The Birth and Demise of the ISApl1-mcr-1-ISApl1 Composite Transposon: the Vehicle for Transferable Colistin Resistance.

The origin and mobilization of the ~2,609-bp DNA segment containing the mobile colistin resistance gene mcr-1 continue to be sources of uncertainty, but recent evidence suggests that the gene originated in Moraxella species. Moreover mcr-1 can be mobilized as an ISApl1-flanked composite transposon (Tn6330), but many sequences have been identified without ISApl1 or with just a single copy (single ended). To further clarify the origins and mobilization of mcr-1, we employed the Geneious R8 software suite to comprehensively analyze the genetic environment of every complete mcr-1 structure deposited in GenBank as of this writing (September 2017) both with and without associated ISApl1 (n = 273). This revealed that the 2,609-bp mcr-1 structure was likely mobilized from a close relative of a novel species of Moraxella containing a chromosomal region sharing >96% nucleotide identity with the canonical sequence. This chromosomal region is bounded by AT and CG dinucleotides, which have been described on the inside ends (IE) of all intact Tn6330 described to date and represent the ancestral 2-bp target site duplications (TSDs) generated by ISApl1 transposition. We further demonstrate that all mcr-1 structures with just one ISApl1 copy or with no ISApl1 copies were formed by deletion of ISApl1 from the ancestral Tn6330, likely by a process related to the "copy-out-paste-in" transposition mechanism. Finally, we show that only the rare examples of single-ended structures that have retained a portion of the excised downstream ISApl1 including the entire inverted right repeat might be capable of mobilization.IMPORTANCE A comprehensive analysis of all intact mcr-1 sequences in GenBank was used to identify a region on the chromosome of a novel Moraxella species with remarkable homology to the canonical mcr-1 structure and that likely represents the origin of this important gene. These data also demonstrate that all mcr-1 structures lacking one or both flanking ISApl1 were formed from ancestral composite transposons that subsequently lost the insertion sequences by a process of abortive transposition. This observation conclusively shows that mobilization of mcr-1 occurs as part of a composite transposon and that structures lacking the downstream ISApl1 are not capable of mobilization.

INTRODUCTION

Over the past decades, the serious threat to public health posed by the development and rapid spread of antibiotic resistance (Abr) has become increasingly clear to the medical community and to the general public (1). This dissemination of resistance genes is facilitated by their frequent sequestration into transposable genetic elements (TE), small DNA segments capable of moving from one place in their host genome to another. Insertion of TE into transmissible plasmids subsequently facilitates their colonization of other bacteria of the same species and, depending on the plasmid, of other species and genera.

There are several ways in which Abr genes may be sequestered into TEs (see reference 2). Among the important Abr gene sources are composite or compound transposons frequently found in nature (3). Establishment of compound structures is thought to result from random insertion of two IS copies on either side of an Abr passenger gene. However, there are examples of structures which include an entire flanking IS copy at one side and a surrogate IS end located at some distance at the other (4–7). The way in which such structures arise and subsequently decay has not been documented. However, the massive increase in DNA sequencing power has now provided us with the tools necessary to investigate transposon populations and the way in which transposons and plasmids change over time within and between bacterial populations (and their hosts). This allows us to address issues of short-term evolution of these highly plastic mobile genetic elements: how they arise and how they decay over time.

A recently identified example of the birth and decay of a compound transposon concerns the mcr-1 gene responsible for resistance to the last-resort antibiotic, colistin (polymyxin E). Identification of this transferable phosphoethanolamine (pEtN) transferase gene in November 2015 caused great concern (8). Prior to the discovery of mcr-1, colistin resistance mediated through pEtN modification of lipid A was attributed primarily to mutations in regulators of intrinsic pEtN transferases such as the PmrAB system in Acinetobacter baumannii (9) or the PhoPQ system in Klebsiella pneumoniae (10). Although they perform the same function, there is considerable diversity among pEtN transferases (8).

The origin of mcr-1 is still uncertain. In most mcr-1 copies identified to date, the region encompassing this gene is an ~2,609-bp DNA segment containing mcr-1 and a putative 765-bp open reading frame (ORF) gene encoding a protein similar to a PAP2 superfamily protein (8). The initial description of MCR-1 indicated that it aligned to pEtN from several species, including Moraxella catarrhalis, with which it shared 59% amino acid identity (8). More recently, AbuOun and colleagues reported a chromosomal region on Moraxella sp. MSG13-C03 that shared 96.6% nucleotide identity (NI) with the 2,609-bp mcr-1-pap2-containing DNA segment (11). However, irrespective of the origins of this gene, the initial mobilization of mcr-1 appears to be closely associated with insertion sequence (IS) ISApl1 (12), a member of the IS30 insertion sequence family, most likely via the composite transposon Tn6330 (13).

