Selective Pressure of Antibiotic Pollution on Bacteria of Importance to Public Health

1Antibiotic pollution may facilitate the development and spread of antibiotic resistance. Antibiotics used in human and veterinary medicine can enter the environment via wastewater treatment plant effluents, hospital and processing plant effluents, application of agricultural waste and biosolids to fields, and leakage from waste-storage containers and landfills. One of the difficulties of relating increased levels of resistance in the environment to antibiotic pollution, however, is the fact that antibiotic resistance genes can be co-released into the environment with antibiotic compounds. The question then is whether an observed increase in resistance emerged as a result of the selective pressure of the antibiotic in the environment or if it emerged within the treated host.

2Antibiotic resistance genes and antibiotic compounds are different pollutants that have different modes of action and are subject to different fate processes in the environment. They are also likely to respond differently to treatment processes designed to remove them from environmental compartments and from liquid and solid wastes. Estimating the relative contribution of pollution by antibiotic resistance genes and antibiotic compounds to increased levels of antibiotic resistance is important, as this knowledge may be used to improve the effectiveness of counteractive measures.

3Except for the the report by, who found excessive concentrations of antibiotics and concurrent high levels of resistance in streams receiving effluents from a drug-production plant in India, there is limited evidence as to whether environmental concentrations of antibiotics can enhance the development and spread of resistance in the environment (e.g., ). Current guidelines on the environmental risk assessment of medicinal products for human and veterinary use in the European Union, for example, do not explicitly address the effect of antibiotics on the prevalence of antibiotic resistance in the environment [European Medicines Agency (EMEA) 2006, 2008].

4From an environmental health perspective, the selective pressure that antibiotic pollution may exert on clinically important bacteria is of particular concern. Several clinically relevant bacteria, such as Escherichia coli and the enterococci, occur and are able to grow in different environments. In the presence of environmental concentrations of antibiotics, they may face a selective pressure leading to a gradual increase in the prevalence of resistance.

5In the present study we used bacterial species sensitivity distributions derived from a comprehensive set of minimum inhibitory concentration (MIC) distributions of antibiotics to model bacterial sensitivities and characterize the selective pressure that antibiotic pollution may exert on bacteria of importance to public health that are found in the environment. The study was carried out under the premise that antibiotics will primarily increase the prevalence of resistance by favoring the selection of resistant phenotypes via the inhibition of sensitive ones. Although there is evidence to suggest that subinhibitory concentrations of antibiotics may indirectly favor resistance, the use of bacterial inhibition as an assessment end point provides a standardized response across taxa that can be directly linked to a selective pressure favoring an increase in the prevalence of resistance.

6We derived species sensitivity distributions for three antibiotics from publicly available MIC distributions and determined the fraction of inhibited bacterial taxa at antibiotic concentrations that were measured in different environments and used as action limits in environmental risk assessment. To our knowledge, this is the first study in which measured environmental concentrations of antibiotics are examined in regard to the antibiotic sensitivity of clinically relevant bacteria.

7MIC distributions. We obtained MIC distributions for ciprofloxacin, erythromycin, and tetracycline from the European Committee on Antimicrobial Susceptibility Testing (EUCAST) MIC and zone diameter distribution website (EUCAST 2010; ). Distributions are based on data collated from > 20,000 different worldwide sources and encompass the variability within species and between researchers, methods, and geographic areas.

8Ciprofloxacin, erythromycin, and tetracycline were selected from among a list of approximately 150 compounds in the database for three reasons: a) they represent three distinct classes of antibiotics of importance to human and veterinary medicine; b) the number of bacterial taxa represented in their MIC distributions was higher than in most of the other compounds in the database; and c) they have been measured in different environmental compartments.

9Phylogenetic analysis. To place bacteria represented in the EUCAST MIC distributions in a wider phylogenetic context, we conducted a brief phylogenetic analysis. 16S rRNA sequences from bacterial taxa represented in the MIC distributions were obtained from the All-Species Living Tree Project (LTP; ), March 2011 release, and imported into ARB software. We selected bacterial taxa represented in the MIC distributions of ciprofloxacin, erythromycin, and tetracycline to create a pooled 16S rRNA sequence alignment comprising the species represented in the MIC distributions of all three antibiotics. Some bacterial taxa were not represented in the LTP database; therefore, the alignment contained a subset of the taxa represented in each original MIC distribution [see Supplemental Material, Table S1 (http://dx.doi.org/10.1289/ehp.1104650)]. This alignment, along with the entire LTP alignment for the domain Bacteria, was imported into the ape package (version 2.6-2) of the R environment for statistical computing; we then calculated an evolutionary distance matrix for each alignment using Kimura’s two-parameter substitution model. An unrooted phylogenetic dendrogram was estimated from the evolutionary distance matrix of the pooled antibiotic alignment using the neighbor-joining method, and confidence was assessed by bootstrapping with 1,000 permutations. We plotted histograms of pairwise evolutionary distances covered by the pooled 16S rRNA alignment of species represented in the MIC distributions of all three antibiotics and by the entire LTP alignment for the domain Bacteria to assess the range of evolutionary distances covered by our data set in relation to that of all sequenced type strains of bacteria.

