Unravelling the mechanisms of antibiotic and heavy metal resistance co-selection in environmental bacteria

Abstract The co-selective pressure of heavy metals is a contributor to the dissemination and persistence of antibiotic resistance genes in environmental reservoirs. The overlapping range of antibiotic and metal contamination and similarities in their resistance mechanisms point to an intertwined evolutionary history. Metal resistance genes are known to be genetically linked to antibiotic resistance genes, with plasmids, transposons, and integrons involved in the assembly and horizontal transfer of the resistance elements. Models of co-selection between metals and antibiotics have been proposed, however, the molecular aspects of these phenomena are in many cases not defined or quantified and the importance of specific metals, environments, bacterial taxa, mobile genetic elements, and other abiotic or biotic conditions are not clear. Co-resistance is often suggested as a dominant mechanism, but interpretations are beset with correlational bias. Proof of principle examples of cross-resistance and co-regulation has been described but more in-depth characterizations are needed, using methodologies that confirm the functional expression of resistance genes and that connect genes with specific bacterial hosts. Here, we comprehensively evaluate the recent evidence for different models of co-selection from pure culture and metagenomic studies in environmental contexts and we highlight outstanding questions.


Introduction
The pr olifer ation of antibiotic r esistance genes (ARGs) amongst bacterial pathogens underpins the problem of antimicrobial resistance .T his is becoming a leading cause of mortality, associated with 4.95 million deaths globally in 2019 and projected to become 10 million deaths annually in 2050 (Murray et al. 2022 ).The widespread use and misuse of antibiotics are known to cause a large part of the antimicrobial resistance problem, but beyond the clinical setting, an interplay between resistance genes for antibiotics, other selectiv e a gents like metals, and mobile r esistance elements (MGEs) is also occurring.This warrants a 'One Health' a ppr oac h to this problem (Durso and Cook 2019 ).The topic of antibiotic resistance is frequently reviewed, and the reader is dir ected to r ecent r e vie ws that focus on the mor e envir onmental aspects (Skandalis et al. 2021 , Larsson andFlach 2022 ).What is less well-understood, and the subject of this r e vie w, is how other selectiv e a gents, namel y hea vy metals , impact the acquisition and transmission of ARGs.
In the broadest sense, co-selection is where exposure to one selectiv e a gent allows ada ptation to a second selectiv e a gent, in this context where one antimicrobial selects for a resistance mechanism for both itself and also another antimicrobial (Di Cesare et al. 2016a ).Ther e ar e thr ee r ecognized genetic models of co-selection: co-r esistance, cr oss-r esistance, and co-r egulation (Baker-Austin et al. 2006 ); these are detailed in subsequent sections.A fourth model, co-selection via biofilm formation, is not a genetic model and does not form part of this r e vie w.
Ther e ar e se v er al potential co-selectiv e a gents for antibiotic r esistance including detergents (Grenni and Corno 2019 ), biocides (Wales and Davies 2015 ), pol yar omatic hydr ocarbons, pol yc hlorinated biphenyls (Goro vtso v et al. 2018 ), and nanoparticles (Chen et al. 2019b, Zhang et al. 2019b ).Here, we will focus just on the role of metals in the co-selection of antimicr obial r esistance since metals have the largest sphere of influence due to their diverse anthropogenic and natural sources, and their persistence in envir onmental nic hes.Individual r esistance genes ar e common in envir onmental bacteria irr espectiv e of anthr opogenic pollutant exposure (Lo Giudice et al. 2013, Farias et al. 2015 ) and certainly predate anthropogenic pollution due to natur al exposur e to antibiotics and metals (Mindlin et al. 2005, Petr ov a et al. 2011 ).Howe v er, the increased dissemination and abundance of these resistance genes over human history correlates with the anthropogenic use of mercury (Poulain et al. 2015 ), copper (Staehlin et al. 2016 ), and antibiotics (Knapp et al. 2010 ).Moreover, parallel responses in the increased persistence and diversity of resistance genes, and in the assembly of multiple resistance genes into single MGEs , ha ve been observ ed in r ecent decades (Hughes andDatta 1983 , v an Hoek et al. 2011 ).
This r e vie w compr ehensiv el y e v aluates an outstanding ga p in knowledge , i.e .what is the evidence supporting different mechanisms of heavy metal and antibiotic co-selection?Building on previous seminal r e vie ws (Baker-Austin et al. 2006, Seiler and Berendonk 2012, Pal et al. 2017, Vats et al. 2022 ), we offer an in-depth pictur e of co-selectiv e mec hanisms.Knowing these mec hanisms is important for ameliorating the rise in resistance, as several clinical or environmental solutions to this problem have been proposed based on co-selection theories (Vats et al. 2022 ).Following an introduction of k e y conce pts including heavy metal contamination, r esistance mec hanisms, horizontal gene tr ansfer (HGT) and MGEs, and co-selective models and concepts, liter atur e will be discussed that support each of the models.Special focus will be given to research on cadmium, copper, and zinc since these metals are often associated with co-selection.To emphasize modern molecular a ppr oac hes, the r e vie w is focused on recent papers (2010 onw ar ds) ( Table S1 ), with the inclusion of some older studies that made fundamental advances.Common findings including correlational networks across the literature have been summarized to point out potential links that warr ant mec hanistic inv estigation.This is concluded with a discussion of what we do know and gaps in knowledge that are yet to be unravelled.

Heavy metal sources
Heavy metals are defined as naturally occurring metals of an atomic number greater than 20 and density greater than 5 g cm −3 (Ali and Khan 2018 ).As this r e vie w is concerned with eco-toxicity, this definition will be extended to include toxic metalloids such as arsenic and tellurium and the nonmetal selenium.
Heavy metals can enter the environment from natural sources, such as geothermal activity, fires, weathering, and erosion, but these are minor contributors compared to anthropogenic sources, suc h as industry, a gricultur e, and healthcar e (Swaine 1994 , Gillings andPaulsen 2014 ).Mineral-based industries, including mining, metal refining, and fossil fuel combustion, are obvious anthropogenic sources of heavy metal contamination.The waterwa ys , sediments , and soils near these industries often have high le v els of arsenic, cadmium, c hr omium, copper, lead, manganese, mercury, nickel, and zinc (Rowe et al. 2002, Thomas et al. 2020, Yang et al. 2021 ).In a gricultur e, zinc and copper ar e common liv estock feed supplements for infection control and growth promotion (Yu et al. 2017, Yue et al. 2020 ).Furthermore , hea vy metals are common components of fertilizers , pesticides , and fungicides (Yu et al. 2017 , Grenni andCorno 2019 ).In healthcare , hea vy metal use has lar gel y been supplanted b y antibiotics, ho w e v er, it is still retained for some topical treatments and antimicrobial coatings (Pal et al. 2017 ).The waste streams from these sectors are increasing, and are the major source of metals in w aterw a ys , sediments , and soils (Seiler and Berendonk 2012, Silva et al. 2021, Palm et al. 2022 ).
Metal and antibiotic waste streams often overlap, posing significant co-contamination and microbial toxicity in w aterw a ys , sediments , and soils .Later sections will discuss this ov erla p also with respect to resistance genes.Akin to heavy metals, the manufacture of pharmaceutical antibiotics have introduced a source of antibiotic emissions of unprecedented concentration and scale (Larsson and Flach 2022 ).In China, which is the largest antibiotic producer and consumer globally, the antibiotics most in use b y w eight ar e macr olides , β-lactams , fluoroquinolones , tetracyclines, and sulfonamides (Zhou et al. 2017, Zhao et al. 2018 ).An estimated 53 800 tonnes of antibiotics were released into the environment after human or animal use in China alone in 2013 (Zhang et al. 2015 ).It is well known that the disc har ge of active antibiotics in waste streams imposes selective pressur es on surr ounding ecosystems (Zhang et al. 2015 , Larsson andFlach 2022 ).

The microbial toxicity of metals
Hea vy metals ha v e a dynamic r ole in biology, and man y ar e essential for cellular function (Nies 1999 ).Essential metals such as copper, zinc, cobalt, and nickel are toxic at higher concentrations, and others such as cadmium and mercury have no known biological function (Seiler and Berendonk 2012 ).Metal toxicity triggers the release of reactive oxygen species, which cause DNA mutation and other cellular damage (Yu et al. 2017 ).
The microbial impact of metal pollution is dependent on numerous factors, including concentration, ion valency, bioavailability, and en vironmental context.P olyvalent metal ions have different solubilities, r eactiv e potentials and hence toxicities.For example, Cr 6 + and As 3 + are more soluble, reactive and toxic than Cr 3 + and As 5 + (Nies 1999 ).The toxic impacts of metals are often ameliorated in solid matrices (sediments and soils) since most metal ions in these niches are bound to anionic residues on the soil surface and are not readily bioavailable (Peltier et al. 2010, Song et al. 2017, Hung et al. 2022 ).In general, the solubility of free metal ions and organometals is inversely proportional to the pH, which dir ectl y impacts uptake into the bacterium, and ther efor e toxicity (Olaniran et al. 2013 ).T here are exceptions , ho w ever, e.g.hydroxozinc, cadmium, or nickel groups are soluble at higher pH (Olaniran et al. 2013 ).Bioavailability is further influenced by the mineral content (Goro vtso v et al. 2018 ), redox potential, oxygenation, and or ganic content (Sc hulz-Zunkel and Krueger 2009 ).These factors m ust be consider ed when determining the toxic or selective effects of metals on microbes.
Heavy metals potentially have a larger sphere of influence than antibiotics .Many hea vy metals (arsenic , cadmium, c hr omium, copper, iron, lead, manganese, nickel, and zinc) are released at concentr ations se v er al orders of ma gnitude higher than antibiotics (Baker-Austin et al. 2006, Yang et al. 2021 ), they are more widespread, and they do not biodegrade (Grenni and Corno 2019 ).Furthermore, some metals are known to have lo w er sorption potential than antibiotics , e .g. zinc and copper compared to tetracycline (Song et al. 2017 ), thus boosting their bioavailability in soils and sediments (Schulz-Zunkel and Krueger 2009 ).