We showed previously (12) that mcr-1 sequences deposited in public databases have three general structures (Fig. 1): that of ISApl1 composite transposon Tn6330 with directly repeated flanking IS copies (13) (Fig. 1A); those with just a single, upstream ISApl1 copy (single ended) (Fig. 1B); and sequences lacking ISApl1 altogether (Fig. 1C). However, in some of the latter two types of structure, the remnants of an ancestral ISApl1 were still present (12). Recently, Poirel et al. demonstrated mobilization of the mcr-1 fragment by an engineered derivative of Tn6330, Tn6330.2, experimentally confirming that Tn6330 is a primary vehicle for mcr-1 transposition (14).

FIG 1 

Four representative sequences showing the four general mcr-1 structures identified to date. (A) The composite transposon Tn6330 (13). (B) pHNSHP45, a single-ended structure with an upstream copy of ISApl1 only. (C) pSLy1, a structure lacking both copies of ISApl1. (D) pSh418-m3, a single-ended structure with a downstream copy of ISApl1 only. The ~2,609-bp region comprising the mcr-1 region is shown flanked by the conserved, ancestral TSD dinucleotides AT and CG on the IE. The last 27 bp of the mcr-1 region that comprise the 3′ end of the putative pap2 gene are indicated for clarity. ISApl1, light gray box; transposase, long white arrow; terminal inverted repeats, dark gray triangle; mcr-1, red arrow; pap2, short white arrow. The same scheme is used for all of the figures.

Much of the uncertainty surrounding the role of ISApl1-mediated mcr-1 mobilization has centered on the single-ended variants, with some studies suggesting that mcr-1 can be mobilized by this structure (15, 16). Gao and colleagues postulated that this is achieved as a consequence of the presence of a pseudo-ISApl1 inverted right repeat (IRR) that encompasses the last 26 bp of the pap2 gene (Fig. 1) (15). Putative target site duplications (TSDs) were identified in a number of single-ended variants, but it is unclear whether these are true TSDs. Since ISApl1 generates only a short, 2-bp TSD upon integration, determining whether these represent true TSDs or are simply serendipitous is difficult (12). Indeed, recent experiments failed to detect mcr-1 transposition by this type of single-ended variant with the pseudo-ISApl1 IRR (14), although the frequencies obtained for transposition of Tn6330.2 itself were extremely low and the frequency of transposition of a transposon with a pseudoend might be expected to be even lower.

In the present study, we conducted a comprehensive analysis of all mcr-1-containing sequences deposited in GenBank to date (n = 273; see Table S1 in the supplemental material). Our data suggest that Tn6330 was generated by ISApl1 capture of a mcr-1 gene from a close relative of Moraxella sp. MSG13-C03 whose genomic sequence was recently deposited in GenBank (11). Furthermore, our data indicate that single-ended variants of Tn6330 are the result of ISApl1 loss and appear capable of transposition only when a fully intact IRR of this ancestral ISApl1 is present. Finally, the results show that ISApl1 is exceptionally recombinogenic in natural populations. Tn6330 has a propensity to lose one or both flanking ISApl1 copies, resulting in loss of transposability and stabilization of the mcr-1 gene. In addition to the three structures described previously (12), we have now observed a rare fourth structure in which the upstream copy of ISApl1 is deleted while the downstream copy is retained (Fig. 1D). This type of deletion activity had also been observed for the founding member of this IS family, IS30, in the laboratory (17). Finally, to explain the frequent ISApl1 deletions, we provide a model that is based on the transposition mechanism of IS30 family members (18) and which invokes a process of abortive transposition.

Information concerning the sequences analyzed. Column 1, GenBank accession numbers of the sequences; column 2, location of the sequence (in either plasmid or chromosome); column 3, plasmid name (where applicable); column 4, plasmid incompatibility group (Inc) (where known); column 5, sequencing platform; column 6, source of the sequence, organism, and strain; column 7, type of sequence (composite transposon with two flanking IS; double deletion in which both flanking IS have been excised; single deletion in which the left or right IS copy has been excised; empty site where an identical sequence without the transposon insertion was identified); column 8, the 2-bp target site duplication (TSD); column 9, unique insertion site code; column 10, specific comments (where applicable). Download TABLE S1, XLSX file, 0.03 MB.

RESULTS

To gain greater insight into how ISApl1 has mobilized mcr-1, we undertook an extensive in silico analysis of all interpretable mcr-1 sequences deposited in GenBank as of this writing (n = 273; September 2017) (see Table S1 in the supplemental material). All sequences share an ~2,609-bp DNA segment (Fig. 1A) (8) containing mcr-1 and a putative 765-bp orf, pap2, encoding a Pap2 superfamily protein. There is some minor variability in the length of this mcr-1 region due to the presence of small insertions/deletions and mutations at the extremities, but, apart from this, the sequences are highly conserved. Variability in the immediate flanking regions is primarily due to the presence or absence of ISApl1. In the analysis presented below, we address how these structures may have been formed and their relationship to each other.

The birth of Tn6330: ISApl1-mcr-1-ISApl1.

