Between March 2013 and February 2014, one conventional and one organic batch of three fruits and three vegetables were purchased, once a month, as available (Table 1). Produce were grown and marketed in Portugal and acquired at retail stores located in the region of Lisbon and Tagus Valley. The specimens were meant to represent the most consumed fresh produce in Portugal. Organic products were either bought at certified specialized stores, or were differentiated from conventional produce by the presence of specific labels, indicating that the producer respects specific regulation. Overall, we evaluated 144 products, including 74 vegetables and 70 fruits. Of these, 24 were green leaf lettuces (Lactuca sativa var. romana), 26 tomatoes (Solanum lycopersicum), 24 carrots (Daucus carota), 20 pears (Pyrus communis), 26 apples (Malus spp.) and 24 were strawberries (Fragaria spp). The exact same number of samples was collected from conventional and organic products. All products were immediately transported to the laboratory and processed: 50 g of each product was selected at random without washing or peeling, diluted 1:5, homogenized (Stomacher 80 Biomaster®, Seward, UK), labeled, enriched 12 h at 37°C, and frozen at −80°C (Ruimy et al., 2010).

Portions of the resultant fluids were spread-plated onto violet red bile glucose agar (VRBG) plates each containing specific concentrations of different antibiotics: 100 mg/L of amoxicillin, 2 mg/L of cefotaxime, 2 mg/L of ceftazime, 4 mg/L of ertapenem, 2 mg/L of imipenem, 10 mg/L of tetracycline, 20 mg/L of chloramphenicol, 50 mg/L of nalidixic acid, 4 mg/L of ciprofloxacin or 2 mg/L of gentamicin, and incubated at 37°C for 18–24 h. Control plates without antibiotic and with each antibiotic were used to assess total recovery of each sample and the integrity of the antibiotic selection protocol. The selection of 333 resistant Gram negative bacteria (Table 1, Tables S1, S2) was performed as previously described (Jones-Dias et al., 2016). Briefly, colonies obtained with each commodity were grouped by color, texture and size prior to representative numbers of them being streaked onto fresh VRBG agar to obtain an array of discrete colonies. Bacteria displaying a persistent, homogenous morphotype were considered pure strains, as previously reported (Bezanson et al., 2008; Jones-Dias et al., 2016). Individual colonies were selected based on their morphology so that no putative duplications were included, and all perceivable morphologically distinguishable colony types were sampled.

The isolates were identified through the amplification of the 16S ribosomal RNA (rRNA) gene, as previously described (Jones-Dias et al., 2016). PCR products were then purified with ExoSAP IT (USB Corporation, Cleveland, OH), and further sequenced directly, on both strands, using automatic sequencer ABI3100 (Applied Biosystems, Warrington, UK). The resulting sequences were then analyzed using the Bionumerics software (Applied Maths, Sint-Martens-Latem, Belgium) and assigned to respective identification, using the tools available on the NCBI website (http://blast.ncbi.nlm.nih.gov/Blast.cgi).

Among the 333 isolates selected, antibiotic susceptibility was analyzed for 320 isolates due to the inexistence of specific breakpoints to evaluate non-susceptibility results from Aeromonas spp., Comamonas spp., and Delftia spp. (n = 12). Isolates from genera Stenotrophomonas (n = 13) were only evaluated toward sulfamethoxazole/trimethoprim, which corresponds to the only breakpoint available for that bacterial species. The antibiotic susceptibility of remaining isolates (n = 308), distributed among Enterobacteriaceae, Acinetobacter spp. (family Moraxellaceae), and Pseudomonas spp. (family Pseudomonadaceae) was performed by disc diffusion method against the following antibiotics (Biorad): amoxicillin (25 μg), ampicillin (10 μg), cefotaxime (5 μg), ceftazidime (10 μg), cefepime (30 μg), cefoxitin (30 μg), ertapenem (10 μg), imipenem (10 μg), amoxicillin with clavulanic acid (20 + 10 μg), nalidixic acid (30 μg), ciprofloxacin (5 μg), gentamicin (15 μg), amikacin (30 μg), chloramphenicol (75 μg), tetracycline (30 μg), and trimethoprim with sulfamethoxazole (1.25 + 23.75 μg). Interpretation of results was performed according to the cut-off values recommended by the European Committee on Antimicrobial Susceptibility Testing (EUCAST, http://mic.eucast.org/Eucast2/); however, for Acinetobacter spp., four clinically relevant antibiotics were interpreted according to the Antibiogram Committee of the French Society of Microbiology (CA-SFM, http://www.sfm-microbiologie.org/) due to non-existence of EUCAST breakpoints (Table 2). Isolates were considered multidrug resistant when presenting non-susceptibility to three or more structurally unrelated classes of antibiotics (Magiorakos et al., 2011). Double disk synergy test was used to phenotypically detect the presence of specific groups of β-lactamases in the 333 isolates: amoxicillin (25 μg) and amoxicillin plus clavulanic acid (20 + 10 μg) for the phenotypic detection of penicillinases, cefotaxime (5 μg) and amoxicillin clavulanic acid (20 + 10 μg) for ESBL, cefoxitin (30 μg) and cloxacillin (750 μg) for AmpC β-lactamases, and imipenem (5 μg) and imipenem plus dipicolinic acid (750 μg) for metallo-β-lactamases. Escherichia coli ATCC® 25922™ was also tested as a control of this technique (Jones-Dias et al., 2014).

