Amphotericin primarily kills yeast by simply binding ergosterol

Abstract

Amphotericin B (AmB) is a prototypical small molecule natural product that can form ion channels in living eukaryotic cells and has remained refractory to microbial resistance despite extensive clinical utilization in the treatment of life-threatening fungal infections for more than half a century. It is now widely accepted that AmB kills yeast primarily via channel-mediated membrane permeabilization. Enabled by the iterative cross-coupling-based synthesis of a functional group deficient derivative of this natural product, we have discovered that channel formation is not required for potent fungicidal activity. Alternatively, AmB primarily kills yeast by simply binding ergosterol, a lipid that is vital for many aspects of yeast cell physiology. Membrane permeabilization via channel formation represents a second complementary mechanism that further increases drug potency and the rate of yeast killing. Collectively, these findings (i) reveal that the binding of a physiologically important microbial lipid is a powerful and clinically validated antimicrobial strategy that may be inherently refractory to resistance, (ii) illuminate a more straightforward path to an improved therapeutic index for this clinically vital but also highly toxic antifungal agent, and (iii) suggest that the capacity for AmB to form protein-like ion channels might be separable from its cytocidal effects.

Amphotericin B (AmB) represents an important prototype for small molecules that can self-assemble into membrane-permeabilizing ion channels in eukaryotic cells (1). An advanced understanding of how this natural product interacts with living systems thus stands to enable efforts to develop small molecules that serve as surrogates for deficient or dysfunctional protein ion channels that underlie currently incurable human diseases (2–7). In addition, in stark contrast to most antimicrobial agents (8, 9), resistance to AmB has remained exceptionally rare despite extensive worldwide utilization to treat life-threatening systemic fungal infections for more than half a century (10). Understanding how this natural product kills yeast cells therefore also stands to guide the vital search for new types of resistance-refractory antimicrobial agents. Moreover, although AmB remains the last line of defense in the treatment of life-threatening systemic fungal infections, it suffers from serious dose-limiting side effects (11). Efforts to improve the therapeutic index of this drug would benefit substantially from a more complete molecular understanding of its mode of action.

Despite extensive investigation for the past 50 years, however, the mechanism(s) of action of AmB have remained unclear. In the now widely accepted leading model, AmB kills yeast via ion channel-mediated membrane permeabilization (12–18). This channel model has been extended to also rationalize the unique lack of resistance to AmB as well as its dose-limiting toxicity (12–16). This perceived understanding has stimulated extensive research toward the goals of developing new channel-forming compounds as putative resistance-refractory antimicrobial agents (19) and/or less toxic derivatives of AmB that more selectively form channels in fungal cells versus human cells (12–15). Moreover, this model suggests that the capacity of AmB to form ion channels and exert cytocidal effects are inextricable, which challenges the notion that ion channels formed by AmB or one of its derivatives might have therapeutic benefit in the setting of protein ion channel deficiencies.

Importantly, it has long been known that lipid bilayers containing sterols are uniquely vulnerable to permeabilization by AmB, but it had remained unclear if this effect was due to indirect sterol-mediated global changes in membrane properties or direct sterol binding. We recently determined that AmB directly binds membrane-embedded ergosterol, the main sterol found in yeast cells, in a manner that requires the mycosamine appendage at C19 (2). Deletion of this appendage yields a derivative, amphoteronolide B (AmdeB) (Fig. 1A), which cannot bind ergosterol, is unable to form ion channels, and has no antifungal activity (2, 3). These findings caused us to consider two possible conclusions regarding the mechanism(s) by which AmB kills yeast: (i) ergosterol binding-dependent channel formation is required for antifungal activity, or (ii) ergosterol binding alone is a critical mechanism of antifungal activity, and channel formation represents a second complementary mode of action (Fig. 1B).

Fig. 1.

Chemistry and biology of polyene macrolides. (A) Molecular structures of the polyene macrolide natural products AmB, natamycin, and a series of their functional group-deficient derivatives prepared via chemical synthesis. (B) Representations of two mechanisms for the antifungal activity of AmB: ergosterol binding and membrane permeabilization. (C) Representations of the two leading models for the structure of the AmB ion channel. Both models predict that the hydroxyl group at C35 is critical for ion channel self-assembly. (D) Iterative cross-coupling, a small molecule synthesis strategy, analogous to iterative peptide coupling, in which bifunctional building blocks are sequentially linked using only one reaction recursively.

