**Using a Chemical Genetic Screen to Enhance Our Understanding of the Antibacterial Properties of Silver**

**Natalie Gugala 1,†, Joe Lemire 1,†, Kate Chatfield-Reed 1,†, Ying Yan 2, Gordon Chua <sup>1</sup> and Raymond J. Turner 1,\***


Received: 7 June 2018; Accepted: 3 July 2018; Published: 6 July 2018

**Abstract:** It is essential to understand the mechanisms by which a toxicant is capable of poisoning the bacterial cell. The mechanism of action of many biocides and toxins, including numerous ubiquitous compounds, is not fully understood. For example, despite the widespread clinical and commercial use of silver (Ag), the mechanisms describing how this metal poisons bacterial cells remains incomplete. To advance our understanding surrounding the antimicrobial action of Ag, we performed a chemical genetic screen of a mutant library of *Escherichia coli*—the Keio collection, in order to identify Ag sensitive or resistant deletion strains. Indeed, our findings corroborate many previously established mechanisms that describe the antibacterial effects of Ag, such as the disruption of iron-sulfur clusters containing proteins and certain cellular redox enzymes. However, the data presented here demonstrates that the activity of Ag within the bacterial cell is more extensive, encompassing genes involved in cell wall maintenance, quinone metabolism and sulfur assimilation. Altogether, this study provides further insight into the antimicrobial mechanism of Ag and the physiological adaption of *E*. *coli* to this metal.

**Keywords:** silver; silver toxicity; silver resistance; Keio collection; *Escherichia coli*; antimicrobials

### **1. Introduction**

For centuries, metal compounds have been deployed as effective antimicrobial agents [1]. The use of silver (Ag) for antimicrobial purposes is a practice that dates back thousands of years [2] and is still implemented for medical purposes in an effort to curtail the rise of antimicrobial resistant pathogens [3–6], a threat that has once again surfaced as a clinical challenge [7–10].

Applications of Ag-based antimicrobials include: wound dressings [11] and other textiles [12], antiseptic formulations [13], nanoparticles [14], coatings [15], nanocomposites [16], polymers [17], and part of antibiotic combination therapies [18]. Many of these approaches have proven to be effective in controlling and eradicating pathogenic microorganisms.

Presently, research in this field focuses on finding new formulations and utilities for Ag-based antimicrobials. Despite this, the identity of the cellular targets that are involved in Ag antimicrobial activities are known to a far lesser degree [19]. This current knowledge gap hinders the potential utility of Ag-based antimicrobials, and in turn the expansion of this metal as a therapeutic agent.

Previous studies examining the mechanisms of Ag resistance and toxicity have not provided a complete understanding of the global cellular effects of Ag exposure on the bacterial cell. Further, several studies fail to build upon preceding work and the literature is replete with contradicting reports, in part due to non-standardized conditions of study. Furthermore, it has been demonstrated that the speciation/oxidation state of Ag has substantial influence on toxicity, a factor that is dependent on the source of Ag ions [4], growth conditions, and is further complicated by the organism (species and strain) of interest [6].

Proposed mechanisms of metal toxicity include the production and propagation of reactive oxygen species through Fenton chemistry and antioxidant depletion, the disruption of iron-sulfur clusters, thiol coordination and the exchange of a catalytic/structural metal that leads to protein dysfunction, interference with nutrient uptake, and genotoxicity [19]. Microorganisms are able to withstand metal toxicity through several mechanisms such as reduced uptake, efflux, extracellular and intracellular sequestration, repair, metabolic by-pass and chemical modification [20]. Whether these mechanisms are solely responsible for cell death or resistance has yet to be determined. Still, what is understood is that metals demonstrate broad-spectrum activity and decreased target specificity [19] when compared to conventional antimicrobials.

In this work, we hypothesized that Ag exerts its effects on multiple targets both directly and indirectly, and thus various cellular systems may be altered by Ag exposure. To test this, we performed a genotypic screening workflow of a mutant library composed of 3985 strains, each containing a different inactivated non-essential gene in *Escherichia coli*. Using a comparable genome-wide workflow [21] and by use of transcriptomic profiling [22,23], similar approaches have been implemented in order to study the mechanisms of action caused by Ag. Despite this, genes conferring resistance to Ag when absent have been studied and compared to a far lesser degree than those that result in sensitivity when absent. Further, many previous approaches aimed at studying Ag toxicity and resistance have primarily examined the effects of Ag shock or rapid pulses of exposure, followed by the evaluation of gene expression. Hence, as a means of complementing existing work, we have identified a number of genes that are implicated in prolonged Ag resistance and/or toxicity, and mapped their metabolic function to their respective cellular system.

### **2. Materials and Methods**

### *2.1. Escherichia coli Strains and Storage*

The Keio collection [24]—a mutant library of 3985 single-gene *E. coli* BW25113 mutants (*lacI*q, *rrnB*T14, Δ*lacZ*WJ19, *hsdR*514, Δ*ara*BADAH33, Δ*rha*BADLD78)—was obtained from the National BioResource Project *E. coli*, (National Institute of Genetics, Shizuoka, Japan). All strains were initially stored at −80 ◦C in vials containing Lysogeny Broth (LB) media (VWR International, Mississauga, ON, Canada) with 30% glycerol (VWR International). For chemical genetic screening, the Keio collection was transferred, and subsequently arrayed into 96-well microtiter plates containing LB medium with 30% glycerol. Construction of the arrayed Keio collection and pre-culturing of the *E. coli* strains was carried out on LB agar (1.0%). For chemical genetic screening, M9 minimal media (6.8 g/L Na2HPO4, 3 g/L KH2PO4, 1 g/L NH4Cl, 0.5 g/L NaCl, 4 mg/L glucose, 0.5 mg/L MgSO4 and 0.1 mg/L CaCl2) containing Noble agar (1.0%) with and without silver nitrate (AgNO3) was used for treatment and control testing, respectively (all obtained from VWR International).

Although the Keio collection strains are engineered with a kanamycin resistance cassette in place of the gene of interest [24], for our experiments here, we did not include this antibiotic in the growth media. Synergistic antimicrobial effects were found when bacterial cells were grown in the presence of Ag and kanamycin (data not included).

### *2.2. Stock Ag Solution*

Silver nitrate (AgNO3) was obtained from Sigma-Aldrich (St. Louis, MO, USA). Stock solutions of Ag were made at equivalent molarities of Ag in distilled and deionized (dd)H2O and stored in glass vials for no longer than two weeks.

### *2.3. Determination of the Minimal Inhibitory Concentration and Controls*

The minimal inhibitory concentration was determined using a known Ag sensitive strain (*cusB*) and negative control strains (*lacA* and *lacY*). CusB is a part of the CusCFBA copper/silver efflux system [25]; therefore, it was anticipated that the absence of this gene would confer toxicity, denoted as a Ag sensitive hit. Further, LacA and LacY, are not expected to be involved in Ag resistance or toxicity. The aforementioned strains, along with the parent strain (wild type (WT)) were grown at 37 ◦C on M9 minimal media and Noble agar (1%) in the presence or absence of Ag at varying concentrations. The Ag concentration found to visibly decrease colony size in the *cusB* mutant and demonstrate no changes in colony size in the *lacA* and *lacY* mutants was selected. Furthermore, the latter mutants and the WT were grown in the presence of 100 μM ionic nitrate to ensure growth was not impeded by the accompanying counter ion. The full chemical genetic screen was challenged at time zero of inoculation in the presence of 100 μM AgNO3.

### *2.4. Screening*

M9 minimal media Noble agar (1%) plates were prepared two days prior to use. Colony arrays in 96-format were produced and processed using a BM3 robot (S&P Robotics Inc., Toronto, ON, Canada). The strains were spotted using a 96-pin replicator, allowing for uniform application. Cells were transferred from the arrayed microtiter plates using the replicator onto LB agar plates. These plates were then grown overnight at 37 ◦C. Once grown, the colonies were spotted using the replicator onto two sets—with and without 100 μM AgNO3—of M9 minimal media Noble agar plates, and subsequently grown overnight at 37 ◦C. Images of both sets of plates were acquired using the spImager (S&P Robotics Inc.) and colony size, which is a measure of fitness, was determined using integrated image processing software. For each 96-colony array, four technical trials per strain were combined onto a single plate in 384-colony array format and three biological trials were performed. Therefore, each strain was tested a total of 12 times.

### *2.5. Normalization*

Experimental factors such as incubation time and temperature, local nutrient availability, colony location, gradients in the growth medium and neighboring mutant fitness were all considered as independent variables that could contribute to systematic variation, and subsequently affect colony size. As a result, the colonies were normalized and scored using Synthetic Genetic Array Tools 1.0 (SGATools) [26]. Firstly, all of the plates were normalized to establish identical median colony size working on the assumption that most colonies exhibited WT fitness. Next, to ensure the colonies were directly comparable, colonies were rescaled, a factor that is primarily important for colonies close to the edge of the plate. Further, spatial smoothing accounted for partialities in each plate owing to inconsistencies, such as the thickness of the agar. Very large colonies, likely an indication of contamination among other factors, and those that were different from the corresponding technical replicates were removed. Lastly, colonies that were larger than anticipated and located next to colonies that were found to be smaller than anticipated were marked as potential false-positive hits.

Following this normalization, the colonies were scored. Here, paired evaluation was completed by comparing the colony size (in the presence of Ag) to a matched control (in the absence of Ag). Fitness values were established, and the subsequent scores represented deviation from the fitness of the WT strain. Once normalized and scored, colonies displaying a reduction in size were indicative of a Ag sensitive hit and those displaying an increase in colony size qualified as a Ag resistant hit. Finally, the *p*-value was calculated as a two-tailed *t*-test and significance was determined using the Benjamini-Hochberg procedure, as a means of lowering the false discovery rate, which was selected to be 0.1.

### *2.6. Data Mining and Analyses*

Subsequent analyses were conducted using Pathway Tools Omics Dashboard, which surveys against the EcoCyc database [27]. This allowed for clustering of the hits into systems, subsystems, component subsystems, and lastly, into individual objects. It is important to note that genes can be found in multiple systems, since many are involved in a number of cellular processes.

Further, in order to identify biological processes most prominent under Ag challenge, enrichment analyses were conducted for the Ag resistant and Ag sensitive hits. To analyze the gene list, the Database for Annotation, Visualization and Integrated Discovery (DAVID) bioinformatics resource was utilized [28,29]. Lastly, as a means of exposing the direct (physical) and indirect (functional) protein-protein connectivity between the gene hits, the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) database was used [30]. Interactive node maps based on experimental, co-expression and gene fusion studies were generated based on genes defined in our chemical genetics screen.

### **3. Results and Discussion**

### *3.1. Genome-Wide Screen of Ag-Resistant and Ag-Sensitive Hits*

The chemical genetic screen completed in this work provided a method for genome-wide probing of non-essential genes involved in Ag-sensitivity or -resistance in *E. coli*. A total of 3810 non-essential genes were screened for growth in the presence of 100 μM AgNO3 (Supplementary Materials, Table S1). 3073 mutants displayed little change in colony size in the presence of Ag with a normalized fitness score between ±0.1 (Figure 1). The statistical colony size cutoff that indicated a significant difference in fitness was selected to be ±0.15, or two standard deviations from the mean. This resulted in 225 gene hits, which represents approximately 5% of the open reading frames in the *E*. *coli* genome. The remaining gene hits were not regarded as significant hits in this work based solely on the cut offs selected. In general, the normalization was performed on the assumption that Ag does not specifically interact with the deleted gene but rather impedes growth due to environmental stress. In short, those displaying hits between the cut off values were assumed to have non-specific or neutral interactions with Ag.

It is important to note that when reflecting on the data generated from our chemical genetic screen, it is the absence of the gene that imparts the Ag-resistant or -sensitive phenotype. Upon Ag exposure, an increase in colony size (>0.15) is suggestive of an Ag-resistant hit, and therefore the presence of this gene is proposed to confer toxicity. On the contrary, a decrease in colony size (<−0.15) was designated to be an Ag-sensitive hit, and therefore the presence of this gene is proposed to confer resistance. In total, the deletion of 106 and 119 genes resulted in Ag-resistant and -sensitive hits, respectively (Tables 1 and 2). These gene hits were mapped to their corresponding cellular systems using EcoCyc (Figure 2 and Table A1). In short, genes were found in multiple cellular systems, validating our hypothesis that Ag cytotoxicity and the corresponding physiological responses of *E. coli* involve a number of cellular mechanisms.

**Figure 1.** Synthetic Array Tools (version 1.0) was used to normalize and score the silver (Ag)-resistant and -sensitive gene hits as a means of representing the growth differences in *Escherichia coli* K12 BW25113 in the presence of 100 μM silver nitrate (AgNO3). Only those with a score greater or less than ±0.15, respectively, were selected for further analysis. Hits between ±0.15 were regarded as having neutral or non-specific interactions with Ag. The *p*-value was a two-tailed *t*-test and significance was determined using the Benjamini-Hochberg procedure; false discovery rate was selected to be 0.1. Each individual score represents the mean of 12 trials.


**Table 1.** Ag-resistant hits organized according to system and subsystem mined using the Omics Dashboard (Pathway Tools), which surveys against the EcoCyc Database; genes represent resistant hits, each with a score >0.15 and a false discovery rate of 0.1 1,2.


**Table 1.** *Cont.*

<sup>1</sup> Each individual score represents the mean of 12 trials—three biological and four technical; <sup>2</sup> Two-tailed *t*-test and significance was determined using the Benjamini-Hochberg procedure; <sup>3</sup> Gene hits can be mapped to more than one system and subsystem.


**Table 2.** Ag-sensitive hits organized according to system and subsystem mined using the Omics Dashboard (Pathway Tools), which surveys against the EcoCyc Database; genes represent resistant hits, each with a score <−0.15 and a false discovery rate of 0.1 1,2.


**Table 2.** *Cont.*

<sup>1</sup> Each individual score represents the mean of 12 trials—three biological and four technical; <sup>2</sup> Two-tailed *t*-test and significance was determined using the Benjamini-Hochberg procedure; <sup>3</sup> Gene hits can be mapped to more than one system and subsystem.

**Figure 2.** Ag-resistant and -sensitive gene hits mapped to component cellular processes. The cutoff fitness score implemented was −0.15 and 0.15 (two standard deviations from the mean) and the gene hits with a score less or greater than, respectively, were chosen for further analyses. The hits were mined using the Omics Dashboard (Pathway Tools), which surveys against the EcoCyc Database. Several gene hits are mapped to more than one subsystem. The *p*-value was calculated as a two-tailed *t*-test and significance was determined using the Benjamini-Hochberg procedure; false discovery rate was selected to be 0.1. Each individual score represents the mean of 12 trials.