During our analysis of mcr-1 sequences for this study, we identified a recent GenBank submission of a nucleotide sequence termed mcr1.10 (MF176238) that showed 97.6% identity to the canonical mcr-1 gene (11). The gene was found in Moraxella sp. MSG13-C03 and was present in a 2,606-bp chromosomal region that shared a strikingly high level of nucleotide identity (96.6%) with the 2,609-bp mcr-1 region (Fig. 2). Furthermore, as shown by calculation of the average score of all substitutions in the alignment using a BLSM62 substitution-scoring matrix (19), mcr-1 aligned to this region with a pairwise positive-identity level of 99.3%. Moraxella sp. MSG13-C03 was isolated in April 2014 in the United Kingdom from the cecal contents of a healthy pig (11). Analysis indicates that Moraxella sp. MSG13-C03 represents a novel species whose closest relatives are Moraxella porci DCM 25326 (GenBank accession number NZ_MUYV00000000.1) and Moraxella pluranimalium CCUG 54913 (GenBank accession number NZ_MUYU00000000.1), with which it shares average nucleotide identities (ANI) of 82.3% and 81.3%, respectively (Fig. S1). The 2,606-bp Moraxella sp. MSG13-C03 region encompassing the mcr-1 and pap2 genes has 3 indels compared to the canonical 2,609-bp mcr-1 structure, but the missing bases are all in noncoding regions: One occurs upstream of mcr-1, while the others occur between mcr-1 and the start of pap2 (Fig. 2). Of particular note, this 2,606-bp region is flanked upstream by the dinucleotide AT and downstream by the dinucleotide CG (11) (Fig. 2, bold text). The same nucleotides were previously reported to occur at the inside ends (IE) of the flanking ISApl1 copies in Tn6330, likely representing the ancestral TSDs formed during the initial mobilization of this gene by ISApl1 insertion (12).

FIG 2 

Alignment of Tn6330 and the homologous region on the Moraxella sp. MSG13-CO3 chromosome. The consensus sequence at the end of both regions is provided, with the conserved AT and CG dinucleotides that are found on the IE of all Tn6330 highlighted in bold. Mutations and deletions are indicated by vertical black lines and colons, respectively, between the two images.

Phylogenetic tree generated from the core genome sequences of various Moraxella species using RAxML (34). The genome of Moraxella sp. strain MSG13-C03 is highlighted in blue. Branch lengths are proportional to the number of nucleotide changes with 100 bootstrap replications. Download FIG S1, TIF file, 2.5 MB.

The presence of these dinucleotides bounding a region of such high homology to the mcr-1 region strongly suggests that the initial mobilization of mcr-1 occurred from a Moraxella species closely related to Moraxella sp. MSG13-C03 by insertion of ISApl1 upstream and downstream. ISApl1 insertion into these T-A-rich target sites flanking the 2,606-bp region (Fig. 2) presumably generated the 2-bp TSDs AT and CG at the directly repeated upstream and downstream ISApl1 copies, respectively. This would have constituted the ancestral Tn6330. In further transposition events, the ancestral TSDs located at the Tn6330 outside ends (OE) would be replaced by new TSDs generated by Tn6330 insertion into a new target site. The resulting ISApl1 composite transposon, Tn6330, would retain both internal conserved 2-bp fingerprints (AT and CG).

ISApl1-mcr-1-ISApl1 composite transposons (Tn6330) have target site duplications.

As of this writing (September 2017), 31 mcr-1 sequences with complete directly repeated flanking ISApl1 copies forming the composite transposon Tn6330 had been deposited in GenBank (Table S1). Twenty-two were located on plasmids encompassing a variety of incompatibility (Inc) classes and 9 on host chromosomes (Table S1). There is some redundancy in this collection, as only 20 different insertion sites were identified from the 31 sequences. In each case, the conserved ancestral AT and CG dinucleotides are evident at the IE of the flanking ISApl1 (Fig. 1A). In the majority of cases, Tn6330 is flanked by 2-bp TSDs, characteristic of ISApl1 transposition. Six Tn6330 derivatives lack these TSDs, but all are associated with plasmid rearrangements that have resulted in TSD deletion after Tn6330 insertion (Fig. S3). Furthermore, the sequences of Tn6330 were >99% identical in all of the structures and therefore appear to be active representative copies of a single ancestral transposon.

Additional genetic events within Tn6330 were also evident among seven sequences from five unique insertions (Table S1). Tn3 and ISKpn26 have inserted into Tn6330 in pMCR_1511 and pls1, respectively, generating characteristic 5-bp (Tn3; GTAAA) and 4-bp (ISKpn6; CTAG) TSDs (data not shown). Similarly, IS1294 has inserted into Tn6330 in five sequences representing 3 different insertion sites (Table S1). No TSDs are evident with these insertions because IS1294 transposes via a rolling-circle mechanism and does not generate TSDs upon integration (20).

Tn6330 derivatives which retain a functional downstream ISApl1 IRR.

