Synergism of imipenem with fosfomycin associated with the active cell wall recycling and heteroresistance in Acinetobacter calcoaceticus-baumannii complex

The carbapenem-resistant Acinetobacter calcoaceticus-baumannii (ACB) complex has become an urgent threat worldwide. Here, we determined antibiotic combinations and the feasible synergistic mechanisms against three couples of ACB (A. baumannii (AB250 and A10), A. pittii (AP1 and AP23), and A. nosocomialis (AN4 and AN12)). Imipenem with fosfomycin, the most effective in the time-killing assay, exhibited synergism to all strains except AB250. MurA, a fosfomycin target encoding the first enzyme in the de novo cell wall synthesis, was observed with the wild-type form in all isolates. Fosfomycin did not upregulate murA, indicating the MurA-independent pathway (cell wall recycling) presenting in all strains. Fosfomycin more upregulated the recycling route in synergistic strain (A10) than non-synergistic strain (AB250). Imipenem in the combination dramatically downregulated the recycling route in A10 but not in AB250, demonstrating the additional effect of imipenem on the recycling route, possibly resulting in synergism by the agitation of cell wall metabolism. Moreover, heteroresistance to imipenem was observed in only AB250. Our results indicate that unexpected activity of imipenem on the active cell wall recycling concurrently with the presence of heteroresistance subpopulation to imipenem may lead to the synergism of imipenem and fosfomycin against the ACB isolates.

. The minimum inhibitory concentrations (MICs) of six ACB isolates to imipenem (IPM), meropenem (MEM), fosfomycin (FOF), amikacin (AMK), and colistin (CT) were determined by agar dilution method. The results were interpreted as susceptible (S), intermediate resistant (I), or resistant (R). The presence of carbapenemase genes was performed by PCR. The clonality was performed by MLST Oxford scheme as seven allelic numbers and a sequence type number.  Overexpression of efflux pumps. Multidrug efflux pumps play one of the essential roles in the antibiotic resistance of ACB isolates. The overexpression of efflux pump phenotype for carbapenems was characterized by using CCCP. No ACB isolate showed the positive phenotype of efflux pump overexpression to carbapenems (Table 2). Therefore, we determined the expression level of efflux pump genes. In this study, A. baumannii carried adeB gene, whereas A. pittii carried adeE and adeY genes ( Table 2). Neither of these genes was found in A. nosocomialis isolates ( Table 2). Overexpression of adeB was observed in A. baumannii A10, which had a high level of carbapenem MICs (128-256 mg/L) ( Fig. 1G and Table 1). A. pittii AP1 showed slightly overexpressed adeE ( Fig. 1H) with twofold carbapenem MICs above these of AP23 (Table 1). A. pittii AP1 and AP23 equally displayed the adeY expression (Fig. 1I). The p-values were calculated using unpaired two-tailed t-test (*p-value ˂0.05; **p-value < 0.01; ***p-value < 0.001 and ns, non-significant). (G) The relative mRNA expression of adeB among A. baumannii was evaluated by RT-PCR and normalized to 16S rRNA expression. The relative mRNA expression of adeE (H) and adeY (I) was evaluated by RT-PCR and normalized to 16S rRNA expression. All experiments were performed in triplicate. Mean values of the relative mRNA expression were plotted with error bars representing the standard error of the mean (n = 3). The p-values were calculated using unpaired two-tailed t-test (*p-value ˂0.05; **p-value < 0.01; ***p-value < 0.001 and ns, nonsignificant). www.nature.com/scientificreports/ In conclusion of carbapenem resistance mechanisms, the major mechanism found in all isolates was carbapenemase production. OXA-23 production was present in AP23, AN4, and AN12. OXA-24 production was present in AB250 and A10. The production of OXA-58 with IMP was found in AP1. For the reduction of porins, reduced OprD was present in A10, AP23, and AN12. Reduced 33-36 kDa porin was present in AB250, AP23 and AN12. Reduced CarO was present in AB250 and AP23. For efflux pumps, overexpression of adeB and adeE was present in A10 and AP1, respectively. Activity of antibiotic combinations against six ACB isolates. The in vitro activities of imipenem or meropenem in combination with either amikacin, colistin, or fosfomycin against A. baumannii, A. pittii, and A. nosocomialis isolates were determined by checkerboard assay. The MICs of antibiotic combinations that used for the fractional inhibitory concentration (FIC) index calculation in Eq. (1), are present in Supplementary Table S2. The most effective combination was imipenem with fosfomycin that exhibited synergism (FICI ≤ 0.5) against all six ACB isolates (Table 3). Secondly, meropenem plus fosfomycin and imipenem plus amikacin were potential combinations against A. nosocomialis and A. pittii (Table 3). No synergism was observed in imipenem plus colistin, meropenem plus amikacin, and meropenem plus colistin. From the results of fosfomycin susceptibility, fosfomycin alone had inadequate potency against all ACB isolates ( Table 1). The combination results differed in that fosfomycin had a synergistic activity with carbapenems, especially imipenem.