Comparable numbers of Ag-resistant and -sensitive hits were mapped in the systems 'Response to stimulus'—starvation, heat, cold, DNA damage, pH, detoxification, osmotic stress, and other, 'Cellular processes'—cell cycle and division, cell death, genetic transfer, biofilm formation, quorum sensing, adhesion, locomotion, viral response, response to bacterium, host interactions with host, other pathogenesis proteins, and 'Degradation'—amino acids, nucleotide, amine, carbohydrate/carboxylate, secondary metabolite, alcohol, polymer and aromatic, the cell exterior,

and regulation. A greater number of Ag-resistant than -sensitive hits were mapped to the processes 'Biosynthesis'—amino acids, nucleotides, fatty acid/lipid amines, carbohydrate/carboxylates, cofactors, secondary metabolites, and other pathways and 'Other pathways'—detoxification, inorganic nutrient metabolism, macromolecule modification, activation/inactivation/interconversion and other enzymes.

In total, 49 and 73 resistant and sensitive hits, respectively, were found to be a part of the 'Cell exterior'—transport, cell wall biogenesis and organization, lipopolysaccharide metabolism, pilus, flagellar, outer and inner membrane, periplasm, and cell wall components. Compared to the latter cellular processes, non-essential genes comprising 'Energy' processes—including glycolysis, the pentose phosphate pathway, the tricarboxylic acid (TCA) cycle, fermentation, and aerobic and anaerobic respiration were found to be involved in Ag toxicity or resistance the least, by more than seven-fold when compared to genes mapped to the 'Cell exterior'.

Based on the fold enrichment, metal binding proteins were affected to the same degree in both Ag-resistant and -sensitive groups, displaying an enrichment score <5 (Figure 3). However, when examining proteins involved with specific metals in more detail, such as zinc and magnesium, fold enrichment values were >5, but only for the Ag-sensitive hits (Figure 3). Cellular and anaerobic respiration were represented by the Ag-sensitive hits only, while processes involved in amino acid biosynthesis were heavily enriched for by the Ag-resistant hits.

A number of hits were found to be involved with the cell membrane using EcoCyc's system of classification, but this was not detected in the fold enrichment analysis. Here, cell membrane proteins were affected three-fold less than the most highly represented clusters, which were amino acid biosynthesis and phosphoproteins for the Ag-resistant and -sensitive hits, respectively (Figure 3).

**Figure 3.** Functional enrichment among the Ag-resistant and -sensitive gene hits. The DAVID gene functional classification (version 6.8) database, a false discovery rate of 0.1 and a score cutoff of −0.15 and 0.15 (two standard deviations from the mean) were used to measure the magnitude of enrichment against the genome of *E. coli*. Processes with a *p*-value < 0.05, fold enrichment value ≥3 and gene hits >3 are included only. Each individual score represents the mean of 12 trials.

### *3.2. Ag-Resistant Gene Hits*

### 3.2.1. Regulators of Gene Expression

When examining processes of the 'Central dogma'—systems involved in replication, and transcription to translation—in more detail, each subsystem had a mean score between 0.211 and 0.294 (Figure 4a). Despite this consistency, transcription and RNA metabolism contained the greatest number of Ag-resistant hits, 12 and 16, respectively. The protein EttA—energy-dependent translational throttle protein [31], can be found within the subsystems translation and protein metabolism. EttA is sensitive to the energy state of the cell. This protein represses translational elongation in response to high ADP/ATP, stimulating dipeptide bond synthesis in the presence of ATP (cell high energy state) and vice versa. As a result, EttA may inhibit translation in Ag-treated cells due to the occurrence of high ADP/ATP ratios. The absence of EttA might allow for increased translation of proteins, such as RecA [32] or CusB [25], which may result in Ag resistance. Furthermore, six proteins involved in proteolysis were found to confer resistance when absent, such as Prc. This enzyme is a periplasmic protease that processes and degrades specific proteins, has been found to provide resistance against a number of small hydrophilic antibiotics and causes the leakage of periplasmic proteins when absent [33]. Antibiotic resistant mechanisms have been compared to those of metal ions, drawing on similarities such as substrate modification or sequestration. The leakage of the periplasmic proteins in *prc* mutants may result in Ag sequestration, thereby causing metal resistance.

**Figure 4.** *Cont.*

**Figure 4.** Ag-resistant gene hits plotted against respective cellular processes. Y-axis representative of the normalized score, smaller circles represent the individual hits and the larger circles represent the mean of each subsystem. The *p*-value was calculated as a two-tailed *t*-test and significance was determined using the Benjamini-Hochberg procedure; false discovery rate was selected to be 0.1. Each individual score represents the mean of 12 trials. (**a**) Central Dogma; (**b**) Cell exterior; (**c**) Biosynthesis; (**d**) Degradation; (**e**) Other pathways; (**f**) Energy; (**g**) Cellular processes; and (**h**) Response to stimulus. Plots constructed using Pathway Tools, Omics Dashboard.

### 3.2.2. Cell Membrane Proteins

It has been demonstrated that Ag may exert toxicity and potentially impede growth by acting on the cell membrane [34,35]. In this study, 49 coding genes that resulted in Ag resistance when absent were determined to be a part of the 'Cell exterior', which includes proteins of the cell membrane, periplasm and extracellular structures (Figures 2 and 4b). Of these, 25 genes coded for plasma membrane proteins, and while Ag was observed to enter bacterial cells [23], the exact mechanism of import has yet to be determined. Loss of the porin genes *ompC* and *ompF* was observed to confer resistance to Ag [36]. While these two genes were not detected within our cut offs, we did recover two additional porin genes (*ompA* and *ompG*) as conferring Ag resistance when deleted. Relative to this, it has been demonstrated that a mechanism of entry for zinc into the cell is co-transport with low molecular weight metabolites via transport proteins found within the membrane [37]. Further, ExbB, a Ag-resistant hit with a score of 0.241, is part of the energy transducing Ton system that transports iron-siderophore complexes and vitamin B12 across the outer membrane [38]. Collectively, these findings provide insight into possible mechanisms of Ag import, such as entry through porins, co-transport with metabolites or the replacement of Ag with other ions predetermined for import. The enrichment analysis offered further evidence for this hypothesis, as a number of ion transport proteins and proteins pertaining to the cell membrane were involved in Ag resistance when absent (Figure 3). Furthermore, MngA, a permease that simultaneously phosphorylates 2-*O*-α-mannosyl-D-glycerate in a process called group translocation, contains two putative phosphorylation sites His87 and Cys192 [39]. Thiols are regarded as soft bases, and according to the hard-soft acid base theory, which is key to the reactivity and coordination of metals [40], cysteine, and to a lesser degree methionine and imidazole chemically interact with Ag(I) with high affinity. Therefore, proteins with key structural or catalytic thiols/imidazols are possible Ag interacting sites.

### 3.2.3. Biosynthetic Enzymes

Eight hits were found to be involved in the biosynthesis of amino acids and 10 hits were found to be involved in cofactor/prosthetic group/electron carries catabolism (Figure 4c). When examining the functional enrichment analysis, serine, glycine, threonine, arginine and proline biosynthetic processes were highly enriched, on average three-fold more than the remaining cellular processes (Figure 3). The third step in the synthesis of NAD+ from L-aspartate occurs via the enzyme NadC—quinolinate phosphoribosyltransferase [41] and based on our data the absence of this protein confers resistance in *E. coli.* In fact, the genes coding for the first and second steps of de novo NAD<sup>+</sup> synthesis, NadB—L-aspartate oxidase and NadA—quinolinate synthase, respectively, were also found to be Ag resistant hits. NadA contains a [4Fe-4S] cluster that is required for activity [42]. Soft metals have the capacity to inactivate dehydratases in vitro via iron-sulfur cluster degradation, possibly leading to the bridging of the sulfur atoms [43]. As a result, proteins with iron-sulfur centers are of possible interest when examining the interactions of Ag with cellular biomolecules. Furthermore, it has been demonstrated that H2O2 formation is diminished via the addition of precursors involved in the synthesis of NAD+ [44]. The absence of one gene involved in NAD<sup>+</sup> biosynthesis may result in metabolite accumulation since there is no evidence of negative precursor feedback inhibition. Therefore, there is a possibility that deletion of the *nadA*, *nadB* or *nadC* may confer resistance if H2O2 is generated in the presence of Ag.

Using the STRING database, several points of interaction were revealed. Among the Ag-resistant hits, the latter genes involved in de novo NAD<sup>+</sup> production were connected to proteins a part of amino acid biosynthesis, including *trpB*, *aroC*, and *metL* (Supplementary Materials, Figure S2).

### 3.2.4. Catabolic Enzymes

Genes encoding enzymes functioning in the catabolism of metabolites, such as amino acids, fatty acids, carbohydrates and polymers, were underrepresented compared to anabolism (Figures 2 and 4d). In fact, in the functional enrichment analysis, degradation processes were not represented within the cutoffs selected (Figure 3). The gene *idcA*—L,D-carboxypeptidase, a component of secondary metabolite and polymer degradation, had an elevated score of 0.311. IdcA is essential for murein turnover [45]. Murein processing is an important energy-conserving activity that transports cell wall components from the exterior of the cell to the cytoplasm [46]. Evidence has demonstrated that during logarithmic growth, the *idcA* mutant strain displays a decrease in the overall cross-linkage of murein, causing a reduction in turnover and the abundance of murein being transported into the cell. In turn, this may result in the transport of fewer Ag ions, which may have bound to the cell wall, into the cell thereby prompting increased resistance in the *idcA* mutant strain. Metal nanoparticles have been proposed to target the outer membrane regions of bacteria due to strong electrostatic interactions and co-coordination of the metal with the lipopolysaccharide or similar cell wall structures [47]. The particles are proposed to release ionic Ag, likely triggering toxicity through membrane damage and facilitating the entry of excess Ag ions.

### 3.2.5. Sulfur Metabolism Proteins

Within the subsystem inorganic nutrient metabolism, a part of 'Other pathways', which also includes processes such as macromolecule modification and activation/inactivation/interconversion (Figure 4e), one pathway was found to be affected by Ag exposure—sulfur metabolism. CysH—phosphor-adenylsulfate reductase is involved in assimilatory sulfate reduction by catalyzing the reduction of 3 -phospho-adenylylsulfate to sulfite and adenosine 3 ,5-biphospahte (PAP). This protein contains highly conserved cysteine residues that become oxidized to form a disulfide bond [48]—possible targets based on the affinity of Ag for sulfur. Moreover, the *cysC*, *cysD* and *cysI* genes, also involved in the pathway sulfate reduction I (assimilatory) via phosphorylation, adenylation and reduction, respectively, were also Ag-resistant hits. These sulfate assimilatory proteins are linked

to the Ag-resistant hit CysQ, which is involved in the recycling of PAP and has been experimentally determined to be the main target of lithium toxicity [49] (Supplementary Materials, Figure S2). The protein CysI, contains a siroheme and one [4Fe-4S] cluster per polypeptide chain [50]. Comparably, it has been demonstrated that the exposure of Ag nanoparticles upregulates the expression of several genes involved in iron and sulfate homeostasis [22], including those aforementioned. A decrease in the activity of this pathway reduces the amount of hydrogen sulfide required for processes such as L-cysteine biosynthesis, and since Ag interacts with sulfur compounds well, such as hydrogen sulfide—the final product of sulfate reduction I—fewer Ag targets may be available when genes of this pathway are deleted. CysH had the highest score of 0.360 out of all four sulfur assimilatory genes, and since this protein interacts with thioredoxin, the absence of CysH may free reduced thioredoxin, thus providing elevated resistance in presence of reactive oxygen species that may arise under Ag stress.

### 3.2.6. Biofilm Formation

In total 19 genes in the 'Cellular processes' system, which includes subsystems such as genetic transfer, quorum sensing, adhesion and locomotion, were found to confer resistance when absent (Figure 2). Three hits were involved in cell cycle and division, and two were found to be involved in biofilm formation (Figure 4g), such as CsgF, which is an outer membrane protein that initiates curli subunit polymerization, and therefore involved in the colonization of surfaces and biofilm formation [49]. In the absence of CsgF, less biofilm is formed, and according to our results, Ag resistance is generated. Biofilms commonly provide resistance in the face of fluctuating or threatening environments [51]; however, studies have shown that bacterial residence within a biofilm does not always provide enhanced resistance against metals [6,52,53], an observation supported by this work. An explanation for this may reside in the ability of biofilms to sequester Ag ions by attracting them to varying components of the extracellular polymeric matrix. While this may provide resistance, it may also concentrate ions within a localized area, thereby causing greater sensitivity. Similarly, Ag nanoparticles have been shown to inhibit *E. coli* biofilm formation by potentially targeting curli fibers [54], therefore the absence of curli fibers may promote Ag resistance. Previous studies, which have found that biofilm formation are a source of Ag resistance [32], were completed under differing culture conditions, therefore direct comparisons are challenging.

### 3.2.7. DNA Damage and Repair

The effect of Ag exposure on DNA damage and repair in *E. coli* has been inconsistent from several studies involving gene deletion strains. Radzig et al. showed that several deletion strains lacking in the ability to excise DNA bases were sensitive to Ag exposure, but not the Δ*recA* strain, which is involved in SOS repair [36]. In contrast, the Δ*recA* deletion strain showed Ag sensitivity in a previous study [32]. From our list of Ag resistant hits, six mutants were identified within the DNA damage subsystem (Figure 4h) including the Δ*dam* strain. Dam is methyltransferase that functions in mismatch DNA repair in *E. coli* and may also play a role in controlling oxidative damage. Based on this protein's function, we expect that the deletion strain of the gene would exhibit Ag sensitivity, potentially due to a deficiency in DNA repair of oxidative damage. However, the *dam1*Δ strain exhibits an upregulation of RecA and constitutive SOS activity which may be the nature of the Ag resistance exhibited in this mutant [36]. Moreover, we also identified several other Ag-resistant strains from our screens (*purF*, *damX*, *dcd*, *ruvC* and *ompA*) that are also known to possess RecA-mediated constitutive SOS activity [32].

### *3.3. Ag-Sensitive Hits*

### 3.3.1. Central Dogma and Cell Exterior Proteins

Within the 'Central dogma', 56 mutants resulted in Ag sensitivity (Figure 5a). For example, *ruvA*, a gene found to be involved in DNA repair, had a normalized score of −0.430 [55]. Direct DNA damage has not been attributed to Ag exposure; however, in the presence of reactive oxygen species potentially triggered by Ag exposure, the propagation of Fenton active iron may cause DNA damage [19,56].

