In Vitro Activity of Amphotericin B in Combination with Colistin against Fungi Responsible for Invasive Infections

The in vitro interaction of amphotericin B in combination with colistin was evaluated against a total of 86 strains comprising of 47 Candida species (10 Candida albicans, 15 Candida auris, five Candida glabrata, three Candida kefyr, five Candida krusei, four Candida parapsilosis and five Candida tropicalis), 29 Aspergillus species (five Aspergillus flavus, 10 Aspergillus fumigatus, four Aspergillus nidulans, five Aspergillus niger, and five Aspergillus terreus), and 10 Rhizopus species (seven Rhizopus arrhizus, one Rhizopus delemar and two Rhizopus microsporus) strains. For the determination of the interaction, a microdilution checkerboard technique based on the European Committee on Antimicrobial Susceptibility Testing (EUCAST) reference method for antifungal susceptibility testing was used. Results of the checkerboard technique were evaluated by the fractional inhibitory concentration index (FICI) based on the Loewe additivity model for all isolates. Different inhibition endpoints were used to capture both the interaction at MIC and sub-MIC levels. Additionally, checkerboard technique results for Candida species were evaluated by response surface analysis based on the Bliss independence model. Against common Candida species, the combination was synergistic for 75% of the strains by FICI and for 66% of the strains by response surface analysis. For C. tropicalis, the interaction was antagonistic for three isolates by FICI, but antagonism was not confirmed by response surface analysis. Interestingly, synergistic and antagonistic FICIs were simultaneously present on checkboard microplates of all three strains. Against C. auris the combination was synergistic for 73% of the strains by response surface analysis and for 33% of the strains by FICI. This discrepancy could be related to the insensitivity of the FICI to detect weak interactions. Interaction for all other strains was indifferent. For Aspergillus and Rhizopus species combination exhibited only indifferent interactions against all tested strains.


Introduction
Fungal infections are a leading cause of mortality, especially in immunocompromised patients. Both yeast and filamentous invasive fungal infections are associated with poor outcomes and high mortality rates. In Europe, aspergillosis and mucormycosis are the two

Drugs and Microplate Preparation
European Committee on Antimicrobial Susceptibility Testing (EUCAST) guidelines for antifungal susceptibility testing of yeasts and molds with modifications for broth microdilution checkerboard procedures were used in this study [33,34]. Nunclon TM delta surface 96-wells microtiter plates for adherent cells (Thermo Fisher Scientific, Darmstadt, Germany) were used. Drugs tested in combination were amphotericin B (Merck), and colistin (Merck). Final concentrations tested ranged from 0.03 to 16 µg/mL, and from 1 to 64 µg/mL for amphotericin B and for colistin, respectively. Before the addition of the inoculum, each well contained 100 µL of double strength RMPI medium with 1% (v/v) of DMSO.

Inoculum Preparation and Inoculation of Microplates
All strains were subcultured twice from frozen stocks on Sabouraud dextrose agar slants supplemented with chloramphenicol and gentamycin (Bio-Rad Laboratories, Feldkirchen, Germany) at 35 • C and 95% humidity. Incubation time was 24 h for Candida spp. and 7 days for filamentous fungi in accordance with EUCAST recommendations for slow growing molds [33]. Suspensions were counted in a hemocytometer and adjusted to the final inoculum size of 2 × 10 5 colony forming units (CFU)/mL in water for yeasts, and water containing 0.1% (v/v) of Tween 80 for molds, which should prevent fungal growth on surfaces of the wells [35]. After the distribution of 100 µL of the final inoculum into each well, microplates were incubated at 35 • C, with 95% humidity. Incubation time was 24 h for Candida and Rhizopus species and 48 h for Aspergillus species. After incubation optical densities were read spectrophotometrically at a wavelength of 530 nm using a MultiSkan FC spectrometer (Thermo Fisher Scientific). Before the reading, microplates containing yeast inocula were shaken for 2 min at 1100 rpm with a PMS-1000 Microplate Shaker (Grant Instruments, Shepreth, UK). All experiments were run in duplicate.

