Using the hormone-responsive MCF-7/vBOS (abbreviated as MCF-7) breast cancer cell line, we further developed the original Cell Painting (CP) method4,5 into the Cell Painting PLUS (CPP) assay. When conducting CP in this comparative study, we followed a typical imaging set-up using four laser lines and captured dyes with largely overlapping spectral ranges, i.e., RNA/ER and Actin/Golgi (AGP), in the same imaging channel (Fig. 1A). In contrast, in the CPP assay, all dyes were captured in separate imaging channels, providing more specific information for these organelles (Fig. 1B). Notably, CPP also included the additional staining of lysosomes.
With regard to the selection and evaluation of suitable dyes for CPP, we systematically investigated spectral crosstalk (emission bleed-through and cross-excitation) and the signal stability of dyes over time. Emission bleed-through was observed for the RNA dye and, to a weaker extent, for the DNA dye when excited with the 488 nm or 405 nm imaging lasers, respectively (Supplementary Fig. 1A). Therefore, each dye was sequentially imaged in the CPP assay to avoid effects of emission-bleed-through on staining specificity. The RNA dye also showed some cross-excitation when excited with the Mito channel laser at 561 nm (Supplementary Fig. 1A). This property of the RNA dye leading to weak emission in the Mito channel, could not be fully mitigated by adapting imaging routines but was considered to be minor due to the high signal-to-noise ratio of the specific Mito signal. However, as a precautionary measure, the Mito signal was only analyzed in the staining second cycle as summarized below.
All CPP dyes were well detectable over the course of 4 weeks after staining, but the staining intensities of all dyes remained sufficiently stable only until day 1 (deviation of less than ± 10% compared to day 0) (Supplementary Fig. 1A). Prominent differences in the relative signal intensities over time were observed for the Lyso dye (decreasing) and the ER dye (increasing) already from day 2 after staining. This may be related to pH-dependent changes in the fluorescence properties of the dye itself, or due to changes in the binding of the organelle-targeting moiety. The LysoTracker™ dye accumulates in lysosomes due to their acidic pH (~4.5–5.0) and is therefore being applied to live cells. In turn, concanavalin A is used in fixed cells and binds to specific carbohydrate structures, such as mannose and glucose residues, that are enriched at the ER. After cell fixation by cross-linking proteins and other cellular components with paraformaldehyde (PFA), the cellular morphology is preserved, and the ability of fluorescent dyes to diffuse within the cell should be significantly limited, but is not fully inhibited. Under the staining conditions used in this study, the Lyso and ER dyes may take longer (up to day 2) to fully reach equilibrium. Whereas the signal intensity of the ER dye is maintained at this plateau, the intensity of Lyso may have quickly dropped again to lower levels after day 2. This emphasizes that thoroughly characterizing the fluorescence properties of the dyes used in the specific experimental setting is crucial for interpreting the results, especially when using dyes in different cellular compartments with distinct pH levels, such as in the highly acidic environment of lysosomes. For that reason, imaging was conducted in the CPP assay within 24 h after staining to ensure robustness of phenotypic profiling data. The concentrations of the CPP dyes and corresponding exposure times were chosen to achieve a balanced compromise between dye cost and total imaging time, while maintaining an optimal signal intensity range (Supplementary Fig. 1B, Supplementary Data 1). The dye concentrations used in CPP and in the original4 or recently updated5 CP method were similar indicating comparable screening costs in CPP and CP per single dye used (Supplementary Data 1). Thus, the additional reagent costs of the CPP assay are mainly due to the inclusion of the Lyso dye (Supplementary Data 1), but may decrease if alternative Lyso dyes compatible with fixed cell staining become available in the future.
To allow iterative staining of cells, we developed a dye elution buffer that efficiently removed the staining signals but preserved subcellular compartment and organelle morphologies (Fig. 1C, D). The development and optimization of suitable elution buffers for each dye involved extensive testing of various buffer components and parameters in combination including different pH, reducing agents, chaotropic agents (ionic strength), temperatures, and elution times. The optimal elution buffer compositions for each dye are summarized in Supplementary Data 1 to guide other laboratories in implementing and customizing their own CPP assay depending on their specific needs. Notably, the CPP elution buffer (0.5 M L-Glycine, 1% SDS, pH 2.5), which was used in this study for all phenotypic profiling screens, was designed to efficiently remove the signals of all dyes except for the Mito dye (Fig. 1C). This was intended to be able to use the Mito channel in the image analysis workflow as a reference channel for combination (registration) of individual image stacks from multiple staining cycles into a single multi-channel image stack (Fig. 2B). Notably, the final CPP elution buffer demonstrated superior efficacy for dye elution in comparison to other published buffers28,29,30 that were used for antibody elution (Supplementary Fig. 1C). Thus, the CPP elution buffer provided not only an efficient, but also a time- and cost-effective elution procedure in the CPP protocol (Supplementary Data 1).