Albeit in stark contrast with the leading model, several lines of evidence led us to the hypothesis that the latter dual mechanism is operative and that ergosterol binding alone is the primary mechanism of action. First, a dissociation between the membrane-permeabilizing and antifungal activities of AmB has been observed in some strains of yeast (20). To rationalize these findings, a second mechanism involving oxidative membrane damage caused by the putatively redox active heptaene motif of AmB is often invoked (9, 10, 12). However, we found that AmdeB (Fig. 1A), which contains the complete heptaene motif found in AmB and binds very readily to lipid bilayers, is devoid of antifungal activity (2). Thus, an alternative and previously unidentified major mechanism of action for AmB seemed likely. Second, another antifungal polyene macrolide natural product, natamycin (Fig. 1A), was recently found to directly bind ergosterol but not to cause membrane permeabilization (21). Third, ergosterol is critical for many aspects of yeast cell physiology, including vacuole fusion (22), endocytosis (23), pheromone signaling (24), membrane compartmentalization (25), and the proper functioning of membrane proteins (26). Thus, binding alone of this multifunctional lipid should be sufficient for killing yeast cells. Fourth, structural modification and/or decreased expression of ergosterol via mutations in the sterol biosynthesis pathway substantially diminish fungal pathogenicity in vivo (27). Therefore, a primary mechanism involving direct ergosterol binding and sequestration would help explain why clinically significant resistance to AmB is so rare.

Results

Testing the hypothesis that sterol binding is paramount to the antifungal activity of AmB required a derivative of this natural product that retains the ability to bind ergosterol but lacks the capacity to form ion channels. The structures of the AmB/ergosterol complex and the multimeric AmB-based ion channel are unknown, thus making the rational design of such a derivative challenging. However, given its recently established critical role in sterol binding (2), it was clear that the mycosamine appendage should be retained. Alternatively, in both of the leading models for the structure of the AmB ion channel (Fig. 1C) the hydroxyl group at C35 (Fig. 1A) is predicted to be critical for channel formation. Specifically, in the single barrel model (Fig. 1C, Left), an aggregate of AmB molecules spans a dimpled membrane, and the C35 hydroxyl groups anchor the channel to one side of the lipid bilayer (28). In the competing double barrel model (Fig. 1C, Right), two molecular aggregates, each spanning one half of the lipid bilayer, are united into a membrane-spanning channel via hydrogen bonds between the C35 hydroxyl groups on AmB molecules in opposing leaflets (28). Supporting an important role for this hydroxyl group in channel formation, in vitro studies in model liposomes demonstrated that a doubly modified derivative of AmB, in which the C41 carboxylic acid is protected as a methyl ester and the C35 hydroxyl group is removed, caused only weak potassium efflux at high concentration (10 μM) and no efflux at low concentration (1 μM) (29). Our own studies suggested that the weak efflux observed in the former case was likely an artifact of the relatively sensitive nature toward membrane permeabilization of liposomes versus live yeast cells (SI Appendix, Fig. 1) and/or unique biophysical properties resulting from the net positive charge of C41 methyl ester derivatives (SI Appendix, Fig. 2). Thus, with the goal of retaining ergosterol binding, completely eliminating the capacity for membrane permeabilization, and enabling direct comparisons with the biophysical and biological properties of the zwitterionic natural product, we targeted the synthesis of a derivative of AmB that retains the mycosamine appendage at C19, lacks a hydroxyl group at C35, and retains the anionic carboxylate group at C41 (C35deOAmB) (Fig. 1A).

The synthesis of polyene macrolides is challenging because of the sensitivity of these molecules to light, oxygen, and many organic reagents, including the basic and/or acidic conditions often required at the end of a pathway to remove various protective groups. These sensitivities can be magnified considerably when specific functional groups have been deleted from the polyene macrolide skeleton, which has previously precluded the synthesis of some targeted AmB derivatives, including C35deOAmB (29). To improve the efficiency and flexibility with which such complex small molecules can be prepared, we have recently developed a simple and modular synthesis strategy, analogous to iterative peptide coupling, in which building blocks having all of the required functional groups preinstalled in the correct oxidation states and with the desired stereochemical relationships are sequentially linked via iterative application of one mild reaction (30–38). Enabling this approach, we discovered that N-methyliminodiacetic acid (MIDA) is a highly effective ligand for reversibly attenuating the reactivity of a boronic acid and thereby permitting the controlled sequential assembly of bifunctional haloboronic acids (Fig. 1D).

