Morphological profiling of small molecules

Profiling approaches such as gene expression or proteome profiling generate small-molecule bioactivity profiles that describe a perturbed cellular state in rather unbiased manner and have become indispensable tools the evaluation of bioactive small molecules. Automated imaging image analysis can record morphological alterations are induced by molecules through detection hundreds features high-throughput experiments. Thus, has gained growing attention academia pharmaceutical industry it enables compound collections broader biological context early stages development drug-discovery process. may be used successfully to predict mode action targets unexplored compounds uncover unanticipated activity for already characterized Here, we review reported kind detected so far and, thus, predicted. A phenotype unites observable characteristics an organism cell, protein expression, morphology, biochemical properties, results from interaction genotype environment (Nussinov et al., 2019Nussinov R. Tsai C.J. Jang H. Protein ensembles link phenotype.PLoS Comput. Biol. 2019; 15https://doi.org/10.1371/journal.pcbi.1006648Crossref Scopus (18) Google Scholar). Cell morphology been linked specific states processes predictive value genetic, chemical, disease-related perturbations. However, often not obvious human eye, which able discern subtle changes independent visualization tools, thus calling reliable methods. There high demand detailed mapping space, i.e., targets, off-targets, (MoA) general more importantly, drug candidates particular. Whereas available detect mostly address known drug-target e.g., G-protein-coupled receptors (GPCRs), kinases, enzymes, general, “omics” transcriptomics, proteomics, epigenomics, metabolomics enable collecting all measurable parameters obtain holistic view given state. Although omics studies rarely provide direct proof inherently rich data they deliver inform about numerous altered traits between two states, particular when different combined. Moreover, date, generic methods target identification available, is labor- time-intensive (Saxena, 2016Saxena C. Identification binding partners using label-free methods.Expert Opin. Drug Dis. 2016; 11: 1017-1025Crossref PubMed (12) Scholar; Wilkinson 2020Wilkinson I.V.L. Terstappen G.C. Russell A.J. Combining experimental strategies successful deconvolution.Drug Discov. Today. 2020; 25: 1998-2005Crossref (8) Ziegler 2013Ziegler S. Pries V. Hedberg Waldmann Target molecules: finding needle haystack.Angew. Chem. Int. Ed. Engl. 2013; 52: 2744-2792Crossref (325) Complementary employ comparison structural similarity (Awale Reymond, 2019Awale M. Reymond J.L. Web-based polypharmacology prediction.Methods Mol. 1888: 255-272Crossref Byrne Schneider, 2019Byrne Schneider G. In silico prediction molecules.Methods 273-309Crossref (20) Scholar), (Lamb 2006Lamb J. Crawford E.D. Peck D. Modell J.W. Blat I.C. Wrobel M.J. Lerner Brunet J.P. Subramanian A. Ross K.N. al.The connectivity map: gene-expression signatures connect molecules, genes, disease.Science. 2006; 313: 1929-1935Crossref (3178) 2017Subramanian Narayan Corsello S.M. D.D. Natoli T.E. Lu X. Gould Davis J.F. Tubelli A.A. Asiedu J.K. al.A next generation L1000 platform first 1,000,000 profiles.Cell. 2017; 171: 1437-1452 e1417Abstract Full Text PDF (878) selective cell toxicity deduce hypotheses (Rees 2016Rees M.G. Seashore-Ludlow B. Cheah J.H. Adams D.J. Price E.V. Gill Javaid Coletti M.E. Jones V.L. Bodycombe N.E. al.Correlating chemical sensitivity basal reveals mechanism action.Nat. 12: 109-116Crossref (299) 2015Seashore-Ludlow Rees Cokol Soule C.K. al.Harnessing large-scale dataset.Cancer 2015; 5: 1210-1223Crossref (314) Similar modulation similar (Keiser 2007Keiser Roth B.L. Armbruster B.N. Ernsberger P. Irwin J.J. Shoichet B.K. Relating pharmacology ligand chemistry.Nat. Biotechnol. 2007; 197-206Crossref (1208) Scholar, Keiser 2009Keiser Setola Laggner Abbas A.I. Hufeisen S.J. Jensen N.H. Kuijer M.B. Matos R.C. Tran T.B. al.Predicting new molecular drugs.Nature. 