Three Dimensional Checkerboard Synergy Analysis of Colistin, Meropenem, Tigecycline against Multidrug-Resistant Clinical Klebsiella pneumonia Isolates

The spread of carbapenem-non-susceptible Klebsiella pneumoniae strains bearing different resistance determinants is a rising problem worldwide. Especially infections with KPC (Klebsiella pneumoniae carbapenemase) - producers are associated with high mortality rates due to limited treatment options. Recent clinical studies of KPC-blood stream infections revealed that colistin-based combination therapy with a carbapenem and/or tigecycline was associated with significantly decreased mortality rates when compared to colistin monotherapy. However, it remains unclear if these observations can be transferred to K. pneumoniae harboring other mechanisms of carbapenem resistance. A three-dimensional synergy analysis was performed to evaluate the benefits of a triple combination with meropenem, tigecycline and colistin against 20 K. pneumoniae isolates harboring different β-lactamases. To examine the mechanism behind the clinically observed synergistic effect, efflux properties and outer membrane porin (Omp) genes (ompK35 and ompK36) were also analyzed. Synergism was found for colistin-based double combinations for strains exhibiting high minimal inhibition concentrations against all of the three antibiotics. Adding a third antibiotic did not result in further increased synergistic effect in these strains. Antagonism did not occur. These results support the idea that colistin-based double combinations might be sufficient and the most effective combination partner for colistin should be chosen according to its MIC.


Three dimensional checkerboard assays
Bacterial cells were cultivated overnight at 35°C at constant rotation (200 rpm) in MH broth. The overnight cultures were adjusted to 0.5 McFarland (equivalent to 10 8 CFU/mL) and diluted 1:100 with broth to obtain a 10 6 CFU/mL suspension. For the checkerboard assay, the broth microdilution method was modified by including some additional antibiotic concentrations. Into each well of a standard microwell plate 100 μL of the 10 6 CFU/mL bacterial suspensions were transferred and mixed with an equal volume of antimicrobial solution. For each strain and antibiotic the selected concentration ranges depended on previously determined MICs. In total, 11 dilution steps of meropenem, 7 dilution steps of tigecycline and 6 dilution steps of colistin were analyzed (Fig 1). The microwell plates were incubated at 35°C for 16-20 h and interpreted according to EUCAST breakpoints for Enterobacteriaceae. Each test was performed at least in duplicate and included a growth control without addition of any antibiotic. The fractional inhibitory concentration index (FICI) for each double (Eq 1) or triple (Eq 2) antibiotic combination was calculated as follows: The FICIs were calculated using the concentrations in the first non-turbid well found in each row and column along the turbidity / non-turbidity interface and expressed as the median value. We used the median averaged FICI values instead of the lowest value to avoid over-interpretation of the synergism due to methodological (one-well) errors of the double dilution method. In this regard, we refer to the FICI interpretation proposed by [14]: FICI < 0.8 synergy, 0.8 < FICI < 4 additive effects or indifference, and FICI ! 4 antagonism. We recorded the lowest FICI values of the combinations ( Table 1, indicated in brackets) as some authors prefer those, but we did not discuss these vales.

Combined disk diffusion test
The disk diffusion test was preformed according to EUCAST rules for antimicrobial susceptibility testing. Bacterial suspensions of 0.5 McFarland were spread over the surface of the MH agar plates (diameter 90 mm) using a sterile swab. On sterile cellulose discs (diameter 6 mm) 10 μL of each antibiotic solution were applied to obtain 40 μg meropenem, 20 μg colistin and 15 μg tigecycline per disc. The three disks where placed at different distances to examine single and combination effects of the antibiotics on bacterial growth. Synergism was defined as extended edge of the inhibition zone of one antibiotic towards the disc of another antibiotic.

Analysis of porin genes
The porin genes ompK35 and ompK36 were amplified and sequenced by using primers described previously [15]. The obtained nucleotide sequences were compared with wild-type ompK sequences of K. pneumoniae strain JM45 (accession number CP006656) available at the NCBI database.