There are 10 single-ended derivatives in the library (Table S1) with a deletion within the downstream ISApl1 but where a significant segment of the right ISApl1 end is retained, including an entire 27-bp IRR presumably essential for transposase recognition and activity. One of these had been identified previously (12) in the Escherichia coli RL465 chromosome (LT594504). This retained the last 90 bp of the IS. This structure is flanked by a putative 2-bp TSD (CA). A second, identified in the E. coli S51 chromosome, retained 42 bp of the ISApl1 right end, including the entire IRR (Table S1) (12). This structure is also flanked by a probable 2-bp TSD (TC). We had previously speculated that these may be mobile due to the presence of the intact IRR ends and the putative TSDs (12). Since then, 7 additional sequences, representing 5 unique insertion sites, which also retain the last 42 bp of an ancestral downstream ISApl1 have been deposited in GenBank (Table S1) (Fig. 3). Three are present on host chromosomes, and 4 are located on IncI2 or IncK2 plasmids (21, 22). Four have putative TSDs flanking the entire structure, and at least three of these appear to be legitimate based on a comparison with corresponding empty site sequences (i.e., identical sequences with no mcr-1 insertion). In the remaining 3 structures, no TSDs are present, but deletions and rearrangements similar to those observed with some of the composite transposon structures are evident in the surrounding area which likely removed the TSDs. On the basis of these additional data, it is possible that single-ended variants retaining the IRR of an ancestral downstream ISApl1 are capable of transposition, though experimental verification would be required to address this definitively.

FIG 3 

Schematic representation of single-ended mcr-1 cassettes that have retained the last 42 bp (A) or 90 bp (B) of an ancestral ISApl1 that includes the entire IRR. The conserved, ancestral CG dinucleotide on the IE of the downstream ISApl1 is indicated with a black triangle. The bases upstream and downstream of the deletion that are retained after ISApl1 loss are highlighted in red and blue, respectively. The deletion joints upstream and downstream of the ISApl1 are encased in a black rectangle. The bases upstream and downstream of the deletion that are retained after ISApl1 loss are highlighted in red and blue, respectively, while the remaining copy of the deletion joint that is retained after the two ends are joined following ISApl1 excision is highlighted in green and encased in a black rectangle. The same labeling scheme is used throughout the figures. Putative 2-bp TSDs in 5 of the 7 sequences are highlighted in bold.

ISApl1-mcr-1: single-ended Tn6330 variants were created by the loss of a downstream ISApl1.

The members of a second major class of 59 ISApl1-associated mcr-1 genes have only a single, upstream ISApl1 copy (Fig. 1B). Fifty sequences are located on plasmids representing a variety of Inc types, 8 on host chromosomes, and 1 in a putative phage. There is significant redundancy within these 59 sequences, however, and just 18 unique insertion sites are present (Table S1). In all 59 sequences, the ~2,609-bp regions encompassing mcr-1 and pap2 are >99% identical except for a concentration of nucleotide variability at the 3′ end of pap2.

Putative 2-bp TSDs could be identified flanking 29 sequences representing 10 unique insertions sites, and this has led to speculation that these structures are mobile, ostensibly due to an imperfect IRR encompassing the last 26 bp of pap2 (Fig. 1B) (15). There are two plausible hypotheses to explain these: (i) they are mobile and have transposed using the imperfect IRR (15) or (ii) they represent Tn6330 decay products in which the downstream ISApl1 copy has undergone deletion.

A notable feature of all single-ended variants, irrespective of whether putative TSDs are present, is a concentration of nucleotide polymorphisms at the 3′ end of the pap2 gene (12) (Fig. 4; bold text). For example, an alignment of the single-ended variants in many IncI2 plasmids suggests that a GA TSD flanks the entire structure (Fig. 4; black rectangle). However, when the sequences directly upstream of this putative TSD were aligned, some contained a CAAG tetranucleotide, others had a TAAG, and others appeared to be missing one or more bases (Fig. 4; bold text). This pattern is repeated throughout other single-ended variants even in plasmids that are almost identical. In analyzing these base changes, we noted a pattern in the nucleotide differences that was consistent with the second hypothesis proposed above: deletion of an ancestral downstream ISApl1. In the majority of single-ended variants, an ancestral downstream ISApl1 was found to have been removed from a position between an ~20-bp upstream region that encompasses the 3′ end of pap2 and an ~20-bp downstream region encompassing the 3′ end (i.e., the IRR) of an ancestral downstream ISApl1 (Fig. 4B; bold blue). This fully explains all nucleotide variation in single-ended variants. Excision of ISApl1 between these regions occurs at deletion joints that range in length from 1 to 4 bp and which we hypothesize to be the point at which the two ends are joined after ISApl1 excision.

FIG 4 

(A) Consensus alignment of multiple mcr-1 regions from a variety of plasmid backbones with one (single-ended) flanking copy or no flanking copy of ISApl1. The minor sequence variation at the 3′ end of the pap2 gene in all structures and at the 5′ end of mcr-1 in structures lacking the upstream ISApl1 is highlighted in bold text, with absent bases indicated by bold dashes. To preserve the alignment, the additional TTAA tetranucleotide in pSCE516-1 has been placed above the sequence, with a line indicating the correct position. (B) Alignment of the last 27 bp of the pap2 gene and the 26 bp that constitute the IRR of ISApl1. The nucleotides that form the basis for the sequence variability noted in structures that have lost the downstream ISApl1 are highlighted in blue.