In total of 73 genes were mapped to the system 'Cell exterior' (Figure 5b). The gene *ygiZ*, which codes for a putative inner membrane protein, had a score of −0.751, the lowest value of any protein in this screen. A common resistance mechanism employed by microbes is the export of the challenge from the periplasm or interior of the cell to the extracellular space [20]. The fold enrichment analysis supported this finding—cell membrane proteins and those involved in the ion transport were highly enriched (Figure 3). In total, 13 transport proteins conferred Ag sensitivity when absent, such as *cusB*, which encodes for a component of the copper/silver export system CusCFBA in *E. coli*, and contains several methionine residues important for function [57]. In the absence of this protein, sensitivity is anticipated, since the cell is unable to expel Ag ions. Another Ag sensitive hit was Lpp, considered to be the most abundant protein in *E*. *coli* [56]. Cells lacking Lpp have been found to be hypersensitive to toxic compounds [56], potentially because there is less protein available to sequester the incoming threat. In addition, the protein TolC was an Ag resistant hit. This protein is required for the function of a number of efflux systems including the AcrAB multidrug efflux system, which is involved in the export of a number of toxic exogenous compounds [58]. In contrast to efflux proteins, we identified the *cysA* and *cysP* genes—thiosulfate and sulfate permeases—to be sensitive hits when absent. CysA and CysP function in the first step of cysteine biosynthesis, which may be important in Ag resistance since this metal may target cysteine residues via thiol side chains [32].

**Figure 5.** *Cont.*

**Figure 5.** Ag-sensitive gene hits plotted against respective cellular processes. Y-axis representative of the normalized score, smaller circles represent the individual hits and the larger circles represent the mean of each subsystem. The *p*-value was a two-tailed *t*-test and significance was determined using the Benjamini-Hochberg procedure; false discovery rate was selected to be 0.1. Each individual score represents the mean of 12 trials. (**a**) Central Dogma; (**b**) Cell exterior; (**c**) Biosynthesis; (**d**) Degradation; (**e**) Other pathways; (**f**) Energy; (**g**) Cellular processes; and (**h**) Response to stimulus. Plots constructed using Pathway Tools, Omics Dashboard.

### 3.3.2. Lipopolysaccharide Biosynthetic Genes

In total, 18 Ag-sensitive hits were mapped to 'Biosynthesis processes' (Figure 5c). Processes associated with lipopolysaccharide biosynthesis were highly represented in the enrichment analysis (Figure 3). FabF, a key protein involved in fatty acid biosynthesis, and *clsB*—cardiolipin synthase B were found to be Ag-sensitive hits. If Ag targets the cellular membrane, lipid biosynthesis/regeneration could serve as a mechanism of Ag resistance and consequently, Ag toxicity would be increased if either of these processes were compromised *via* the deletion of these candidate genes.

Processes of biomolecule degradation were affected to a lesser degree than biosynthesis (Figure 2). Only nine hits were mapped to this system (Figure 5d). The mutant *hcaD* had the second lowest score of −0.707 in this screen. This protein is a predicted ferredoxin reductase subunit that is involved in the degradation of aromatic acids as carbon sources.

### 3.3.3. Three Ag-Sensitive Hits Comprise the ATP Synthase Fo Complex

Seven hits were mapped to 'Energy processes' (Figure 5f). Of these, three are components of the ATP synthase Fo complex—AtpB, AtpE and AtpF. Ag has been suggested to damage the respiratory chain of *E*. *coli* [59], thereby preventing the efficient pumping of protons across the membrane. Small disruptions to the Fo complex may amplify this consequence and render this biological process hypersensitive. If this mechanism is correct and the cytoplasmic membrane becomes more permeable to protons, than the cell will attempt to compensate for this increase in acidity via several mechanisms, one being the reversal of ATP synthase in order to pump protons outward (if ATP is not limiting) and decrease cytoplasmic proton concentrations [60]. If the ATP synthase complex exhibits decreased activity due to disruptions in any of the subunits, this resistant mechanism may be unable to function properly, resulting in greater Ag sensitivity.

Several nodes of interaction based on the STRING connectivity maps were made evident within this cluster of proteins such as the association of *atpF* and *atpB* to *gshB* and several putative membrane proteins, *tufB*—elongation factor Tu and *ppgL*—a putative zinc peptidase (Supplementary Materials, Figure S3).

### 3.3.4. Oxidative Stress Response Genes

Out of the 31 proteins mapped to 'Response to Stimulus', 24 were involved in mediating DNA damage and other processes (Figure 5h). The gene coding for glutathione synthetase—*gshB* was found to be an Ag-sensitive hit. Strains overexpressing either GshA or GshB are more resistant to oxidative damage, and this system has been shown to mediate metal resistance [61]. As a result, the deletion of either gene is anticipated to cause Ag sensitivity. Furthermore, the putative Fe+2 trafficking protein, YggX was found to have a score of −0.450. This protein is proposed to play a role in preventing the oxidation of iron-sulfur clusters [62], a proposed mechanism of Ag toxicity. The absence of this protective protein may result in sensitivity since it can be found at elevated concentrations in vivo and it is involved in mediating oxidative damage [63]. Further, the protein Hmp, a flavohemoglobin with nitric oxide dioxygenase activity [64] had a score of −0.254. This protein has been shown to protect respiratory cytochromes in *E*. *coli* [37], which is a possible mechanism of Ag toxicity [59,65].

### **4. Conclusions**

In this work, a chemical genetic screen of a mutant library was performed as a means of drawing insight into the mechanisms of Ag toxicity and resistance in bacteria. In total, 3810 mutant strains containing single deletions of non-essential genes in *E. coli* were screened, and subsequent hits were bioinformatically evaluated in order to highlight processes and pathways that are affected by Ag exposure. This systematic mutant screen involved a low but prolonged concentration of Ag exposure on solid minimal media to avoid indirect secondary and acute responses, while also attempting to directly target relevant genes. Here, resistant hits represented genes involved in enhancing the cytotoxicity of Ag, while in contrast sensitive hits represented genes functioning in tolerance to Ag including physiological responses that mitigate toxicity.

In short, processes involved with the cell exterior and the central dogma were found to be affected by Ag exposure to a greater extent than other processes analyzed. However, when further examining the fold enrichment, the cell membrane and transport were involved in Ag exposure to a lesser degree. In fact, proteins involved in amino acid biosynthesis (Ag sensitivity), phosphoproteins and metalloproteins (Ag resistance) were most densely represented as hits in this work—trends that were supported by the protein-protein interaction networks.

Our work supports many previously proposed mechanisms of Ag toxicity—disruption of iron-sulfur cluster containing proteins and certain cellular redox enzymes, and DNA damage, and Ag resistance—toxin export and sequestration. However, the data presented here also demonstrates that the activity of Ag within the bacterial cell is more extensive than previously suggested, involving genes a part of the cell wall structure, quinone metabolism, ATP synthesis and sulfur reduction.

The use of Ag as an antimicrobial is a practice garnering considerable popularity, as the introduction of Ag-based compounds, such as combination treatments, nanomaterials, and formulations make way. In order to continue the development of this metal as a therapeutic agent, it is imperative that we gather more understanding into the accompanying mechanisms of Ag toxicity and resistance. This study provides a vast number of biomolecular mechanistic hypotheses to the community investigating the mechanisms of action of Ag and other metals.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2073-4425/9/7/344/s1, Figure S1: Resistant and sensitive hits involved in regulation; Figure S2: Connectivity map displaying silver resistant hits; Figure S3: Connectivity map displaying silver sensitive hits; Figure S4: Silver resistant metabolic overview; Figure S5: Silver sensitive metabolic overview; Table S1: Normalized and scored gene list.

**Author Contributions:** Conceptualization, J.A.L., K.C.-R. and R.J.T.; Data curation, N.G., K.C.-R. and Y.Y.; Formal analysis, N.G., K.C.-R. and Y.Y.; Funding acquisition, J.A.L., G.C. and R.J.T.; Investigation, J.A.L. and K.C.-R.; Methodology, J.A.L. and K.C.-R.; Resources, G.C. and R.J.T.; Supervision, G.C.; Visualization, J.A.L. and K.C.-R.; Writing—original draft, N.G.; Writing—review & editing, N.G., J.A.L., K.C.-R., G.C. and R.J.T.

**Funding:** Natalie Gugala was funded by an Eyes High Doctoral Scholarship from the University of Calgary. Joe A. Lemire. was funded by the Banting Postdoctoral Fellowship Program, Alberta Innovates Health Solutions Postdoctoral Fellowship, and a MITACS Accelerate Fellowship. Raymond J. Turner was funded by a project bridge grant from the Canadian Institutes of Health Research (CIHR bridge: PJT-149009). Raymond J. Turner. and Gordon Chua were also funded by Discovery grants from the Natural Science and Engineering Research Council (NSERC DG: RGPIN/04811-2015) of Canada.

**Acknowledgments:** The authors of this paper would like to acknowledge Iain George and Connor Westersund for their assistance with the BM3 robot from S&P Robotics Inc. and experimental set-up, respectively.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **Appendix**

**Table A1.** Systems and comprising subsystems cited in this study. The Ag resistant and sensitive hits were surveyed against the EcoCyc database permitting the clustering of the hits into systems, subsystems, component subsystems, and lastly into individual objects 1,2.



**Table A1.** *Cont.*

<sup>1</sup> Each individual score represents the mean of 12 trials—three biological and four technical; <sup>2</sup> Two-tailed *t*-test and significance was determined using the Benjamini-Hochberg procedure.

### **References**


© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Article*

## **Using a Chemical Genetic Screen to Enhance Our Understanding of the Antimicrobial Properties of Gallium against** *Escherichia coli*

### **Natalie Gugala 1, Kate Chatfield-Reed 2, Raymond J. Turner <sup>1</sup> and Gordon Chua 1,\***


Received: 7 November 2018; Accepted: 4 January 2019; Published: 9 January 2019

**Abstract:** The diagnostic and therapeutic agent gallium offers multiple clinical and commercial uses including the treatment of cancer and the localization of tumors, among others. Further, this metal has been proven to be an effective antimicrobial agent against a number of microbes. Despite the latter, the fundamental mechanisms of gallium action have yet to be fully identified and understood. To further the development of this antimicrobial, it is imperative that we understand the mechanisms by which gallium interacts with cells. As a result, we screened the *Escherichia coli* Keio mutant collection as a means of identifying the genes that are implicated in prolonged gallium toxicity or resistance and mapped their biological processes to their respective cellular system. We discovered that the deletion of genes functioning in response to oxidative stress, DNA or iron–sulfur cluster repair, and nucleotide biosynthesis were sensitive to gallium, while Ga resistance comprised of genes involved in iron/siderophore import, amino acid biosynthesis and cell envelope maintenance. Altogether, our explanations of these findings offer further insight into the mechanisms of gallium toxicity and resistance in *E. coli*.

**Keywords:** *Escherichia coli*; gallium; antimicrobial agents; metal toxicity; metal resistance; metal-based antimicrobials

### **1. Introduction**

The therapeutic capabilities of gallium(III) (Ga) have been and continue to be exploited for a number of clinical applications, which include: the treatment of cancer, autoimmune and infectious diseases, for the localization of tumors, inflammation and infection sites, and the reduction of accelerated bone resorption [1,2]. At the nuclear level, certain characteristics of this abiogenic metal permit essential metal mimicry, owing its similarities to Fe. In particular, the pharmacological characteristics of Ga are likely a result of its Fe(III)-like coordination chemistry and its ability to form stable six-coordinated complexes through ionic bonding [3]. This metal is trivalent and a hard acid in solution, according to the hard-soft acid-base theory [4], binding well with strong Lewis bases. As a result, Ga tends to form bonds with oxygen predominantly forming Ga(OH)4<sup>−</sup> (gallate) at pH 7.4 [5].

Despite Ga's similarities to the essential metal Fe, these metals share two main differences: (i) Ga cannot be reduced under biologically relevant reduction potentials, whereas Fe can be readily changed to and from a reduced state; and (ii) the concentration of unbound Fe(III) in solution is extremely low, localized primarily as a neutral complex with organic compounds, whereas gallate, which is anionic, can exist at significant concentrations [6].

As an Fe(III) mimetic, Ga(III) can incorporate itself into proteins and enzymes replacing Fe and effectively halting several essential metabolic processes [7–14]. Since the bioavailability of Fe is scarce, organisms, such as bacteria, have produced a variety of biomolecular chelating scavenging systems including siderophores and Fe-chelating proteins. Cells rapidly multiplying are more susceptible to Ga toxicity due to their high Fe demands [0]. As a result, this metal is approved by the US Food and Drug Administration for the treatment of cancer-associated hypercalcemia (Ganite®, Genta, NJ, USA) and has been tested as an antimicrobial agent against a variety of organisms including *Mycobacterium tuberculosis* [15,16], *Pseudomonas aeruginosa* [9,10,17], *Staphylococcus aureus* [18], *Rhodococcus equi* [19], *Acinetobacter baumannii* [20], and *Escherichia coli* [21].

In general, proposed mechanisms of toxicity for metal-based antimicrobials include the production and propagation of reactive oxygen species (ROS), the disruption of Fe-sulfur centers, thiol coordination, the exchange of a catalytic or structural metal, which in turn may lead to protein dysfunction, obstructed nutrient uptake, and genotoxicity [22]. The route by which Ga enters the cells is unknown, although, it is predominantly assumed that this metal crosses the cytoplasmic membrane by exploiting Fe-uptake routes, such as siderophores [23]. Several studies have explored the use of Fe-chelators as "Trojan horses" as a means of improving the delivery and toxicity of this metal in bacterial cells [14]. Still, there is insufficient research demonstrating that complexes of Ga and Fe-chelators/siderophores, such as Ga-citrate, increase the antibacterial abilities of this metal mainly since the import of this metal is not suggested to be the limiting step [23]. Furthermore, Ga exposure has been demonstrated to trigger the production of ROS in vitro [7,8]. Upon the cytoplasmic replacement of Fe with Ga, the available Fe pool is thought to increase, in turn fostering Fenton chemistry [22].

Bacteria have developed mechanisms of resistance as a means of withstanding metal toxicity. Some mechanisms include extracellular and intracellular sequestration, efflux, reduced uptake, repair, metabolic by-pass, and chemical modification [24]. Microbial resistant mechanisms associated with Ga have been studied to a far lesser degree, nonetheless, studies have shown that Ga is not as effective as postulated. For example, Ga resistance in *P. aeruginosa* and *Burkholderia cepacia* has been identified, suggested to be the result of decreased Ga import and the formation of bacterial biofilms [25,26].

