Ten fish markets located in different urban and rural sections of the Mwanza region were randomly selected for the study. These markets receive fishes obtained from different fishing sites within Lake Victoria. A total of 196 Nile tilapia (Oreochromis niloticus) fish from randomly picked vendors were sampled with an average of 15 to 20 fish at each market between July and September 2015. Fish were first washed with saline, following which a sterile surgical blade was used to open the carcass and to make a longitudinal incision along the gut. The incision was opened and the gut contents were swabbed using a sterile swab.

In addition, at ten sites including the Ngerengere River that crosses the city, drain waste and possible sewage from households located in the surrounding hills were sampled (environmental samples). Environmental samples included dirty muddy water samples from different location in the city (Figure 1). For each site between six to seven samples were taken from different locations (Figure 1). About 3 ml of each sample was collected using sterile 25 ml falcon tube (BD, Nairobi, Kenya). A sample was mixed with sterile 0.9% saline at ratio of 1:1 and vortexed to produce a homogenous solution.

Using swabs, MacConkey agar (Oxoid, Basingstoke, UK) plates supplemented with 2 mg/L cefotaxime (Medochemie Ltd, Limassol, Cyprus) were inoculated to screen for ESBL-producing Enterobacteriaceae. Enteric bacilli were identified using colony morphology and differentiated based on lactose fermentation on MacConkey agar. Single colonies from predominant lactose fermenting bacteria were picked for further identification using several biochemical tests (Triple Sugar Iron Agar, Simmons' citrate Agar, Sulfur-Indole-Motility test and Urease test) (Murray et al., 1995). In case of ambiguous results, the VITEK® 2 system (BioMérieux, Marcy l'Etoile, France) was used to confirm identification. A confirmed ESBL-producing E. coli isolate was used as a positive and E. coli ATCC 25922 as a negative control. ESBL isolates were stored as glycerol cultures at −80°C and used for further characterization.

Antimicrobial susceptibility testing was done by disk diffusion method as recommended by the Clinical Laboratory Standards Institute (CLSI) guidelines (Wayne, 2012). A bacterial suspension of 0.50 McFarland standard turbidity was prepared from pure culture. An inoculum was then plated on Mueller-Hinton Agar plates (HiMedia, Mumbai, India) and the following antibiotic disks were set: Tetracycline (2 μg), ciprofloxacin (5 μg), gentamicin (10 μg), or trimethoprim/sulphamethoxazole (1.25/23.75 μg) (Oxoid, Hampshire, UK). The plates were incubated aerobically at 37°C for 18–24h. The diameters of the respective zone of inhibitions were measured and interpreted following CLSI 2012 guidelines (Wayne, 2012). Disk approximation method based on CLSI guidelines was used to confirm ESBL production and for selected isolates ESBL production was further identified using VITEK® 2 system (BioMérieux, Marcy l'Etoile, France) and in addition the MIC for cefepime, carbapenems and colistin were determined in all selected isolates.

Using STATA version 11 the two-sample test of proportion was done to compare the rates of resistance between ESBL isolates from fish and those from environment. A p-value < 0.05 was used to indicate a significant difference at 95% confidence interval.

Eleven ESBL-producing isolates obtained from fish and 13 E. coli from the environmental samples were chosen for whole genome sequencing (WGS). DNA was isolated using the Purelink Genome DNA Mini kit (Invitrogen, Darmstadt, Germany) according to the manufacturer's instruction. WGS was carried out using an Illumina Nextera XT library with 2x300bp paired-end reads on an Illumina MiSeq instrument (Illumina, San Diego, CA, USA). The raw data was assembled using SPAdes (version 3.0) (Bankevich et al., 2012). Contigs from E. coli isolates were ordered by using MAUVE (Rissman et al., 2009) and E. coli MG1655 (accession number U00096.3) were chosen as a reference for all E. coli isolates, Citrobacter freundii strain P10159 for Citrobacter braakii isolates (accession number CP012554; no complete genome available for C. braakii), Klebsiella pneumoniae strain ATCC BAA-2146 (accession number CP006659) for K. pneumoniae and Enterobacter cloacae strain 34977 (accession number CP010376) for E. cloacae isolates. Pseudogenomes were created and whole genome phylogenetic analysis was subsequently performed by using ParSNP package of Harvest Suite (Treangen et al., 2014). The raw sequencing data of the sequenced isolates are available at the European Nucleotide Archive (ENA) under the project number PRJEB12361.

Sequences were analyzed for their multi locus sequence types, transferrable resistance genes, plasmid replicon types and pMLST using MLST 1.8, ResFinder, Plasmidfinder and pMLST software from the Center for Genomic Epidemiology (Larsen et al., 2012; Zankari et al., 2012; Carattoli et al., 2014). Search for plasmid-encoded heavy metal resistances and detergence resistance was performed using blastn with the references given in Supplementary Table 1.

