INTRODUCTION Earlier studies showed that concentrations of antibiotics up to 230-fold below the MIC for susceptible bacteria (MICsusc) can both enrich preexisting resistance in a population and select for de novo resistant mutants (1, 2). Previous work focused on resistance caused by mutations or resistance genes located on the chromosome, and it remains unclear if the same selective effect of very low antibiotic concentrations ( 1 MSC). MSC is conditional and depends on the fitness cost of resistance. To further elucidate the dependence of MSC on the fitness cost of resistance, three of the resistance cassettes, tetRA (tetracycline resistance), dhfr (trimethoprim resistance), and the mph operon (erythromycin resistance), were transferred from the pUUH239.2 plasmid to the Escherichia coli chromosome by genetic recombineering (59). The results of the competitions between these strains and isogenic susceptible strains are shown in Fig. 3. These data demonstrate that the MSC is directly dependent on the fitness cost of the resistance. Thus, when the fitness cost conferred by the rest of the plasmid is eliminated, the selection coefficient curve is shifted upwards, thereby reducing the MSC value. For tetracycline, the MSC value was 1/25 of the MICsusc (Fig. 3A); for trimethoprim, the value was less than 1/100 (Fig. 3B); and for erythromycin, the value was less than 1/60 (Fig. 3C). These values represent absolute antibiotic concentrations of 30 ng/ml (tetracycline), less than 2 ng/ml (trimethoprim), and less than 200 ng/ml (erythromycin) for the chromosomally carried resistance compared to 45 ng/ml (tetracycline), 33 ng/ml (trimethoprim), and 3,000 ng/ml (erythromycin) when the corresponding resistance was present on the plasmid. As these resistance genes on the chromosome do not confer any significant fitness costs, the MSC values approach zero, making the MSC values for trimethoprim and erythromycin uncertain in this low range. FIG 3 Selection coefficients as function of antibiotic concentrations during competition experiments between strains carrying chromosomally carried resistance genes and susceptible strains at low levels of antibiotics. Competitions in the presence of tetracycline (A), trimethoprim (B), and erythromycin (C). Data can be found in Table S7 in the supplemental material. Standard deviations are indicated. DISCUSSION In this study, we investigated the selective effect of four clinically relevant antibiotics as well as two heavy metals on the maintenance of the ESBL plasmid pUUH239.2. This plasmid confers resistance not only to β-lactams but also to aminoglycosides, tetracycline, trimethoprim, sulfonamides, and erythromycin, as well as biocides such as quaternary ammonium compounds and the heavy metals copper, silver, and arsenic. Results show that selection of the multidrug resistance plasmid can occur at concentrations far below the MIC of a susceptible strain and that this general pattern is observed for many different classes of antibiotics and heavy metals. To exclude the possibility that any genetic changes might occur in the plasmid or chromosome during the course of the competition experiments (40 generations of growth), we performed whole-genome sequencing (WGS) of 12 independent lineages of plasmid-carrying bacteria that were grown for 40 generations in the presence of low levels of the different antibiotics and heavy metals. Results showed that none of the 12 lineages acquired any mutations in the chromosome or plasmid (see Table S9 in the supplemental material). The strongest selection in this study was observed with arsenic, where the level required for selection, the MSC, is close to 140-fold lower than the MIC of the susceptible wild-type strain. The MSC value of arsenite determined here corresponds to 7 mg/liter arsenic, a level that has been found in many types of environments. For example, it often occurs in certain sites that have been contaminated by mining (28, 60, 61). Furthermore, there are large areas around the world where the groundwater naturally contains high levels of inorganic arsenic, such as in Bangladesh (28, 62, 63), where millions of people get their drinking water from wells containing up to 300 µg/liter arsenic (64). Arsenic compounds have also been used in large quantities as growth promotion agents in animal feed, mainly as roxarsone, the organoarsenic compound 