Evolution of a Heavy Metal Homeostasis/Resistance Island Reflects Increasing Copper Stress in Enterobacteria

Abstract

Copper homeostasis in bacteria is challenged by periodic elevation of copper levels in the environment, arising from both natural sources and human inputs. Several mechanisms have evolved to efflux copper from bacterial cells, including the cus (copper sensing copper efflux system), and pco (plasmid-borne copper resistance system) systems. The genes belonging to these two systems can be physically clustered in a Copper Homeostasis and Silver Resistance Island (CHASRI) on both plasmids and chromosomes in Enterobacteria. Increasing use of copper in agricultural and industrial applications raises questions about the role of human activity in the evolution of novel copper resistance mechanisms. Here we present evidence that CHASRI emerged and diversified in response to copper deposition across aerobic and anaerobic environments. An analysis of diversification rates and a molecular clock model suggest that CHASRI experienced repeated episodes of elevated diversification that could correspond to peaks in human copper production. Phylogenetic analyses suggest that CHASRI originated in a relative of Enterobacter cloacae as the ultimate product of sequential assembly of several pre-existing two-gene modules. Once assembled, CHASRI dispersed via horizontal gene transfer within Enterobacteriaceae and also to certain members of Shewanellaceae, where the original pco module was replaced by a divergent pco homolog. Analyses of copper stress mitigation suggest that CHASRI confers increased resistance aerobically, anaerobically, and during shifts between aerobic and anaerobic environments, which could explain its persistence in facultative anaerobes and emergent enteric pathogens.

Introduction

Humans have employed copper over the last eight millennia for tools, disinfection, plumbing, manufacturing, animal husbandry, crop protection, and preservation of perishable commodities (Hobman and Crossman 2015). This usage has exposed many bacteria in the environment and animal microbiomes to unprecedented levels of copper stress. Intracellular copper stress in animal-associated Enterobacteria such as Escherichia coli is modulated by both chromosomal and plasmid-encoded copper trafficking and export mechanisms (Finney and O'Halloran 2003; Nies 2003; Rensing and Grass 2003). The combined activity of these systems maintains homeostasis and compartmentalization of this essential, but potentially toxic, metal. The Es. coli chromosome contains a monovalent copper export system, cus (copper sensing copper efflux system; Munson et al. 2000; Franke et al. 2003), which combines with the muticopper oxidase (MCO) and P-type ATPase of the chromosomal cue system (Rensing et al. 2000; Grass and Rensing 2001; Stoyanov et al. 2001) to detoxify and export copper ions from the cell during aerobic and anaerobic copper stress (Outten et al. 2001; Fung et al. 2013). The cus system, which is particularly important for copper resistance in growth under anaerobic conditions (Outten et al. 2001), shares sequence and functional homology with the silver-resistance system sil in the plasmid pMG101 (Gupta et al. 1999; Silver 2003), isolated from Salmonella typhimurium growing on burns treated with silver nitrate (McHugh et al. 1975). cus is also known to contribute to infection in pathogenic Es. coli (Huang et al. 1999) and is associated with neonatal Cronobacter meningitis but not other Cronobacter strains (Kucerova et al. 2010).

Another copper resistance gene cluster, pco (plasmid-borne copper resistance system), was discovered on plasmid pRJ1004 following isolation from Es. coli growing in Australian piggery effluent on feedlots where copper supplements were employed in growth stimulation and antibiotic roles (Tetaz and Luke 1983). pRJ1004 confers a 3-fold increase in copper resistance over endogenous homeostasis mechanisms to Es. coli cells grown on nutrient agar supplemented with CuSO₄ (Rouch et al. 1985). Analyses of pRJ1004 in Es. coli revealed that inducible copper resistance on this plasmid (Rouch et al. 1985) is mediated by a cluster of seven pco genes (Bryson et al. 1993; Brown et al. 1995), which are homologous to the cop system from plasmid pPT23D in Pseudomonas syringae and the genomic and plasmid-encoded systems in Xanthomonas spp. (Cooksey 1987; Lee et al. 1994; Mellano and Cooksey 1988a; Voloudakis et al. 1993) and contribute to inducible copper resistance under a range of copper stress conditions (Brown et al. 1995; Rouch and Brown 1997).

Here we describe physiological, genomic, and phylogenetic evidence for a transposable element-flanked island of 19 genes that includes adjacent cus and pco clusters, which we call the Copper Homeostasis and Silver Resistance Island (CHASRI). In order to better understand how this compound resistance mechanism emerged, we investigated the patterns of CHASRI sequence diversification and the functional advantage it confers. Phylogenetic analyses suggested the CHASRI was derived from a single event involving the linkage of the two copper resistance mechanisms, likely in a close relative of Enterobacter cloacae. The constituent resistance mechanisms (cus and pco) are themselves the products of associations of multiple pre-existing modules that largely correspond to functional subcomplexes within the copper export systems (fig. 1 and supplementary table S1, Supplementary Material online). Since its assembly, CHASRI has realized a spotty dispersal by horizontal transfer among emergent pathogens in Enterobacteria and other bacteria families under both natural and anthropogenic copper stress. We find that CHASRI confers copper tolerance aerobically, anaerobically, and during transitions between aerobic and anaerobic growth. Molecular dating of CHASRI sequences suggested that increases in anthropogenic copper have influenced its dispersal in bacterial pan-genomes.

Fig. 1.—

Proposed integrated function of CHASRI module genes in homeostasis. The three major mechanisms of copper detoxification encoded in the CHASRI are the periplasmic detoxification pco/CopABCD system, the P-ATPase (silP) involved in efficient cytoplasmic detoxification, and the cus/sil tricomponent cation efflux mechanism. The putative function of pcoA = multicopper oxidase, pcoB = outer membrane transport, pcoC = periplasmic metal binding, pcoD = inner membrane transport, pcoE = periplasmic metal binding, pcoF = periplasmic metal binding, pcoG = M23 metallopeptidase, cusA = RND efflux pump (inner membrane), cusB = RND efflux pump (periplasm), cusC = RND efflux pump (outer membrane), cusF = periplasmic metal binding, silE = periplasmic metal binding. DUF411 and RS modules (pcoRS and cusRS) involved in regulatory processes are not shown. Homology with other study systems is detailed in supplementary table S1, Supplementary Material online.