To complete the synthesis of C35deOAmB via this iterative cross-coupling strategy, it was critical to develop a collection of chemically robust building blocks having all of the required functional groups preinstalled in such a way that they can be liberated at the end of the synthesis without affecting the structural or stereochemical integrity of the sensitive polyene macrolide skeleton. This problem turned out to be very challenging. However, after extensive experimentation this goal was achieved in the form of building blocks BB₁, BB₂, and BB₃ (Fig. 2). These three building blocks were all efficiently constructed by using a suite of versatile MIDA boronate reagents and associated methods recently developed in our group (SI Appendix, Figs. 3–5) with the majority of complex building block BB₁ obtained via selective degradation of the natural product (13, 39).

Fig. 2.

Synthesis of C35deOAmB via iterative cross-coupling. Inset shows three building blocks for the synthesis of C35deOAmB having all of the required functional groups preinstalled in the correct oxidation states and with the required stereochemical relationships. A single cross-coupling reaction was used in an iterative fashion to sequentially assemble BB₁, BB₂, and BB₃ with complete stereocontrol. Reagents and conditions are, as follows: a, pinacol, NaHCO₃, MeOH, 45 °C; b, BB₂, 10 mol % PdCl₂dppf, K₃PO₄, DMSO, 23 °C, 53% over two steps; c, BB₃, 5 mol % Pd(OAc)₂, 10 mol % XPhos, aqueous NaOH, THF, 45 °C; d, LiOH, THF : MeOH : H₂O, 35 °C; e, MNBA, DMAP, DCM, 23 °C, 56% over three steps; f, TBAF, THF, 23 °C; g, HF/pyridine, MeOH, 40 °C, 15% over two steps; h, HCl, MeCN : H₂O, 0 °C, 32%; and g, penicillin G amidase, H₂O, 37 °C, 28%. Bn, benzyl; DCM, dichloromethane; DMAP, 4-dimethylaminopyridine; dppf, 1,1′-bis(diphenylphosphino)ferrocene; MNBA, 2-methyl-6-nitrobenzoic acid; PA, phenylacyl; PMP, para-methoxyphenyl; Pyr, pyridine; TBAF, tetrabutylammonium fluoride; TBS, tert-butyldimethylsilyl; THF, tetrahydrofuran; TMSE, trimethylsilylethyl; XPhos, 2-dicyclohexylphosphino-2′,4′,6′-triisopropylbiphenyl.

With these three building blocks in hand, they were readily assembled using a single reaction in an iterative fashion (Fig. 2). Specifically, deprotection of BB₁ followed by selective Suzuki–Miyaura cross-coupling with bifunctional iodotrienyl MIDA boronate BB₂ provided intermediate 1. Subsequent in situ boronate deprotection and a second Suzuki–Miyaura cross-coupling with BB₃, which lacks a C35 hydroxyl group, yielded advanced intermediate 2 possessing the full carbon skeleton of the targeted derivative. Finally, macrocyclization and deprotection of all protic functional groups completed the synthesis of C35deOAmB. In addition to enabling access to this key compound, the successful application of this iterative cross-coupling strategy to such a complex target suggests that this approach has substantial potential to improve synthetic access to a broad range of other complex molecules en route to advanced understanding and/or optimization of their functions.

To further probe the putatively general primary role of ergosterol binding in the antifungal activity of the polyene macrolides, we also prepared via site-selective degradation of natamycin the corresponding natamycin aglycone (Fig. 1A and SI Appendix, Fig. 6) (40). As described above, AmdeB completely lacks the capacity to bind ergosterol, and we expected that the same would be true for this derivative of natamycin lacking its mycosamine appendage.

With all targeted probes in hand (Fig. 1A), we first determined the capacity of each compound to bind membrane-embedded ergosterol via an isothermal titration calorimetry-based assay (Fig. 3A). Consistent with our recent report (2), when a solution of AmB was titrated with either sterol-free or ergosterol-containing 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) large unilamellar vesicles (LUVs), a substantial increase in the net exotherm was observed in the latter experiment, indicating a direct binding interaction between AmB and ergosterol. This assay also confirmed that AmdeB, which lacks the mycosamine appendage, completely lacks the capacity to bind ergosterol. The same series of results were observed for natamycin and natamycin aglycone, demonstrating that the mycosamine appendage is also critical for the ergosterol binding capacity of this natural product. Most importantly, when this pair of titrations was repeated with C35deOAmB, an increase in exotherm equivalent to that observed with AmB was recorded. Thus, the capacity for ergosterol binding is retained in this C35 deoxygenated AmB derivative.

Fig. 3.