To investigate whether the CPP elution step may influence subcellular compartment and organelle morphologies or interfere with dye binding, we re-stained the fixed cells of staining cycle 1 after elution with exactly the same dye in a staining cycle 2, except for the live-cell Lyso and Mito dyes. All dyes but not the Actin dye could be re-stained and the respective morphologies of subcellular compartments and organelles were preserved (Fig. 1D). The Mito dye was not eluted and showed very comparable signal intensities in staining cycles 1 and 2. Overall signal intensities of some re-stained dyes slightly differed between the two staining cycles. Regarding the absence of Actin dye signals after re-staining, testing of other Phalloidin dyes and an anti-Actin antibody revealed that the CPP elution step might generally inhibit the binding of Phalloidin-based dyes to actin fibers rather than disrupting the actin cytoskeleton (Supplementary Fig. 1D). Thus, in the final CPP staining-elution workflow used in this study (Fig. 1E), the two live-cell Lyso and Mito dyes were included in staining cycle 1 along with the Actin dye (to avoid re-staining issues after elution) and the RNA dye (slightly higher elution efficiency than the Golgi dye). Accordingly, the DNA, Golgi, and ER dyes were assigned to staining cycle 2 and analyzed together with the remaining Mito signal (to avoid cross-excitation of the RNA dye) from cycle 1.
To demonstrate transferability of this staining-elution approach and the broad applicability of the CPP assay, we tested three additional human cell lines that are widely used as surrogate models for relevant target organs including the U2OS (bone) and HepG2 (liver) cancer cell lines as well as post-mitotic, differentiated RPTEC-TERT1 (abbreviated as RPTEC) primary kidney cells (Fig. 1F). For each cell line, the seeding density was optimized to achieve ~80% confluency for HepG2 and U2OS cells or full confluency for MCF-7 and RPTEC at the time of imaging. For the epithelial cell lines MCF-7 and RPTEC, cells were imaged at full confluency to more closely reflect physiologically-relevant conditions resulting in a smaller size and more compact cytoplasm as compared to HepG2 and U2OS cells, which were less densely seeded and grew more flatly to maintain comparability to publicly available CP data. The adaptation of CPP to different cell lines did not need any further modifications to the staining protocol or the image/data analysis pipelines, indicating the versatile applicability of CPP to diverse cellular and biological contexts.
In order to compare CPP with the original CP method, we established a reference compound plate comprising 13 diverse compounds (drugs, biotoxins, plant alkaloids, and industrial chemicals) that target relevant organelles and cell functions17,31,32,33,34,35,36,37,38,39,40,41,42,43,44, as well as two compounds (saccharin and sorbitol) serving as negative controls along with the DMSO solvent control (Fig. 2A). Eight concentrations (at half-log dilution) of each reference compound were distributed across the reference compound plate in three blocks of technical replicates and alternating order to evaluate potential plate position effects (Supplementary Data 2).
Using this reference compound plate, we conducted small-scale CPP screens with 48 h exposure time in MCF-7, HepG2, U2OS, and RPTEC cells. For direct comparison of the CPP profiling data with CP, we also conducted the reference compound screen using the CP method in MCF-7 cells to explore its specific added value for phenotypic profiling. The data was analyzed using customized image and data analysis workflows as outlined for CPP in Fig. 2B. The CPP and CP image analysis procedures included the registration and segmentation of images (Supplementary Fig. 2A), followed by the extraction of cell features using the commercial Harmony software (Revvity Inc.) (CPP/Harmony: 894 features, CP/Harmony: 558 features), representing the primary image analysis software that was applied to all cell lines in this study. For comparison, we additionally analyzed CPP images in MCF-7 cells in an analogous way using the open-source Cell Profiler software45 (CPP/Cell Profiler: 3648 features) as an alternative image analysis method. The extracted cell features were subsequently fed into the data analysis procedure, which included the standardization and visualization of the cell feature data using the open-source KNIME process automation software46 to generate compound activity profiles (see Fig. 2) and conduct benchmark concentration (BMC) modeling (see Fig. 3). For CPP/Cell Profiler, standardization of cell features was followed by an additional step of feature selection/reduction to align with common practice47 (CPP/Cell Profiler: 345 selected features).