The resistome
The clinical definition of 'r esistance' r efers to the growth of a bacterial strain at higher than the antimicrobial breakpoint value, the br eakpoint v alue being the concentr ation that determines whether a strain is considered resistant or susceptible.'Tolerance' is clinically defined as an increase in minimum bactericidal concentration without an increase in minimum inhibitory concentration (MIC) (Wales and Davies 2015 ).These terms cannot be used in their strict clinical sense for most heavy metals, as breakpoint values and MICs may not exist.So it is important to note that when we use the word 'resistance' here, this has a more general meaning, i.e. being able to grow at a heavy metal concentration that would normally inhibit growth (Wales and Davies 2015 ).
Antibiotics and heavy metals exert a po w erful selectiv e pr essur e on micr oor ganisms, whic h has led to the e volution of v arious resistance and homeostatic mechanisms (Staehlin et al. 2016, Squadrone 2020 ).The collection of resistance genes in a given environment is known as the 'resistome'.The antibiotic and heavy metal resistomes are diverse and ARGs and metal resistance genes (MRGs) are common in most en vironments , with a range of 10 −5 -10 −1 ARGs per copy of 16S rDNA in bacterial genomes and meta genomes (P al et al. 2016, Zhao et al. 2018, Thomas et al. 2020 ), and 86% of complete genomes contain potential MRGs (Pal et al.Borremans et al. ( 2001Borremans et al. ( ) 2015 ) ).While ARGs gener all y ar e bona fide r esistance genes, man y MRGs may also play roles in maintaining the homeostasis of essential metals, only providing resistance at higher metal concentrations.It is the latter capacity that we will focus on here .T he mor e pr edominant MRGs and ARGs ar e shown in Tables 1 and 2 , r espectiv el y.

Parallels in resistance mechanism: metals versus antibiotics
The mechanisms of metal resistance (Nies 1999 ) and antibiotic resistance (van Hoek et al. 2011 ) have been thor oughl y r e vie wed, but it is worth highlighting here the similarities between mechanisms, namel y efflux, c hemical modification, and sequestr ation (Baker-Austin et al. 2006 ).In contrast, some mechanisms that solely apply to antibiotics include modified membrane permeability (Perron et al. 2004 ), target modification (Timms et al. 1992 ), or target r eplacement (Deur enber g and Stobberingh 2008 ).These mec hanisms are not seen for heavy metals as metals pr esent br oad toxicity to many cellular systems whereas antibiotics have more specific modes of action and targets in the bacterial cell.Other resistance mechanisms such as general stress responses (e.g.DNA repair and ROS scavengers) do assist in cellular resistance to metals and antibiotics, ho w e v er, this r e vie w will just deal with mec hanisms specific to metal ions and antibiotics.The effector proteins for MRGs and ARGs have striking similarities, although ther e ar e differ ences in terms of gene r egulation.Efflux pumps are the most common MRG (Squadrone 2020 ), and are very common ARGs as well (Poole 2005 ).These export unwanted molecules to the cell exterior.An MRG efflux pump example is the CzcCBA resistance nodulation division (RND) divalent cation efflux pump (Nies 1995 ).The two-component regulator CzcRS senses cadmium, zinc, and cobalt ions, whic h upr egulates the expression of czcCBA (Fig. 1 A).The inner membrane transporter (CzcA), periplasmic linker (CzcB), and outer membrane protein (CzcC) are structurally homologous to the AcrAB-TolC antibiotic efflux system.AcrB forms an inner membrane transporter with the AcrA periplasmic linker to efflux antibiotics through the TolC outer membrane protein following the alleviation of AcrR repressor by global transcriptional activator MarA (Fig. 1 B) (White et al. 1997, Chowdhury et al. 2019 ).
Chemical modification is emplo y ed as a detoxification mechanism in both antibiotic and heavy metal resistance systems, although here systems are functionally analogous rather than posing an y structur al homology.Amelior ating metal toxicity thr ough reduction is limited to metals with a redox potential within the physiological range of cells (Nies 1999 ).For example, in the copper resistance system Cue, originally characterzied in Escherichia coli , sensing of Cu 1 + causes CueR to upregulate CopA and CueO.
CopA exports Cu 1 + to the periplasm, where CueO oxidizes it to Cu 2 + (Fig. 1 C) (Rensing and Grass 2003 ).In contrast, antibiotics offer mor e div erse options for degr adation or modification via hydr ol ysis or additions of various chemical groups.An example is the hydr ol ytic cleav a ge of β-lactams by CTX-M (Fig. 1 D) (King et al. 2016 ).
Another method used for both metal and antibiotic resistance is intracellular or extracellular sequestration.In Synechococcus , the detection of high le v els of zinc by the SmtB r egulator r esults in expression of the metallothionein SmtA, which binds zinc via internal thiol groups (Fig. 1 E) (Blindauer et al. 2001 ).An analogous system for antibiotics is the sequestration of mitomycin by Mrd protein in Streptomyces lavendulae .Efflux is later emplo y ed for mitomycin by the Mct transporter as there is no homeostatic need to store mitomycin in the cell as opposed to zinc and other metals (Fig. 1 F) (Sheldon et al. 1999 ).
A final point worth noting when comparing antibiotic and metal r esistance str ategies in bacteria is the difference in the typical regulatory systems in volved.T he MRG effector genes tend to be activ el y r egulated (induced or r epr essed), while some of the ARG examples above are constitutively expressed (Sheldon et al. 1997, Cantón et al. 2012 ).This may reflect the fact that for the metals, the cell is attempting to maintain homeostasis across envir onments containing v ariable metal concentr ations, while for antibiotics, the cell is just attempting to r emov e all of the chemicals at all times.

The mobilome
Many ARGs and MRGs have environmental origins (Gillings andPaulsen 2014 , Larsson andFlach 2022 ), but how are these genes mobilized from environmental bacteria into clinical pathogens?Answering this question r equir es a detailed understanding of the 'mobilome', which is defined as the total genetic information stored in MGEs in a microbial community.MGEs are discrete sections of DNA that can transfer independent of c hr omosomal replication.These include mobile plasmids, insertion sequences (IS), tr ansposons, integr ons, integr ativ e conjugativ e elements, and phages .T he mobilome clearly overlaps with the resistome discussed abo ve .
The speed and range of resistance acquisition in bacterial comm unities ar e lar gel y gov erned by plasmids .T hese extr ac hr omosomal elements transmit DN A betw een cells, resulting in HGT within or across taxonomic barriers including from Gramnegativ es into Gr am-positiv es (Wang et al. 2020 ) and niche barriers such as from environmental to clinical contexts (Martins et al. 2014, Silv eir a et al. 2014 ).This is typicall y ac hie v ed via conjugation (direct cell-cell plasmid transfer) (Zhang et al. 2018b ), transformation (direct uptake of 'naked DNA') (Xu et al. 2017 ), or transduction (via viruses as intermediates) (Mašla ňová et al. 2016 ).Most clinicall y r ele v ant r esistance genes hav e been found on conjugativ e plasmids (Palm et al. 2022 ), but mobilizable plasmids (can transfer in the presence of a conjugative plasmid) and nonmobilizable plasmids (vertical inheritance only) also contribute via transformation.Resistance plasmids can carry multiple ARGs and/or MRGs, whic h incr ease the bacterial host's ability to r esist m ultiple antimicr obial str esses.Some c hr omosomal MGEs can also catalyse HGT in a similar manner to plasmids; these include genomic islands (Arsene-Ploetze et al. 2018 ) and integr ativ e conjugativ e elements (Song et al. 2013 ).
Other distinct groups of MGEs have roles in the intracellular mov ement of r esistance genes, suc h as IS, tr ansposons, and integrons .T hese elements are abundant in bacteria, with one study reporting their prevalence in soil metagenomes as 10 −3 -10 −2 copies per 16S rRNA gene (Zhao et al. 2019 ).Transposons encode tr ansposase and/or r ecombinase enzymes that facilitate either cut-and-paste or copy-and-paste insertion of the whole element at new locations in the genome (Babakhani and Oloomi 2018 ).Integrons enable the site-specific excision, integration, shuffling, and expression of their partner elements, known as gene cassettes  (Labbate et al. 2009 ).Class 1 integrons are very common in Gramnegative clinical isolates (Labbate et al. 2009 ), but they can also be r eadil y detected in envir onmental bacteria, e v en in r elativ el y pristine locations (Gillings et al. 2008 ).
While IS elements, transposons and integrons cannot selftransfer between cells, their interaction with plasmids enables this to occur.ARGs and MRGs can thus move between the c hr omosomes of unrelated bacteria via the agencies of multiple interacting MGEs .T he case of tr ansposon Tn 21 in plasmid NR1 pr ovides an excellent example of genetic linkage and HGT of ARGs and MRGs (Liebert et al. 1999 ).

