Correlated transcriptional responses provide insights into the synergy mechanisms of the furazolidone, vancomycin and sodium deoxycholate triple combination in Escherichia coli

Effective therapeutic options are urgently needed to tackle antibiotic resistance. Furazolidone (FZ), vancomycin (VAN), and sodium deoxycholate (DOC) show promise as their combination can synergistically inhibit the growth of, and kill, multidrug-resistant Gram-negative bacteria that are classified as critical priority by the World Health Organization. Here, we investigated the mechanisms of action and synergy of this drug combination using a transcriptomics approach in the model bacterium Escherichia coli. We show that FZ and DOC elicit highly similar gene perturbations indicative of iron starvation, decreased respiration and metabolism, and translational stress. In contrast, VAN induced envelope stress responses, in agreement with its known role in peptidoglycan synthesis inhibition. FZ induced the SOS response consistent with its DNA damaging effects, but we demonstrate that using FZ in combination with the other two compounds enables use of lower dosages and largely mitigates its mutagenic effects. Based on the gene expression changes identified, we propose a synergy mechanism where the combined effects of FZ, VAN, and DOC amplify damage to Gram-negative bacteria while simultaneously suppressing antibiotic resistance mechanisms. Importance Synergistic combinations of existing antibacterials against multidrug-resistant “superbugs” are an alternative strategy to costly and arduous development of novel antibacterial molecules. The synergistic combination of nitrofurans, vancomycin and sodium deoxycholate shows promise in inhibiting and killing multidrug-resistant Gram-negative bacteria. We examined the mechanism of action and synergy of these three antibacterials, and proposed a mechanistic basis for their synergy. Our results highlight much needed mechanistic information necessary to advance this combination as a potential therapy.

translational stress. In contrast, VAN induced envelope stress responses, in agreement with its 23 known role in peptidoglycan synthesis inhibition. FZ induced the SOS response consistent with 24 its DNA damaging effects, but we demonstrate that using FZ in combination with the other two 25 compounds enables use of lower dosages and largely mitigates its mutagenic effects. Based on 26 the gene expression changes identified, we propose a synergy mechanism where the combined 27 effects of FZ, VAN, and DOC amplify damage to Gram-negative bacteria while simultaneously 28 suppressing antibiotic resistance mechanisms. 29 30 Importance 31 Synergistic combinations of existing antibacterials against multidrug-resistant "superbugs" are 32 an alternative strategy to costly and arduous development of novel antibacterial molecules. The 33 synergistic combination of nitrofurans, vancomycin and sodium deoxycholate shows promise in 34 inhibiting and killing multidrug-resistant Gram-negative bacteria. We examined the mechanism 35 of action and synergy of these three antibacterials, and proposed a mechanistic basis for their 36

Introduction 40
Antimicrobial resistance is one of the biggest public health crises at present. With the traditional 41 discovery and development of new antibiotics unable to keep pace with the emergence of 42 resistance (1), alternative strategies are urgently needed to tackle multidrug-resistant bacteria. 43 One promising approach is combining two or more drugs, especially if they are synergistic or 44 have an enhanced combined effect (2, 3). Synergistic combinations can lead to better pathogen 45 clearance, may slow down or prevent resistance development, and can lower the doses needed 46 for each of the components, which in turn can mitigate adverse effects (3,4). Repurposing 47 existing drugs approved for human use can also be a faster way of bringing new therapies into 48 the clinic in comparison to the development of novel antibacterial compounds (5). 49 Our recent studies have demonstrated the synergistic interaction of the existing antibiotics 50 nitrofurans and vancomycin (VAN) with the secondary bile salt sodium deoxycholate (DOC) (6). 