Experimental Evolution of Copper Resistance in Escherichia coli Produces Evolutionary Trade-Offs in the Antibiotics Chloramphenicol, Bacitracin, and Sulfonamide

The rise in antimicrobial resistant bacteria have prompted the need for antibiotic alternatives. To address this problem, significant attention has been given to the antimicrobial use and novel applications of copper. As novel applications of antimicrobial copper increase, it is important to investigate how bacteria may adapt to copper over time. Here, we used experimental evolution with re-sequencing (EER-seq) and RNA-sequencing to study the evolution of copper resistance in Escherichia coli. Subsequently, we tested whether copper resistance led to rifampicin, chloramphenicol, bacitracin, and/or sulfonamide resistance. Our results demonstrate that E. coli is capable of rapidly evolving resistance to CuSO4 after 37 days of selection. We also identified multiple de novo mutations and differential gene expression patterns associated with copper, most notably those mutations identified in the cpx gene. Furthermore, we found that the copper resistant bacteria had decreased sensitivity when compared to the ancestors in the presence of chloramphenicol, bacitracin, and sulfonamide. Our data suggest that the selection of copper resistance may inhibit growth in the antimicrobials tested, resulting in evolutionary trade-offs. The results of our study may have important implications as we consider the antimicrobial use of copper and how bacteria may respond to increased use over time.


Introduction
The emergence of multi-drug resistant (MDR) bacteria is a global public health concern, resulting in a need for antibiotic alternatives. One promising approach includes the use of metal-based antimicrobials that possess potent antibacterial abilities that are capable of killing a wide range of bacteria, including MDR strains [1,2]. The use of antimicrobial metals dates back thousands of years, although a general decline was observed when traditional antibiotics were introduced and started being used in the early 20th century [3]. However, the approach of a post-antibiotic era has prompted renewed interest in the use of antimicrobial metals.
The antimicrobial use of copper (Cu) and its alloys are of particular interest as a means of controlling bacterial infections within clinical and commercial settings. In February 2008, Cu became the first solid antimicrobial metal to be registered by the United States Environmental Protection Agency [4]. Generally, studies have indicated that the implementation of

E. coli K-12 MG1655 Adapts to Excess CuSO 4
After 37 days of selection in CuSO 4 , the Cu-selected populations displayed superior 24 h growth, according O.D. 600nm measurements, in comparison to the control and ancestor populations in CuSO 4 ( Figure 1). Greater differences in growth were observed at 8-62 mg/L of CuSO 4 . There was a further difference between the ancestor and control populations, in that the control populations demonstrated slightly better growth in varying CuSO 4 concentrations, but not population growth. F-statistics and their p-values for all comparisons are shown in Table 1.
To ensure that the effects observed in the Cu-selected populations were due to excess Cu and not to SO 4 , we grew all the populations in varying concentrations of MgSO 4 ( Figure 2) . We did not observe any differences between the ancestral, control, and Cuselected populations in MgSO 4 . To ensure that the effects observed in the Cu-selected populations were due to excess Cu and not to SO4, we grew all the populations in varying concentrations of MgSO4 (Figure 2). We did not observe any differences between the ancestral, control, and Cu-selected populations in MgSO4.  Table 1. General linear model (GLM) results for phenotypic assays comparison for each population (Cu-selected, control, and ancestor) in each tested antimicrobial. The general linear model is an analysis of variance to determine the effect of the population (Cu-selected, control, and ancestral), the concentration (range tested for each antimicrobial), and their interaction. The GLM computes an Fstatistic and the probability of achieving that F-statistic by chance (p value). NS means not significant.

Antimicrobial
Concentration   Table 1. General linear model (GLM) results for phenotypic assays comparison for each population (Cu-selected, control, and ancestor) in each tested antimicrobial. The general linear model is an analysis of variance to determine the effect of the population (Cu-selected, control, and ancestral), the concentration (range tested for each antimicrobial), and their interaction. The GLM computes an Fstatistic and the probability of achieving that F-statistic by chance (p value). NS means not significant.

