Model of artificially evolved aminoglycoside phosphotransferase (3 9 )-IIIa, based upon crystal structure (1L8T) of the wildtype enzyme [ 23 ] rendered in PyMOL. The putative beneficial mutations are colored orange, the catalytic D190 residue is green), adenosine diphosphate is yellow and magnesium is pink. Kanamycin is blue, while the ‘‘extra’’ moiety that differentiates amikacin from kanamycin is red. doi:10.1371/journal.pone.0076687.g001 

Contexts in source publication

Context 1

... form [21]. The evolved his 6 -4.1 APH(3 9 )-IIIa exhibited substantially higher k cat and lower K M in reactions with amikacin (and detectable substrate inhibition, K i . 2 mM), when compared to the his 6 -wildtype APH(3 9 )-IIIa. The second order rate constant ( k cat / K M ) was similar to that of the wild-type enzyme in reactions with kanamycin. The evolved enzyme retains some catalytic activity in reactions with kanamycin, but its steady state enzyme kinetics could not be fit to the Michaelis-Menten equation or to any substrate inhibition model [34]. The interaction between his 6 -4.1 APH(3 9 )-IIIa and kanamycin is likely complex; we speculate that multiple non-productive binding modes compete. We noticed during our directed evolution experiments that some of the selected clones formed fewer colonies than did the isogenic ancestral strain. Fresh Inv a F’ cells were transformed with aph(3 9 )-IIIa -pQBAV3c plasmids encoding the wild-type, evolved 2.3, 3.1 or 4.1 alleles (or pBC or pACYC Duet as controls). The transformants were propagated in parallel under non-selective conditions (liquid LB supplemented with 34 micrograms/mL chloramphenicol or 50 micrograms/mL kanamycin). The optical density (600 nm) of each culture was measured; the cultures were serially diluted, then spread on LB agar plates supplemented with either chloramphenicol or kanamycin. We observed significant and reproducible differences among isogenic transformants in growth rates (during log phase in liquid culture) and colony forming ability (Table 5). Two to eight-fold decreases in fecundity would almost certainly be decisive in nature, so we consider them worthy of further study. We first wondered whether the proliferation of untransformed cells in liquid culture, those that absorb light at OD 600 but fail to form colonies on agar plates containing chloramphenicol, could explain the observed differences in fitness. If mutations in the aph(3 9 )-IIIa gene could affect plasmid stability, a disparity in colony forming ability would be revealed by growing E. coli Inv a F’ containing pQBAV3c on plates both with and without selection for plasmid retention. We observed little or no plasmid loss in cells expressing the wild-type or 4.1 variants of APH(3 9 )-IIIa (Table 6). This result suggests that the fitness differences we observed (Table 5) are consequences of sequence differences in the aph(3 9 )-IIIa alleles themselves, rather than of indirect effects upon plasmid stability. We wondered whether the fitness costs correlate with improve- ments in activity against amikacin. We already have circumstantial evidence against this hypothesis. The intermediate 2.3 aph(3 9 )-IIIa pQBAV3c imparts decreased fitness (relative to isogenic cells carrying its wild-type aph(3 9 )-IIIa -pQBAV3c ancestor), while mutant 3.1 did not (Table 5). Furthermore, the fitness associated with that APH(3 9 )-IIIa variants that we tested was unaffected by kanamycin (Table 5), suggesting that active-site occupancy apparently does not affect the fitness. To investigate our hypothesis more decisively, a single point mutation was made in the 190th residue of the wild-type and 4.1 APH(3 9 )-IIIa variants, changing the catalytic aspartic acid into alanine. This well characterized mutation has no significant effect on the structure or stability of APH(3 9 )-IIIa, but it abrogates detectable catalytic activity [20]. As expected, E. coli Inv a F’ expressing the D190A or 4.1 + D190A APH(3 9 )-IIIa failed to grow in LB medium supplemented with kanamycin (Table 5) or amikacin (data not shown). To our surprise, however, the fitness effects of the D190A mutation were context-dependent. Cells expressing the D190A APH(3 9 )-IIIa protein were significantly less fit than those expressing the wild-type protein under non-selective conditions. In contrast, cells expressing the 4.1 + D190A APH(3 9 )-IIIa were much fitter than isogenic cells expressing the 4.1 variant (Table 5). We don’t know whether the D190A and 4.1 proteins debilitate fitness through different biochemical mechanisms, or whether these protein variants act through a common non-catalytic mechanism that is sensitive to epistatic interactions among mutations. Many have observed that chromosomal mutations that impart resistance to other antibiotics come with a fitness cost. The population biology of these phenomena may be broadly similar [35,36,37,38,39], but the biochemical mechanisms are almost certainly idiosyncratic. The evolved 4.1 APH(3 9 )-IIIa contained nine amino acid replacements relative to its wild-type ancestor: E24V, I40T, R120K, C156R, K176R, S194R, I196F, Y219H and K255R. All but one, Y219H, first appeared in previous rounds of selection, suggesting that they are beneficial with respect to amikacin recognition. Two, namely I40T and S194R, are apparently beneficial in isolation. Amikacin is structurally identical to kanamycin, except that it contains an extra bulky modification (represented as red sticks in Figure 1) that creates a steric clash with the dynamic aminoglycoside binding loop (residues 147–170), at least in its kanamycin-binding conformation. The E157, N158 and E160 residues in that loop form hydrogen bonds with the amine groups in middle saccharide ring of kanamycin (including the one modified in amikacin) [23]. We therefore hypothesize that the C156R mutation increases the conformational flexibility of the loop, enabling it to accommodate amikacin. Two other mutations, Y219H and K255R, occur in alpha-helices that interact with the binding loop, and could therefore influence its conformation. The D190 residue, which we mutated, is in different active site loop (residues 188–195), and forms a hydrogen bond with the hydroxyl group that the enzyme later phosphorylates [23]. The S194R and I196F mutations in that loop could increase its conformational flexibility, so that amikacin can bind in an orientation different than that of kanamycin. The E24 residue, located in yet another active-site loop (residues 22–29), forms a hydrogen bond with neomycin B, but not with kanamycin. We hypothesize that the E24V mutation destabilizes an unproductive binding mode. Other mutations in the evolved 4.1 APH(3 9 )-IIIa, namely I40T, I120K and K176R, occurred in residues more distant from the active-site, so it is more difficult to speculate about their effects upon amikacin recognition. Most wild-type proteins are only marginally stable. Most amino acid changes are destabilizing, so the evolvability of most proteins is limited by conformational stability [40]. Mutations that alter the molecular recognition properties of an enzyme are particularly likely to be destabilizing [41]. Two active-site mutations that we observed, S194R and I196F, probably destabilize the active conformation by introducing new steric clashes with adjacent residues. The other seven amino acid changes occurred in surface residues, so their effects upon thermostability, if any, are less obvious. Global suppressor mutations, such as M182T in the TEM-1 beta-lactamase [16], can offset the destabilizing effects of beneficial mutations, but we made no deliberate effort to select for such mutations [42] nor did we see evidence for any. These hypotheses could be tested by calorimetric measurement of the thermodynamic parameters of the wild-type and mutant proteins. The aph(3 9 )-IIIa gene has appeared in clinical samples tested for resistance to kanamycin and other aminoglycosides [43,44], but not amikacin [45,46,47]. We showed here, however, that four rounds of mutation and selection were sufficient to direct the evolution of an APH(3 9 )-IIIa variant (4.1) that conferred resistance to 60 times higher concentrations of amikacin than did the wildtype. The MIC exceeded 1200 micrograms/mL, substantially higher than the highest serum level (35 micrograms/mL) recommended in humans [48]. The aminoglycoside modifying enzymes diverged to adapt to different aminoglycosides in nature [49], but no mutants of these enzymes have been identified in clinical isolates [19]. It is tempting to speculate that the fitness costs observed in our study, in combination with the plethora of existing aminoglycoside modifying enzymes that efficiently confer resistance to amikacin, including AAC(6 9 )-1ad, ANT(4 9 )-II, and APH(3 9 )-VI [45,46,47] collude to prevent the adaptive evolution of APH(3 9 )-III in the ...