10Bacterial species sensitivity distributions. End point selection. Species sensitivity distributions were derived using the median MIC (MIC50) and the no observed effect MIC (NOEC) of each taxon. MIC tests are performed using double-dilution steps of antibiotic concentrations, and the data they generate is interval censored. Therefore, we considered the conservative MIC50 of each species to be the antibiotic concentration immediately below the observed 50th percentile, and the NOEC to be the antibiotic concentration immediately below the lowest MIC observed in each taxon. MIC50 and NOEC values were aggregated within cogeneric species by using the arithmetic mean to minimize the lack of independence between individual observations [see Supplemental Material, Section 1 and Figure S1 (http://dx.doi.org/10.1289/ehp.1104650)]. To derive species sensitivity distributions, we used pooled MIC50 and NOEC values only from genera for which there was evidence to suggest that—under certain conditions—the genera could grow in an environmental compartment (e.g., soil, sewage, fresh water) (for details regarding the selection of genera, see Supplemental Material, Section 2 and Table S2).

11Linking end points to resistance. The MIC50 was calculated including data beyond the wild-type cut-off value (COWT). The COWT separates microorganisms with (i.e., non–wild-type) and without (i.e., wild-type) acquired resistance mechanisms and represents an antibiotic concentration above which only bacteria with acquired resistance mechanisms can grow A comparison of MIC50 values with the wild-type MIC range of species in the EUCAST distributions indicates that the MIC50 is an adequate estimate of the wild-type MIC (i.e., it falls within the wild-type MIC range in those species that have one), except for a few cases in which it falls above the wild-type MIC range. Concentrations of antibiotics ≥ MIC50 are therefore likely to inhibit approximately half of the wild-type population. Assuming equal growth rates of wild-type and resistant populations, this causes an increase in the prevalence of resistance in the remaining active populations (Figure 1). In contrast, the NOEC represents a minority of isolates across taxa whose MIC values are sometimes below the MIC range representative of the wild-type population; we used the NOEC as a means to assess the lower limit of antibiotic sensitivity represented in the MIC distributions.

12Bootstrap regression. We derived species sensitivity distributions by bootstrap regression, as described by. The MIC50 and NOEC vectors of each antibiotic were resampled 5,000 times. To each of these bootstrap resamples, a log-logistic model was fitted by maximum likelihood estimation of the distribution parameters and direct optimization of the log-likelihood function following the method of and using the fitdistrplus package, version 0.1–4, of the R environment for statistical computing. The distribution parameters α (i.e., location) and β (i.e., scale) from each fitted curve were used to derive 5,000 replicate estimates of antibiotic concentrations associated with a potentially affected fraction between percentiles 0.01 and 0.99 at 0.01-step intervals. From these, the bootstrap estimate and 95% bootstrap confidence intervals (CIs) were calculated. The MIC distributions used in this study, along with an R script to replicate the analysis, are available from the authors.

13We determined the potentially affected fraction of bacterial genera by all three antibiotics at the aquatic and soil VICH phase I action limits [International Cooperation on Harmonisation of Technical Requirements for Registration of Veterinary Medicinal Products (VICH) 2000] and at measured environmental concentrations reported in the literature (e.g., ). This was complemented with a direct comparison of measured environmental concentrations and VICH action limits with the COWT of species represented in the MIC distributions of each antibiotic. Although ciprofloxacin is not approved for use in veterinary medicine, it is the major active metabolite of enrofloxacin in different species. In the absence of data for enrofloxacin, we used ciprofloxacin as a representative of the fluoroquinolones in comparisons of potentially affected fractions and VICH phase I action limits. We obtained measured environmental concentrations of ciprofloxacin, erythromycin, and tetracycline from,, and, and from data collated by. Antibiotic concentrations are expressed as parts per billion to facilitate analysis and comparisons.