Models of co-selection
As mentioned prior, there are three recognized co-selection models: co-r esistance, cr oss-r esistance, and co-r egulation.Cor esistance involv es r esistance genes being physicall y linked, most often on the same MGE, resulting in simultaneous inheritance as one pac ka ge (Fig. 2 A).Cr oss-r esistance involv es a single mec hanism conferring resistance to both agents, for example an efflux pump removing both antibiotics and metals (Fig. 2 B).Finally, co-r egulation involv es r esistance genes that shar e a pr omoter or Some r ele v ant concepts for understanding co-selection between metals and antibiotics include the MIC, the minimum selectiv e concentr ation (MSC), and the minim um co-selectiv e concentration (MCC).The MIC is the lo w est concentration of an agent that inhibits the growth of a specific microbe (Seiler and Berendonk 2012 ).The MSC is the lo w est concentration of an agent, wher e at whic h the fitness benefit of r esistance outweighs the fitness cost (Yu et al. 2017 ).The MCC is the lo w est concentration of one agent that will select for resistance to w ar ds another agent (Arya et al. 2021 )

Methodological advances in the study of co-selection
The methodology used to study co-selection over the years has become v astl y mor e sensitiv e and scalable .T he first evidence for co-selection between heavy metals and antibiotics was published in 1964 and described the co-transduction of penicillin and mercury resistance in Staphylococcus , such that the vast majority of transductants under penicillin selection also acquired mercury resistance, and vice versa (Richmond and John 1964 ).Since then, many studies using diverse methods (genetic mutation, transformation, and conjugation) have revealed examples of linked antibiotic and metal resistance in different bacterial lineages including Salmonella (Ghosh et al. 2000 ), Enterococcus (Hasman and Aarestrup 2002 ), and Burkholderia (Hayashi et al. 2000 ).This led to the seminal r e vie w by Baker-Austin et al. ( 2006 ), whic h r ecognized the importance of metals in antibiotic co-selection and outlined the coselection models, but the r elativ e contributions of particular coselection mechanisms in specific environments is not clear, and r equir es further study.
Advances in molecular biotechnologies such as highthroughput DNA sequencing have enabled a more mechanistic understanding of co-selection.The genetic basis of co-selection can now extend beyond genetically characterized lab strains to the culture-independent analysis of whole microbial communities (Li et al. 2020 , Huang et

Evidence for co-resistance
Most evidence for the co-resistance of antibiotics and metals is correlational.Finding ARGs in close proximity to MRGs in a genome is consistent with these genes being selected as a combined entity.The co-resistance model also implies a vehicle of resistance gene transfer, so correlations with integrons, transposons, or plasmids lend support to the model.Although correlational evidence is far from conclusive, it is a good starting point to r e v eal what the likel y r ele v ant ARG and MRG gene combinations are that warrant closer attention.To help identify these ARG and MRG combinations of special interest, the correlations from the liter atur e ( Table S1 ) hav e been pr esented as a meta genomic network analysis (Fig. 5 ).
The network analysis emphasizes consistent connections between czcA and sul1 , sul2 , or tet (G); copA or pcoA and bla TEM ; merA and sul1 ; and pcoA and tet (G).These combinations warrant further mechanistic investigations to confirm co-resistance since all these studies are impacted by biases of sampling and study design, such as the environments sampled, the culturing and/or molecular methods used, and the antibiotic or metal agent(s) chosen for study.Particular resistance genes may be more numerous across the bacterial community, or are easier to study, both of whic h would r esult in positiv e corr elations .As an example , consider the correlation of merA and sul1 (Fig. 5 ); while merA homologues are found in both Gram-positives and Gram-negati ves, the y ar e str ongl y corr elated with sul1 onl y in Gr am-negativ es (Sköld 2001, Christakis et al. 2021 ).T hus , a study on a Gr am-negativ e collection is likely to find this corr elation, wher eas a wider study may not.Proof of co-resistance ideally requires the experimental introduction of MRGs and ARGs into recipient test bacteria.This functional confirmation is important, especially for MRGs, many of which may serve homeostatic rather than resistance functions.

Pristine environments
ARGs and MRGs are common in environmental bacteria, even in environments with minimal anthropogenic exposure.In metalenriched water surrounding deep-sea hydrothermal vents, sequenced plasmids carrying MRGs were prevalent, and many of these plasmids also carried ARGs, including β-lactam and fluor oquinolone r esistance genes (Farias et al. 2015 ).At least one of these deep-sea resistance plasmids possessed conjugation genes, indicating strong potential for HGT, which was also supported by a phylogenetic comparison of the str ains.Meta genomic studies of natur all y metal-ric h high-altitude lakes in the Himalayas (Sharma et al. 2022 ) and Andes (Perez et al. 2020 ) have also yielded extensiv e corr elations between MRGs and ARGs, with r elativ e abundance comparable to urban w astew ater.In a survey of subsurface soils ( > 1 m depth) from diverse locations including both pristine and contaminated en vironments , high concentrations of metals (copper, lead, and zinc) increased ARG and MRG r elativ e abundance and diversity more than any other factor (Wang et al. 2021b ).A linear positiv e corr elation of MRGs with ARGs was seen, consistent with associated inheritance .T hese studies collectiv el y support the hypothesis that environments that are naturally rich in metals select for MRGs and ARGs but r equir e further v alidation for co-resistance within a single organism.

Agricultural environments
Fertilized (either manure or inorganic-based) agricultural soils contain both heavy metals and antibiotics (Yu et al. 2017 ), and so are potential hot spots for co-selection.The agricultural context is also critical for study from a One Health perspective since this is a location where plants , animals , humans , and their feeds Figure 4. Methodologies for studies investigating co-selection.Environmental samples or environmental simulations (microcosms and mesocosms) are sources of communities or pure cultures for study.qPCR measures relative gene abundance, while the culturable resistant isolates can be enumerated via plating.Pure cultures can have MIC , MSC , or MCC profiled through disk diffusion, microbroth dilution, or E tests.Growth rates and yields under varying selective pressures can be quantified.Gene knockouts or knockdowns can assign genotypes to phenotypes .T he occurrence and genetic context of resistance genes can be r e v ealed thr ough DNA sequencing, while expr ession le v els of these genes can be tested via transcriptomics, RT-qPCR, micr oarr a ys , proteomics , enzyme assa ys , ELISA, or western blots.Plasmid ca ptur e and/or plasmid curing experiments can help to infer the function of plasmid-encoded genes.Finally, bioinformatic data can be mined for resistance genes and MGEs, which in turn informs experimental c har acterization.Icons wer e cr eated with BioRender .comand smart.servier.com. and wastes intersect.The metagenomic total ARG distribution in manure-fertilized soil at poultry farms was influenced by metals, particularly cadmium, with 5.7% of ARG variance explained due to metals alone and 32% by metals in combination with other factors (Mazhar et al. 2021 ).Notably, these numbers are higher than the ARG variances attributed to antibiotics (0.7% alone and 10% combined with other factors).The co-resistance model was supported in the poultry farm study via the fr equent corr elation of ARGs with MGEs ( intI1 , IS 613 , and Tn 24 ).Copper-enriched soils from oliv e tr ee farms sho w ed statisticall y significant co-occurr ences of resistance genes to copper, tetracycline, and β-lactams and zntA co-occurring with tet (C) in r ecov er ed isolate genomes (Glibota et al. 2020 ).The exact genomic context was not pr ovided her e, but this is a step up fr om meta genomic corr elations as it was established that these genes reside in the same genome .Furthermore , intI1 was present in resistant isolate genomes, offering a potential mechanism for the acquisition of the resistance genes.In livestock w astew ater, metagenomic qPCR uncovered a strong correlation between the r elativ e abundance of copA and plasmid-borne mcr-1 , which is consistent with these genes being on the same plasmid (Yuan et al. 2018 ).