Interpretation of the Results by Fractional Inhibition Concentration Index
After subtraction of the blank plates, the optical density values from the microplates were transformed into a percentage of growth compared to the growth control. For yeast, MICs of amphotericin B were determined as the concentration that resulted in an inhibition of 90% [34], and MICs for colistin or in combination that resulted in an inhibition of 50% compared to the growth control (primary inhibition endpoint). Additionally, FICIs for the endpoints of 90% and 50% of inhibition for both drugs and in combination were calculated (additional inhibition endpoints). For molds, a 90% of inhibition endpoint for drugs alone and in combination was chosen. High off-scale MICs were converted to the next log 2 dilution. If the lowest fractional inhibition concentration index (FICI) on the microplate was ≤0.5, or >0.5 to 4 synergy or indifference (no interaction) were assumed, respectively. If a FICI was >4.0, antagonism was concluded [36].

Interpretation of the Results by Response Surface Analysis
The major advantage of the Bliss independence model is its independence of MIC endpoints and MIC definitions, as it compares the effects of drugs alone, or in combination, instead of concentrations. Based on the hypothesis that drugs act independently from each other, the indifference of the combination is achieved, when the sum of the effects of the drugs alone is equal to the effect of the combination. The effect of the combination can be synergistic or antagonistic when the observed effect is better or worse compared to the expected indifferent interaction. Briefly, from the data of the microplates consisting of the percentage of growth compared to the growth control, a dose-response curve for each drug alone is generated. These dose-response curves serve to calculate a theoretical response surface of an indifferent interaction of the two drugs. This surface was then compared to the experimental surface and the synergy distribution was calculated. All calculations were performed by the Combenefit software (Windows v2.02) [37]. The synergy distribution was evaluated using three metrics: the SYN-SUM, the ANT-SUM, and the SUM-SYN-ANT. This later metric consists of the sum of synergy and antagonism observed by comparison of the two surfaces. To determine the threshold of the metric, a response surface with an indifferent interaction was determined experientially. Therefore, the combination of an antifungal with itself (amphotericin B + amphotericin B) was tested by checkerboard in triplicate. Based on the results of the experimental plates, synergy was assumed when the SUM-SYN-ANT was ≥43.8%, and antagonism was assumed when ≤−43.8%. Between -43.8 and 43.8%, indifference was concluded [38]. To determine the SUM-SYN-ANT of the different strains, the results of both runs were combined.

Results
The interactions of amphotericin B with colistin were evaluated by checkerboard against all fungal species. Interpretation of the results by FICI or by response surface analysis against strains of Candida spp. and C. auris are presented in Tables 1 and 2, respectively. A comparison of FICI and response surface analysis for selected C. tropicalis strains is presented in Figure 1. The additionally calculated FICIs using 50% or 90% of inhibition are presented in Tables 3 and 4 for Candida spp. and C. auris, respectively. Interpretation of the results by FICI of strains of Aspergillus spp. and Rhizopus spp. are presented in Tables 5 and 6, respectively. A summary of all results is presented in Figure 2.          