High-throughput platform for yeast morphological profiling predicts the targets of bioactive compounds

Abstract

Morphological profiling is an omics-based approach for predicting intracellular targets of chemical compounds in which the dose-dependent morphological changes induced by the compound are systematically compared to the morphological changes in gene-deleted cells. In this study, we developed a reliable high-throughput (HT) platform for yeast morphological profiling using drug-hypersensitive strains to minimize compound use, HT microscopy to speed up data generation and analysis, and a generalized linear model to predict targets with high reliability. We first conducted a proof-of-concept study using six compounds with known targets: bortezomib, hydroxyurea, methyl methanesulfonate, benomyl, tunicamycin, and echinocandin B. Then we applied our platform to predict the mechanism of action of a novel diferulate-derived compound, poacidiene. Morphological profiling of poacidiene implied that it affects the DNA damage response, which genetic analysis confirmed. Furthermore, we found that poacidiene inhibits the growth of phytopathogenic fungi, implying applications as an effective antifungal agent. Thus, our platform is a new whole-cell target prediction tool for drug discovery.

Fig. 1. Schematic illustration of the new platform for morphological profiling in yeast.

a Selection of genes deleted in a drug-hypersensitive strain. Haploid mutant strains in which each gene was individually deleted in a drug-hypersensitive strain with a triple-deletion genetic background were constructed for 1637 non-essential genes with GO annotation and used for the morphological profiling and functional annotation of chemical compounds (blue circle in top left panel). Of the 4718 non-essential genes in the yeast genome (orange circle in top left panel), 2378 genes (green circle in top left panel) were selected as morphologically important genes. Among 1982 quadruple mutants (cyan circle in top left panel) successfully constructed by gene deletion in a drug-hypersensitive strain (3∆; pdr1∆ pdr3∆ snq2∆, middle panel), 1637 genes had GO annotation. b High-throughput microscopy system coupled with the image-processing system CalMorph. Fluorescence microscope images of the yeast quadruple deletion collection and drug-treated yeast cells were acquired by HT microscopy and subjected to the image-processing program CalMorph (bottom left panel) to quantify morphological features of 501 traits. c Application of a generalized linear model to predict biological processes targeted by the bioactive compounds. Morphological data of the quadruple mutants and drug-treated cells were normalized and subjected to analysis with a GLM to predict biological processes targeted by the bioactive compounds (bottom right panel).

Fig. 2. Morphological phenotype of pdr1 pdr3 snq2.

a Cells of a drug-hypersensitive 3∆ strain and its parental strain. The cell wall (green), nuclear DNA (blue), and actin cytoskeleton (red) were stained with FITC-ConA, DAPI, and rhodamine-phalloidin, respectively. Scale bars indicate 5 μm. b Illustration of the morphological phenotype of 3∆. Color legend is the same as (a). c PCA plot showing the mutual relationship of 3∆ (red), 2∆ (green), 1∆ (blue), and their parental strain (black). Open and closed circles indicate biological replicates (n = 5) and mean of each strain.

Fig. 3. Effectiveness of bioactive compounds in the drug-hypersensitive strain.

a Number of morphological traits changed by drug treatment in the drug-hypersensitive strain (solid bar) and its wild-type strain (his3Δ, shaded bar). HU, ECB, and TMN indicate 8 and 10 mM of HU for the drug-hypersensitive strain and his3Δ, respectively, 2.0 μg/mL ECB for both strains, and 100 ng/mL TMN for both strains. b Drug-treatment-induced changes in holistic morphological abnormality in the drug-hypersensitive strain (solid bar) and his3Δ (shaded bar). Error bars indicate standard deviations. Asterisks indicate a significant difference at p < 0.01 after Bonferroni correction by the likelihood-ratio test. c PCA plot showing morphological profiles of the drug-hypersensitive strain (circle) and his3Δ (square) induced by HU, ECB, and TMN. The drug-hypersensitive strain (cross) and his3Δ (plus) without drug treatment are also shown. Directions of morphological changes induced by HU, ECB, and TMN are shown as arrows in the drug-hypersensitive cells (solid) and his3Δ (dashed). Percentages in parentheses indicate the proportion of variance explained by each PC. d Similarity of morphological profiles between the drug-hypersensitive strain and his3Δ after treatment with HU, ECB, and TMN. PC scores of 259 PCs were plotted between his3Δ (x-axis) and 3∆ (y-axis) after chemical treatment. R and p indicate the Pearson product-moment correlation coefficient and the p-value calculated by the t-distribution.