Ergosterol binding and membrane-permeabilizing activities of polyene macrolides. (A) Total exotherm, as measured by ITC, for solutions of each polyene macrolide titrated with either sterol-free or 10% ergosterol-containing POPC LUVs. AmB, natamycin, and C35deOAmB all bind ergosterol, whereas AmdeB and natamycin aglycone do not. Values represent the mean of at least three experiments ± SD. * p ≤ 0.035; NS, not significant. (B, C) Potassium efflux from 10% ergosterol-containing POPC LUVs (B) and live S. cerevisiae cells (C) after treatment with polyene macrolides. Percent of potassium release was monitored via a potassium-sensitive electrode and expressed relative to the total potassium released upon the addition of Triton X-100 (B) or digitonin (C) at the end of the experiment. In both LUVs and live yeast cells, robust membrane permeabilization is observed with AmB, whereas AmdeB, natamycin, natamycin aglycone, and C35deOAmB are all inactive.

We next determined whether each of these small molecules can permeabilize lipid bilayers (Fig. 3B). AmB at a concentration of 1 μM produced a rapid efflux of potassium ions from ergosterol-containing POPC liposomes (2). In contrast, AmdeB, natamycin, and natamycin aglycone all showed no ion permeabilizing activity at the same concentration. Importantly, C35deOAmB was also devoid of permeabilizing activity in this in vitro experiment, even at an elevated concentration of 10 μM. We further tested the capacity of all five of these compounds to permeabilize live Saccharomyces cerevisiae cells (Fig. 3C). Again, AmB produced rapid and substantial efflux of potassium ions, whereas AmdeB, natamycin, and natamycin aglycone were completely inactive. Most importantly, C35deOAmB also caused no membrane permeabilization in this in vivo experiment, even at the very elevated concentration of 30 μM.

Collectively, these studies revealed that C35deOAmB retains the capacity to bind ergosterol but completely lacks the capacity to permeabilize membranes, making this derivative a powerful probe for the hypothesis that sterol binding is paramount for the antifungal action of AmB. Moreover, the ability to eliminate sterol binding via deletion of the mycosamine appendages from both AmB and natamycin further enabled evaluation of the putatively dominant and general role of ergosterol binding in the antifungal activity of these polyene macrolide natural products.

We therefore next determined the relative capacities of our five probes to kill S. cerevisiae cells in a standardized broth microdilution assay (Fig. 4A). AmB demonstrated a minimum inhibitory concentration (MIC) of 0.5 μM, whereas its non-sterol-binding counterpart, AmdeB, was completely inactive (2, 3), even at 1,000-fold higher concentration (500 μM). Natamycin was only fourfold less potent than AmB (MIC = 2 μM), whereas, like AmdeB, the natamycin aglycone demonstrated no antifungal activity. Most importantly, C35deOAmB, which like natamycin binds ergosterol but cannot permeabilize membranes, demonstrated substantial antifungal activity (MIC = 3 μM), i.e., this derivative is only sixfold less potent than AmB. Moreover, the MICs for C35deOAmB and natamycin are remarkably similar. A very similar series of results were obtained when these same five compounds were tested against Candida albicans, the species of yeast that is the primary cause of systemic fungal infections in humans. Killing kinetics studies (41) further demonstrated that, like AmB, C35deOAmB is a powerful fungicidal agent against both strains of yeast with small decreases in killing rates (Fig. 4A). Finally, we determined via alkaline extraction and HPLC quantification (42) the number of ergosterol molecules per yeast cell employed in these assays. At the MICs for AmB, there is an order of magnitude more AmB relative to ergosterol for both S. cerevisiae and C. albicans (Fig. 4C). Therefore, there is an overwhelming amount of AmB available at the MICs to bind the ergosterol found in these lipid bilayer membranes and thereby kill the yeast.

Fig. 4.

Antifungal activities of polyene macrolides. (A) MIC as measured by broth dilution. Relative to AmB against S. cerevisiae, C35deOAmB, which cannot permeabilize membranes but retains the capacity to bind ergosterol, is only six times less active. Also note that the MICs for C35deOAmB and natamycin are very similar. A similar series of results is observed with C. albicans. (B) Killing kinetics of S. cerevisiae (Left) and C. albicans (Right) cells exposed to various antifungal agents. In contrast to the fungistatic agent ketoconazole, both AmB and C35deOAmB are powerful fungicidal agents. All compounds were tested at four times their MICs. (C) The number of molecules of AmB and ergosterol, as determined by HPLC, per yeast cell employed in the broth dilution experiments. At the MICs for both S. cerevisiae and C. albicans there is ample AmB available for binding ergosterol and thereby killing the yeast cells. CFU, colony forming units. * p ≤ 0.0003.