Genomic Results
Whole-genome sequencing and analysis was conducted on all populations and compared to the E. coli K-12 MG1655 reference genome (NC_9000913) to determine genetic mutations that may be associated with Cu resistance. Genomic changes associated with the ancestor were reported in our previous studies and can be found in Appendix A, Table A1. Single nucleotide polymorphisms and selective sweeps were observed in the Cu-selected populations and reported in Table 2, along with the descriptions of the de novo mutations. We found six hard sweeps in the Cu-selected populations, with an increase in frequency from the ancestor (0.000) that exceeded 0.500 to 1.000 (fixation) in five genes, yidL, yfcZ, rho, cpxP, and cpxA. Of the hard sweeps, two were found in replicate Cu2, three were found in Cu4, one was found in Cu6 and one near-sweep was found in Cu8. Eleven cpx mutations were found in seven of the nine replicates, with seven found in cpxA, two found in cpxR, and two more found in cpxP, suggesting significance for adapted Cu resistance. We also identified two mutations in the rpoD gene, one at fixation and another major variant found only in Cu6. Three rho mutations (one hard selective sweep, one major variant, and another minor variant) were identified in Cu5, Cu6, and Cu9. Other hard sweeps were identified in Cu4, yfcZ/fadL, a UPF0381 family protein/long-chain fatty acid outer membrane transporter (UniProtKB-P10384), and yidL, an uncharacterized transcription regulator (UniProtKB-P31449). All other variants were found in the adhE, yddA, yidK, yidL, and fis genes.

Functional Annotation Analysis of Upregulated and Downregulated Genes
The DAVID functional annotation clustering tool grouped upregulated genes found in four or more Cu-selected replicate populations into one significant functional cluster-ABC transport and ATP binding-with a significant enrichment score of 1.3 (Table 3). Downregulated genes clustered into three significant functional clusters. The reported clusters were associated with bacterial flagellum, iron transport, and cell division ( Table 3). The top two clusters each had an enrichment score of 1.3 or above and may be important for Cu resistance.

Differences in Gene Expression for Upregulated and Downregulated Genes
Hierarchical cluster analysis of the differentially expressed genes was displayed using heatmaps that were generated using R version 3.5.3. The rows and columns of the heatmaps were clustered based on similar observations. Genes that were upregulated in at least four replicates with a fold-change of 2 or above that were not expressed in the controls are shown in Figure 3. Three genes involved in ABC (adenosine tri-phosphate binding cassette) transport, mdtA, ybbP, and acrD, were upregulated in at least half of the Cuselected replicates. Both acrD and ybbP were upregulated in all eight of the replicates. AcrD is an aminoglycoside membrane efflux pump (UniProtKB-P24177); however, YbbP is uncharacterized. Another gene that was upregulated in all eight replicates was xynR, which encodes a transcriptional regulator of xylonate catabolism (UniProtKB-P77300). The most notable of the genes found in four replicate populations was mdtA, which encodes a multi-drug resistance-ABC transporter that requires the TolC efflux pump for function (UniProtKB-P76397). moeA involved in metal binding was upregulated in seven replicates. The most highly upregulated gene was bioA, adenosylmethionine-8-amino-7-oxononanoate aminotransferase, which is associated with biotin biosynthesis (UniProtKB-P12995). bioP, which is also responsible for biotin uptake (UniProtKB-P0ADP5), was upregulated in five replicates. The remaining genes, yigA, hemF, spoT, uvrA, and ycjF, were upregulated in at least four of the Cu-selected replicate populations.
There were many downregulated genes in four or more replicates ( Figure 4). Most of the downregulated genes were associated with inner/outer membrane integrity, iontransport, and flagella biosynthesis. Of the downregulated genes, wbbL was downregulated in all replicate populations. wbbL, which encodes for a rhamnosyltransferase, was involved in outer-membrane biogenesis (UniProtKB-P36667).

24 h Growth Curves for Antibiotics
The 24 h growth measurements for each population (Cu-selected, control, and ancestor) were performed in varying concentrations of rifampicin, chloramphenicol, bacitracin, and sulfonamide to assess whether Cu resistance confers resistance to other antimicrobials ( Figure 5A-D). F-statistics and p-values for each phenotype comparison are shown in Table 1. Generally, adaptation to Cu did not provide cross-resistance to the antibiotics tested in this study. R PEER REVIEW 9 of 24 There were many downregulated genes in four or more replicates ( Figure 4). Most of the downregulated genes were associated with inner/outer membrane integrity, iontransport, and flagella biosynthesis. Of the downregulated genes, wbbL was downregulated in all replicate populations. wbbL, which encodes for a rhamnosyltransferase, was involved in outer-membrane biogenesis (UniProtKB-P36667).

