The proteomic response of Aspergillus fumigatus to amphotericin B (AmB) reveals the involvement of the RTA-like protein RtaA in AmB resistance

Abstract

The polyene antimycotic amphotericin B (AmB) and its liposomal formulation AmBisome belong to the treatment options of invasive aspergillosis caused by Aspergillus fumigatus. Increasing resistance to AmB in clinical isolates of Aspergillus species is a growing concern, but mechanisms of AmB resistance remain unclear. In this study, we conducted a proteomic analysis of A. fumigatus exposed to sublethal concentrations of AmB and AmBisome. Both antifungals induced significantly increased levels of proteins involved in aromatic acid metabolism, transmembrane transport, and secondary metabolite biosynthesis. One of the most upregulated proteins was RtaA, a member of the RTA-like protein family, which includes conserved fungal membrane proteins with putative functions as transporters or translocases. Accordingly, we found that RtaA is mainly located in the cytoplasmic membrane and to a minor extent in vacuolar-like structures. Deletion of rtaA led to increased polyene sensitivity and its overexpression resulted in modest resistance. Interestingly, rtaA expression was only induced by exposure to the polyenes AmB and nystatin, but not by itraconazole and caspofungin. Orthologues of rtaA were also induced by AmB exposure in A. lentulus and A. terreus. Deletion of rtaA did not significantly change the ergosterol content of A. fumigatus, but decreased fluorescence intensity of the sterol-binding stain filipin. This suggests that RtaA is involved in sterol and lipid trafficking, possibly by transporting the target ergosterol to or from lipid droplets. These findings reveal the contribution of RtaA to polyene resistance in A. fumigatus, and thus provide a new putative target for antifungal drug development.

Introduction

Human health is seriously threatened by fungal infections. These eukaryotic pathogens kill about 2.5 million people per year and infect billions of people globally, ranking on par with well-known bacterial or protozoan pathogens like those that cause tuberculosis or malaria (Brown et al. 2012, Denning 2024). Fungal infections are notoriously challenging to treat. Amphotericin B (AmB), a polyene macrolide antibiotic initially isolated from the soil bacterium Streptomyces nodosus, has been the preferred antifungal medication for treating various systemic mycotic infections over the past six decades (Carolus et al. 2020). The drug is administered intravenously, either as a conventional form complexed to sodium deoxycholate or as a liposomal formulation called AmBisome, which enhances its solubility and availability. AmB exerts its antifungal activity by binding to ergosterol, a key component of fungal cell membranes, which leads to pore formation, ergosterol sequestration, and ultimately cell death due to oxidative stress and loss of membrane integrity (Cavassin et al. 2021). Compared to other antifungal agents, AmB has demonstrated lower rates of resistance due to its fungicidal properties and its ability to target ergosterol, a metabolic product, rather than specific enzymes involved in its biosynthesis (Goncalves et al. 2016, Carolus et al. 2020). The development of resistance to AmB often incurs a significant cost to fungal fitness, as seen in previous studies (Vincent et al. 2013). The most frequent cause of acquired AmB resistance is alterations in the sterol composition of the fungal cell membrane, which results in the loss of the drug target ergosterol. In Candida species, resistance has been linked to genetic mutations in enzymes involved in ergosterol biosynthesis (Martel et al. 2010, Vincent et al. 2013, Ahmad et al. 2019).

AmB and AmBisome are widely used for empirical and salvage treatment of invasive pulmonary aspergillosis (IPA), one of the most severe fungal infections, with mortality rates that can reach up to 90% (Arastehfar et al. 2021). IPA is predominantly caused by inhalation of Aspergillus fumigatus spores (conidia), a saprophytic mold commonly found in the environment. In immunocompromised individuals, inhaled conidia can germinate and cause fatal invasive disease. Despite recent reports indicating high resistance rates in clinical isolates of Aspergillus species (A. terreus, A. flavus, A. niger, A. lentulus, and A. fumigatus), the mechanisms of AmB resistance in Aspergilli are not well understood, raising concerns about the efficacy of AmB as a therapeutic option for IPA (Ashu et al. 2018, Reichert-Lima et al. 2018, Imbert et al. 2020, Fakhim et al. 2022). Therefore, understanding the mechanisms of AmB resistance in A. fumigatus is a pressing need to enable the creation of novel resistance-evading treatment approaches. However, research on this issue has been limited. Manavathu et al. (1998) generated A. fumigatus AmB-resistant isolates by UV irradiation but did not identify the responsible genetic drivers. Furukawa et al. (2020) recently demonstrated that deletion of the negative cofactor 2 transcriptional regulators encoded by nctA and nctB leads to AmB resistance, but the underlying mechanisms are not fully understood. Gautam et al. (2008) used a transcriptomic and 2D gel-based proteomic approach to identify differentially expressed genes and proteins, respectively, in A. fumigatus following 24 h exposure to AmB. However, the obtained findings, such as an increase of ergosterol biosynthetic enzymes, were not further examined for their potential to confer resistance.

To address this knowledge gap, we employed advanced analytical tools to investigate the proteomic response of A. fumigatus to both AmB and AmBisome, since we were also interested in whether the drug formulation affects the proteomic response. We then employed genetic and molecular methods to examine the most strongly elevated proteins. Our approach led to the identification of a novel resistance gene, Afu3g12830 (rtaA), encoding a member of the RTA-like protein family putatively belonging to fungal lipid-translocating exporter proteins (Manente and Ghislain 2009). Our findings reveal that this gene plays a hitherto unknown role in generating AmB resistance, providing a new target for future antifungal drug development and shedding new light on a protein family that has hardly been studied in filamentous fungi.

Materials and methods

Strains and growth conditions

Aspergillus fumigatus strains were grown on Yeast Extract (YAG) solid medium [0.5% (w/v) yeast extract, 1% (w/v) dextrose, 0.01 M MgSO₄, trace elements solution, vitamin mix, 1.5% (w/v) agar] or solid Aspergillus minimal medium (AMM), for 48–72 h at 37°C. Conidia were harvested after 72 h in 0.02% (v/v) Tween 20 in water, then resuspended in double distilled water (DDW), and counted with a hemocytometer. Minimal inhibitory concentrations (MICs) were determined by CLSI M38-A2 broth microdilution methodology in Roswell Park Memorial Institute-3-(N-Morpholino)propansulfonsäure (RPMI–MOPS) broth [10% (v/v) RPMI × 10, 3.45% (w/v) MOPS], or in AMM (Brakhage and Van den Brulle 1995). Spot assays were performed on AMM or RPMI–MOPS agar plates. Transformation of A. fumigatus protoplasts was performed on Yeast extract Peptone Glucose Saccharose (YPGS) agar plates using hygromycin B selection [2% (w/v) yeast extract, 0.5% (w/v) peptone, 2% (w/v) d-glucose, 1 M sucrose, 1.5% (w/v) agar for plates or 0.7% (w/v) for top agar, 200 µg/ml hygromycin B, and pH = 6]. 0.5% (w/v) xylose (Sigma-Aldrich, Rehovot, Israel) was added to MIC and spot experiments to induce the overexpression of rtaA in the RtaA^PXylP and 1% (w/v) xylose (AppliChem, Darmstadt, Germany) to induce the overexpression of rtaA-gfp in the RtaA-GFP^PXylP strain. The A. fumigatus strains used in this study were all constructed in strain ΔKU80 (da Silva Ferreira et al. 2006), here referred to as wild type (WT). Strains are listed in Tables S1 and S2, and their construction and verification are displayed in Tables S3 and S4, and Figs S1–S3.

The A. terreus strains 2023–0030, with low AmB MIC and 2023–1134 with high AmB MIC were grown on solid AMM for 72 h at 37°C. Aspergillus lentulus strains 2018–031 and 2019–599 with low or high AmB MIC were grown on solid AMM for 7 days at 37°C. All strains were provided by the NRZMyk (National Reference Centre for Invasive Fungal Infections, Jena, Germany) and MIC values were tested and are listed in Table S2. Conidia from all A. terreus and A. lentulus strains were harvested as described above. For expression analysis, A. terreus strains were grown in liquid AMM for 40 h and A. lentulus strains were grown for 24 h in liquid AMM.

Scattering-based growth curves were measured using the Cell Growth Quantifier (CGQ; Aquila Biolabs, Baesweiler, Germany) in order to identify concentrations of AmB and AmBisome, which did not lead to a total growth arrest, but significantly inhibited growth. Scattering was measured continuously by the CGQ and the cells were treated with compounds, or solvent for controls, at 20 h of growth. Growth measurements were continued until the culture reached the stationary growth phase.