To better clarify this, we identified complete Tn6330 copies on plasmid backbones that were highly similar to those carrying the single-ISApl1 sequences and that were also inserted into the same location. Furthermore, we also identified highly similar plasmids with corresponding empty sites allowing us to accurately examine the nucleotide variation present at the 3′ end of the single-ended variants.

The top of Fig. 5 shows multiple examples of a variety of insertion sites and plasmids with representative composite transposons deposited in GenBank. In addition, we inferred the sequence of a hypothetical parental composite transposon insertion for pHNSHP45-2 (marked with an asterisk), which has had no corresponding composite transposon deposited to date. This was accurately achieved by comparing the single-ended sequence, its corresponding empty site, and known sequences of other composite transposons. The 2-bp TSDs flanking the downstream IRR that was generated upon insertion of the original composite transposon (prior to the loss of the downstream copy) are indicated (black, solid rectangle above the dinucleotide). The conserved, ancestral CG dinucleotide abutting the left inverted repeat (IRL) on the IE of the downstream ISApl1 is indicated (black triangle). The bottom of Fig. 5 illustrates the single-ended variants formed through loss of the downstream ISApl1 identified in our library. These are compared to the real (and, for pHNSHP45-2, “hypothetical”) composite transposons described above, with the deletion junction encased in a black rectangle and highlighted in green. Nucleotides highlighted in red represent the original sequences upstream of the excision point that are retained in the single-ended variant, while those highlighted in blue represent the sequences retained downstream of the excision. The corresponding empty site for each sequence is highlighted in gray, with the 2-bp TSD that was generated upon insertion of the ancestral Tn6330 highlighted by a black, solid rectangle above the dinucleotides. The nucleotides that form the basis for the variation noted above at the 3′ end of the single-ended structures are highlighted in bold and can be consistently explained by small variations in the region excised during loss of ISApl1. It is noteworthy that, although there are some minor variations in the excision point, the length of the excised region ranges from 1,064 bp to 1,074 bp, with an average of 1,070 bp (standard deviation [st. dev.], 2.2 bp; Fig. S2), the same length as ISApl1. For example, if the excision occurs further into the pap2 gene (i.e., further in the 3′ direction), the point of the downstream excision has a tendency to shift in the 3′ direction to compensate, and vice versa.

FIG 5 

Alignment showing the decay of multiple different instances of Tn6330 (top) into the corresponding single-ended structures formed (bottom of figure), aligned with their respective empty sites (bottom of figure, highlighted in gray; see the text for more details). “Hypothetical” Tn6330 insertions constructed from known empty site plasmids, the corresponding single-ended structures, and the sequence of Tn6330 are indicated with an asterisk. The labeling scheme is identical to that described for Fig. 3.

Distribution of the different deletion sizes following loss of the downstream ISApl1 in all 19 sequences shown in Fig. 5 and 6. The average deletion size is 1,069.8 bp (standard deviation, 2.4). Download FIG S2, TIF file, 7.9 MB.

FIG 6 

Alignment showing the decay of multiple different instances of Tn6330 (top) into the corresponding double-deletion structures formed (bottom of figure), aligned with their respective empty sites (below, highlighted in gray; see the text for details). “Hypothetical” Tn6330 insertions constructed from known empty site plasmids, the corresponding double-deletion structures, and the sequence of Tn6330 are indicated with an asterisk. Dashes were inserted to maintain sequence alignment.

Comparison of plasmids with Tn6330 insertions to the corresponding “empty site” plasmids, indicating the various events that have removed the TSD in some Tn6330 structures. (A) The deletion of ~6,380 bp from the pECJP-59-244 plasmid backbone compared to empty site plasmid p180-PT54. (B) The deletion of ~1,480 bp from the P14408_M1 and pSA-MCR-1 plasmid backbones with a concomitant inversion of the downstream ISApl1 gene compared to empty site plasmid p180-PT54. (C) The deletion of ~1,450 bp from the plasmid backbone in pS38 with a corresponding movement of the last 36 bp of the downstream ISApl1 directly upstream of the IRL of the upstream ISApl1. The mcr-1 and pap2 genes are represented by a red block arrow and a white block arrow, respectively. Truncated genes are represented by ragged ends. The solid, rotated triangles at the ends of ISApl1 represent the IRL (left) and IRR (right) sequences. Download FIG S3, TIF file, 5.5 MB.

ISApl1 loss resulted in mcr-1 regions entirely devoid of ISApl1.

By far the largest numbers of mcr-1 sequences deposited in GenBank lack any copy of ISApl1. However, there was considerable redundancy in these sequences, with just 8 unique insertion sites identified from the 181 different submissions (Table S1). Every sequence is plasmid borne, with IncI2 and IncX4 plasmids being the most common groups (60% and 36%, respectively). Similarly to the single-ended variants, the sequences encompassing mcr-1 and pap2 are >99% identical except for a concentration of base changes at both the 5′ and 3′ ends of the region (Fig. 4). We conducted the same analysis of these sequences as for the single-ended variants described above. In all examples, as seen with the loss of the upstream ISApl1, the loss of the downstream ISApl1 copy could be explained in a manner identical to that employed for the single-ended variants.