3-nitro-4-hydroxyphenylarsonic acid (65). The level of roxarsone in feed can be as high as 45.4 mg/kg, which corresponds to a molar amount of arsenic approximately 3-fold above the MSC for arsenite determined here (7 mg/liter). Thus, bacteria present in animal intestines might be exposed to arsenic levels that are sufficiently high to selectively maintain bacteria carrying a multidrug resistance plasmid, similar to the one studied here. Furthermore, most of the roxarsone is excreted unchanged from the animals, and litter from poultry houses using roxarsone-containing feed has been shown to contain approximately 30 mg/kg of arsenic. It is estimated that the broiler manure produced in the United States alone during the year 2000 contained roxarsone levels equivalent to 2.5 × 105 kg of arsenic, eventually ending up contaminating soil and water when manure from the animals was spread onto agricultural fields as fertilizer (29). It is likely that also other metals to which bacteria are known to carry resistance genes, such as silver, copper, mercury, and cadmium, would constitute a similar selective force. This is problematic since metals of various kinds are widely applied in industries, agriculture, construction, health care, and other areas from which they are released into the environment (66 – 75). The combined effect of mixtures of substances such as metals and antibiotics that are present in the environment is an additional complicating factor regarding selection (46, 76). The fact that combined effects have been demonstrated here is of great concern since it shows that even very small amounts of several compounds, where each individual compound is not sufficiently high to confer selection, could in combination add up and result in selection of resistance. The combined effects of the compounds tested here vary from synergistic to less than additive. Thus, the combination of the antibiotics erythromycin and trimethoprim shows a synergistic effect, where the presence of one of the drugs lowers the MSC of the other. Likewise, combining various concentrations of arsenite, tetracycline, and trimethoprim results in stronger selection, but here, the effect is less than additive. The difference between these two mixtures of drugs may be the result of the specific mechanism of action of each substance interfering with the other, and other combinations of metals and antibiotics with different mechanisms of action could therefore result in either greater or lesser combination effects. However, based on these experiments it seems that each new compound added to the mixture will lower the MSC of the others. Because of the high number of compounds present in many environments, this combination effect could be a contributing factor to the persistence of multidrug resistance plasmids, even though each individual compound is present at levels below the minimal selective concentration. Finally, our data show that the MSC for a specific resistance is conditional and depends on the fitness costs of the resistance mutation/gene. Thus, when the cost associated with the resistance was reduced (here achieved by moving the resistance genes from the plasmid to the chromosome), the MSC was also reduced correspondingly. For the three antibiotics examined, the reduction in MSC associated with the transfer of the resistance from the plasmid to the chromosome was 2- to 15-fold. An important implication from this finding is that a cost-free resistance could in principle be maintained in a population by an infinitesimally low concentration of antibiotic, provided that there are no threshold effects with regard to the inhibitory effect of antibiotics. As the initial cost of resistance mutations can often be efficiently compensated by second-site mutations without loss of resistance, it is expected that cost-free resistances are common in natural settings (77). MATERIALS AND METHODS Strain construction. All strains in this study were derived from the wild-type E. coli MG1655 strain and are listed in Table 1. Single-cell tracking of the different strains during the competition assays was enabled by integration of chromosomal copies of either a blue (mTagBFP2) (47) or yellow (SYFP2) (48) fluorescent protein gene. Both genes were codon optimized for expression in E. coli and synthesized by DNA 2.0. The two genes were inserted into galK using the λ red recombineering system as