Proteomic analysis

Fungal mycelium was harvested by filtration with miracloth (Merck Millipore, Darmstadt, Germany), washed with ultrapure water, frozen in liquid N₂ and disrupted by grinding with mortar and pestle. Protein was extracted from the ground mycelium by mixing and sonication in lysis buffer [1% (w/v) SDS, 150 mM NaCl, 100 mM TEAB with cOmplete Ultra Protease Inhibitor Cocktail and PhosSTOP (one tablet per 10 ml, Roche Diagnostics, Mannheim, Germany) and 0.33 U/µl Benzonase (Novagen, Merck, Darmstadt, Germany)] for 30 min at 37°C. Cell debris was removed by centrifugation at 14 000 × g for 15 min at 4°C. The protein concentration of the supernatant was determined using the Direct Detect System (Merck Millipore). 100 µg of protein in 100 µl of lysis buffer were reduced with 2 µl of 500 mM TCEP in 100 mM TEAB for 1 h at 55°C and alkylated with 2 µl alkylation buffer (625 mM iodoacetamide in 100 mM TEAB) for 30 min at room temperature (RT) in the dark. Protein samples were further purified by MeOH–chloroform–water precipitation using the protocol of Wessel and Flügge (1984). Dried protein pellets were solubilized in 100 mM TEAB and subsequently digested with 4 µg Trypsin/LysC protease mix (Promega, Walldorf, Germany) for 18 h at 37°C. Peptides were resolubilized in 0.05% trifluoroacetic acid (TFA)/2% acetonitrile (ACN) in water, filtered through a 10-kDa molecular weight cut-off spin filter (PTFE; Merck Millipore) and stored at −20°C until analysis.

Tryptic peptides were analysed on an UltiMate 3000 RSLCnano system with a QExactive HF mass spectrometer (Thermo Fisher Scientific, Waltham, MA, USA). The peptides were trapped on an Acclaim Pep Map 100 column (2 cm × 75 µm, 3 µm; Thermo Fisher Scientific) for 4 min with a flow rate of 5 µl/min and separated on an Acclaim Pep Map RSLC nano column (50 cm × 75 µm, 2 µm; Thermo Fisher Scientific). Mobile phase gradient elution was performed as follows, with solvent A consisting of 0.1% (v/v) formic acid in water and solvent B of 0.1% (v/v) formic acid in 90:10 acetonitrile/water: 0−5 min at 4% B, 20 min at 7% B, 60 min at 10% B, 100 min at 15% B, 140 min at 25% B, 180 min at 45% B, 200 min at 65% B, 210−215 min at 96% B, and 215.1−240 min at 4% B. A Nanospray Flex Ion Source (Thermo Fisher Scientific) with a stainless steel emitter with 2.2 kV spray voltage was used to generate positively charged ions. The ions were measured in Full MS/data-dependent MS2 (Top 15) mode. Precursor ions were scanned at m/z 300–1500 at a resolution of up to 120 000 full width at half maximum (FWHM), with a maximal injection time of 120 ms and an AGC target of 3 × 10⁶. Precursor ions with a charge state of z = 2–5 were filtered at an isolation width of m/z 1.6 amu and fragment ions were generated using N₂ in the higher-energy C-trap dissociation cell at 30% normalized collision energy. Fragment ions were scanned in a data-dependent manner with a dynamic exclusion of 30 s at 15 000 FWHM and an automated gain control target of 2 × 10⁵; with a maximal injection time of 120 ms. A fixed first mass of m/z 120 amu was used.

MS/MS data were searched against the A. fumigatus Af293 genome of the Aspergillus Genome Database using Proteome Discoverer (PD) 2.2 and the algorithms of Mascot 2.4.1, Sequest HT (PD2.2 version), and MS Amanda 2.0. Two missed cleavages were allowed for tryptic peptides, the precursor mass tolerance was set to 10 ppm, and the fragment mass tolerance was set to 0.02 Da. The dynamic modification was oxidation of Met and acetylation of the N-terminus; static modification was cysteine carbamidomethylation. At least two peptides per protein and a strict target false discovery rate (FDR) of <1% were required for positive protein hits. Label-free protein quantification was based on the Minora algorithm of PD2.2 using a signal-to-noise ratio >5. Normalization was performed by using the total peptide amount method.

Positive protein hits identified in at least two out of three biological replicates of an experiment were further analysed. Proteins with altered abundances in AmB- or AmBisome-treated versus untreated cultures were compared using InteractiVenn (Heberle et al. 2015). Functional annotation assignment, gene ontology (GO) categorization and enrichment analysis of identified proteins were conducted in FungiFun 2.2.8 BETA (Priebe et al. 2015). Volcano plots were created with the aid of the web app VolcaNoseR (Goedhart and Luijsterburg 2020).

For the comparison of the proteome of the WT (CEA17 ΔakuB^KU80) with the ΔrtaA strain, the protocol was modified in the following way:

Sample preparation was performed as mentioned above, except cysteine thiols were reduced and carbamidomethylated in one step for 30 min at 70°C by addition of each 2 µl of 500 mM TCEP [tris(2-carboxyethyl)phosphine] and 625 mM 2-chloroacetamide per 100 µg of total protein in 100 µl. Liquid chromatography tandem mass spectometry (LC–MS/MS) analysis was performed on an UltiMate 3000 RSLCnano system connected to a Orbitrap Exploris 480 mass spectrometer (both Thermo Fisher Scientific) with FAIMS. Peptide trapping for 5 min on an Acclaim Pep Map 100 column (2 cm × 75 µm, 3 µm) at 5 µl/min was followed by separation on an µPACneo 110 column. Mobile phase gradient elution of eluent A [0.1% (v/v) formic acid in water] mixed with eluent B [0.1% (v/v) formic acid in 90/10 acetonitrile/water] was performed using the following gradient: 0 min at 4% B and 750 nl/min, 10 min at 6.5% B and 750 nl/min, 12 min at 7% B and 300 nl/min, 85 min at 25% B and 300 nl/min, 130 min at 50% B and 300 nl/min, 135 min at 96% B and 300 nl/min, 138–140 min at 96% B and 750 nl/min, and 140.1–150 min at 4% B and 750 nl/min. Positively charged ions were generated at spray voltage of 2.2 kV using a stainless steel emitter attached to the Nanospray Flex Ion Source (Thermo Fisher Scientific). The quadrupole/orbitrap instrument was operated in Full MS/data-dependent MS2 mode. Precursor ions were monitored at m/z 300–1100 at a resolution of 120 000 FWHM using a maximum injection time (ITmax) of 50 ms and 300% normalized automatic gain control (AGC) target. Precursor ions with a charge state of z = 2–5 were filtered at an isolation width of m/z 4.0 amu for further fragmentation at 28% HCD collision energy. MS2 ions were scanned at 15 000 FWHM (ITmax = 40 ms, AGC = 200%). Each sample was measured in triplicate with a different compensation voltage (−42, −57, and −72 V).

Tandem mass spectra were searched against the UniProt pan proteome database of A. fumigatus (https://ftp.uniprot.org/pub/databases/uniprot/current_release/knowledgebase/pan_proteomes/UP000002530.fasta.gz; 2024/07/04, YYYY/MM/DD) using PD 3.1 (Thermo) and the database search algorithms (threshold search engine scores in parenthesis) Mascot 2.8 (>30), Comet (>3), MS Amanda 3.0 (>300), Sequest HT (>3) with Precursor Detector node with and without INFERYS Rescoring, and Chimerys (>2). Two missed cleavages were allowed for the tryptic digestion. The precursor mass tolerance was set to 10 ppm and the fragment mass tolerance was set to 0.02 Da. Modifications were defined as dynamic Met oxidation, phosphorylation of Ser, Thr, and Tyr, protein N-term acetylation with and without Met-loss as well as static Cys carbamidomethylation. A strict FDR <1% (peptide and protein level) was required for positive protein hits. The Percolator node of PD 3.1 and a reverse decoy database was used for q-value validation of spectral matches. Only rank 1 proteins and peptides of the top scored proteins were counted. Label-free protein quantification was based on the Minora algorithm of PD 3.1 using the precursor abundance based on intensity and a signal-to-noise ratio >5. Normalization was performed by using the total protein amount method. Imputation of missing quan values was applied by using abundance values of 75% of the lowest abundance identified per sample. Differential protein abundance was defined as a fold change of >2, P-value/ABS(log4ratio) < .05 (P-value/log4ratio) and at least identified in two of three replicates of the sample group with the highest abundance. Statistics and data visualization were performed with R 4.4.0 and RStudio 2024.04.0.

