To re-create satellite colony forming conditions we placed isogenic XL1-Blue E. coli, from a single colony, on agar plates with several different low concentrations of amp. Under our laboratory conditions the MIC (minimal inhibitory concentration, giving no detectable growth in liquid culture following an overnight incubation) was 2.5 μg/ml amp. We observed, however, that there was significant cell death even at sub-MIC antibiotic concentrations. For example, only 21% (+/-3%) of cells survived and formed colonies on plates with 1 μg/ml amp.
This simple experiment demonstrates the existence of selectable phenotypic variation in an isogenic population. A significant fraction of cells survived, while most died, similar to what is seen in the formation of satellite colonies. The key question addressed in this study is why, within a population of E. coli, will some individuals survive to form colonies while genetically identical neighbors die?
It is important to note that percentage of cells surviving 1 μg/ml amp, at about 20%, is too high to be accounted for by the occurrence of random spontaneous mutations. Frequencies of stable mutation to even low levels of antibiotics are far lower than this. For example, the rate of selection of stable mutant low-level quinolone resistant mutants in P. aeruginosa
ranges from 1.2 × 10-6
to 4 × 10-10
, depending on the concentration of quinolone used [1
]. While it is true that the frequency of mutation to antibiotic resistance is dependent on many variables, including the number of target genes, and the numbers of positions within those genes that can be mutated to confer resistance, nevertheless the observed high rate of 20% argues very strongly against an explanation based on stable DNA mutations.
As expected, the percent survivors dropped dramatically as the amp levels increased, with only approximately 1/105 cells surviving on the MIC concentration of 2.5 μg/ml amp (Fig. ). Survival rates were routinely determined by performing parallel serial dilution platings on LB agar with and without antibiotic.
Figure 1 E. coli ampicillin survival curves. All E. coli are isogenic, recently derived from a single colony. Triangles show survival rates for cells previously grown in the absence of ampicillin. Resistance rates were ramped higher by prior exposure to ampicillin. (more ...)
We next asked if the survival rate could be ramped higher by successive exposures to increasing concentrations of amp. We found that this was indeed the case. For example, cells surviving an initial exposure to 1 μg/ml amp were then found to show over 50% survival on a second plating on 1 μg/ml amp, compared to the original 20%, and cells surviving 2.5 μg/ml amp gave survival rates of ~3% when re-plated on 2.5 μg/ml, an improvement of over 1000 fold compared to the 1/105 for cells not previously exposed to amp (Fig. ). Through such successive exposures to increasing antibiotic concentrations it was possible to evolve E. coli that survived on amp concentrations as high as 10 μg/ml, yet showed very high reversion rates, indicating the absence of stable mutations conferring resistance.
The high frequency of survival on low antibiotic concentrations suggested an epigenetics based resistance. Perhaps some cells within the population were by chance over expressing one or more genes that conferred low level amp resistance. Epigenetic inheritance of this amp resistance gene expression pattern would allow such a cell to divide and form a colony, while neighboring cells without this advantage would die.
To test this model we examined reversion rates. Amp resistance due to changes in DNA sequence would be extremely stable. Reversion to antibiotic sensitivity would generally require back mutation of the precise base originally altered, a very infrequent event, estimated to be on the order of 10-9
or less [2
]. Epigenetic inheritance, on the other hand, is based on more unstable processes, such as maintenance of certain chromatin configurations and/or DNA methylation states, and would give much higher reversion rates.
We observed that the high frequency low level amp resistance was indeed extremely unstable, showing very high reversion rates to antibiotic sensitivity.
For example, over 95% of cells taken from a single colony growing on a 2.5 μg/ml amp plate were unable to grow when immediately transferred to a new plate with the same 2.5 μg/ml amp concentration (Fig. ). This immediate reversion rate of over 95% to amp sensitivity argues strongly in favor of an epigenetic inheritance based mechanism, and against the involvement of DNA mutation.
These results are quite striking. When the cells of a colony on a plate with 2.5 μg/ml amp were quickly resuspended in LB broth and immediately re-plated in parallel on plates with amp (2.5 μg/ml amp again), and without amp to determine viable cell count, it was observed that over 95% of the cells placed on the amp plate failed to form a colony. That is, over 95% of the cells reverted to antibiotic sensitivity over the brief time span of the experiment. It is clear from these results that this antibiotic resistant state is extremely unstable.
It is also interesting to note that this instability disappeared as the time under selection, and antibiotic concentrations were increased. The cells surviving 30 μg/ml amp showed variable but generally low reversion rates, suggesting that most of these cells had now acquired stable DNA mutations conferring amp resistance.