Currently, research in this field is directed toward discovering novel utilities for this metal, still, the expansion of Ga as a therapeutic antimicrobial has been delayed compared to other metal-based antimicrobials, such as silver and copper. In short, it is essential that the mechanisms of Ga action in microbes are explored to greater degree in order to further the development of this antimicrobial agent.

In this work, we hypothesized that Ga exerts toxicity on multiple targets. Furthermore, we believe that there are several mechanisms of resistance that are fundamental to an organism's adaptive response under sub-lethal concentrations of Ga. To evaluate this, we performed a genotypic screening workflow of an *E. coli* mutant library composed of 3985 strains. Each strain contains a different inactivated non-essential gene. Genome-wide toxin/stressor-challenge workflows have been used to study silver [27–30], copper [31,32], cadmium [33], cobalt [33], and zinc [34]; however, no such study has been implemented to examine the effects of Ga. Therefore, as a means of complementing existing work, we have identified a number of genes that may be involved in Ga toxicity or resistance and mapped their biological processes to their respective cellular system in *E. coli*.

### **2. Materials and Methods**

All methods are as described previously by Gugala et al. [30] and all chemicals were obtained from VWR International, Mississauga, Canada, unless otherwise stated.

### *2.1. Escherichia coli Strains*

The Keio collection [35] consisting of 3985 single gene *Escherichia coli* BW25113 mutants (*lacI*<sup>q</sup> *rrnB*T14 Δ*lacZ*WJ19 *hsdR*514 Δ*ara*BADAH33 Δ*rha*BADLD78), was obtained from the National BioResource Project *E. coli* (National Institute of Genetics, Shizuoka, Japan).

### *2.2. Determination of the Minimal Inhibitory Concentration and Controls*

The sublethal inhibitory concentration, a concentration below the minimal inhibitory concentration that is found to visibly challenge selected mutants under prolonged metal exposure, was determined using Δ*recA*, Δ*lacA* and Δ*lacY* strains from the Keio collection. The protein RecA is involved in a number of processes, including homologues recombination and the induction of the SOS response in reaction to DNA damage [36]. Evidence may suggest that Ga causes the formation of ROS, although the precise mechanism of production is unknown. As a result, the absence of this gene was anticipated to confer the Ga sensitive phenotype, implied by a decrease in colony formation, since it is thought to be involved in mitigating ROS stress. Further, the protein products of *lacA* and *lacY* were not anticipated to be involved in Ga resistance or toxicity, therefore mutant strains of these genes were used as negative controls. Strains Δ*recA*, Δ*lacA,* and Δ*lacY*, and the wild-type (WT) were grown overnight at 37 ◦C on M9 minimal media plates (6.8 g/L Na2HPO4, 3.0 g/L KH2PO4, 1.0 g/L NH4Cl, 0.5 g/L NaCl, 4.0 mg/L glucose, 0.5 mg/L MgSO4 and 0.1 mg/L CaCl2) containing Noble agar (1.0%) in the presence and absence of Ga at varying concentrations. The concentration of Ga that visibly decreased colony formation in the *recA* mutant and produced no growth changes in the negative control strains was selected as the sublethal inhibitory concentration. Furthermore, Δ*recA*, Δ*lacA*, and Δ*lacY* and the WT strain were grown overnight in the presence of ionic nitrate at the equivalent molarity as the sublethal inhibitory concentration to ensure growth was not influenced by the accompanying counter ion. In order to identify Ga-sensitive and -resistant genes in this study, the Keio collection was exposed to 100 μM Ga(NO3)3 (Ga). Gallium nitrate was obtained from Sigma–Aldrich, St. Louis, MO, USA. Stock solutions of Ga were prepared with deionized H2O and stored in glass vials for no longer than two weeks.

Similarly, Δ*recA*, Δ*lacA*, and Δ*lacY* and the WT strain were grown on M9 minimal media plates in the presence of varying concentrations of hydroxyurea (HU), obtained from USBiological Salmen, MA, USA, or sulfometuron methyl (SMM) obtained from Chem Service, West Chester, PA, USA, dissolved in ddH2O and dimethyl sulfoxide, respectively. Select mutants from the Keio collection were exposed to a final concentration of 5.0 mg/mL HU and 5.0 μg/mL SMM in the presence and absence of 100 μM Ga(NO3)3.

### *2.3. Screening*

M9 minimal media and Noble agar (1.0%) plates, with and without the addition of Ga, were prepared two days prior to use. Here, Ga was added directly to the liquid agar and swirled before solidification. Colony arrays in 96-format were produced and processed using a BM3 robot and spImager (S&P Robotics Inc., Toronto, ON, Canada), respectively. Cells were transferred from the arrayed microtiter plates using a 96-pin replicator onto Luria-Bertani (LB) media agar plates and grown overnight at 37 ◦C. Colonies were then transferred using the replicator onto two sets of M9 minimal media Noble agar plates, with and without 100 μM Ga(NO3)3. Plates were then grown overnight at 37 ◦C. All images were acquired using the spImager and colony size, a measure of Ga sensitivity or resistance, was determined using integrated image processing software. Three biological trials were conducted and each of these trials included four technical replicates originating from the 96-colony array, which were combined and expanded onto a single plate in 384-colony array format; *n* (trials) ≥ 9. Strains presenting less than nine replicates were excluded (see Section 2.5).

Select mutants were exposed to HU or SMM at sublethal inhibitory concentrations. Identical conditions were maintained to enable direct comparisons between mutants grown in the presence of Ga only, and those grown in the presence of Ga and either HU or SMM. Here HU or SMM were added to the M9 minimal media plates directly before solidification.

### *2.4. Normalization*

In this study, incubation time and temperature, nutrient availability, colony location, agar plate imperfections, batch effects, and neighboring mutant fitness were considered independent variables that could influence colony size and subsequently cause systematic variation. As a result, the colonies were normalized and scored using Synthetic Genetic Array Tools 1.0 (SGATools) [37,38], a tool that associates mutant colony size with fitness, thereby enabling quantitative comparisons. All the plates were normalized to establish average colony size, working on the assumption that the majority of the colonies would exhibit WT fitness since the concentration of Ga used in this study was below the minimal inhibitory concentration.

Mutant colony sizes in the presence (challenge) and absence (control) of Ga were quantified, scored, and compared as deviation from the expected fitness of the WT strain. This assumes a multiplicative model and not an additive effect originating from the challenge. Once scored, mutants displaying a reduction in colony size were indicative of a Ga sensitive hit and those displaying an increase in colony size were recovered as Ga resistant hits. Finally, the *p*-value was calculated as a two-tailed *t*-test and significance was determined using the Benjamini–Hochberg procedure, as a method of lowering the false discovery rate, which was selected to be 10%.

### *2.5. Data Mining and Analyses*

Data mining was performed using Pathway Tools Omics Dashboard, which surveys against the EcoCyc database [39] and Uniport [40]. This allowed for the clustering of the Ga resistant and sensitive data sets into systems, subsystems, and individual objects (Table A1). Here, genes can be found in multiple systems since many are involved in a number of cellular processes.

Enrichment analyses were performed using the DAVID Bioinformatics Resource 6.8 [41,42]. Moreover, as a means of revealing the direct (physical) and indirect (functional) protein interactions amongst the gene hits, the STRING database [43] was utilized. Node maps based on experimental, co-expression, and gene fusion studies were generated using the Ga resistant and sensitive hits found in our screen.

### **3. Results and Discussion**

### *3.1. Genome-Wide Screen of Ga Resistant and Sensitive Hits*

In this work, the chemical genetic screen provided a method for the identification of the non-essential genes that may be involved in Ga resistance or sensitivity. A total of 3985 non-essential genes were screened for growth in the presence of 100 μM Ga(NO3)3 and from here, 3641 hits, in which *n* ≥ 9, were used for subsequent statistical analyses (Figure 1 and Supplementary Table S1). The statistical cutoff that suggested a significant difference in fitness when compared to the WT, indicated by a change in colony size, was selected to be two standard deviations from the mean or a normalized score of +0.162 and −0.154. This resulted in 107 gene hits, which represents approximately 2.5% of the open reading frames in the *E. coli* K-12 genome. In general, the normalization was performed with the assumption that hits presenting scores within two standard deviations from the mean had non-specific or neutral interactions with Ga. Therefore, the remaining hits were not regarded as significant based exclusively on the cutoffs selected.

**Figure 1.** Synthetic Array Tools (version 1.0) was used to normalize and score the Gallium(III) (Ga) resistant and sensitive hits as a means of representing the growth differences in *Escherichia coli* K12 BW25113 in the presence of 100 μM Ga(NO3)3. Each individual score represents the mean of 9–12 trials.

In this work, the absence of the gene was inferred to give rise to the Ga resistant or -sensitive phenotype. A decrease in colony size (normalized score < −0.154) signified a Ga-sensitive hit, which implied that the presence of this gene increased Ga resistance. Here, 58 genes were found to cause Ga sensitivity when absent (Table 1). Likewise, an increase in colony size (normalized score > 0.162) signified a Ga-resistant hit, therefore the presence of this gene may suggest an increase in toxicity. Comparably, 49 genes were found to impart resistance when absent (Table 2), within the cutoffs applied.

**Table 1.** Ga sensitive hits organized according to system and subsystem mined using the Omics Dashboard (Pathway Tools), which surveys against the EcoCyc Database; genes represent sensitive hits with scores < −0.154.



**Table 1.** *Cont.*


**Table 1.** *Cont.*

<sup>1</sup> Gene hits can be mapped to more than one system and subsystem. <sup>2</sup> Each individual score represents the mean of 9–12 trials. <sup>3</sup> Two-tailed *t*-test and significance was determined using the Benjamini–Hochberg procedure; false discovery rate 10%.


**Table 2.** Ga-resistant hits organized according to system and subsystem mined using the Omics Dashboard (Pathway Tools), which surveys against the EcoCyc Database; genes represent resistant hits with scores >0.162.


**Table 2.** *Cont.*

<sup>1</sup> Gene hits can be mapped to more than one system and subsystem. <sup>2</sup> Each individual score represents the mean of 9–12 trials. <sup>3</sup> Two-tailed *t*-test and significance was determined using the Benjamini–Hochberg procedure; false discovery rate 10%.

Using Pathway Tools, which surveys against the EcoCyc database, a number of gene hits were mapped to more than one system and subsystem (Tables 1 and 2). In general, comparable number of hits were mapped to the system "Response to stimulus", "Cellular processes", "Energy", and "Biosynthesis" (Figure 2). Still, "Regulation", "Degradation", and proteins of the "Cell exterior" contained more resistant hits. Whereas "Other pathways" and proteins involved in processes of the "Central dogma" were represented by the Ga-sensitive hits at least two-fold more than the Ga resistant hits (Figure 2). Proteins residing or involved in maintaining cell envelope homeostasis were not enriched in the resistant hits; however, two-fold more hits were mapped to the system "Cell exterior" using EcoCyc's system of classification when compared to the sensitive hits (Figure 2).

**Figure 2.** Ga-resistant and -sensitive gene hits mapped to component cellular processes. Several gene hits are mapped to more than one subsystem. The cutoff fitness score selected was two standard deviations from the mean and recovered gene hits with a score outside this range were chosen for further analyses. The hits were mined using the Omics Dashboard (Pathway Tools), which surveys against the EcoCyc database. Each individual score represents the mean of 9–12 trials.

Despite similar numbers of resistant and sensitive hits scored in this screen, a greater number of categories were enriched for by the resistant hits, such as the biosynthesis of the vital coenzyme biotin—when surveyed using the DAVID gene functional classification (Figure 3).

**Figure 3.** Functional enrichment among the Ga-resistant and -sensitive gene hits. The DAVID gene functional classification (version 6.8) database, a false discovery rate of 10% and a cutoff score two standard deviations from the mean was used to measure the magnitude of enrichment of the selected gene hits against the genome of *E. coli* K-12. Only processes with gene hits ≥3 were included.

In addition, a number of amino acid biosynthetic processes and cytosolic proteins were enriched in the resistant hits, whereas proteins involved in the processing of 20S pre-rRNA and malate metabolic processes were enriched in the sensitive hits (Figure 3). In general, the enrichment profile of the resistant and sensitive hits provides insight into the dissimilarities between the mechanisms of Ga toxicity and resistance since there was no overlap in enrichment (Figure 3). Based on previous reports [44,45] several mutants belonging to the Keio collection, such as those involved in the synthesis of amino acids, did not grow in M9 minimal media, contrary to what we observed in this study. We attribute this observation to the presence of residual resources, such as amino acids, that were carried over from the LB media agar plates onto to the M9 minimal media agar plates. Once these resources are exhausted, dying cells may provide a source of nutrients for surviving cells. Furthermore, previous studies have provided cutoff values as markers of growth, such as one-third the average OD [45]. Mutants displaying growth below the cutoff are regarded as non-growers despite possible survival. As a follow up, we grew a number of mutants overnight, including *leuC*, *metA*, *proA, ilvB*, *trpD*, *lacA,* and the WT strain in liquid M9 minimal media from existing culture stocks (see Section 2.1) and transferred 20 μL onto M9 minimal agar plates in the absence and presence of Ga. These strains were then grown overnight. Growth was only observed for the WT strain and the *lacA* mutant. When the same procedure was completed with liquid LB medium and agar plates, colony formation was evident for each mutant tested. As a result, in this study we were able to test mutants that have otherwise been reported to not grow on minimal media due to the lack of essential nutrients, such as amino acids.

### *3.2. Ga Sensitive Systems*

### 3.2.1. Iron Homeostasis and Transport, and Fe–Sulfur Cluster Proteins

Gallium(III) has been shown to disrupt the function of several enzymes containing Fe–sulfur clusters, likely by competing for Fe-binding sites [7]. *Escherichia coli* contains over 10 Fe-acquisition systems, encoded by over 35 genes [46], providing an abundance of Ga potential targets, such as the sensitive hit *fdx* (ferredoxin). The protein product of *fdx* serves as an electron transfer protein in a wide variety of metabolic reactions, including the assembly of Fe–sulfur clusters [47], consequently, Ga resistance is probable if this metal is damaging Fe–sulfur centers. Ferredoxin may also serve as

a binding site since the exchange of Fe may cause Ga sequestration. Furthermore, the sensitive hit *lipA* (lipoyl synthase) codes for an enzyme that uses ferredoxin as a reducing source, and catalytic Fe–sulfur clusters to produce lipoate [48]. This protein's requirement for ferredoxin may provide an explanation for the two-fold score decrease observed in the *lipA* mutant when compared to the *fdx* mutant. Furthermore, our screen recovered the hit *ygfZ*, which codes for a folate-binding protein that is implicated in protein assembly and the repair of Fe–sulfur clusters [49]. The loss of *ygfZ* results in sensitivity to oxidative stress likely due to the generation of ROS, subsequently, this may lead to the inhibition of Fe–sulfur cluster assembly or repair [50]. In addition, the disruption of Fe–sulfur clusters has been found to downregulate the uridine thiolation of particular tRNAs as a means of decreasing sulfur consumption [51]. This process appears to be important in coupling translation with levels of sulfur-containing amino acids. We recovered *trmU*, which encodes a tRNA thiouridylase as a Ga-sensitive hit in this study.