24 h Growth Curves for Antibiotics
The 24 h growth measurements for each population (Cu-selected, control, and ancestor) were performed in varying concentrations of rifampicin, chloramphenicol, bacitracin, and sulfonamide to assess whether Cu resistance confers resistance to other antimicrobials ( Figure 5A-D). F-statistics and p-values for each phenotype comparison are shown in Table 1. Generally, adaptation to Cu did not provide cross-resistance to the antibiotics tested in this study. tor and control populations showed superior growth compared to the Cu-selected populations in concentrations of 4-125 mg/L ( Figure 5C). However, there was no difference observed in concentrations of 250 mg/L to 1000 mg/L. Similar results were found in the presence of sulfonamide, where the ancestors also had greater growth at mild concentrations (16-25 mg/L) when compared to the controls and the Cu-selected populations (Figure 5D). Generally, there was not a significant difference in growth when comparing the Cu-selected and control populations in the presence of sulfonamide.  In the presence of chloramphenicol, the ancestors demonstrated slightly better growth when compared to the controls and the Cu-selected populations at 4-16 mg/L ( Figure 5A). However, there was not a clear difference in growth between the controls and the Cu-selected populations in chloramphenicol ( Figure 5A. In rifampicin, the controls demonstrated better growth when compared to the ancestor and Cu-selected populations ( Figure 5B). The Cu-selected populations did, however, demonstrate slightly better growth compared to the ancestors. When comparing O.D. values in bacitracin, the ancestor and control populations showed superior growth compared to the Cu-selected populations in concentrations of 4-125 mg/L ( Figure 5C). However, there was no difference observed in concentrations of 250 mg/L to 1000 mg/L. Similar results were found in the presence of sulfonamide, where the ancestors also had greater growth at mild concentrations (16-25 mg/L) when compared to the controls and the Cu-selected populations ( Figure 5D). Generally, there was not a significant difference in growth when comparing the Cu-selected and control populations in the presence of sulfonamide.