14Phylogenetic and environmental overview of MIC distributions. Seventy-nine species from the ciprofloxacin, erythromycin, and tetracycline MIC distributions were represented in the LTP 16S rRNA database [Figure 2; see also Supplemental Material, Table S1 (http://dx.doi.org/10.1289/ehp.1104650)]. Major bacterial groups in Figure 2 (appearing in order from top to bottom in the dendrogram) include staphylococci, enterococci, streptococci, a few representatives of the Actinobacteria (e.g., Clostridium spp., Mycobacterium spp.), Bacteroides, pseudomonads (e.g., Pseudomonas spp., Burkholderia spp.), and the enterics (e.g., Escherichia spp., Enterobacter spp., Proteus spp., Klebsiella spp.). The range of evolutionary distances covered by these species spans the range of evolutionary distances represented in the entire LTP database, highlighting it as a rather diverse phylogenetic group (Figure 3).

15Bacterial genera used to derive the species sensitivity distribution of each antibiotic represent commensal and pathogenic bacteria that occur, and may grow, to a larger or lesser extent in the environment (Table 1). Among the 27 genera included in the species sensitivity distributions are some known to be widely distributed in the environment, such as Pseudomonas, Acinetobacter, Burkholderia, and Chryseobacterium, as well as others for which growth in the environment has either been reported or for which there is evidence to suggest that under certain conditions it is likely to occur [see Supplemental Material, Table S2 (http://dx.doi.org/10.1289/ehp.1104650)].

16Inhibitory effects at environmental concentrations. The log-logistic model had a good fit to the NOEC and MIC50 vectors, explaining ≥ 90% of the variance in the original data (Figure 4). Table 2 shows that the potentially affected fraction of bacterial genera at measured environmental concentrations of ciprofloxacin, erythromycin, and tetracycline in water environments—including surface water, sewage treatment plant (STP) effluents, raw sewage, and groundwater—is low, with upper 95% CIs of 7% and 3.2% in raw sewage for ciprofloxacin using the NOEC and MIC50 species sensitivity distributions, respectively. At the low range of concentrations measured in surface waters and STP effluents, the practical difference between potentially affected fraction estimates is minimal using the NOEC and MIC50 species sensitivity distributions. For erythromycin, the NOEC and MIC50 species sensitivity distributions overlap at the lower tail of the distributions, causing the MIC50 species sensitivity distribution to estimate slightly higher potentially affected fractions than the NOEC species sensitivity distribution for concentrations measured in surface waters and STP effluents. Given that the potentially affected fractions of both species sensitivity distributions are well below 1% at this range of concentrations, this discrepancy was not significant for our assessment of effects.

17Potentially affected fractions in river sediments, swine feces lagoons, liquid manure, and farmed soil are markedly higher than those in aquatic compartments (Table 2). Concentrations of ciprofloxacin, erythromycin, and tetracycline measured in river sediments are ≥ MIC50 values of ≤ 60%, 7.6%, and 6% of the bacterial genera, respectively (i.e., upper 95% CIs in Table 2). Estimated concentrations of these three antibiotics in the sediments of a swine feces lagoon are ≥ MIC50 values of ≤ 92%, 21%, and 56% of the bacterial genera. The extremely high concentration of tetracycline in liquid manure reported by is ≥ MIC50 of 100% of the bacterial genera. In contrast, the high concentration of tetracycline measured in farmed soil is ≥ MIC50 for ≤ 30% of genera. We estimated that the tetracycline concentration reported by for farmed soil inhibits at least some isolates in up to 94% of the bacterial genera (i.e., NOEC species sensitivity distribution). Some environments, such as soil and sediments, are likely to contain more bacterial genera of clinical relevance than others. Thus, for example, a potentially affected fraction of 30% for tetracycline in farmed soil may inhibit more bacterial genera than a potentially affected fraction of 100% in liquid manure and therefore have larger public health implications.

18For bacterial taxa represented in the MIC distribution of each antibiotic, we compared MIC values ≥ COWT with measured environmental concentrations (Figure 5).Concentrations > COWT for a given bacterial taxa completely inhibit the wild-type population, increasing the prevalence of resistance in the remaining active population to 100%. The measured environmental concentration of ciprofloxacin in swine feces lagoon sediment is > COWT for 14 bacterial taxa belonging to nine genera of predominantly enteric bacteria (Figure 5A). The concentration of tetracycline measured in liquid manure is > COWT for all but one bacterial taxa, and the concentration measured in swine feces lagoon sediment is borderline with the COWT for Staphylococcus and Streptococcus (Figure 5C). Measured environmental concentrations of erythromycin are < COWT for all taxa (Figure 5B).