All isolates were investigated for the presence of class 1, 2, and 3 integrase-encoding genes, through PCR amplification using primers reported elsewhere (Leverstein-Van Hall et al., 2002; Corrêa et al., 2014; Manageiro et al., 2015).

All integrase-positive isolates were characterized by whole genome sequencing. Briefly, genomic DNA was extracted using DNeasy Blood and Tissue Kit (Qiagen, Aarhus) and quantified using Qubit 1.0 Fluorometer (Invitrogen, Waltham). The Nextera XT DNA Sample Preparation Kit (Illumina, San Diego, CA) was used to prepare sequencing libraries from 1 ng of genomic DNA according to the manufacturer's instructions. Paired-end sequencing of 250 bp reads was performed on a MiSeq (Illumina). Analysis of the integrase-producing isolates was carried out as described elsewhere (Jones-Dias et al., 2015). Briefly, PathogenFinder 1.1, ResFinder 2.1 (90% identity and 40% minimum length), and PlasmidFinder 1.3 (< 98% homology) were used to estimate the number of pathogenicity determinants, antibiotic resistance genes and plasmids, respectively, within the genome (Zankari et al., 2012; Cosentino et al., 2013; Carattoli et al., 2014). The NCBI prokaryotic genome automatic annotation pipeline (PGAAP) was used for annotation (http://www.ncbi.nlm.nih.gov/genome/annotation_prok). Specific analysis of class 1, 2, and 3 integrons was also carried out with CLC genomics workbench version 8.5.1 (Qiagen). Contigs carrying integrons were manually assembled and annotated whenever was necessary. Isolates showing novel Multilocus Sequence Typing (MLST) alleles combinations were assigned to sequence types (STs) through the respective MLST websites (http://pubmlst.org/ecloacae/ and https://enterobase.warwick.ac.uk/).

This Whole Genome Shotgun project has been deposited at DDBJ/EMBL/GenBank under the accession PRJNA311932. The versions described in this paper are the following: LSUR00000000 (Enterobacter cloacae INSali2), PRJNA311932 (E. cloacae INSali10), LSRK00000000 (E. coli INSali25), LSUT00000000 (E. coli INSali38), LSUU00000000 (E. coli INSali92), LSUV00000000 (Raoultella planticola INSali127), LSUW00000000 (R. planticola INSali133), LSUX00000000 (Morganella morganii INSali207), LSUY00000000 (E. coli INSali370), LSUZ00000000 (Pseudomonas putida INSali382) and LSVA00000000 (Klebsiella pneumoniae INSali390).

The integron sequences were submitted to the INTEGRALL database (http://integrall.bio.ua.pt) for annotation and integron number assignment.

Three hundred and thirty three Gram negative antibiotic resistant isolates, cultured from 72 conventionally and 72 organically produced fruits (n = 70) and vegetables (n = 74) were obtained upon selection with the antibiotics previously referred (Table 1, Tables S1, S2). Among the 144 samples from fresh produce, 78 samples carried Gram negative bacteria, among which, 39 were conventionally and 39 were organically produced. Globally, 195 and 138 bacterial isolates were recovered from fruits and vegetables conventionally and organically produced, respectively.

16S rDNA profiling revealed that the isolates were represented by 20 bacterial genera distributed among six main families of Gram negative bacteria, among which the predominant was Enterobacteriaceae (55.3%; Figure 1). The remaining groups comprised Moraxellaceae (Acinetobacter, 26.7%), Pseudomonadaceae (Pseudomonas, 10.5%), Xanthomonadaceae (Stenotrophomonas, 3.9%), Comamonadaceae (1.8%) [Delftia (0.9%) and Comamonas (0.9%)], and Aeromonadaceae (Aeromonas, 1.8%; Figure 1).