Time-killing curves of imipenem with fosfomycin against six ACB isolates. As a result of the checkerboard assay, we, therefore, verified the synergism of imipenem with fosfomycin against six ACB isolates by time-killing assay. In every ACB isolate, the growth control curves were normal S-curves, which reached the log phase (exponential phase) at 2 to 4 h of incubation (Fig. 2). Both 0.5× and 1× imipenem MICs were unable to kill A. baumannii AB250 ( Fig. 2A) and A10 (Fig. 2D). In the presence of 0.5× fosfomycin MIC, AB250 was able to grow (Fig. 2B), but it was killed by 1× fosfomycin MIC for 4 h and regrew subsequently (Fig. 2B). Both fosfomycin concentrations (0.5× and 1× MICs) killed A10 for 6 h before the regrowth (Fig. 2E). No combination was able to achieve the synergistic activity with AB250 (Fig. 2C). AB250 was not killed by either 0.5× or 1× imipenem MIC combined with 0.5× fosfomycin MIC (Fig. 2C). Although imipenem in combination with 1× fosfomycin MIC eliminated AB250, the regrowth occurred after 4 h, resulting in no synergism. In contrast, most combinations were able to eradicate A10 leading to synergism, except 0.5× combination that achieved regrowth and no synergism (Fig. 2F).
In A. nosocomialis, both concentrations of imipenem alone killed AN4 for 12 h ahead of regrowth occurring in the 0.5× imipenem MIC (Fig. 2M). In contrast to AN12, 0.5× imipenem MIC could not kill, whereas the regrowth after 12 h was observed in the 1× imipenem MIC (Fig. 2P). All fosfomycin concentrations alone were able to kill AN4 before regrowth appeared at 4 h (Fig. 2N). Killing and inhibition of AN12 were observed by 1× and 0.5× fosfomycin MIC, respectively, before regrowth (Fig. 2Q). AN4 and AN12 were killed by all combinations, resulting in synergistic activity (Fig. 2O,R), except 0.5× imipenem with 0.5× fosfomycin MIC that regrowth appeared in AN12 after 6 h (Fig. 2R).
MurA amino acid sequences among six ACB isolates. To understand fosfomycin resistance mechanisms in ACB isolates, the amino acid sequences of the fosfomycin target, MurA, were determined and analyzed. Amino acid sequences of the MurA among six ACB isolates are shown in Supplementary Fig. S2. All MurA sequences were wild-type (WT) that displayed no mutation associated with fosfomycin resistance, including Cys116, Lys22, Arg121, Arg398, Asp370, and Leu371 (arrows in Supplementary Fig. S2).
Expression of murA gene in six ACB isolates. Interestingly, no MurA mutation was found in all ACB isolates that intermediate or resistant to fosfomycin. To determine whether MurA was associated with fosfomycin resistance in ACB isolates, murA expression was evaluated in the presence of fosfomycin for 2 h. Fosfomycin did not affect murA expression in most isolates, including A. baumannii AB250 ( Overexpression of efflux pump induced by fosfomycin. Another mechanism of fosfomycin resistance reported in A. baumannii is the overexpression of the efflux pump, AbaF. Firstly, the phenotype of efflux pump overexpression was performed using CCCP. Unfortunately, all isolates showed negative phenotypes of efflux pump overexpression (Table 2). Therefore, the level of abaF expression was determined by RT-PCR. Fosfomycin exhibited the downregulation of abaF in A. baumannii AB250 (Fig. 3G). Overexpression of abaF was observed when using a low concentration of fosfomycin against A. baumannii A10 (Fig. 3H). These results indicate that overexpression of abaF may involve fosfomycin susceptibility in a strain-specific manner in A. baumannii.