previously described (59) and moved by P1 transduction into the wild-type strain. In strains DA26735 to DA26738, the fluorescent genes are expressed by the promoter PLlacO (78), while in the remaining fluorescent strains, they are expressed from the weaker promoter J23101 (79) (GenBank accession numbers KM018299 to KM018302). The strains carrying the plasmid pUUH239.2 (GenBank accession number CP002474.1) were constructed by conjugation from a clinical isolate into a strain carrying the λ red recombineering system (59) on the chromosome. To eliminate the risk of plasmid conjugation during the competitions, the gene controlling the transfer operon of the plasmid, traJ (49 – 53), was replaced with a chloramphenicol resistance marker using recombineering. The pUUH239.2 ΔtraJ::cat plasmid was then conjugated into the strains carrying chromosomally integrated fluorescent markers by overexpression of a cloned copy of traJ in trans from a second plasmid (GenBank accession number KM018297) in the donor strain. TABLE 1 Strains and genotypes Strain Genotype Source or reference DA5438 E. coli MG1655 wild type (parent) Strain collection DA25916 /pUUH239.2 This study DA26735 ΔlacIZYA::FRT galK::mTagBFP2-amp This study DA26736 ΔlacIZYA::FRT galK::SYFP2-amp This study DA26737 ΔlacIZYA::FRT galK::mTagBFP2-amp/pUUH239.2 ΔtraJ::cat This study DA26738 ΔlacIZYA::FRT galK::SYFP2-amp/pUUH239.2 ΔtraJ::cat This study DA28200 galK::SYFP2-FRT This study DA28202 galK::mTagBFP2-FRT This study DA28893 galK::SYFP2-FRT ΔbglGFB::FRT This study DA28895 galK::mTagBFP2-FRT ΔbglGFB::FRT This study DA28702 galK::SYFP2-FRT ΔbglGFB::dhfr This study DA28704 galK::mTagBFP2-FRT ΔbglGFB::dhfr This study DA28706 galK::SYFP2-FRT ΔbglGFB::mph This study DA28708 galK::mTagBFP2-FRT ΔbglGFB::mph This study DA28759 galK::SYFP2-FRT ΔbglGFB::tetRA This study DA28761 galK::mTagBFP2-FRT ΔbglGFB::tetRA This study The strains carrying chromosomally integrated resistance genes from pUUH239.2 were constructed by first replacing the genes bglGFB with a chloramphenicol marker flanked by FLP recombination target (FRT) sites as well as terminators to ensure transcriptional insulation from the surrounding genes. In a second recombineering step, the chloramphenicol marker was replaced with resistance genes amplified from the pUUH239.2 plasmid (dhfr, tetRA, and the mph operon). The chromosomally integrated pUUH239.2 resistance genes were then moved into the fluorescent strains by P1 transduction. The ΔbglGFB::cat marker was also moved into the fluorescent strains using P1 transduction, and the chloramphenicol marker was removed using FLP recombinase to generate the ΔbglGFB::FRT strains DA28893 and DA28895. Primers used are listed in Table S2 in the supplemental material. MIC measurements. Susceptibilities of the strains to different antibiotics (Table 2) were determined using Etests according to the manufacturer’s instructions (bioMérieux, Marcy l’Étoile, France). Susceptibility to copper and arsenic was determined using broth microdilution with an initial bacterial inoculum of 106 cells and an incubation time of 18 to 20 h at 37°C without shaking. The MIC of sodium arsenite was determined in LB medium and defined as the lowest concentration where no visible growth was observed. MIC determinations of copper(II) sulfate were conducted in minimal medium (M9 glucose) with limited access to oxygen. TABLE 2 MICs, MSCs, and relative fitness values for the different strains used a Strain Resistance Fitness Arsenite Cu(II) sulfate Tet Trm Ery Kan MSC (µM) MIC (µM) MSC (ng/ml) MIC (ng/ml) MSC (ng/ml) MIC (ng/ml) MSC (ng/ml) MIC (ng/ml) MSC (µg/ml) MIC (µg/ml) MSC (µg/ml) MIC (µg/ml) DA5438 None (wt) 1.0 NA 12,500 NA 1,300 NA 750 NA 190 NA 12 NA 0.75 DA25916 pUUH239.2 0.96 90 25,000 90 >80,000 45 24,000 33 >32,000 3 96 0.47 24 DA28759 tetRA 0.99 NA NA NA NA 30 48,000 NA NA NA NA NA NA DA28702 dhfr 1.0 NA NA NA NA NA NA 32,000 NA NA NA NA DA28706 mph 1.0 NA NA NA NA NA NA NA NA 256 NA NA a Abbreviations: MSC, minimal selective concentration; Tet, tetracycline; Trm, trimethoprim; Ery, erythromycin; Kan, kanamycin; NA, not applicable; wt, wild type. Competition assays. Competition experiments were performed between a susceptible strain expressing mTagBFP2 or SYFP2 and an isogenic resistant strain expressing the other marker. Overnight cultures were mixed 1:1 and serially passaged with a 1,000-fold dilution