19Inhibitory effects at VICH phase I action limits. Potentially affected fractions at the VICH phase I aquatic action limit suggest that the action limit is protective of major inhibitory effects on bacteria by all three antibiotics (Figure 4), although a minority of sensitive individuals could be inhibited in up to 12% of genera (i.e., upper 95% CI in ciprofloxacin NOEC species sensitivity distribution) (Table 2). The 1-ppb VICH phase I aquatic action limit refers to an environmental introduction concentration (i.e., concentration in an effluent); thus, exposure concentrations in receiving water bodies are further reduced by dilution.

20Concentrations of ciprofloxacin and erythromycin at the VICH phase I soil action limit are estimated to be ≥ MIC50 values of ≤ 76% and 25% of bacterial genera, respectively (Figure 4, Table 2). Figure 4 shows that the VICH phase I soil action limit is below the erythromycin and tetracycline COWT for all species, indicating that at this concentration these antibiotics are not expected to inhibit 100% of the wild-type population in any species. Conversely, the ciprofloxacin MIC distributions (Figure 5A) show that the 100-ppb soil action limit is > COWT for five bacterial taxa and borderline with that for nine other taxa.

21It is also illustrative to consider these action limits in relation to the empirical MIC50 of individual species. The MIC50 for ciprofloxacin, for example, is 8 ppb in E. coli (n = 17,877), 8 ppb in Enterobacter cloacae (n = 2,354), 64 ppb in Acinetobacter lwoffi (n = 262), 125 ppb in Pseudomonas aeruginosa (n = 27,387), and 125 ppb in Campylobacter coli (n = 2,532). The MIC50 for erythromycin is 32 ppb in Streptococcus pneumoniae (n = 40,452) and 125 ppb in Staphylococcus aureus (n = 36,038). The MIC50 for tetracycline is 32 ppb in Clostridium difficile (n = 832) and 125 ppb in Str. pneumoniae (n = 13,813). These MIC50 are between 12.5 times lower to slightly higher than the 100-ppb soil action limit. Several species also have NOECs that are orders of magnitude lower than the 100-ppb soil threshold.

22In this study we found that environmental concentrations of antibiotics, as well as concentrations representing action limits used in the environmental risk assessment of veterinary medicines, may be high enough to inhibit growth in bacteria of clinical importance occuring in different environments. By completely or partially inhibiting the growth of wild-type bacterial populations, antibiotics cause a selective pressure that will increase the prevalence of resistance. The potentially affected fractions for ciprofloxacin, erythromycin, and tetracycline at measured environmental concentrations of river sediments, swine feces lagoon sediments, liquid manure, and farmed soil suggest that these environments are likely to be hot spots for the selection of resistance. In this regard, the comparison of measured environmental concentrations of ciprofloxacin and tetracycline and their respective COWT values in different bacteria is striking (Figure 5) because it shows that wild-type populations of certain species are completely inhibited at these concentrations in vitro. In swine feces lagoons, liquid manure, and soil amended with manure, concentrations of certain antibiotics can build up to levels that may act to extend the antibiotic selective pressure that started for some bacteria within their treated hosts and exert a new selective pressure on other bacteria. Interestingly, in a study of soil from the 1970s to the present, found a significant increase in tetracycline resistance genes that mimicked the use of tetracyclines in agriculture in the Netherlands. Although studies have generally failed to find a significant effect of tetracyclines on resistance levels in soil (e.g., ), some evidence suggests that it could have contributed to the persistence and prevalence of resistance genes. Our results indicate that tetracycline concentrations in soil may build up to levels high enough to exert a significant selective pressure on clinically relevant bacteria.

23The extrapolation of MIC data to the field has inherent limitations that must be considered in the interpretation of our results. Physicochemical and biological conditions of MIC tests, for example, are not representative of those generally encountered by bacteria in the environment. MIC tests are also acute tests, whereas the exposure to antibiotics in the environment is mainly chronic and will exert a constant selective pressure over extended periods of time. Chronic exposure provides a longer temporal window for the selective enrichment of resistance and will favor stepwise transitions from low-level to high-level clinical resistance. In this study we did not formally address the bioavailability of antibiotics in the environment. However, tetracyclines and macrolides have been shown to retain their bioactivity and inhibit bacterial growth even when tightly adsorbed by clay particles; there is also evidence to suggest that fluoroquinolones retain part of their activity when sorbed to solids. In the environment, bacteria are likely exposed to multiple antibiotics and other substances, such as metals and disinfectants, which will affect the selection of resistance. Synergistic and/or antagonistic interactions between combinations of antibiotics, for example, may significantly influence the evolution of resistance. Because many antibiotic resistance genes are associated with mobile genetic elements carrying multiple antibiotic resistance genes and genes conferring resistance to heavy metals and/or disinfectants, any of these factors may select for multidrug resistance. An example relevant to this study is the broad host range transposon Tn1545, which encodes resistance to tetracycline, erythromycin, and kanamycin. In all, these factors emphasize the complexity of relating antibiotic pollution to the prevalence of antibiotic resistance in the environment, and inevitably introduce a degree of uncertainty in our results. Despite these limitations, however, our results provide a means to grasp the potential effect of antibiotic pollution on the prevalence of resistance in clinically relevant bacteria in the environment by putting measured environmental concentrations in perspective with bacterial sensitivities.