The redox pair Fe(II)/Fe(III) is well suited for a number of redox reactions and electron transfers. Accordingly, bacteria have developed a number of Fe-acquisition systems, such as siderophores and Fe-chelating proteins [52]. Siderophores, such as enterobactin are synthesized internally and exported extracellularly to scavenge Fe(III) from the environment [53]. The ferric-siderophore complex is imported into the cell and then degraded to release Fe(III) [53] and since Ga is an Fe mimetic [54], this metal has been demonstrated to bind certain siderophores [23]. The protein TolC is an outer membrane carrier required for the export of the high-affinity siderophore enterobactin from the periplasm to the external environment [55]. The Ga sensitivity of the Δ*tolC* strain may be due to the periplasmic accumulation of Ga-enterobactin complexes. If TolC is inactivated, then less enterobactin is exported outside the cell in turn providing more Ga targets, and as a result, Ga-enterobactin complexes may accumulate inside the cell. Further, EvgA is part of the EvgAS two-component system involved in the transcriptional regulation of *tolC* [56]. Loss of *evgA* is expected to display a similar defect in enterobactin export as would a *tolC* mutant, thus resulting in Ga sensitivity. Finally, *bfr* (bacterioferritin) was recovered as a sensitive hit in this work. This protein, which binds one heme group per dimer and two Fe atoms per subunit, functions in Fe storage and oxidation [57]. The sensitivity phenotype of the Δ*bfr* strain may be associated with a failure to mitigate Fe-mediated ROS production due to the disruption of Fe homeostasis in the presence of Ga (see Section 3.2.2).

### 3.2.2. Oxidative Stress

The production of ROS has been shown to be a mechanism of metal toxicity. Exposure to hydrogen peroxide or other agents that catalyze the production of ROS, such as superoxide, causes DNA and protein damage to macromolecules including proteins, lipids, nucleic acids and carbohydrates [58]. This in turn causes the upregulation of genes encoding ROS-scavenging enzymes [58]. An increase in cytoplasmic Fe intensifies ROS toxicity by catalyzing the exchange of electrons from donor to hydrogen peroxide [22]. Consequently, this may require the assistance of cellular antioxidants such as glutathione, and enzymes such as catalase, superoxide dismutase and peroxidase [59]. Gallium(III) is Fenton inactive, and therefore the induction of ROS in the presence of Ga is likely to result in the release of Fe in the cytoplasm. One study observed higher levels of oxidized lipids and proteins in Ga exposed *Pseudomonas fluorescens* [7]. In turn, the oxidative environment stimulated the synthesis of nicotinamide adenine dinucleotide phosphate (NADPH) via the overexpression of NADPH-producing enzymes, invoking a reductive environment.

In this screen, several sensitive Ga hits effective in ROS protection were recovered, including γ-Glutamate-cysteine ligase, or *gshA*. Strains lacking this gene have been shown to be hypersensitive to thiol-specific damage generated through mercury and arsenite exposure [60]. Similarly, strains lacking glyoxalase II (*gloB*), also a sensitive hit in this study, accumulate *S*-lactoylglutathione and demonstrate depleted glutathione pools [61]. If this antioxidant is depleted, then the potential for ROS-mediated protection is lowered. Furthermore, the gene *grxD*, which codes for a scaffold protein that transfers intact Fe–sulfur clusters to ferredoxin, was also recovered as a Ga-sensitive hit. The presence of this abundant protein is further upregulated during stationary phase [62] and one study demonstrated, using the Keio collection, that a *grxD* mutant is sensitive to Fe depletion [63]. Based on this observation, Ga exposure may prompt toxicity via Fe exhaustion, or the introduction of this toxin may result in ROS production thereby leading to Fe–sulfur damage. Finally, bacterioferritin (*bfr*) was also identified as a sensitive hit in this work. This protein acts to prevent the formation of hydrogen peroxide from the oxidation of Fe(II) atoms [57]. The sensitivity phenotypes of the Δ*gshA*, Δ*gloB*, Δ*grxD,* and Δ*bfr* strains may be associated with Fe-mediated ROS production upon the disruption of Fe-homeostasis in Ga exposed cells.

The sensitive hit *ubiG,* involved in the production of ubiquinol-*8*, a key electron carrier used in the presence of oxygen or nitrogen, was recovered in this screen. The production of ubiquinol from 4-hydroxybenzoate and trans-octaprenyl diphosphate necessitates the use of six enzymes and UbiG twice [64]. Mutant strains deficient in ubiquinol demonstrate higher levels of ROS in the cytoplasmic membranes, a threat lessened via the addition of exogenous ubiquinol [65]. Furthermore, the Δ*ubiG* strain exhibited reduced fitness when exposed to oxidative stress [65]. Altogether, the presence of this hit may be explained by the exacerbation of the production of ROS due to Ga exposure alongside the compromised oxidative stress response of the Δ*ubiG* strain.

### 3.2.3. Deoxynucleotide and Cofactor Biosynthesis, and DNA Replication and Repair

Compounds targeting ribonucleotide reductase (RNR), a key enzyme involved in the synthesis of deoxynucleotides from ribonucleotides, have long been regarded as cancer therapeutics [66]. In mammalian cells, Ga targets RNR through at least two mechanisms. These mechanisms include the inhibition of cellular Fe uptake resulting in decreased Fe availability at the M2 subunit of the enzyme [67] and direct inhibition of RNR activity [68], leading to a reduction in the concentration of nucleotides in the cell. This mechanism is not limited to mammalian cells. Gallium(III) has been shown to inhibit RNR and aconitase activity in *M. tuberculosis* [16]. If RNR inhibition is in fact a mechanism of Ga toxicity, then we predict that gene deletions resulting in decreased deoxynucleotide levels may cause hypersensitivity. Consequently, the deletion of the gene *purT*, which is involved in purine nucleotide biosynthesis [69], resulted in Ga sensitivity in this study.

Chromosomal replication is delayed in *E. coli* cells when the deoxynucleotide pool is depleted upon the inhibition of RNR [70]. If this is the case, then a defect in DNA replication may result in hypersensitivity to Ga. Our observation that the loss of the DNA polymerase III subunits HolC and HolD causes Ga sensitivity appears to support this hypothesis. Another potential consequence of RNR inhibition is an increase in stalled replication forks, which are prone to DNA strand breakage [70]. Resumption of stalled replication forks and double strand breaks due to defective RNR function require the activity of recombination repair enzymes such as the RuvABC, RecBCD and RecA [71,72]. Our results support these observations since the deletion of *recA*, *recD* or *ruvC* triggered the Ga sensitive phenotype. It is important to note that genes involved in base and nucleotide excision repair were not retrieved as Ga sensitive hits suggesting that DNA damage associated with Ga exposure may be predominantly in the form double stranded breaks.

A number of sensitive hits were mapped to the subsystem "Biosynthesis of cofactors, prosthetic groups and electron carriers". Processes affected include folate, lipoate, quinol, quinone, ubiquinol and thiamine biosynthesis. The gene products of *pabA* and *pabC*, which encode an aminodeoxychorismate synthase and an aminodeoxychorismate lyase, respectively, are involved in the biosynthesis of p-aminobenzoic acid [73], a precursor of folate. In both prokaryotes and eukaryotes, folate cofactors are necessary for a range of biosynthetic processes including purine and methionine biosynthesis (Figure 4) [74]. Folate biosynthesis has long served as an antibiotic target in prokaryotes since this cofactor is synthesized only in bacteria yet actively imported by eukaryotes using membrane associated processes [75]. Similar to *purT*, the Ga sensitivity of Δ*pabA* and Δ*pabC* strains may be a result of the reduction in deoxynucleotide levels caused by the inactivation of RNR.

**Figure 4.** Connectivity map displaying the predicted functional associations between the Ga-sensitive gene hits; disconnected gene hits not shown. The thicknesses of the lines indicate the degree of confidence prediction for the given interaction, based on fusion, curated databases, experimental and co-expression evidence. Figure generated using STRING (version 10.5) and a medium confidence score of 0.4.

To test the potential connection between Ga and RNR activity, we exposed the *holC*, *holD*, *recA*, *recD*, *ruvC* and *purT* mutants to hydroxyurea (HU), which is a known inhibitor of RNR activity [76]. Further, we included a number of mutants involved in DNA synthesis, such as *ruvA* and *recR*, that were not uncovered in our initial screen. In *E. coli*, HU has been shown to increase ribonucleotide pools and decrease total deoxyribonucleotide concentrations, thus negatively affecting the synthesis of DNA [77]. We exposed these mutants to sublethal concentrations of HU and normalized the cellular effect of this agent. Using this reagent, the sensitivity of the *holC*, *ruvC* and *recD* mutants in the presence of HU and Ga was found to increase (Table 3). Furthermore, *ruvA*, which assists in recombinational repair together with *ruvB* [78], was also found to be a sensitive hit in the presence of this inhibitor. The genes *purT* and *holD* were not uncovered as either sensitive or resistant hits based on the cutoffs applied and no changes in the sensitivity or resistance of either *lacA* or *lacY*, negative controls in this work, were statistically identified.


**Table 3.** Hydroxyurea sensitive and gene hits involved in the synthesis of DNA, normalized to include only the effects of Ga exposure; those with a score two deviations from the mean are included.

<sup>1</sup> Each individual score represents the mean of 9–12 trials. <sup>2</sup> Two-tailed *t*-test and significance was determined using the Benjamini–Hochberg; procedure; false discovery rate 10%. HA: Hydroxyurea

### *3.3. Systems Involved in Ga Resistance*

### 3.3.1. Fe Transport Systems

In *E. coli*, the mechanisms by which Ga is transported into the cell have yet to be identified. In this screen, we identified a number of transport proteins that confer resistance against Ga when absent. Metal resistance mechanisms may involve decreased import or enhanced export of the toxin. Therefore, loss of a gene in which the product mediates import of the toxin into the cell would prevent its accumulation and result in resistance. Both FepG and TonB are proteins that demonstrate close interaction (Figure 5) and fit the latter criterion, both involved in the import of Fe-siderophores. The protein FepG is an inner membrane subunit of the ferric enterobactin ATP-binding cassette transporter complex. When *fepG* is inactivated, *E. coli* cells lose ferric enterobactin uptake abilities [79,80]. The protein product of *tonB* is a component of the Ton system which functions to couple energy from the proton motive force with the active transport of Fe-siderophore complexes and Vitamin B12 across the outer membrane [81]. Since Ga entry into the bacterial cell can occur through siderophore binding and since this metal is an Fe mimetic [23,54], we hypothesize that in the absence of *fepG* and *tonB* Ga import and intracellular accumulation is reduced.

**Figure 5.** Connectivity map displaying the predicted functional associations between the Ga-resistant gene hits; disconnected gene hits not shown. The thicknesses of the lines indicate the degree of confidence prediction for the given interaction, based on fusion, curated database, experimental and co-expression evidence. Figure generated using STRING (version 10.5) and a medium confidence score of 0.4.

OmpC is a promiscuous porin that permits the transport of 30+ molecules, and is postulated to be a transporter of copper(I) and copper(II) [82] and potentially other metal species [83]. It has been hypothesized that Ga can cross the membrane of *E. coli* via porins [23]. While this hypothesis has not been demonstrated in *E. coli* directly, other works have confirmed findings in *P. aeruginosa* [9], *Mycobacterium smegmatis* [84] and *Francisella* strains [12]. Further evidence for the importance of OmpC in Ga resistance can be visualized using the STRING map (Figure 5). Here, OmpC is connected to two proteins that comprise the ATPase complex through the periplasmic protein TolB. TolB has been shown to physically interact with porins such as OmpC and is required for their assembly into the outer membrane of *E. coli* cells [85]. The resistance recovered in the Δ*tolB* strain may be due to a

disruption in OmpC function, thereby hindering Ga import. In addition, CysU, which is involved in the uptake of sulfate and thiosulfate was also recovered as a resistant hit [86]. According to the hard-soft acid-base theory, Ga coordinates well with sulfate or thiosulfate [4]. A reduction in the uptake of these metabolites may prove useful against Ga stress due to decreased toxin import.

Genes involved in Fe import in other organisms have been shown to confer Ga resistance when deleted, or Ga sensitivity when overexpressed. A three-fold increase in Ga resistance was displayed upon the deletion of the gene *hitA*, which codes for a Fe-binding protein in *P. aeruginosa* [25]. The *Haemophilus influenzae* proteins FbpABC, which are involved in the delivery of Fe from the periplasm to the cytoplasm, were expressed in *E. coli* as a means of investigating their impact on Ga import, which increased in the presence of these genes [87]. Furthermore, earlier studies have examined the use of metal-chelators as antimicrobial enhancements. Although the majority of studies regarding Ga import have been performed in *P. aeruginosa*, some findings can be compared. For example, it has been demonstrated that the siderophore complex Ga-deferoxamine was slightly more effective at killing cells than Ga alone [10] and more promising results have been made with the complex Ga-protoporphyrin IX [88]. Altogether, these studies and our work suggest that Ga enters the cell via siderophore transport systems or Fe-binding transporters.

### 3.3.2. Amino Acid Biosynthesis

Ga resistant hits were functionally enriched for the synthesis of amino acids (Figure 3), classified in the subsystem, "Amino acid biosynthesis" (Table 2) and highly connected in the functional map (Figure 5). The genes recovered were found to be mainly involved in the biosynthesis of branched (*ilvB*, *ilvY*, *leuA* and *leuC*) and aromatic (*aroF*, *trpB*, and *trpD*) amino acids, methionine (*metA* and *metR*), and proline (*proA* and *proB*). The demand for NADPH in biosynthetic pathways of branched and aromatic amino acids, as well as methionine and proline, are among the highest [89]. It is plausible that a defect in the synthesis of these amino acids may increase levels of NADPH, which has been shown to neutralize the oxidative stress elicited from Ga exposure [7].