2009; 462: 175-U148Crossref (1190) Scholar) patterns provoked with common targeted pathways By analogy, growth inhibition broad cell-line panel employed target/MoA challenge at heart biology how (synthesized) possible various simultaneously optimization. Omics per se well suited this purpose; however, amenable screening. The advent automated high-content algorithms pattern recognition, feature extraction, paved way additional pillar suite approaches, profiling, research discovery. High-content assays usually designed monitor process limited number parameters. extraction exploration small-molecule-related manner. For (Figure 1), components stained fluorescent dyes antibodies upon perturbation Alternatively, tags genetically introduced into cells directly markers. capture images each component, segmentation identifies subcellular regions (Bougen-Zhukov 2017Bougen-Zhukov N. Loh S.Y. Lee H.K. Loo L.H. Large-scale image-based screening phenotypes.Cytometry 91: 115-125Crossref (32) Hundreds numerical phenotypic descriptors then extracted phenotypes (as compared non-perturbed state) Boutros 2015Boutros Heigwer F. Laufer Microscopy-based screening.Cell. 163: 1314-1325Abstract (178) Caicedo 2017Caicedo J.C. Cooper Warchal Qiu Molnar Vasilevich A.S. Barry J.D. Bansal H.S. Kraus O. al.Data-analysis profiling.Nat. Methods. 14: 849-863Crossref (234) Grys 2017Grys B.T. Lo D.S. Sahin O.Z. Morris Q. Boone Andrews B.J. Machine learning computer vision profiling.J. 216: 65-71Crossref (79) These related size shape organelles, intensity, texture, among others. Profiles agents action, where profile implies MoA target. Dimensionality reduction performed convert high-dimensional low-dimensional removing redundant irrelevant while retaining variance dataset Clusters MoA, mapped unsupervised machine learning, expected output unknown (Grys Phenotypic categories defined prior supervised train models will assign existing power MoAs annotated compounds, bioactivity, biologically uncharacterized This aims give overview currently according system components, assessed these (see Table S1), lessons learned their application. It does focus on computational Scheeder 2018Scheeder discovery.Curr. Syst. 2018; 10: 43-52Crossref (70) We lay emphasis Painting Assay (CPA) (Bray 2016Bray M.A. Singh Han C.T. Borgeson Hartland Kost-Alimova Gustafsdottir Gibson C.C. Carpenter A.E. Painting, assay multiplexed dyes.Nat. Protoc. 1757-1774Crossref (211) 2013Gustafsdottir Ljosa Sokolnicki K.L. Anthony Wilson Walpita Kemp M.M. Petri Seiler K. Carrel H.A. Golub T.R. Schreiber S.L. al.Multiplex cytological measure diverse states.PLoS One. 8: e80999Crossref (105) require genetic manipulation, variety lines, found acceptance application community increasingly industry. major compartments (endoplasmic reticulum [ER], Golgi, mitochondria, lysosomes, endosomes, actin tubulin cytoskeleton, nucleoli, nucleus) was Boland Murphy, 2001Boland M.V. Murphy R.F. neural network classifier capable recognizing structures fluorescence microscope HeLa cells.Bioinformatics. 2001; 17: 1213-1223Crossref (375) cells. Upon staining one DNA, 84 were which, example, monitored patterns, object distance measures respect center, size, stain overlap nucleus, DNA standardization consistent cells, localization referred landmark. pioneering work techniques differ type visualization, addressed following sections. Morphobase, database employs bright-field srcts-NRK treatment (Futamura 2012Futamura Y. Kawatani Kazami Tanaka Muroi Shimizu T. Tomita Watanabe Osada encyclopedic database, its use identification.Chem. 2012; 19: 1620-1630Abstract (69) clusters several classes tubulin, actin, synthesis, histone deacetylase (HDAC), heat-shock proteins (HSPs) (Table S1). Importantly, HDAC proteasome inhibitors could distinguished srcts-NRK, whereas RNA synthesis separated both lines considered, activities clearly differentiated. Morphobase mitochondrial complex I inhibitor rotenone cyclin-dependent kinase (CDK) 3-ATA, polypharmacological resveratrol synthesis. As observed had previously associated findings emphasize need complete annotation reference only nominal target, most commonly (Moret 2019Moret Clark N.A. Hafner Wang Lounkine E. Medvedovic Gray Jenkins Sorger P.K. Cheminformatics analyzing designing optimized libraries.Cell 26: 765-777Abstract (25) public databases vendors’ websites. Further identified also combination ChemProteobase Futamura 2013Futamura Aono Nogawa novel fungal metabolite, pyrrolizilactone, systems.Chembiochem. 