Discussion
Little attention has been paid to the de novo evolution of copper resistance and the possible dissemination of copper-resistant bacteria. Even less attention has been paid to how evolved copper resistance may produce correlated resistances to other antimicrobials. Previous studies demonstrated that copper-in ionic and nanoparticles forms-is antibacterial, with MICs ranging from 32 µL/mL to 960 µg/mL [26][27][28]. Here, we showed that Cu resistance evolved rapidly in E. coli after 37 days of selection at a sub-inhibitory concentration of 75 mg/L CuSO 4 , via de novo mutations and gene expression changes. The Cu-selected populations grew better-i.e., they had higher O.D. values-than the control and ancestral populations at concentrations up to 62 mg/L of CuSO 4 ( Figure 1). Although the Cu-selected populations demonstrated superior growth from 2-62 mg/L, Cu selection did not improve growth in 125-1000 mg/L of CuSO 4 . This suggested that superior growth in Cu is limited to a selection concentration at or below 75 mg/L of CuSO 4 . However, the Cu-selected populations may be capable of adapting to higher concentrations of CuSO 4 if this experiment were extended beyond 37 days of selection. In addition, we observed that the Cu-selected populations demonstrated superior growth in the absence of Cu, as shown in Figure 1. However, we did not explicitly investigate whether Cu selection leads to a selective advantage in a non-Cu environment, as that was beyond the scope of this study.
The most notable of our genomic results were the eleven mutations identified in cpx, which encompassed almost half of all the mutations identified. CpxAR makes up a two-component response system (TCRS) that responds to envelope stress [29], while CpxP inhibits the CpxAR response in the absence of that stress [30]. Previous studies have shown that CpxAR is activated in response to Cu toxicity and that excess Cu levels lead to the expression of genes regulated by CpxAR [22,29]. It has also been reported that CpxAR-controlled transcription may play a role in Cu tolerance in Salmonella by preventing Cu toxicity and restoring redox balance in the Salmonella envelope [31]. Additionally, adaptation to Cu stress in E. coli and S. enterica requires activation of the envelope stress response regulators [32,33], which is monitored by CpxAR [33,34]. The results of our study confirm that cpxA mutations are important for Cu adaptation caused by envelope stress stimulated by excess Cu [35]. This conclusion is supported by the fact that these mutations were only observed in the Cu-selected populations.
Populations with mutations in CpxR (PDB 4UHK) [36] include either an L20R mutations or a D104N mutation. As shown in Figure 6A, these two residues are located opposite one another at the ends of helices 1 and 5, respectively. Each of these mutations change the charge on the residue, likely disrupting helix stability within the structure of the protein.
The Q128* mutation detected in CpxP results in a premature stop codon that would prevent translation of an entire helix (Figure 6B red). CpxP (PDV 3QZ3) [30] carries two conserved LTXXQ motifs (green) and Q128 (blue) in the second of these conserved motifs. Based on this mutation, which leads to a premature stop codon, it is very likely that this mutation disrupts the structure sufficiently so that it results in a non-functional protein. Finally, we were unable to map all CpxA (PDB 5LFK) [36] mutations onto structures available in the PDB, as only the cytoplasmic transmitter core domain was crystallized. Of the unmodeled mutations, three-R93H, V94A, and K121R-are located in the periplasmic sensor domain. Interestingly, previous work has shown that the deletion of residue 93-124 results in a constitutively active protein that upregulates the cpx regulons [37]. L178R is located within the second transmembrane helix and K188Q is the first residue of the HAMP domain directly out of the phospholipid bilayer. Unmodeled K194T and modeled P215L are also both located within the HAMP domain, and the modeled A262E mutation is located in the DHp domain [38]. These domains are important for structural rearrangement during activation and dimerization; all three mutations result in major changes in both size and charges of the residues, indicating that the function of this protein will likely be altered, although from current data we cannot yet predict whether that will be an increase or a decrease in activation/activity. Conversely, we identified multiple rpo mutations found in one Cu-selected replicate and in the control populations. We suggest that the rpo mutations found in our study may be important for adaptation to DMB, since rpo mutations were observed in the control populations of our previous studies [39][40][41][42][43]. under the control of the stress response system CpxAR, which regulates the expression of genes involved in outer membrane biogenesis. Although not previously described, we suggest that the downregulation of waaL and waaO may be important for Cu resistance.
Although the results of our gene expression studies are interesting, future studies are needed to better assess our hypothesis, and proteomic studies may provide further insight into how these genomic changes impact Cu resistance. The potential cross-resistance across different antimicrobials, such as metal antibiotic antimicrobials, may hinder long-term use of antimicrobial metals. To test this, the control, ancestor, and Cu-selected bacteria were grown in various concentrations of antibiotics. Generally, Cu-adapted populations grew worse in the presence of the antibiotics, chloramphenicol, bacitracin, and sulfonamide, when compared to the control and ancestor populations ( Figure 5), resulting in evolutionary trade-offs. However, our findings contrast with previous reports [50][51][52]. For example, Yu et al. (2022) found that Cu hydroxideresistant E. coli were also resistant to chloramphenicol and tetracycline [50]. Other studies have shown that Cu exposure co-or cross-selects for resistance to antibiotics [53,54]. Generally, those studies isolated bacteria from Cu-contaminated environmental sites and attributed copper and antibiotic resistance to exogenous genes [50][51][52][53][54]. Here, we selected for Cu resistance using a laboratory strain without plasmids; accordingly, all Cu-associated mutations and correlated responses were the result of changes found on the chromosome. Additionally, the results of this study may have important implications for infectious disease control. For example, previous studies on antimicrobial copper primarily Bacteria utilize metal transport systems to maintain metal homeostasis by allowing for the uptake of trace metals and the expulsion of excess metal, to avoid toxicity [44]. In bacteria, ABC transporters transport metal ions across the cell membrane to maintain metal homeostasis [45]. Among these transporters are P1B-type ATPases that export toxic levels of Cu from the cytosol [46]. It was predicted that the majority of transition metal transporting P1B-type ATPases responsible for cytoplasmic metal efflux are important for resistance to metal toxicity [46]. Thus, it was not surprising that a large percentage of the upregulated genes identified in multiple replicates were found in multiple ABC transporters pathways (Figure 3). In addition, according to our DAVID functional annotation cluster analysis, genes that fell within the category of ABC-transport/ATP-binding had the highest enrichment score (see Table 3). Two genes that were upregulated in multiple replicants-acrD, which encodes the ABC transporter aminoglycoside efflux pump, and mdtD, a multidrug efflux transporter gene-were shown to contribute to Cu resistance in Salmonella [47]. We suspect that the upregulation of acrD and mdtD may also contribute to Cu resistance in E. coli.