Materials and Methods

Plasmid Sequencing

Using the pco operon as a starting point, a polymerase chain reaction (PCR)-based gene-walking strategy was used to recover over 17,000 bp of plasmid pRJ1004 (GenBank accession X83541.1) upstream of pcoA. Twenty-nine overlapping primer sets were used to generate sequences using an ABI 3730 High-Throughput DNA Sequencer, which were then assembled manually (GenBank accession # KC14966). This region was then searched for predicted open reading frames (ORFs) using NCBI ORF Finder (http://www.ncbi.nlm.nih.gov/gorf/gorf.html, last accessed March 2, 2016). The contents of the pRJ1004 plasmid were further investigated using an Illumina HiSeq strategy. Plasmid DNA was sheared via sonication using a Bioruptor (Diagenode) at high setting for 15 min in cycles of 30 s on and 30 s off. Library preparation was carried out as previously described (Gibbons et al. 2009) with the following modifications: i adapter ligated plasmid DNA was run out on a 2.5% agarose gel (Fisher’s Nucleic Acid Recovery Agarose) in 80% glycerol loading buffer at 80 V for 2 h and ii the library was amplified for 12 cycles of PCR. The final library was quality controlled for integrity, size, concentration, and molarity on a Bioanalyzer (Agilent). The library was run, with other samples for 57 cycles, on a single lane of an Illumina HiSeq 2000 at the Vanderbilt Genome Sciences Resource (NCBI BioProject # PRJNA217802).

Detection of Related Clusters and Plasmids

Using copper cluster-associated amino acid sequences as queries, we searched for homologous protein models in a database of 1,150 prokaryotic proteomes (supplementary table S2, Supplementary Material online), including both plasmids and chromosomes, using BLASTP (Altschul et al. 1997), retaining sequences that were greater than 45% similar at the amino acid level and between 40% and 150% the length of the query. Protein families were delimited using OrthoMCL (version 1.4) (Li 2003) at an inflation value of 2.0. Clustering of cus and pco gene homologs was inferred when two or more homologs were separated by no more than six intervening genes as described previously (Campbell et al. 2012).

Similarity between the pRJ1004 genome and other plasmids was investigated with a mapping strategy in which 51 bp reads were used as BLASTN queries against 2,556 sequences of the plasmid database at NCBI (supplementary table S3, Supplementary Material online), and locations with greater than 90% identity and overlap were considered homologous to the query.

Phylogenetic Methods

Gene family amino acid sequences were aligned using mafft (version 6.847) (Katoh et al. 2002; Katoh and Toh 2008), and maximum likelihood (ML) analysis was performed in RAxML (version 7.2.8) (Stamatakis 2006) under the PROTGAMMAJTT model of amino acid substitution. Phylogenetic analysis, which included additional sequences retrieved from NCBI nr via BLASTN, was also performed on nucleotide sequences of the region corresponding to silE, cusSRCFBA, and silP in RAxML under the GTRCAT model of evolution. Support for amino acid and nucleotide analyses was assessed with 100 bootstrap replicates in RAxML. A bacterial species phylogeny was constructed with a database-wide alignment of RNA polymerase, beta prime subunit subjected to 100 neighbor-joining bootstrap replicates in PAUP* 4.0b (Swofford 2003). Correlations between gene family phylogenies were measured by multiplying the proportions of shared reciprocal best-BLAST hits (rbh) between pairs of proteins out of the total number of rbh genetic element (chromosome or plasmid) pairs for each protein in the comparison. We performed a molecular clock analysis on the nucleotide alignment in BEAST v. 1.7.1 (http://beast.bio.ed.ac.uk/beast, last accessed March 2, 2016) with the following assumptions: An uncorrelated relaxed molecular clock, lognormal rate distribution, HKY model of nucleotide substitution with four gamma categories, normally distributed priors, a Yule process tree model, and UPGMA starting tree. Shewanella putrifaciens strain CN32, isolated from shale sandstone 230 m below ground level in New Mexico (Fredrickson et al. 1997), has been estimated to be 2,060 to 3,500 years removed from contact with the surface given the local rate of ground water movement (Walvoord et al. 1999). We used a groundwater age estimate of 2,060 years as a minimum divergence time (normally distributed with a standard deviation of 100 years) between the deep subterranean S. putrifaciens CN-32 and Es. coli A2363. We separately applied the minimum divergence time to the split between the deep subterranean and estuary Shewanella species ANA-3. We ran eight independent Markov chain Monte Carlo (MCMC) simulations for ten million generations each and assessed convergence of log-likelihood values in Tracer, discarding the first 10–40% of generations. We combined trees in LogCombiner version 1.7.1 and summarized 95% highest posterior density range distributions for node ages in TreeAnnotator version 1.7.1. Effective sample sizes exceeded 200 for all parameters. A lineage through time plot was computed in the APE package v3.0-4 (http://ape-package.ird.fr, last accessed March 2, 2016) from the maximum clade credibility tree drawn in TreeAnnotator v1.7.1 using median node heights.

Copper Challenge Experiments

We compared the copper tolerance of cells containing either the entire CHASRI gene cluster, as found on plasmid pRJ1004 (Tetaz and Luke 1983), or part of the cluster, from constructed plasmid pCOIV239-B1 (Munson et al. 2000), by growth assays on solid media and by determining growth rates in liquid media. As described in the Results section, the incomplete CHASRI cluster in pCOIV239-B1 contains pcoABCDRSE but is missing pcoFG and the plasmid-encoded region of cus. The parent vector for pCOIV239-B1 is derived from the pSC101 family, so the low copy number (Cohen and Chang 1977) should be comparable with that of a large, naturally occurring plasmid-like pRJ1004. Plasmids pRJ1004 and pCOIV239-B1 were transformed into Es. coli strains BW25113 and W3110 and successful transformants were selected by plating on lysogeny broth (LB) media (Difco) plus agar (BD) with the appropriate selection agent (170 mg/L Chloramphenicol for pCOIV239-B1, 7.5 mM CuSO₄ under aerobic growth in BW25113 and 2 mM CuSO₄ under anaerobic growth in W3110 for pRJ1004). For aerobic challenges, LB media plus agar and a 100 mM CuSO₄ solution (adjusted to pH 7.0 with triethanolamine to avoid precipitation and metal complexation; Magyar and Godwin 2003) were combined to make 0–50 mM CuSO₄ solid media in nontreated 96-well plates (BD). A 5 ml starter culture of each construct in LB media was grown with shaking at 37 °C for 6 h and 5 µl of culture was used to inoculate designated wells on each plate. The plates were incubated overnight at 37 °C. After 12 h, the wells were examined and any colony growth was considered successful tolerance of the included copper level (Pontel and Soncini 2009). In anaerobic copper challenge experiments, LB media was purged of oxygen through three cycles of vacuum evacuation/nitrogen replacement and transferred to a glove bag (I2R Model X-37-37) purged with nitrogen. Starter cultures in LB media of each construct were transferred into a glove bag and used to inoculate test tubes of 5 ml LB media. The test tubes were sealed with rubber septa and incubated overnight at 37 °C. LB media plus agar and a 100 mM CuSO₄ solution (adjusted to pH 7.0 with triethanolamine) were combined to make 0–50 mM CuSO₄ solid media in nontreated 96-well plates. The wells were inoculated with 5 µl aliquots from 5 ml anaerobic starter cultures in LB media, the plates were sealed inside a Desi-Vac container (Fisher) with an AnaeroPack (MGC), and the entire setup was stored at 37 °C. To ensure anaerobic growth, O₂ concentration in the sealed container was monitored with a ToxiPro O₂ sensor (Honeywell) and was shown to be below 0.1% for the duration of the experiment. After 48 h, the plates were examined to determine the highest level of copper tolerated by each construct.