A "whole plate" selection scheme was primarily used (see Methods). At each step of antibiotic selection we pooled a few hundred colonies from a single plate, by adding five ml of LB broth to the surface of the plate, scraping the cells into the LB broth, and resuspending the cells by pipeting. The cells were then immediately re-plated, in serial dilutions, in the absence of antibiotic to determine viable cell count, and on plates with different antibiotic concentrations, to create survival curves, determine reversion rates, and to select for increased antibiotic resistance.
Variable Gene Expression
To further investigate the basis of the antibiotic resistance we examined gene expression patterns in multiple E. coli populations surviving selection with 0, 1, 2.5, 5, 10 or 30 μg/ml amp, using a total of 35 Affymetrix E. coli version 2.0 microarrays (see Methods). Altered gene expression patterns in resistant cells might reveal epigenetic based survival mechanisms.
Biological replicates were used for the microarray analysis, with each microarray representing an independently generated biological sample. The data was first examined with GeneSpring software, performing an ANOVA analysis, producing a list of differentially expressed genes. A heat map provides a visual display of the results (Fig. ). There is some scatter in the data, perhaps reflecting the inherent variability in the selection response and the possible existence of multiple resistance mechanisms.
Figure 2 Heat map showing changing gene expression patterns in ampicillin resistant E. coli. Each column represents one population grown in the absence of amp (C), or in the concentration of amp shown at the bottom. Each row shows the expression level across populations (more ...)
Of particular interest, the list of genes with differential expression included an endogenous E. coli
gene with known β-lactamase activity. This gene, AmpC, is present in most laboratory strains of E. coli
. A close relative of E. coli
, E. cloacae
, carries an AmpR gene important in the induction and repression of AmpC. It has been shown that the AmpR gene was deleted from the E. coli
genome following the divergence of E. coli
and E. cloacae
from their common ancestor [3
]. This results in very low-level constitutive expression of AmpC in E. coli
, allowing the use of amp selection in cloning experiments, for example. Amp resistance in E. coli
has previously been associated with stable DNA mutations resulting in novel AmpC promoters, weakened attenuators or AmpC gene duplications [4
]. Of interest, however, we observed elevated expression of AmpC not only in the stable amp resistant populations (resistant to 30 μg/ml amp), but also in many of the unstable amp resistant populations (resistant to 1, 2.5, 5 and 10 μg/ml amp) (Fig. ).
Figure 3 Expression levels of endogenous β-lactamase gene in E. coli surviving selection at different ampicillin concentrations. Transcript abundances were determined using a total of 29 Affymetrix E. coli 2.0 oligonucleotide microarrays. Expression levels (more ...)
These results suggest that AmpC can contribute to amp resistance through three distinct mechanisms. In E. cloacae the presence of amp can induce elevated AmpC expression through the action of the AmpR gene. That is, the combination of the AmpR and AmpC genes provide a survival mechanism that is induced in the presence of amp. In E. coli, without the AmpR gene, DNA mutations can result in elevated AmpC expression and stable amp resistance, as shown previously. In addition, in this report, we show that unstable resistance to low levels of amp in E. coli can be mediated through epigenetic events, likely stochastic in nature, yet semi-stable, that result in elevated AmpC expression.
To test for functionality in conferring amp resistance we picked a total of eight genes with elevated expression following amp selection, after examining the microarray data, and subcloned their coding sequences into an expression vector plasmid, which was then introduced into E. coli. The over-expression of the endogenous E. coli AmpC gene did indeed confer dramatically increased amp resistance, as expected, allowing all cells to survive on up to 30 μg/ml amp, but not 100 μg/ml. None of the other genes tested (see Methods) gave significant survival benefits. These results argue that the observed increase in expression of the endogenous AmpC gene was responsible, at least in some cases, for the observed amp resistance.
It is not surprising that many of the genes with elevated expression following amp selection were unable to confer amp resistance. It has been previously shown that very low, subinhibitory concentrations of antibiotics can cause significant changes in bacterial gene expression patterns. For example, it was shown in Salmonella typhimurium
, using a promoter-Lux reporter library, that many promoters responded to low levels of erythromycin and rifampicin, below the MIC [9
In an attempt to remove some of the induced genes unrelated to antibiotic resistance from the list of differentially expressed genes we re-analyzed the microarray data, only this time including the data from E. coli exposed to very low levels of amp (1 μg/ml) in the control group. This more specifically looked for genes that showed increasing change in expression concordant with survival in the presence of increasing levels of amp. A pairwise t-test was performed using GeneSifter software. The top five genes on the resulting list, ranked by fold change, were as follows. (1.) Glutamate decarboxylase (GadA), up 7.2 fold, p = 0.021. (2.) The second gene on the list encodes a separate glutamate decarboxylase (GadB) up 6.1 fold, p = 0.014. It is interesting that the top two genes on the list are located at very different genomic positions, yet encode isozymes very similar in sequence and function. (3.) The third gene, GadC, encodes an APC transporter that mediates export of gamma-aminobutyrate in exchange for glutamic acid. The GadC gene is located on the same operon as GadB, providing a measure of cross-validation. (4.) The fourth gene, AmpC, was up-regulated 4.4 fold, with a p value of 0.0034 (the lowest p value). (5.) The fifth gene, hdeB, was up 4.0 fold, p = 0.011, and encodes an acid stress chaperone.