2456-2463Crossref (23) Minegishi 2015Minegishi Fukashiro Nakamura Methyl 3-((6-methoxy-1,4-dihydroindeno[1,2-c]pyrazol-3-yl)amino)benzoate (GN39482) polymerization MorphoBase ChemProteoBase methods.J. Med. 58: 4230-4241Crossref (28) Markers (5-ethynyl-3-deoxyuridine [EdU]) mitosis (phospho-histone H3) along (Young 2008Young D.W. Bender Hoyt McWhinnie Chirn G.W. Tao C.Y. Tallarico J.A. Labow Mitchison T.J. al.Integrating ligand-target identify 2008; 4: 59-68Crossref (268) information-rich stemmed nuclear stain, nucleus quantity. Profile revealed around vacuolar ATPase, antimitotics, corticosteroids, progesterone-related compounds. Interestingly, cell-cycle-arrest-related cluster featured cardiac glycosides, Na+/K+-ATPases, translation emetine cycloheximide, steroid hormones progesterone danatrol. authors nature based similarity. Overall, better correlated predicted than structures. addition visualizing microtubules Golgi stainings five (SKOV3, A549, SF268, DU145, HUVEC [human umbilical vein endothelial cell], see S1) (Adams 2006Adams C.L. Kutsyy Coleman D.A. Cong Crompton A.M. Elias K.A. Oestreicher D.R. Trautman Vaisberg Compound classification phenotypes.Methods Enzymol. 440-468Crossref (42) 2005Tanaka Bateman Rauh Ramachandani Zhang Hansen K.C. Burlingame A.L. Shokat K.M. al.An morphology-based screen new, active molecules.PLoS 2005; 3: e128https://doi.org/10.1371/journal.pbio.0030128Crossref (197) Subpopulation group populations depending cell-cycle phase employing information content, condensation. Different structurally Src PP2 hydroxy-PP, A549 HUVECs, effects SKOV3 similar. Like PP2, hydroxy-PP inhibited kinases Fyn, p38-α, B, explain lack HUVECs stem levels proteins. line set represents tissues backgrounds beneficial changes. Carragher's DNA-, actin-, tubulin-staining four (Ovcar3, MiaPaCa2, MCF7, MCF7 truncated, dominant negative TP53 mutant) (Caie 2010Caie P.D. Walls R.E. Ingleston-Orme Daya Houslay Eagle Roberts Carragher N.O. response across distinct cancer cells.Mol. Cancer Ther. 2010; 9: 1913-1926Crossref (94) They even subpopulations cell-shape epithelial mesenchymal differentiated proteases, modulators, Aurora Eg5 2013Ljosa Caie Ter Horst E.L. Genovesio al.Comparison responses treatment.J. Biomol. Screen. 18: 1321-1329Crossref (84) Of note, tubulin-targeting stabilizing (Taxol, epothilone B) destabilizing (colchicine, nocodazole) agents. resistance selectivity some other lines. Ovcar3 resistant classes, MiaPaCa2 sensitive perturbations phospho-histone H2A.X classify (Twarog 2016Twarog N.R. Low Currier D.G. Miller Chen Shelat Robust minimalist microscopy multidimensional trajectory analysis.PLoS e0149439https://doi.org/10.1371/journal.pone.0149439Crossref (9) approach kinase, proteasome, topoisomerases, HDAC, antimetabolites, HSP90. mammalian rapamycin (mTOR) phosphatidylinositol 3-kinase (PI3K) classified due low assay, contrast (Reisen 2015Reisen Sauty de Chalon Pfeifer Gabriel Selzer Linking modes fingerprints.Assay Dev. Technol. 13: 415-427Crossref (40) Genotype-dependent 12 isogenic parental HCT116 (Breinig 2015Breinig Klein F.A. Huber W. chemical-genetic map 846https://doi.org/10.15252/msb.20156400Crossref (43) harbored deletions oncogenic mutations knockouts Both genotype-dependent genotype-independent compound-treated drug-gene interactions varied panel. Drug-gene-phenotype 15% provided crosstalk signaling potential synergy. clustered MEK, p38, glucocorticoid receptor, alkylation, proton gradient. connected iron chelator, antifolates, methyltransferase inhibition. Uncoupling gradient Cδ (PKCδ) rottlerin 2A), 2001 (Soltoff, 2001Soltoff S.P. Rottlerin uncoupler decreases ATP indirectly blocks C delta tyrosine phosphorylation.J. 276: 37986-37992Abstract addition, matrix aldehyde dehydrogenase (ALDH) disulfiram, factor receptor (EGFR) tyrphostin AG555, κB (NF-κB) caffeic acid phenyl ester (CAPE), experimentally validated Figure 2B). physiological setup membrane integrity patient-derived organoids (PDOs) 19 patients colorectal clinical (Betge 2019Betge Rindtorff Sauer Rauscher Dingert Gaitantzi Herweck Miersch Valentini Hauber al.Multiparametric phenotyping patient derived organoids.bioRxiv. : 660993https://doi.org/10.1101/660993Crossref (0) Analysis compound-induced PI3K/AKT/mTOR, glycogen synthase 3 (GSK3), EGFR, CDK MEK PDO AKT GSK3 subset PDOs. mechanistic insights mutations. multiple sets parallel processed samples spectrally separable seminal Altschuler Wu groups explored 100 drugs sets: (1) splicing SC35 cytokinesis anillin; (2) β-tubulin actin; (3) phospho-p38 phospho-extracellular signal-regulated (ERK); (4) p53, c-Fos; (5) phosphoadenosine 3,5-monophosphate element-binding (CREB) calmodulin (Perlman 2004Perlman Z.E. Slack M.D. Feng L.F. Multidimensional microscopy.Science. 2004; 306: 1194-1198Crossref (513) strong responses, changed differently concentrations, implying differences concentrations engagement affinities. camptothecin inhibits concentration topoisomerase causes S-phase arrest, transcription further processes. poorly austocystin inhibitors. Already report made essential observations: displayed irrespective