The compound activity profiles of the CPP reference compound screen revealed distinct concentration-dependent activity patterns for the four cell lines (Fig. 2C, Supplementary Fig. 2B, Supplementary Data 3, Supplementary Data 12–18). Very strong cellular responses spanning multiple channels often correlated with cytotoxic concentration ranges, i.e., compound concentrations that led to a reduction of the relative cell number below 53% compared to the DMSO solvent control (Fig. 2C, Supplementary Data 8–11). This 53% threshold for cytotoxicity was set to ensure that the maximum non-cytotoxic concentrations of the reference compounds were identical between CPP and CP in order to enable direct comparisons of the two methods in MCF-7 cells. The four cell lines showed different cytotoxicities for the reference compounds, with post-mitotic RPTEC cells being the least sensitive. As expected, the activity profiles of the negative control compounds saccharin and sorbitol did not indicate relevant activities across all cell lines in CPP and CP (Fig. 2C, D, Supplementary Fig. 2B). Notably, the activity profiles of fluphenazine, siramesine as well as tetrandrine displayed prominent effects in the Lyso channel at non-cytotoxic concentrations across all cell lines. Further, the weaker activity profile of tetrandrine at the highest tested concentration indicated potential compound solubility issues across all cell lines. The resulting lower exposure levels were also reflected in the corresponding relative cell number plots (Supplementary Data 3–6), consequently leading to the exclusion of this concentration from subsequent analysis. The profile of cytochalasin D was pronounced in the Actin channel and the profile of berberine chloride in the Mito channel, particularly in HepG2, U2OS, and RPTEC cells. Overall, CPP and CP showed largely consistent compound activity profiles for the DNA and Mito channels common to both methods (Supplementary Fig. 2C), with the Mito activity profiles of fluphenazine and tetrandrine showing higher robust z-score magnitudes in CP than in CPP (Fig. 2C, D, Supplementary Fig. 2C). In addition to the observed Lyso-related MoA of the two compounds (see Fig. 2A), those Mito activities were also consistent with published CP profiles of MCF-7 cells17. Notably, tetrandrine also induced prominent responses in the RNA/ER and AGP channels in CP, which could be more specifically allocated to the ER and Golgi channels, respectively, when using CPP (Fig. 2D). This observation is one example of the usefulness of separating the ER and RNA channels, which provides qualitatively different information. This is further supported by the direct comparison of the features extracted for all compounds from only the cytoplasmic region of the RNA and ER channels (CPP) with the merged RNA/ER channel (CP) (Supplementary Fig. 2C). Together, these examples illustrate that CPP expands the multiplexing capacity of the original CP method and, importantly, expands the diversity of the phenotypic profiles.
To assess the robustness of the CPP assay, we analyzed the reproducibility of the robust z-scores that were determined for the three intra-plate/technical (TRep) and the four inter-plate/biological (BRep) replicates in the CPP reference compound screen (Fig. 2E). All cell lines generally showed high Pearson correlation scores in the CPP assay for 20x or 40x magnification, which were similar to the Pearson correlation scores obtained when using the CP method in MCF-7 cells (Fig. 2F; Supplementary Fig. 3A). To directly compare the total variabilities of CPP across the four cell lines and between CPP and CP in MCF-7 cells, we further summed up the relative differences of the robust z-scores of all features for each individual TRep, BRep, and AllRep (Supplementary Data 4). The total variabilities between all replicate experiments of each cell line were similar. Those results were consistent with published CP performance metrics19. Notably, in MCF-7 cells, the CPP assay showed overall smaller total variabilities compared to the CP method. Our CP and CPP analyses further show that the coefficient of variation of feature data between cells depends mainly on the number of cells being imaged but not on the number of imaging fields per se, indicating that capturing ~ 2500 cells is a sufficient number to ensure statistical robustness of the data (Supplementary Fig. 1F). In conclusion, running the assay in single technical but at least four biological replicates using 20x magnification was also preferred in CPP to ensure its robustness, reproducibility, and applicability to HTPP platforms.