Using the primary inhibition endpoint, the 32 Candida strains (except C. auris) exhibited MICs for amphotericin B alone ranging from 0.25 to 0.5 µg/mL (Table 1) with a MIC50, MIC90, and geometric mean MIC of 0.25, 0.5, and 0.35 µg/mL, respectively. Amphotericin B MICs ranged from 0.25 to 0.5 µg/mL for C. albicans, C. glabrata, C. parapsilosis, C. tropicalis and C. kefyr and were 0.5 µg/mL for C. krusei. Colistin showed activity against certain species or strains of the 32 Candida strains (except C. auris) tested. MICs for colistin ranged from 16 to >64 µg/mL (128 µg/mL used as the high-off scale MIC) with a MIC50, and a geometric mean MIC of >64, and 74.48 µg/mL, respectively. The best activity of colistin was seen against C. tropicalis with MICs ranging from 16 to 32 µg/mL. MICs of C. krusei and C. kefyr were 64 µg/mL, except for one strain of each species (MIC of 32 µg/mL). Against C. albicans and C. parapsilosis, colistin was almost inactive, only one strain of each species had a MIC of 64 µg/mL, all other strains had higher MICs. Colistin showed no activity against C. glabrata, all MICs were >64 µg/mL. Between experiments, amphotericin B and colistin MICs were within +/− 1 log 2 dilutions in 100% of the cases for all Candida species tested (data not shown). Interpretation of the results by fractional inhibitory concentration index showed that interaction was synergistic for 75% of the strains with FICIs ranging from 0.1328 to 0.375 with a geometric mean FICI of 0.2312. Synergy was obtained for 40, 60, 67, 80, 90% and 100% of C. tropicalis, C. krusei, C. kefyr, C. glabrata, C. albicans and C. parapsilosis strains, respectively ( Figure 2). All other interactions were indifferent, except for 3 C. tropicalis strains. For these strains, the interaction was antagonistic. Interestingly, synergistic and antagonistic interactions were found on the same plate (Figure 1) with lowest FICIs of 0.1563 and twice 0.3125. The geometric mean FICI for all strains was 0.343, despite the inclusion of the high FICIs from the antagonistic strains.
Analysis of the checkerboard data of the 32 Candida strains (except C. auris) by the response surface approach led to similar results compared to the FICI results. Overall synergy and antagonism were obtained for 66% and none of the strains, respectively (Table 1). The SUM-SYN-ANT metric for the synergistic strains ranged from 45.14 to 87.84, with a mean of 64.74.
Synergy was obtained for 20, 40, 60, 67, 75, 100% of C. tropicalis, C. glabrata, C. krusei, C. kefyr, C. parapsilosis and C. albicans. The geometric mean SUM-SYN-ANT metric for all strains was 48.35. When comparing the results of the FICI with the response surface approach, synergy was obtained for the majority of the strains by both techniques for C. krusei, C. kefyr, C. parapsilosis and C. albicans. Interpretation of the results for C. glabrata was synergistic (three of five strains) by FICI and indifferent by surface analysis (three of five strains). One major difference between the interpretation techniques was that interaction against C. tropicalis was antagonistic (three of five strains) by FICI and indifferent by response surface analysis (four of five strains). Although the SUM-SYN-ANT metric did not reach the determined threshold, there was a trend for an antagonistic interaction by response surface analysis as shown in Figure 1.
Using the primary inhibition endpoint, the 15 C. auris strains exhibited slightly higher MICs for amphotericin B alone than the other Candida spp. ranging from 0.5 to 1 µg/mL (Table 2) with a MIC50, and a geometric mean MIC of 1, and 0.83 µg/mL, respectively. Colistin alone showed no activity against C. auris, all MICs were >64 µg/mL. Between experiments, amphotericin B and colistin MICs were within +/− 1 log 2 dilutions in 100% of the cases for all strains tested (data not shown). Interpretation of the results by fractional inhibitory concentration index led to synergistic interactions for 33% of the strains with FICIs ranging from 0.1563 to 0.375 with a geometric mean FICI of 0.2378. The geometric mean FICI for all strains was 0.3943. The geometric mean MIC for colistin in combination with the synergistic isolates was 5.28, and 1.7 µg/mL for all strains.
Response surface analysis for the 15 C. auris strains led to synergistic interactions for 73% of the strains ( Table 2). The SUM-SYN-ANT metric for the synergistic strains ranged from 47.90 to 84.31, with a geometric mean of 56.38. All other interactions were indifferent. The geometric mean SUM-SYN-ANT metric for all strains was 47.46. When comparing the results of the FICI with the response surface approach, synergy was more frequently obtained (73 vs. 33%) (Figure 2).