The location of bla_CTX-M-15 was determined by analyzing the contigs harboring bla_CTX-M-15 using blastn. The whole genome sequences were compared with the plasmid pPGRT46 (accession number KM023153.1) using BRIG and blastn (Alikhan et al., 2011; Fortini et al., 2015). Quinolone resistance-determining regions (QRDR) mutations were identified by silico analysis by comparing the sequence using a reference sequence from a quinolone-susceptible Enterobacteriaceae strain (Weigel et al., 1998; Liu et al., 2012).

The protocol of this study was approved by CUHAS/BMC joint ethics and scientific review committee with reference CREC/019/2014.

A total of 26 (13.3%) lactose-fermenting ESBL-producing bacteria were isolated from gut samples of 196 wild Nile tilapia fish from Lake Victoria. Diverse bacterial species were detected and included C. braakii (11/26, 42.3%), E. cloacae (5/26, 19.2%), K. pneumoniae (5/26, 19.2%), E. coli (4/26, 15.4%) and Klebsiella oxytoca (1/26, 3.9%). From 73 environmental samples 39 (53.4%) ESBL-producing enteric bacteria were isolated. Of these 39 isolates, 20 (51.3%) were E. coli and 19 (48.7%) were K. pneumoniae.

All isolates had MICs for cefotaxime and cefepime of >32 and >8 μg/ml respectively. High resistance rates to co-trimoxazole (n = 19, 73.1%), ciprofloxacin (n = 19, 73.1%), and gentamicin (n = 19, 73.1%) and tetracycline (n = 16, 61.5%) were observed among the 26 ESBL-producing isolates from fish. Environmental isolates were resistant to co-trimoxazole (n = 34, 87.2%), ciprofloxacin (n = 14, 38%), and gentamicin (n = 18, 46.1%). ESBL-producing isolates from fish were significantly more resistant to gentamicin and ciprofloxacin (p < 0.01: Table 1). In addition, all isolates were sensitive to meropenem and imipenem (MIC < 0.25 μg/ml), and all were sensitive to colistin.

Each of the bacterial genome-sequenced fish and environmental isolates carried up to four different β-lactamase genes (ESBL, AmpC and other β-lactamases, Table 2). The most common ESBL gene was bla_CTX-M-15, present both in fish (9/11) and environmental isolates (12/13). The non-ESBL β-lactamase gene bla_TEM-1 was present in 7/11 fish isolates and 9/13 environmental isolates, whereas bla_OXA-1 was present in 6/11 fish isolates and 1/13 environmental isolates. K. pneumoniae isolates from fish were found to harbor bla_SHV-11 and bla_SHV-1 due to the fact that K. pneumoniae chromosomally possesses bla_SHV-1/-11. In addition, the fish isolates also harbored a number of AmpC type β-lactamase genes (bla_ACT-15, bla_MIR-3, bla_CMY-37, bla_CMY-49), which were not present in the environmental isolates. The latter two results might simply reflect the fact that while different types of enterobacterial species were isolated from fish, only E. coli were obtained from the environmental samples studied.

Commonly occurring aminoglycoside resistance genes detected in both fish and environmental isolates were aac(6′)-Ib-cr (6/24), strA/strB (15/24), aac(3)-IId (5/24), and aadA1 (4/24) (Table 2). Other aminoglycoside resistance genes were present either only in fish isolates (aadA2, aac(3)-IIa) or in environmental samples (aadA5). Quinolone resistance genes were detected in 8/11 (72.7%) of fish isolates and in 7/13 (53.7%) of environmental isolates that were sequenced. Quinolone resistance genes detected in both populations comprised of the aac(6′)-Ib-cr (6/24), qnrB1 (5/24), and qnrS1 alleles. Other quinolone resistance genes were only present in fish isolates (qnrB29, qnrB48, oqxA, oqxB).

Of the 24 sequenced isolates, 20 (83.3%) harbored sulfonamide resistance genes (sul1 or sul2), and 19 of these isolates harbored in addition a resistance gene encoding trimethoprim resistance (dfrA14, n = 11; dfrA17, n = 3; dfrA1, dfrA18, dfrA30, dfrA5, dfrA7, each n = 1). Tetracycline resistance genes, [tet(A), n = 15; tet(D), n = 3], were present in 18/24 isolates. Altogether, there was a direct correlation between resistance phenotype and resistance genotype in all 24 isolates.

Using in silico analysis parC mutations were detected in 7 E. coli, only 2 isolates had S80I mutation that is associated with fluoroquinolone resistance, while gyrA mutations were observed in 9 isolates (2 E. cloacae and 7 E. coli). S83 (S83T, S83L, and S83A) mutations that are associated with fluoroquinolone resistance were observed in 5/9 isolates. One isolate had both S83L and D87N.