To further test this hypothesis, we exposed a number of the resistant hits mapped to branched amino acid biosynthesis to sublethal concentrations of Sulfometuron methyl (SMM), an inhibitor of acetolactate synthase [90], a key enzyme involved in the synthesis of branched amino acids. The resistance score of *ilvY* and *leuA* increased in the presence of SMM (Table 4). Sulfometuron methyl inhibits acetolactate synthase, which in turn may increase the liable NADPH pool. In fact, *ilvY* is a positive regulator of *ilvC* [91], which encodes a reductoisomerase and is the only enzyme in this pathway that directly uses NADPH. Here, *ilvB* and other genes involved in branched amino acid biosynthesis did not make the statistical cutoffs owing to large standard deviations. Finally, no changes in the sensitivity or resistance of *lacA* or *lacY*, negative controls in this work, were statistically identified.

**Table 4.** Sulfometuron methyl resistant gene hits, involved in the synthesis of amino acids, normalized to include only the effects of Ga exposure; only those with a score two deviations from the mean are included.


<sup>1</sup> Each individual score represents the mean of 9–12 trials. <sup>2</sup> Two-tailed *t*-test and significance was determined using the Benjamini–Hochberg; procedure; false discovery rate 10%. SMM: sulfometuron methyl

It has been postulated that the oxidation of amino acids is a common and damaging effect of metal-induced oxidative stress [92]. Certain side chains, such as Arg, Cys, His, Lys and Pro residues are major targets, leading to protein damage and intra/inter-crosslinking [92,93]. If Ga targets amino acids, both free and within proteins, a possible explanation for the recovery of amino acid gene resistant hits in this study may rest in the cell's requirement to repair or replace damaged amino acids. If these

genes are absent fewer Ga targets remain and the cell expends less energy rebuilding these targeted biomolecules, while directing more energy elsewhere, such as scavenging and importing required metabolites. Furthermore, the oxidation of these amino acid side chains may lead to the propagation of ROS, and therefore a deficiency in amino acids may minimize damage by slowing the advancement of amino acid metal-induced oxidative stress.

### 3.3.3. Lipopolysaccharides and Peptidoglycan

The *E. coli* envelope is composed of lipopolysaccharides (LPS), which surround and protect the cytoplasm, and the cross-linked polymer peptidoglycan (PG), which is the primary stress-bearing biomolecule in the cell [94]. In this study, a number of genes involved in LPS or PG biosynthesis/maintenance were observed to cause Ga resistance when absent. These genes include *cpsG* and *rfaC* (LPS), and *alr*, *env* and *mrcB* (PG). Many of these genes are RpoS-regulated and participate in maintaining membrane integrity in response to pressure [95]. Loss of *mrcB*, which encodes for an inner membrane enzyme functioning in transglycosylation and transpeptidation of PG, has been shown to result in reduced surface PG density when absent [96]. The protein RfaC is essential in LPS production [97] and cells lacking this gene contain defects in the core heptose region [98]. The protein EnvC, which is a divisome-associated factor has been shown to have PG hydrolytic activity and result in decreased cell envelope integrity when deleted. Furthermore, the protein product of *tolB*, which plays a role in maintaining the structure of the cell envelope, was also a Ga-resistant hit. Cells deficient in *tolB* have been shown to release periplasmic proteins into the extracellular space [99]. An explanation for the appearance of *mrcB*, *envC* and *tolB* in this study may reside in the ability of PG to bind metals. Metal ions are known to bind the LPS or PG layer of Gram-negative and Gram-positive bacteria [100], and the presence of anionic groups such as carboxylic acids [101] and other hard acids within the cell envelope, provide suitable binding sights for free metal ions like Ga. Although the major ionic form of Ga is Ga(OH)4 −, free Ga ions produced through equilibrium may be quickly bound by hard acids such as alcohols, carboxylates, and hydroxyls, which comprise the bulk of the PG. Despite their presence at low concentrations these species may further impede cell health and cause toxicity. However, if the LPS or PG layer is reduced, as would be the case in the absence of *mrcB*, *rfaC*, *envC* and *tolB*, then a reduction in Ga-cell envelope binding may occur. In the case of the Δ*tolB* strain, the potential release of periplasmic proteins with Ga-binding sites into the extracellular space may also provide protection via sequestration, which is a common bacterial resistance mechanism [24]. Another possible explanation for Ga resistance associated with LPS and PG genes may include the structural alteration of the cell envelope, which may disrupt Fe import systems. Inhibition of lipid biosynthesis prevents proper assembly and insertion of porins into the outer membrane since LPS-porin interaction sites have been shown to be important in their biogenesis [102,103]. Therefore, compromised function of siderophore receptors or porins in these mutants could decrease Ga import and mitigate toxicity.

### **4. Conclusions**

In this study, the Keio collection was used as a means of drawing insight into the mechanisms of Ga toxicity and resistance in *E. coli* BW25113. In total, 3895 non-essential genes were screened and 3641 of these were normalized and scored. Genes demonstrating resistance or toxicity were mined to highlight processes and pathways affected by Ga exposure. Mutants demonstrating an increase in colony formation were considered resistant hits, in that the presence of the gene results in Ga sensitivity. In contrast, a decrease in colony size was regarded as a Ga-sensitive hit, consequently it was assumed that the presence of this gene would impart the resistant phenotype and mitigate the toxicity of prolonged Ga exposure.

Overall, comparable numbers of resistant and sensitive hits were mapped to each subsystem using Pathway Tools, which surveys against the EcoCyc Database. When examining the fold enrichment data, no biological process was enriched comparably between the two data sets. One general observation made evident from the latter conclusion is that distinct pathways are affected by Ga when comparing

the mechanisms of toxicity and resistance since no overlap in functional enrichment was uncovered. Still, one significant exception was found: Fe-metabolism. Based on this study, and previous reports, there is a relationship between Ga and Fe-metabolism. The genes that code for TonB and FepG were two resistant hits highlighted in this work. On the contrary, Fdx, Bfr and LipA, proteins also involved in Fe-metabolism, gave rise to sensitivity when absent. Therefore, we propose that Fe-metabolism may serve as a mechanism of resistance and toxicity in *E. coli*. Here, the complexity of Ga exposure is made further apparent, fostering more questions regarding the interaction of this metal with microbes. What is clear however, is that the mechanism of Ga action is likely a result of a number of direct and indirect interactions, an observation made evident by the wide array of hits uncovered in this work.

Few studies have explored the mechanisms of adaptive resistance in *E. coli* under sub-lethal concentrations of Ga. In response, we have presented a number of genes that are implicated to be involved in adaptive survival. For example, genes involved in preventing oxidative damage and DNA repair were emphasized as sensitive hits, as such that their presence gives rise to resistance. In short, preventing and repairing DNA damage, a mechanism that has yet to be demonstrated in vivo, and redox maintenance may provide tools by which microbial organisms mitigate metal stress.

The use of Ga for the treatment of diseases and infections is gaining considerable attention. Still, to further the development of this metal as an antimicrobial agent it is imperative that we determine the associated mechanisms of toxicity and resistance. Further work must be completed to specifically test the various hypotheses we have presented here, such as determining the mode of Ga entry, the levels of ROS produced in the cell and the specific influence of Ga on Fe-metabolism. Nonetheless, this study provides a significant number of biomolecular mechanistic hypotheses to the community investigating the mechanisms of Ga action in *E. coli* and other microbes.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2073-4425/10/1/34/s1, Table S1: *Escherichia coli* gallium resistant and sensitive gene hits determined using a chemical genetic screen.

**Author Contributions:** Conceptualization, K.C.-R. and R.T.; Data curation, N.G. and K.C.-R.; Formal analysis, N.G. and K.C.-R.; Funding acquisition, R.T. and G.C.; Investigation, N.G.; Methodology, N.G.; Resources, G.C.; Supervision, R.T. and G.C.; Visualization, N.G., K.C.-R., R.T. and G.C.; Writing—original draft, N.G.; Writing—review & editing, N.G., K.C.-R., R.T., and G.C.

**Funding:** N.G. was funded by an Alexander Graham Bell Scholarship from the Natural Science and Engineering Research Council of Canada and an Eyes High Doctoral Scholarship from the University of Calgary. R.J.T. was funded by a project bridge grant from the Canadian Institutes of Health Research (CIHR bridge: PJT-149009). R.J.T. and G.C. were also funded by Discovery grants from the Natural Science and Engineering Research Council (NSERC DG: RGPIN/04811-2015) of Canada.

**Acknowledgments:** The authors of this paper would like to acknowledge Joe A. Lemire and Iain George for their assistance with experimental design and the BM3 robot from S&P Robotics Inc., and Ying Yan for his assistance with the statistical analyses.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **Appendix A**

**Table A1.** The Gallium(III) (Ga) resistant and sensitive hits were surveyed against the EcoCyc database permitting the clustering of the hits into systems, subsystems, component subsystems, and lastly into individual objects.


<sup>1</sup> Genes can be found in multiple systems and subsystems.

### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Integrase-Controlled Excision of Metal-Resistance Genomic Islands in** *Acinetobacter baumannii*

### **Zaaima AL-Jabri 1,2,3, Roxana Zamudio 1, Eva Horvath-Papp 3, Joseph D. Ralph 1, Zakariya AL-Muharrami 2, Kumar Rajakumar <sup>3</sup> and Marco R. Oggioni 1,\***


Received: 5 June 2018; Accepted: 16 July 2018; Published: 20 July 2018

**Abstract:** Genomic islands (GIs) are discrete gene clusters encoding for a variety of functions including antibiotic and heavy metal resistance, some of which are tightly associated to lineages of the core genome phylogenetic tree. We have investigated the functions of two distinct integrase genes in the mobilization of two metal resistant GIs, G08 and G62, of *Acinetobacter baumannii*. Real-time PCR demonstrated integrase-dependent GI excision, utilizing isopropyl β-D-1-thiogalactopyranoside IPTG-inducible integrase genes in plasmid-based mini-GIs in *Escherichia coli*. In *A. baumannii*, integrase-dependent excision of the original chromosomal GIs could be observed after mitomycin C induction. In both *E. coli* plasmids and *A. baumannii* chromosome, the rate of excision and circularization was found to be dependent on the expression level of the integrases. Susceptibility testing in *A. baumannii* strain ATCC 17978, A424, and their respective ΔG62 and ΔG08 mutants confirmed the contribution of the GI-encoded efflux transporters to heavy metal decreased susceptibility. In summary, the data evidenced the functionality of two integrases in the excision and circularization of the two *Acinetobacter* heavy-metal resistance GIs, G08 and G62, in *E. coli*, as well as when chromosomally located in their natural host. These recombination events occur at different frequencies resulting in genome plasticity and may participate in the spread of resistance determinants in *A. baumannii*.

**Keywords:** copper resistance; genomic island; integrase; *Acinetobacter baumannii*; mobile genetic element

### **1. Introduction**

Genomic islands (GIs) are discrete gene clusters most of which are found as DNA segments within the chromosome. The GIs were originally known as pathogenicity islands (PAIs) by Hacker et al. in late 1980s, when they were examining the genetic virulence mechanisms in *Escherichia coli* [1]. Genomic islands are variable in size, ranging from 10 to 200 kb, and are usually detected during comparative genomic analysis of different closely related strains [2]. Genomic islands generally harbor genes coding for an integrase or recombinase, but could also carry insertion sequences or transposons within the element contributing to their movement [3,4]. Genomic islands are by definition part of the accessory gene pool of a species and allow for mobilization of whole pathways and gene clusters, thus acting as reservoirs for genetic diversity [5].

In the last few decades, *Acinetobacter baumannii* has been recognized as a multi-drug resistant opportunistic pathogen as a result of being burdened with massive use of broad-spectrum antibiotics [6]. The presence of *A. baumannii* with closely related Gram-negative bacteria in clinical settings has contributed to the development of new resistance mechanisms on top of their own intrinsic factors [7]. *Acinetobacter baumannii* and the other Gram-negative pathogens are known for their genome plasticity and capability of evolving mainly due to acquiring new virulence determinants carried on mobile genetic elements [8]. This issue has rendered antibiotic therapy ineffective in life-threatening *A. baumannii* infections, and in some cases even with the last line of combination therapy of high doses of antibiotics [9].

In *A. baumannii,* the first GI identified was a large GI of 86 kbp in size, during the sequencing of the epidemic strain AYE, and was named "AbaR1" harboring 45 resistance determinants [10]. Similar GIs were later recognized in *A. baumannii* and named from AbaR0 to AbaR27 [10–12]. Despite being similar in the backbone structure, AbaR-like GIs are often variable in terms of size and genetic composition. For example, the multiple-antibiotic resistance region (MARR) of the AbaR GI harbor a set of genes encoding antibiotic, heavy metal, and antiseptic resistance and efflux determinants [10]. This diversity in AbaR GIs compositions were a result of several events of recombination like integration, excision, and rearrangements [13].

Comparative genomic analysis showed extensive synteny throughout the genome and identified 63 DNA regions, ranging in size from 4–126 kb, all exhibiting certain features of GIs including a group of resistance GIs with different genes encoding resistance to antibiotics and heavy metals which are grouped in clusters [14]. For example, the *aadA1* (streptomycin-resistance encoding) gene, flanked by *satR* (streptothricin-resistance encoding) and *dhfr* (trimethoprim-encoding resistance) genes were found in GIs in clusters. Moreover, genes involved in mercury resistance (*merRCAD* cluster) were found to be located in a separate cluster, and a 4.5 kb DNA segment containing *feoAB* (ferrous iron transport operon), *czc* (tricomponent proton/cation antiporter efflux system), and *ars* (arsenite transporters) genes were co-existing as a group, next to the *cus* (copper resistance) genes conserved in the same chromosomal locations of certain GIs [14]. However, these genes differ in sequence and the overall arrangement from other homologous GIs in *A. baumannii*. This supports the notion that the set of accessory genes had been independently acquired by the different strains.