Antifungal susceptibility testing: determination of MICs

MICs were determined by the CLSI M38-A2 broth microdilution methodology of the CaLS Institute (2008). Briefly, stock solutions of either voriconazole (VRZ), AmB, liposomal amphotericin B (AmBisome), or menadione (Mnd) were obtained from Sigma-Aldrich. AmB, Mnd, and VRZ were dissolved in dimethyl sulfoxide. AmBisome was dissolved in sterile water. The antifungal agents were tested over a range of final concentrations (0.015–8 µg/ml for AmB, VRZ, and AmBisome, 0.39–200 µg/ml for MnD). MICs were defined as the lowest concentration of each treatment completely inhibiting fungal growth (observed by light microscopy).

Spot assay

Freshly harvested conidia were serially diluted in DDW to concentrations of 10⁷, 10⁶, 10⁵, and 10⁴/ml. 10 µl of conidial dilutions were spotted on AMM or RPMI–MOPS agar plates (with/without xylose where required) in the presence of the specified stress-inducing agents as indicated. Thermotolerance was tested at 50°C and hypoxic conditions were established at 0.2% (v/v) O₂ and 5% (v/v) CO₂ in an H35 Hypoxystation (Don Whitley Scientific, Bingley, UK). Growth was documented after incubation for 48 or 60 h at 37°C depending on the growth rate.

Ergosterol and sphingolipid analysis by LC–TQ

Ergosterol was separated using the Prominence high performance liquid chromatography system consisting of the controller CBM-20A, autosampler SIL-20AC, column oven CTO-20A, pump LC-20AD, and degasser LC-20A5R (Shimadzu, Duisburg, Germany). Collected biomass was inactivated in 2 ml HCCl₃. After addition of 1 ml H₂O, 1 ml methanol and 10 µl 10 mM cholesterol as internal standard (Merck KGaA, Darmstadt, Germany), samples were mixed in a multitube vortexer (VWR, Darmstadt, Germany) for 10 min and subsequently centrifuged for 5 min at 3000 × g. The lower HCCl₃ phase was transferred into a new glass centrifuge tube. The extraction was repeated with additional 2 ml HCCl₃. The combined HCCl₃ phases were evaporated and redissolved in 1 ml HCCl₃/methanol [1:4 (v/v)]. Triplicates of 10 µl portions were diluted to a final volume of 100 µl in sample vials. 10 µl were applied on a MultoHigh 100 RP18-3 60 × 2 mm column (CS Chromatographie Service GmbH, Langerwehe, Germany) maintained at 35°C. The column was equilibrated with 90% of a 1% aqueous formic acid solution and 10% (v/v) methanol at a flow rate of 400 µl/min. The eluent changed to 100% methanol after injection. The flow rate gradually increased to 800 µl/min between 3 and 5 min. Mass spectrometric determination of ergosterol (m/z 379 > 69) and the internal standard cholesterol (m/z 369 > 161) was performed using the QTrap triple quadrupole mass spectrometer (Sciex, Darmstadt, Germany) run in the multiple reaction monitoring mode using positive atmospheric pressure chemical ionization at 450°C. Data analysis was performed using Analyst 1.6.3 (Sciex). Sphingolipid metabolites were determined as described in Bode and Gräler (2012).

Mouse model of IPA

Six-week-old ICR female mice were supplied by Envigo (Indianapolis, IN, USA). Mice were immunocompromised by two subcutaneous injections with 300 mg/kg cortisone acetate (CA), given 3 days before infection, and on the day of infection (CA model) or injected subcutaneously with cyclophosphamide (CP; 150 mg/kg) and CA (150 mg/kg) 3 days prior to infection, and with CP (150 mg/kg) on the day of infection, 3- and 6-days postinfection (CP model). Mice were housed in sterile cages with sterile bedding and provided with sterile feed and drinking water. The mice were anaesthetized by intraperitoneal injection with 100/10 mg/kg ketamine + xylazine and infected intranasally with 5 × 10⁵ conidia of WT, rtaA^C, and ΔrtaA strains, suspended in 20 µl of 0.2% (v/v) Tween 20 in saline solution [0.9% (w/v) NaCl]. Endpoints for sacrifice were defined by a drop of >15% in body weight and/or signs of acute distress. Mice were monitored for 28 days. The study was carried out in accordance with the recommendations of the Ministry of Health (MOH) Animal Welfare Committee, Israel. The protocol was approved by the MOH Animal Welfare Committee, Israel (protocol number MOH 01–17-035).

LC–MS/MS analysis of natural products

Extraction and detection of natural products were carried out as described previously (Stroe et al. 2020). Identification of fumicyclines (also referred to as neosartoricins) was achieved by comparison with the authentic reference neosartoricin (Biomol, Hamburg, Germany).

Fluorescence microscopy

For fluorescence microscopy of filipin and Nile red-stained hyphae, ibidi channel slides (ibidi GmbH, Gräfelfing, Germany) were inoculated with 1 × 10⁴ conidia in AMM in the presence or absence of 0.04 or 0.08 µg/ml AmB and incubated overnight at 37°C in a CO₂ incubator. On the next day, the mycelium was fixed for 20 min at room temperature with 3.7% (v/v) formaldehyde/phosphate-buffered saline (PBS) followed by a washing step. Afterwards the samples were stained with 50 µg/ml of the sterol stain filipin-III (Sigma-Aldrich, Taufkirchen, Germany) and 10 µg/ml of the fluorogenic lipid dye Nile red (Sigma-Aldrich) for 10 min at room temperature. After a final washing step, the slides were analysed using a Zeiss LSM 780 confocal laser scanning microscope (Carl Zeiss GmbH, Jena, Germany) and evaluated by a bioinformatic deep learning (DL) approach as described in the following.

A DL-based approach has been employed to identify A. fumigatus hyphae using transmitted light (TL) images. To train the DL network, 3D images of A. fumigatus WT hyphae labelled with Concanavalin A-FITC were generated using a Zeiss LSM 780 confocal laser scanning microscope (Carl Zeiss GmbH). In parallel, A. fumigatus was visualized via TL bright field microscopy. Images were acquired as z-stacks with a bit depth of 16. Each image covered an area of 224.92 µm² × 224.92 µm² at a pixel resolution of 0.109 µm × 0.109 µm. Since the required number of z-layers of the stacks depended on the vertical growth of the hyphae, each stack consisted of the proper number of slices to cover the entire structure, resulting in ~13–46 slices per z-stack with 0.75 µm axial slice distance. In total, 16 z-stacks of WT samples were provided for training the DL model.

For the intensity-based analysis, 2D images of Nile red-stained intracellular lipid vesicles and filipin-stained ergosterol were recorded alongside the TL images. Three to five images were acquired per sample, each originating from separate, nonoverlapping regions of the sample. The dimensions of each image were 224.92 µm² × 224.92 µm², capturing details at a resolution of 0.219 µm × 0.219 µm.

The software Huygens Professional (SVI, Hilversum, Holland) was used to correct for chromatic aberration in the 3D images, with an estimated aberration value of zero (for more details about the DL approach see the Supplementary material and below).

For microscopy analysis of the RtaA-GFP^PXylP strain, 8-well chamber µ-slides (ibidi GmbH) were inoculated with 1 × 10³  A. fumigatus conidia in AMM containing 1% (w/v) xylose in the presence or absence of 0.04 µg/ml AmB and incubated overnight at 37°C in ambient air. After incubation, the mycelium was either stained with 10 µg/ml Nile red (Sigma-Aldrich) for 10 min at room temperature or with 5 µg/ml FM 4–64 (Thermo Fisher Scientific, Dreieich, Germany) for 10 min on ice. The stained slides were analysed on a Zeiss LSM 780 confocal laser scanning microscope (Carl Zeiss GmbH).