51 In terms of efficacy and dose-reduction, we have shown that combining these three antibacterials 52 is superior to the previously reported double combination synergy of nitrofuran and DOC (7) or 53 nitrofuran and VAN (8). The triple combination is synergistic against a range of Gram-negative 54 bacteria, including the critical priority pathogens, carbapenem-resistant Enterobacteriaceae and 55 Acinetobacter baumannii (6,9). We have characterized the nitrofuran, VAN,and DOC synergy 56 in vitro, although the mechanism of synergy remains unknown. 57 Nitrofurans and DOC have variable effects on Gram-negative bacteria, but their exact 58 mechanisms of action are not fully understood. Nitrofurans are prodrugs (10) whose reactive 59 FZ-treated E. coli compared to the control were members of the SOS response (Fig.1A,Cluster 116 3) and respiration (Cluster 5), while those clusters that were most highly altered by DOC were 117 involved in iron import, translation, and amino acid transport and synthesis (Fig. 1A,Clusters 1 118 and 4). Particularly striking was the major overlap of gene perturbations by FZ and DOC, and 119 that the FVD combination resulted in the same pattern of gene clusters dysregulation, albeit 120 sometimes less pronounced (i.e. smaller fold change relative to the control). FZ's, DOC's, and 121 FVD's upregulated genes were involved in iron import ( Fig. 2A), ribosome assembly and 122 translation (Fig. 2B), whereas downregulated genes appear in the respiratory/electron transport 123 chain (ETC) (Fig. 2C) and central carbon metabolism (Fig. 2D). 124 A closer look at the DEGs showed that the expression patterns in response to FZ and DOC 125 indicate an iron starvation response; specifically, the upregulation of iron uptake, siderophore 126 synthesis, Fe-S cluster assembly, and downregulation of iron storage and utilization (e.g. ETC, 127 TCA cycle) (25, 26). The majority of these genes are either directly or indirectly controlled by 128 the transcriptional repressor Ferric Uptake Regulator (Fur), which is inactivated by iron 129 depletion (27-29). These results therefore suggest that FZ and DOC disturbs iron homeostasis 130 either by signaling or triggering an iron starvation response. Other genes induced by FZ, DOC, 131 or FVD include oxidative stress genes (sodA, soxS, ahpF, ahpC, gor, grxA) and those involved in 132 increased antibiotic tolerance (Fig. 2F). The multiple antibiotic resistance genes marA and marB, 133 of which the former is a master regulator of large number of genes involved in resistance, were 134 both upregulated. Similarly, expression patterns for genes that encode efflux pumps and porins 135 are suggestive of the inhibition of entry and accumulation of the antibiotics, such as the 136 upregulation of efflux pump component genes acrA and acrB and downregulation of the outer 137 membrane porin genes ompC and ompF (Fig. 2F). Genes involved in the SOS response were significantly upregulated in response to FZ, while 147 other conditions, including the FVD combination, did not show the same level of SOS response 148 gene upregulation (Fig. 2E), indicating the absence of severe DNA damage. To assess DNA 149 damage levels induced by the antibacterials alone and in combination, we determined mutation 150 frequencies in E. coli by quantifying mutant clones that gained resistance to rifampicin. 151 Rifampicin is an RNA polymerase inhibitor, and resistance can arise through single base 152 substitutions in the RNA polymerase gene rpoB (30). Under nonstress conditions (no-153 antibacterial control), the spontaneous mutation frequency of E. coli K1508 is around 7 mutants 154 per 10 8 cells (Fig. 3). Expectedly, given the DNA-damaging effects of FZ (11,12), this number 155 significantly increased upon FZ treatment (the mean frequency more than doubled). The other 156 two single compounds (DOC and VAN) and the FVD combination, on the other hand, did not 157 result in a significant change in rifampicin mutation frequency compared to the control. In 158 comparison to the single antibacterials, the SOS gene induction (Fig. 2E) and rifampicin 159 mutation frequency (Fig. 3) by FVD is considerably lower than those by FZ alone and more 160 similar to DOC, suggesting that the use of the combination, in which the FZ concentration is 161 lower than when used alone, can mitigate nitrofuran mutagenicity. the assessment of physiological changes at the level of cellular respiration. Since oxygen is the 175 major electron acceptor of the E. coli ETC (31), we investigated the overall effect on respiration 176 by measuring oxygen consumption using a Clark-type oxygen electrode (Fig. 4). Exposure to 177 IC 50 FZ, DOC, and Van for 1 h significantly decreased the oxygen consumption rate, causing a 178 1.6-fold, 1.7-fold, and 1.15-fold decrease, respectively, compared to the no-antibacterial control. 