Fecal samples were collected from four Herring Gulls in a breeding colony at the Shoals Marine Laboratory on Appledore Island, ME (42° 58′ N, 70° 36′ W) located approximately 10 km off the coast of mainland United States. Approximately 650 pairs of Herring Gulls nest on the island (Ellis, unpublished data). We collected fresh samples by observing individual birds until defecation occurred and immediately collecting samples from the ground using a sterile syringe. For comparison, one soil and two gull samples were also collected from the mainland (Rochester, NH, USA). Samples were immediately placed on ice. DNA extraction was performed with the FastDNA Spin for Soil kit (MP Biomedical, CA, USA) using the protocol of the manufacturer with the following changes. Cells were lysed using the MP Biomedicals FastPrep24 (setting 5.0 twice for 45 s and once for 20 s, cooling on ice in between). The precipitation step was carried out twice with 250 μl protein precipitation solution (PPS) from the FastDNA Spin for Soil kit (MP Biomedical, CA, USA). We used 1.5 ml binding matrix and inverted for 4 min. We discarded 1 ml of the supernatant. The final pellet was resuspended in 75 μl of the provided water.

Three micrograms of genomic DNA from gull feces was sheared to a target size of 3 kb with a Covaris S2 instrument (Covaris Inc., MA, USA). The 3-kb fragments were end-repaired with the DNA Terminator End Repair kit (Lucigen, WI, USA). The DNA was loaded on a 1% low melt agarose gel and fragments between 1 and 3 kb were extracted with Zymoclean Gel Recovery kit (Zymo Research, CA, USA). DNA was eluted in 10 μl TE elution buffer and ligated into pSMART–HCkan vector (accession number AF532107) using the Clonesmart blunt end cloning kit (Lucigen, WI, USA). Ligation was carried out overnight at room temperature. E. coli 10G electrocompetent cells (Lucigen, WI, USA) were transformed by electroporation using a BTX electroporator and 1.0 mm gap cuvettes. Electroporation conditions were 25 μF, 200 Ω, and 1800 V. The transformation was incubated for 1 h at 37°C and, after adding 30 μg/ml kanamycin, the transformation mixture was incubated another 2 h at 37°C. We plated 0.1 μl of the transformation mixture on kanamycin LB plates for titer counts and 100–200 μl on ampicillin (50 μg/ml), amoxicillin (20 μg/ml), penicillin (50 μg/ml), and tetracycline (8 μg/ml) containing plates. Screened clones were picked and grown up over night without shaking in LB + 10% glycerol. Sequencing was done using the SL1 forward primer (5′-CAGTCCAGTTACGCTGGAGTC-3′) and SR2 reverse primer (5′-GGTCAGGTATGATTTAAATGGTCAGT-3′; as suggested by Lucigen) and sequenced at Beckman Coulter Genomics.

Antibiotic resistance clones were tested in duplicate for growth in the following concentrations of ampicillin: 60, 100, 200, 350, 500, 800, and 1000 μg/ml. Five microliters of live culture was inoculated into 96-well plates containing 225 μl LB Broth plus kanamycin (30 μg/ml) and the above concentrations of ampicillin. We also included a “no-cell” control for contamination and background OD. The cultures were incubated at 37°C and growth was measured at OD₆₀₀ every 15 min for 24 h.

To identify the AR gene in selected inserts, we randomly inserted a Tn5 transposon in the clones using the EZ-TN5 <TET-1> Insertion Kit (Epicentre, CA, USA). We used a molar ratio Tn5 to AR plasmid of one to minimize multiple insertion events. One microliter of a transposon reaction was electroporated (BioRad BTX) into 50 μl of E. Cloni 10G Electrocompetent Cells (Lucigen). Cells were plated on LB Agar containing kanamycin (30 μg/ml) and tetracycline (10 μg/ml). Colonies were next screened for Tn5 “knockouts” based on lack of ampicillin resistance in LB Broth plus kanamycin (30 μg/ml), tetracycline (10 μg/ml), and ampicillin (60 μg/ml). A replicate control 96-plate was also inoculated with the same colony in LB Broth plus kanamycin (30 μg/ml) and tetracycline (10 μg/ml). Both plates were grown at 37°C (without shaking) overnight. Sequencing was done using the TN5 <TET-1> forward and reverse primers (provided in Epicentre kit).