Surprisingly, bioA was upregulated in all eight Cu-selected replicate populations. The bioA gene encodes the protein adenosylmethionine-8-amino-7-oxononanoate aminotransferase, which catalyzes the transfer of the alpha-amino group from S-adenosyl-Lmethionine (SAM) to 7-keto-8-aminopelargonic acid (KAPA) to form 7,8-diaminopelargonic acid (DAPA) (UniProt-P12995), which is a part of the biotin biosynthesis pathway. In biotin synthase, abstraction of a hydrogen atom from the C9 position of DTB leads to the formation of a dethiobiotinyl radical, which reacts with a nearby [2Fe-2S] 2 + cluster, generating a new carbon-sulfur bond [48]. It has been proposed that Cu toxicity causes damage to Fe-S clusters [16], perhaps upregulating bioA-a biotin synthesis pathway gene is a way that Cu-selected strains are accommodating for Cu ion-induced damage to biotin's Fe-S clusters-but this proposal remains to be further elucidated. We also identified two additional genes associated with Fe-S clusters and biotin synthesis, hemF and bioP, to be upregulated in multiple replicates.
We also identified many downregulated genes in our gene expression studies. Based on our DAVID functional annotation clustering analysis for downregulated genes, the top cluster was associated with bacterial flagella structural components and synthesis (see Table 3). Flagellum is a motility organelle that participates in protein export, adhesion, and biofilm formation. In E. coli, the regulation of flagella genes is important for the transition from the planktonic state to the biofilm state [49]. Although flagella are needed for the initial formation of biofilms, previous studies suggested that the inhibition of motility may promote biofilm formation [49]. Flagella synthesis genes flgM (UniProt-P0AEM4), a negative regulator of flagellin synthesis, and flgN (UniProt-P43533, an initiator of filament assembly biosynthesis, were downregulated in multiple Cu-selected populations. In addition, structural components flagella-basal body genes, flgB and flgG were downregulated for multiple replicates (UniProt-P0ABW9, UniProt-P0ABX5). It is well known that biofilms are inherently more resistant to metals than planktonic cells. Perhaps downregulation of flagella genes may be important for Cu adaptation.
The genes waaO and waaL both of which serve important roles in lipopolysaccharide synthesis, were downregulated in all eight replicates. Lipopolysaccharides are molecules located in the outer membrane of Gram-negative bacteria, which are important for the structural integrity of the bacterial cell. WaaL(rfaL) is involved in the LPS core biosynthesis pathway, which is part of the bacterial outer membrane biogenesis (UniProtKB-P27243). The WaaL protein adds the O-antigen on the glucose group of LPS. The O antigen provides protection from the antimicrobial action of certain antibiotics. WaaO (rfaI) is an LPS 1,3 galactosyltranferase involved in the synthesis of the R core lipopolysaccharide (UniProtKB-Q9ZIS5). Cytoplasmic stress response is activated upon the block of LPS, under the control of the stress response system CpxAR, which regulates the expression of genes involved in outer membrane biogenesis. Although not previously described, we suggest that the downregulation of waaL and waaO may be important for Cu resistance. Although the results of our gene expression studies are interesting, future studies are needed to better assess our hypothesis, and proteomic studies may provide further insight into how these genomic changes impact Cu resistance.
The potential cross-resistance across different antimicrobials, such as metal antibiotic antimicrobials, may hinder long-term use of antimicrobial metals. To test this, the control, ancestor, and Cu-selected bacteria were grown in various concentrations of antibiotics. Generally, Cu-adapted populations grew worse in the presence of the antibiotics, chloramphenicol, bacitracin, and sulfonamide, when compared to the control and ancestor populations ( Figure 5), resulting in evolutionary trade-offs. However, our findings contrast with previous reports [50][51][52]. For example, Yu et al. (2022) found that Cu hydroxide-resistant E. coli were also resistant to chloramphenicol and tetracycline [50]. Other studies have shown that Cu exposure co-or cross-selects for resistance to antibiotics [53,54]. Generally, those studies isolated bacteria from Cu-contaminated environmental sites and attributed copper and antibiotic resistance to exogenous genes [50][51][52][53][54]. Here, we selected for Cu resistance using a laboratory strain without plasmids; accordingly, all Cu-associated mutations and correlated responses were the result of changes found on the chromosome. Additionally, the results of this study may have important implications for infectious disease control. For example, previous studies on antimicrobial copper primarily focused on the effect of incorporating copper surfaces in commonly touched areas in clinical settings. Generally, these studies indicated that the incorporation of copper-based surfaces reduces bacterial growth to levels that discourage nosocomial infections in healthcare facilities [5][6][7][8][55][56][57]. However, research that focused on the potential of bacteria to evolve resistance to antimicrobial copper in clinical settings over time is limited. Results indicate that Cu-resistance occurs rapidly and should be further explored. We also showed that Cu-resistance increased chloramphenicol, bacitracin, and sulfonamide sensitivity, suggesting that copper may provide a nonpharmacological approach to increasing antibiotic susceptibly. However, further studies are needed to investigate this hypothesis.
Although cpx mutations have been associated with antibiotic resistance in E. coli [58], we suspect that metal-antibiotic cross-resistance was not observed, as cpx mutations found in our study may impact downstream processes that are associated with copper homeostasis. However, further investigations are needed to determine the validity of this hypothesis.
Interestingly, the control populations showed superior growth in the presence of rifampicin. We suspect that rpo mutations resulting from adaptation to the DMB media may have contributed to this result.
As antibiotic resistance continues to rise, it is important to understand how bacteria may adapt to antibiotic alternatives. As a result, understanding how background resistances impact bacteria's ability to develop future resistance may provide a greater fight in creating sustainable antimicrobials. Overall, the results of our study demonstrate that E. coli rapidly evolves resistance to excess Cu via genomic changes, and that Cu adaptation limits its ability to persist in the presence of the other antimicrobials tested. These results may have important implications for the use of Cu as an antimicrobial.