Nucleic acid hybridization

Gene deletion of rtaA as well as green fluorescent protein (GFP) and xylose-promoter integration of RtaA-GFP^PXylP was verified by Southern hybridization. Briefly, genomic DNA of A. fumigatus was extracted from ground mycelium using the MasterPure Complete DNA and RNA purification Kit (Lucigen, LGC group, Teddington, UK) and digested by restriction enzymes as detailed in the Supplementary material. Resulting DNA fragments were separated by agarose gel electrophoresis and transferred onto Hybond-N+ membranes (GE Healthcare Bio-Sciences, Munich, Germany) by capillary blotting. Gene-specific DNA probes were generated by polymerase chain reaction (PCR), including digoxigenin (DIG)-labelled dUTPs (Jena Bioscience, Jena, Germany). DIG-labelled probes were hybridized in 20 ml hybridization buffer [19 ml 5x SSC, 0.1% (w/v) SDS, 50 g/l dextrane sulfate, and 1 ml Western Blocking Reagent (Roche Diagnostics)] and detected using anti-DIG AP fab fragments (Roche Diagnostics) and CDP-Star as chemiluminescence substrate (Roche Diagnostics). The blot was imaged by an Octoplus QPLEX CCD camera (NH DyeAGNOSTICS, Halle, Germany). Oligonucleotides used for amplifying DNA probes are listed in Table S3.

For northern hybridizations, RNA was isolated from A. fumigatus mycelium using the RNeasy Plant Mini Kit (Qiagen, Hilden, Germany). 10 µg of total RNA were separated on a denaturing agarose gel [1.2% (w/v) agarose, 40 mM MOPS, 10 mM sodium acetate, 2 mM EDTA, 2% (v/v) formaldehyde, pH 7)]. Blotting, hybridization, and detection were done as described above. The primers used to generate the DNA probe specific to the 5′ region of the gene rtaA are listed in Table S3.

RT-qPCR

For real time quantitative PCR (RT-qPCR) analysis of rtaA expression in different Aspergillus species, A. terreus, A. lentulus with low or high AmB MIC as well as A. fumigatus WT and A. fumigatus RtaA^PXylP strains were grown in the presence or absence of 0.125 µg/ml AmB. RNA was extracted from ground mycelium using the Universal RNA Kit (Roboklon, Berlin, Germany) followed by DNase digest utilizing the TURBO DNA-free Kit (Thermo Fisher Scientific). cDNA was synthesized using the RevertAid first strand-cDNA-Synthesis Kit (Thermo Fisher Scientific). All qPCR reactions were carried out on a CFX Duet Real-Time PCR System (BioRad, Feldkirchen) equipped with the CFX Maestro Software 2.3, using the Luna Universal qPCR Master Mix (New England Biolabs, Frankfurt am Main). The A. fumigatus rtaA orthologues ATEG_04110 (A. terreus) and IFM58399_04074 (A. lentulus) were used for species-specific primer design. Three biological replicates for every strain used were measured in three technical replicates each. All genes were normalized to the expression of the reference gene β-actin using the 2^−∆∆Ct method (Livak and Schmittgen 2001). All RT-qPCR primers had an efficiency between 90% and 110%. Primers for target amplification, as well as their efficiency, are listed in Table S3.

Amino acid alignments and phylogenetic analysis

The amino acid sequence of RtaA (Afu3g12830) was analysed for protein family classification and domain prediction using InterPro (Paysan-Lafosse et al. 2023). Pairwise sequence alignment with Saccharomyces cerevisiae RTA1 was performed using EMBOSS Stretcher (Myers and Miller 1988) and transmembrane topology prediction using DeepTMHMM (Hallgren et al. 2022). For phylogenetic analysis of orthologues of Afu3g12830 and all fungal proteins containing RTA1 domain were identified using OrthoMCL release 6.21 (Chen et al. 2006). For the phylogenetic reconstruction of all RTA1 domain-containing fungal proteins, orthogroup sequences were included in the alignment if the orthogroup contained >4 sequences and >97% of the sequences in the orthogroup contained an annotated RTA1 domain (PFAM number PF04479). Amino acid sequences were aligned using MUSCLE v3.8 (Edgar 2004). The resulting alignment was trimmed using ClipKit v2.3.0 (Steenwyk et al. 2020) in smart-gap mode. A maximum-likelihood phylogeny was inferred using IQ-TREE v2.2.0.3 (Minh et al. 2020). ModelFinder in IQ-TREE was used to identify Q.pfam+G4 as the best substitution model for both phylogenies based on Bayesian information criterion. Bootstrapping was performed using 1000 ultrafast bootstraps. The phylogeny was visualized using ggtree 3.11.1 (Yu 2020).

Image data analysis

To identify hyphae based on the TL, a DL segmentation model was trained on pairs of TL images and their corresponding Concanavalin A-stained channel (Fig. S10A). The dataset included 16 z-stack images, totaling 400 2D image slices. Of the 16 z-stacks, 15 3D images (corresponding to 357 2D slices) were used for training and validation. The remaining one z-stack was reserved for model testing and evaluation. Given the challenges of working with a limited dataset in training DL models, data augmentation methods, such as image rotation, flipping, translation, cropping, and blurring were applied (Mikołajczyk and Grochowski 2018, Shorten and Khoshgoftaar 2019). This approach expanded the training dataset to 5712 2D images.

To obtain the hyphal structure masks, several preprocessing steps were applied to the Concanavalin A-stained fungus, including contrast enhancement and noise reduction. Subsequently, the enhanced images were converted into a binary mask (Fig. S10B). For the training and validation stages, the EfficientNetB5-UNet++ model was chosen due to its known efficiency and precision in segmentation tasks (Ronneberger et al. 2015, Zhou et al. 2018 , Tan and Le 2019, Lourenço-Silva et al. 2022). The model showed significant accuracy, with a Dice score of 83.08% on the test data (Fig. S10C).

The hyphal masks were generated from the TL channel by the trained DL model (Fig. S10D). Subsequently, the binary masks were used to record the corresponding average intensity values of Nile red and filipin across the entire mask. Moreover, Otsu thresholding was applied to filter out low-intensity values of the foreground pixels, thereby emphasizing regions characterized by higher illumination associated with lipid concentration (Fig. S10E). More details are provided in the Supplementary material.

Statistical analysis

Data and statistical analysis were analysed with the GraphPad Prism 8 software package (GraphPad Software, Inc, San Diego, CA, USA) or the Microsoft Excel software package (Microsoft Corporation, Redmond, WA, USA). The ANOVA test was used for significance testing of two groups. Differences between the groups were considered significant at P ≤ .05. Mortality results were analysed by the log-rank test for Kaplan–Meyer survival curve, but t-test was used for significance testing of two groups. Differences between the groups were considered significant at P ≤ .05. Statistical analysis for intensity-based image analysis was conducted using SciPy v1.10.1 library in Python v3.10. To compare multiple independent samples, Kruskal–Wallis test with Benjamini–Hochberg adjustment was employed. Significance was defined as a P-value < .05. Statistical test results were included in the figure legends as follows: *P < .05, **P < .01, and ***P < .001. Effect sizes were calculated as Hedges’ g, where a value of ~0.2 indicates a small effect, 0.5 suggests a moderate effect, and 0.8 or higher indicates a large effect.

Results

Proteomic analysis of A. fumigatus response to AmB and AmBisome

To gain a better understanding of the proteomic response of A. fumigatus to AmB and its liposomal formulation AmBisome, we first determined a suitable, sublethal concentration for both antifungals, which induced a robust growth inhibition, but did not lead to a complete growth arrest. Therefore, growth curves of A. fumigatus cultures treated with different concentrations of AmB and AmBisome in comparison to Dimethyl sulfoxide (DMSO) controls were measured, using backscattered light detection. Based on the results, for subsequent proteomic analyses concentrations of 0.125 µg/ml AmB and 7 µg/ml AmBisome were selected, which met the criteria (Fig. S4). To further investigate the proteomic response of A. fumigatus to AmB, we conducted additional experiments in which we exposed the fungus to 0.125 mg/l AmB for different time intervals and analysed the resulting proteomic changes. Specifically, we performed three technical replicates each at time points of 1, 2, and 4 h and set a threshold of ≥4-fold due to the limited number of replicates. The 4-h exposure time resulted in the highest number of proteins (97 out of 3604 proteins) with significant changes in abundance, whereas only a small number of proteins showed exclusively significant changes after 1 and 2 h of AmB exposure, but not at 4 h. (Fig. S5). For this reason, we investigated the proteomic response of A. fumigatus to AmB and AmBisome after 4 h exposure. The LC–MS/MS-based proteomic analyses identified a total of 3680, 3683, and 3448 different proteins in the AmB-treated, AmBisome-treated, and untreated A. fumigatus samples, respectively (proteins present in at least two out of three biological replicates). We observed increased levels of numerous A. fumigatus proteins (≥2-fold) in response to AmB, including 202 and 193 proteins upon AmB and AmBisome exposure, respectively (Fig. 1A, Table S7). The level of 94 proteins increased significantly under both conditions in comparison to untreated A. fumigatus cultures (Fig. 1B, Table S7). In contrast, the number of proteins showing reduced abundance in response to antifungal exposure was considerably smaller, with 70 for AmB and 83 for AmBisome, and an overlap of 30 proteins for both drug formulations (Fig. 1A and B, Table S7).