We trimmed the paired-end sequences using Phrap and analyzed the inserts using blast against both GenBank nr and the Antibiotic Resistance Genes Database (ARDB; Liu and Pop, 2009). For β-lactamases, the ARDB classification scheme generally follows other systems for organizing resistance genes (Bush and Jacoby, 2010). However, here we rely entirely on a sequence-based comparison. Using ARDB, genes are divided into the four major β-lactamase classes (A–D), then into specific ARDB types, and finally into sequence variants within each type. We detected some sequences that were identical to β-lactamases used in common cloning vectors; because we cannot determine whether these sequences are contaminants we removed them from the analysis.

Sequences from each ARDB type were assembled into contigs using Geneious (Biomatters). For selected inserts, we aligned and translated the sequences and built phylogenetic trees with Phylip (100 bootstraps) using neighbor-joining (protein distance matrix with Jones–Taylor–Thornton correction) and maximum likelihood methods (Jones–Taylor–Thornton correction and constant rate of change among sites; Felsenstein, 2006). The phylogenetic trees were visualized with the Interactive Tree of Life (iTOL, http://itol.embl.de/index.shtml; Letunic and Bork, 2007). Sequences from this study have been submitted to GenBank under accession numbers JM426721–JM427844.

To target AR in gulls, we sampled feces from four gulls on Appledore Island, Maine, and two gulls in Rochester, NH, USA. For comparison, we also sampled soil in Rochester without any gulls nearby (Table 1). We extracted and sheared the DNA, then cloned 3 kbp DNA fragments into E. coli. The number of resulting clones ranged between 1 × 10⁵ and 4.4 × 10⁶ per library. Based on these numbers, we estimate that up to 5% of the bacteria in gull feces are resistant to antibiotics (Table 1). We found ampicillin, amoxicillin, and penicillin resistance genes in all gulls assayed, whereas we detected tetracycline AR genes in only two out of six gulls. Although it is difficult to compare across antibiotics due to differences in potency and concentration, resistance to ampicillin was found in higher occurrence than amoxicillin and penicillin.

Gulls carried an extensive genetic diversity of both known and previously undescribed AR genes. To characterize this diversity in the metagenomic libraries, we paired-end sequenced the resistant clones and first classified the genes using ARDB (Liu and Pop, 2009; Tables 2 and 3). Overall, we found that similar genes conferred resistance to ampicillin, amoxicillin, and penicillin, including nine different class A and C β-lactamase ARDB categories. The most common category was bl1_ec, which was 100% similar to a β-lactamase in E. coli strain 1.1.43. Another common category detected was the extended-spectrum class A β-lactamase (ARDB category: bl2b_tem1). However, the bl2b_tem1 types were not identical to known variants in the database and on average shared only 92% sequence similarity. Combined, these two ARDB categories constituted more than 50% of the AR clones (Table 2). The last common ARDB category was the class A CTX-M type. However, these genes only shared 72% amino acid sequence similarity to CTX-M genes in E. coli (Table 3). A broader search in GenBank and a phylogenetic analysis revealed that the gene was 100% similar to the β-lactamase variant RAHN-2 in the Gram-negative enteric bacterium Rahnella aquatilis Y9602 (Ruimy et al., 2010) and therefore, was likely not a CTX-M variant (Figure 1).

Beyond the class A and class C β-lactamases, we also found ampicillin- and penicillin- resistant clones encoding for efflux pumps (acrA and macB type) and a membrane carboxypeptidase (ARDB category: pbp1a). Finally, we observed a high proportion of clones (38%) that had no clear match in ARDB.

The tetracycline resistant genes detected belonged to several ARDB categories including tetC, tetJ, and tetL efflux pumps and tetM, tetO, and tetW ribosomal modification proteins. These genes were generally very similar to known genes (Table 3) and likely do not constitute new variants.

For comparison, we also analyzed the genetic diversity of AR in a nearby soil sample. Keeping in mind that we only analyzed one soil sample, we found a lower frequency of AR genes in the soil sample compared to the gulls. With the exception of tetW, all ARDB categories observed in soil were also detected in gulls. Furthermore, we did not detect the RAHN-2/CTX-M category in soil. Thus, it appears that many AR genes in gulls are also found in soil (and vice-versa), with perhaps some exceptions.