Industrial environments
Industrial environments are typically metal-enriched but devoid of antibiotic pollution (with the notable exception of the pharmaceutical industry).T hese hea vily impacted en vironments ha ve pr ovided man y opportunities to study the effects of metals on ARGs , MRGs , and MGEs .Mining is an acute source of heavy metal contamination and may promote the mobilization of resistance genes.In soil impacted by a gold mine, metagenomic community qPCR found that ARG abundances were correlated to levels of copper, manganese, nickel, and zinc (Yan et al. 2020 ).This and F igure 5. Netw ork analysis of correlations between MRGs and ARGs.Analysis based on correlations of ARGs and MRGs from co-selection studies in environmental contexts from 2010 onw ar ds.Size of nodes is proportional to the number of primary studies investigating that specific resistance gene.Weight of edges is proportional to the number of studies finding a positive correlation.Only positive correlations are shown.Blue indicates ARG, and red indicates MRG.Network visualized with Gephi v 0.9.7.
another study (Huang et al. 2023 ) r e v ealed that Actinom ycetota and Pseudomonadota were associated with a range of ARGs , MRGs , and MGEs (IS 26 , istA5 , tnpA2 , and ISRj1), which is consistent of cotransfer of ARGs and MRGs by HGT.One river polluted with mine tailings yielded a Lysinibacillus isolate with plasmid-borne copper and stre ptom ycin resistance (Chihomvu et al. 2015 ).This assessment was made through curing experiments rather than plasmid sequencing, so details of the involved plasmids were not determined.One interesting metagenomic study used a clone library a ppr oac h with DNA from an acid mine drainage site to find a DNA fr a gment that conferred antibiotic resistance in E. coli (Arsene-Ploetze et al. 2018 ).This metagenomic DNA encoded a CusAB-like efflux pump that increased the MIC for gentamicin, kanamycin, and rifampicin upon heterologous expression in E. coli .The authors could not confirm that this gene also facilitated copper resistance, so it is unclear whether this is an example of co-resistance or cr oss-r esistance.
Heavy metal contamination from industries other than mining also facilitates metal co-resistance.Lake sediment cores from a heavily industrialized area in the UK found a correlation between sedimentary zinc concentration and the proportion of culturable isolates with zinc r esistance, oxacillin r esistance, and trimethoprim r esistance (Dic kinson et al. 2019 ).These resistant phenotypes were additionally correlated to metagenomic intI1 abundance measured via qPCR, suggesting integron involvement.In electr oplating waste water, whole genome sequencing of a nickelr esistant Shew anella sp.r e v ealed the close pr oximity of IS elements with nickel and cobalt MRGs in addition to blaR1 and erythromycin ARGs (Cai et al. 2019 ).A clear example of plasmidmediated co-resistance was seen in a Pseudomonas aeruginosa isolate obtained from freshwater near chemical industries .T his isolate carried a conjugative plasmid carrying tetracycline and copper resistance genes tet (A), copA , and copB (Martins et al. 2014 ), which was able to conjugate into E. coli .
Waste water tr eatment plants and se wa ge sludge hav e been heavily studied for their role in propagating MRGs , ARGs , and MGEs, and are another hotspot for co-selection.Positive correlations of MRGs with ARGs and MGEs in w astew ater are common (Lin et al. 2019, Murray et al. 2019 ), with intI1 correlated with czcA , sul2 , arsB (Di Cesare et al. 2016b ), and mpbH (Yuan et al. 2018 ) revealed by metagenomic qPCR.Integrons are correlated with MRGs at many sites, but these correlations must be interpreted with care .MRGs ha ve never been seen as integron gene cassettes, the mor e likel y hypothesis is that integrons and MRGs are carried by the same transposons and plasmids.
Consistent correlations of MRGs with ARGs and MGEs have been observed across entire catchment systems .T he Xiangjiang river in China has been investigated by several studies measuring ARGs at mining disc har ge points (Xu et al. 2017(Xu et al. , 2019 ) ). Bacteria isolated from the river contained numerous β-lactamase genes and MRGs, and their MIC for ampicillin increased up to eightfold near the mining disc har ge points (Wang et al. 2021a ).Two co-resistance plasmids were discovered in Bacillus megaterium and Shewanella oneidensis .One plasmid contained czcD , qnrA , and qnrB and the second contained copB , merR , tet (A), and tet (W) (Xu et al. 2017 ).In a metagenomic qPCR study of Indian and UK metalimpacted waters and sediments, total metal concentration was positiv el y corr elated to MRGs , ARGs , and integr on r elativ e abundance, to the extent that it explained 83% of resistance gene distribution (Gupta et al. 2022 ), and 92% of bacterial community composition (Gupta et al. 2023 ).In those studies, network analysis of metagenomic qPCR data suggested that resistance genes tet (W), bla TEM , mefA , zntA , and c hrA wer e likel y r esiding in the same bacterial host, but the nature of any MGEs involved is unknown.Genome analysis of a Comamonas isolate from Melbourne sediment r e v ealed a class 1 integron and kanamycin gene cassette adjacent to a c hr omate r esistance tr ansposon.The sediment meta genome str ongl y corr elated class 1 integr ons with heavy metals (Rosewarne et al. 2010 ).This is a nice example of the inter play of differ ent MGEs , ARGs , and MRGs that can be involved in co-selection.
One unusual industrial environment notable for co-resistance is dense atmospheric urban smog.In a fascinating study by Pal et al . ( 2016 ), the meta genomic div ersity of both ARGs and MRGs in smog from Beijing was found to be higher than in any other sampled en vironment.T he abundance of ARGs and MRGs was equivalent to w astew ater, which is generally considered to be the most significant hotspot for resistance gene spread.Based on the high abundances and diversity of MRGs and ARGs detected, including notable carbapenem ARGs, the smog microbiome seems likely to be another important location where co-resistance in bacteria can arise.

Microcosms
Characterization of the resistome and mobilome before and after controlled heavy metal or antibiotic exposure in microcosms allows the gathering of mor e rigor ous e vidence for co-r esistance than correlations and genomic sequencing.One striking finding to emer ge fr om suc h studies is that high concentr ations of metals can promote HGT, and this effect can manifest r a pidl y.For example, r elativ e pr oportions of antibiotic-r esistant bacteria in a biofilter comm unity incr eased after just 6 hours of copper (100 mg l −1 ) exposure.In addition, metagenomic qPCR revealed an increase in the relative abundance of cusCBA , tet (B), tet (G), tet (L), mexF , sedB tr ansposons, and integr ases (Zhang et al. 2018a ).Although cr oss-r esistance w as suggested b y the authors, there w as no evidence for copper-export genes other than cusCBA , so the responsible mechanism is still under question.Concerningly, in this study, acquir ed r esistance to v ancomycin, erythr omycin, and lincomycin was maintained in the absence of selective agents for at least se v en da ys .
Studies using combinations of metals and antibiotics can give useful insights on possible synergy but are unfortunately rare in the liter atur e.In waste water micr ocosms exposed to copper or zinc (1 mg l −1 ) plus tetracycline or ampicillin (0.5 mg l −1 ), se v er al ARGs and MRGs increased in absolute abundance under all exposure combinations.Metagenomic qPCR revealed that the tr ansposase gene tnpA incr eased in absolute abundance and was str ongl y corr elated with r esistance genes copA and c hrB , whic h may mean these MRGs are carried by transposons (Zhao et al. 2021 ).This study also linked specific bacterial types to specific MRGs and ARGs through correlational network analysis, with one notable candidate being Mycobacterium spp.carrying bla TEM , copA , copB , pcoD , and zntA .Experiments using microcosms of wetland sediments demonstrated through metagenomic qPCR that doxycycline (50 mg l −1 ) and cadmium (0.5-5 mg l −1 ) exposure for 3 months increased MRG and ARG relative abundance more than either agent on their own (Yu et al. 2022 ).In this study, both agents wer e positiv el y corr elated with individual r esistance genes and bacterial genera, with Acinetobacter being one notable predicted host for MRGs , ARGs , and integrons .Both of these studies r e v eal clinically important genera possessing MRGs and ARGs.
Microcosm studies of w astew ater or sludge have revealed close relationships between metals , ARGs , and MRGs .A common finding in bacterial communities is that heavy metals (Tan et al. 2023 ) or MRGs (Zhang et al. 2019a ) have a m uc h gr eater impact than antibiotics on the abundance and types of ARGs, although community composition is typically the most dominant factor in determining ARG types (Zhang et al. 2016(Zhang et al. , 2017 ) ).The r elativ e abundance of ARGs and MRGs in the metagenome of w astew ater micr ocosms hav e been corr elated to intI1 via qPCR, r e v ealing intI1 correlations with both ARGs and MRGs , again consistent with coresistance (Zhang et al. 2016, 2019a, Tan et al. 2023 ).In soil microcosms, 10 mg kg −1 sulfamethoxazole was determined to be the MSC for MRGs , ARGs , and MGEs at a comm unity le v el (Li et al. 2023 ).In that study, each MGE was correlated with multiple ARGs, consistent with co-carriage.

