PREreview of Butyrolactol A is a phospholipid flippase inhibitor that potentiates the bioactivity of caspofungin against resistant fungi

This study from Chen et al. addresses the rise of antifungal resistance, which is of increasing biomedical concern due to our limited number of antifungal drugs. The authors leveraged a natural product screen on microbial extracts of resistant fungal pathogens to identify butyrolactol A as a potentiator for echinocandins, a cell wall biosynthesis disruptor. Using a combination of single-particle cryo-EM and functional assays, the authors demonstrate that the polyketide butyrolactol A blocks phospholipid flippase Apt1-Cdc50, leading to perturbed membrane asymmetry and stalling of endocytic pathways. This work elegantly demonstrates how natural product screens combined with structural analysis and functional assays can reveal new avenues to pursue in developing antifungals, and presents lipid flippases as a novel drug target.

The study is well-written and of high methodological quality, with the data rigorously supporting most of their interpretations. Given this study’s valuable addition to the field of antifungal drug development and broad pharmaceutical relevance, we find this work appropriate for publication, provided the following points have been addressed:

Major issues

• The authors should describe their reasoning for choosing a methanolic extract protocol and address the aspect of excluding compounds by their insolubility in the chosen solvent. Additionally, they should mention the limitations of their single-concentration approach in their potentiator assay.

• While of interest from a metabolic engineering perspective, the section regarding BLB and successful identification of a suitable methyltransferase is somewhat of a distraction. Moving this section to a supplemental note or removing it altogether would clarify the focus of the study for readers. Alternatively, the authors should more clearly justify why this excursion benefited this study by showing a difference in caspofungin potentiation between BLB and BLA.

• While investigating the effect of deleting ergosterol biosynthesis components, the authors should emphasize that this was only shown for C. albicans. Having shown a difference between the response to BLA treatment on reducing ergosterol (Fig. 3F) between the investigated cells, this experimental limitation should be added by the authors or proven with a similar set of deletions for C. neoformans.

• The linear – sequence-based – description of mutations (Fig. 4 / p.7) doesn’t do much to demonstrate the importance & localization of these mutations in APT1.  The authors should map these mutations onto an AlphaFold prediction model of Apt1 to provide a structural context.

•  Conventional fluorescence microscopy does not provide sufficient resolution to unambiguously assign DsRed-labeled Apt1 to the plasma membrane (Fig. 4G). TIRF microscopy would need to be performed to confirm this initial observation, so the conclusions from these studies should be softened.

• Identifying and building ligands into high-resolution structures (even at better than 3 Å) is not always conclusive. To demonstrate this issue, we include an image showing that CHS could accommodate the majority of the assigned ligand density. The authors should explore the possibility that alternative ligands (lipids) may have co-purified with the complex by comparing the BLA-assigned reconstruction with the other reconstructions resulting from their 3D classification – are any of these reconstructions ligand-free, and how much variability is present? Additional experiments like MS analyses of BLA-excluding mutants compared to wt would confirm the absence of BLA.

• Based on the processing scheme Fig. S4, only about one quarter of the particles show clear ligand density after providing a very tight mask during 3D classification. Concerns have been raised about the application of tight masks in cryoSPARC’s 3D classification, invoking the “Einstein-from-Noise”-problem. To validate the presence of the ligand, the particles from the final selected class should be reprocessed without providing 3D alignments (starting with an ab initio reconstruction) followed by the author’s additional processing steps.

• A panel including the entire cryo-EM map used for modeling should be added to Fig. 5, highlighting the quality of the map being used for modeling and interpretation. Along these lines, the ligand-binding site in Fig. 5A is hard to see/interpret. Please add a 2D ligand plot to visualize ligand-residue interactions with software like LigPlot+. Further, the authors have "zoned" the cryoEM map to only display voxels within a certain radius of the ligand to show that EM density is present for the model. The density for surrounding residues should be included in this figure to demonstrate the quality of the map in this region.

• Superimposing the Apt1 E2P and E1 states would be a more informative structural comparison than the side-by-side surface representation alone (Fig. 5). 

• In-vivo point mutations of Apt1 in C. neoformans are crucial, but should be accompanied by analytical SEC of recombinant protein, which would show that the impact of mutations does not destabilize the complex. The changes in relative growth (Fig. 5F) without BLA indicate such an effect.

• The authors reference a designed peptide to block Cdc50 (p.16). Since the authors reference AlphaFold in the manuscript, a predicted structure of the peptide binding to Cdc50 and a comparison with the binding mode of their resolved complex would add to the discussion.

• The authors uploaded DeepEMhancer-manipulated maps to the EMDB as the primary map for the validation reports. Only B-factor sharpened maps should be used for these reports, since the way that DeepEMhancer modifies the input densities in unpredictable and these types of maps have not been thoroughly characterized by the community. Further, this program is known for improperly handling non-protein densities due to a lack of training data.

Minor issues

• The results of the NMR experiments to identify and characterize BLA alongside the MS data are not shown in Fig. S1F. Please add these results.

• A gene cluster annotation is missing in Fig S2A. Please add ORF numbers according to Suppl. Table 1.

• Please highlight the yeast Dnf1/2 in the phylogenetic tree (Fig. S3B).

• The authors used LMNG/CHS as a membrane mimic for their cryo-EM experiments. Please rationalize this choice, reflecting the importance of sterols for stabilizing many complexes extracted from the plasma membrane.

• The authors mention specific chemical groups and interactions in BLA. Please refer to Fig. 3D.

• Reference and PDB identifier are needed for the comparison to PC-bound Dnf1 (E2P & E1, Fig. 5).

• The reference about TOP1 inhibition by camptothecin is missing (p.12).

• The authors have modeled a magnesium that doesn’t appear to have sufficient density to confirm its position, or it is improperly modeled (see attached image). This should be addressed.

• In the cryo-EM image processing methods section:

  a.     There appear to be two frame-alignment / dose-weighting steps – one prior to CTF estimation, and then another after manual inspection. Please explain how and why this was performed. Please also add information regarding any settings used for curating movies.

  b.     The processing section should be written so that someone else could repeate the steps. Include all non-default parameters used for classification and refinement (mask diameters, # of classes, initial low pass filter, # of iterations, etc.)

  c.     The atomic modeling methodology also should be expanded to be more descriptive, including any relevant Phenix refinement parameters and any constraints (secondary structure, Ramachandran, etc.)

  d.     The supplementary materials should include local resolution plots and model-to-map FSC with the resolution at a 0.5 FSC cutoff denoted for each structure.

Competing interests

The authors declare that they have no competing interests.