Antibiotic resistance genes are known to move around between hosts via lateral gene transfer, thus the AR gene sequence itself is not a good indicator of the organism with the gene. Therefore, we analyzed the flanking sequences around the AR genes (Figure 2; Tables 2 and 3). Data regarding the genomic neighborhood can give information about the host organism and/or whether the gene was located on a mobile genetic element. Our attempts to assemble clones belonging to the same ARDB category yielded good contigs, indicating that the detected AR genes shared flanking sequences. An example of that was the common class C β-lactamase belonging to the ARDB category bl1_ec. All clones associated with this category were located on a genomic fragment syntenous with E. coli strain K-12 substr. DH10B and several other E. coli strains (Figure 2A). Similar to DH10B, the gene was flanked by fumarate reductase subunits frdB, frdC, and frdD, an outer membrane lipoprotein (Blc), a quaternary ammonium compound-resistance protein (sugE), and the toxin entericidin (Durfee et al., 2008). Overall, this suggests that the most common class C β-lactamase found in the gulls was located on a genomic fragment similar to that found in E. coli. Similarly, the β-lactamase variant RAHN-2 also assembled into one contig which suggested a fragment syntenous with the genome from R. aquatilis Y9602 (Figure 2B). This contig including a chemotaxis gene and a hypothetical protein. This supports the hypothesis that this variant is indeed located in a Rahnella genome.

In contrast to the bl1_ec and RAHN-2 variants, clones matching the TEM class A β-lactamases (bl2b_tem1) were found in a range of genetic neighborhoods. The genes flanking bl2b_tem1 sequences were in some cases most closely related to Gram-positive bacteria (e.g., Lactobacillus, Enterococcus, and Clostridium) and in other cases, to Gram-negative bacteria (e.g., E. coli, Shigella, and Salmonella).

There were two major types of tetracycline resistance genes in our samples – efflux pumps (tetC, tetJ, and tetL) and ribosomal modification proteins (tetM, tetO, and tetW; Table 2). With the exception of tetJ, these were all associated with Gram-positive bacteria including Bacillus, Enterococcus, and Clostridium. Several of them were located on what appeared to be mobile elements including plasmids and transposons.

In addition to detecting multiple classes of AR genes, we found several DNA polymorphisms in the bl1_ec type. To further explore this result, we aligned the sequences matching this ARDB type and built a phylogenetic tree (Figure 3). All gull bl1_ec sequences grouped with E. coli 1.1.43 with high bootstrap support. In contrast, sequences retrieved from the soil samples formed three separate clades, with only one soil clone clustering with the gull variants. Instead, the one clade of soil AR sequences were 100% similar to E. coli H120, whereas the other genes branched away from known E. coli genes. Overall, this indicates that the bl1_ec diversity is related to its environmental origin. We also analyzed the sequence variation among clones from the RAHN-2 β-lactamase genes but only observed one variant.

In addition to the known AR gene categories in ARDB, we detected 31 gene types not previously described (Table 3). The novel AR genes were associated with a wide range of phylogenetic lineages including Pseudomonas, Bacteroides, Gram-positives, Verrucomicrobia, and different genera affiliated with Enterobacteriaceae. To further characterize these, we first measured the minimum inhibitory concentration (MIC) for ampicillin. All ampicillin/penicillin/amoxicillin resistant clones had an MIC above 1000 μg/ml ampicillin and thus clearly conferred resistance to this antibiotic.

Two of these clones assembled to a 2800-bp DNA fragment (unk15 in Table 3). One end of the fragment shared 84% amino acid similarity to the hypothetical protein VspiD_21980 in Verrucomicrobium spinosum from the phylum Verrucomicrobia. The other part of the fragment was 45% similar to a C-terminal protease in Gemmata obscuriglobus from the phylum Planctomycetes.

We next identified the genomic region in the clone that encoded for AR by randomly inserting a Tn5 transposon followed by testing for ampicillin resistance and DNA sequencing of the clones. The Tn5 “knockout” revealed that insertions in both genes interrupted the resistance whereas several insertions between the two genes had no effect. This represents a completely novel AR gene system that confers resistance in E. coli, but at present, we do not know the specific mechanism. To the best of our knowledge, this is the first study to identify an AR gene in either the phylum Verrucomicrobia or Planctomycetes.

A second previously uncharacterized AR gene (unk1) shared 85% amino acid similarity to a protein in the Gram-positive Carnobacterium sp. AT7 that was annotated as a putative DNA repair protein. A Tn5 insertion in this gene interrupted the resistance. Next to this putative AR gene was a gene 85% similar to pilT in another Gram-positive lineage, Enterococcus faecalis. Another previous unrecognized AR gene (unk31) was 92% similar to a DNA repair gene of the ImpB/MucB/SamB family in Lactobacillus coryniformis subsp. torquens KCTC 3535. Multiple Tn5 insertions in this gene also interrupted the resistance. None of these DNA repair genes have previously been shown to confer AR, so the exact mechanism of resistance is unclear.