Another set of experiments assessed aerobic growth following anaerobic copper challenge. LB media was purged of oxygen through three cycles of vacuum evacuation/nitrogen replacement and transferred to a glove bag purged with nitrogen. Starter cultures in LB media of each construct were taken into a glove bag and used to inoculate test tubes of 5 ml LB media. The test tubes were sealed with rubber septa and incubated overnight at 37 °C. LB media and a 100 mM CuSO₄ solution (adjusted to pH 7.0 with triethanolamine) were purged of oxygen through three cycles of vacuum evacuation/nitrogen replacement and transferred to a glove bag purged with nitrogen. Under anaerobic conditions, LB media and a 100 mM CuSO₄ solution (adjusted to pH 7.0 with triethanolamine) were combined to make 90 µl solutions of CuSO₄ media in nontreated conical bottom 96-well plates (Granier). The wells were inoculated with 10 µl aliquots from 5 ml anaerobic starter cultures in LB media (bringing the final sample volume to 100 µl and the final copper concentration from 0 to 10 mM), the plates were sealed inside a Desi-Vac container with an AneroPack, and the container was incubated at 37 °C. After 90 min, the plates were removed from anaerobic conditions and spun down at 2,500 rpm for 15 min in an Allegra 6R centrifuge (Beckman Coulter). The supernatant was removed and each pellet was resuspended in 100 µl LB or LB + CuSO₄ media. The plates were incubated at 37 °C with shaking for 10 h, with optical density readings at 600 nm taken every hour on a Synergy 2 microplate reader (Bio-Tek) as a proxy for cell number. Bacterial growth was fitted with the equation Y = Y_M × Y₀/((Y_M − Y₀)e^{(−k(x−x0))} + Y₀) (Vandermeer 2010) and growth rate constant (k) and growth lag (x₀) were compared using the extra sum-of-squares F test, implemented in GraphPad Prism version 5.04 for Windows (GraphPad Software, La Jolla, CA).

Results

A Multipart Resistance Island on pRJ1004 Confers Resistance to Copper during Changes in Environmental Oxygen

Targeted sequencing of the pRJ1004 plasmid in Es. coli K-12 revealed a Copper Homeostasis and Silver Resistance Island (CHASRI) of 19 genes, 6 of which are homologous to the genomic Es. coli cus cluster (cusABFCRS) (Munson et al. 2000) and two more that complete homology with the sil heavy metal export system cluster (silE and silP) (Gupta et al. 1999). The 11 remaining genes include the known pco gene cluster (pcoABCDRSE) (Brown et al. 1995) and 2 additional pco genes (pcoF and pcoG) transcribed together from the P_pcoF promoter in a copper-inducible manner and regulated by PcoRS (Munson 1997). BLAST-based mapping of Illumina short reads (fig. 2) revealed that pRJ1004 is most similar overall to plasmid ColIb-P9 from Es. coli K12 (NC_002122) and related plasmids in Escherichia, Shigella, and Salmonella. The pRJ1004 reads that mapped to the CHASRI did not show any similarity with ColIb-P9 or the other similar plasmids, but the island-mapped reads were very similar to isolated regions in nine unrelated plasmids ranging in size from 54 (NC_013285) to 279 kbp (NC_008573). The plasmid-encoded CHASRI in En. cloacae is 82.5% identical to the icus over the alignable region.

Fig. 2.—

Depth of reads mapping to the plasmid database at NCBI. (A) The nine plasmids to which the most 50-bp Illumina HiSeq reads mapped that also mapped to the CHASRI region of pRJ1004. (B) The nine plasmids to which the most total pRJ1004 reads mapped. ¹Plasmid overall synteny type. ²Depth of reads prefiltered to map to CusPco region of pRJ1004 is shown above axis. ³Depth of total mapped reads from pRJ1004 are shown below axis. ⁴GenBank accession number of plasmid. ⁵Bacterial strain of plasmid isolation (where published). Reads are shown for 1,000-bp bins.

In order to determine potential selective advantages conferred by CHASRI, we conducted several copper challenge experiments. The maximum tolerable concentration (MTC) for two Es. coli strains transformed with plasmids containing complete or partial CHASRI genes was established under aerobic and anaerobic growth conditions. Escherichia coli strains BW25113 and W3110, wild type strains that display dissimilar copper tolerance profiles (Outten et al. 2001; Pontel and Soncini 2009), were transformed with either plasmid pRJ1004, containing all of CHASRI (Tetaz and Luke 1983), or plasmid pCOIV239-B1, containing only pcoABCDRSE (Munson et al. 2000), and grown on copper-containing LB media to determine resistance levels (table 1). For strain BW25513, plasmids pRJ1004 and pCOIV239-B1 conferred aerobic copper resistance 40% (P = 0.0011) and 90% (P = 0.00011) greater than wild-type copper tolerance, respectively. Under anaerobic growth conditions, none of the three constructs experienced inhibition of growth up to the highest concentration tested, 50 mM CuSO₄. In W3110, neither plasmid improved on the 15 mM wild-type aerobic copper resistance. Although our plasmid aerobic copper resistances were comparable with previously reported values (Tetaz and Luke 1983), our wild-type W3110 resistance consistently exceeded published levels (Grass and Rensing 2001), potentially affecting the significance of the comparison. Under anaerobic growth, both pRJ1004 and pCOIV239-B1 increased resistance to copper 50% over wild-type levels (P < 0.01 and P < 0.01, respectively). When these MTC experiments were repeated on minimal media (Fung et al. 2013) supplemented with copper, no significant difference was detected between the wild type and transformed strains across the range of copper concentration tested. Although a variety of point mutations and inversions in the W3110 genome are known to give rise to deleterious growth rate phenotypes in minimal media (Jensen 1993), similar variations in BW25113 have not been described and the lack of a significant difference between our constructs deserves further consideration.