For Enterobacteriaceae (n = 184), β-lactam non-susceptibility ranged between 0% for imipenem/cefepime and 45.1% for ampicillin, in isolates from conventionally produced fruits and vegetables (Table 2). Variations regarding β-lactam non-susceptibility were detected among the modes of produce production for Enterobacteriaceae: an average of 9% for organic and 11.9% for isolates recovered from conventionally produced items. Globally, values of non-susceptibility for the two quinolones tested ranged between 0% and 4.9% for conventional produce (Table 2). Non-susceptibility values for Enterobacteriaceae against aminoglycosides (0.5–3.8%), chloramphenicol (3.8%) and sulfonamides (0.5–1.1%) showed similar results between isolates recovered from organic and conventional fruits and vegetables (Table 2).

For Acinetobacter spp. (n = 89), non-susceptibility to β-lactam antibiotics ranged between imipenem/cefepime (0%) for both types of production, and cefotaxime (63.0%) for isolates recovered from conventionally produced fruits and vegetables. Non-susceptibility to other assessed antibiotics (ciprofloxacin, gentamicin, amikacin, and tetracycline) showed 0–1.1% of non-susceptibility in bacteria from this genus recovered from conventional and organic production, respectively (Table 2).

When Pseudomonas spp. (n = 35) isolates were analyzed, the minimum values of antibiotic non-susceptibility fluctuated between 0% for ceftazidime/ciprofloxacin and ciprofloxacin/gentamicin/amikacin for organic and conventional produce, respectively. Maximum non-susceptibility values corresponded to 8.6% for ceftazidime/imipenem for isolates recovered from conventionally produced items (Table 2).

Stenotrophomonas malthophilia isolates (n = 13) recovered from lettuces (n = 11), carrots (n = 1) and tomatoes (n = 1) were all susceptible to trimethoprim/sulfamethoxazole.

Overall, E. coli (n = 2), M. morganii (n = 1), Acinetobacter spp. (n = 1), Enterobacter spp. (n = 1) and K. pneumoniae (n = 1) showed multidrug resistance.

Synergy tests identified presumptive phenotypic production of ESBL, AmpC β-lactamase, metallo-β-lactamases, penicillinases, and co-production of those β-lactamases (Figure S1). Their detection was associated with the origin of the samples to determine that 60.3% and 32% of the isolates recovered from conventionally and organically produced foodstuffs, respectively, were presumptive AmpC producers; 0.8% isolates from conventionally produced items were identified as phenotypically positive for ESBL production by the presence of synergy between cefotaxime and clavulanic acid, ceftazidime, and clavulanic acid or both (Figure S1). Overall, 80.1% of the isolates were presumptive producers of penicillinases. Metallo-β-lactamases were also presumptively produced by 1.6% of the isolates, distributed among conventional and organic recovered isolates grown exclusively on the soil (Figure S1).

The main statistics obtained with the de novo assembly of the 11 genomes are displayed in Table S3. The draft genomes varied among 3.8 Mb and 3688 protein-coding genes for M. morganii, and 6.5 Mb and 5938 protein-coding genes for P. putida, respectively (Table S3).

Overall, among the 333 isolates studied, we have detected 11 isolates harboring 12 integrons (Table 3). A single K. pneumoniae isolate harbored a class 1 and a class 3 integron within its genome. Integrons were similarly recovered along the whole period of sampling (March 2013 to February 2014), distributed among organic (n = 6) and conventional (n = 5) food products, and were predominant among lettuces (n = 9), but were also present in carrots (n = 1) and strawberries (n = 1; Table 3).

Besides the presence of typically chromosomal resistance genes such as bla_ACT−type and bla_DHA−type in E. cloacae, and M. morganii (data now shown), respectively, 10 out of the 11 isolates showed diversity of acquired resistance genes (Table 3). These genes may be able to confer resistance to multiple antibiotic classes, such as β-lactams, aminoglycosides, quinolones, chloramphenicol, and sulfonamides, as represented in the susceptibility profile displayed in Table 3. Among the acquired resistance mechanisms, we highlight the presence of the recently described plasmid-mediated colistin resistance gene mcr-1 in E. coli, and the macrolide inactivation gene mphA in M. morganii, both from lettuce samples, as well as ESBL-encoding gene bla_GES−11 and PMQR-encoding gene oqxAB in K. pneumoniae recovered from organic strawberries.