Cell wall recycling pathway in six ACB isolates. Our results suggest other mechanisms that bypass the MurA-dependent cell wall synthesis pathway. We screened several enzyme genes that play a role in the cell wall recycling pathway. All A. baumannii and A. nosocomialis carried all tested genes, including ampG, nagZ, anmK, amgK, and murU, whereas amgK did not detect in both A. pittii isolates ( Table 4).
The initial step of cell wall recycling is the uptake of shedding peptidoglycan into the cytoplasm through the AmpG transporter. Therefore, the expression level of ampG was determined in the presence of fosfomycin for 2 h. Fosfomycin dose-dependently downregulated ampG expression in A. baumannii AB250 ( Another essential protein in cell wall recycling is MurU, the last enzyme producing the cell wall precursor that bypasses the MurA-dependent pathway. Downregulation of murU by fosfomycin was found in most ACB isolates (Fig. 4G,I-L) except A. baumannii A10, which overexpressed murU by a low level of fosfomycin (Fig. 4H). In summary, according to ampG and murU expression, fosfomycin did not upregulate cell wall recycling at 2 h of exposure in ACB isolates (Fig. 4). This is in accordance with the time-kill results, which demonstrated that no increase of bacterial cells in 2 h of fosfomycin exposure (Fig. 2). Except for AP1, fosfomycin induced ampG expression (Fig. 4C), but downregulated murU expression (Fig. 4I). Subsequently, we focused on A. baumannii AB250 and A10, which were non-synergistic and synergistic strains, respectively, by the combination of fosfomycin and imipenem. To evaluate the difference between these isolates, the expression of the additional genes in the cell wall recycling, including nagZ, murU, and anmK, was conducted at 4 and 12 h of exposure to fosfomycin, the rebound in growth (regrowth) occurred (Fig. 2B,E). The different expression patterns were significantly observed at 4 h (Fig. 5). The downward trend of expression was found in AB250, in which fosfomycin slightly downregulated nagZ ( To evaluate the role of imipenem combination on cell wall recycling, the expression of cell wall recycling was determined after 12 h exposure to 0.5× fosfomycin MIC with 1× imipenem MIC, which had synergistic activity in A10 (Fig. 2F) but not in AB250 (Fig. 2C), compare to either fosfomycin or imipenem alone. Imipenem in the combination did not affect murA expression in AB250 (Fig. 6A) but downregulated in A10 (Fig. 6B). The transporter gene, ampG, was upregulated in the imipenem combination in AB250 ( Fig. 6C) but had no effect in A10 (Fig. 6D). In AB250, although imipenem combination induced murU expression, its expression level was slightly lower than that in the absence of imipenem (Fig. 6E). In contrast, the combination reduced murU expression in A10 nearly to that in the absence of any antibiotic (Fig. 6F). Interestingly, imipenem combination  www.nature.com/scientificreports/ significantly downregulated nagZ (Fig. 6H), anmK (Fig. 6J), and amgK (Fig. 6L) in A10 compared to control and either single fosfomycin or imipenem. However, the combination showed a few effects on these gene expressions in A250 (Fig. 6G,I,K). These results indicate that imipenem may affect at least in part of cell wall recycling resulting in synergism with fosfomycin. The summary of cell wall recycling and the proposed synergistic mechanism of fosfomycin and imipenem in AB250 and A10 are present in Fig. 7. Imipenem synergistically reduced the expression of cell wall recycling in A10 (red symbols in Fig. 7A), indicating dwindling cell wall synthesis that may result in cell death. Differently, imipenem showed a minor effect on the alteration of cell wall recycling (red symbols in Fig. 7B), indicating a sluggish but functional and adequate cell wall synthesis that may result in cell growth.
