The highly evolvable nature of the antibiotic efflux protein TolC limits use of phages and bacterial toxins as antibacterial agents

Bacteriophages and bacterial toxins are promising antibacterial agents to treat infections caused by multidrug resistant (MDR) bacteria. In fact, bacteriophages have recently been successfully used to treat life-threatening infections caused by MDR bacteria [1–3]. One potential problem with using these antibacterial agents is the evolution of resistance against them in the long term. Here, we studied the fitness landscape of the Escherichia coli TolC protein, an outer membrane protein that is exploited by a pore forming toxin called colicin E1 and by TLS-phage [4, 5]. By systematically assessing the distribution of fitness effects (DFEs) of ~9,000 single amino acid replacements in TolC using either positive (antibiotics and bile salts) or negative (colicin E1 and TLS-phage) selection pressures, we quantified evolvability of the TolC. We demonstrated that the TolC is highly optimized for the efflux of antibiotics and bile salts. In contrast, under colicin E1 and TLS phage selection, TolC sequence is very sensitive to mutation. Our findings suggest that TolC is a highly evolvable target limiting the potential clinical use of bacteriophages and bacterial toxins.


Creation of a TolC mutant library 91
We measured the evolvability of TolC by quantifying DFEs of all possible single amino 92 acid replacements in the presence of four physiological stress factors: antibiotics 93 (piperacillin-tazobactam), bile salts, colicin E1, and TLS phage ( Figure 1C-H). First, we 94 generated a tolC deletion strain (E. coli-∆tolC, Methods) which became more sensitive 95 to both antibiotics and bile salts (Figures S1 and S2) relative to its wild-type parent 96 strain (BW25113) [22,23]. The tolC deletion strain was also more resistant to both 97 colicin E1 and TLS phage relative to its wild-type parent ( Figure S1). We reintroduced 98 the tolC gene into this strain using a plasmid that has a constitutively active promoter 99 (pSF-OXB14, Oxford Genetics) and rescued both the antibiotic and bile salt resistance 100 and the colicin E1 and TLS phage sensitivity of the E. coli-∆tolC strain (Figure S1 and 101 Figure S2). We mutated all residues except the start codon (471 residues in the mature 102 TolC protein, and the 21 residue-long signal peptide) of TolC and generated a pool of 103 ~9,841 (492 sites x 20 aa and a stop codon) mutants ( Figure 1C, Figure S3). We cloned 104 the mutated tolC genes into the pSF-OXB14 plasmid and then transformed the E. coli-105 ∆ tolC strain with this pool of plasmids carrying mutated tolC genes. We randomly 106 selected 30 amino acid positions from our library and using Sanger sequencing, 107 confirmed that all 30 of the mutated sites were randomized and these tolC alleles did 108 not have unintended mutations at other sites (Figure S3). For amplicon sequencing, we 109 pooled mutants into five sub-libraries and carried out parallel selection and sequencing 110 experiments (Methods). We deep-sequenced the tolC genes in each sub-library by 111 utilizing the Illumina MiSeq platform and verified that 98.9% of possible amino acid 112 replacements in the mutant library yielded at least 10 counts when sequenced (Figure 113 1D-E), with ~1,800 reads per residue or an average of ~90 reads per amino acid 114 replacement ( Figure 1D-E). We also confirmed that frequencies of the mutations in the 115 tolC library did not change significantly when the library was grown in growth media 116 without selection (Figure 1F We measured fitness effects of TolC mutations under selection using a liquid based 120 assay ( Figure 1C). In brief, we grew mutant libraries in growth media to saturation, 121 diluted them to an OD600 of 0.001, and then grew these cultures in the presence of one 122 of the four selection factors for three hours. Cells were then washed and grown in 123 nonselective media for six hours. Finally, we harvested plasmids carrying tolC mutants 124 and performed amplicon sequencing to count the surviving tolC variants(Methods). The 125 duration of selection and recovery periods were optimized to maximize the dynamic 126 range of the measurements and to minimize the chances of losing some alleles during 127 plasmid harvesting (Methods, Figure S1 and S4). Of note, all concentrations used in 128 these assays were above the minimum concentrations sufficient to kill wild-type E. coli, 129 except the bile salts. The maximum soluble amount (50 mg/ml) of bile salts in our 130 selection experiments inhibited growth of the E. coli-∆tolC strain but not the wild-type E. 131 coli strain ( Figure S2). A control experiment with no selection was performed in parallel, 132 to decouple fitness effects due to growth defects. 133 For calculating fitness, we determined the enrichment of each mutation by comparing 134 mutation frequencies with and without selection (݁ represents the enrichment, and ݂ the frequency, of mutation ݅ , Figure 1G). We used the 136 average fitness of synonymous wild-type mutations as a reference point for defining 137 relative fitness values (s) of each mutation with respect to the wildtype (WT) TolC 138 Figure 1G, green bins). As a control, we compared the 139 fitness effects of early stop codons ( Figure 1G, pink bins) with the E. coli-∆tolC strain 140 supplemented with the tolC gene ( Figure 1G, green bins) and confirmed that the results 141 we obtained using our sequencing-based assay matched our observations in batch 142 culture ( Figure S4) both qualitatively and quantitatively. By comparing the enrichments 143 of mutations in the absence of selection (Methods) relative to the frequencies of 144 mutations in the library before any growth or selection, we confirmed that TolC 145 mutations did not have significant fitness effects in the absence of selection ( Figure 1F, 146 H). 147 Figure 2A-B shows the fitness effects for a subset of single amino acid replacements in 148 TolC in the presence of antibiotics (6 µg/ml), and TLS phage (2.5 x 10 8 pfu/ml). Figure  149 S5 summarizes the fitness effects of the entire mutation library under all four selection factors. We found that the fitness effects of mutations in the presence of antibiotics or 151 bile-salts were mostly neutral (Figure 2A and Figure S5A-B , white pixels) except a 152 group of mutations increasing sensitivity to antibiotics or bile salts (Figure 2A blue 153 pixels, Figure 2 C-D insets). Figure 2C and Figure S6 shows the corresponding DFEs. 154 When we repeated the same assay using 10 times lower dose of antibiotics (0.6 µg/ml, 155 which is still higher than the MIC value of piperacillin-tazobactam for wild-type E. coli, 156 Figure S2A) and bile salts (5 mg/ml), we saw that the DFE in bile salt selection did not  Table S1). We repeated these measurements 169 using ten times higher concentrations of colicin E1 and TLS phage and showed that the 170 DFEs under these conditions were still wide ( Figure 2F has the potential to evolve resistance and resides at a sub-optimal fitness state as the 173 TolC sequence is very sensitive to mutations. 174