24The link between MIC50 and resistance is based on the assumption that antibiotic concentrations ≤ COWT may increase the prevalence of resistance to < 100% by inhibiting a fraction of the wild-type population. Variation within the wild-type part of MIC distributions is normally in the order of 3–5 log2 MIC steps. Although this variation may reflect inherent variation in antibiotic sensitivity and other biological features that influence the MIC, it may also reflect method variability. In environmental compartments such as those discussed in this study, it is reasonable to expect some degree of inherent variation in the antibiotic sensitivity of wild-type populations. Environments such as sewage, river sediments, and agricultural soil act as transient or permanent sinks in which wild-type populations of the same species, from different sources, and with slightly differing sensitivities may physically converge. These environments are also likely to have microgradients of physicochemical variables, such as pH and nutrients, that are known to affect the MIC of bacteria in vitro. Collectively, these factors may provide enough variation in antibiotic sensitivity to enable the differential inhibition of wild-type populations under equal measured environmental concentrations of antibiotics.

25The VICH phase I guidance document (VICH 2000) informs environmental risk assessment of veterinary medicines and has been implemented in the regulatory scheme in the European Union, United States, Japan, and Australia. Under VICH phase I guidance, the environmental risk assessment of a veterinary medicine—except for parasiticides—is discontinued if the medicine’s introduction concentration into the aquatic environment is < 1 ppb (i.e., aquatic action limit). For terrestrial environments, environmental risk assessment is discontinued if the predicted environmental concentration in soil is < 100 ppb (i.e., soil action limit). Our results suggest that the VICH phase I soil action limit for veterinary medicines does not protect background antibiotic resistance levels. Certain antibiotics at concentrations < 100 ppb may inhibit a significant fraction of clinically relevant bacteria in the environment (Figure 4); the high potentially affected fractions for erythromycin and ciprofloxacin at the 100-ppb soil threshold are a clear example of this. In a critique of action limits, used MIC data from 13 soil microorganisms and 22 antimicrobials to construct a substance–species sensitivity distribution; on the basis of this distribution, he determined that the aquatic and soil action limits should be set at 4 × 10–4 and 1 ppb, respectively, if they were to be protective for all compounds. Similarly, our results suggest that VICH phase I action limits leave an ample margin for antibiotics to exert a selective pressure on bacteria of clinical importance in the environment.

26Current knowledge on the presence and mechanisms of bacterial resistance of clinical and environmental origin clearly indicate that the resistome of pathogens is and will continue to be inevitably linked with the environment. Moreover, as this study shows, the prevalence of resistance in bacteria of importance to public health has the potential to be increased by antibiotic pollution in the environment. Therefore, to minimize the potential effect of antibiotic pollution on antibiotic resistance, resistance—or a proxy thereof—should be considered in environmental risk assessment of human and veterinary antibiotics.

27MIC distributions are at the center of clinical microbiology. In conjunction with drug pharmacokinetics, MIC distributions are used to establish clinical breakpoints for the effective treatment of infectious diseases. Similarly, we suggest that MIC distributions can be used to explicitly link environmental concentrations of antibiotics with the prevalence of resistance, and can therefore provide a cogent framework to address the potential effects of antibiotics on antibiotic resistance in the initial phase of a risk assessment. Just as pharmacokinetics provides information on the fate of antibiotics in the body, environmental exposure assessment can be used to further refine the assessment of effects. If necessary, MIC distributions may be used to set breakpoints to protect background resistance levels in the environment.

28Antibiotics are present in different environments as a result of their use in human and veterinary medicine; concentrations currently found in different environments and concentrations representing action limits used in environmental risk assessment may be high enough to exert a significant selective pressure on clinically relevant bacteria. The potentially affected fraction of bacterial genera at concentrations of antibiotics measured in river sediments, liquid manure, and farmed soil suggests that these environments are likely to be hot spots for the development of resistance. Both explicit consideration of antibiotic resistance in the environmental risk assessment of antibiotics and efforts to reduce the input of antibiotics into the environment—by limiting use and/or improving the treatment of liquid and solid wastes—are crucial to maintaining background resistance levels.