The two GIs of interest in this study, G08 and G62, harbor a set of putative heavy metal resistance conferring genes which are identical (Figure 1A) [14]. Only a few studies have described the G62 island [14–16], for example the presence of a similar resistance island has been shown in an *A. baumannii* hyper-virulent and outbreak-associated isolate, LAC-4 in China [15]. In LAC-4 clinical isolate, the G62 harbors the exact set of resistance genes and was referred to as a "copper resistance gene cluster"; however, in that strain, G62 was found to be sandwiched between two copies of IS*Aba26* element [15]. The ATCC17978 genome has been extensively analyzed in previous studies [17] and 13 putative zinc/copper resistance efflux pumps have been identified, including the efflux pumps present in G08 and G62 [16] (Figure 1A). The chromosomal region harboring zinc and/or copper efflux genes are likely to have been acquired laterally on mobile genetic elements, with the G62 of ATCC 17978 and LAC-4 being the largest of these elements [16]. Comparative analysis of putative zinc and/or copper efflux systems in *A. baumannii* and *A. baylyi* (strain ADP1) other than the ones in G08 and G62, identified a number of genes ranging between eight (strain SDF) and 18 (strain AB6870155) in each of the strains examined, all of which were chromosomally located [14,18]. Further BLAST search revealed that five strains harbored more than ten genes encoding putative zinc and/or copper efflux components, including ATCC 17978, ATCC 19606T, AB0057, AB6870155, and ACICU. On the other hand, the G08 island is found more frequently in *A. baumannii* including in strains AB0057, AB6870155, and AYE, where the element is inserted into the *dusA* locus encoding for the enzyme tRNA-dihydrouridinesynthase A, catalyzing the post-transcriptional reduction of uridine to dihydrouridine in tRNA [19].

The analysis of the distribution of genomic islands and other genes belonging to the accessory genome has shown in many species that many so-called mobile elements or accessory genes cluster tightly with specific lineages on a phylogenetic core genome tree [5,10,20]. This raises questions on this association which could be due to some positive selection or more likely due to loss of function of the mobile elements. In this work, we aim to test the hypothesis that the two metal-resistant related genomic islands G08 and G62 of *A. baumannii* are still functional.

**Figure 1.** Schematic map of the *Acinetobacter baumannii* strain genomic islands G08 and G62 and phylogenetic tree of genomic island (GI) integrases. (**A**) G62 island is taken from strain ATCC 17978 (accession CP000521; updated refseq NC\_009085.1) and G08 from both strains AB0057 (NC\_011586.2/CP001182.2) and AYE (NC\_010410.1/CU459141.1). The annotated sequences were aligned and visualized by the Easyfig tool [21]. The genes involved in encoding copper efflux systems are shown in blue, all *czc*-like genes *czcA*, *czcB*, and *czcC* encoding cadmium, zinc, and cobalt resistance are represented in green color, genes encoding ferrous iron transport proteins are shown in red (*feoA* and *feoB*), and a putative further heavy metal efflux system A1S\_2929 is shown in pink. The integration sites (*att* sites) for both GIs are shown as black vertical lines. The genes encoding the integrases are colored in orange and all other genes in grey. Locus tags of relevant genes are shown above or below the genes. The image is drawn in scale and the percentage of DNA identity between various regions is shown by gradient shading. (**B**) Phylogenetic tree of tyrosine recombinases from strains ATCC 17978, AB0057, and AYE. The evolutionary relationship between phage integrase family proteins (NCBI Reference Sequence) detected in three strains of *A. baumannii* was inferred using the Neighbor-Joining method [22]. Integrases were labelled according to the genomic island number, strain name, followed by Refseq accession numbers. In certain instances, some GIs have not been given a specific number, but instead were indicated by their accession. When two strains are mentioned next to a GI, this means that the same GI is present in both strains. Identical proteins in different genomes which yield identical Refseq numbers are shown only once. The three major branches are indicated by numbers on the right side of the figure. G08 and G62 are indicated by black boxes.

### **2. Materials and Methods**

### *2.1. Bacterial Strains and Cultivation*

A set of *A. baumannii* strains from different geographical origins were used. Hundred strains were collected from clinical samples at Sultan Qaboos University Hospital (SQUH, Oman) between 2012 to 2013. These strains were collected from various body sites of patients admitted in the internal medical wards in SQUH. The rest of the strains were from the collection at the Department of Infection, Immunity and Inflammation of the University of Leicester (Table 1). All strains were stored at −80 ◦C in 30% glycerol. Strains were re-streaked in LB agar (BD) or broth for liquid cultures.


**Table 1.** List of *A. baumannii* strains.

ST = Sequence type [24].

### *2.2. Genome Analysis*

The whole genome phylogenetic single nucleotide polymorphism (SNP) tree was build using the FFP (Feature frequency profile) version 3.19 suite of programs (http://sourceforge.net/projects/ ffp-phylogeny/) [25]. As input, the 101 complete *A. baumannii* genomes deposited in GenBank (8 July 2018) were used with the addition of our own two strains A424 (GCA\_003185755.1) and KR3831 (GCF\_003185745.1/GCA\_003185745.1). The matrix of integrase presence in the genomes was generated using as a query the integrase genes present in AB0057, AYE, and ATCC 17978 (shown also in Figure 1B). The matrix was generated using command line BLAST (90% identity; 95% coverage) and the R platform to generate the output.

### *2.3. Colony Genotpying*

The polymerase chain reaction was performed on crude cell extracts from serially diluted suspensions of *E. coli* and *A. baumannii* cells. Thirty μL aliquot of the dilution was boiled for 5 min, and 2 μL of supernatant was used as a polymerase chain reaction (PCR) template. PCR was conducted in a 20 μL reaction volume, containing 1 μL DNA (50 ng/μL), 5 μL 10× reaction buffer (Promega, Madison, WI, USA), 1 μL dNTPs (10 mM), 1 μL of each 10 mM primer, and 0.2 μL Go*Taq* DNA polymerase (Promega). Amplification was performed in a thermal cycler (Mastercycler gradient, Eppendorf, Hamburg, Germany) with an initial denaturation at 95 ◦C for 2 min, followed by 25 cycles of 95 ◦C for 1 min, 57 ◦C for 1 min, and 72 ◦C for 2 min, and a final extension at 72 ◦C for 10 min. The PCR products were imaged from a 1% TAE-agarose gel.

### *2.4. Real-Time PCR*

The real-time PCR reactions had a total volume of 20 μL containing 5 μL of template DNA, 10 μL of the SensiMixPlus SYBR Green mastermix (Bioline, London, UK), and 0.5 μL of each 15 μM primer (F-G08-exc and R-G08-circ) or (F-G62-exc and R-G62-circ). Since there was no positive control used, every run included a negative control without target DNA, and all reactions were performed in triplicate. The reactions were performed in an Applied Biosystems Prism (Foster City, CA, USA) model 7500HT Sequence Detection System with the following settings: 40 cycles of 20 s at 95 ◦C and 1 min at 60 ◦C. Determinations of cycle threshold (Ct), or the PCR cycle where fluorescence first occurred, were performed automatically by the Sequence Detection Systems software of the instrument (version 2.3; Applied Biosystems, Foster City, CA, USA).

### *2.5. RNA Extraction and Retrotranscription*

Following induction, 5 mL of cells were harvested at predetermined time points by centrifugation for 5 min at 4000× *g*, and resuspended in 1 mL RNALater (Invitrogen, Carlsbad, CA, USA), and stored at 4 ◦C. The total RNA was extracted using the Geneflow total RNA purification kit protocol (Norgen, Thorold, ON, Canada). The RNA was eluted from the column into a 1.5 mL microcentrifuge tube by addition of 30 μL RNase-free water and stored at −20 ◦C. Total RNA was quantified by spectrophotometry at A260 (Nanodrop 2000; Fisher ThermoScientific, Waltham, MA, USA), and cDNA was created by taking 1 μg of RNA per each reverse transcription reaction which was 20 μL, and the procedure was completed according to the High Capacity RNA-to-cDNA kit (Applied Biosystems).

### *2.6. Construction of Plasmids with G08 and G62 Mini-GIs*

To test the functionality of the integrase genes of G08 and G62 in excising their respective GIs, two plasmid constructs were created. Mini-islands were generated by creating smaller circular molecules with precise site-specific excision via the attachment sites *attL*/*attR* included within the left and right flanking regions, and integrase coding gene cloned in a plasmid under an inducible promoter. The new fragments generated were later cloned into pUC18 vector. The primer pairs F-LF-08/R-LF-08 and F-RF-08/R-RF-08 (Table S1) were used to amplify the left and right G08 flanking regions including the *att* sites from strain A424 (Genbank accession: GCA\_003185755.1). Similarly, the left- and right-flanking regions including the *att* sites of G62 GI were amplified from the strain ATCC 17978 using the primer pairs F-LF-62/R-LF-62 and F-RF-62 and R-RF-62 (Table S1), respectively. The two integrase genes G08*int* (A424\_1287 from A424) and G62*int* (A1S-2927 from ATCC 17978) were separately amplified by PCR using primer pairs F-G08int/R-G08int and F-G62int/R-G62int, respectively. Amplicons containing the integrases were ligated in the HindIII within the multiple cloning site (MCS) to be expressed under the *lacZ* promoter in the final recipient vector. The three PCR fragments (LF, RF, and integrase) were finally joined by fusion PCR resulting in a recombinant DNA product.

### *2.7. Construction of Inducible Plasmids for A. baumannii*

The vectors carrying the min-islands of G08 and G62 were sub-cloned into pWSK129, a lowcopy-number plasmid carrying aminoglycoside 3 -phosphotransferase (*aphA1*) gene conferring kanamycin-resistance (KmR) [26]. As this plasmid turned out to be non-functional in *A. baumannii*, we amplified the origin of transfer from pWH1277, a cryptic plasmid from an *A. lwoffii* strain fragment of pWH1266 (kindly donated by Philip Rather, Emory University, USA), using the primer pair PR3136 and PR3137 (Table S1). These pWSK129-WH plasmids were successfully transferred into competent *A. baumannii* knock-out strains (A424 and ATCC 17978) by conjugation.

### *2.8. Suicide Vector-Based Allelic Exchange for Mutant Construction in A. baumannii*

Deletion mutants of the GIs were constructed in *A. baumannii* using the suicide vector pJTOOL-3 [27], containing 500 bp long fragments of each of the borders of either G08 or G62. For transformation, *E. coli* CC118λ*pir* and S17.1λ*pir* were used as a host for replication and as a conjugative strain, respectively. Plasmid single cross-over insertion into *Acinetobacter* was selected by gentamicin and the double cross over by plating on 6% sucrose containing to check for the loss of the levansucrase *sacB* gene of pJTOOL-3. The expected genotype was obtained in all three randomly selected colonies that possessed the expected chloramphenicol-sensitive and gentamicin-resistant phenotype.

### *2.9. IPTG and Mitomycin C Induction*

For isopropyl β-D-1-thiogalactopyranoside (IPTG), 5 mL overnight culture of *A. baumannii* or *E. coli* were diluted at 1:100 into fresh LB and then incubated at 37 ◦C in the shaking incubator at 200× *g*, until OD600nm = 0.2 is reached. IPTG was added at concentration of 1.0 mM, and the cultures were then incubated, with 500 μL of the culture removed at time points 0, 4, 8, and 24 h for DNA preparation and qPCR analysis. In *Staphylococcus*, mitomycin was shown to induce excision of genomic islands [28]. For mitomycin C, induction 5 mL of overnight cultures of *A. baumannii* strains were treated with sub-lethal concentrations of mitomycin C MIC (0.5 times the MIC) for 2 h. Mitomycin C MIC of AYE, AB0057, A424, KR3831, and ATCC 17978 was found to range from 32–64 μg/mL. Non-induced cultures were run alongside in each occasion under identical conditions.

### *2.10. Metal Susceptibility Testing*

For testing of susceptibility to the heavy-metal salts, analytical-grade salts of CdCI2·H2O, CoCl2·6H2O, NiSO4·6H2O, and ZnSO4·7H2O, CuSO4·5H2O, FeSO4·7H2O, MnSO4·H2O. and As2SO3 (Sigma–Aldrich, Gillingham, UK) were used to prepare 1.0 M stock solutions, which were dissolved in ultrapure distilled water and later filter-sterilized and added to the medium at final concentrations of 1 mM. MIC and MBC assays to heavy metals was performed as described by the Clinical and Laboratory Standards Institute (CLSI) guidelines using a broth microdilution method [29]. Briefly, starting inocula of <sup>1</sup> × 105 CFU/mL of all *A. baumannii* strains were aliquoted in 96-well plates containing serial dilutions of each metal compound in the range 0.02–10 mM using MHB (Oxoid Ltd., Basingstoke, UK).

### **3. Results**

The distribution of integrases, as proxies of their genomic islands, varies widely in different lineages of a whole genome phylogenetic SNP tree constructed on all complete deposited *A. baumannii* genomes (Figure 2). One of the integrases not associated to genomic islands (*int1* ABAYE\_RS10930) present in almost all isolates, some others such as G08, G13, G16, and G42 are detected in many lineages and may be present only in a subgroup of isolates of a given ST. Other integrases like G09, G31 or G62 are present only in single or very few STs. The integrase of the metal resistance associated genomic islands G08 and G62 are representatives of this latter groups being G08 present in ST1, 25, 26, 52, 79, 81, 126, 138, 229, 422, and 638m while G62 only in ST10 and 437 (Figure 2).

**Figure 2.** Integrase distribution on an *A. baumannii* phylogenetic tree. The whole genome phylogenetic SNP tree was built using the FFP and input from the 105 complete *A. baumannii* genomes deposited in GenBank (08/07/2018) with the addition of strains A424 and KR3831. Strain names include the MLST "Pasteur" sequence type. The matrix of 16 integrases on the right part of the figure includes all integrase genes present in AB0057, AYE, and ATCC 17978 (red absent; blue present). The integrases of the GI are numbered according to Di Nocera et al. [14]. Three further integrases present in the three stains but not associated to genomic islands were included in the analysis and named *int1* (ABAYE\_RS10930), *int2* (AUO97\_RS03560), and the p3ABAYE integrase *int3* (ABAYE\_RS00155).

In order to test the hypothesis that these islands, even if present only in defined lineages are still mobile, we selected the genomic islands G08 and G62 respectively in strains AB0057 and ATCC 17978 [14] (Figure 1A). Phylogenetic analysis of the respective GI integrases confirmed that the G08*int* and G62*int* are not related, and each belong to a separate clade within the *Acinetobacter* GI-related tyrosine recombinases (Figure 1B). Both GIs carry the *copABCD* and *copRS* copper resistance genes [16], and G62 carries in addition the cadmium, zinc, iron, and cobalt resistance genes (Figure 1A). To check the distribution of G08 and G62, we screened by PCR a 100 sample collection of *A. baumannii* clinical isolates obtained from SQUH, Oman from 2012 to 2013 using primers for conserved sequences flanking the target region. The PCR screening analysis yielded a possibly occupied G08 in only one clinical isolate, and the presence of G08 was confirmed by WGS (strain KR3831, accession GCF\_003185745.1, GCA\_003185745.1). To test the contribution of G08 and G62 to metal susceptibility phenotypes, we constructed deletion mutants respectively in strain A424 and ATCC 17978 (Table 2). In the G08 knock out mutant, only the minimal inhibitory concentration (MIC) for manganese (MnSO4) decreased form 1 μg/mL to 0.5 μg/mL, while the copper MIC remained unchanged. Deletion of G62 in ATCC 17978 resulted in a decrease of the MIC of zinc (ZnSO4 from 4 to 2 μg/mL), cobalt (CoCl2 4 to 2 μg/mL), cadmium (CdCl2 4 to 2 μg/mL), and nickel (NiSO4 4 to 2 μg/mL), and again no decrease in the MIC of copper was detected (CuSO4 8 μg/mL). No differences were observed in susceptibility to iron (FeSO4) and arsenic (As2O3). Four independent ko mutants were assayed against the wild type in all tests and the difference in MICs found to be statically relevant.