PAP assay. In addition to antibiotic resistance mechanisms, the population phenotypes of all ACB isolates were evaluated. The phenotypes of populations were determined by using the PAP study, which displayed the frequency of bacteria growing on agar supplemented with various concentrations of tested antibiotics, which calculated by using Eq. (2). The positive of heteroresistant subpopulation was defined as the presence of bacterial frequency that grows above 10 -7 on the agar supplemented with eightfold above the antibiotic concentration of the main population. For imipenem, AB250 exhibited the heteroresistant subpopulation in which the frequency of bacteria was above 10 -7 at eightfold above the antibiotic concentration of the main population (32 mg/L) (Fig. 8A). This phenotype was called resistant with heteroresistant subpopulation to imipenem (Fig. 8A). In the case of A10, the frequency of the growth at eightfold above the resistance level of the main population to imipenem (512 mg/L) was less than 10 -7 , indicating no heteroresistant subpopulation (Fig. 8B). Therefore, A10 was resistant without heteroresistant subpopulation to imipenem (Fig. 8B). No heteroresistant subpopulation to imipenem was also observed in A. pittii AP1 (Supplementary Fig. S4A), AP23 ( Supplementary Fig. S4B), A. nosocomialis AN4 (Supplementary Fig. S4C), and AN12 ( Supplementary Fig. S4D).
For fosfomycin, all isolates exhibited the resistant frequency upper the cut-off point (10 -7 ) at eightfold above the antibiotic concentration of the main population of each isolate (Fig. 8C,D, Supplementary Fig. S4E-H). Therefore, all ACB isolates were resistant with heteroresistant subpopulations to fosfomycin. In summary, among all isolates, A. baumannii AB250 was the only isolate that imipenem and fosfomycin combination had no synergistic activity and was resistant with heteroresistant subpopulation to imipenem. These results may indicate the feasible association between synergism and the heteroresistant subpopulation. However, further study is required to understand the impact of heteroresistance in antibiotic synergism.

Discussion
ACB complex, generally considered saprophytes, has been regarded as a critical multidrug-resistant nosocomial pathogen in clinical settings within the last two decades 18 . Although A. pittii and A. nosocomialis are lower grade pathogens than A. baumannii, their abilities to resistant to antibiotics, particularly carbapenems, have been reported 17,19,20 . Moreover, the emergence of colistin-resistant ACB is rising worldwide 21 ; fortunately, no colistinresistant isolate was observed in our study. A couple of A. baumannii and A. pittii in our study belonged to new STs submitted to the PubMLST 22 , indicating unique clones emerging in our hospital. Carbapenem-resistant A. nosocomialis belonged to ST958 which is similar to carbapenem-resistant A. baumannii isolated from a patient in Uruguay, indicating a close relation among the ACB complex 23 . The vast majority of ACB complex produce class D carbapenemases, including OXA-23, OXA-24, OXA-58, and OXA-51. The latter is an intrinsic carbapenemase in A. baumannii; however, OXA-51-type carbapenemase is transferred to non-baumannii Acinetobacter via horizontal transfers particularly, plasmid 24 . Reduction of 33-36 kDa OMP and CarO was found in A. pittii AP23 and A. nosocomialis AN4, respectively, whereas the loss of OprD is reported in A. nosocomialis isolated from Taiwan, and no reduction of OMP is found in non-baumannii Acinetobacter isolated from South Korea 25,26 . Interestingly, not only involved in antibiotic resistance, but these porins also act as virulence factors 27 . Moreover, we found overexpression of adeB and adeE in A. baumannii and A. pittii, respectively. A report from Taiwan revealed that carbapenem-resistant A. nosocomialis exhibited overexpression of adeB, leading to tigecycline resistance 28 . A. baumannii may act as the coach supporting antibiotic resistance determinants and virulence machines to other species in the ACB complex. Not surprisingly, non-baumannii Acinetobacter is turning into a potential pathogen.