Efflux assay
A novel fluorescent arylidenehydantoin piperazine dye BM-27 was used for the assay (Bohnert JA and Handzlik, J, manuscript in preparation), which is virtually non-fluorescent in aqueous solution but fluoresces strongly upon binding to periplasmic phospholipids. BM-27 was developed in a set of N3-aminealkyl arylidenehydantoin derivatives [16] that are substrates and inhibitors of the MDR efflux pump AcrAB-TolC [17]. 20 mL of the overnight cultures were centrifuged and washed twice with potassium phosphate buffer (PPB) containing 12.5 mM K 2 HPO 4 , 7.8 mM KH 2 PO 4 and 1 mM MgCl 2 (pH 7). The bacterial pellets were resuspended in PPB and adjusted to an optical density of 0.5 at 600 nm. The efflux assay was performed in flat, transparent 96-well plates (Greiner, Germany) using Infinite M200Pro spectrometer (Tecan, Switzerland). In each well 200 μL bacterial suspensions and 15 μM carbonyl cyanide 3-chlorophenylhydrazone (CCCP) (Sigma Aldrich) were mixed and incubated for 8 min at room temperature to disrupt the residual proton gradient of the cells. Bacterial cells were labeled with 50 μM of BM-27. Fluorescence intensity was determined at 400 nm excitation and 457 nm emission. The treatment time varied depending on the labeling of the respective strains and was completed by reaching a constant fluorescence signal (steady-state). By adding glucose to a final concentration of 50 mM the efflux of the BM-27 dye was initiated and the time-resolved decrease of the fluorescence signal was measured. The obtained curves could be best fitted by applying the general fit for one-phase exponential decay (Eq 3) using GraphPad Prism 6.00 (GraphPad Software, La Jolla California USA): Where y is the measured fluorescence signal, x represents the time ordinate, F max represents the fluorescence signal at x = 0; F min represents the fluorescence signal at infinite time. The efflux half-time (EHT) corresponds to the term 0.6932/K and was used as ratio to compare the efflux properties. The labeling efficiency ΔF was given by fluorescence difference of F max to F min .

Statistical analyses
Differences between two groups (e.g. ompK mutation absent or present) for several semi-quantitative FICIs (FICI MER/TGC , FICI MER/CST , FICI TGC/CST and FICI MER/TGC/CST ) were investigated by exact non-parametric Wilcoxon-Mann-Whitney tests and quantified by median differences and 95% confidence intervals (CIs). Similarly we determined non-parametric Spearman's rank correlation coefficients ρ (and 95% CIs) for semi-quantitative variables MICs and efflux with FICIs. We applied a two-sided significance level of 5% and report nominal two-sided p-values. Wilcoxon-Mann-Whitney tests were conducted using the statistical language R version 3.0.3 and the Spearman correlations were calculated using SAS version 9.3.

Analysis of porin genes ompk35 and ompk36
Sequence analyses revealed that all strains exhibited mutations in the ompK genes. Three strains lost OmpK35 and three other strains lost OmpK36 due to nonsense mutations in the respective genes. One strain exhibited only a modified OmpK36. One strain lost both porins whereas 10 strains lost the OmpK35 but harbored modified OmpK36 (Fig 2).
Further sequence modifications within the ompK36 gene were found in some isolates (S2 Table) but the consequences for the function of the proteins were not studied in this work. Amino acid substitutions or in-frame insertions or deletions may lead to altered functions but are unlikely to result in total loss of the porin.

Characterization of the efflux properties
The novel fluorescent dye BM-27 was used to measure the efflux properties of the strains. Contrary to other known membrane probes like Nile red [20], BM-27 is readily water soluble so no organic solvents with potentially antibacterial effects are needed. Since no comparable data exist for K. pneumoniae, we first analyzed and interpreted the distribution of the maximal labeling efficiency by the fluorescent dye as well as the distribution of the efflux half-time (EHT) including ESBL K. pneumoniae isolates with meropenem susceptible phenotype. In general, it is more difficult to label K. pneumoniae strains with any dye compared to e.g. Escherichia coli due to the stronger outer membrane charge [21]. In the present study only one meropenem resistant strain (RKI 412/11) could not be suitably labeled for the efflux assay and therefore we cannot make any conclusions about its efflux properties. For the other strains labeling of > 900 fluorescence units were obtained and efflux curves were evaluable (S1 Fig). The influx and efflux distributions did not differed between meropenem resistant and susceptible isolates (p values 0.247 and 0.879, respectively). The influx properties seemed to be independent of the OmpK-porins since no correlation could be observed between ΔF and the respective porin mutation indicating that BM-27 might use other channels or mechanism to penetrate the cell wall.
Synergism of all three colistin-based combinations (MEM / CST, TGC / CST and MEM / TGC / CST) was found in 5 isolates (Fig 3A-3C). For one additional isolate, only MEM / CST and for another one TGC / CST and MEM / TGC / CST synergies were found. A clearly visible reduction of the FICI MEM/TGC/CST compared to FICI TGC/CST could not be observed for the isolates with synergism (Table 1) Table. The MIC for at least one antibiotic could be strongly reduced in combination for highly resistant strains (Fig 4, S2 and S3 Figs): For the combination of colistin and tigecycline both MICs could be decreased below the EUCAST breakpoints.
Within the six isolates exhibiting synergistic effects (Table 1, FICIs in bold letters), four were found with loss of OmpK36 porin due to two non-senses mutations, one IS5-insertion and one frame shift mutation in the ompK36 gene sequence, respectively. The two remaining isolates showed identical alterations of OmpK36 due to the insertion of glycine and aspartic acid in the L3 structure. Within these two isolates carried only narrow-spectrum and ESBL βlactamases but non carbapenemase.