Table 1.

Level of CuSO₄ Resistance Conferred by Constructs in Escherichia coli

Construct	Copper Modules Present	Aerobic Copper Tolerancea (mM)	Anaerobic Copper Tolerance (mM)
BW25113	cue, icusb	6.7 ± 0.3	≥50 ± 0.0
BW25113/pRJ1004	cue, icus, pco, cus	9.3 ± 0.3c	≥50 ± 0.0
BW25113/pCOIV239-B1	cue, icus, pcoABCDRSE	12.7 ± 0.3c,d	≥50 ± 0.0
W3110	cue, icus	15 ± 0.0	2.0 ± 0.0
W3110/pRJ1004	cue, icus, pco, cus	15 ± 0.0	3.0 ± 0.0c
W3110/pCOIV239-B1	cue, icus, pcoABCDRSE	15 ± 0.0	3.0 ± 0.0c

^aCopper tolerance is determined by the presence of colony formation units on LB agar + CuSO₄ (adjusted to pH 7.0 with triethanolamine) plates. Number reported is an average of three replicates.

^bicus refers to the independent chromosomally inherited cus cluster, whereas cus refers to the CHASRI-associated cus cluster.

^cSignificant increase in copper resistance over wild type (P < 0.05).

^dSignificant increase in copper resistance over pRJ1004 (P < 0.05).

This initial result demonstrated an advantage of CHASRI over wild-type copper resistance, but to understand the importance of the entire gene island over its constituent parts, we tested the same constructs from the MTC experiment in a growth trial that included a switch in oxygen availability during the experiment. When BW25113 cells were grown anaerobically and then switched to aerobic growth (fig. 3), there is a significantly slower growth rate for cells transformed with pCOIV239-B1, which contains only the pcoABCDRSE component of CHASRI, relative to those transformed with pRJ1004, which contains the intact CHASRI. We conclude that the intact CHASRI has an advantage over a partial construct under this combination of copper and oxygen stress, with one caveat: While both plasmids have low copy numbers, we cannot exclude other contributions from the different parent vectors. Compared with wild-type BW25113 cells and cells with pRJ1004, cells with pCOIV239-B1 showed a significant growth lag (P < 0.0001 and P < 0.0001, respectively) but increased growth rate (P = 0.0420 and P = 0.0093, respectively) with no copper added to the media (supplementary table S4, Supplementary Material online). When BW25113 cells were grown anaerobically in LB media containing 5 mM CuSO₄ then switched to aerobic growth in LB with no copper added, cells with pCOIV239-B1 again showed a significant lag in growth compared with both wild-type cells and cells with pRJ1004 (P < 0.0001 and P < 0.0001, respectively) and again demonstrated a significantly faster growth rate than wild-type cells and cells with pRJ1004 once the initial lag was overcome (P < 0.0001 and P = 0.006, respectively). Finally, when BW25113 cells were grown anaerobically in LB media containing 5 mM CuSO₄ then switched to aerobic growth in LB with 5 mM CuSO₄ added, cells with pCOIV239-B1 showed both a significant lag in growth compared with both wild-type cells and cells with pRJ1004 (P < 0.0001 and P < 0.0001, respectively) and a slower growth rate following the lag period (P < 0.0001 and P < 0.0001, respectively).

Fig. 3.—

Copper tolerance of wild-type Escherichia coli BW25113 with resistance plasmids pRJ1004 and pCOIV239-B1. Growth curves mark 10-h aerobic growth after 90 min anaerobic incubation, with error bars indicating standard error of the mean for three replicates.

In parallel experiments using the W3110 strain, cells containing pCOIV239-B1 showed no growth lag but had a slower growth rate than wild-type cells or cells with pRJ1004 (P = 0.003 and P = 0.0013, respectively) when grown in LB media with no copper added (supplementary fig. S1, Supplementary Material online). W3110 cells with pCOIV239-B1 showed delayed growth relative to wild-type cells and cells with pRJ1004 (P = 0.0033 and P = 0.002, respectively) and also grew more slowly than the wild-type and pJR1004 cells (P = 0.0018 and P = 0.0434, respectively) when grown in LB media containing 2 mM CuSO₄ anaerobically then grown aerobically in LB with no copper added. W3110 cells were also grown anaerobically in LB media containing 2 mM CuSO₄ then switched to aerobic growth in LB with 2 mM CuSO₄. Under these conditions, cells with pRJ1004 had the highest growth rate, but this difference was only statistically significant in comparison with cells containing pCOIV239-B1 (P < 0.0001). Additionally, cells with pCOIV239-B1 showed a longer lag period when compared with wild-type cells and cells with pRJ1004 (P = 0.0333 and P = 0.0088, respectively). These trials revealed a statistical equivalence between wild-type cells and cells with pRJ1004 but a significant growth lag for cells with pCOIV239-B1 following a switch from anaerobic to aerobic growing conditions in both BW25113 and W3110 wild-type strains. These results suggest that the advantage of the complete CHASRI over a partial module in response to copper stress is most apparent in growth conditions where the availability of oxygen changes and show that the extent of the advantage can be dependent on the strain background. Combined with the results from the MTC experiments, these studies reveal a CHASRI phenotype that is robust to copper stress under a range of oxygen environments. However, the specific role of each component in copper and silver tolerance under different growth conditions remains to be resolved. In order to understand the evolutionary processes underpinning the formation these functions, we investigated the patterns of CHASRI gene diversification and synteny.