179 Similarly, the FVD combination reduced oxygen consumption by ~1.7-fold. The pattern of 180 dysregulation of the aerobic ETC genes by FZ, DOC, and FVD potentially reflects an 181 intracellular iron starvation signal ( Fig. 2A), which is known to activate iron-sparing 182 mechanisms (28). In this case, the non-iron utilising NADH dehydrogenase II (ndh) was possibly 183 upregulated to compensate for the downregulation of iron-rich NADH dehydrogenase I 184 (nuoABCEFGHIJKLMN) (Fig. 2C). In this assay, however, it was shown that the overall effect is 185 a decrease in aerobic respiration. of magnesium, iron, and manganese and an increase in copper levels. Lastly, the effects of the 202 FVD combination are the same as DOC, indicating that DOC drives most of the metal 203 homeostasis changes. However, the fold changes caused by FVD are much more pronounced 204 than those by DOC, reflecting the synergistic effect, such as the 18-fold decrease in magnesium 205 (vs. 9-fold by DOC), 2-fold decrease in iron (vs. 1.6-fold by DOC), 5-fold decrease in manganese 206 (vs. 2-fold by DOC) and a 6.5-fold increase in copper (vs. 2.5-fold by DOC) (Fig. 5). There was 207 no significant change in the total intracellular zinc and nickel levels by any of the treatments, and 208 VAN did not affect the total intracellular levels of any of the metals analysed in this study. 209 The observed changes in metal levels could partly explain the observed transcriptional response 220 to FZ and DOC. For example, copper excess has been reported to degrade Fe-S clusters, block 221 Fe-S cluster assembly, and stimulate an iron starvation response in E. coli and other bacteria (33-222 36). Therefore, copper toxicity could be a contributing factor that results in an iron starvation 223 response to DOC or FVD. Similarly, during iron starvation or oxidative stress, manganese import 224 is upregulated to replace iron as a cofactor in essential enzymes or prevent oxidative protein 225 damage (37, 38), and could therefore explain increased manganese levels by FZ. Surprisingly, 226 even though DOC induced the expression of oxidative stress genes of the OxyR regulon (e.g. 227 ahpC, ahpF, katG, dps, grxA, trxC, oxyS) (Fig. 2F) indicating H 2 O 2 stress (39), it caused a 228 decrease in total intracellular manganese levels. Taken together, FZ and DOC, besides affecting 229 iron homeostasis, also result in the dysregulation of other essential metals, including manganese, 230 magnesium, and copper. 231

Possible inhibition of SOS response by VAN 232
Most of the 17 DEGs triggered by Van were found to belong to multiple stress response 233 regulons, most frequently those involved in envelope stress: Rcs, Cpx, and Bae (Fig. 6). 234 Upregulation of these genes is consistent with the effects of peptidoglycan synthesis inhibitors 235 that are effective against Gram-negative bacteria (40, 41) and supports a VAN mechanism of 236 action identical to that in Gram-positive bacteria (42). Similarly, DOC and FVD upregulated 237 most of the envelope stress genes induced by VAN (Fig. 6). This is not surprising since DOC is 238 known to disrupt biological membranes (15,17). It is possible that membrane disruption by DOC 239 allows more Van to enter and exert its effect leading to synergy.  (44)), but we did not 252 use a fold change cut-off, as in that paper, so as not to omit any information from lowly 253 expressed genes. Biological process GO term enrichment of significantly downregulated genes 254 (FDR < 0.01, log 2 FC < 0) showed the SOS response to be overrepresented (Fig. S2). The SOS 255 response genes downregulated by VAN in the O'Rourke et al. dataset are umuC,umuD,uvrB,256 uvrC,uvrD,recN,dinB,dinG,polB,cho,and sulA. Downregulation of these genes could thus 257 explain the synergy between nitrofuran and VAN reported by Zhou et al. (8), who hypothesized 258 that VAN must increase DNA-damaging effects when combined with DNA-damaging agents, 259 such as nitrofurantoin or trimethoprim. If VAN exerts the same inhibition of SOS/DNA repair in 260 wildtype E. coli, this effect will contribute to the triple synergy by amplifying the DNA-261 damaging effects of FZ and DOC, in such a way that it decreases DNA damage adaptation and 262 survival through mutagenicity, which increases lethality. 