Studies with plasmids
Plasmids are the primary transfer vehicle for MRGs and ARGs, and co-residence of ARGs and MRGs on the same tr ansfer able plasmid is a mechanism of co-resistance .Furthermore , metals can stimulate plasmid transfer.The frequency of conjugation of heavy metal and antibiotic co-resistance plasmids from B. megaterium (Xu et al. 2017 ) or Pseudomonas monteilii (Wang et al. 2021a ) into E. coli HB101 increased by one to two orders of magnitude in the presence of copper (0-20 μg l −1 ) or zinc (0-30 μg l −1 ).In a clear display of acquired co-resistance, the recipient E. coli gained resistance to both heavy metals and antibiotics upon transformation or conjugative uptake of the plasmid (Xu et al. 2017, Wang et al. 2021a ).In one of the few studies investigating the MSCs for both metals and antibiotics, E. coli MG1655 hosting a 220-kb extended spectrum β-lactamase (ESBL) resistance plasmid was exposed to subinhibitory le v els of metals and antibiotics (Gullberg et al. 2014 ).When the bacteria were exposed to various combinations of arsenite , tetracycline , and trimethoprim, the MSC decreased with all agents when compared to individual exposure .T he reported MSCs were at environmentally relevant concentrations, giving weight to the argument that even trace residues of antimicrobials may potentiall y hav e pr ofound impacts on selection.
Genotypic or phenotypic verification of plasmid recipients demonstr ates co-r esistance in action.The tr ansfer ability of r esistance to mercury, β-lactams, and quinolones was examined in ESBL-containing isolates from the Yam una Riv er in Ne w Delhi (Siddiqui et al. 2020 ).Conjugation from the arsenic and mercuryresistant isolates into E. coli r e v ealed sim ultaneous acquisition of metal and antibiotic resistance, with transconjugants receiving ESBL, qnrS, merB , merP , and merT .Transfer of ARGs and MRGs together has also been observed in Gram-positives, such as in Enterococcus lab strains which acquired tcrB , cueO , aadE , erm (B), tet (L), tet (M), and v anA fr om mating assays with copper r esistant envir onmental Enterococcus spp.(Silv eir a et al. 2014 ).Subsequent transconjugants inherited phenotypic resistance to both copper and multiple antibiotic families.In plasmid ca ptur e experiments, it is important to phenotypically verify resistance acquisition, since the transfer of an ARG or MRG alone does not guarantee a resistance phenotype.For instance, a m ultir esistance plasmid carrying copAB in P. aeruginosa did not confer copper resistance to E. coli , despite detection of copAB in the transconjugants (Martins et al. 2014 ).Although these authors did not probe this finding, it could be due to the m ultigene natur e of the cop system, assuming the other genes r equir ed ( copC and copD ) (Pal et al. 2017 ) were not co-tr ansferr ed with copAB .

Bioinformatic studies of co-resistance
The expanding size of genome databases gives us increasing po w er to analyse the relationships between MRGs and ARGs.In one study of 2522 complete genomes from the 2014 NCBI bacterial genome database (Pal et al. 2015 ), 17% contained both ARGs and MRGs and genomes containing an MRG were 10-fold more likely to contain an ARG compared to those with no MRGs.Two years later, the same database contained 5436 complete bacterial genomes, with half containing both ARGs and MRGs (Li et al. 2017 ).This seemingly dramatic shift is lar gel y due to the increased r epr esentation of clinicall y significant taxa (especiall y Enterobacteriaceae ) in bioinformatic databases .T he m ultir esistance genomes wer e found m uc h mor e often in human or animal microbiota and pathogens rather than in the environment (Pal et al. 2015, Li et al. 2017 ).Based on analysis of the 2014 NCBI database, 5% of plasmids contained both ARGs and MRGs (Pal et al. 2015 ), with those resistance plasmids tending to be large (median 76 kb), conjugati ve, and contain to xin-antito xin systems, indicating that the y can both self-transfer and fix themselves into new hosts.
Mining of genome databases suggests that MRG-ARG pairings are frequent and that some specific pairs of MRG-ARG are more common than others .T his fa vours co-resistance o ver crossresistance as a mechanism, because in cross-resistance, the expectation would be that a few individual genes with a broad substr ate r ange would dominate.Instead, it seems that gr oups of diverse genes with more specific activities have been assembled.The most common associations seen in c hr omosomes ar e mercury MRGs with aminoglycoside, phenicol, sulfonamide, and tetracycline ARGs (Pal et al. 2015 ); zinc MRGs with β-lactam, bacitr acin, and pol ymyxin ARGs; copper MRGs with β-lactams, kasugamycin, and bacitracin ARGs; and arsenic MRGs with β-lactam, bacitracin, and fosfomycin ARGs (Li et al. 2017 ).Some of these corr elations differ fr om those shown in Fig. 5 .This is likel y due to the different states of the literature and databases in 2016 versus 2023.Nonetheless, many similar dominant correlations are seen, such as mercury ARGs with sulfonamide and tetracycline ARGs; copper MRGs with β-lactam ARGs; and zinc MRGs with β-lactam and polymyxin ARGs.
In the context of MGEs, recurring associations of MRGs and ARGs seen in plasmids include the cadmium resistance gene cadD with aminoglycoside or macrolide resistance genes; mercury resistance genes with sulfonamide , aminoglycoside , β-lactam, phenicol, or trimethoprim resistance genes; and bacitracin with copper or zinc resistance genes (Pal et al. 2015, Li et al. 2017 ).In-tegr ons ar e mor e associated with MRG and ARG pairs than transposons (Li et al. 2017 ).Integron and IS CR genes are on 10% and 7% of plasmids carrying resistance genes, respectively, and intI1 and IS CR had a strong correlation with mercury, aminoglycoside, phenicol, sulfonamide, and tetracycline resistance genes (Pal et al. 2015 ).

Evidence for cross-resistance
Cr oss-r esistance exists where a single mechanism provides resistance to both a metal and antibiotics.Robust evidence for crossresistance can be derived from experimental characterization in pur e cultur es, but the majority of studies lac k this le v el of rigour and just report correlations .T he limited functional information about most open reading frames in genome databases further exacerbates this issue.Despite these problems, we can still navigate a useful path by being guided by correlational studies as indicators for which metal/antibiotic/ARG/MRG combinations deserve more detailed mechanistic investigation.Zinc and copper resistances dominate studies of cr oss-r esistance, with consistent correlations of these metals to aacA4 , bla OXA , bla TEM , erm (B), erm (F), sul1 , sul2 , tet (M), and tet (W) (Fig. 6 ).Other metals which correlate highly to ARGs include arsenic, cadmium, chromium, manganese, nickel, and lead.As expected, metal and ARG corr elations ar e not universal, and some yield negative correlations, so cross-selection is limited to certain chemical and enzyme combinations.Further indir ect e vidence for cr oss-r esistance comes fr om the pr edominance of broad-spectrum efflux pumps in the resistome (Thomas et al. 2020, Liu et al. 2021, Wang et al. 2021b, Furlan et al. 2022 ), these provide efflux of multiple diverse targets.
Further speculative evidence for cross-resistance comes from the many studies that re port discre pancies between detected resistance genes and the resistance phenotypes.Some of these may r epr esent cases where resistance genes have a broader range of action than pr e viousl y documented.For example, in a study on marine biofilms growing on boat hulls treated with copper and zinc-based paint, tetr acycline-r esistant colonies fr om the metaltr eated surfaces wer e significantl y mor e numer ous compar ed to nontreated surfaces despite a reduction in the relative abundance of tet genes (Flach et al. 2017 ).In this study, metagenomic sequencing r e v ealed copper and zinc MRGs did incr ease four-fiv efold, whic h r aises the question whether the MRGs may be r esponsible for the tetracycline resistance.Similar unexplained metal r esistance phenotypes, wher e ARGs hav e been detected, but not the expected MRGs also occur in the liter atur e (Vignar oli et al. 2018, Zhou et al. 2019, Jia et al. 2021 ).These cases may also indicate the presence of novel resistance genes that have yet to be identified.There are still many hypothetical proteins even in well-c har acterized genomes like E. coli that await assignation of functions, and the proportion of these hypothetical open reading frames is higher yet on MGEs (Hatfull 2008 ).