To determine the specific concentrations at which the reference compounds elicited phenotypic responses, we used the benchmark concentration (BMC) modeling approach as previously described for CP8,11,12,17. For each feature, the individual BMC corresponded to the concentration at which the phenotypic response exceeded the defined benchmark response (BMR) cutoff as described before48. For example, fulvestrant treatment induced clear concentration-dependent effects for specific Mito channel features (Fig. 3A). Again, the Pearson correlation scores of the BMC values obtained for CPP and CP in MCF-7 cells were very similar, with all cell lines generally showing lower Pearson correlation scores for the BMC values than the robust z-scores (Fig. 2F; Fig. 3C; Supplementary Fig. 3A, B), indicating that the BMC curve fitting increased data variability to some extent. To facilitate analysis of the BMC modeling data, we further assigned the individual features to biologically meaningful feature categories representing specific combinations of channels, cell regions, and analysis modules (Supplementary Data 5) in a similar way as previously described for CP8,11,12,17.
Using the BMC data and the feature category assignments, we visualized the relative number (proportion) of features that showed significant responses in each feature category. Those Proportion BMC profiles enabled direct comparison of the relative compound activities within and between the four cell lines (Fig. 3B, Supplementary Fig. 3C). Just as described for the compound activity profiles (Supplementary Fig. 2C), the Proportion BMC profiles were generally similar between CPP and CP (Fig. 3B). Notably, this comparison at the feature category level was hampered due to the differences in the number of imaging channels and number/type of extracted features. Several compounds such as rotenone showed broad activities across all feature categories and cell lines, with a large number of single features showing a significant response particularly in MCF-7 and RPTEC cells. Other compounds also showed broad but more cell-line specific activities. For example, the cytostatic drug etoposide showed activities only in proliferating HepG2, U2OS, and MCF-7 cancer cell lines but not in the post-mitotic, differentiated RPTEC primary kidney cells, which is in line with studies showing that etoposide sensitivity decreases with cellular differentiation49. Notably, treatment with the cytostatic drug fulvestrant resulted in phenotypic responses exclusively in MCF-7 cells reflecting its function as a specific inhibitor of the estrogen receptor alpha signaling pathway, which is a specific trait of the MCF-7 breast cancer cell line50. In addition to the described broad activities, the Proportion BMC profiles also enabled the identification of compounds showing activities that were more specific to different feature categories. For example, treatment with siramesine and tetrandrine caused pronounced responses in Lyso-related feature categories in MCF-7, HepG2, and U2OS cells, consistent with the activities described for these compounds41,42. These data illustrate that the visualization of relative compound activity profiles using biologically meaningful feature categories can support the characterization of the specific MoA of compounds.
To further distinguish between low-concentration (most sensitive, potentially primary) and high-concentration (less sensitive, potentially secondary) compound effects, we further investigated the specific sequence of feature category responses as well as the maximum sizes of those effects for CPP (Fig. 3D; Supplementary Fig. 4A–O) and CP (Supplementary Fig. 5A–O) using accumulation and magnitude plots as previously described for CP8,11,12,17. As illustrated by the example of fulvestrant (Fig. 3D), each data point in the BMC accumulation plot represents a different feature category in a ranked manner to enable the straightforward identification of sensitive feature categories that showed responses at lower concentrations. The corresponding BMC magnitude plot provides more detailed information about the BMCs and the effect sizes for all individual features of those feature categories that were displayed in the accumulation plot. For fulvestrant-treated MCF-7 cells, the BMC accumulation plot of CPP revealed concentration-dependent responses for 56 of the 62 feature categories, concordant with its broad activity shown in the Proportion BMC profile (Fig. 3B). In particular, CPP was able to distinguish between RNA and ER features showing maximum effect sizes, highlighting the advantage of separating RNA and ER channels to more precisely determine different MoA. The following three use cases describe some examples of applying the CPP assay and the described data analysis, visualization, and interpretation approaches to investigate specific research questions, thereby illustrating its added value.
Nucleoli are dynamic nuclear condensates that play a pivotal role in ribosome biogenesis and serve as important stress sensors that, when disrupted, trigger p53-dependent cell cycle arrest51. Changes in nucleoli number and/or morphology are associated with cancer, neurodegenerative disorders, and aging, thereby providing a relevant biomarker and therapeutic target52. Although being visualized by CP in the RNA channel, nucleoli have not yet been directly analyzed as a distinct cellular compartment in CP. Therefore, nucleoli were added in this study to the CPP and CP image and data analysis pipelines to profile potential compound effects on nucleoli-related features (Supplementary Data 3, Supplementary Data 5) such as changes in nucleoli numbers, which already served in genetic screens as a readout to identify regulators of ribosome biogenesis and cell cycle progression53,54.