CHASRI Has a Spotty Distribution among Enterobacteria

We detected 1,099 related clusters (supplementary table S5, Supplementary Material online) of at least 2 homologs of CHASRI genes from a database of 1,150 complete bacterial genomes including their corresponding plasmids (supplementary table S2, Supplementary Material online). CHASRI has a sparse but taxonomically broad distribution (fig. 4) in the Enterobacteriaceae (gammaproteobacteria) and is found at least as often on plasmids as on chromosomes (figs. 5 and 6). However, the constituent cus and pco clusters are more likely to be located on chromosomes. The island was found in Es. coli, En. cloacae, Klebsiella pneumoniae, Cronobacter turicensis, Cronobacter sakazakii, and Serratia marescens strains, all but one strain of which (Es. coli 8739 origin not known) are facultative anaerobic human or animal pathogens. CHASRIs detected in two Shewanella spp. isolated from the environment have a slightly different assemblage of genes, but were included as part of the overall distribution based on phylogenetic analysis results, which suggest that the original pco genes were replaced by an alternative set of genes including highly divergent homologs of pcoA, pcoB, pcoC, and pcoD. Shewanella islands also differ by containing a single two-component regulatory (RS) module, unlike other islands which contain RS modules associated with both the cus and the pco components. Transposon sequences usually directly flank (within two ORFs) both ends of CHASRIs (supplementary table S6, Supplementary Material online), with notable exceptions: 1) The Shewanella islands have three intervening sequences, one of which (YP_001181790) is annotated as a transposase that adjoins the expected recombination breakpoint; and 2) one of the two islands in En. cloacae ATCC 13407 is characterized by three transposon domains, which flank the pco cluster and also a complex of the cus cluster joined to a mercury resistance cluster on both sides.

Fig. 4.—

Distribution and structure of CHASRIs. (A) Distribution of CHASRI and icus among proteobacteria. (B) To scale alignment of CHASRIs and recent ancestral modules detected in complete gammaproteobacteria proteomes. Genes with multiple colors are homologous to other genes in the cluster with any matching colors (i.e., pcoE, pcoF, and silE are homologous sequences).

Fig. 5.—

Diversity of CHASRI proteins. (A) ML amino acid phylogenies of CusB. (B) ML amino acid phylogenies of PcoB. Support values out of 100 ML bootstrap replicates are shown when ≥70. Color key displays clustering with functionally related genes. (C) The left panel is a heatmap of correlation of gene phylogenies expressed as the product of the proportion of shared reciprocal best BLAST hits for each gene, normalized by row. The right panel shows the number of homologs detected and analyzed for each gene.

Fig. 6.—

The association between plasmid encoding and aggregation of modules. Graph of the number of clusters shown in center panel of pco (left panel) and cus (right panel) genes. CHASRI clusters may not be of the exact synteny shown, but correspond to those in figure 4.

The cus cluster, cus-S-RCFBA (genes transcribed in reverse indicated with “-”), is usually found independent of pco genes and restricted to the chromosomes of species in the Enterobacteriaceae (figs. 2 and 4). Independent cus clusters (icus) are fairly conserved within this lineage, with the notable exceptions of five Shigella spp. where it has experienced insertion, deletion, or complete loss, and also its absence from Salmonella, Dickeya, Pantoea, and Erwinia. A similar cluster, czcBA.RS (up to six intervening genes indicated with “.”), involved in cobalt, zinc, and cadmium resistance (van der Lelie et al. 1997), is found in several lineages of betaproteobacteria and gammaproteobacteria, sometimes associated with homologs of cusF (a periplasmic metal-binding protein hypothesized to be involved in sequestration) (Xue et al. 2008) and/or cus/czcC (an outermembrane cation exporter) (Kim et al. 2010). In Enterobacteriaceae, canonical pco clusters (pcoABCDRS) are restricted to CHASRI, although pcoABCD is more widely distributed in the proteobacteria, often as pco-S-RABCD.

CHASRI Resulted from Sequential Combinations of Pre-existing Gene Modules

Phylogenetic analyses (fig. 5 and supplementary fig. S2A–S, Supplementary Material online) suggest that several collinearly transcribed constituent gene modules (pcoAB, pcoCD, cusBA, cus/pcoRS) were associated prior to the formation of the CHASRI and its constituent clusters. The most common association, cusBA, codes for an inner-membrane transporter and adapter (Rensing and Grass 2003) and is supported by these analyses to be homologous to czcBA modules. Homologs of cus/czcC, which complete the tripartite efflux complex as the outer membrane component for metal export (Su et al. 2011), are usually found upstream of cusBA homologs. pcoABCD contains a more widely distributed association, pcoAB, homologs of which are found in both beta and gammaproteobacteria. Evidence of alternative pairing of pcoAB homologs with other genes can be found; in the Acinetobacter baumannii AB0057 genome (NC_011586) this module is adjacent to a cus/czcC homolog (YP_002318156.1), of the cation efflux family cl00316, which is involved in the export of Co, Zn, and Cd (van der Lelie et al. 1997). pcoCD function has not been well characterized, but its requirement in copper resistance has been shown to be variable between cop and pco homologs (Brown et al. 1995). Homologs of two-component sensor–regulator modules (RS) found in both cus and pco clusters are also found in multiple genomic contexts.

The distribution of putative module-associated genes shows bias toward presence in these modules and the modules show bias toward presence in clusters (supplementary table S5, Supplementary Material online). For example, 43 of 45 pcoD homologs detected are part of a precise pcoCD module and 32 homologs are part of a precise pcoABCD cluster. Similarly, out of 456 homologs of cus/pcoS detected, 410 homologs are part of precise RS modules, and out of 599 cus/czcB homologs detected, 454 homologs are part of precise cus/czcBA modules. Out of 101 cusF homologs detected, 79 homologs are clustered with cus/czcBA modules.

Although the modules inferred here are more often found on chromosomes than plasmids, plasmid occurrence rises with increased clustering. CHASRIs are enriched for plasmid occurrence (55% when including Shewanella islands) compared with all constituent modules.

CHASRI Evolution Is Distinct from Its Constituent Modules

Phylogenies of genes participating in the same putative module are highly correlated (fig. 5A and B). This is also supported by strong associations by shared reciprocal best BLAST hits between homologs of cusA and cusB, pcoA and pcoB, pcoC and pcoD, and cus/pcoR and cus/pcoS (fig. 5C). Associations are not as strong between pairs of genes belonging to alternative putative modules.