Molecular genetic studies in pure cultures
Mor e r obust e vidence fr om cr oss-r esistance stems fr om heter ologous expression or knockout methods, which link specific resistance genes to phenotypes.Making knoc k outs in environmental bacteria is not easy, and expression in standard hosts has its own technical difficulties (Rosano andCeccarelli 2014 , Kaur et al. 2018 ), thus, onl y a fe w studies hav e taken these mor e rigor ous molecular a ppr oac hes for studying cr oss-r esistance, as described below.While a variety of metal and antibiotic cr oss-r esistance mec hanisms exist, efflux pumps seem to play the largest role.F igure 6. Netw ork analysis of correlations between heavy metals and ARGs.Analysis based on correlations of ARGs and MRGs from co-selection studies in environmental contexts from 2010 onw ar ds.Size of nodes is proportional to the number of primary studies investigating that specific resistance gene.Weight of edges is proportional to the number of studies finding a positive correlation.Only positive correlations are shown.Blue indicates ARG, and red indicates heavy metal.Network visualized with Gephi v 0.9.7.
One efflux system known to provide cross-resistance is the RND pump MdtABC.Copper and zinc induce this efflux system, which is involved in the export of antibiotics ( β-lactams, novobiocin) and metals (copper and zinc) in Salmonella and E. coli as demonstrated via by an MIC alteration following gene ov er expr ession or deletion (Nishino et al. 2007 , Wang andFierke 2013 ).This was confirmed by the significantly higher intracellular zinc ion content following mdtA or mdtC deletion compared to the wild type (Wang and Fierke 2013 ).Other known or likely cross-resistance pumps identified by knoc k out methods include MacAB in Agrobacterium tumefaciens , the deletion of which caused an accumulation of intracellular arsenite and a 2-16 fold MIC reduction to erythromycin, various penicillins and arsenite (Shi et al. 2019 ).As another example, deletion of efflux transporter MdrL in Listeria monocytogenes decreased MICs 2-10-fold to w ar ds erythrom ycin, josam ycin, clindamycin, cefotaxime, cobalt, c hr omate, and zinc (Mata et al. 2000 ).
Heter ologous expr ession also pr ovided e vidence for cr ossresistance in the case of MacAB above (Shi et al. 2019 ), with expression of this gene in E. coli giving a two-fold MIC increase to macrolides , penicillins , and arsenite .Similarl y, heter ologous expression of the GesAB gold efflux pump from Salmonella in E. coli significantl y incr eased MICs to phenicols and β-lactams (Conroy et al. 2010 ).Ov er expr ession of the E. coli copper and silv er RND efflux pump system, cusCFBA , provided three-fold enhanced resistance to fosfomycin (Nishino and Yama guc hi 2001 ) and moderate resistance to sulfamethoxazole (Conroy et al. 2010 ).T hus , data fr om man y bacterial gener a confirm that ther e ar e div erse efflux pump systems that give resistance to both antibiotics and metals.
Other cr oss-r esistance mec hanisms include extr acellular pol ysacc haride pr oduction and enzyme-mediated detoxification.The extr acellular pol ysacc haride pr oduced by Enterobacter cloacae P2B can sequester both metals and antibiotics (Naik et al. 2012 ), r esulting in m ultir esistance to lead, cadmium, mercury, β-lactams , macrolides , chloramphenicol, and sulfamethoxazoletrimethoprim.This pol ysacc haride pr oduction is compar able to the c hemical a gent pr otection offer ed by biofilms, and hence biofilm production can also be viewed as a form of crossresistance .F ew chemical modification enzymes have shown evidence for cross-resistance, but one example is DsbA-DsbB, a periplasmic disulfide bond oxidoreductase.Deleting this enzyme system in Burkholderia cepacia decreased their MIC to cadmium, zinc, and a range of antibiotics (Hayashi et al. 2000 ).It is possible that this effect of DsbA-DsbB is because both cations and antibiotics may react with thiol residues, ho w ever, this mechanism was not verified.

Evidence for co-regulation
Co-regulation is the least frequently reported metal and antibiotic r esistance mec hanism (Bazzi et al. 2020 ).This mechanism is indir ect, wher e exposur e to one a gent triggers r egulatory e v ents that result in resistance to another.Many instances of co-regulation also encompass elements of co-and cr oss-r esistance.Due to this complexity, determining a co-regulatory mechanism of resistance is difficult with metagenomic or microcosm data and requires pur e cultur e manipulations.

Co-regulation in pure cultures
In E. coli MG1655, exposure to zinc (0.2-1 mM) or copper (2 mM) led to a global regulatory response affecting 122 genes, with BasRS (Lee et al. 2005 ) and BaeRS (Nishino et al. 2007 ) as potential global r egulators.Upr egulated genes included the efflux pumps mdtABC , discussed above also in the context of cr oss-r esistance.Exposur e of P. aeruginosa to zinc (5 mM) activated the CzcR-CzcS regulatory system, which led to both the expression of the efflux pump Czc-CBA and the reduced expression of membrane porin OprD, with the latter conveying carbapenem resistance (Perron et al. 2004 ).
Comparison of the gr owth r ates of cultur es exposed to metals, antibiotics, and combinations implies co-regulation if the metal ameliorates the inhibitory growth effects of the antibiotic.One study that took this a ppr oac h with an Enterobacteriaceae isolate found that growth was faster in media containing tetracycline plus subinhibitory arsenate, copper, or zinc compared to tetracycline alone, and arsenate was found to induce the tet (34) and emrD genes (Chen et al. 2015 ).Another kind of co-regulation can arise if metals stim ulate gener al str ess r esponses that minimize the toxicity of the antibiotic.Evidence for this comes from studies with Pseudomonas fluorescens , which exhibited higher growth rates in media containing cefradine (1 mg l −1 ) and zinc ( < 160 mg l −1 ) than with cefradine alone (Xu et al. 2015a ).This was proposed to be due to differences in r eactiv e oxygen species management in the cultures, including superoxide dismutase (Xu et al. 2015a ) and nitric oxide synthase activities (Xu et al. 2015b ).Care must be taken to differentiate co-regulation from other kinds of antagonism between metal and antibiotics.For example, zinc acts as a cofactor for metallo-β-lactamases (Gupta et al. 2023 ).

Plasmid-focused studies of co-regulation
One intriguing area of study has been the impact of metals and antibiotics on the movement of MGEs , e .g. conjugation.This represents an intersection of co-resistance and co-regulation.The response of E. coli to metals appears to facilitate higher conjugation rates via increased cell membrane permeability and the upregulation of Omp porin proteins (Zhang et al. 2018b, 2019b, Wang et al. 2020, Pu et al. 2021 ).In addition, str ess r esponse genes including the SOS pathway, soxRS and oxyR are upregulated in response to metal exposure .T his is an inter esting observ ation since suc h systems are also inducers for (or induced by) MGE movement, such as in conjugation and integr on r ecombination (Guerin et al. 2009, Baharoglu et al. 2010 ).
The broad host range 60 kb IncP-1 α antibiotic resistance plasmid RP4 (also known as R18, R68, RK2, and RP1; P ansegr au et al. 1994 ) has been used for studying the effect of metals on conjugation.Transfer of RP4 is stimulated by environmentally relevant le v els (10-300 μg l −1 ) of zinc, c hr omate, silv er, and copper in both E. coli (Zhang et al. 2018b ) and Pseudomonas putida (Zhang et al. 2019b ).Experiments using whole communities of freshwater bacteria also sho w ed increased RP4 conjugation at 5-100 μg l −1 of zinc, copper, and lead (Wang et al. 2020 ).Inter estingl y, cadmium gav e some what differ ent effects, with high le v els (10-100 mg l −1 ) r equir ed to stimulate conjugation in freshwater bacteria (Pu et al. 2021 ).In the IncP-1 plasmids like RP4, metal exposure indir ectl y decr eases expr ession of the korA and korB r epr essor genes, whic h switc hes on expr ession of pr oconjugation tr aC , tr aF , trbB , and trfA pilus and relaxosome genes .T his increased conjugation frequency is likely attributed to reactive oxygen species stimulated DNA repair responses and increased membrane permeability (Zhang et al. 2018b, 2019b, Pu et al. 2021 ).A k e y thing to k ee p in mind here is that if plasmid conjugation frequency is increased by metals, this means that all the other MGEs, ARGs, and MRGs embedded within the plasmid are also tr ansferr ed at increased frequency.
Promotion of conjugation by metals has also been seen in IncP-1 ε resistance plasmids in anaerobic sludge exposed to arsenic (0.1 mM), mercury (5 μM), or lead (0.5 mM) (Lin et al. 2019 ).These plasmids also carried two-component transcriptional factors, secretion systems and efflux pumps implicated in heavy metal resistance.It is important to note that the stimulation of conjugation by metals is not a universal effect, and decreased conjugation rates were observed for IncP-1 ε plasmids in Pseudomonas , Aeromonas , Esc heric hia , and Enterobacter in the presence of cadmium, copper, or zinc (Lin et al. 2019 ).Similar findings were also seen for IncF plasmids in E. coli treated with copper, arsenate, or zinc (Palm et al. 2022 ) .Neither of these studies tested whether the transconjugants had enhanced MICs, so it is hard to say whether this r epr esents an example of co-r egulation in the sense of metals impacting antibiotic resistance.

An updated understanding of co-selection
Mechanistic insights on co-selection have progressed in recent decades, but more combinatorial approaches to genetic and phenotypic assessments are required to attain a comprehensive understanding.The distinction of co-r esistance r equir es an a ppr eciation of genomic context, MGEs, and the conditions that facilitate HGT.An assessment of cr oss-r esistance demands c har acterization of multiple substrates that a resistance gene enacts upon.Co-r egulation is e v en mor e complex to rigor ousl y identify, r equiring detailed molecular studies of gene expression.It is acknowledged that many studies fall short of the level of evidence required to confirm mechanisms of co-selection in microbial communities and the environmental conditions that facilitate this (Pal et al. 2016, Yue et al. 2020 ).Renewed efforts are needed since a better understanding of co-selection mechanisms will inform predictions and control strategies for antimicrobial resistance.