The assessment of STs, showed that one of the E. cloacae and one of the E. coli isolates were assigned to new allele combinations, that were registered in the respective MLST databases as ST636 and ST5981 (Table 3). In addition, other STs were detected for the remaining typable species: E. cloacae (ST90), E. coli (ST345, ST1716, and ST2522) and K. pneumoniae (ST15) (Table 3).

According with PlasmidFinder, the majority of the isolates harbored known Enterobacteriaceae plasmids from different incompatibility groups, among which we highlight the predominance of IncQ (INSali10, INSAli25, INSali38, INSali92, INSali370, and INSali390). Globally, 1–6 intact prophage regions were detected within the genomes of these Gram negative isolates. For E. coli isolates INSali25, INSali38, INSali92, and INSali370, virulence factors gad, lpfA, capU were differentially detected, as displayed in Table 3.

According with PathogenFinder, the majority of the isolates showed relevant similarity with human pathogens: the values have leveled off between 66.6% for M. morganii INSali207 and 94.2% for the clinically relevant E. coli isolates INSali38 and INSali92. P. putida INSali382 was the only integron-positive isolate predicted as a non-human pathogen (8.8%), showing also no acquired antibiotic resistance genes (Table 3).

Among the eight class 1 integrons analyzed, two were assigned to new integron numbers due to the detection of novel gene cassette arrays.

Globally, an assortment of promoters, gene cassette arrays and associated transposable elements were detected. Several Tn402-like class 1 integrons were recovered in this study. The 12,038 bp sequence of isolate E. cloacae INSali2 (Figure 2A) was assigned to In369, and enclosed an array containing gene cassettes drfA1b and aadA1b, encoding resistance to trimethoprim and aminoglycosides, respectively. A single complex integron was recovered (In293::ISCR1::qnrA1) in E. cloacae INSali10 (Figure 2B). The integron showed a variable region 1 (vr-1) comprised of aadB and aadA2 genes, and variable region 2 (vr-2) included a qnrA1, an ampR and a ΔhybF, which encode a zinc-containing protein associated with NiFe hydrogenase. Integron from E. coli INSali38 (Figure 2C) exhibited an intI1 interrupted by the insertion of a cluster of three resistance genes (sul2-strA-strB). In this case, integron variable region enclosed only one gene cassette with an identifiable open reading frame and attendant attC site: aadA1a. Similarly to INSali38, 11,395 bp Tn402-like integron harbored by E. coli INSali92 was preceded by repAC and sul2-strA-strB gene cluster, interrupting intI1 gene (Figure 2D). The gene cassette array of this In369, which was identical to INSali2, included drfA1b and aadA1b antibiotic resistance genes. In1259 was firstly detected in M. morganii INSali207 due to the presence of a new gene cassette array (Figure 2E). It is worth noting that the downstream flanking region was composed by an IS6100, a chromate transporter-encoding gene chrA, and by macrolide inactivation gene cluster mphA-mrx-mphR. E. coli INSali370 and K. pneumoniae INSali390 harbored similar Tn402-like integrons. However, while In27 enclosed dfrA12, gcuF and aadA2 in the variable region (Figure 2F), in the new In1258 (Figure 2G) aadA2 was substituted by the novel derivative aadA2br. The class 1 integron present in the genome of P. putida INSali382 contained a variable region for which no significant homology could be found in existing integrons (Figure 2H). In this In0, gene cassettes for known antibiotic resistance determinants were not present, neither the traditional 3′CS region encoding qacΔE1, sul1 and orf5. This Tn402-like integron was limited by a complete Tn402 transposition module – tniABQR – which was followed by a mer operon, encoding resistance to mercury.

Different class 2 integrons were detected in E. coli INSali25 (In2–35), and in R. planticola INSali127 and INSali133 isolates (In2–11) (Figures 3A–C). Just like the majority of class 2 integrons, the gene cassette arrays included genes coding for resistance to aminoglycosides and streptothricin, as follows: estX12-sat2-aadA1y for E. coli INSali25 and sat2-aadA1y for R. planticola INSali127 and INSali133 (Figures 3A–C).

The new class 3 integron In3–13 was detected in K. pneumoniae INSali390 isolate (Figure 4) that also carried the new In1258 (Figure 2G). The 8,873 bp length region showed 100% identity with In3–9, whose sequence was recently submitted to Genbank (KT984195), after being detected in Citrobacter freundii from a hospital effluent in France. The variable gene array enclosed dihydrofolate reductase type B-encoding gene dfrB3, which was followed by the ESBL-encoding gene bla_GES−11. The array of acquired antibiotic resistance genes was followed by genes encoding proteins essential for plasmid replication (repC, repA) and mobilization (mobC, mobA) (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.