Protein set enrichment analysis (PSEA) indicated that enzymes involved in oxidation–reduction processes, secondary metabolite biosynthesis, and aromatic amino acid metabolism were significantly enriched upon exposure to AmB, whereas nitrate transport and nitrate reduction activity were notably decreased (Fig. 2A). The addition of AmBisome resulted in a similar proteomic response compared to that of AmB. However, in contrast to AmB, the addition of AmBisome also resulted in an enrichment of proteins involved in the response to iron starvation and ergosterol biosynthesis (Afu2g00320, Erg3) among the proteins with increased abundance (Fig. 2B, Table S6). Interestingly, the level of reactive oxygen species (ROS)-detoxifying enzymes did not increase under both conditions with the exception of a predicted glutathione S-transferase Afu2g17300. Catalase CatB even decreased in abundance by 1.5–2-fold in response to both amphotericin formulations.

The PSEA data revealed that AmB and AmBisome induce the production of secondary metabolites in A. fumigatus (Table S6). The level of four out of six proteins (FccB, FccC, FccA, and FccD) encoded by the fumicycline gene cluster, significantly increased by a factor of 3–23-fold after AmB and AmBisome exposure (Table S6). The induction of fumicycline production by AmB was also verified at the metabolome level by LC–MS/MS analysis (Fig. 2C). Notably, the fumicycline gene cluster of A. fumigatus is known to be induced by interaction with the Gram-positive soil bacterium Streptomyces rapamycinicus (König et al. 2013). Likewise, the expression of enzymes involved in the biosynthesis of the copper-binding isocyanide xanthocillin (XanA, XanB, XanE, XanF, and XanG) and the iron(III)-chelating complex HAS (HasB, HasC, and HasH) was found to be significantly upregulated (Table S6) (Yin et al. 2013, Lim et al. 2018). Collectively, these findings showed that the antifungal AmB derived from the secondary metabolite produced by S. nodosus triggers the activation of secondary metabolite biosynthesis gene clusters in A. fumigatus.

Conversely, exposure to AmB and AmBisome resulted in a significant decrease in the levels of Pyr6 (Afu6g13980) and Pyr1 (Afu6g13920), which are enzymes involved in the biosynthesis of the secondary metabolite pyripyropene A (Itoh et al. 2010) (Table S7). The AMP-activated protein kinase Afu6g04500, which is part of the Snf1-complex involved in regulating adaptation to glucose limitation in S. cerevisiae, was also found to have significantly decreased in abundance (Shashkova et al. 2015).

Among the 14 proteins with the highest increase in abundance following exposure to AmB and AmBisome, it is evident that several secondary metabolite biosynthesis enzymes were included, such as the xanthocillin biosynthesis enzymes XanB and XanE, as well as the alcohol dehydrogenase Afu5g10220, which is part of an uncharacterized NRP-like cluster (Table 1, Table S7). In addition to secondary metabolite biosynthesis enzymes, the list of the top 14 most abundant proteins included membrane transporters (Afu1g05010/Mfs56, Afu1g12620, and Afu6g03690/Ena1) and a putative glycosylphosphatidylinositol (GPI) transamidase responsible for attaching GPI-anchors to proteins (Afu4g08200/PIG-U). The corresponding orthologous gene GAB1 has been demonstrated to be essential in S. cerevisiae (Grimme et al. 2004). Moreover, the list of the top 14 most abundant proteins also includes a putative antimicrobial peptide (Afu8g00710), the aforementioned glutathione S-transferase (Afu2g17300), which is generally involved in detoxification of ROS and xenobiotics, and several uncharacterized proteins (Afu4g09230, Afu5g00800, and Afu6g11790). The gene hppD, which encodes the enzyme 4-hydroxyphenylpyruvate dehydrogenase (Afu2g04200), is part of a gene cluster (Afu2g04200–Afu2g04262) known to be involved in the degradation of tyrosine (Schmaler-Ripcke et al. 2009). Additionally, the enzymes HmgA, FahA, and MaiA, which are further involved in this degradation pathway, were also activated 3.6–10.1-fold after AmB treatment (Table S7). A remarkably substantial increase in abundance of more than 300-fold was observed for an RTA-like protein encoded by the gene Afu3g12830 after AmB or AmBisome exposure and designated here as RtaA. To investigate the possible involvement of these proteins in AmB resistance, we selected nine proteins for further analysis (Table 1, in bold) and generated knockout mutant strains of the corresponding genes (for more details see the Supplementary material).

The ΔrtaA mutant strain is more sensitive to AmB and AmBisome

To assess the susceptibility of the deletion strains to AmB and AmBisome, we conducted an initial evaluation in comparison to VRZ and Mnd controls using the CLSI M38-A2 broth microdilution methodology of the CaLS Institute (2008). Table 2 presents the MIC values in µg/ml of AmB, AmBisome, VRZ, and Mnd after 48 h of incubation at 37°C in RPMI–MOPS medium for both the WT strain and the deletion mutants. The results show that only the ΔAfu3g12830 (ΔrtaA) mutant strain exhibited a 4-fold increased sensitivity to AmB and AmBisome, with MIC values of 0.125 µg/ml and 0.03 µg/ml, respectively, compared to 0.5 µg/ml and 0.125 µg/ml for the WT strain. The complemented rtaA null mutant (rtaA^C) exhibited WT MICs for AmB and AmBisome, indicating that the reintroduction of the rtaA gene restored the WT phenotype. We further assessed the susceptibility of these strains to AmB by point inoculation after 60 h of incubation at 37°C in AMM agar (Fig. 3). In agreement with the broth microdilution results, ΔrtaA exhibited increased AmB susceptibility compared to the parental and reconstituted rtaA^c strains. These findings suggest that RtaA contributes to the A. fumigatus stress response to AmB and its absence leads to AmB hypersensitivity. In contrast, RtaA has no effect on the sensitivity of the fungus to other antifungal agents tested here. To further characterize the role of RtaA in A. fumigatus, we assessed the growth of the ΔrtaA strain under various stress conditions, compared to the WT and the rtaA^c-complemented strain (Fig. 4). For this purpose, the effects of cell wall- (Congo red, calcoflour white, caspofungin, and tunicamycin) and membrane-interfering compounds [SDS, phytosphingosine, α-tomatin, miltefosine, and itraconazole (ITZ)], and of oxidative stress (4 mM H₂O₂), hypoxia [0.2% (v/v) O₂], alkaline growth conditions (pH 8.0), heat (50°C), osmotic stress (1 M NaCl), iron depletion (-Fe plus 0.2 mM BPS), and reductive stress (10 mM DTT) were monitored. We did not observe any differences in growth between the ΔrtaA and the WT strain under any of these conditions. However, a 2-fold increase in susceptibility to the polyene antifungal agent nystatin was visible for the ΔrtaA strain (MIC of 0.8 µg/ml) compared to the WT (MIC of 1.6 µg/ml), indicating that RtaA specifically protects A. fumigatus to some extent against polyene antifungals such as AmB, AmBisome, and nystatin.

The protein sequence of A. fumigatus RtaA contains seven transmembrane domains and shows sequence similarity to the S. cerevisiae protein Rta1

RtaA belongs to the RTA-like protein family (PF04479/IPR007568), which consists of fungal proteins with multiple transmembrane domains. Some members of this protein family are known to provide resistance against toxic compounds and are also designated as fungal lipid-translocating exporter proteins (Manente and Ghislain 2009). To compare RtaA with another RTA-like protein, we performed a pairwise sequence alignment with the S. cerevisiae protein Rta1p/YGR213C using EMBOSS Stretcher (Fig. 5A). Rta1p is known to confer resistance against 7-aminocholesterol (Soustre et al. 1996). Our results showed that RtaA shared 42.4% sequence similarity with Rta1p, with an identity of 29.1%, and that both proteins contained seven transmembrane domains. The paralogous S. cerevisiae protein YLR06C also exhibited equal sequence similarity with the RtaA protein of A. fumigatus. Interestingly, the A. fumigatus genome encodes 23 different members of the RTA-like protein family (InterPro entry IPR007568; https://www.ebi.ac.uk/interpro/) highlighting the potential importance of this protein family for the physiology and adaptation ability of this fungus (Fig. S6). Nevertheless, none of the other 22 different RTA-like proteins of A. fumigatus was among the 4329 proteins identified in our study, which suggests that they are mainly produced under specific stress or growth conditions.