Figure 6 depicts multiple examples of the IRL and IRR and surrounding sequences of the upstream ISApl1 from a variety of insertion sites and plasmids. The composite transposon MTY17668-MCR1.5 is an example of the ancestral sequence that gave rise to the double-deletion variants (see below). In addition, we inferred the sequence of hypothetical parental single-ended variants for pH226B and pMCR16_PO53 (marked with an asterisk), as no corresponding single-ended variants of these double-deletion structures have been deposited to date. The 2-bp TSDs flanking the hypothetical upstream IRL (IE) that would have been generated upon insertion of the original composite transposon (prior to the loss of the upstream copy) are indicated (black, solid rectangle above the dinucleotide). The conserved, ancestral AT dinucleotide abutting the IRR (IE) of the upstream ISApl1 is also indicated (black triangle). The aligned sequences presented at the bottom of Fig. 6 illustrate the actual double-deletion variants formed through the loss of the upstream ISApl1 in the real and “hypothetical” single-ended variants described above. The deletion joint is encased in a black rectangle and highlighted in green. Nucleotides highlighted in red represent the original sequence upstream of the excision point retained in the double deletion variant, while those highlighted in blue represent the retained downstream sequence. The corresponding empty site for each sequence is highlighted in gray, with the 2-bp TSD that would have been generated upon insertion of the hypothetical ancestral Tn6330 highlighted in black (solid rectangle above the dinucleotides). As noted for the single-ended variants, the nucleotides that form the basis for the variation at the 5′ end of the double-deletion structures (highlighted in bold) can be consistently explained by the presence of small variations in the region excised during loss of the upstream ISApl1 and the point at which the two ends are joined after the excision. Therefore, loss of the upstream copy appears to occur in a way similar to that seen with the downstream ISApl1 copy.

mcr-1-ISApl1: single-ended Tn6330 variants were created by the loss of an upstream ISApl1.

In addition to the three general mcr-1 structures identified previously (Fig. 1A, B, and C) (12), we also identified a rare fourth structure that includes only a downstream ISApl1 copy (Fig. 1D). Only three such sequences were available. They represent 2 unique insertion sites (Table S1) in IncI2 and IncX4 plasmids. In two of the structures (KY363997 and KY363999), IS1A has a 59-bp sequence inserted upstream of the mcr-1 gene, creating a characteristic IS1A 9-bp TSD (AAAAAATTG). Analysis performed using the procedures outlined above with corresponding empty sites (and intact Tn6330 insertions) demonstrated that loss of the upstream ISApl1 copy can also explain these structures (data not shown).

The demise of Tn6330.

We identified a set of related incompatibility group IncI2 plasmids that clearly showed consecutive in situ stages in Tn6330 acquisition and decay (Fig. 7). pChi7122-3 is an example of an IncI2 plasmid with an empty site. Insertion of Tn6330 into this plasmid at a GA dinucleotide, next to nikB (relaxase gene), generated a close relative of pMCR-M15049 and a 2-bp GA TSD. Deletion of a sequence between a TT dinucleotide (encased in a black rectangle in the figure) within the downstream ISApl1 IRR and a second upstream TT (encased in a black rectangle) between the end of the pap2 gene and the IRL of the IS copy yielded a close relative of pHNSHP45 that formed when the ends (highlighted in red and blue) joined at the TT dinucleotide (highlighted in green and encased in a black rectangle) after excision of the downstream ISApl1. A subsequent excision of the upstream ISApl1 occurred between an AAGA tetranucleotide (encased in a black rectangle) abutting the upstream ISApl1 IRL and a second AAGA tetranucleotide (encased in a black rectangle) spanning the IRR. This generated a close relative of plasmid pSLy1 formed when the ends (highlighted in red and blue) joined at the AAGA tetranucleotide (highlighted in green and encased in a black rectangle), and that was thus devoid of both flanking ISApl1 copies. Notably, the plasmid backbones of pMCR-M15049, pHNSHP45, and pEC006 differ by only 10 to 11 single nucleotide polymorphisms (SNPs), strongly suggesting that these plasmids indeed represent successive periods of decay of Tn6330 as a consequence of loss of ISApl1.

FIG 7 

Alignment showing the formation and successive decay of Tn6330 in a set of four highly related IncI2 plasmids. The labeling scheme is identical to that described for Fig. 3.

Together, these results provide a convincing scenario explaining the sequestration of the mcr-1 gene into an ISApl1-based composite transposon, Tn6330, from a genome similar to that of Moraxella sp. MSG13-C03 and the transmission of this transposon to various plasmid replicons and its decay into a series of nontransposable derivatives.

DISCUSSION

A comprehensive analysis of 273 intact mcr-1 sequences representing 38 unique insertion sites (see Table S1 in the supplemental material) allowed us to formulate a model for the formation and subsequent decay of Tn6330, the primary vehicle for the mobilization of colistin resistance gene mcr-1 (13, 14). A previous analysis of an initial set of mcr-1 sequences revealed that mcr-1 was part of an ~2,609-bp region flanked by one copy, two copies, or no copies of ISApl1 (12). In this report, we demonstrate that all mcr-1 structures can be explained by loss of one or both copies of ISApl1 from an ancestral Tn6330 and provide a scenario that maps this course of decay.