**Table 2.** Metal susceptibility testing wild type (wt) and GI mutants.

\* MIC in μg/mL.

To test the functionality of integrase genes of the G08 and G62 islands in excising their respective GIs in a heterologous *E. coli* background, mini-islands were generated. Mini-GIs were obtained by cloning the integrase coding genes G08*int* (locus\_tag A424\_1287, strain A424, NC\_011586.2/CP001182.2) and G62*int* (locus\_tag A1S-2927, strain ATCC 17978, NC\_009085.1/CP000521) under control of the *lacZ* promoter and flanked by *attL* and *attR* sites [30] (Figure 3). The resultant plasmids pUC18-G08*int* and pUC18-G62*int* carrying the mini-islands of G08 and G62 were complemented in their respective knock out strains to be tested for integrase dependent excision using sets of divergent primers. The *dusA*–associated integrases have been shown to excise as circular elements with the restoration of the junction [19]. Our data showed IPTG-dependent mini-GI excision, by amplification of both the reconstituted target site and the junction of the circular form, for both the G08 and G62 constructs over the whole growth phase in liquid medium (Figure 4A,B). No or only marginal excision of the mini-GIs was detected without IPTG induction of the plasmid-carried mini-GIs (Figure 4A,B). To test for the excision of mini-GIs in *A. baumannii* background, the constructs were transferred on pWH1266 and pWK129 shuttle vectors. The IPTG-induced excision of the mini-GIs in *A. baumannii* was detected by amplification of the circular intermediates of the G08 and G62 mini-GIs using the same primer sets as in *E. coli* (Figure 4C).

**Figure 3.** G08 and G62 mini-GI vectors for *E. coli* and *A. baumannii*. pUC18 (**A**) used as a vector harboring *bla* gene conferring ampicillin resistance (green), *E. coli* origin of replication *ori* (grey), multiple cloning site MCS shown as green triangle, to clone mini-GIs containing upstream and downstream flanking borders with *attL/attR* sites (red) of the respective islands as well as integrase genes (**B**) G08*int* (olive) and (**C**) G62*int* (blue). The IPTG-inducible P*lac* promoter was fused by PCR in front of the integrase genes during construction of the plasmids (**B**,**C**). Plasmids used to assess integrase activity in *A. baumannii* (**D**–**F**) were constructed by fusing pUC18-based G08 and G62 mini-island vectors with pWSK129 carrying the *aphA1* conferring kanamycin resistance (orange) and pLG339 replication initiation protein (*repP*, light green). To allow for stable transfer of the constructs to *Acinetobacter* pWSK129-WH-G08 (**G**) and pWSK129-WH-G62 (**H**) were constructed. pWSK129-kan plasmids were used to clone the *Acinetobacter* origin of Replication (*oriR*) from pWH1266 (black color) resulting in pWSK129-WH-derived new constructs compatible with *A. baumannii* strains.

To test the dynamics of excision and reconstitution of the chromosomal target site of the G08 and G62 islands in *Acinetobacter*, we amplified the circular intermediates and targets in our four G08-positive AB0057, AYE, A424, and KR3831 strains, as well as the G62-positive strain ATCC 17978. To test GI excision, bacteria were grown to mid log phase and either tested directly or after exposure to 38 μg/mL of mitomycin C for 2 h. The junction and the circular forms were sequenced by Sanger sequencing to map the *att* sites of G08 and G62. In ATCC 17978, the *att* sites of G62 were identical at both ends with the consensus "AATAACTTTAAAGATTAA" [14]. However, our data show that in all examined strains, G08 was flanked by two 17 bp semiconserved attachment sequences [19], which showed variation in the *attR/attL* and *attP/attB* of a single nucleotide (SNP) in strains AYE and KR3831, and of two SNPs in strain AB0057 compared to ATCC 17978 (Figure 5). To examine the reason for these differences in the *att* sites among G08-harbouring strains, the database was searched for *att* sites of strains devoid of G08 and showed that two variable alleles of the *attB* sites in *dusA* gene exist in two G08-negative strains ATCC 17978 and AB307-0294. The ATCC 17978 had the more frequently occurring allele with a single SNP, whereas AB307-0294 had two SNPs similar to those

seen in AB0057. This could mean that either of the *att* sites could be recognized by the integrases as preferable integration/excision sites during mobilization of the G08.

**Figure 4.** Integrase induction drives excision of G08 and G62 mini-islands in *E. coli* and *A. baumannii*. Gel images of circularized G08 (from AB0057) (**A**) and G62 (from ATCC 17978) (**B**) mini-islands respectively, with (+) and without (−) IPTG induction. Mini-GI excision was tested by amplification of circular intermediates using outwards facing primer sets yielding 567 bp product for G08 (**A**) and a 489 bp product for G62 excision (**B**). Different time points (in hours) after IPTG induction in early exponential phase are shown (legend above gel). Integrase dependent excision of G08 and G62 mini-islands in *A. baumannii* is shown in panel C. Gel images of circularized G08 and G62 mini-islands 8 h after IPTG induction with the same primers as in *E. coli* (panel **A** and **B**). The non-induced samples in lane 3 for G8 and in lane 5 for G62 in lanes 4 and 5 of panel C. The marker is Gene Ruler 1 kb and the size of some bands in bp is given on the left of each gel.

### A


**Figure 5.** Allelic variation in the *attB* site of *A. baumannii* strains. (**A**) Shows the variability of *att* sequences in AB0057 and AYE and KR3831 compared to ATCC 17978. The *att* sites sequenced in this study are abbreviated as (seq). (**B**) Shows two variable alleles of *attB* sequences in *dusA* in two *A. baumannii* strains ATCC 17978 and AB307-0294 lacking the G08 island. The *attB* site is underlined. The variable alleles differing from the consensus sequence by two or three SNPs are shown in red.

Real-time PCR was performed to quantify the circular forms after excision and variation in number of excised elements between samples was corrected by arbitrarily setting the values of strain AB0057 to 1 and expressing the data in the other strains and after mitomycin C treatment at fold change. The primers' efficiencies were checked by performing serial dilutions (Figure S1). Without any induction, the excision of the elements was low and no strong baseline variation between the G08 carrying strains AB0057, AYE, A424, and KR3831 were seen for both the detection of the reconstituted target site and the circular intermediates (Figure 6A). After exposure of *A. baumannii* cells to mitomycin C, the detection of G08 circular intermediates increased in all strains significantly 4- to 8-fold (Figure 6B). Similarly, when testing excision of the G62 element in strain ATCC 17978, we detected a significant increase of about 4-fold in the formation of circular intermediates after exposure to mitomycin C (Figure 6C).

To test whether the increased excision of the G08 and G62 elements after mitomycin C treatment in *A. baumannii* was integrase mediated, we tested the expression of the integrases G08*int* (A424\_1287 from A424) and G62*int* (A1S-2927 from ATCC 17978). This was done by real-time PCR with primers internally to G08*int* and G62*int*. Data show significant upregulation of integrase expression after mitomycin C exposure of about 5-fold for G08*int* and 6-fold for G62*int* (Figure 6D).

**Figure 6.** Quantification of G08 and G62 island excision and integrase expression in *A. baumannii* strains with and without mitomycin C induction. The data in (**A**) represent the real-time PCR detection of reconstituted target sites after excision of G08 (excision) and of the circular intermediates (circularization) without any induction. The G08 positive strains are AB0057 strain (black, accession NC\_011586.2/CP001182.2). AYE (white, accession NC\_010410.1/CU459141.1), A424 (striped, accession GCA\_003185755.1), KR3831 (grey, accession: GCF\_003185745.1. GCA\_003185745.1), and data are normalized to AB0057. (**B**) Reports the variation in G08 excision by detecting circular intermediates with and without mitomycin C induction (0.75 × MIC) for 2 h. The asterisks represent the significance of change in each strain when compared to AB0057. Error bars represent SEM of three independent replicates. (**C**) Repost real-time PCR quantification of G62 circular intermediates with and without mitomycin C induction. (**D**) Shows the expression G08*int* (A424\_1287 from strain A424, protein ID: PRJNA473420:DMB35\_05960, Genbank accession: GCA\_003185755.1) and G62*int* (A1S-2927 from ATCC 17978) measured by real-time PCR without or with mitomycin C exposure (grey). Error bars indicate the SEM of three independent replicates in each experiment as analyzed by two-way ANOVA test. \*\* *p* < 0.01, \*\*\* *p* < 0.001.

### **4. Discussion**

Previous comparative genomic analysis of *A. baumannii* explored the chromosomal loci of 63 GIs including the two GIs objects of this study, G08 and G62, within seven strains belonging to different genotypes ST1, ST2, ST25, ST77, and ST78 [14]. The genomic alignment revealed GIs of various functions such as those encoding for surface components and transport systems, as well as resistance to drugs and heavy metals. More recently, data of a pan-genome analysis of 50 *A. baumannii* isolates and 249 previously sequenced *A. baumannii* strains were compiled [31], and their dataset confirmed the diversity of gene pools found within the GIs identified as an adaptive response of the *A. baumannii* strains to facilitate their survival in a nutrient-deficient environment. In many instances, the integrases of these GIs were found to be non-functional in various species due to frameshift and nonsense-mutations [32–34]. This resulted in most of these GIs being permanently positioned in their chromosomal location [35]. Therefore, this work aimed to check whether the G08 and G62 integrases are functional and contribute to excision of the islands by generating mini-GIs carrying the essential components for mobilization of integrases and *att* sites. In addition, direct excision from the chromosomal host was tested via mitomycin C induction. This approach has been previously employed in other *A. baumannii* studies [36] as well as other GIs circularization and have successfully demonstrated excision after the use of modified protocols [37–39]. The response to mitomycin C that was studied and research showed that pathogenicity island excision was facilitated by mitomycin C which induces an SOS response [28,40]. The data presented here have confirmed that the use of mitomycin C can effectively induce the excision of the GIs G08 and G62 via the visualization of the bands in gel electrophoresis. Similar observations were reported on other *dusA/dusB* associated integrases [19]. Real-time PCR data supported our observation and even demonstrated excision and circular events occurring at later cycles without induction. Sequencing of the circular intermediates and the chromosomal junctions after excision showed the possibility of having multiple *att* sites in AB0057, which probably can lead to having multiple insertions occurring at different frequencies depending on the most prevalent or preferable sites.

It could also be argued that the importance of such variability in attachment sites is probably minor, due to the low excision frequencies under laboratory conditions. Studies in which integrase activity was assessed in *A. baumannii* background are limited, most of which were performed in integron studies in the closely related non-pathogenic species *Acinetobacter baylyi* due to the ease of genetic modification and transformations [41–43]. In this context, our study demonstrated integrase-dependent excision and circularization in both tested metal-resistance GIs that were hypothesized to be non-mobile in the majority of the cases.

Moreover, the contribution of these two GIs towards metal susceptibility phenotypes was addressed by generating deletion mutants. Susceptibility data of both G08 and G62 mutants respectively of ATCC 17978 and A424 showed a significant, but minor decrease in the MIC for zinc (ZnSO4), cobalt (CoCl2), cadmium (CdCl2), and nickel (NiSO4), whereas the MIC remained unchanged for copper (CuSO4), iron (FeSO4), and arsenic (As2O3). These GI deletion mutants showed slight phenotypic changes as compared to their wild type counterparts, and their tolerance to the rest of the metals could be attributed the presence of other chromosomal efflux transports. Putative efflux pumps for copper and zinc have been previously analyzed in *A. baumannii* ATCC 17978 by TransAAP [44]. Thirteen efflux systems were identified that belong to either the CDF family, P-type ATPase family, CorA metal ion transporter family, HME family of RND transporters, or CopBtype family of Cu exporters. Transcriptional data by qPCR revealed that some of these putative efflux genes were induced by addition of either or both zinc and copper [16]. The MIC/MBC data of broth microdilution were identical in both A424 and ATCC 17978 wild-type and mutants for copper, cobalt, iron, and nickel. This observation was partially explained by the presence of *cnrCBA* mediating resistance to cobalt-nickel, as well as *czcCBA* genes which are cobalt-zinc-cadmium resistance determinants in this bacterial strain [45,46]. Similar iron susceptibility data in both wild-type and mutants of ATCC 17978 and A424 could be due to the presence of putatively non-functional FeoB in

ATCC 17978, and the tolerance could be attributed to another iron efflux transporter. The additive value of metal resistance carried on mobile elements, for example the copper resistance conferred by the *Staphylococcus aureus* COMER element in USA3000, still confers the strain's increased resistance to copper-related macrophage killing, and showed significant higher virulence, despite the weak phenotypes detected in vitro [47].

Collectively, this work reveals that metal resistance GIs in *A. baumannii* are of clinical significance as they confer metal resistance phenotypes, and their mobility could be demonstrated by the integrase assays. This issue can raise concern as these metal-resistance GIs could be readily transferred among strains (and patients) in clinical settings, and could be viewed as vehicles disseminating resistance as well as other potential virulence genes. The use of sub-lethal doses of antimicrobials and metal-containing compounds not only accelerate their resistance, but could also potentiate their virulence and spread in hospital environments.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2073-4425/9/7/366/s1, Table S1: List of primers, Figure S1: Standard curves for the qPCR.

**Author Contributions:** Z.A.-J. performed the experiments and wrote the manuscript, R.Z., performed genome analysis and contributed to revision of the manuscript, E.H.-P. contributed to the constructions of the knock out strains J.D.R. contributed to the genome analysis and the manuscript Z.A.-M. was responsible for the strain collection, K.R. designed and supervised the initial phases of the project, M.R.O. supervised the work and wrote the manuscript.

**Funding:** Z.A.-J. was funded by a PhD fellowship of the Sultan Qaboos University, Oman. This research received no further external funding.

**Acknowledgments:** Authors thank Philip Rather, Emory University, USA for plasmid pWH1266.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Article*