Relationship between strength of selection and fitness effects 176
We measured fitness effects of TolC mutations using different doses of colicin E1 in 177 order to measure the relationship between mutational sensitivity and selection strength. 178 In these experiments, we used increasing concentrations of colicin E1 (0, 5pM, 0.1 nM, 179 and 2 nM, Figures 3A-E). In addition, we measured fitness effects of TolC mutations in 180 the presence of TLS phage particles ( Figure 3F) and bile salts ( Figure S6C). These measurements were done using the Illumina NovaSeq platform and yielded nearly 182 hundred-fold higher number of reads compared to the MiSeq platform. As the NovaSeq 183 platform provided large number of sequencing reads, we did not observe any extinct 184 mutations and we were able to quantify fitness values with greater confidence. We  When we excluded stop codon mutations, there were still many (372 mutations 197 spanning 168 residues for colicin E1 selection, 408 mutations spanning 184 residues for 198 TLS phage selection) resistance-conferring mutations suggesting that the TolC 199 sequence was only one mutation away from developing resistance to colicin E1 or TLS 200 phages ( Figure 3G). Using phages or colicin E1 in combination with antibiotics may 201 potentially reduce the rate of evolution to some extent as early stop codon mutations will 202 be eliminated by the use of antibiotics. However, many resistance conferring mutations 203 will still be available and extended use of phages or colicin E1 in clinical settings will 204 lead to selection of resistant TolC mutants, limiting the success of these therapies in the 205 The average fitness effects of TolC mutations under bile salt and antibiotic selection 207 were both very small and weakly correlated ( Figure S6D surface of the beta-barrel domain, residues near the so-called "equatorial domain", and 220 residues in the periplasmic pore opening [24]. While positional sensitivity to TLS and 221 colicin E1 selection showed strong overlap, there were several residues that were more  Looking at individual mutations, rather than the mean effects averaged over each 230 position, provided additional insight. We considered mutations that had fitness effects 231 larger than three standard deviations from the mean (Methods) and excluded early stop 232 codon mutations that were equivalent to loss of the tolC gene. For this analysis, we also 233 excluded mutations that did not have consistent fitness effects between experimental 234 replicates (Methods). We grouped TolC mutations that passed these criteria as 235 summarized and highlighted mutated residues on the TolC monomer ( Figure S7). 236 Interestingly, many of these residues clustered together suggesting that they might 237 induce similar changes on the TolC structure due to their physical proximity. We found a 238 cluster of residues that made E. coli cells resistant (Figure S7 :kan E. coli strain following the protocol in reference [22]. This strain is referred as the Δ tolC strain throughout the manuscript. We whole-genome sequenced both the wild-type (BW25113) and the Δ tolC E. coli strains and confirmed that no other mutations besides the tolC deletion were present in the Δ tolC strain.
Saturation mutagenesis assay for the tolC gene pSF-Oxb14 plasmid was obtained from Oxford Genetics (OGS557, Sigma). This plasmid contained a kanamycin resistance cassette and an Oxb14 constitutively open promoter region. The tolC gene was PCR amplified from the BW25113 (wild-type) strain using 5'ATTCAAAGGAGGTACCCACCATGAAGAAATTGCTCCCCATTC-3' (forward), and 5'AGAAATCGATTGTATCAGTCTCAGTTACGGAAAGGGTTATGAC-3' (reverse) primers. It was then cloned into the pSF-Oxb14 plasmid using the NEBuilder HiFi DNA Assembly Kit (E5520, New England Biolabs), following the protocol provided by the manufacturer. Bold and underlined nucleotides in primer sequences overlap with the plasmid sequence. The integrated tolC gene was confirmed to have no mutations by Sanger sequencing.
Whole gene saturation mutagenesis was performed by two PCR reactions individually for each codon in the tolC gene, including the first 22 amino acid long signal sequence. First PCR reaction amplified a portion of the tolC gene in the pSF-Oxb14-tolC plasmid and randomized the targeted codon with a primer that contained a randomized NNS nucleotide sequence (N stands for A, C, G, or T nucleotides and S stands for G or C nucleotides) for the targeted codon (this PCR product is referred as insert). Second PCR reaction amplified the rest of the pSF-Oxb14-tolC plasmid (this PCR product is referred as backbone). Our custom software for designing mutagenesis primers is available at https://github.com/ytalhatamer/DMS_PrimerDesignTool. Inserts were cloned onto the backbones using the NEBuilder HiFi DNA Assembly Kit (E5520, New England Biolabs and assembled plasmids were transformed into NEB-5-alpha (C2987, New England Biolabs) cells. Plasmid extraction from these cells was done using Nucleospin Plasmid kit (740588, Macharey-Nagel). As this assay produced libraries per each residue, plasmid concentrations were measured and then equimolar amounts of each library were pooled into five sublibraries for 2x250bp paired-end MiSeq Finally, these pooled sublibraries were transformed into ∆ tolC strain for selection experiments. All growth and selection assays with the library were done using 50 µg/ml kanamycin in minimal M9 media.