To investigate the evolutionary relationship of RtaA, we identified 176 orthologues that were present in the Ascomycota subphylum Pezizomycotina and the classes Eurotiomycetes, Sordariomycetes and Dothiomycetes, including some of the most important human- or plant-pathogenic fungi. Phylogenetic analysis revealed that RtaA has the closest similarity to orthologues of other Aspergillaceae, namely many Aspergillus species and Penicilliopsis zonata (Fig. 5B; Supplementary data). No sequences from the Saccharomycetes class were among the identified orthologues, indicating that S. cerevisiae Rta1 is not an orthologue of the RtaA protein. To analyse the phylogenetic relationships of all 23 RTA-like proteins of A. fumigatus, we constructed a phylogenetic tree from more than 2500 RTA-like domain-containing sequences in fungi (Fig. S6; Supplementary data). This tree clearly showed that S. cerevisiae Rta1 and A. fumigatus RtaA are only distantly related to each other and that the RTA-like protein Afu6g09550 is the most closely related A. fumigatus protein to Rta1.

RtaA localizes primarily to the cytoplasmic membrane

To investigate the cellular distribution of RtaA–GFP, we generated an A. fumigatus strain, which expresses an RtaA–GFP fusion protein under the control of the xylose-inducible xylP-promoter PXylP in the ΔrtaA background. Previous attempts with the native rtaA promoter resulted in strains exhibiting only weak GFP signals, even with the addition of sublethal concentrations of AmB (data not shown). During hyphal growth, the RtaA–GFP fusion protein, whose production was induced by xylose in the medium, was visible at the cell periphery of hyphae and septa and in larger roundish structures. There were occasional colocalizations of the GFP-signal with Nile red-stained lipid droplets (indicated by arrows in Fig. 5C). However, there was no observable colocalization with FM4-64-stained endosomes (Fig. 5C, arrows show specific endosomes), what speaks against the involvement of RtaA in endocytic membrane transport processes. The addition of 0.04 µg/ml AmB did not have a significant impact on the distribution of RtaA (Fig. S7). These data confirmed the cytoplasmic membrane localization of RtaA, but also indicate a partial accumulation around structures that represent lipid droplets (Fig. 5C).

Inducible overexpression of rtaA results in AmB resistance

To investigate the effect of rtaA overexpression on the susceptibility of respective strains on antifungal drugs, we generated an rtaA-inducible strain of A. fumigatus using the xylose-inducible promoter PXylP. RtaA expression levels were verified by RT-qPCR. The expression of rtaA was significantly higher in the RtaA^PXylP strain after the addition of xylose to the medium in comparison to the AmB-induced expression of rtaA in the WT (Fig. S2C). Additional supplementation with AmB led to expression levels comparable to the rtaA expression level in the WT exposed to AmB (Fig. S2C). After 48 h of incubation at 37°C in RPMI–MOPS medium with or without xylose, we determined the MIC values (µg/ml) of AmB, AmBisome, VRZ, and Mnd for the RtaA^PXylP strain, the ΔrtaA strain, and the WT strain. The results are summarized in Fig. 6(C). Overexpression of rtaA in the presence of 0.5% (w/v) xylose resulted in a 2-fold increase in MIC to AmB and AmBisome (MIC values 1 µg/ml and 0.25 µg/ml, respectively) compared to the parental strain (0.5 µg/ml and 0.125 µg/ml, respectively), with no alteration in sensitivity to VRZ and Mnd. Consistent with our expectations, RtaA^PXylP behaved similarly to ΔrtaA in the absence of xylose.

To further assess the susceptibility of the strains to AmB, point inoculation assays were performed as described before, and the results are presented in Fig. 6. After 60 h of incubation at 37°C in RPMI–MOPS medium with/without xylose, the susceptibility of RtaA^PXylP was compared to that of ΔrtaA and the WT strain in both solid (Fig. 6A) and liquid (Fig. 6B) media. Notably, the overexpression of rtaA in the presence of 0.5% (w/v) xylose resulted in an increased resistance to AmB in the RtaA^PXylP strain, as compared to the parental strain. As expected, in the absence of xylose, the AmB susceptibility of strain RtaA^PXylP was similar to that of ΔrtaA. These findings provide further evidence that the overexpression of rtaA confers modest AmB resistance in A. fumigatus.

The expression of rtaA orthologues in the Aspergillus species A. terreus and A. lentulus is also activated by AmB

To determine if other Aspergillus species with more frequently observed AmB resistance also induce the expression of rtaA orthologues upon AmB treatment, we quantified the expression levels of the rtaA orthologues ATEG_04 110 (A. terreus) and IFM58399_04 074 (A. lentulus) via RT-qPCR in patient isolates of A. terreus and A. lentulus (Fig. 7). We used both susceptible strains with low AmB MIC and strains with elevated AmB MICs. In all strains, independent of the AmB susceptibility exhibited a significantly higher levels of the respective rtaA orthologue when treated with 0.125 µg/ml AmB as compared to untreated controls of the different strains. The highest change in abundance was observed in the A. lentulus strains (Fig. 7B). The fold change was comparable to the fold change seen in A. fumigatus (Fig. 7A). This indicates that closely related Aspergillus species like A. lentulus utilize similar mechanisms in the response to AmB, and even the more distantly related Aspergillus species A. terreus showed a similar regulation pattern. No significant difference in the expression of rtaA was determined between susceptible and resistant strains.

rtaA expression is specifically induced by exposure to AmB and nystatin

As described above, the spotting assays conducted with the ΔrtaA mutant demonstrated that RtaA confers resistance to AmB while exhibiting no resistance to other clinically significant antifungal agents. To corroborate these observations, we investigated the induction of rtaA gene expression by sublethal concentrations of AmB, ITZ, and caspofungin using northern blot analysis. (Fig. 8). Under standard growth conditions in AMM (control, Ctrl), the expression of rtaA transcript was below the detection limit. However, exposure to sublethal concentrations of AmB (0.125 µg/ml) for 4 h resulted in a significant increase in the expression of rtaA, while no changes in the expression levels of rtaA were observed when exposed to sublethal concentrations of ITZ (0.25 µg/ml) or caspofungin (0.1 µg/ml) (Fig. 8A). We also tested two other antifungal polyene macrolides, namely natamycin (2 µg/ml) and nystatin (0.25 µg/ml), of which only nystatin induced the expression of rtaA. As a control, we determined the expression of rtaA in A. fumigatus upon exposure to the antibacterial macrolide erythromycin (64 µg/ml), the polyene antibiotic linearmycin (8 µg/ml), and the ROS-inducing compound Mnd (2 µg/ml). None of these three compounds led to an increased expression of rtaA (Fig. 8B). This finding confirms that RtaA specifically mediates resistance to AmB and the structurally similar glycosylated polyene macrolide nystatin and that its expression is induced by sublethal concentrations of these two antifungal drugs, which are both produced by Streptomyces bacteria.

Deletion of rtaA alters modestly the lipid content of fungal hyphae

Lipid flippases are membrane proteins that play a crucial role in maintaining the lipid asymmetry of cellular membranes by selectively translocating specific lipids between the two membrane leaflets. This lipid asymmetry is critical for a wide range of cellular processes, including membrane trafficking, lipid metabolism, and signalling (Rizzo et al. 2019). There are various models that attempt to explain how fungal flippases may be involved in drug resistance. One proposed mechanism involves the S. cerevisiae protein Rta1p, which confers resistance to 7-aminocholesterol. According to this model, Rta1p induces membrane internalization via the endocytic pathway and further transports the drug to vacuoles (Manente and Ghislain 2009). Another possibility is that deletion of a flippase gene leads to changes in the membrane composition resulting in an incorrect localization of the drug target, which increases sensitivity to the corresponding antifungal compound or disrupts membrane repair. Additionally, lipid flippases are necessary for intracellular vesicle trafficking, and any alterations in their activity or expression may affect various cellular processes such as membrane trafficking, lipid metabolism, and signalling (Shor et al. 2016). However, clear experimental evidence for a flippase function of Rta1p is still lacking.