263 If the SOS response plays a role in the synergy of FVD, for example through the inhibition of 264 SOS response by VAN, deletion of recA, which makes E. coli unable to mount an SOS response 265 (45), is expected to disrupt the FVD combination's synergy mechanism, therefore decreasing the 266 synergy in the mutant strain. Expectedly, recA deletion increased the susceptibility to FZ by 32-267 fold relative to the wildtype. Likewise, this deletion also decreased DOC MIC to 80000 g/mL 268 from more than 80000 g/mL, while VAN MIC (250 g/mL) remained unchanged. In the 269 checkerboard assay to investigate the interaction of FZ, VAN, and DOC, deletion of recA caused 270 a slight increase in the interaction index (FICI) of FVD (FICI < 0.22) compared to the wildtype 271 (FICI < 0.13), indicating only a slight decrease in synergy (Fig. S3). For the two-drug 272 interactions, only FZ and VAN combination showed a significant change in the FICI in the recA 273 mutant (Fig. 7). The deletion of recA resulted in a shift to indifferent interaction (FICI = 1) 274 instead of the synergy observed in the wildtype (FICI < 0.38) (Fig. 7A). Taken together, these 275 findings support the hypothesis that the SOS response is an interacting point for the synergy 276 between FZ and VAN. In terms of the triple combination synergy, however, the SOS response 277 contributes to the synergy, but, given that deletion of recA still results in a synergistic interaction, 278 other factors contributing to synergy are present. of their synergy. This is in line with studies that found a higher likelihood of synergy occurring 291 in drug combinations that induce either very similar or opposite gene perturbations (46, 47). In 292 particular, FZ and DOC both resulted in the upregulation of genes encoding iron uptake systems 293 and downregulation of those encoding iron storage and iron-utilizing proteins. This response is 294 indicative of Fur inactivation, which usually occurs during iron starvation (29), and is consistent 295 with previous reports of FZ and DOC increasing the expression of iron import genes (48, 49). An 296 E. coli fur mutant grown in iron-rich conditions has been shown to result in a 2-fold iron 297 decrease, while growth in the iron-depleted conditions showed a 14-fold reduction (50). Through 298 measurements of the total intracellular iron levels, we determined a less than 2-fold intracellular 299 iron decrease in the cultures containing FZ and DOC that corresponds to iron-rich conditions, 300 thereby ruling out that these drugs cause real iron starvation. 301 The mechanisms by which FZ or DOC inactivate Fur and induce an iron starvation-like response 302 are currently unknown. The mechanism may be as simple as reactive oxygen species oxidizing 303 useable Fe 2+ to unusable Fe 3+ via the Fenton reaction (51) or damaging the Fe-S clusters (52). If 304 so, iron import would be upregulated to supply iron to the labile iron pool and Fe-S cluster 305 machinery. The Fur-Fe 2+ complex has also been shown to be inactivated by nitric oxide (53). 306 Incidentally, nitroheterocyclic drug reduction has been proposed to result in nitric oxide 307 byproduct (54), though direct evidence for nitric oxide production during nitrofuran activation 308 has not yet been reported. Despite the overall decrease of iron content in the cells, the 309 inactivation of Fur will likely lead to an increase in the labile iron pool inside the cell that can 310 increase oxidative damage and stress via the Fenton reaction (55). 311 FZ and DOC also caused gene perturbations usually observed in bacteriostatic translation 312 inhibitors, such as downregulation of the central carbon metabolism and respiration, along with 313 the upregulation of ribosomal proteins to compensate for the translational stress (44,56,57). 