Colicin E1 purification
A colicin E1 expression vector with IPTG inducible T7 polymerase promoter was kindly provided by Dr. William A. Cramer (Purdue University). Only Colicin E1 was amplified and put back to an empty pET24a plasmid to remove immunity protein. Plasmids were then transformed into BL21-DE3 cells for expression and purification. Cells were grown in TB broth media and colicin E1 was purified first with a size exclusion chromatography. Elutes corresponding to the size of Colicin E1 (~57 kDa) were further purified using a cation exchange chromatography in Sodium borate buffer with a salt gradient of 0-0.3M (NaCl). All fractions are collected and analyzed by SDS-PAGE. Elutes with right band sizes pooled and concentrated using Amicon Centrifugal filters with 30K pore size (UFC803024, Milipore).

TLS Phage Harvesting
TLS phage strain was kindly provided by Dr. Joe Fralick (Texas Tech University). Phage propagation and purification were done following the protocol described in [25]. Briefly, overnight grown bacterial cells were diluted hundred times in 100 mL of LB medium with 5 mM CaCl 2 and incubated 2-3 hours till optical density reached 0.4-0.6. Phage particles were added to the culture and the culture was shaken (at 37˚C) until the culture became optically clear. Cell lysates were spun down in 50 mL falcon tubes at 4000 x g and for 20 minutes. Supernatant was filter sterilized using 0.22 µm filters. Chloroform was added to the filtered phage solution (10% v/v final chloroform concentration) and the solution was vortexed shortly and incubated at room temperature for 10 minutes. Finally, the phage lysate and chloroform mixture were centrifuged at 4000 x g for 5 minutes. Supernatant was removed, aliquoted, and stored at 4˚C.