Fig. 4. Morphological profiling of bortezomib-treated cells.

a BTZ-treated cells exhibited the most similar morphology to rpn10Δ. The RPN10 gene belongs to the significantly enriched functional gene group represented as “proteasome regulatory particle (green text)”. The red lines in the top panel represent 1486 groups of the functionally related genes, which are ordered by the similarity of gene functions. Briefly, the distance between genes was calculated using a Boolean matrix with genes as rows and GO as columns, hierarchical cluster analysis was performed based on the obtained distance matrix, and the gene-deleted strains were arranged in the order of the tree diagram of this cluster analysis (see “Materials and Methods”). The y-axis indicates average similarities of functionally related genes to BTZ-treated cells as expressed by −log10p (likelihood ratio test). Dark green texts indicate the GO terms enriched in the significant top three groups of the lowest p value by the likelihood ratio test among the detected gene groups. The horizontal dashed line indicates a threshold p-value of 0.05 after Bonferroni correction. Enriched GOs in the detected gene groups (p < 0.05 after the Bonferroni correction) are listed in Supplementary Data 2. The lower panel shows the most similar mutant and the mutant belonging to the functionally related gene shown in the top panel (black), 1635 quadruple mutants (grey), and 749 replicates of 3Δ (orange, ordered by the date of data acquisition). The morphological similarity (y-axis) is represented by −log10p, which was calculated by Gaussian distribution fitted to the t-distribution of correlation coefficients from the wild type (n = 749). The horizontal solid line indicates threshold p-values at FDR = 0.05. The morphological similarity was calculated using five independent biological experiments with bortezomib-treated cells. Significantly similar mutants (FDR = 0.05) are listed in Supplementary Data 1. b Triple-staining images of cells of 3Δ treated with BTZ and rpn10Δ in the 3Δ background (rpn10). The cell wall (green), nuclear DNA (blue), and actin cytoskeleton (red) are shown. Scale bars indicate 5 μm. c Correlation coefficient of morphological profiles between BTZ-treated cells and rpn10Δ. PC scores of 259 PCs were plotted between BTZ-treated cells (x-axis) and rpn10Δ (y-axis) after chemical treatment. R, p; the redacid dilactone was made from ferulic acid by line indicates the Pearson product-moment correlation coefficient, p-value calculated by the t-distribution of 3Δ replication (n = 749), and the linear regression line, respectively.

Fig. 5. Reproducibility of morphological phenotypes.

a Distribution of correlation coefficients between quadruple mutants and single mutants. Correlation coefficients were calculated between each pair of the 1982 deletion mutants in quadruple mutants and single mutants (red) and between arbitrary pairs of 3∆ (n = 749) and his3∆ (n = 109) (grey), and are expressed using density plots. Vertical red lines indicate a threshold at FDR = 0.05 by the one-tailed t-distribution test. b Percentages of gene-deletion mutants showing reproducibility. Among 1982 common gene-deleted strains (left), 27.0% were detected as having significantly similar morphological phenotypes (FDR = 0.05 by the one-tailed t-distribution test) between single and quadruple mutants. Among 321 quadruple mutants with significantly high HMA, 69.2% were detected as having significant phenotypic similarity (FDR = 0.05 by the one-tailed t-distribution test). c Comparison of average HMA in the genes belonging to the same gene function groups. Orange dashed lines indicate averages of the parental strains. Black line indicates the same HMA between quadruple and single mutants. Green and red circles indicate the gene groups with significantly high average HMA in the single- and quadruple-deletion mutants (Wald test, FDR = 0.05), respectively. Yellow circles indicate the gene groups with similar HMA. d Venn diagram of the gene groups; the color legend is the same as in (c).