In CPP, nucleoli were clearly detectable in the RNA channel based on their spot-like structure in the nucleus region (Fig. 4A), with MCF-7 and RPTEC exhibiting ~ 2 ± 1, HepG2 ~ 3 ± 2, and U2OS ~ 4 ± 2 nucleoli on average per cell. Rapamycin, a highly specific mTOR inhibitor, caused a strong effect on the number of nucleoli per cell (Fig. 4A; Supplementary Fig. 4I), consistent with the role of the mTOR pathway in controlling cell cycle progression through regulating ribosome biogenesis and nucleoli numbers52. The compound activity plots (Fig. 4A) and BMC accumulation plots (Fig. 4C) revealed that rapamycin showed activities already at lowest concentrations (3 nM) among the tested reference compounds in MCF-7, HepG2, and U2OS cells, but interestingly no relevant responses in RPTEC cells. As the lowest tested concentration of rapamycin was still too high for calculating a reliable BMC for the nucleoli number feature category, it was set to the next lower concentration (as described in the methods section). In fact, this suggests that nucleoli responses to rapamycin will be observed even in the sub-nanomolar range, corresponding to the published range of the IC50 value.
Furthermore, several perturbations in cell functions, such as impairment of the actin cytoskeleton or generation of reactive oxygen species (ROS), can cause changes in nucleolar number or morphologies and thus alter nucleolar function or trigger nucleolar stress response mechanisms55,56. Indeed, treatment with latrunculin B, a disruptor of the actin cytoskeleton, led to a clear increase of nucleoli numbers in MCF-7 cells but to a decrease in RPTEC cells (Fig. 4A–C, Supplementary Fig. 4G). Other compounds such as the Mito inhibitor rotenone, which led to mitochondria-mediated ROS generation in cell culture models57,58, caused prominent responses in both number and morphology of nucleoli (Fig. 4A–C, Supplementary Fig. 4J). Particularly in RPTEC and MCF-7 cells, rotenone treatment led to the observation of enlarged nucleoli signals at high but non-cytotoxic concentrations, which actually matched the signals in the DNA channel, indicating a severe disruption of the nucleolus structure.
In summary, addition of nucleoli as a distinct cellular compartment to the CPP and CP image and data analysis pipelines enabled comprehensive phenotypic profiling of compound effects on nucleoli and, thus, may provide relevant insights when studying mechanisms of ribosome biogenesis and the nucleolar stress response in basic research. Considering that increased number and size of nucleoli has been shown to correlate with elevated cancer cell proliferation and poor prognosis59, this may further support development of biomarkers and therapeutic intervention in a clinical context.
The actin cytoskeleton is a dynamic network of bundles of actin filaments (F-actin) that controls the shape and motility of cells but also determines the morphology of cellular organelles such as the Golgi apparatus and mitochondria60,61,62. Cytochalasin D (fugal biotoxin) and latrunculin B (marine biotoxin) both inhibit actin polymerization and disrupt actin filament organization, but act through different specific mechanisms63,64.