Due to the limitation of antibiotic usage for carbapenem-resistant ACB, many studies focused on antibiotic combinations, most of which were colistin-based combinations such as imipenem or meropenem plus colistin www.nature.com/scientificreports/ www.nature.com/scientificreports/ and sulbactam plus colistin [29][30][31] . Interestingly, sulbactam is a beta-lactamase inhibitor, not an antibiotic, showed synergism with colistin against carbapenem-resistant ACB 29 . A report revealed that sulbactam inhibits penicillinbinding proteins (PBPs), including PBP1 and PBP3 32 . Another non-traditional antibiotic being effective in the combinations is fosfomycin. Fosfomycin displayed synergism with sulbactam and colistin against carbapenemresistant A. baumannii 33,34 . Moreover, the synergy of fosfomycin in the combination with imipenem was revealed in our previous study 16 . Therefore, we determined the in vitro activity of fosfomycin plus imipenem and other combinations and characterized the resistance mechanisms to unveil the plausible synergistic mechanism. Apart from A. baumannii, the synergistic activity was also found against A. pittii and A. nosocomialis. This study demonstrates that imipenem with fosfomycin could be used for combating carbapenem-resistant ACB complex. However, the effectiveness of the combination should be clinically evaluated further. Fosfomycin inhibits peptidoglycan synthesis by covalently binding to MurA (Fig. 7). Generally, fosfomycin is recommended for UTIs caused by certain Enterobacteriaceae, notably E. coli 10 . Nevertheless, E. coli is resistant to fosfomycin by various mechanisms. Firstly, the alteration of drug target, MurA, generally occurs at the active site, Cys115, and the ligand-binding site, including Lys22, Arg120, and Arg397 ( Supplementary Fig. S2) 35 . This mechanism has never been found in A. baumannii, so MurA has been inspired to be a new target 36 . Unfortunately, A. baumannii normally has a high level of fosfomycin susceptibility with WT MurA, suggesting the intrinsic resistance by the MurA-independent pathway 37 . The second mechanism is the mutations of fosfomycin transporters (GlpT and UhpT), leading to decreased uptake of fosfomycin, but this mechanism has not been revealed in A. baumannii 37 . Thirdly, the production of fosfomycin-modifying enzymes (such as FosA, FosB, FosC, and FosX) is the most frequently found mechanism in both Gram-negative and Gram-positive bacteria 37 . fosA has been deposited on 2% and 7.8% of the A. baumannii and A. pittii genomes, respectively, demonstrating that FosA production is not an intrinsic resistance mechanism in ACB 38 . Another mechanism reported in A. www.nature.com/scientificreports/ baumannii, is the efflux transporter, AbaF 13 . The deletion of abaF confers the reduction of eightfold fosfomycin MIC in A. baumannii 13 . In our study, both A. baumannii had nearly equal fosfomycin MICs, but overexpression of abaF was found in one isolate, indicating a minor role of the AbaF. An additional mechanism being debate is the cell wall recycling pathway that bypasses the de novo synthesis via MurA (Fig. 7) 14 . The cell wall recycling pathway has been well characterized in E. coli, but it differs from that of other Gram-negative bacteria, including A. baumannii that is the MurA-bypass pathway 15,39,40 .
Although fosfomycin induces overexpression of murA in E. coli 41 , most ACB isolates, whose MurA exhibited wild-type enzymes that susceptible to fosfomycin, were not induced murA expression by fosfomycin except at a high concentration (1× MIC) in A10 (Fig. 3B). These results with the presence of genes encoding cell wall recycling enzymes strongly suggest the cell wall recycling pathway that is a MurA-independent pathway in the ACB complex. In spite the fact that AmpG transporter upregulates at a high level of substrates, indicating an active turnover of the cell wall 42 , most ACB isolates were no change or downregulation of ampG together with murU after 2 h treated by fosfomycin, except the low fosfomycin concentration in A10 (Fig. 4B) and AP1 (Fig. 4C). Downregulations of these genes indicate the downward trend of cell wall syntheses that are related to inhibition of cell growth in the time-kill curves (Fig. 2).