Bacterial Strains and Culture Media
E. coli K-12 MG1655 (ATCC #47076) was used for this study. It was chosen because of its sensitivity to metals and its lack of metal or antibiotic resistance. E. coli K-12 MG1655 does not contain any plasmids and its chromosome has 4,641,652 nucleotides (GenBank:NC_000913.3). All growth experiments in this study were carried out using Davis Minimal Broth (DMB, Difco TM , Sparks, MD, USA) supplemented with 1 g per liter dextrose as a sole carbon source, with thiamine hydrochloride 0.1% (Thiamine Hydrochloride, Fisher Scientific, Fair Lawn, NJ, USA).

Experimental Evolution
The experimental evolution by serial transfer method used in this study was modeled after our previous studies [39][40][41][42][43]. To start the evolution experiment, the E. coli K-12 MG1655 stock strain was cultured in 10 mL of DMB, serial-diluted, and plated onto DMB agar plates. Four individual colonies were picked and grown overnight in standard DMB media. To determine the minimum inhibitory concentration (MIC), overnight cultures were diluted to 0.05 O.D. 600nm and added to 96-well plates in triplicate in 2× fold increasing concentrations of CuSO 4 (0-5000 mg/L) diluted in autoclaved H 2 O. O.D. 600nm measurements were taken at 0 and 24 h. The 0-h O.D. 600nm measurement readings were subtracted from the 24 h readings and averaged for statistical analysis. An aliquot of the original culture was stored at −80 • C in 50% glycerol to be used as the ancestral strain. Once the MIC was determined, a sub-inhibitory concentration of CuSO 4 (II) (75 mg/L) was used to start the selection experiment. Ten CuSO 4 selected (Cu-selected) populations and five control populations originated from independent colonies from the same culture as the E. coli K-12 MG1655 ancestor. Each population was maintained in 50 mL Erlenmeyer flasks (Pyrex, Corning New York, NY, USA) that contained 10 mL (0.1 mL of each culture into 9.9 mL of DMB) of total culture were placed in a shaking incubator at 250 rpm (I2500 series, New Brunswick Eppendorf, Hamburg, Germany), at 37 • C for 24 h. Cultures were propagated by daily transfers. The selection experiments were continued until Cu resistance was confirmed by phenotypic assays (37 days of selection). The Cu-selected replicate populations were maintained with the addition of 75 mg/L of CuSO 4 , while controls were maintained in standard DMB only. Aliquots of each population were stored at −80 • C with 50% glycerol once a week to be used as a "fossil record". There was a total of ten Cu-selected replicate populations; however, one replicate went extinct at approximately 14 days, leaving a total of nine replicates (Cu1, Cu2, Cu3, Cu4, Cu5, Cu6, Cu8, Cu9, and Cu10) and five controls (C1, C2, C3, C4, and C5).