It is quite interesting that the two genes showing the greatest upregulation in more amp resistant cells encode isozymes that both decarboxylate glutamate (Fig. ). There are reasons to suspect that this overexpression of glutamate decarboxylase could be functionally related to amp resistance. Glutamate decarboxylase removes a carboxyl group from glutamic acid (which is neuro-excitatory in mammals), producing gamma-aminobutyric acid (GABA, neuro-inhibitory in mammals). There is a surprising connection between penicillin and glutamate-GABA in mammals. Penicillin is a GABAA
receptor blocker and can be used to induce convulsions in animal models of epilepsy [10
]. This ability of penicillin to interact with a GABA receptor suggests a steric similarity between penicillin and glutamate-GABA. Of interest, both penicillin, (as well as ampicilliin), and glutamate do carry a carboxyl group in a comparable chemical setting (Fig. ). Indeed the structural similarity between a region of ampicillin (including the carboxyl group) and the backbone of a peptide chain have been used to explain the mechanism of action of β-lactam antibiotics. This suggests that glutamate decarboxylase might be capable of removing the carboxyl group from ampicillin, thereby inactivating it.
Elevated expression of GadA and GadB genes in ampicillin resistant populations. Both genes encoding the two glutamate decarboxylase isozymes show significantly elevated expression in ampicillin resistant E. coli populations.
Structures of glutamate and ampicillin. Similarities between the lactam ring and the peptide backbone have been proposed to be responsible for the antibiotic function of ampicillin.
Previous work also suggests another possible role for glutamate decarboxylase in antibiotic resistance. The decarboxylation reaction consumes a proton, contributing to membrane potential difference, which in turn is used by the AcrAB multi-drug efflux pump [13
]. The overexpression of the GadA and GadB genes may therefore provide increased power to the AcrAB pump.
To test GadA for possible function in providing amp resistance we made a GadA expression plasmid and introduced it into the XL1-Blue E. coli. We observed that the XL1-Blue E. coli with the GadA expression construct showed somewhat improved survival rates, five fold higher colony formation on 1.75 μg/ml amp compared to cells without (P = 0.014). These results suggest that over expression of GadA, and likely the isozyme GadB, can contribute to survival under amp selection conditions. Therefore the increased expression of GadA and GadB could be the result of stochastic variation in expression levels of these genes coupled with amp selection for cells with higher expression. It is also interesting to note the possibility that just as penicillin can perturb glutamate signaling in the mammalian brain, perhaps through its structural similarity to glutamate, penicillin (or ampicillin) can induce GadA and GadB expression in E. coli.
Nalidixic Acid and Tetracycline
To study the epigenetic based evolution of antibiotic resistance further, and in particular to determine if this was a general phenomenon or restricted to ampicillin, we performed similar selection experiments using two additional antibiotics, nalidixic acid and tetracycline. These three antibiotics exhibit distinct modes of action, with ampicillin disrupting cell wall biosynthesis, nalidixic acid inhibiting DNA synthesis and tetracycline blocking protein synthesis.
Interestingly, the results for the nalidixic acid and tetracycline selection experiments closely mirrored those observed using ampicillin, showing the generality of epigenetic inheritance mediated evolution of antibiotic resistance. For example, again starting with isogenic XL1-Blue E. coli, we established a nalidixic acid survival curve, finding survival rates that varied from about 40% on 20 μg/ml nalidixic acid to about 1 in 100,000 on 80 μg/ml nalidixic acid (Fig. ). For reference, the nalidixic acid MIC for XL1-Blue E. coli was 40 μg/ml. As for ampicillin, the survival rates under low concentration antibiotic concentration selection were too high to be accounted for by spontaneous mutation. This was confirmed by testing reversion rates, which with rare exception were again very high, indicating that the resistance was not the result of a stable change in DNA sequence (Fig. ).
Figure 6 E. coli nalidixic acid survival curves. As observed for ampicillin, the survival rates at low nalidixic acid concentrations are too high to be accounted for by spontaneous mutation. Cells surviving prior antibiotic exposure show dramatically improved (more ...)
The XL1-Blue E. coli carry a stable mutation in the gyrase gene (gyrA96), conferring significantly greater nal resistance than found in wild type E. coli, which typically would not survive even 10 μg/ml nalidixic acid. In this series of experiments, therefore, we started with E. coli showing moderate levels of nal resistance, and examined the evolution of increased resistance to still higher antibiotic concentrations.