According to the time-kill curves with fosfomycin, the log phase of growth shifted to 4-12 h instead of 2-4 h. Therefore, the cell wall expression in A. baumannii AB250 (no synergistic strain) and A10 (synergistic strain) were focused at 4 and 12 h. Downward expression of the recycling was found in non-synergistic strain, whereas the upward trend was observed in synergistic strain, indicating the more active recycling in the synergistic strain. However, both strains exhibited upregulation of the recycling in 12 h compared to that of 4 h. Unexpectedly in both isolates, fosfomycin equal to the MICs barely induced murU expression (Fig. 5C,D), but the viability of growth did not affect (Fig. 5B,E), possibly due to achieving the steady-state of the recycling at that time. In addition to being the last enzyme in the cell wall recycling, MurU is believed that plays an important role in the preservation of a steady-state of a MurNAc pool and the suppression of an anhMurNAc pool 43 . According to these hypotheses, the bacterial growth with shutting down of murU at 12 h exposure to fosfomycin is probably www.nature.com/scientificreports/ caused by the excess of cell wall materials. It is the limitation of our study that the bacterial metabolites were not determined. In combination of 0.5× fosfomycin with 1× imipenem that did not affect bacterial growth ( Fig. 2A-F), imipenem showed a tiny role on the recycling expression in non-synergistic strain, indicating an inert response of AB250. In contrast, the significant downregulation of the recycling was observed in synergistic strain, suggesting enhance effect of imipenem in perturbation of the recycling of A10. Therefore, imipenem may be synergistic with fosfomycin at least in part downregulation of the cell wall recycling.
The targets of imipenem are PBPs, not the cell wall recycling, thus imipenem indirectly plays a role on the recycling. The cell wall synthesis is composed of the precursor production (the de novo or recycling bypath) and peptidoglycan crosslinking (by the PBPs). Both processes are inevitably related and sophisticated. Many studies are supporting the hypothesis that imipenem not only inhibits PBPs, but also interrupts other cellular metabolisms 44,45 . For instance, mecillinam, whose targets are PBPs, simultaneously blocks the PBPs and enhances a cycle of cell wall synthesis and turnover via the Rod system, resulting in depleting PG precursor pools in E. coli. Similar to a report in A. baumannii 46 , imipenem affects not only PBP2 and PBP1a, but also perturbs the Rod www.nature.com/scientificreports/ system. Therefore, imipenem in combination with fosfomycin may complicatedly disturb cell wall metabolism, at least in some parts, resulting in decreasing of cell wall recycling. An additional factor that may additionally affect the activity of antibiotic combination, is heteroresistance. The heteroresistance is a phenotype of a subpopulation that displays a greater potency of antibiotic resistance than that in the main population. The higher-level resistance in the heteroresistance is caused by the mutation of antibiotic resistance determinants 47 . Therefore, the ACB isolates may differently express the cell wall recycling to generate the fosfomycin heteroresistance. Furthermore, the bacterial regrowth in the time-kill curves may be the growth of fosfomycin and imipenem heteroresistance subpopulations. However, regrowth may be due to loss of antibiotic stability. Imipenem has a half-life of 0.7 h in serum in vitro 48 , whereas fosfomycin has a half-life of 5.7 h in plasma 9 . There are various methods for the detection of the heteroresistant subpopulation. The gold standard method is the PAP assay 49 . In this study, the PAP assay was used with the addition of the frequency of heteroresistant subpopulations as recommended by Andersson et al. 47 . Among six ACB isolates, only A. baumannii AB250, non-synergistic strain, exhibited heteroresistant to imipenem (Fig. 8A), whereas all isolates had heteroresistant subpopulations to fosfomycin (Fig. 8C,D, Supplementary Fig. S3E-H). Notably, all isolates resistant with heteroresistant subpopulations to fosfomycin showed similar patterns of the time-kill curves that regrowth occurred after 4 h, possibly due to subpopulation change. However, all ACB isolates exhibited unstable heteroresistance phenotypes in which subcultures with antibiotic-free media (> 50 generations) 47,50 showed the loss of heteroresistance (Supplementary Table S3). Therefore, the unstable heteroresistance may lead to failure treatment by combination therapy and is difficult to detect by the routine method.