Using the additional inhibitions endpoints, synergy was equally or less frequently seen, compared to the primary inhibition endpoint. Combination exhibited synergy for 33 or 13% of the strains using the 50 or 90% of inhibition endpoint, respectively.
The 29 Aspergillus strains exhibited MICs for amphotericin B alone ranging from 0.5 to 4 µg/mL (Table 5) with a MIC50, MIC90, and geometric mean MIC of 2, 4, and 2 µg/mL, respectively. Amphotericin B MICs ranged from 0.5 to 1 µg/mL for A. niger, from 2 to 4 µg/mL for A. flavus, A. nidulans and A. terreus, and were 2 µg/mL for A. fumigatus. Colistin alone showed no activity against Aspergillus species, all MICs were >64 µg/mL. Between experiments, amphotericin B and colistin MICs were within +/− 1 log 2 dilutions in 100% of the cases for all Aspergillus species tested (data not shown). Interpretation of the results by fractional inhibitory concentration index led to indifferent interactions for all the strains tested ( Figure 2). The 10 Rhizopus strains exhibited MICs for amphotericin B alone ranging from 0.5 to 1 µg/mL (Table 6) with a MIC50, MIC90, and geometric mean MIC of 0.5, 1, and 0.57 µg/mL, respectively. Amphotericin B MICs for R. arrhizus and R. delemar were 0.5 and were 1 µg/mL for R. microsporus. Colistin alone showed activity against Rhizopus species with MICs ranging from 16 to 32 µg/mL with a geometric mean MIC of 18.38 µg/mL. Between experiments, amphotericin B and colistin MICs were within +/− 1 log 2 dilutions in 100% of the cases for all Rhizopus species tested (data not shown). Interpretation of the results by fractional inhibitory concentration index let to indifferent interactions for all the strains tested (Figure 2).

Discussion
Colistin is an antibiotic drug of last resort with good penetration of the lungs used for the treatment of pulmonary infections due to multidrug-resistant gram-negative bacteria, such as Pseudomonas aeruginosa, Klebsiella pneumoniae, or Acinetobacter baumanii [39]. Its bactericidal activity is evoked by its ability to target the external membrane, leading to membrane alteration and resulting in increased membrane permeability [40]. Apart from its activity against gram-negative bacteria, cytoplasmic membrane damage has also been demonstrated in C. albicans and R. arrhizus [41,42]. We previously showed in vitro synergy of colistin in combination with isavuconazole for A. nidulans, A. niger and C. auris [30,31], which makes the antibiotic an interesting partner to explore combinations with other antifungals. Therefore, in this study amphotericin B was tested in vitro in combination with colistin against fungi responsible for invasive infections.
Amphotericin B MICs were in the same ranges as previously reported for Rhizopus species [43], C. auris [44], and the different Candida species [45,46]. For Aspergillus species, amphotericin B MICs were in the same range for A. flavus, A. nidulans, A. niger and A. terreus, but not for A. fumigatus [46]. In this study, all A. fumigatus MICs were 2 µg/mL in both runs (data not shown). According to the newest EUCAST breakpoint definition for A. fumigatus from 2020, after the elimination of the status intermediate susceptibility, a MIC of 2 µg/mL would identify an amphotericin B resistant isolate, while in the old definition a MIC of 2 µg/mL would have identified an isolate with an intermediate susceptibility [47]. As it has been shown that spectrophotometric reading is a good alternative for visual reading [48], using 90% or 95% of inhibition as an endpoint compared to the growth control [49], quality controls were within the target range for amphotericin B (Table 1), and that it is unlikely all tested A. fumigatus strains are resistant to amphotericin B, it remains unclear how the interpret the MICs of 2 µg/mL of these isolates.