Thus, in stark contrast to the now widely accepted channel model, AmB like natamycin (21) primarily kills yeast by simply binding ergosterol. Membrane permeabilization via channel formation represents a second complementary mechanism that further increases potency and the rate of yeast cell killing. Conservation of the mycosamine appendage in the very large family of polyene macrolides combined with the finding that this subunit is required for the sterol binding and antifungal activities of both AmB and natamycin further suggests that mycosamine-mediated ergosterol binding accounts primarily for the antifungal activities of this entire family of natural products.

Discussion

AmB represents the archetype for both ion channel-forming small molecules and resistance-refractory antimicrobial agents. For decades it has been widely believed that these two features are inextricably linked, i.e., that channel formation is a necessary condition for the fungicidal activity of this natural product. The herein disclosed discovery that sterol binding is actually paramount to the antifungal action of AmB has implications in several important areas.

First, microbial resistance to most antibiotics has been observed within a few years of their introduction into clinical medicine, resulting in an emergent global public health crisis (8, 9). Interestingly, microbial resistance long predates modern medicine (43), which suggests that the quest for resistance-refractory mechanisms of antimicrobial action also has a rich natural history. In this vein, despite extensive utilization in the treatment of life-threatening systemic fungal infections for more than half a century, clinically significant resistance to AmB remains exceptionally rare (10, 12, 13). Thus, clinically relevant antimicrobial mechanisms that can evade resistance exist, and the mode of action(s) of AmB is one of them.

In this context, the discovery that AmB kills yeast primarily by binding ergosterol suggests that the simple binding of a functionally vital lipid is an antimicrobial mode of action with substantial potential for evading resistance in the clinical setting. Interestingly, our results support the conclusion that sterol binding is also primarily responsible for the antifungal action of natamycin (21), and several antibacterial peptides exert their effects via specific lipid binding as well (44–47). Collectively, these findings suggest that the potential generality of this mode of action is substantial. Thus, as accelerating advances in lipid biology continue to illuminate many specific lipids that are required for microbial cell physiology (48), such lipids can be viewed as promising new targets in the search for superior antimicrobials that may be less vulnerable to resistance. It is also likely that the possession of a dual mode of action, i.e., lipid binding and membrane permeabilization, which also appears to be more general (46), further contributes to the resistance-refractory nature of AmB. Gaining a more advanced understanding of how all of these processes are related at the molecular level thus also represents a frontier challenge of substantial importance.

Second, the classic mechanistic model in which membrane permeabilization is paramount to the biological activities of AmB has focused extensive efforts to improve the therapeutic index of this clinically vital antifungal agent on increasing its capacity to selectively form ion channels in yeast cells versus human cells (12–14, 16). The discovery that ergosterol binding is in fact primarily responsible for yeast cell killing reveals a much simpler path forward, i.e., increasing the selective binding of ergosterol versus cholesterol, the main sterol found in human cells to which AmB also binds (2), should be sufficient. To further enable this path, more advanced characterization of this small molecule/small molecule interaction, including determination of the binding stoichiometry, will be the subject of future investigations. Moreover, the herein demonstrated capacity for the iterative cross-coupling strategy to efficiently and stereospecifically assemble the polyene core of AmB from readily accessible MIDA boronate building blocks provides a powerful platform for systematic modifications of this substructure in the pursuit of more sterol-discriminating derivatives.

Finally, the prototypical capacity for AmB to form discrete ion channels in eukaryotic cells suggests that small molecules may possess untapped potential to replicate the functions of deficient protein ion channels that underlie a wide range of currently incurable human diseases. Realizing this potential, however, will require the development of ion channel-forming small molecules that are not cytocidal. The discovery that membrane permeabilization represents a relatively minor contributor to the fungicidal action of AmB suggests that such a goal might be achievable. For example, derivatives of AmB that do not bind endogenous sterols yet maintain the capacity for ion channel formation now represent particularly promising targets. Thus, the herein disclosed advanced understanding of how AmB interacts with living systems is cause for increased optimism regarding the prospect of replacing deficient or dysfunctional protein ion channels with small molecule surrogates.

Experimental Methods

All reactions were performed under an inert atmosphere using dry solvents in anhydrous conditions, unless otherwise noted. Full experimental details and characterization for all compounds are included in the SI Appendix. Isothermal titration calorimetry (ITC) studies were performed as described (2) using a Nano ITC instrument with a 190 μL reference cell. Potassium efflux (2), MIC (2), killing kinetics (41), and ergosterol quantification (42) assays were performed as described. Full experimental details and associated references are included in SI Appendix.