24 h Growth Assay in CuSO 4
All the phenotypic growth assays used archived glycerol stocks that were grown overnight in 10 mL DMB supplemented with 75 mg/L of CuSO 4

24-h Growth Assay in MgSO 4
To ensure that the Cu-selected populations evolved resistance to Cu, and not SO 4 ions, we measured the 24 h growth of the ancestor, control, and Cu-selected populations in

Whole-Genome Sequencing and Sequence Analysis
To identify single nucleotide polymorphisms and insertion element localization associated with Cu resistance in E. coli K-12 MG1655, we employed whole-genome resequencing after resistance was observed at 37 days of selection. DNA samples from the Cu-selected, control, and ancestral populations were extracted and purified via the Nucleic Acid Sample Preparation for Downstream Analysis: Principles and Methods Kit (GE Health Sciences 2014).
Genomic libraries were prepared using the following standard protocol, as described in the manufacturer's instructions (Illumina, Nextera DNA Flex Library Prep Reference Guide Part # 20,018,704, 20,018,705). The genomic libraries were sequenced using the Illumina MiSeq platforms at the Joint School of Nanoscience and Nanoengineering (JSNN). The depth of sequencing was between 40-80× for all samples. Ancestral mutations associated with our sample of this strain were removed from the analysis and the reads were mapped to the reference genome for E. coli K-12 MG1655 (NC_000913). Sequence alignment and variant calling from the samples was achieved by use of the breseq 0.30.0 pipeline [59]. The breseq pipeline uses three types of evidence to predict mutations, which include read alignments (RA), missing coverage (MC), and new junctions (JC). Any reads that indicate a difference between the sample and the reference genome that cannot be resolved to describe precise genetic changes are listed as "unassigned" [59]. These unassigned reads were not described or interpreted here. This analysis was used to identify genomic differences between control and Cu-selected bacteria, suggesting which physiological systems may be contributing to Cu resistance.

RNA Extraction
Once adaptation was observed at 37 days of CuSO 4 selection, eight Cu-selected replicate populations (Cu1, Cu3, Cu4, Cu5, Cu6, Cu8, and Cu9) and four controls (C1, C2, C3, and C5) were prepared for RNA-sequencing. Cu4 and C1 were omitted due to low RNA quality. The gene expression data of the ancestral population is not shown, as it is not adapted to Cu or to the Davis Minimum Broth (DMB). Each replicate population was cultured under two conditions, Davis Minimum Broth (DMB) only and DMB with a sub-inhibitory concentration (75 mg/L) of CuSO 4 in a shaking incubator at 37 • C at 250 rpm. A total of 1 × 10 6 cells were harvested in mid-log phase at an O.D. 600 of 0.50. Total RNA was extracted, according to the manufacturer's instructions, using the Monarch Total RNA Miniprep Kit. We further evaluated the RNA samples with a QuantiFluor RNA system, using the manufacturer's standard protocol. Total RNA was stored at −80 • C prior to shipping to Novogene Corporation Inc. (Sacramento, CA, USA) for library construction and RNA-sequencing.
RNA degradation and contamination were monitored on 1% agarose gels. The RNA concentration and integrity number (RIN) were measured on an Agilent Bio Analyzer 2100 system. Only RNA samples with a RIN > 6 were used for cDNA library construction. Therefore, eight Cu-selected and four controls were cultured in a sub-inhibitory concentration of CuSO 4 and four Cu-selected and one control cultured in DMB were used in this experiment. All cDNA libraries were sequenced using a paired-end strategy (read length 250-300 bp) on an Illumina NovaSeq 6000 platform. Novogene stringently analyzed the sequencing data to ensure accuracy and reliability, including, quality, error rate, Q20, Q30, and adapter contamination rate. Raw reads with low quality, containing adapter sequences, or with >10% of "N" base calls were removed to obtain the clean reads.