MICs of colistin alone determined by EUCAST methodology for Aspergillus species and C. auris were the same as previously reported [30,31,50]. Colistin MICs for Rhizopus species by EUCAST methodology have not been determined before, but CLSI methodology MICs were in the same range [41]. Colistin combination MICs for common Candida species ranged from 1 to 2 µg/mL (except for C. tropicalis), which would be in the range of peak serum levels reported in patients with cystic fibrosis [51], and critically ill patients [52]. The geometric mean MIC of colistin in combination with the synergistic C. auris isolates was slightly higher (5.28 µg/mL), but was still in the range of the achievable serum levels.
In this study, we analyzed the checkerboard data of Candida species by interpretation of the results by FICI, or response surface analysis. One of the disadvantages of the FICI technique is its dependence on the MIC endpoints. Another problem is the definition of the endpoint itself, as 50% or 90% of growth inhibition compared to the growth control can be used, using either can lead to completely different conclusions [53,54]. For combination studies, no standardized methods exist, especially if one of the partners belongs to another drug class (in our case antifungal and antibiotic). In this study, we have chosen 90% of inhibition for amphotericin B for Candida species as recommended by EUCAST [34], and 50% for colistin and in combination. EUCAST recommends using 50% of inhibition for all other antifungals except amphotericin B, but of course, colistin is not comparable to other antifungals. To overcome these limitations of the FICI approach, we additionally interpreted the checkerboard results by response surfaces analysis. The great advantage of this approach is its independence of MIC endpoints and definitions, as it compares the effects of drugs alone, or in combination, instead of concentrations [38].
The use of different endpoints for drugs alone and for the combination has already been reported in previous studies [55,56]. The influence of the reading endpoint has also been evaluated in previous studies [57], and showed that using a 90% inhibition endpoint led to less detection of synergy. In the present study, the use of 90% inhibition for amphotericin B and 50% for colistin and combination for the FICI calculation showed the best agreement with the response surface analysis results (75% synergy by FICI and 66% by response surface analysis for Candida spp.) and has, therefore, been chosen as primary inhibition endpoint. The additionally evaluated inhibition endpoints of 50 or 90% of inhibition for both drugs and in combination globally exhibited less synergistic and equal or more antagonistic interactions. Synergy was detected for 56 or 25% of the tested strains, and antagonism for three of five, or five of five C. tropicalis strains, when 50 or 90% of inhibition was used as an endpoint, respectively. The different results obtained with the different endpoints or methods (FICI vs. response surface analysis) could be explained by the fact that using a 50% inhibition endpoint, or response surface analysis may capture interactions at the sub-MIC level that are not captured with the FICI when using a 90% inhibition endpoint.
Apart of the two studies from our laboratories mentioned above [30,31], synergy of colistin in combinations with antifungals has been reported for yeasts [42,50,[58][59][60][61], and filamentous fungi [42,60], but indifference [42,50,61,62], and antagonism [62,63] have also been reported. In this study, using the primary inhibition endpoint, we found synergy of the combination of amphotericin B and colistin for common Candida species except for C. tropicalis by both approaches (75% for FICI and 66% for response surface analysis). Two studies showed synergy for the combination of amphotericin B with colistin, but each study tested only one C. albicans strain [42,61]. These results are in accordance with our study. As previously suggested [42], the membrane damage probably induced by colistin could be enhanced by the known permeabilization of the membrane by amphotericin B, and could, therefore, explain the synergistic effect observed when these two drugs are combined. Another study evaluated the combination of liposomal amphotericin B and colistin against five Candida strains belonging to different species. Unfortunately, only amphotericin B combination MICs were shown, and not colistin combinations MICs, which makes an interpretation of the results impossible [60]. Against C. albicans, the combination of colistin with caspofungin or fluconazole was synergistic in vitro and in vivo in Galleria mellonella [58,64]. Echinocandins were also found synergistic in combination with colistin, but the number of strains tested was limited [61].