In contrast to the CP method, Actin and Golgi signals were captured in separate channels in CPP, which enabled the differentiation between Actin and Golgi responses of cells to compound treatment. In untreated cells, phalloidin staining revealed the characteristic network of actin filaments, e.g., with prominent cortical F-actin portions at the plasma membrane in MCF-7 cells or F-actin bundles forming stress fibers in U2OS and RPTEC cells (Fig. 5A). The corresponding Golgi staining showed the typical compact Golgi structures in close proximity to the nucleus in all cell lines, with some weaker signals at the plasma membrane. Upon exposure to cytochalasin D (Fig. 5A) and latrunculin B (Supplementary Fig. 6A), the actin cytoskeleton collapsed into intensely stained clumps and aggregates, which was accompanied by dispersed fragmentation and clumping of Golgi structures. Notably, these phenotypic changes occurred at different concentrations in cytochalasin D treated cells, with Actin features showing responses at lower concentrations than Golgi features, which was particularly evident in U2OS cells (Fig. 5B, Supplementary Data 23). In cells treated with latrunculin B, separation of Actin and Golgi effects were not that pronounced and generally occurred at higher concentrations (around 1 µM) as compared to cytochalasin D (around 0.1–0.3 µM) (Supplementary Fig. 6B, Supplementary Data 23). The BMC accumulation and magnitude plots further indicated a higher potency of cytochalasin D than latrunculin B with regard to the onset and the maximum effect size of the observed phenotypic changes of Actin and Golgi. Concordant with the essential role of the actin cytoskeleton in mitochondrial fusion and fission61, the BMC plots also revealed clear effects of cytochalasin D and latrunculin B on Mito features in both CPP (Fig. 5B; Supplementary Fig. 6B) and CP (Supplementary Fig. 5C, G). Importantly, when using the CP method, the BMC plots confirmed the effects of cytochalasin D and latrunculin B on the AGP channel (Fig. 5B, Supplementary Fig. 5G), with the caveat that differentiation between Actin and Golgi responses at different concentrations was not possible. This caveat was also evident when comparing the total number of BMC features responding to cytochalasin D and latrunculin B treatment in either the AGP channel (CP) or Actin and Golgi (CPP) in the BMC bar plots (Supplementary Data 23). Therefore, the specific sequence of these concentration-dependent responses to compound treatment that can be distinguished with the CPP assay suggest that perturbation of the actin cytoskeleton, e.g., caused by cytochalasin D, represents an event at lower concentrations leading to subsequent secondary effects on other subcellular compartments and organelles such as Golgi and Mito.
In order to identify reference compounds with activity profiles similar to cytochalasin D and latrunculin B, we used Spearman correlation and hierarchical clustering to generate profile similarity plots for each cell line (Fig. 5C, Supplementary Fig. 6C). These plots show the overall similarity of compound effects (robust z-score, feature level, see Fig. 2C, Supplementary Fig. 2B) at each highest non-cytotoxic concentration. To enable direct comparison of the compound activity profiles of CPP/Harmony, CP/Harmony, and CPP/Cell Profiler in MCF-7 cells, the Lyso features were excluded from this analysis as Lyso was not part of the CP dye set. In MCF-7 cells, the active reference compounds grouped into three major clusters (#4-6) comprising compounds with annotated Mito-, Lyso-, and Actin-related MoAs (see Fig. 2A, Fig. 5C). The “Mito cluster” (#5) comprised three compounds with the Mito inhibitors berberine chloride and rotenone showing the highest profile similarity in this group (dashed orange squares), which was concordant between CPP/Harmony, CP/Harmony, and CPP/Cell Profiler. In the other cell lines, the two compounds grouped into larger clusters (Supplementary Fig. 6C). The size of the “Lyso cluster” (#6) in MCF-7 cells varied more between CPP/Harmony, CP/Harmony and CPP/Cell Profiler but concordantly included the Lyso modulators fluphenazine and tetrandrine, which showed high profile similarities (dashed red squares). Since Lyso features were excluded from those plots, the clustering of the fluphenazine and tetrandrine was essentially based on their additional activities on ER and Golgi in CPP or RNA/ER and AGP in CP, as detailed in case study 3.
Interestingly, despite their common Actin-related MoA, no clear “Actin cluster” comprising cytochalasin D and latrunculin B was consistently observed in the profile similarity plots of the different cell lines (Fig. 5C, Supplementary Fig. 6C). This indicates that their overall compound activity profiles were rather distinct from each other, which was probably also due to the more pronounced Actin-related responses of cytochalasin D compared to latrunculin B, as shown in the corresponding BMC plots (Supplementary Fig. 5C, G) and the BMC bar plots (Supplementary Data 23). In fact, latrunculin B was part of a major cluster (#4) along with nocodazole and etoposide in both CPP/Harmony and CP/Harmony, and showed a higher correlation to the microtubule inhibitor nocodazole than to cytochalasin D in MCF-7, HepG2 and U2OS cells (dashed cyan squares). In contrast, the clustering of cytochalasin D clearly differed between CPP/Harmony and CP/Harmony in MCF-7 cells. In CP/Harmony, cytochalasin D was part of the “Lyso cluster” (#6), with a high correlation to fluphenazine and tetrandrine that both exerted AGP and prominent RNA/ER responses as shown by the corresponding BMC bar plots (Supplementary Data 23, 24). Similarly, cytochalasin D also showed considerable responses in those two CP channels, which contributed to the observed clustering. When using CPP/Harmony, those responses of fluphenazine and tetrandrine could be more clearly separated into stronger Golgi than Actin and stronger ER than RNA responses (Supplemental material 18 and 19). Importantly, in CPP/Harmony, cytochalasin D showed much stronger Actin than Golgi responses, which contributed to its clear separation from the other compounds in the profile similarity plots line (Fig. 5C), which was also the case in HepG2 and U2OS cells (Supplementary Fig. 6C).