every 24 h, resulting in 10 generations of growth per serial passage. The ratios between the two competing strains were measured by counting 105 cells at each serial passage using flow cytometry (BD FACSAria IIu). The selection coefficients were calculated using the regression model s = {ln[R(t)/R(0)]}/[t], as previously described (80), where R is the ratio of resistant to susceptible strain. Control experiments were performed to determine the relative cost of expressing the mTagBFP2 marker compared to the SYFP2 marker, and this difference in cost was used for compensation of the calculated selection coefficients (see Table S1 in the supplemental material). Competitions were terminated if nonlinear slopes were detected in plots of the ratio of resistant to susceptible bacteria over time. Nonlinearity indicates that a beneficial adaptive mutation occurred in one of the populations (periodic selection). This phenomenon is impossible to completely avoid (81) since medium adaptation mutations will always occur at some frequency (82). In our experiments, only 4 out of 525 competitions were discarded (see Tables S3 to S8) and the adaptive mutations appearing in these competitions were not further investigated. The competitions using arsenic and using arsenic plus tetracycline were performed in Luria-Bertani broth, the competitions using copper were performed in minimal medium (M9 plus 0.2% glucose) with limited access to oxygen, and the remaining competitions were performed in Mueller-Hinton medium (similar results were obtained in these two media). WGS. To investigate whether the bacteria undergo any genetic changes during the competition experiments, 12 independent cultures of bacteria carrying the plasmid (strain DA26737) were analyzed using whole-genome sequencing (WGS) before and after being passaged in the presence of low levels (2× MSC) of all the different antibiotics and heavy metals. Two independent cultures per antimicrobial agent were passaged for 40 generations and analyzed (see Table S9 in the supplemental material). Genomic DNA was prepared from 3 ml of overnight cultures using the Genomic-tip 100/G columns and the Genomic DNA buffer set (Qiagen, Netherlands), according to the manufacturer’s instructions. The genomic DNA was then sequenced using the Illumina sequencing technology with 500-bp paired-end libraries by the BGI sequencing facility (Hong Kong). Mutations were identified by assembling the sequencing reads to the reference genome sequence in CLC Genomics Workbench 7.5 (CLC Bio, Denmark). Illumina reads of each strain were individually mapped onto the reference genome of the wild-type MG1655 and the reference pUUH239.2 plasmid sequenced earlier in our laboratory. The WGS data show that no mutations or rearrangements (83) were found in the plasmid or chromosome during this time frame. Nucleotide sequence accession numbers. Nucleotide sequence accession numbers are as follows: Klebsiella pneumoniae plasmid pUUH239.2, CP002474.1; conjugation activation plasmid pMH3CIq-PT5lac-traJ for pUUH239.2, KM018297; synthetic fluorescent protein expression cassette cat-J23101-mTagBFP2, KM018299; synthetic fluorescent protein expression cassette cat-J23101-SYFP2, KM018300; synthetic fluorescent protein expression cassette amp-PLlacO-mTagBFP2, KM018302; synthetic fluorescent protein expression cassette amp-PLlacO-SYFP2, KM018301. SUPPLEMENTAL MATERIAL Figure S1 Competitions with low initial fractions of resistant bacteria. Competition experiments at different concentrations of arsenite and different starting fractions of resistant bacteria carrying the pUUH239.2 plasmid. (A) Initial ratio of susceptible to resistant mutants, 10:1. (B) Initial ratio of susceptible to resistant mutants, 100:1. Data can be found in Table S8. Standard deviations are indicated. Download Figure S1, TIF file, 0.1 MB Table S1 Costs of fluorescence markers. Table S1, XLSX file, 0.04 MB. Table S2 Primers used in this study. Table S2, DOCX file, 0.1 MB. Table S3 Data for Fig. 1 Table S3, DOCX file, 0.1 MB. Table S4 Data for Fig. 2A Table S4, DOCX file, 0.1 MB. Table S5 Data for Fig. 2B Table S5, DOCX file, 0.1 MB. Table S6 Data for Fig. 2C Table S6, DOCX file, 0.1 MB. Table S7 Data for Fig. 3 Table S7, DOCX file, 0.1 MB. Table S8 Data for Fig. S1 Table S8, DOCX file, 0.04 MB. Table S9 Summary of whole-genome sequencing data. Table S9, DOCX file, 0.1 MB.