Mec hanisms dri ving co-selection
The studies discussed abov e demonstr ate that co-resistance, cr oss-r esistance, and co-r egulation ar e r eal phenomena, not just in contr olled labor atory settings , but in situ in the en vironment also.The evidence for co-resistance to date is greater than for cr oss-r esistance or co-regulation, ho w ever, this may be because the evidence for the latter two is more difficult to obtain.The notable MRGs czcA , copA , pcoA , and ARGs aacA4 , blaOXA , blaTEM , ermB , ermF , sul1 , sul2 , tetM , and tetW fr equentl y exhibit corr elations with one another.Although individual studies ma y ha ve certain biases, these MRG-ARG combinations do stand out as probable true cases of co-r esistance.Man y studies hav e focused onl y on a limited number of MRGs and ARGs (Fig. 5 ).This needs to be expanded with high-throughput methods.Plasmids are frequently found containing both ARG and MRG combinations in metal-contaminated en vironments , and the HGT of these is enhanced under metal str ess.Incr easing e vidence suggests intI1 may play a role in co-resistance as well, despite the fact that a specific integron gene cassette conferring metal resistance is yet to be found.Cr oss-r esistance and co-r egulation ar e less r eported in the liter atur e, although pr oof of concept examples hav e been wellc har acterized suc h as MacAB (Shi et al. 2019 ) and Czc with OprD (Perron et al. 2004 ).
We do not know how co-selection operates in complex communities, as most studies involve species in isolation (Karkman et al. 2018 ).Complex communities are more difficult to study, since one species may modify the effect of a metal on another, e.g.oxidation/r eduction r eactions that c hange the metal's toxicity and/or bioa vailability.T he use of mixed bacterial communities allows HGT to be studied, but only a fe w studies hav e explor ed this (P al et al. 2017 ).In some community studies, bacterial community structure is the factor with the greatest impact on ARGs, implying that r esistance genes ar e still taxonomicall y segr egated in some cases, despite manifold correlations between MRGs and ARGs (Yan et al. 2020, Sun et al. 2021, Yang et al. 2021, Huang et al. 2022 ).This offers an alternative explanation for co-selection, where taxonomic shifts within communities caused by heavy metal exposure promote species that incidentally possess ARGs (Pal et al. 2017 ).This kind of indirect selection may also underpin positive or negative corr elations r eported in other studies.
In order to unr av el the mechanisms at play, and de v elop useful pr edictiv e models, studies that measur e c hanges to the r esistome and mobilome after a ppl ying pr ecise selectiv e pr essur es ar e needed (Grenni andCorno 2019 , Li et al. 2020 ), and these should aim to obtain multiple lines of evidence at different molecular le v els (DN A, RN A, and/or pr otein).Micr ocosm experiments offer a practical and effective way to address many of these gaps.Ideally, these experiments need to be long-term (months or years) to effectiv el y document the players involv ed, their inter actions, and the effects of selection.Metagenomic analyses should extend to more than just qPCR as this technique offers little information on genomic context or genetic expression.Genomele v el r esolution is needed not only to ascribe co-resistance functions but also to determine the likely stability, mobility and expr ession le v els of genes .T he mo v ement of genes fr om plasmids into c hr omosomes tends to incr ease their stability (Gullber g et al. 2014 ), but the r e v erse pr ocess tends to increase the strength of resistance through higher copy number (Shen et al. 2020 ).Expr ession measur ements (e.g.micr oarr ays, tr anscriptomics, and RT-qPCR) ar e especiall y useful as a counterpart to 'omics' studies and can enable insights into whether a putative resistance gene is serving homeostatic or resistance functions.Further confirmation of MRG and ARG expression can be obtained with pr oteomic tec hniques suc h as enzyme activity assa ys , western blots, or ELISA (Fig. 4 ).Targeted functional assessments on putativ e r esistance genes can be performed thr ough plasmid ca pture and characterization, heterologous expression, or gene knock out/down a ppr oac hes .T hese a ppr oac hes ar e critical for expanding our knowledge of gene functions of novel environmental resis-tance genes .T his is particularly pertinent for efflux pumps , which are often hypothesized to be contributors to cr oss-r esistance, but onl y r ar el y hav e complete substr ate r ange experimentall y verified.
To complement the experimental a ppr oac hes, ur gent r e vision of bioinformatic databases is r equir ed.T he en vir onmental r esistome and mobilome are highly diverse and abundant (Pal et al. 2016 ), but only a tiny fraction of this diversity is represented in w ell-curated and w ell-annotated database entries (Pal et al. 2015, 2017, Perez et al. 2020 ).The many mismatches between expected genotypes and phenotypes that can be found in the co-selection liter atur e illustr ate that our current knowledge of resistance is far fr om compr ehensiv e. Hits to r esistance genes in such databases are further complicated by the fact that many MRGs are part of normal homeostasis machinery and may not be serving a resistance function.Resistance genes that arise from point mutations may be hard to distinguish from wild-type genes.Unfortunatel y, man y r esistance gene databases ar e either no longer curated (ARDB, MEGARes, and BacMet), or they do not distinguish between experimentall y v erified vs. pr edicted r esistance genes (C ARD and ARG-ANNO T) (Bengtsson-Palme et al. 2017 ).It is likely that there are many undetected or misannotated resistance genes and MGEs in public databases.

Envir onmental conditions pr omoting co-selection
The frequencies and impacts of resistance gene acquisition depend on a complex interplay of factors including the setting (clinical, environmental, urban, and rural) (Pal et al. 2016, Xu et al. 2017 ), physicochemical factors (Chen et al. 2019a, Zhong et al. 2021 ), bacterial taxa present (Ma et al. 2019, Huang et al. 2022 ), types of MGEs present (Pal et al. 2015, Mazhar et al. 2021 ), and of course, whic h selectiv e a gents ar e pr esent, and their concentr ations (Gao et al. 2015, Xu et al. 2015a ) Although ther e ar e str ong contenders for heavy metals that hav e co-selectiv e ca pacity, further r efinement on the concentr ation and conditions r equir ed to ac hie v e is needed.Studies on environmental bacterial communities frequently conclude that exposure to heavy metals has a greater selective effect on ARGs than exposure to antibiotics themselves (Ji et al. 2012, Hubeny et al. 2021, Mazhar et al. 2021 ).Ho w e v er, this assessment does have the complication that many antibiotics degrade rapidly, so their effect may last far longer than their detectable presence .T his can be addressed in comparative simulations, where one microcosm is exposed to only heavy metals and the other to only antibiotics.The co-selective potential of zinc and copper from agricultural or industrial sources has fr equentl y been cited, but there is still no consensus on their r elativ e contributions or importance under real-world conditions .Meanwhile , arsenic , cadmium, and manganese ar e compar ativ el y understudied but ar e consistentl y correlated with ARG dissemination (Knapp et al. 2011, Yan et al. 2020, Zhao et al. 2020 ) (Fig. 6 ).The impact of co-selection is concentration dependent (Zhang et al. 2018b ), but the MSC or MCC (Arya et al. 2021 ) for most metals is unknown.Furthermore, the MSC is subject to other factors such as bioavailability (Peltier et al. 2010, Zhong et al. 2021 ), pH (Sui et al. 2019, Chen et al. 2019a ), matrix structure (Zhang et al. 2019a, Hung et al. 2022 ), and bacterial community composition (Yan et al. 2020, Huang et al. 2022 ).Most studies do not measure the bioavailable fraction of heavy metals; this can be highly variable depending on the sample type (Olaniran et al. 2013 ).This makes it hard to compare the relationships between ARGs and heavy metals between studies.Laboratory-based studies typically focus on single selectiv e a gent tr eatments, how-e v er, this is not an accurate reflection of real-world conditions, so should be complemented with selective agents combinations to analyse the potential synergistic interactions (Gullberg et al. 2014, Zhao et al. 2018 ) that can impact bioavailability and MCCs (Song et al. 2017, Arya et al. 2021 ).
Identifying environmental reservoirs of resistance genes and determining how they are maintained and transmitted is fundamental for improving antimicrobial stew ar dship.Sediments are now recognized as fertile environments for the co-accumulation of heavy metals, MRGs, and ARGs (Nguyen et al. 2019 ).In agricultural soils, the addition of animal manure or w astew ater sludge adds both antibiotics and metals (Grenni and Corno 2019 ).This makes these a potential hotspot for co-selection, but it also makes it difficult for us to distinguish the impact of the two separate kinds of agents.To unravel these factors, it would be helpful to have more data from metal-contaminated environments that have minimal antibiotic co-contamination, e.g.industrial and mining wastes .T his would help to r edr ess the balance of r esearc h, which is currently tipped heavily to w ar ds clinical en vironments , antibiotics , and ARGs .
Knowing which kinds of bacteria harbour specific resistance genes is critical for identifying bacterial agents of concern (Pal et al. 2015, Karkman et al. 2018 ).The MRG content and MSC/MCC for heavy metals are not known for many bacteria (Gullberg et al. 2014, Song et al. 2017, Arya et al. 2021 ).Gr am-negativ e bacteria ar e gener alized to be mor e metal-r esistant (Seiler andBerendonk 2012 , Nguyen et al. 2019 ), ho w e v er, lar ge v ariations in r esistance r ates hav e been observ ed within individual species (Seiler and Berendonk 2012 ).The Enterobacteriaceae hav e r eceiv ed considerable focus due to their ease of culture, their role in human disease, and their known carriage of resistance plasmids and class 1 integrons (Li et al. 2017, Nguyen et al. 2019 ).T his o v er-r epr esentation is in some w ays w arranted, but it has ske wed meta-anal yses that aim to determine the origins and mechanisms of spread of resistance genes.
Meta genomic inv estigations can r e v eal the r esistome of both unculturable and culturable bacteria, ho w ever, attributing gene to the host can be challenging for metagenomes.Actinomycetota and Pseudomonadota phyla harbour the largest resistome in environmental communities (Yan et al. 2020, Zhao et al. 2021, Wang et al. 2021b, Huang et al. 2023 ) (see also Tables 1 and 2 ), but resistance genes are found throughout the bacterial domain, notably in Bacteroidota , Bacillota , Planctomycetota , Verrucomicrobiota , Acidobacteriota , Gemmatimonadota , and Chloroflexota (Kothari et al. 2019, Stalder et al. 2019, Zhao et al. 2021, Wang et al. 2021b, Tan et al. 2023 ).Although many metagenomic analyses do not identify the hosts of resistance genes, it has been estimated that the proportion of plasmids containing both ARGs and MRGs is higher for uncultured bacteria than culturable genera (Pal et al. 2015 ).
Linking resistance genes to hosts in a nonculture-dependent way can be done with molecular techniques such as epicPCR (em ulsion pair ed-isolation and concatenation PCR), Hi-C, or fluor escent-activ ated cell sorting (FACS).EpicPCR segregates single cells and then performs single-cell PCR that amplifies both 16S rDNA and target gene and then fuses them together (Spencer et al. 2016 ).This has been used to taxonomically link ARGs and class 1 integrons in w astew ater (Hultman et al. 2018 ) and sediments (Roman et al. 2021 ).OilPCR acts in a similar way via single-cell segregation and has enabled the linkage of β-lactamase containing plasmids to specific hosts (Diebold et al. 2021 ).Hi-C cross-links adjacent DNA, enabling plasmids to be physically bound to c hr omosomes .T his has been applied in environmental communities to link ARGs, plasmids and class 1 integrons to hosts (Stalder et al. 2019 ).If the MGE and/or host can be genetically manipulated, fluorescent markers may be incorporated and then the movement of the MGE, ARGs, and MRGS can be monitored with FACS.Sequencing of the sorted fluor escent fr action may yield novel recipients that cannot be cultured.Indeed, this technique has discovered ARG-carrying IncP plasmids being transferred to 11 different phyla (Klümper et al. 2015 ).
Along with knowing which bacterial taxa harbour particular ARGs and MRGs, identifying the MGE vectors for these genes is also important.Our understanding of the types and functions of MGEs in the environment is limited, and databases and bioinformatic methods need to be impr ov ed to help organize the vast diversity of MGEs in genomes and meta genomes (Jor gensen et al. 2015, Delaney et al. 2018 ).It is important to be aware that some biases exist in sequence databases, especially those arising fr om meta genomic studies.Small, high-copy-number plasmids ar e pr obabl y ov er-r epr esented because the y survi v e DNA extr action better, are easier to assemble, and are preferentially amplified by multiple displacement amplification (Jorgensen et al. 2015, Kothari et al. 2019 ).The sequencing technologies can also introduce biases, such as high GC-rich regions impacting sequencing methods (Chen et al. 2013 ).Our understanding of most plasmid biology arises from the study of plasmids that are compatible with standard lab hosts .T his systematic bias warrants more work to de v elop an accur ate pictur e of the plasmidome of environmental bacterial communities, so this can be better utilized to combat resistance.