Recently, a 2,606-bp chromosomal region containing mcr-1.10 from Moraxella sp. MSG13-C03 was identified which showed ~97% nucleotide identity with the 2,609-bp mcr-1 region from Tn6330 (Fig. 2) (11). The region contains 3 indels compared to the mcr-1 structure, but the missing bases are all in noncoding regions (Fig. 2, colons). Importantly, this 2,606-bp region is flanked by the ancestral dinucleotide ISApl1 target sites (upstream, AT; downstream, CG) (Fig. 2, bold text) previously identified at the inside ends (IE) of the flanking ISApl1 copies in Tn6330. Although we have yet to identify presumed intermediates in Tn6330 formation (i.e., the chromosomal target mcr-1 region with a single upstream or downstream ISApl1 insertion), the presence of the conserved dinucleotide embedded in consensus ISApl1 target sequences together with the extremely high level of identity between the Moraxella sp. MSG13-C03 chromosome sequence and that of Tn6330 leave little doubt as to the ancestry of Tn6330.

In the current library, there are 30 examples of complete Tn6330 copies, 23 of which exhibited 2-bp TSDs; 59 examples of structures with an upstream ISApl1; only 3 examples with a downstream ISApl1; and 180 examples with no associated IS (Table S1). Analysis of the mcr-1 sequence environment presented here strongly supports the idea that, once acquired, Tn6330 has a strong tendency to decay by undergoing deletion which removes part or all of both downstream and upstream IS copies, resulting in an immobile mcr-1 region trapped in its vector plasmid (Fig. 5 and 6).

Ten structures with an upstream IS copy also included a small segment of the downstream IS copy with the entire right downstream ISApl1 terminal IR (IRR). These were flanked by a 2-bp duplication typical of the target site duplication (TSD) generated by IS30 family members. This suggested that these might be so-called “single-ended” transposons that have been found with different resistance genes and other IS types (4, 5, 23). These sequences represent 5 unique insertion sites (Table S1) and are potentially competent in transposition. In other examples, the downstream IS segment was absent and a sequence showing only limited similarity to IRR, but possessing a potential 2-bp TSD, was present. This raised the formal possibility that these structures could continue to transpose (12, 15, 16, 24). A third type of structure contained the mcr-1 region but no accompanying ISApl1 sequences.

That the structures with only an upstream ISApl1 might transpose stemmed from the observations that some are flanked by typical IS30 family consensus target sequences (12), that they possess an apparent TSD (12), and that, in some, a short sequence abutting the potential TSD resembled the terminal IRR of ISApl1. Analysis of the upstream and downstream sequences in the extended library suggested that all these examples were generated by deletion between a downstream ISApl1 IRR and a deletion joint located in the 3′ end of the pap2 gene (Fig. 4B). A comparison of these derivatives with Tn6330 and empty sites revealed a “ragged” junction between the 3′ end of the mcr-1 region and the area immediately adjacent to the downstream ISApl1 (Fig. 5). The junctions varied from isolate to isolate, implying that the downstream ISApl1 has repeatedly undergone a form of imprecise excision to generate this degree of sequence diversity. Similar conclusions could be drawn for the three examples in which the downstream IS was retained but the upstream was deleted (Table S1). Such behavior had previously been observed for IS30 itself under controlled laboratory conditions (17).

In all cases, deletion involves flanking nucleotides ranging in size from 1 to 4 bp, and a single copy of these remains at the deletion junction (Fig. 5, 6, and 7; highlighted in green). One way of understanding how these events might occur is suggested by the “copy-out–paste-in” transposition mechanism first demonstrated for IS3 family member IS911 (25) and also shown to apply to members of the IS30 family (18) (Fig. 8). A first chemical step in this pathway is cleavage of one end of the IS to generate a 3′OH. In a second step, the free 3′OH attacks the opposite IS end at a small distance (generally the length of the TSD), creating a single-strand bridge (Fig. 8, left) and generating a 3′OH on the donor flank. Note that we have indicated a 2-bp sequence between the abutted IRs (26) (Fig. 8, right) although 1-bp and 3-bp spacers have also been reported (A. Arini, M. P. Keller, W. Arber, unpublished data; cited in reference 17). The 3′OH is then used as a primer in replication (27) to generate a circular IS copy that subsequently integrates into a suitable target. The bridged intermediate shown in Fig. 8 uses the sequence of plasmid pMCR-M17059 (IncI2) as an example and shows how pSCS23 (Fig. 5) might have been generated by template switching during the copy-out transposition replication step. The observation that the vast majority of ISApl1 deletions have a length similar to that of the IS would support a model involving anchoring the two IS ends in close proximity (Fig. S2) followed by abortive transposition in which replication would fail to traverse the IS and generate the circular transposition intermediate.