Context 2

... density at 600 nm of each culture was measured in a plate reader (Biotek Synergy 2). Each culture was serially diluted three times, each time with 20 microliters of the sample being added to 980 microliters of LB broth; 50 microliters of the final serial dilution were spread on LB plates containing 34 micrograms/mL chloramphenicol or 50 micrograms/mL kanamycin. Plates were incubated at 37 u C overnight, and the visible colonies were counted. E. coli Inv a F’ containing the aph(3 9 )-IIIa -pQBAV3c plasmid with was grown overnight to saturation in 2 mL LB broth with 34 micrograms/mL chloramphenicol or 50 micrograms/mL kanamycin. Each culture was serially diluted twice, each time with 20 microliters of the sample being added to 980 microliters of LB broth. A 200 microliter sample of each culture was transferred into a clear, flat bottom 96 well plate. The plate was agitated at a medium speed at 37 u C for 24 hrs in a Biotek Synergy2 microtiter plate reader; the optical density at 600 nm was measured every 30 minutes. APH(3 9 )-IIIa is well-suited for our study of adaptive enzyme evolution because, as we explained above, its weak activities against a range of antibiotics enables facile selections in Escherichia coli [25]. We chose amikacin as the ‘‘novel’’ substrate because it is nearly identical in structure to its ‘‘native’’ (most reactive) substrate, kanamycin, from which it is chemically synthesized [33]. The aph(3 9 )-IIIa gene was amplified and randomly mutated in an error-prone PCR; the resulting library cloned into pQBAV3c [27], which also encodes chloramphenicol acetyltrans- ferase. E. coli strain Inv a F’ was transformed with the plasmid- borne library; 10,000 colony-forming units were spread on 10 6 (100 6 15 mm) Petri dishes containing LB agar supplemented with chloramphenicol and 18 micrograms/mL amikacin. Isogenic control cells transformed with the ancestral aph(3 9 )-IIIa -pQBAV3c plasmid do not grow under these conditions, but 43 colonies formed among the approximately 10,000 that expressed mutant APH(3 9 )-IIIa proteins. The 43 selected aph(3 9 )-IIIa alleles were sequenced (Table 2); most contained just 1 or 2 nucleotide mutations in the open reading frame (ORF) (27/43 mutants). Nearly all alleles also contained mutations in the upstream 5 9 region, including those without ORF mutations. Each mutant was restruck on fresh plates containing higher concentrations of amikacin (22–50 micrograms/mL) in order to measure the minimum inhibitory concentration (MIC, Table 2). The 43 variant aph(3 9 )-IIIa -pQBAV3c plasmids were pooled, along with the ancestral plasmid, and used as templates for staggered extension process (StEP) recombination [29]. The resulting recombinant library was ligated back into the pQBAV3c plasmid. E. coli Inv a F’ was transformed with the library and spread on LB agar plates supplemented with 80 micrograms/mL amikacin. Seven colonies out of approximately 10,000 formed under these more stringent conditions (Table 3). We sequenced these aph(3 9 )-IIIa alleles and found that this small population was dominated by two new mutations (I40T and D193N) and two others (S194R and K255R) that were selected in the first round. Most of the selected mutants unexpectedly contained only single amino acid changes (4/7 mutants). It is possible that the increase in amikacin resistance of these mutants in this round was due to mutations outside of the ORF; mutants 2.4 and 2.5 had the same ORF mutation as 1.18 but additional non-ORF mutations (most notably 2 58(T-C) and 2 35(T-C)). It is also possible that the amikacin resistance can increase from the elimination of slightly deleterious mutations; mutant 2.3 shares a mutation with mutants 1.27 and 1.33, but is missing some mutations unique to them. The seven selected aph(3 9 )-IIIa alleles were pooled, amplified, and mutated in an error-prone PCR. The mutated genes were cloned back into the pBAV3c plasmid. Inv a F’ cells were transformed with the cloned library, and spread on LB agar plates supplemented with 220 micrograms/mL amikacin. Four colonies (out of approximately 10,000 transformants) formed under these selection conditions. DNA sequencing of the four associated alleles showed that each contained the S194R and three of the four mutants (3.1, 3.3 and 3.4) also had a mutation in the 40th residue, although there were two different mutations (I40V and I40T) at this position. The D193N mutation, carried by two mutants from the second round of evolution went extinct in the third. The four selected plasmids, and their ancestor, which encodes the wild-type aph(3 9 )-IIIa gene, were pooled and used as templates for StEP recombination. The recombinant library was cloned and used to transform Inv a F’. The transformants were challenged with LB agar supplemented with 1200 micrograms/mL amikacin. One colony (out of approximately 10,000 transformants) formed. The DNA sequence of mutant 4.1 (E24V, I40T, R120K, C156R, K176R, S194R, I196F, Y219H, K255R; Figure 