Disk diffusion test
Disk diffusion tests were performed to supplement the results of the checkerboard assay. We selected 7 isolates exhibiting synergistic effects and 2 isolates that showed no synergism.  Compared to the checkerboard test similar synergistic pattern were found for four strains (RKI 563/13, RKI 178/11, NRZ 04322 and RKI 551/13). For the strains NRZ 00246, RKI 84/14 and RKI 85/14 no clear synergism between colistin and the other antibiotics was visible, but synergistic effects of meropenem and tigecycline were found for these strains (Fig 5). Synergism of meropenem and tigecycline was also visible for two exemplary strains revealing no synergism in the checkerboard assay (NRZ 06142 and RKI 318/11). In this test, growth of mutants inside the inhibition zone of the respective antibiotic (in case of RKI 563/13: meropenem and colistin; RKI 85/14 and NRZ 06142: meropenem; and NRZ 04322 and RKI 84/14: colistin) was suppressed by the activity of the synergistic agonist tigecycline.

Discussion
The number of K. pneumoniae isolates with reduced susceptibility to carbapenems increases annually and often this phenotype is carbapenemase independent. In the present study, to date rarely addressed porin defects were observed in all analyzed K. pneumoniae isolates (Table 1). Carbapenemase producers (KPC, OXA-48 or VIM-1) were consistently highly resistant to meropenem but also the MIC MEM of five isolates harboring only ESBL variant CTX-M-15 or additionally the narrow-spectrum β-lactamase OXA-1 varied from 4 mg/L to 32 mg/L. The increased MICs for meropenem are most probably caused by the loss of the porins OmpK35 and/or OmpK36 as described previously [8,22,23]. Based on our results the proportional contribution of both OmpK-porins to the MIC MEM remains unclear since no explicit pattern, due to the isolated loss of OmpK35 or OmpK36, could be evaluated. However, other studies demonstrated that both porins contribute to the carbapenem resistance in K. pneumoniae [24]. While loss of Omp35 alone was shown to have weak effects, OmpK36 is stronger associated with increased cephalosporine and carbapenem MICs of ESBL and AmpC producers and loss of both porins was shown to strongly elevate the resistance to most cephalosporines and carbapenems [22,23]. Interestingly, isolate RKI 84/14 exhibit an exceptional high MIC MEM of 32 mg/L. Since this isolate showed similar omp-mutation and similar β-lactamase variants compared to isolate RKI 83/14, the reduced meropenem susceptibility seems to be triggered by further unknown factors. In this context, loss of the porins LamB [25] or PhoE [26] that have been shown to contribute to carbapenem resistance or increased expression of the CTX-M gene might be involved but were not investigated in this work. The efflux properties of all study isolates could not been directly correlated to the MIC MEM or MIC CST or MIC TGC .
Using the checkerboard assay synergism of double and triple antibiotic combinations was identified in 30% of the K. pneumoniae isolates with phenotypic carbapenem resistance but was not significantly increased for the triple combination. Similar to a recently published work, no significant correlation between the synergistic effect of colistin / meropenem and loss or modifications of OmpK36 porin could be identified [27].
Retrospective clinical data indicated improved outcome for KPC K. pneumoniae infections for treatments including at least one drug with in vitro confirmed activity [28]. In the present study, synergistic effects were only observed for isolates with high MIC MEM (> 8 mg/L) that simultaneously exhibited high MIC CST (> 8 mg/L) and / or MIC TGC (> 2 mg/L) and thus were classified as resistant to all three antibiotics. It has to be mentioned that the predictive value of in vitro antimicrobial susceptibility testing (AST) is often limited by various methodological and individual factors (focus of infection, co-morbidities or co-infections, antibiotic blood and tissue levels) [29]. Nevertheless, in vitro synergy testing might be helpful to elucidate the bestperforming therapeutic partners. In this study the synergism was dominantly observed in colistin-based combinations, which may be explained as follows: The β-lactam antibiotics act within the periplasmatic space and primary pass though the outer membrane via the major porins OmpK35 and OmpK36 [30]. Tigecycline acts in the cytoplasm by inhibiting the 30S subunit of the ribosome [31] and has to cross both membranes. Colistin is a amphiphilic polymyxin and is known to interact with lipoid compounds and to induce instability and pore-formation in bacterial membranes [32], and thus it might promote the membrane translocation of meropenem and tigecycline. To overcome the increased resistance against periplasmatic or intracellular active antibiotics, higher concentrations of these antibiotics have to be achieved in the respective cell compartment. The colistin enhanced membrane translocation might be only realized in colistin resistant strains that can accumulate higher amounts of colistin without timely killing of the bacteria. This supports the idea that there is a relation between the permeability of the cell wall and membrane and the restored antimicrobial efficacy of meropenem and tigecycline even against resistant isolates by disregarding of the resistance mechanisms due to saturated antibiotic concentrations. A similar mechanism was proposed for the colistin-doripenem-ertapenem combination [27].