FIG 8 

A mechanism for ISApl1 deletion. The left hand panel shows part of the ISApl1 transposition cycle. The IS is shown in green and the donor plasmid backbone in black. The terminal IRs are indicated by black boxes and the tips of the IS as green circles. (A to C) The IS ends in the donor plasmid (A) undergo synapsis (B), and one or other end undergoes cleavage to generate a 3′OH which then attacks the opposite end at a position several nucleotides into the donor backbone to generate a molecule in a figure eight formation in which the two IRs are joined by a single-strand bridge (C). (D) A 3′OH generated on the donor plasmid is then used to replicate the IS, generating a double-strand circular transposition intermediate. A strong promoter is generated by the juxtaposition of the two IS ends which drives high levels of transposase expression, facilitating insertion into a suitable target DNA (not shown). The right panel (top) shows the double-strand sequence of the IS ends in IncI2 plasmid pMCR-M17059 as an example. The middle panel presents the structure of the single-strand bridged molecule (as described for panel C) that is shown in the left panel. IS ends are boxed in green. The 3′OH generated in the donor plasmid DNA is indicated by a red dot and the corresponding 5′ phosphate at the other IS end by a black dot. The blue arrow indicates the direction of transposition-associated replication. The deletion joint is shown in blue. The sequence remaining after deletion (bottom) representing plasmid pSCS23 is composed of the bold black characters together with one of the blue tetranucleotide sequences.

Szabó et al. observed with IS30 that this type of deletion occurred as a minor component of all deletions (17). However, they observed similar products, albeit at a 103-fold-lower frequency, when an IS30 derivative with an ablated transposase gene was used. However, in these cases, the vast majority of events proved to be more complex and included large deletions or unidentified plasmid rearrangements, suggesting that the spectrum of events was different in the absence of a functional transposase.

It is interesting that the single-ended Tn6330 variants that retain an intact IRR segment of the ancestral downstream ISApl1 were generated by deletions whose lengths were significantly different from those of the typical 1,070-bp deletions noted for the other variants: 1,030 bp and 979 bp in the 42-bp and 90-bp variants, respectively (Fig. 3). Analysis of these ends revealed that the lengths of the flanking nucleotides where the two strands are joined after ISApl1 excision increased to 7 bp and 10 bp in the 42-bp and 90-bp variants, respectively, offering a possible reason for the existence of these structures (Fig. 3).

We are aware that the flanking sequences giving rise to deletion joints in certain deletion events are not robust and can be as short as a single base pair. However, on the basis of the single-strand bridged structure (Fig. 8), it seems possible that these sequences are not essential but simply assist in resolving a template switching event provoked by an error at the replicative (copy-out) transposition step. This would therefore reflect nonproductive transposition. In this light, if ISApl1, like IS30, can generate spacers of 1 and 3 nucleotides (nt) as well as the typical 2-nt bridge (Arini et al., unpublished; cited in reference 17), it is possible that intermediates with atypical spacers which would change the architecture of the bridged intermediate are less efficient in the “copy-out” process and that it is these that increase the propensity for template switching. With the present facility in DNA sequencing, it would be useful to revisit these issues to determine the host factors involved in this behavior in vivo.

It has been pointed out by Szabó et al. that such deletions “can result in fixation of marker genes” by preventing further transposition (17). ISApl1 appears to be extremely active and was observed to undergo robust transposition in serial mcr-1-positive Escherichia coli isolates from a patient over a period of a month (28). Deletion of the flanking ISApl1 copies would therefore prevent further plasmid rearrangements.

In summary, the results presented here provide a strong framework for understanding how a chromosomal copy of the mcr-1 gene may have been sequestered by ISApl1 to form the composite transposon Tn6330 and how this transposon had subsequently decayed by loss of the upstream and downstream IS copies, reducing the probability of plasmid destruction by ISApl1-mediated rearrangement and also resulting in a stabilized (nontransposable) copy of the resistance gene (Fig. 9).

FIG 9 

The birth and demise of Tn6330. A schematic representation of the birth of Tn6330 following insertion of two copies of ISApl1 into the chromosome of a species closely resembling Moraxella sp. MSG13-C03 followed by successive decay of the transposon due to the loss of one ISApl1 copy and then both ISApl1 copies is shown. The conserved AT and CG dinucleotides that were formed during the first sequestration of the mcr-1 region, and that are now found on the IE of all instances of Tn6330, are highlighted in bold.

MATERIALS AND METHODS

All mcr-1-containing sequences deposited in GenBank through September 2017 were downloaded from the nucleotide and Whole Genome Shotgun (WGS) databases. Only entries with >1,000 bp of flanking sequence on both ends of the mcr-1 gene (n = 273) were subjected to further analysis. Comparative genomic analysis was performed using the Geneious 8 software package (Biomatters Ltd., Auckland, New Zealand). Sequence features were annotated using the Geneious 8 annotation transfer tool, and sequences were aligned using a combination of BLAST (29), MUSCLE (30), progressive MAUVE (31), and LASTZ alignment tools. Average nucleotide identity data were calculated using BLAST (ANIb) and JSpecies (32), core genome alignments were performed using PanSeq (33), and phylogenetic trees were constructed using RAxML (34).