An analysis for the presence of plasmid-located heavy metal- and detergent- resistance genes revealed, that 8/24 isolates harbored the qacEdelta gene, conferring resistance to tertiary ammonium compounds. Seven (29.2%) of isolates sequenced carried a mercury resistance operon originally described in plasmid R478, and 4/24 (16.6%) isolates had genes mediating resistance to silver. Only one isolate carried genes involved in resistance to copper together with nickel/cobalt efflux system (Supplementary Table 2).

Analysis of the whole genome sequences was used to identify and map the location of bla_CTX-M alleles. For 8/24 isolates, the bla_CTX-M-15 gene was located in the chromosome. In two isolates bla_CTX-M-15 was located on a phage-like plasmid, similar to the E. coli phage-like plasmid p1303_95 isolated from wound swab (isolate SO007, accession number of the reference: CP009168.1) and pECOH89 (isolate SO053, accession number of pECOH89: HG530657.1) (Falgenhauer et al., 2014). The isolate SO069 harbored bla_CTX-M-15 on an IncI1 plasmid similar to pESBL-EA11 (Ahmed et al., 2012). Six of 24 sequenced isolates (one fish and five environmental isolates, F044, SO005, SO008, SO025, SO037, SO063, all E. coli) harbored a resistance cassette similar to the one present in plasmid pPGRT46 (accession number KM023153.1) that included both qnrS1 as well as bla_CTX-M-15. The other isolates displayed bla_CTX-M-15 or bla_CTX-M-55 containing resistance cassettes with similarities to other ESBL-encoding plasmids (F025: pEC_L46, accession number GU371929.1; F102: pSTm-A54650, accession number LK056646.1; SO042: pSKLX3330, accession number KJ866866.1; SO060: pCA14, accession number CP009231.1). The isolates F016 and F017 did not harbor any ESBL resistance gene. It should be noted that because conjugation experiments were not done plasmid location could only be suggested.

In 13 of 22 ESBL-encoding isolates, the environment of the ESBL gene was characterized by a Tn3 transposon deletion in the vicinity of the bla_CTX-M gene (Figure 2). This included four isolates harboring a chromosomally located bla_CTX-M-15 allele. The other isolates, including two chromosomally located bla_CTX-M-15 and the single bla_CTX-M-55 isolate, did not have any Tn3-related sequences. All ampC genes detected were located in the chromosomes (F006, F009, F016, F025, F102).

E. coli isolates sequence type (ST) ST-38 and ST-5173 were present in both fish and environmental isolates. Among the 13 E. coli environmental isolates 12 different STs were observed. The fifteen E. coli isolates, 13 from the environment and 2 from fish, grouped into five clusters using genome-based phylogenetic analysis. The tree was generated on the basis of a core genome accounting for 78% (3.62 Mbp) of the reference genome (E. coli MG1655). The two E. coli isolates from fish have 100% similarity with respective ST isolates derived from the environment (SO025 and F044, ST-38; F080 and SO038, ST-5137; Figure 3). E. cloacae isolates from fish displayed three different STs (ST-91 n = 2; ST-422 n = 1; ST-500 n = 1), based on an analysis of core genome content of around 68% (3.33 Mbp) using E. cloacae strain 34977 as reference. The two ST-91 isolates were more closely related to each other, than the other two isolates. K. pneumoniae isolates were typed as ST-37 (2/3) with a single isolate as ST-280 (1/3). The core genome with K. pneumoniae strain ATCC BAA-2146 (accession number CP006659) as reference accounted for 88% (4.78 Mbp) of its genome size. The ST-37 and ST-280 are quite different from one other. The C. braakii isolates could not be assigned to any ST, because there is no typing scheme available. Based on 82% (4.17 Mbp) core genome size using C. freundii strain P10159 as a reference genome, these two isolates are very distinct from each other (Figure 3). E. coli were grouped in different phylogroups based on scheme of Clermont et al. (2000). A total of 7/15(46%) of E. coli isolates belong to the newly described phylogroups E (Table 3).

Of the eleven fish isolates tested, seven (63.6%) were found to carry IncF plasmid replicon types as compared to 5/13 (38.5%) of environmental isolates (Table 4). The IncF pMLST types detected in fish isolates were F-: A-:B36, K4:A-:B- and K5:A-:B-. For the environmental isolates IncF pMLST types detected were F2: A-:B-, F29:-A-:B10 and F31:A4:B1. The IncI1 plasmids detected in E. coli from environment were classified using pMLST either as ST-31 and unknown pMLST while that in E. coli from fish had an unknown IncI1 pMLST. Two out of four E. cloacae isolates from fish harbored IncF plasmids with identical pMLST (F-:A-:B36). We extracted contig sequences of the F044, SO005, SO008, SO025, SO037, SO063, carrying the resistance cassette present in pPGRT46 (Fortini et al., 2015) (accession no. KM023153) and compared them to this plasmid to examine for overall homology. These isolates showed an overall overlap of between 76 and 90% in their nucleotide sequences with plasmid pPGRT46 (Figure 4).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.