In summary, although, most ACB isolates possessed cell wall recycling pathway, their response to fosfomycin were quite difference, indicating strain-specific responses. The synergistic strain (A10) exhibited more active of the cell wall recycling than no synergistic strain (AB250). Imipenem, in the combination, significantly downregulated the cell wall recycling in the synergistic strain, indicating the additional action apart from inhibition of PBPs. Therefore, the feasible synergistic mechanism of imipenem and fosfomycin was an unexpected function of imipenem that affects at least in part of cell wall recycling resulting in synergism via downregulation of cell wall recycling concurrently without heteroresistance subpopulation. Nevertheless, the role of heteroresistance in the synergism of imipenem and fosfomycin is still unclear and needs further investigation. Moreover, both cell wall metabolism and bacterial response to antibiotics are dynamic and sophisticated. Therefore, the comparative metabolic perturbations of these strains should be further investigated to unveil the synergistic mechanism of imipenem and fosfomycin. This study demonstrates the in vitro synergism of imipenem with fosfomycin against carbapenem-resistant ACB complex.

Methods
Bacterial strains and antibiotic susceptibility testing. Two A. baumannii isolates (AB250 and A10), two A. pittii isolates (AP1 and AP23), and two A. nosocomialis isolates (AN4 and AN12) from our previous study were clinical strains that were isolated from an individual patient at the King Chulalongkorn Memorial Hospital, Bangkok, Thailand 16,17 . All strains were routinely tested for antibiotic susceptibility according to CLSI recommendation 51 . Susceptibility of imipenem (Apollo Scientific), meropenem (Sigma-Aldrich), amikacin (Sigma-Aldrich), and colistin (Sigma-Aldrich) was performed by broth microdilution method using cationadjust Mueller-Hinton broth (CAMHB) (Becton Dickenson BBL) whereas that of fosfomycin was performed by agar dilution method using Mueller-Hinton agar (MHA) (Becton Dickenson BBL) supplemented with 25 mg/L of glucose-6-phosphate (G6P) (Sigma-Aldrich). Bacterial reference strains were E. coli ATCC 25922 and P. aeruginosa ATCC 27853. The antibiotic susceptibility was interpreted according to the CLSI guideline 51 (Supplementary Table S1).
Moreover, phenotype of overexpression of efflux pump against imipenem, meropenem, and fosfomycin was performed by agar dilution method compared with the addition of carbonyl cyanide m-chlorophenyl hydrazone (CCCP) (Sigma-Aldrich), a proton coupler interrupting efflux pump function. The positive phenotype of overexpression of efflux pump was defined as at least fourfold decreased of antibiotic MIC (minimum inhibitory concentration) observed in the presence of CCCP.

MLST.
The clonal relationship among all six ACB isolates was studied by MLST as recommendation of the PubMLST 22 . Briefly, the partial fragments of seven housekeeping genes, including gltA, gyrB, gdhB, recA, cpn60, gpi, and rpoD, were amplified from extracted genomic DNA by PCR using primer recommended by the PubMLST 22 . The PCR products were sequenced. We performed and analyzed the MLST profiles according to the MLST Oxford scheme. The allelic numbers of each gene and the sequence type (ST) numbers were obtained from the PubMLST.
OMP study. OMPs of all isolates were separated by using ultracentrifugation method as our previous study 16 .
Briefly, the mid-log phase bacterial cells were broken by sonication and the membrane fractions were collected by ultracentrifugation at 100,000g for 1 h at 4 °C (Beckman-Coulter). OMPs were extracted by using 2% sodium N-lauryl sarconate (Merck Millipore) and collected by ultracentrifugation again at 100,000g for 1 h at 4 °C. The OMPs were resuspended with phosphate buffer saline (Sigma Aldrich) and were quantified the concentration by using Bio-Rad protein assay (Bio-Rad). The OMP profiles were studied by SDS-PAGE. The concentration of OMPs loaded in each well was 10 µg. The gels were stained with coomassie brilliant blue, dried on cellophane sheets, and captured by using an image scanner. The density of each protein band was determined by using ImageJ. The relative density of each protein was calculated and compared to that of control protein, OmpA, in the same bacterial isolate. Checkerboard assay. In vitro activity of carbapenem (imipenem and meropenem) in combination with either amikacin, fosfomycin, or colistin against all six ACB isolates was performed by checkerboard assay as our previous study 16 . Briefly, the checkerboard assay was conducted in 96-well microtiter plates in which the rows contained CAMHB supplemented with serial dilution of one antibiotic and the columns contained CAMHB supplemented with serial dilution of another antibiotic. The plates were inoculated with the ACB isolates and incubated at 37 °C for 18-24 h. The MICs of each antibiotic in alone and the combination were read by naked eyes. The FIC (fractional inhibitory concentration) index was calculated using the following equation: The results were interpreted as following: synergism as FICI ≤ 0.5, no interaction as 0.5 < FICI ≤ 4, and antagonism as FICI > 4.