Fig. 6. Prediction of the intracellular target of poacidiene.

a Growth inhibition of yeast strain (3∆) by poacidiene (PD) and poacic acid (PA). The solid curve with circles and the dashed curve with plus symbols indicate approximated logistic curves of PD and PA, respectively. Vertical lines indicate IC₅₀ values estimated by logistic regression (PD: 26.41 ± 1.28 μg/mL, PA: 255.2 ± 7.8 µg/mL, mean ± standard error). b Morphological similarity of PD-treated cells with 1637 quadruple mutants. Legends are the same as Fig. 4a. Enriched GOs in the detected gene groups (p < 0.05 after the Bonferroni correction by likelihood ratio test) are listed in Supplementary Data 15. Significantly similar mutants (FDR = 0.05 by one-sample test with Gaussian distribution fitted to the t-distribution of correlation coefficients from the wild type) are listed in Supplementary Data 14. c Similarity of morphological profiles between the PD-treated cells and rad54∆. Legends are the same as Fig. 4c. d Similarity of the morphological profiles among PD-treated cells (PD) and Cul8-RING gene-deletion mutants. Green lines (but not grey) indicate significantly high correlation coefficients at FDR = 0.05. R and p indicate the Pearson product-moment correlation coefficient and p-value calculated by one-tailed t-distribution, respectively.

Fig. 7. Sensitivity of poacidiene in several gene-deletion mutants defective in the recombinational repair pathway.

Inhibitory effects of poacidiene (PD) for cell growth in mutants of a Cul8-RING and b the recombinational repair pathway. Vertical lines indicate estimated IC₅₀ values. Colors correspond to strains indicated in (c). c Distributions of morphological similarity and IC₅₀ values. The dashed lines represent significantly different lines compared to 3∆ (p < 0.05, likelihood ratio test after Bonferroni correction).

Fig. 8. Sensitivity of phytopathogenic filamentous fungi to poacidiene.

a The effects of poacidiene (PD) on the growth of the filamentous fungi R. solani at 300 µg/mL. The radial growth of mycelia was tested (n = 3) after 24 h, and PDA plates containing DMSO were used as control plates. The percentage and ± standard error of the growth inhibition by PD (n = 3) were estimated by maximum likelihood estimation of one-way ANOVA assuming gamma distribution to colony diameter (mm) compared with the growth on the control plates (n = 3). The images with and without a dotted circumference are shown after enhancing the contrast. The raw images are shown in Supplementary Fig. 5. The dotted circle indicates the edge of the colony. The dose-dependent inhibition of b R. solani after 24 h, c A. alternata after 120 h, and d A. solani after 120 h by PD. The radial growth of mycelia was tested (n = 3). Asterisks indicate a significant difference at p < 0.01 by Dunnett’s test. Error bars in the bar plots indicate standard deviations.

Fig. 9. Sensitivity of oomycetes and phytopathogenic fungi to poacidiene.

a The effects of poacidiene (PD) on the growth of the oomycete P. aphanidermatum at 300 µg/mL. The radial growth of mycelia was tested (n = 3) after 15 h, and PDA plates containing DMSO were used as control plates. The percentage and ± standard error of the growth inhibition by PD (n = 3) was estimated by maximum likelihood estimation of one-way ANOVA assuming gamma distribution to colony diameter compared with the growth on the control plates (n = 3). The images with and without a dotted circumference are shown after enhancing the contrast. The raw images are shown in Supplementary Fig. 5. The dose-dependent inhibition of b P. aphanidermatum after 15 h and c P. sojae by PD. Asterisks in (b) P. aphanidermatum indicate a significant difference at p < 0.01 by Dunnett’s test. Error bars indicate standard deviations. The regression curve in (c) P. sojae was estimated by maximum likelihood estimation of multiple linear regression analysis assuming negative binomial distribution to colony diameter (mm, n = 72) with the cultivation time (day) and the concentration (µg/mL) of the PD treatment as explanatory variable (p = 4.39e⁻²³ by likelihood ratio test). Black, brown, red, and orange lines and dots indicate 0, 500, 1000, and 1500 µg/mL of PD.

Scheme 1

Synthetic scheme for producing poacidiene. The bolded bond is that formed during radical coupling (and indicate why this is termed an 8–8 diferulate-derived compound). The ring designations (A and B) and numbering are for NMR assignment purposes (Supplementary Figs. 2–4).