In the clinical context, synergism is only of value if the MICs of at least one combination partner are decreased below the respective breakpoint. Therefore we alternatively plotted MICs of the double antibiotic combinations (Fig 4) and compared the effective combined concentrations to EUCAST breakpoints. In our opinion these plots better illustrate the benefits of the respective combinations against individual strains. This allowed conclusions of the impact of each antibiotic to achieve an effective combination therapy. For example: Strain NRZ 00246 was meropenem, tigecycline and colistin resistant, and for all double combinations the FICIs indicated synergism. The simple plots clearly showed that meropenem combined with tigecycline ( Fig 4A) or colistin (Fig 4B) reduced the required concentration of both antibiotics, but combined low concentrations of tigecycline and colistin were within the therapeutic window (and even within the breakpoints) of both antibiotics. Thus this combination seemed to be more effective (Fig 4C).
We could not determine synergistic effects in strains with a low MIC for one of the three tested antibiotics. It cannot be excluded that some synergism may become unnoticed due to the microdilution technique and the chosen concentration, but we favor the hypothesis that in susceptible strains one effective antibiotic is sufficient for growth inhibition in vitro. The plots in Fig 4D-4F demonstrate the distribution of the combined MICs in a susceptible strain, where no increase in FICI values was observed. Not surprisingly, all MICs are located below the breakpoints for the respective antibiotics.
Studies investigating colistin pharmacokinetic (PK) revealed that much higher dosages than the traditionally used 3 x 1 million units are required to achieve a serum concentration above 2 mg/L [33]. However, since colistin can cause serious side effects, increasing dosages is limited by toxicity [34]. In this study, we could show that even in colistin resistant strains a low colistin dosage (below 2 mg/L) is able to reduce the MIC or even restore the susceptibility against meropenem and tigecycline. The studied isolates were all defective in one or two major porins. These mutations are associated with increased carbapenem resistance in K. pneumoniae [7]. Thus most likely the synergistic mechanism is driven by colistin-induced facilitated translocation of meropenem and tigecycline though the outer and inner membranes. In line with our results, a recent retrospective analysis of carbapenemase-producing K. pneumoniae bloodstream infections did not find an advantage of the carbapenem-tigecycline combination but carbapenem-colistin and colistin-tigecycline combinations [35]. However, improved outcome of patients infected by K. pneumoniae under carbapenem-tigecycline combination therapy was observed in other retrospective studies [36,37] suggesting that in vitro data might poorly reflect the in vivo efficiency. A combination therapy, even if not clearly synergistic in in vitro test, might be more effective than a monotherapy and prevent the emergence of resistant mutants as indicated by the disc diffusion assay in this study.

Conclusion
Combination therapy offers a perspective to threat even highly resistant strains. Our in vitro results suggest that the best appropriate combination therapy to treat carbapenem resistant K. pneumoniae might be colistin / tigecycline. In highly resistant strains relevant synergism was detected resulting in a MIC that would have been clinically achievable by conventional dosing.
What do our observations mean for current clinical practice? Colistin seems to be the most effective part of these combinations. Antagonism even in triple combinations is unlikely. Even low colistin concentrations (< 2 mg/L) were able to restore meropenem and tigecycline susceptibility. Therefore, sufficient dosing of colistin may be more important than adding a third drug to a double combination. Adding a third antibiotic to a colistin based double combination might be only useful in vivo if the MIC of colistin is elevated but data confirming these hypothesis are elusive.