Time-killing assay. Synergism of the most effective combination, imipenem with fosfomycin, was confirmed by time-killing assay as our previous study 16 . The concentrations of 1× MIC and 0.5× MIC of imipenem and fosfomycin were tested as a single agent and in combination. The flask contained CAMHB supplemented with each concentration of single imipenem or fosfomycin or the combinations was inoculated with 10 6 CFU/ mL of ACB isolates. Nine conditions of antibiotic were tested, including growth control (no antibiotic), 0.5× imipenem MIC, 1× imipenem MIC, 0.5× fosfomycin MIC, 1× fosfomycin MIC, 0.5× imipenem + 0.5× fosfomycin MIC, 0.5× imipenem MIC + 1× fosfomycin MIC, 1× imipenem MIC + 0.5× fosfomycin MIC, and 1× imipenem MIC + 1× fosfomycin MIC. The flasks contained fosfomycin were supplemented with 25 mg/L of G6P. All flasks were incubated at 37 °C for 24 h with shaking at 120 rpm. The viable cells at 0, 2, 4, 6, 12, and 24 h after incubation were counted and plotted. The synergism was defined as at least 2log decreased of viable cells compared to the most active single agent after 24 h of incubation. The bactericidal activity was defined as at least 3log decreased of viable cells compared to the viable cells of initial inoculation. This experiment was performed in triplicate.
Population analysis profile (PAP) assay. The characteristics of ACB population that composed of either homogeneous or heterogeneous populations were determined by PAP assay 47 . Briefly, overnight culture of bacterial strains was serially diluted and plated on MHA agar containing different antibiotic concentrations (at 0, 4,8,16,32,64,128,256,512, and 1024 mg/L of imipenem and at 0, 16, 32, 64, 128, 256, 512, 1024, and 2048 mg/L of fosfomycin supplemented with 25 mg/L of G6P). After 24 h of incubation, the viable cells on each plate were counted. The frequency of resistant cells was calculated using the following equation: The PAP assay was performed in triplicate. Mean values of the frequency of resistant cell were plotted with the standard errors of the means represented by error bars. The heteroresistant subpopulation was defined as the frequency of resistant cells ≥ 10 -7 at eightfold above the antibiotic concentration of the main population. The resistant phenotype without subpopulation was defined as the frequency of resistant cells < 10 -7 at eightfold above the antibiotic concentration of the main population.
The stability of the heteroresistance phenotype was evaluated as previously described with a slight modification 49 . Briefly, A single colony grown on agar supplemented with eightfold above the antibiotic concentration of the main population (the heteroresistance) was inoculated in CAMHB and incubated 37 °C for 24 h with shaking at 120 rpm. A subculture (1:1000) was inoculated and incubated as a previous step twice. After 3 days, the PAP was determined at the eightfold antibiotic concentration of the main population of each isolate. The results were interpreted as the stable heteroresistance (the frequency of resistant at eightfold concentration of the main population < 10 -7 ) and the unstable heteroresistance (the frequency of resistant at eightfold concentration of the main population ≥ 10 -7 ). Statistical analysis. Statistical analysis was performed using GraphPad Prism version 5.0. The OMP expressions and mRNA expression of adeB, adeE, and adeY were performed in triplicated. Mean values of the relative expression were plotted with error bars representing the standard error of the mean (n = 3). The expressions of each couple of the species were compared and calculated using the unpaired two-tailed t-test (*, p-value ˂0.05;