The adjacent taxon of CHASRI cus genes in gammaproteobacteria is usually the homolog from the chromosomally inherited cus cluster (icus) in En. cloacae (fig. 5A and supplementary fig. S2L–Q, Supplementary Material online), generally with strong bootstrap support, although the position of the En. cloacae icus sequence is sometimes within the CHASRI clan. Enterobacter cloacae icus genes cusR, cusS, cusC, cusF, cusB, and cusA are monophyletic with CHASRI cus genes with 92%, 100%, 98%, 96%, 100%, and 100% ML bootstraps, respectively, excluding all other instances of icus. icus genes are more evenly distributed among Enterobacteria taxa and do not appear to conflict with the species phylogeny when CHASRI genes are excluded.

cus-R-SCFBA genes in clusters (icus and CHASRI) split from all other assemblages of their constituent genes in ML trees, but genes in pcoABCD clusters appear in six clans in highly correlated pcoA and pcoB phylogenies among both betaproteobacteria and gammaproteobacteria (fig. 5 and supplementary fig. S2, Supplementary Material online) with varying levels of bootstrap support.

The homologs of genes directly participating in CHASRIs consistently form a well-supported clan, with the notable exception of two Shewanella spp. islands, in which the cus cluster, pcoF, pcoG, and silP are nested between an alternatively situated highly divergent pcoABCD cluster with an additional highly divergent silP homolog present in S. putrifaciens CN-32. ML phylogenies of combined nucleotide alignments illustrate that CHASRIs have experienced very little divergence (fig. 5 and supplementary fig. S2, Supplementary Material online), with at most 0.067 substitutions/site in GTR + gamma-optimized ML distance. The taxa represented within this small clan are diverse, including Shewanella, Escherichia, Cronobacter, Enterobacter, Klebsiella, and Salmonella, and within all but Salmonella both plasmid- and chromosome-encoded islands are found. Shewanella sequences formed a well-supported clan with Es. coli plasmid-encoded sequences (100% ML bootstraps). A notable, high-level sequence similarity exists within a clan of CHASRIs from animal and human pathogens, including Salmonella isolated from a cow in 2001 and Cronobacter from neonatal meningitis carried by a milk formula (Kucerova et al. 2010), also isolated in 2001.

Several other metal homeostasis genes are also constituents of the CHASRI. One gene family, containing pcoE, pcoF, and silE, was detected only in Salmonella spp. and CHASRIs (except for one divergent silE homolog adjacent to the icus cluster in Edwardsiella tarda), where it appears to have undergone early diversification from a single original sequence. pcoE, a copper-binding periplasmic protein, exhibits the least sequence variation among taxa, followed by pcoF and silE. pcoG is a member of a large metalloendopeptidase family, for which over 1,000 sequences were recovered in the genome database. pcoG is often associated with silP, a putative copper-translocating P-type ATPase potentially cotranscribed with cusCBFA, for which over 1,100 sequences were recovered in the genome database.

A relaxed molecular clock (fig. 7A) was calibrated with a minimum date of 2,060 years at the divergence of S. putrifaciens strain CN32 and the closest related sequences from Es. coli. Shewanella putrifaciens strain CN32 was isolated (fig. 7B) from shale sandstone 230 m below ground level in the Morrison Formation in northwest New Mexico (Fredrickson et al. 1997) and contains a modified CHASRI that is phylogenetically nested in a clan of CHASRIs detected in farm animal and human-associated strains of Es. coli. Such a transfer of genetic material from the surface to deep subterranean rock microbiota would require an estimated 2,060 to 3,500 years given the local rate of ground water movement (Walvoord et al. 1999). The molecular clock estimates the beginning of CHASRI diversification at 5,477 years ago (4,659–16,370, 95% highest posterior density range [HPDR]), the divergence of Shewanella CHASRIs at 320 years ago (10–2,200 HPDR), and divergence of pco-associated cus from icus in En. cloacae at 6,946 years ago (2,438–58,439 HPDR). These age estimates are consistent with a total mutation rate of 2.5 × 10⁻⁸ mutations per site per generation (assuming 200 generations/year), which is significantly faster than the current estimate for nonsynonymous mutations in the greater Es. coli genome of 8.9 × 10⁻¹¹. Recently, the rate of sequence evolution was shown to scale with decreasing generation time in some bacteria (Weller and Wu 2015), and the observed mutation rate could also reflect horizontal transmission and selection (Wielgoss et al. 2011). A lineage through time plot of the median node heights from the BEAST analysis shows a general pattern of increased rate of CHASRI diversification coinciding with increased rates of copper deposition (Hong et al. 1996). Furthermore, three sharp accelerations and two intervening decelerations observed in CHASRI diversification could correspond to three major peaks of copper production (fig. 7C, Roman Empire, Chinese Sung Dynasty, and post-Industrial Revolution). Diversification among livestock-associated CHASRIs is reconstructed to be within the last century, while human-associated diversifications are both ancient and modern.

Fig. 7.—

Origin and diversification of CHASRIs. (A) Chronogram generated in BEAST version 1.7.1 (Drummond and Rambaut 2007) from a CHASRI multigene nucleotide alignment. The aligned region includes silE, cusSRCFBA, and DUF411. Support values are out of 100 ML bootstraps. Taxon name coloring corresponds to plasmid synteny according to Mauve alignment (supplementary fig. S3, Supplementary Material online). Scale is calibrated with a 2,060-year minimum divergence between subterranean CHASRI and the most closely related surface CHASRI. (B) Description of CHASRI-containing strains in (A). Bold “E”, “A,” and “H” denote isolation in environment, agriculture, and hospital, respectively. Additional metal resistances are as reported in genome publications (parentheses indicate minimum resistance where complete genome was not available at the time of analyses.). (C) Graph of estimated global copper deposition due to human activity, reproduced from Hong et al. (1996). (D) Reconstruction of dispersal of CHASRIs, based on branching order in figure 5 and (A), and plasmid distributions shown in (A). A parsimonious reconstruction suggests that a chromosomal cus became linked with a plasmid pco in Enterobacter, and the resultant cluster was then distributed to new plasmids following insertion into chromosomes. Integration into a Shewanella chromosome resulted in recombination of CHASRI cus with native Shewanella pco, and sil clusters resulted from loss of pco from CHASRI.