Interaction of the combination against C. tropicalis was antagonistic for three isolates and synergistic for two isolates by FICI. Interestingly, synergistic and antagonistic FI-CIs were simultaneously present on checkboard microplates of all 3 antagonistic strains (Figure 1). By definition, if there is at least one FICI ≥ 4, the highest FICI is retained [38]. It is unclear if it has been considered that synergistic and antagonistic interactions can be present on the same microplate when this definition was set-up. Interpretation by response surface analysis showed indifferent interactions for four strains and synergistic for the other. The ANT-SUM of the five C. tropicalis isolates ranges from −8.56 to −16.84, but does not meet the definition of antagonism of −43.8; and certainly not if the SYN-SUM is added. Which interaction of the two approaches represents the reality remains unknown. To answer this question, animal experiments are required.
For C. auris response surface analysis showed 73% of synergy for the combination, while by FICI the combination exhibited synergy for only 33% of the tested strains, using the primary inhibition endpoint. While the geometric mean FICI of all isolates was quite high (0.39), the geometric mean SUM-SYN-ANT was low (47.46). These numbers underline that the synergy of the combination against C. auris is only weak. This could explain the discrepancy between the two approaches, maybe the FICI is not sensitive enough to demonstrate the weak synergy of the combination against C. auris. Two other studies evaluated colistin in combination with antifungals against C. auris. In the first study, the combination of isavuconazole with colistin was synergistic by FICI and response surface analysis, but an agar diffusion assay was not sensitive enough to demonstrate synergy, despite a MIC reduction for the combination of all tested strains compared to the drugs alone [31]. In the second study combination of caspofungin or micafungin with colistin showed synergistic and indifferent interactions, respectively [50].
We found indifferent interactions for the combination against all strains of the tested Aspergillus species using 90% of inhibition for both drugs alone and in combination compared to the growth control. Additionally, sub-MIC evaluation, using an endpoint of 50% of inhibition, showed no significantly different interactions (data not shown). One other study evaluated the combination of liposomal amphotericin B and colistin against three A. fumigatus strains. MICs of amphotericin B in combination were significantly reduced, but it is unclear if combination MICs of colistin were significantly reduced [60]. A combination of colistin with isavuconazole was tested against different Aspergillus species, the synergy of the combination was demonstrated for A. nidulans and A. niger, but agar diffusion assays were not sensitive enough to confirm the synergy. The combination was synergistic for 40% of the tested A. niger strains and indifferent for the rest of the tested A. niger strains, and for all A. nidulans strains tested [30].
One Lichtheimia corymbifera isolate was tested using colistin in combination with amphotericin B or itraconazole. Both combinations exhibited synergy [42]. However, in this study combination of amphotericin B with colistin exhibited only indifference against all Rhizopus species strains tested. As for Aspergillus species, sub-MIC evaluation using an endpoint of 50% of inhibition showed no significantly different interactions (data not shown).
It should be noted that combining two nephrotoxic drugs, such as amphotericin B and colistin may be problematic in patients. Nevertheless, for difficult to treat fungal infections it could be discussed if the benefit of the combination may outweigh the potential toxicity. More importantly, this study is a proof of concept and suggests that drugs active on the bacterial membrane can be synergistic when used in combination with antifungals, and this could stimulate the research and development of new drugs with less nephrotoxicity.
In summary, colistin enhances the in vitro activity of amphotericin B against Candida species, except for C. tropicalis for which the results differed between the interpretation models. Against Aspergillus and Rhizopus species the combination was indifferent for all strains tested. The results of the experiments obtained for the Candida species warrant further in vivo experiments.  Conflicts of Interest: Patrick Schwarz has received research grants from Basilea Pharmaceutica, Gilead and Pfizer, received travel grants from Gilead and Pfizer, and speaking fees from Pfizer. During the past 5 years, Eric Dannaoui has received research grants from MSD and Gilead; travel grants from Gilead, MSD, Pfizer, and Astellas, and speaker's fee from Gilead, MSD, and Astellas. The other authors have none to declare.