Conclusions
A holistic a ppr oac h is needed to understand how ARGs and MRGs interact in bacterial communities to enable better management strategies for multidrug-resistant bacteria.The use and disposal of heavy metals must be critically reviewed to mitigate not only their direct toxicity but also their other effects on the acquisition and persistence of antimicrobial resistance genes .T here are enough correlational studies to warrant further, more mechanistic investigations into heavy metal and antibiotic co-selection.The intricacies of how different combinations and concentrations of metals, bacterial species, and other factors interact need to be explored.It remains unknown to what extent the models of co-r esistance, cr oss-r esistance, or co-r egulation occur in bacterial communities, but it seems plausible that all contribute to coselection.The de v elopment of better models of co-selection requir es a gr eater understanding of the MGEs involv ed and the r ates of different HGT processes, especially as these influence the mobilization of resistance markers from environmental organisms to pathogens.

Figure 1 .
Figure 1.Bacterial sensing and resistance mechanisms for metals and antibiotics.(A) Metal efflux by the Czc of Cupriavidus metallidurans CH34.(B) Antibiotic efflux by AcrAB-TolC from Escherichia coli.(C) Chemical modification by Cue and Cus from E. coli .(D) Cleavage of β-lactam antibiotic by CTX-M in Enterobacteriaceae .(E) Sequestration of metal by Smt of Synechococcus PCC 7942.(F) Sequestration of mitomycin by Mrd, followed by export by Mct in Streptomyces lavendulae .Metal ions are shown as small spheres, coloured grey (Zn 2 + ), blue (Cu + ), or green (Cu 2 + ).The pill symbol (red/white) r epr esents antibiotics, specificall y azithr omycin, cipr ofloxacin, gentamicin, and β-lactams in (B), β-lactams in (D), and mitomycin in (F).Genes are shown as block arrows with the corresponding proteins in the same colour.Icons created with smart.servier.com.

Figure 2 .
Figure 2. Comparison of co-selection mec hanisms.(A) Co-r esistance demonstr ated by ARG and MRG genes residing on the same plasmid, and thus are inherited sim ultaneousl y. (B) Cr oss-r esistance demonstr ated by an efflux pump ejecting both antibiotics and metals.(C) Co-r egulation demonstr ated by an ARG and MRG expressed together after induction by a metal.Grey spheres represent metal ions, and red and white pills represent antibiotics.Icons created with smart.servier.com.
. There are very few examples of defined MSC or MCC values, but in theory, they should be equal to or lo w er than MIC values .T his is significant since although metals or antibiotics r ar el y r eac h the MIC in the environment (Wales and Davies 2015 , Zhang et al. 2018b ), they likely exceed the MSC (Gullberg et al. 2014 , Zhang et al. 2018b ) or MCC (Seiler and Berendonk 2012 , Song et al. 2017 ).The MSC and MCC values will v ary gr eatl y depending on the combination of agents, the type of microbial community and the environmental conditions (Gullberg et al. 2014 , Arya et al. 2021 ).The qualitative relationships between MIC , MSC , and MCC are shown in Fig. 3 .
al. 2023 ), with simultaneous quantification of hundreds of resistance genes(Mazhar et al. 2021, Huang et al. 2022 ), and in-depth c har acterization of the mobilome(Kothari et al. 2019, Perez et al. 2020 ).Greater sequencing ca pabilities hav e massiv el y expanded bioinformatic databases focused on resistance genes and MGEs .T hese databases include CARD(McArthur et al. 2013 ), ResFinder(Florensa et al. 2022 ), MEGARes(Lakin et al. 2017 ), ARG-ANNOT(Gupta et al. 1999 ), ARDB(Liu and Pop 2009 ), Resqu (Resqu 2023 ), BacMet (Pal et al. 2014 ), AMRFinder (Feldgarden et al. 2021 ), ACLAME (Leplae et al. 2004 ), INTEGRALL (Moura et al. 2009 ), ISFinder (Siguier et al. 2006 ), and PlasmidFinder (Carattoli et al. 2014 ).These databases ar e inv aluable for informing subsequent hands-on r esearc h, e.g.gene knoc k outs (Wang and Fierke 2013 , Bisc hofber ger et al. 2020 ), heter ologous expr ession (Conr oy et al. 2010 , Shi et al. 2019 ), and plasmid ca ptur e assays (Fig. 4 ) (Wang et al. 2020 , Pu et al. 2021 ).Metagenomic data on ARGs , MRGs , and MGEs has allo w ed new insights into co-selection.For example, Pal et al .( 2017 ) raised again the important question of how resistance genes are often F igure 3. Relationships betw een MIC , MSC , and MCC as a function of concentration.Selection pr essur e fr om a gent A incr eases with concentr ation from the MSC until it approaches the MIC, after which it rapidly declines due to cytotoxicity.Co-selection pressure on agent B increases from the agent A MCC until the MIC is r eac hed.maintained in the absence of a ppar ent selection, and they challenged the importance of metals as factors driving HGT rather than other inherent properties of the MGEs themselv es.Mor e manipulativ e sim ulations of heavy metal exposur e (suc h as micr ocosms) were recommended to delve into mechanisms that would counterbalance the abundance of environmental case studies.T he en vir onmental origin of r esistance genes and the selective effects of subinhibitory le v els of metals are now well known, and this One Health understanding has been used to propose methods of clinical intervention (metal chelators and efflux blockers) and site rehabilitation (bioremediation or biosorption)(Vats et al. 2022 ).

Table 1 .
Examples of heavy metal resistance mechanisms , genes , targets , and bacterial hosts taxa.Gene names ar e synon ymous with pr otein pr oduct.

Table 2 .
Examples of antibiotics resistance mechanisms , genes , targets , and bacterial host taxa.Protein name only indicates when it differs from the gene name.