314 Protein synthesis inhibition by nitrofurans has been reported and was proposed to be due to non-315 specific binding to ribosomal proteins and rRNA (13,14). However, translation inhibition by 316 DOC has not been demonstrated previously. Given that the Mg content of the cells is 317 dramatically lowered by DOC, a possible connection between Mg homeostasis dysregulation by 318 DOC and translational stress can be proposed: the decrease in total Mg levels by DOC will 319 decrease the number of functioning ribosomes and thus inhibit translation in E. coli due to 320 activation of the stringent response (58). However, due to the absence of (p)ppGpp regulation in 321 the E. coli K1508 strain used in this work (i.e. spoT and relA mutant), translation control based 322 on Mg 2+ levels is possibly absent, which would lead to the unchecked upregulation of ribosome 323 assembly even during Mg 2+ deficiency, a phenomenon which has been demonstrated in 324 Salmonella (58). 325 Our transcriptome analyses showed that FZ and DOC upregulated efflux pump genes (acrA, 326 acrB), downregulated porin genes (ompC, ompF) (Fig. 2F), and induced stress responses that are 327 expected to increase the tolerance to these agents. In this context, it seems contradictory that the 328 combination is synergistic rather than antagonistic. A possible explanation for observed synergy 329 is that some of these resistance mechanisms are somehow being inhibited by the combined action 330 of FZ, VAN, and DOC. A proposed pathway for inhibition of these resistance mechanisms could 331 be through Fur inactivation or downregulation of the central carbon metabolism. Both of these 332 activities will have an inhibitory effect on the ETC (25, 59). FZ, DOC, and FVD downregulated 333 the aerobic ETC genes nuo and cyd, which encode for two of the primary ETC complexes, 334 NADH dehydrogenase I and cytochrome bd-I, that generate the proton motive force (PMF). 335 These findings, along with the demonstrated overall decrease in aerobic respiration, could 336 indicate diminished proton motive force in response to FZ and DOC. Since PMF is directly or 337 indirectly required for the function of efflux pumps, low PMF is conducive for the accumulation 338 of the antibacterials inside the cells by preventing efflux. These findings support our previous 339 study in which deletion of tolC, or acrA efflux pump genes resulted in the loss or reduction of 340 synergy between FZ and DOC, highlighting the importance of efflux in the synergy mechanism 341 (7). 342 VAN treatment of E. coli did not induce much change in overall gene expression compared to 343 the control. Even though only 17 genes were significantly differentially expressed, these genes 344 do however give insight into the initial cellular effects of VAN. The majority of the 17 DEGs are 345 members of envelope stress responses, particularly the Rcs pathway, which has been 346 demonstrated to be induced by peptidoglycan-targeting antibiotics (40). Taken together, these 347 findings indicate that VAN on its own can somehow cross the outer membrane of E. coli, 348 although only in minimal amounts, and inhibit peptidoglycan synthesis. From the known 349 mechanisms of action of VAN (peptidoglycan synthesis inhibitor) and nitrofurans (DNA 350 damage), there does not seem to be an obvious connection as to why these drugs interact 351 synergistically (8). One important mechanism of resistance to nitrofurans is the SOS response to 352 DNA damage. Deletion of recA, making E. coli unable to mount an SOS response, resulted in the 353 loss of synergy between FZ and VAN. Data presented here indicate SOS response's involvement 354 in the mechanism of synergy between FZ and VAN, possibly through inhibition of SOS 355 induction by VAN, which can decrease the resistance to FZ's DNA-damaging effect and 356 therefore increase lethality. The recA deletion also increased the FICI for the triple combination,

Materials and Methods 375
Bacterial strain, growth conditions, and checkerboard assay 376 was grown aerobically in 2xYT medium at 37 ºC. kan R recA deletion mutation from the Keio 378 collection (60) were introduced into E. coli K1508 using phage P1 transduction, as previously 379 described (61). To eliminate potential polar effects on downstream genes in the operon, the FLP 380 recombinase recognition target (FRT)-flanked kan R cassette was excised using FLP-mediated 381 recombination using plasmid pCP20 (62). 382