1) showed the persistence of three mutations that appeared in the first two rounds, I40T, S194R, and K255R (Table 2). We speculate upon the biochemical mechanisms of these adaptations in the Discussion section ( vide infra ). The wild-type and 4.1 aph(3 9 )-IIIa alleles were subcloned into a modified version of pET28a + that encodes the TEM-1 beta- lactamase in place of the usual aph(3 9 )-Ia [18]. The two recombinant plasmids were separately used to transform the E. coli production strain BL21(DE3). The proteins were over- expressed and purified by virtue of N-terminal hexahistidine tags encoded by the pET28a + vector. The kinetic parameters of the two enzyme variants in reactions with the ‘‘native’’ substrate (kanamycin) and ‘‘novel’’ substrate (amikacin) were measured with a coupled pyruvate kinase/lactate dehydrogenase assay (Table 4, Figure 2). The K M and k cat of the his-tagged enzyme, his 6 -APH(3 9 )-IIIa in reactions with kanamycin and amikacin were similar to the published values of the native untagged form [21]. The evolved his 6 -4.1 APH(3 9 )-IIIa exhibited substantially higher k cat and lower K M in reactions with amikacin (and detectable substrate inhibition, K i . 2 mM), when compared to the his 6 -wildtype APH(3 9 )-IIIa. The second order rate constant ( k cat / K M ) was similar to that of the wild-type enzyme in reactions with kanamycin. The evolved enzyme retains some catalytic activity in reactions with kanamycin, but its steady state enzyme kinetics could not be fit to the Michaelis-Menten equation or to any substrate inhibition model [34]. The interaction between his 6 -4.1 APH(3 9 )-IIIa and kanamycin is likely complex; we speculate that multiple non-productive binding modes compete. We noticed during our directed evolution experiments that some of the selected clones formed fewer colonies than did the isogenic ancestral strain. Fresh Inv a F’ cells were transformed with aph(3 9 )-IIIa -pQBAV3c plasmids encoding the wild-type, evolved 2.3, 3.1 or 4.1 alleles (or pBC or pACYC Duet as controls). The transformants were propagated in parallel under non-selective conditions (liquid LB supplemented with 34 micrograms/mL chloramphenicol or 50 micrograms/mL kanamycin). The optical density (600 nm) of each culture was measured; the cultures were serially diluted, then spread on LB agar plates supplemented with either chloramphenicol or kanamycin. We observed significant and reproducible differences among isogenic transformants in growth rates (during log phase in liquid culture) and colony forming ability (Table 5). Two to eight-fold decreases in fecundity would almost certainly be decisive in nature, so we consider them worthy of further study. We first wondered whether the proliferation of untransformed cells in liquid culture, those that absorb light at OD 600 but fail to form colonies on agar plates containing chloramphenicol, could explain the observed differences in fitness. If mutations in the aph(3 9 )-IIIa gene could affect plasmid stability, a disparity in colony forming ability would be revealed by growing E. coli Inv a F’ containing pQBAV3c on plates both with and without selection for plasmid retention. We observed little or no plasmid loss in cells expressing the wild-type or 4.1 variants of APH(3 9 )-IIIa (Table 6). This result suggests that the fitness differences we observed (Table 5) are consequences of sequence differences in the aph(3 9 )-IIIa alleles themselves, rather than of indirect effects upon plasmid stability. We wondered whether the fitness costs correlate with improve- ments in activity against amikacin. We already have circumstantial evidence against this hypothesis. The intermediate 2.3 aph(3 9 )-IIIa pQBAV3c imparts decreased fitness (relative to isogenic cells carrying its wild-type aph(3 9 )-IIIa -pQBAV3c ancestor), while mutant 3.1 did not (Table 5). Furthermore, the fitness associated with that APH(3 9 )-IIIa variants that we tested was unaffected by kanamycin (Table 5), suggesting that active-site occupancy apparently does not affect the fitness. To investigate our hypothesis more decisively, a single point mutation was made in the 190th residue of the wild-type and 4.1 APH(3 9 )-IIIa variants, changing the catalytic aspartic acid into alanine. This well characterized mutation has no significant effect on the structure or stability of APH(3 9 )-IIIa, but it abrogates detectable catalytic activity [20]. As expected, E. coli Inv a F’ expressing the D190A or 4.1 + D190A APH(3 9 )-IIIa failed to grow in LB medium supplemented with kanamycin (Table 5) or amikacin (data not shown). To our surprise, however, the fitness effects of the D190A mutation were context-dependent. Cells expressing the D190A APH(3 9 )-IIIa protein were significantly less fit than those expressing the wild-type protein under non-selective conditions. In contrast, cells expressing the 4.1 + D190A APH(3 9 )-IIIa were much fitter than isogenic cells expressing the 4.1 variant ...