance levels. These findings sound a very big alarm about the potential dangers such pathogens can cause to the general public health and call for the need to scale up the microbiology laboratory capacity as a way of guiding antimicrobial agent prescription. Data from well-established laboratory facilities will shape and strengthen AMR surveillance, IPC protocols within community-based settings and healthcare facilities, and regulation of drug prescriptions from drug outlets and pharmacies.

The largest portion of sequence types isolated from Tanzania belonged to the ST131 accounting for 5/59 (8.4%) of the total isolates while Uganda was represented with 1/34 (2.9%) for ST131 of the total isolates. These sequence types associated with extra-intestinal infections have been reported to be rapidly spreading as high-risk clones in Europe and worldwide due to increased AMR [46,47,48,49]. Some other sequence types like ST206 have been reported to be associated with colistin-resistance conferring isolates from a study in China [50]. These findings reiterate the dangers these organisms impose on the public health system and call for immediate interventions [51].

The presence of virulence factors like Shigella dysenteriae Sd197 (gspD, gspE, gspF, gspG, and gspF), Yersinia pestis CO92 (irp1, ybtU, ybtX, and iucA), Salmonella enterica subsp. enterica serovar Typhimurium str. LT2 (csgF and csgG), Shigella dysenteriae Sd197 (gspC, gspD, gspE, gspF, gspG, gspH, and gspI), and Pseudomonas aeruginosa PAO1 (flhA, fliG, and fliM) in isolates from both Tanzania and Uganda reported by this study depict a classic case of inter-species genetic-determinant element transfer. Multiple studies have reported how the horizontal gene transfer, a process through which genetic information can be acquired from the environment to a bacterium or from one bacterium to another through other mechanisms other than chromosomal inheritance consequently shaping pathogen virulence evolution [52,53,54,55,56]. This study provides strong evidence regarding the acquisition of a set of rather queer virulence factors like Yersinia pestis CO92 among the Escherichia coli isolates originating from a somewhat deadly plague-causing bacteria similar to what has been reported by a study in India [57].

The ingenious advent of bioinformatics platforms like rMAP [10] used to profile all these virulence elements within the isolates in one go provides a comprehensive way of analyzing WGS data because of its easy installation, usage, and applicability, especially in low-income settings where high-performance computing infrastructure is limited. In our opinion, it also bridges and fills the missing link between the rapidly embraced field of WGS and conventional microbiology while providing high-resolution, and shorter result-generating turnaround times for the genomes of MDR pathogens. It is on these grounds that we recommend this tool to be adopted as a continuous monitoring and surveillance software for monitoring the antimicrobial resistance gene trends, plasmids, virulence factors, and MLSTs within community and healthcare settings for Uganda, Tanzania, and Africa as a whole.

Conclusion

This study shows a notable abundance and diversity of the AMR-conferring elements shaping the genomic structure and survival of E. coli isolates from the hospital and clinical settings in Uganda and Tanzania. The findings from this study are provocative and inform the dire need to strengthen the existing infection-prevention controls (IPC) and adoption of WGS alongside conventional microbiology as a way of building genomics surveillance and monitoring capacity in East Africa.

Limitations

Due to financial limitations, our study relied exclusively on WGS short-read data for deductions. Introducing long-read sequencing such as Nanopore sequencing requires a low quantity of input DNA and is suitable for low-concentration samples and in conjunction with short-read might have enriched our understanding of the AMR-conferring elements elucidated in our research. Hybrid long-short-read sequencing approaches offer a powerful tool in microbial genomic research. By combining the strengths of both long-read and short-read sequencing, hybrid approaches offer improved accuracy, read length, and cost-effectiveness.