Determination of MIC and checkerboard assay 383
MIC determinations and checkerboard assays were performed using the broth microdilution 384 method in a 384-well plate according to the CLSI guidelines (63), with minor changes. 2xYT 385 medium was used, the inoculum concentration is 1 × 10 6 CFU/mL, the plates were incubated at 386  VAN(com) MIC VAN where MIC FZ(com) , MIC DOC(com) , and MIC VAN(com)  The concentrations that gave 50% growth inhibition (IC 50 ) at 24 h was chosen for the 396 transcriptomics study. The IC 50 determination was performed in 384-well plates on E. coli K1508 397 at 1 × 10 6 CFU/mL in a total volume of 50 µL. The plate was incubated at 37 o C for 24 h, and 398 optical density at 600 nm (OD600) was determined. The mean % growth inhibition was 399 calculated and the R package drc v3.0-1 (65) was used to plot the concentration-response (% 400 inhibition) curves fitted with a four-parameter log-logistic model to determine the IC 50 . For the 401 combination, the most synergistic combination (i.e. lowest FICI) in a checkerboard assay were 402 first determined (Fig. S2), then fixed-ratio dilutions of these concentrations were used to plot a 403 concentration-response curve. The IC 50 for each antibacterial and combination, which was used 404 in all the assays, are summarized in Table S1. 405 Exponentially growing cultures of the E. coli K1508 at 5 × 10 7 CFU/mL were treated with IC 50 406 antibacterial(s) in a final volume of 25 mL. The DMSO concentration for all treatments was 407 fixed at 0.1%. After incubation at 37 °C with shaking at 200 rpm for 1 h, the cultures were 408 harvested by centrifugation. The resulting pellet was then resuspended in 1 mL of resuspension 409 buffer (20 mM sodium acetate pH 5.5, 1 mM EDTA, 1% SDS) and homogenized by beat 410 beating. The samples were then subjected to phenol-chloroform nucleic acid extraction, as 411 previously described (66)

Mutagenicity Assay 434
Mutation frequencies were measured as described previously (74). Briefly, exponentially-435 growing E. coli K1508 at 1 ×10 7 CFU/mL were treated with IC 50 FZ, VAN,and DOC,alone and 436 in combination, in a final volume of 10 mL in 2xYT medium. The cultures were incubated at 37 437 °C with shaking at 200 rpm for 24 h. The cultures were then centrifuged at 5000 ×g for 10 min 438 and resuspended in maximum recovery diluent (0.1% peptone, 0.85% NaCl). Serial dilutions 439 were plated in triplicate onto 2xYT agar containing 100 μg/mL rifampicin to select for 440 rifampicin-resistant colonies and on non-selective 2xYT agar to count the total number of 441 colonies. The plates were scored after 24 h at 37 °C. The mutation frequency was calculated by 442 dividing the number of rifampicin-positive colonies by the total number of colonies from 9-11 443 biological replicates. 444

Oxygen consumption 445
Oxygen consumption was measured as previously described (75). Briefly, E. coli K1508 culture 446 at OD600 of 0.3 was treated with IC 50 FZ, VAN, and DOC, alone and in combination, at 37 °C 447 for 1 h. Cells were then diluted in air-saturated 2xYT to OD600 of 0.2, and dissolved oxygen 448 was measured in a closed chamber with constant stirring using a Clark-type oxygen electrode 449 (Rank Brothers Ltd.) linked to a chart recorder (Vernier LabQuest Mini). 450

Metal concentration by ICP-MS 451
Antibiotic-treated E. coli cultures in a total volume of 80 mL were processed the same way as for 452 the transcriptomics analyses. After antibiotic treatment, cells were collected and prepared for 453 ICP-MS, as previously described (76). Briefly, cells were harvested by centrifugation (5000 ×g, 454 10 min), then washed twice with 25 mL phosphate-buffered saline (PBS) containing 0.5 mM 455 EDTA, then twice with PBS. All samples were adjusted to a cell number of 2 × 10 9 CFU based 456 on their OD600 values. Washed cell pellets were then digested with 500 μL of 70% (wt/vol) 457 nitric acid (≥99.999% trace metals basis) at 80 °C overnight. Each sample was diluted 1:20 in 458 Milli-Q water (18.2 MΩ), giving a final acid matrix of 3.5%. The samples were then sent to the 459 University of Waikato Mass Spectrometry Facility to analyze metal content by ICP-MS on an 460