Selection Assay
We used Piperacillin-Tazobactam (NDC 60505-0688-4, Apotex Corp), bile salts (B8756, Sigma-Aldrich), colicin E1, and TLS phage in selection assays. TolC mutant sub libraries were separately grown overnight in M9 minimal media supplemented with 50 µg/mL kanamycin. These cultures were diluted to the optical density of 0.001 in 10 mL of M9 minimal media supplemented with 50 µg/mL kanamycin (~5x10 6 cells). Selection agents were added to each sub library and cultures were incubated at 37˚C for three hours. All cultures were spun down at 7000 x g for 2 minutes and pellets were resuspended in fresh M9 minimal media supplemented with 50 µg/mL kanamycin. These cultures were then incubated at 37˚C with shaking for six hours.

Sequence analysis
Paired ended sequencing reads were first merged using the FLASh tool [26] (Customized parameters: -m 40 -M 100). Reads covering primers overlapping with the upstream and downstream of the amplified regions of tolC were excluded. Sequence reads were compared to the wild-type tolC sequence and mutations were listed. Sequence reads that had mutations in more than one residue were excluded from the analysis. Synonymous mutations yielding the same amino acid replacement were grouped together. Frequency of each mutation was calculated by dividing number of counts of for that mutation with number of all reads, including alleles with multiple . For calculating fitness, we first determined enrichment of each mutation by comparing mutation frequencies with and without selection (݁ ); e stands for enrichment and f stands for frequency, Figure 1G). Since randomized mutations of each residue created traceable mutations synonymous to the wild-type protein sequence, we were able to use average fitness of synonymous wildtype mutations for defining relative fitness values of each mutation with respect to the wild-type (WT) TolC sequence ‫ݏ(‬ ൌ ݁ െ ൏ ݁ ௐ ் ; s stands for fitness, Figure 1K, green bins). As a sanity check, we compared the fitness effects of early stop codons with the phenotype of the E. coli-∆tolC strain ( Figure 1G, pink bins) and confirmed that the results we obtained our sequencing-based assay matched our observations in batch culture ( Figure S4) both qualitatively and quantitatively. By comparing the enrichments of mutations in the absence of selection relative to the frequencies of mutations in the library before any growth or selection, we confirmed that TolC mutations did not have significant fitness effects in the absence of selection ( Figure 1F, H).Our source code for data analysis is available at https://github.com/ytalhatamer/DMS_DataAnalysis.   tolC strains without selection pressure. Dark green colored lines represent growth curves of the wild type (BW25113) E. coli strain. Dark red colored lines represent growth curves of the BW25113 E. coli strain with tolC gene deletion (∆tolC). In our fitness assays, we used a duration of three hours for selection (vertical grey dashed line) in order to maximize the fitness difference between the wild type E. coli and E. coli:∆tolC. Horizontal gray dotted line represents the detection limit of the spectrophotometer used (OD 600 : 0.007).  Figure S4. Effects of antibiotic, colicin-E1 and TLS phage selections are quantified with an optical density based assay for wild-type E. coli (green) and E. coli:∆tolC (red). We plotted cell densities with and without selection. (A) Growth values at two concentrations of antibiotics, colicin-E1 and TLS phage were tested. No growth has been observed for ∆ tolC strain in Antibiotics (6 µg/mL) selection and for wild-type strain for colicin-E1 (20nM) selection which are represented as not determined (N.D. with respective colors). Detection limit for spectrophotometer plotted as horizontal gray dotted line. (B) Growth is measured in the presence of antibiotics (piperacillintazobactam) and colicin-E1 for the E. coli:∆tolC strain supplemented with an empty plasmid (red) and with a plasmid carrying the wild type tolC gene (green).    (Table S2) (A) Mutations in pink colored residues confer resistance to colicin-E1 without causing significant fitness changes under other selection conditions. (B) Mutations in pink colored residues confer resistance to TLS phage without causing significant fitness changes under other selection conditions. (C) Mutations in red colored residues increase resistance to both colicin-E1 and TLS phage without disrupting efflux of antibiotics. (D) Mutations in blue colored residues increase sensitivity to both colicin-E1 and TLS phage without disrupting efflux of antibiotics. (E) Mutations in cyan colored residues increase resistance to both colicin-E1 and TLS phage and disrupt efflux of antibiotics. These mutations likely cause misfolding of TolC or blockage of the TolC channel as their fitness effects are reminiscent of the loss of the tolC gene. Table S1: Mean and standard deviation values for DFEs under all selection conditions: antibiotics (0.6 and 6 µg/mL), bile salts (5 and 50 mg/mL), colicin-E1 (2 and 20nM), and TLS phage (2.5x10 8 and 2.5x10 9 pfu/mL). Table S2: Fitness values of mutations represented in Figure S7 are tabulated. There are six sheets in the excel file. First sheet shows the consistent data that has similar fitness values in two different experiments. Remaining five sheets show fitness values for the mutations represented in Figure S7A-E.