Differential Gene Expression Analysis
To analyze gene count data, the STAR output was input into DESeq2 R/Bioconductor package version 1.27.27, with the aim of finding genes that are differentially expressed across populations for the two conditions, DMB only and DMB with CuSO 4 . DEseq2 uses a negative binomial generalized linear model to analyze high-throughput sequencing data [54]. DESeq2 incorporates count data as a two-column table; for each sample, the number of sequence fragments that have been assigned to each gene [60]. The count data table was generated from the STAR ReadsPerGene.out.tab file, where data in the first and the fourth columns were extracted shown in the Appendix A, Table A3.
Each bioconductor software package often has a special class of data object, which contains special slots and requirements [60]. In the case of DESeq2, the data object is DESe-qDataSet, which was used to store read counts and the intermediate estimated quantities during statistical analysis. The simplest design formula for differential expression would be~condition, where each condition is a column in colData (dds) that specifies which group the samples belong to [60]. This was used to identify differentially expressed genes across populations for the two conditions, DMB (Davis Minimal Broth) only and DMB with CuSO 4 .

Fold-Change Analysis
DESeq2 was used to calculate the fold-change differences for four pairwise comparisons, as shown in Appendix A, Table A4 Table A3).

Functional Annotation of Expressed Genes
The Database for Annotation, Visualization, and Integrated Discovery (DAVID), version 6.8, was used to group differences in expressed genes based on functional similarity to enhance biological interpretation. DAVID is comprised of multiple functional annotation tools used to cluster large sets of genes to understand the biological meaning behind large lists of genes. Here, we used the functional annotation clustering tool to classify uniquely upregulated and downregulated genes associated with Cu resistance. This tool suite provides typical batch annotation and gene-GO term enrichment analysis to highlight the most relevant GO terms associated with a given gene list [61]. Genes were considered upregulated if the fol-change was at 2 or above and downregulated if fold-change was at 0.5 and below. Functional annotation clustering integrates the same techniques of kappa statistics to measure the degree of the common genes between two annotations, to classify the groups of similar annotations according to kappa values [61]. Thus, genes with common annotation themes will most likely be grouped together. The functional annotation tool identified four functional clusters for upregulated genes based on the enrichment score (see Table 3). The enrichment score ranks the overall importance of annotation term groups. It is the geometric mean of all the enrichment P-values (EASE scores) of each annotation term in the group [61]. The geometric mean is a relative score instead of an absolute P-value, minus log transformation, which is applied on the average P-values. An enrichment score of 1.3 would be the equivalent to a non-log score of 0.05. Therefore, a higher score for a group indicates that annotation term members in the group are playing a more important (enriched) role in a given study [61].

Differential Gene Expression
Hierarchical cluster analysis of the differentially expressed genes was displayed using heatmaps that were generated using R version 3.5.3. Genes that were upregulated in at least four replicates with a fold-change of 2 and above were considered upregulated, while genes 0.5 or below found in at four least replicates were deemed downregulated. We chose genes that were upregulated or downregulated in at least four replicates to represent the expression of genes found in at least half of the Cu-selected populations.

24-h Growth Assay Antibiotics
To determine whether adaptations to excess CuSO 4 (CuII) lead to cross-resistance in other antimicrobials, we measured the growth of the ancestor, control, and Cu-selected populations in varying concentrations of rifampicin, chloramphenicol, bacitracin, and sulfonamide. Optical density (O.D.) measurements at 600 nm were taken at 0 and 24 h in 0, 4,8,16,31,62,125,250, 500, and 1000 mg/L for each antimicrobial. Archived glycerol stocks for each population were grown and diluted to O.D. 600nm of 0.05 and added to 96-well plates in triplicate, as described in the previous section.

Statistical Analysis
All statistical analysis comparing the 24 h growth rate for the ancestor, control, and Cu-selected populations in Cu and antibiotics were conducted using an ANOVA via the General Linear Method option on IBM SPSS ver. 23. ANOVA was used to determine the effect of population type (Cu-selected, control, ancestor), the concentration, and the interaction of population type and concentration on 24 h growth in varying concentrations of the metals and antibiotics tested.

Conclusions
The widespread use of antimicrobial copper may contribute to the emergence of copper-resistant bacteria. Furthermore, potential cross-resistance across antimicrobials, such as metal-antibiotic resistance, may hinder the long-term use of metals as antimicrobials. The results of this study indicate that copper resistance in E. coli can occur rapidly-i.e., in 37 days-via genomic changes reducing the ability to grow in the presence of the antibiotics tested. Our results may have important implications for the continued and future use of antimicrobial copper.