To investigate whether A. fumigatus RtaA confers drug resistance to AmB via a mechanism similar to that proposed for the S. cerevisiae protein Rta1p, we stained the mycelium of A. fumigatus WT and ΔrtaA mutant hyphae with filipin, a fluorescent sterol marker used to monitor ergosterol content, and Nile red (Fig. 9, Fig. S8). Confocal laser scanning microscopy of both AmB-treated (0.04 and 0.08 µg/ml AmB) and untreated mycelia of A. fumigatus revealed a patchy distribution of filipin-derived fluorescence with higher intensity at septa and some hyphal tips (Fig. S8). A drastic difference in fluorescence intensity between WT and ΔrtaA hyphae could not be visually recognized (Fig. 9A). However, intensity-based image analysis (Fig. 9B) revealed a trend towards reduced filipin staining of ΔrtaA hyphae compared to the WT, which was, however, only significant in the presence of 0.04 µg/ml AmB (P = .01, Hedges’ g = 0.42). We further performed LC–TQ–MS analysis to determine total cellular ergosterol levels. In accordance with the microscopic data, the ΔrtaA strain exhibited a decreased ergosterol content compared to the WT, and ergosterol was increased in the RtaA^PXylP strain upon xylose induction. Nevertheless, these differences were not statistically significant (Fig. 9C). In addition, we analysed the content of phytosphingosin and related sphingolipids, that have been discussed to contribute to AmB resistance (Madaan and Bari 2023). However, no significant differences were observed between the WT and the ΔrtaA mutant strain exposed to 0.125 mg/l AmB for 4 h or without treatment (Fig. S9).

As anticipated, Nile red staining highlighted intracellular lipid droplets, which were uniformly dispersed throughout the fungal hyphae. The treatment of hyphae with AmB led to a decrease in the median intensity of Nile red fluorescence at a higher drug concentration of 0.08 µg/ml AmB, and WT hyphae seemed to exhibit higher Nile red fluorescence signals in comparison to the ΔrtaA mutant, but all differences in Nile red staining were not significant in the intensity-based image analysis (Fig. 9A and B). A potential role of RtaA in lipid storage of mainly triacylglycerols and sterol esters and lipid trafficking is likely but needs further investigations. Alternatively, RtaA could be involved in a vesicle-dependent transport of AmB, the deletion of rtaA may alter the AmB drug response or RtaA simply stabilizes the membrane structure.

Deletion of rtaA alters the proteomic response of A. fumigatus to AmB

To investigate the impact of RtaA on the AmB stress response and to shed light on its possible protective functions, we compared the proteome of strain ΔrtaA exposed to 0.125 µg/ml AmB with the proteome of the WT strain. 399 different proteins decreased, and 264 proteins increased in abundance in comparison to the WT. PSEA revealed no GO term for proteins with higher abundance, but the terms fumagillin biosynthetic process, prenyltransferase activity, signal transduction, and hydrogen peroxide catabolic process for lower abundant proteins (Fig. 10, Table 3, and Table S8). These data suggest that the lack of RtaA leads to an increased disorganization of the cellular membrane resulting in impaired signalling processes and altered secondary metabolite production. Proteins with higher abundance included those which perform their function in the nucleus. In particular, proteins involved in DNA repair should be mentioned here (e.g. DNA repair proteins Swi10/Rad10 and Dds20/Mei5), which suggests that AmB may also induce directly or indirectly DNA damage.

Deletion of rtaA does not affect virulence in infected mice

To evaluate the impact of rtaA deletion on virulence, we employed two distinct pulmonary models of infection in immunocompromised mice. Three groups, each consisting of 10 mice, were infected with either WT, ΔrtaA, or rtaA^c conidia. In the first model, mice were immunocompromised with CA (CA model), while in the second model, mice were rendered neutropenic via administration of CP (CP model). Subsequently, mice were infected intranasally with A. fumigatus conidia. We monitored mouse survival over the course of 28 days. The results indicate that there was no significant difference in virulence between the three strains (P > .10), as shown by the survival curve (Fig. 11). Therefore, it can be concluded that RtaA does not play a role for the virulence of A. fumigatus.

Discussion

Aspergillus fumigatus poses a significant threat to individuals with a weakened immune system and underlying pulmonary conditions such as cystic fibrosis or chronic obstructive pulmonary disease (COPD). While triazoles are commonly used for initial treatment of IPA, an increasing rate of azole resistance has led to a shift towards fallback therapy with AmB, despite its associated side effects. Recent reports have shown a rise in clinical isolates of A. fumigatus exhibiting resistance to AmB (Ashu et al. 2018, Reichert-Lima et al. 2018, Fakhim et al. 2022), but research into the resistance mechanisms of A. fumigatus has been limited (Manavathu et al. 1998, Gautam et al. 2008, Furukawa et al. 2020). To address this gap, we hypothesized that conducting a detailed proteomic analysis following exposure to sublethal concentrations of AmB and AmBisome would reveal proteins involved in mediating resistance against these drugs. We discovered that exposure to sublethal concentrations of AmB or AmBisome is associated with changes in the abundances of proteins involved in oxidation–reduction processes, secondary metabolite biosynthesis, aromatic acid metabolism, and transmembrane transport. Similar effects have previously been documented for other fungi. The activation of transmembrane transport processes upon exposure to antifungals appears to be a common adaptive stress response in many fungi (Ribeiro et al. 2022). Similarly, the modification of amino acid metabolism during AmB adaptation has been reported in yeast (Ribeiro et al. 2022), while the induction of oxidation–reduction processes has been observed in fungi belonging to the Fusarium species complex (Castillo-Castaneda et al. 2020). The only published analysis of the A. fumigatus proteomic response to AmB to date was based on 2D-gel electrophoresis (2D-GE) and only shows a small overlap with our data. This may be explained by the different methods used (2D-GE versus LC–MS/MS) and the much longer incubation time of 24 h instead of 4 h (Gautam et al. 2008). Noteworthy is that the RtaA protein was not detected in this 2D-GE-based study, most probably due to the strong bias of 2D-GE against membrane proteins (Santoni et al. 2000). Nevertheless, transport proteins were also identified as an enriched category in this study. Furthermore, the level of l-ornithine aminotransferase Afu4g09140, which is involved in arginine degradation, increased in both studies. In contrast, Gautam et al. (2008) reported the upregulation of genes and proteins involved in the ergosterol biosynthesis pathway, which is considered as a resistance mechanism. However, in our study, we only observed a 5.7-fold increase in Erg3 expression specifically for AmBisome (Cavassin et al. 2021). Surprisingly, we did not find clear evidence of increased intracellular ROS production following AmB exposure, in contrast to the findings of Gautam et al. (2008) and other studies (Shekhova et al. 2017). The only increased ROS-detoxifying enzyme was a predicted glutathione S-transferase Afu2g17300. It is possible that oxidative stress may become more prominent after a longer exposure time or at higher concentrations of the drug. In agreement with this assumption, by using a fluorogenic ROS probe, significant levels of intracellular ROS were detected after a 45-min exposure to AmB concentrations above the MIC (Shekhova et al. 2017). Collectively, this suggests that ROS-induction of AmB is in particular concentration-dependent.

Interestingly, the proteomic response of A. fumigatus to AmB and AmBisome did not completely overlap. This is demonstrated by the specific increase in the levels of proteins involved in iron homeostasis observed only upon exposure to AmBisome. This difference in proteomic response is likely due to the distinct delivery mechanisms of AmB and AmBisome. AmBisome penetrates the fungal cell wall, enabling direct delivery of AmB to the cell membrane (Walker et al. 2018). By contrast, AmB diffuses through the cell wall matrix, which influences the dynamics of membrane destabilization.