Discussion

A Copper/Silver Resistance Island Arose in a Close Relative of Enterobacter cloacae and Subsequently Experienced Dispersal by HGT

A copper homeostasis and silver resistance island (CHASRI), composed of cus and pco gene clusters and accessory genes situated between transposable elements, can more than triple copper resistance when conjugally transferred via plasmid to Es. coli that already possess chromosomal cus and cue copper resistance mechanisms (Tetaz and Luke 1983; Rouch et al. 1985). Analyses presented here suggest that the CHASRI formed by recombination between a chromosomally inherited cus cluster (icus) in a close relative of En. cloacae and a pco cluster most closely related to sequences found on chromosomes in Pseudomonas spp. These analyses further supported rapid horizontal transfer of the CHASRI following assembly, which we detected by several methods (Zhaxybayeva and Doolittle 2011). Evidence for HGT dispersal of CHASRI is summarized as follows: 1) CHASRI as a cluster has a very spotty distribution compared with icus, which is more commonly found on chromosomes of Enterobacteria; 2) CHASRI phylogenies conflict with corresponding species phylogenies, while icus clusters are consistent; 3) CHASRI genes show very little divergence between different genera, compared with icus clusters, suggesting that CHASRI diversification is more recent than generic divergences in Enterobacteria; 4) an analysis of complete bacterial genomes revealed that homologs of the modules that make up the CHASRI are primarily located on chromosomes, but plasmid-encoded CHASRIs outnumber chromosomal copies; and 5) CHASRIs, like other resistance islands, are often flanked by transposon sequences, which are suggestive of transferability (Paauw et al. 2010).

HGT of the CHASRI entails repeated exchange between plasmids and chromosomes, culminating in its presence on chromosomes in four genera and at least five distinct plasmid types. However, plasmids most similar to pRJ1004, the plasmid interrogated in this study, do not contain CHASRI. Conversely, plasmids with the most similar islands have no other regions of strong similarity with pRJ1004, suggesting CHASRI is not merely a passenger on a successful plasmid, but rather is itself a distinct object of selection. The silver-resistance cluster (sil) on Salmonella typimurium plasmid pMG101 (Silver 2003) and a convergently derived sil-like island on Es. coli A2363 plasmid pAPEC-O2-R (Johnson et al. 2005) arose after loss of pco genes from CHASRIs, which is supported by both phylogenetic analyses and the inclusion of silP and a member of the DUF411 gene family, which are otherwise only associated with cus clusters that are in CHASRIs.

Functional Integration of Interchangeable Modules Leads to Niche-Specific Functions

Analyses of CHASRIs in bacteria revealed several pairs of functionally coupled genes that are phylogenetically correlated and show strong conservation of synteny across a wide range of taxa. We consider these pairs to be functionally integrated modules retained by natural selection. These modules have been shown to have similar roles but different functions when found with different genomic neighbors. For example, the two-component regulatory systems CusRS and PcoRS share a substantial identity (60% for R, 37% for S) but they participate in regulation of different resistance pathways. Furthermore, two-component regulatory systems homologous to Cus/PcoRS in prokaryotes (Albright et al. 1989; Parkinson and Kofoid 1992; Mizuno 1997; Zhou et al. 2003) regulate a wide range of resistance genes and clustering of R and S is usually, but not always, ancestral (Koretke et al. 2000). This suggests that the functions of the two constituent genes are tightly linked, but the module itself is adaptable to diverse genomic and ecological roles. Similarly, resistance-nodulation-cell division (RND) efflux pumps (such as Cus/Czc) have been studied for their ability to detoxify a wide range of substrates. In addition to heavy metals, RND efflux pumps like the AcrAB-TolC system in Es. coli and MexAB-OprM system in Pseudomonas aeruginosa have been shown to export antibiotics including tetracycline and chloramphenicol, β-lactams, fusidic and nalidixic acid, fluoroquinolones, chemotherapeutics, sodium dodecyl sulfate (SDS), Triton X-100, and bile salts (Ma et al. 1995; Tsukagoshi and Aono 2000). In P. aeruginosa, a variety of outer membrane permeases (OMPs) have been shown to be interchangeable with certain AB modules (Murata et al. 2002). This functional interchangeability of modules is consistent with the reduced phylogenetic correlation observed between OMPs and AB modules, suggesting this is an evolutionary mechanism for the emergence of novel efflux pathways.

Although the mechanistic roles of some of the pco constituents are unclear, PcoC and PcoD contribute to copper transport and ensure the full functionality of PcoABCD (Brown et al. 1995), similar to homologs CopC and CopD (Mellano and Cooksey 1988b; Cha and Cooksey 1993). PcoC is a small periplasmic protein that binds both Cu(I) and Cu(II) ions and interacts with MCO PcoA to oxidize Cu(I) to Cu(II) (Huffman et al. 2002; Wernimont et al. 2003), whereas the function of PcoD is still unknown. PcoCD homologs occur as the fusion protein YcnJ in Bacillus subtilis, further supporting the functional linkage of these genes (Chillappagari et al. 2009). The combination of mechanisms that contribute to the spatial clustering of functionally linked genes has been discussed widely (Lawrence and Roth 1996; Hurst et al. 2002; Baquero 2004); our results are consistent with horizontal transfer, linkage, and coexpression as drivers of clustering.

CHASRI Is a “Winning Pattern” Module Combination that is Successful Under Human-Induced Copper Pressure

CHASRI is the most complex aggregation of copper resistance modules known and is highly prone to dispersal by horizontal transfer. The potential for complementation in this aggregation of gene modules may be directly linked to CHASRI success. For example, coordination of periplasmic and extracellular copper transport could explain the increased resistance conferred by CopABCD combined modules over CopAB modules (Mellano and Cooksey 1988a). The cue system, which detoxifies copper with an MCO in the presence of oxygen (Outten et al. 2000; Grass and Rensing 2001), co-occurs with icus systems for efflux under aerobic and anaerobic conditions (Outten et al. 2001) in facultative anaerobes. CHASRIs combine MCO copper detoxification by pco with efflux by cus in a single island. We have shown that the CHASRI on pRJ1004 enhances aerobic copper resistance over wild-type BW25113 cells but provides less resistance than a plasmid that contains only pco. However, cells with the pco-only construct are severely inhibited upon transition from anaerobic to aerobic copper exposure, while CHASRI-containing cells tolerate these transitions as well as control cells containing icus and cue only. This suggests that the combined gene cluster confers a more environmentally robust resistance to copper than the components, cus and pco, in isolation. This type of resistance is likely to provide the greatest benefit to facultative anaerobes that natively lack both cus and pco systems. We suggest that CHASRI is a so-called “winning pattern,” which arose in a copper-stressed environment that favored a range of forms of copper resistance and drove module combination (fig. 7D) through genetic linkage (Baquero 2004). Since its assembly, this resistance island has become highly selectable in copper-rich sites such as mines, agriculture, hospitals, and sewage. All recent diversity of CHASRI appears to involve horizontal transfers, and the rate of HGT increases with integration of copper-resistance modules. This is consistent with HGT-elevated fitness of genes that are found in functionally integrated gene clusters (Lawrence and Roth 1996).