Once again the resistance rates could be elevated by successive antibiotic selection. For example, upon initial exposure to 40 μg/ml nalidixic acid approximately 1/1,000 cells survived to form a colony. When cells from a 40 μg/ml nalidixic acid plate, however, were re-plated on another plate with 40 μg/ml nalidixic acid, then about 20% survived to form colonies, showing a dramatic improvement in survival rate. This also illustrates the high reversion rate, as 80% of cells do not survive a re-plating on 40 μg/ml nalidixic acid.
The ability to ramp antibiotic resistance higher by successive exposures was also illustrated by the survival rates on higher concentrations of nalidixic acid. For example, upon initial exposure to 80 μg/ml nalidixic acid only about one cell in 100,000 survived. But if cells surviving exposure to 40 μg/ml were subsequently selected at 80 μg/ml then almost 1 cell in 100 survived, approaching a 1,000 fold improvement (Fig. ).
For the tetracycline selection experiments we used a different strain of E. coli [XL1-Blue (MRF')], as the standard XL1-Blue cells carry genes conferring strong tet resistance. The XL1-Blue (MRF') cells (Stratagene) are closely related to XL1-Blue cells (see Methods), only with kanamycin resistance instead of tetracycline resistance. The results of tetracycline selection using single colony derived isogenic XL1-Blue (MRF') cells showed a similar pattern to that observed for both ampicillin and nalidixic acid. The tetracycline MIC for XL1-Blue (MRF') E. coli was 1.0 μg/ml. Once again, resistance and reversion rates were far too high to be accounted for by stable DNA mutations, and once again it was possible to ramp up resistance rates by successive exposures to increasing concentrations of antibiotic (Fig. ).
E. coli tetracycline survival curves. Survival ratios for E. coli not previously exposed to tetracycline (triangles), and for E. coli previously surviving exposure to 2 μg/ml tetracycline (circles).
The observed semi-stable epigenetic inheritance could be mediated by DNA methylation [14
], chromatin modifications, superhelical domain configuration [15
], or perhaps other mechanisms. DNA methylation in particular provides a plausible mechanism for the observed metastable antibiotic resistance. The E. coli
deoxyadenosine methyltransferase (DAM), for example, methylates the adenine of the GATC sequence. This sequence occurs approximately 18,000 times in the E. coli
genome, and the two copies of this palindromic sequence opposite each other on DNA are generally both methylated. Following DNA replication, however, the DNA is transiently hemimethylated, and the new strand is then methylated by DAM.
For a few sites in the E. coli
genome, however, the methylation status of the DAM target, GATC, is variable and can impact gene expression. Of particular interest, the flu
gene is a metastable locus encoding the Ag43 protein, which is an outer membrane protein promoting cell-cell aggregation important in biofilm production. Phase variation, in both directions (Ag43+ to Ag43- and vice versa), is metastable and occurs with a frequency of approximately 10-3
]. DAM and OxyR repressor compete for binding to sequences in the flu
promoter, with OxyR binding resulting in repression and blocking of methylation, while DAM methylation is required for full gene activation [17
A previous microarray analysis of gene expression patterns associated with biofilms showed a strong upregulation of the flu
]. Indeed, it was shown that E. coli
with a mutation of this gene were unable to form biofilms [18
]. Because of the metastable binary expression of this gene, and its strong association with biofilms, and the known multidrug antibiotic resistance of biofilm E. coli
compared to E. coli
undergoing planktonic growth, we were particularly interested in the expression levels of the flu
gene in the antibiotic resistant E. coli
in this study. Interestingly, we observed no change in flu
gene expression in the amp resistant cells, arguing against the involvement of a biofilm type genetic program in the antibiotic resistance.
Although the elevated expression of the flu gene, and consequent biofilm production, did not appear associated with the antibiotic resistance described in this report, nevertheless the DNA methylation mediated metastable expression of the flu gene provided a useful model for possible regulation of other genes that might indeed be responsible for the observed selectable antibiotic resistance.
Only a small number of genes were shown to be altered in expression in Dam methylation mutant E. coli
]. The 18 genes with over 2 fold change in DAM mutant cells included representatives of the Csg and Mar operons, which were both down regulated in the absence of DAM methylation.
To test the possible function of DNA methylation in the observed evolution of antibiotic resistance we examined the ER2925 strain of E. coli K12, which is deficient for both DAM and DCM dependent DNA methylation. We generated both DAM positive and DCM positive variants of the ER2925 cells by making DAM and DCM expression plasmids, allowing a direct comparison of cells with and without these specific DNA methylases. We observed that DCM had no apparent effect on antibiotic survival, while DAM methylase did improve the survival rate by a factor of five on nalidixic acid (40 μg/ml).