It is noteworthy that in our study, AmB and AmBisome specifically activated enzymes involved in the biosynthesis of secondary metabolites, such as the prenylated polyphenol fumicycline (also described as neosartoricins), as well as xanthocillin and hexadehydro-astechrome (HAS). The latter two secondary metabolites are known to form complexes with transition metals, and in case of HAS, contribute to the virulence of the fungus (Yin et al. 2013). A similar induction of secondary metabolite production by a natural product-based antifungal has also been reported in response to the lipopetide caspofungin, which triggers the production of fumagillin in A. fumigatus (Conrad et al. 2018). Therefore, it can be hypothesized that bacterial natural products with antifungal activity, such as AmB, have the potential to serve as specific inducers of secondary metabolite production in filamentous fungi. It provides another example that microbial natural products serve as communication signal during cross-kingdom interactions (Netzker et al. 2018, Scherlach and Hertweck 2020). AmB might even trigger the activation of an antimicrobial response program in A. fumigatus. For instance, the copper chelator xanthocillin has antibacterial activity (Raffa et al. 2021), suggesting that it may exert similar effects in response to AmB. Similarly, the iron chelator HAS (Yin et al. 2013) could potentially elicit comparable effects as well. Additionally, the putative antimicrobial peptide Afu8g00710 increased by 20–30-fold. More speculative is the assumption that the activation of the tyrosine and phenylalanine degradation pathway leads to the formation of pyomelanin that confers resistance to ROS and could enhance competitive fitness (Schmaler-Ripcke et al. 2009).

To identify genes involved in mediating resistance to AmB and AmBisome, we prioritized nine proteins with the highest upregulation in response to these drugs. These proteins were chosen as they likely play an active role in resistance against the stress caused by drug exposure, and their deletion may result in increased susceptibility. Of the nine genes studied, only the deletion of rtaA, encoding a putative fungal lipid-translocating exporter protein, resulted in increased susceptibility to AmB and AmBisome.

Our study thus highlights the potential role of RtaA in mediating modest resistance to AmB and AmBisome. The increased susceptibility to these drugs following rtaA deletion, as well as the increased resistance observed with rtaA overexpression, strongly suggest that RtaA is actively involved in conferring modest resistance to these drugs. Notably, this effect appears to be highly selective, as we observed no significant changes in sensitivity to other drugs or stressors. Additionally, our findings regarding the involvement of RTA-family member proteins in polyene resistance appear to be unique in A. fumigatus and related species when compared to other, phylogenetically distinct fungal species. For instance, rta1 deletion in S. cerevisiae and Cryptococcus neoformans increased sensitivity to 7-amino cholesterol, a putative ergosterol biosynthesis inhibitor, without affecting sensitivity to other antifungals targeting sterol biosynthesis, including azoles and morpholine derivatives (Soustre et al. 1996, Smith-Peavler et al. 2022), while in Candida albicans RTA2 and RTA3 deletion led to azole susceptibility (Jia et al. 2008, Whaley et al. 2016).

Studies on rta-like family gene expression have been conducted in various fungal species including S. cerevisiae, C. albicans, and C. glabrata. In S. cerevisiae, rta1 expression is triggered by hypoxia and cytotoxic stress (Kolaczkowska et al. 2012). In C. albicans, RTA2 is induced by a wide range of cell membrane stressors including azoles, polyenes, and allylamines (Thomas et al. 2015), while in C. glabrata, RTA1 induction was shown to be caused by hypoxia and azoles (Kolaczkowska et al. 2013). By contrast, we observed that rtaA expression was specifically induced by AmB and the similar polyene nystatin and not by other antifungal agents, such as ITZ or caspofungin. The lack of induction with the polyene natamycin may be explained with the possible different mode of action by binding specifically to ergosterol without permeabilizing the membrane (te Welscher et al. 2008).

To date, the mechanistic understanding of the RTA-like protein family function remains unclear, although, based on their similarity to the Rsb1 protein in yeast, it is assumed that they function as lipid translocases (Kihara and Igarashi 2004). Our work suggests that RtaA-dependent modulation of AmB sensitivity and resistance may result from changes in the membrane composition mediated via vesicular transport between lipid droplets and cell membrane. Lipid droplets may potentially act as a sink to remove AmB from the cytoplasmic membrane. Alternatively, RtaA may mediate the transport of the AmB target ergosterol or other membrane components between lipid droplets and the cellular membrane. The observation that RtaA localized to the cytoplasmic membrane was also shown for S. cerevisiae (Kolaczkowska et al. 2012), and partly one could observe an association of the RtaA-GFP signal with lipid droplets. No correlation was observed between the GFP signal and endosomal vacuoles, which contradicts the assumption that RtaA is involved in endocytic membrane transport processes. It cannot be ruled out that RtaA has also a protective function for other intracellular membranes (Higuchi 2021) or the overexpression of the GFP-fusion protein leads to mislocalization and degradation in vacuoles. In this context, it is noticeable that the treatment of A. fumigatus with the antifungal compound olorofim caused the formation of vesicle-like structures, which were not stainable with membrane- or vacuole-selective dyes (du Pre et al. 2020). It cannot be ruled out that RtaA accumulates in such structures besides its association with the cell membrane.

Taken together, in this study we conducted a detailed proteomic analysis to investigate the response of A. fumigatus to two commonly used antifungal drugs, AmB and AmBisome. Our results demonstrate that RtaA is the most upregulated protein in response to drug exposure and plays a role in resistance to polyenes, which is supported by the observation of a specific induction of rtaA expression by AmB and nystatin. Specifically, overexpression of RtaA confers resistance to AmB, AmBisome, and nystatin, while deletion of its encoding gene leads to increased sensitivity to these drugs. Our findings also suggest that RtaA is involved in lipid trafficking and storage to maintain lipid homeostasis. Furthermore, our study highlights the general importance of fungal RTA-like proteins in mediating resistance to compounds disturbing the integrity of the cell membrane.

The effect of RtaA deletion or overexpression on AmB drug susceptibility may be considered as modest (4-fold increase in susceptibility), but even small changes in MIC values can have a high impact on the success of an antifungal therapy as shown by a murine model of systemic aspergillosis with clinical isolates of A. flavus exhibiting different susceptibilities to AmB. Here, even small, increased AmB MIC levels of 2–4-fold resulted in failure of AmB and L-AmB treatment (Barchiesi et al. 2013).

An open question remains about how rtaA is regulated. Furukawa et al. (2020) showed that the negative cofactor 2 complex proteins NctA and NctB are key regulators of drug resistance and their loss leads to resistance against antifungal azoles, terbinafine, and AmB. However, it is unlikely that RtaA contributes to the observed AmB resistance, since both knock strains, ΔnctA and ΔnctB, showed reduced transcript levels of rtaA upon exposure to a sublethal concentration of ITZ.

It will be important to assess the clinical relevance of our findings by analysing the rtaA sequence and its expression in emerging AmB-resistant strains of A. fumigatus (Ashu et al. 2018, Reichert-Lima et al. 2018, Fakhim et al. 2022). In a recent study of A. fumigatus Single Nucleotide Polymorphisms in introns and exons of genes and their up- and downstream regions associated with AmB resistance, no correlation was found for the rtaA gene described here (Chen et al. 2023). In addition, we did not find difference in the expression levels of rtaA orthologues between strains of A. lentulus and A. terreus with different AmB-MIC values.

In summary, our data illustrates that RTA-like proteins fulfil an important function in specifically providing fungi with protection against membrane-interfering compounds and there is certainly more to discover. For example, it was recently shown that Botrytis cinerea employs an RTA-like protein to combat membrane-induced stress by plant-derived saponins (You et al. 2024).

Acknowledgements

We thank S. Steinbach for excellent technical assistance. We thank Mike Bromley (Manchester Fungal Infection Group) for providing two A. fumigatus knock-out strains of the genome-wide knockout program (COFUN) library.

Conflict of interest

None declared.

Funding

This work was supported by Deutsche Forschungsgemeinschaft (DFG) within the Collaborative Research Centers (CRC)/Transregio 124 ‘FungiNet’ to A.A.B., M.T.F., and O.K. (A1, B4, and Z2; project ID 210879364), CRC 1127 ChemBioSys (Project-ID 239748522), and CRC 1278 PolyTarget to A.A.B. and M.T.F. (B02 and Z01; project ID 316213987). Further support was received from the Israel-China Science Foundation (IFS) grant 2444/18 to N.O. . The work was further funded by the Federal Ministry of Education and Research (BMBF) in the funding program ‘Zwanzig20–Partnerschaft für Innovation’ (FINAR; FKZ 03ZZ0809A) and in the context of InfectoGnostics2 Research Campus Jena (ADA number 13GW0456B). Furthermore, M.T.F. and A.E.B. were funded by the Cluster of Excellence ‘Balance of the Microverse—Project ID 390713860)’. The work of the German NRZMyk is supported by the Robert Koch Institute from funds provided by the German Ministry of Health (grant number 1369-240).

Data availability

The proteomic data that support the findings of this study are available from the ProteomeXchange Consortium via the PRIDE partner repository (Perez-Riverol et al. 2022) with dataset identifiers PXD042190 and PXD056876.

References

Author notes