The Origin and Diversification of CHASRI Are associated with Long-Term Trends of Human Copper Emissions

The horizontal transfer of a CHASRI from Enterobacteria to Shewanella provided an opportunity to model CHASRI origin and diversification through time using a molecular clock, assuming a minimum surface–subterranean separation date. By applying the 2,060-year calibration to the Es. coli/Shewanella divergence in a relaxed molecular clock analysis of CHASRI nucleotides, we were able to associate CHASRI origin and episodes of accelerated diversification with multimillennial trends in anthropogenic copper emissions. We inferred that the current CHASRI diversity originated in a relative of En. cloacae around 6,000 years ago, which corresponds to current estimates of the beginning of the Bronze Age (Radivojevic et al. 2010), when anthropogenic copper became a factor in environmental stress. The high level of resistance (20 mM) conferred by the CHASRI (Tetaz and Luke 1983) compared with the low level of copper (1.6 × 10⁻⁵ mM) in uncontaminated fresh water (Alberta Environmental Protection 1997) suggests that CHASRI most likely diversified under broad copper stress driven by human metallurgical activity. We speculate that the likely donors of cus (En. cloacae) and pco (P. fluorescens) could have been cocontaminants of dairy fermented in copper alloy vessels by Bronze Age agrarians (Canganella et al. 1999). Alternatively, these resistance mechanisms became clustered under elevated environmental copper conditions and were later acquired from standing diversity by bacteria in animal microbiomes (Sander and Koschinsky 2011).

Greenland ice cores support peaks in atmospheric copper emissions during the time of the Roman Empire 2,000 years ago, again 1,100 years ago in China, and most notably in the last 150 years (Hong et al. 1996). In our model, estimates of copper emissions correspond in number and magnitude to episodic accelerations in CHASRI diversification. The origin of CHASRI in deep subterranean Shewanella in the Southwestern United States may be linked to the pre-Christian Era copper industry (Ehrhardt 2009). Under this model the estimated divergence between surface and subterranean Shewanella CHASRIs corresponds to colonial mining in the American Southwest approximately 300 years ago, which suggests that the modified form of CHASRI was reintroduced to the surface. This model is further supported by two K. pneumoniae CHASRIs, found on nearly identical plasmids 3 years apart, which are estimated to have diverged for less than 10 years. An alternative model that places the subterranean divide at the divergence of Shewanella islands does not match the above dates directly but still suggests recent diversification following an origin ∼9,000 years ago (supplementary table S7, Supplementary Material online). Although we find the visual appearance of cointegration of the CHASRI diversification and ancient anthropogenic copper deposition rate curves to be suggestive of an association, we caution that specific dates inferred by molecular clock analyses are dependent on multiple assumptions including reliable calibration points and the generation times of the bacteria in question, which cannot now be estimated with high certainty.

The current known distribution of CHASRIs is likely driven by tremendous copper pressure exerted by human activities in the 19^th and 20^th centuries. In the late 19^th century, in addition to environmental emissions from smelting, copper was widely used in food preservation and as a disinfectant of sewage, gastroenteritis stools, and wounds (Blyth 1888; Rideal 1895; Young 1898). Since 1955, copper has been used for growth stimulation in large-scale swine and poultry production (Barber et al. 1955), the mechanism of which may involve selection on livestock gut microbiome communities (Shurson et al. 1990; Dunning et al. 1998). Furthermore, elevated copper in tap water due to the corrosion of copper plumbing may give a selective advantage to intestine-associated bacteria with copper-resistance mechanisms. The earliest isolated CHASRIs date to 1890, in En. cloacae strain ATCC13047 isolated from the spinal fluid of a patient with meningitis. Genome analysis revealed that this single strain contains representatives of two major clans of CHASRIs. Highly similar CHASRIs are found in the genomes of pathogenic Enterobacteria on five continents, Es. coli poultry respiratory infections, Cronobacter responsible for neonatal death, gastroenteritic Es. coli, and Klebsiella responsible for pneumonia and meningitis (Munson et al. 2000; Silver 2003; Chen et al. 2004; Gilmour et al. 2004; Johnson et al. 2005; Houdt et al. 2009; Crossman et al. 2010; Su et al. 2011).

An Evolutionary Prospective for Emergent Pathogens

The rapid and widespread dispersal of copper resistance islands illustrates a dramatic flux of bacterial genes by horizontal transfer in response to human influence (Bezuidt et al. 2011). Most bacterial strains in which CHASRIs were detected also show resistance to silver, arsenic, and tellurium, suggesting the island could be associated with multiple selective pressures involving metal toxicity. Human-driven mobilization of silver has been more limited than copper, but medicinal applications of antimicrobial silver (Edwards-Jones 2009) may have contributed to the spread of CHASRI (McHugh et al. 1975). Previous research has linked increased pathogen durability (Walther and Ewald 2004), specifically pathogen metal resistance (Audic et al. 2007; Kucerova et al. 2010; Shafeeq et al. 2011), with increased virulence, suggesting that CHASRI-encoded resistance is part of a survivability mechanism that can be acquired by strains in animal microbiomes transitioning from commensal to pathogenic organisms (Duriez et al. 2001).

Recently, it has been shown that copper is involved in macrophage defense against Salmonella infections (Osman et al. 2010; Achard et al. 2012), highlighting a mechanism for CHASRIs to provide an additional fitness benefit to pathogenic organisms. Because DNA-based resistance islands are not continuously bound to living cells, or even to specific plasmid vectors over the long term, control of resistance dispersal must involve improving upstream management practices to prevent selection for such mechanisms in sympatric bacteria, rather than downstream antibacterial measures, which are likely to exacerbate disease emergence. Together, these results suggest that an ecological disturbance, driven by human activity starting in the Bronze Age, has influenced the contents of gammaproteobacterial pan-genomes, affected bacterial resistance in far-flung environments, and perhaps contributed to the emergence of some modern pathogens.

Supplementary Material

Supplementary tables S1–S7 and figures S1–S3 are available at Genome Biology and Evolution online (http://www.gbe.oxfordjournals.org/).

Note Added in Proof

Recent reviews on related copper resistance (Hao et al. 2015) and silver resistance islands have been published (Randall et al. 2015).