In conclusion, this case study suggests that separating Actin and Golgi channels can be beneficial to enhance the organelle-specificity and diversity of phenotypic profiles for comparative MoA analyses. The example of cytochalasin D and latrunculin B highlights the need for generating phenotypic profiling data across a concentration range and using different analysis methods for detailed MoA analyses. BMC modeling helps to distinguish low-concentration (primary) from higher-concentration (secondary) effects. Investigating maximum effect levels at the highest non-cytotoxic concentrations using profile similarity plots supports clustering of compounds according to overall profile similarities but may miss important responses at different concentrations that contribute to distinct phenotypic profiles.
Lysosomes are essential cellular organelles responsible for degradation and recycling of macromolecules and autophagy. They are typically broadly distributed within cells and can form direct contact sites with the ER and mitochondria. The morphology, positioning, motility, and function of lysosomes are closely linked, making them an intensely studied cellular organelle65. We therefore included lysosomes as relevant organelles in the CPP assay to enable assessment of compound-induced Lyso phenotypes. We further evaluated the use of siramesine (accumulating in lysosomes leading to lysosome leakage and alterations in lysosomal pH)41, fluphenazine (involved in hypoxia-induced cell death leading to lysosome aggregation functional impairment)35, and tetrandrine (lysosomal deacidification agent perturbing autophagic flux)42 as suitable reference compounds for Lyso phenotypes, which were selected based on their published MoA.
Including Lyso features into the profile similarity plots (Fig. 6A, Supplementary Fig. 7C) further increased the correlation of the three “Lyso cluster” compounds tetrandrine, fluphenazine, and siramesine in all four cell lines as expected. However, in contrast to CPP/Harmony, those three compounds with annotated Lyso-related MoA (see Fig. 2A) did not form a separate “Lyso cluster” in the profile similarity plots of CPP/Cell Profiler, neither for the reduced set of features nor for the complete set (without feature selection applied) (Supplementary Fig. 8C). Interestingly, a high profile similarity to the three compounds was also observed for the multi-targeted tyrosine kinase inhibitor sunitinib malate and particularly evident in MCF-7 and U2OS cells, where only the addition of Lyso features led to the inclusion of sunitinib malate into the “Lyso cluster” in CPP/Harmony but not CPP/Cell Profiler, indicating a similar Lyso-related MoA. Indeed, exposure of MCF-7 cells to sunitinib malate, tetrandrine, fluphenazine, and siramesine led to intensely stained clumps and aggregates in the Lyso channel, accompanied by Golgi and ER disruption (Fig. 6B), which was also observed for the other cell lines (Supplementary Figs. 7A–10A). These pronounced Lyso responses were also evident in the corresponding BMC accumulation and magnitude plots, with primarily texture-related features accompanied with less strong phenotypic responses across different other organelles (Fig. 6C; Supplementary Figs. 7B–10B). Lyso responses to tetrandrine, fluphenazine, and siramesine generally occurred at lower concentrations compared to responses of the other organelles. Notably, this was the opposite in cells exposed to sunitinib malate, indicating that Lyso responses might occur as secondary effects. Moreover, the BMC plots further revealed differences in the magnitude of responses of the four cell lines to compound treatment, with particularly pronounced effects on ER and Golgi (Fig. 6C, Supplementary Figs. 7B–10B). However, when using the CP method, distinction of ER and Golgi responses was not possible from the combined RNA/ER and AGP channels (Fig. 6C, Supplementary Fig. 5E, L, N, O, Supplementary Data 23, 24).
Together, these data demonstrate the capacity of the CPP assay to elucidate compound effects on lysosomes and to differentiate between ER and RNA or Actin and Golgi effects, thereby providing valuable insights into the MoA of compounds. Furthermore, despite their diverse MoA, tetrandrine, fluphenazine, and siramesine induced Lyso phenotypes across all cell lines. Those compounds supported the identification of a Lyso activity of sunitinib malate, which is concordant with studies describing its sequestration and accumulation to lysosomes in cancer cells66,67, indicating their usability as suitable reference compounds in the CPP assay.
