Morphological profiling for drug discovery in the era of deep learning.

Morphological profiling is a valuable tool in phenotypic drug discovery. The advent of high-throughput automated imaging has enabled the capturing of a wide range of morphological features of cells or organisms in response to perturbations at the single-cell resolution. Concurrently, significant advances in machine learning and deep learning, especially in computer vision, have led to substantial improvements in analyzing large-scale high-content images at high throughput. These efforts have facilitated understanding of compound mechanism of action, drug repurposing, characterization of cell morphodynamics under perturbation, and ultimately contributing to the development of novel therapeutics. In this review, we provide a comprehensive overview of the recent advances in the field of morphological profiling. We summarize the image profiling analysis workflow, survey a broad spectrum of analysis strategies encompassing feature engineering- and deep learning-based approaches, and introduce publicly available benchmark datasets. We place a particular emphasis on the application of deep learning in this pipeline, covering cell segmentation, image representation learning, and multimodal learning. Additionally, we illuminate the application of morphological profiling in phenotypic drug discovery and highlight potential challenges and opportunities in this field.

Gatorbulin-1, a distinct cyclodepsipeptide chemotype, targets a seventh tubulin pharmacological site.

Tubulin-targeted chemotherapy has proven to be a successful and wide spectrum strategy against solid and liquid malignancies. Therefore, new ways to modulate this essential protein could lead to new antitumoral pharmacological approaches. Currently known tubulin agents bind to six distinct sites at α/ß-tubulin either promoting microtubule stabilization or depolymerization. We have discovered a seventh binding site at the tubulin intradimer interface where a novel microtubule-destabilizing cyclodepsipeptide, termed gatorbulin-1 (GB1), binds. GB1 has a unique chemotype produced by a marine cyanobacterium. We have elucidated this dual, chemical and mechanistic, novelty through multidimensional characterization, starting with bioactivity-guided natural product isolation and multinuclei NMR-based structure determination, revealing the modified pentapeptide with a functionally critical hydroxamate group; and validation by total synthesis. We have investigated the pharmacology using isogenic cancer cell screening, cellular profiling, and complementary phenotypic assays, and unveiled the underlying molecular mechanism by in vitro biochemical studies and high-resolution structural determination of the α/ß-tubulin-GB1 complex.

Discovery and Anti-Inflammatory Activity of a Cyanobacterial Fatty Acid Targeting the Keap1/Nrf2 Pathway.

The monounsaturated fatty acid 7(E)-9-keto-hexadec-7-enoic acid (1) and three structurally related analogues with different oxidation states and degrees of unsaturation (2-4) were discovered from a marine benthic cyanobacterial mat collected from Delta Shoal, Florida Keys. Their structures were elucidated using NMR spectroscopy and mass spectrometry. The structure of 1 contained an α,ß-unsaturated carbonyl system, a key motif required for the activation of the Keap1/Nrf2-ARE pathway that is involved in the activation of antioxidant and phase II detoxification enzymes. Compounds 1-4 were screened in ARE-luciferase reporter gene assay using stably transfected HEK293 cells, and only 1 significantly induced Nrf2 activity at 32 and 10 µM, whereas 2-4 were inactive. As there is crosstalk between inflammation and oxidative stress, subsequent biological studies were focused on 1 to investigate its anti-inflammatory potential. Compound 1 induced Nqo1, a well-known target gene of Nrf2, and suppressed iNos transcript levels, which translated into reduced levels of nitric oxide in LPS-activated mouse macrophage RAW264.7 cells, a more relevant model for inflammation. RNA sequencing was performed to capture the effects of 1 on a global level and identified additional canonical pathways and upstream regulators involved in inflammation and immune response, particularly those related to multiple sclerosis. A targeted survey of marine cyanobacterial samples from other geographic locations, including Guam, suggested the widespread occurrence of 1. Furthermore, the previous isolation of 1 from marine diatoms and green algae implied a potentially important ecological role across marine algal eukaryotes and prokaryotes. The previous isolation from sea lettuce raises the possibility of dietary intervention to attenuate inflammation and related disease progression.

A tetrapeptide class of biased analgesics from an Australian fungus targets the µ-opioid receptor.

An Australian estuarine isolate of Penicillium sp. MST-MF667 yielded 3 tetrapeptides named the bilaids with an unusual alternating LDLD chirality. Given their resemblance to known short peptide opioid agonists, we elucidated that they were weak (Ki low micromolar) µ-opioid agonists, which led to the design of bilorphin, a potent and selective µ-opioid receptor (MOPr) agonist (Ki 1.1 nM). In sharp contrast to all-natural product opioid peptides that efficaciously recruit ß-arrestin, bilorphin is G protein biased, weakly phosphorylating the MOPr and marginally recruiting ß-arrestin, with no receptor internalization. Importantly, bilorphin exhibits a similar G protein bias to oliceridine, a small nonpeptide with improved overdose safety. Molecular dynamics simulations of bilorphin and the strongly arrestin-biased endomorphin-2 with the MOPr indicate distinct receptor interactions and receptor conformations that could underlie their large differences in bias. Whereas bilorphin is systemically inactive, a glycosylated analog, bilactorphin, is orally active with similar in vivo potency to morphine. Bilorphin is both a unique molecular tool that enhances understanding of MOPr biased signaling and a promising lead in the development of next generation analgesics.

Anti-Inflammatory Dysidazirine Carboxylic Acid from the Marine Cyanobacterium Caldora sp. Collected from the Reefs of Fort Lauderdale, Florida.

Dysidazirine carboxylic acid (1) was isolated from the lipophilic extract of a collection of the benthic marine cyanobacterium Caldora sp. from reefs near Fort Lauderdale, Florida. The planar structure of this new compound was determined by spectroscopic methods and comparisons between HRMS and NMR data with its reported methyl ester. The absolute configuration of the single chiral center was determined by the conversion of 1 to the methyl ester and the comparison of its specific rotation data with the two known methyl ester isomers, 2 and 3. Molecular sequencing with 16S rDNA indicated that this cyanobacterium differs from Caldora penicillata (Oscillatoriales) and represents a previously undocumented and novel Caldora species. Dysidazirine (2) showed weak cytotoxicity against HCT116 colorectal cancer cells (IC50 9.1 µM), while dysidazirine carboxylic acid (1) was non-cytotoxic. Similar cell viability patterns were observed in RAW264.7 cells with dysidazirine only (2), displaying cytotoxicity at the highest concentration tested (50 µM). The non-cytotoxic dysidazirine carboxylic acid (1) demonstrated anti-inflammatory activity in RAW264.7 cells stimulated with LPS. After 24 h, 1 inhibited the production of NO by almost 50% at 50 µM, without inducing cytotoxicity. Compound 1 rapidly decreased gene expression of the pro-inflammatory gene iNOS after 3 h post-LPS treatment and in a dose-dependent manner (IC50 ~1 µM); the downregulation of iNOS persisted at least until 12 h.

Fungal Epithiodiketopiperazines Carrying α,ß-Polysulfide Bridges from Penicillium steckii YE, and Their Chemical Interconversion.

Some fungal epithiodiketopiperazine alkaloids display α,ß-polysulfide bridges alongside diverse structural variations. However, the logic of their chemical diversity has rarely been explored. Here, we report the identification of three new (2, 3, 8) and five known (1, 4-7) epithiodiketopiperazines of this subtype from a marine-derived Penicillium sp. The structure elucidation was supported by multiple spectroscopic analyses. Importantly, we observed multiple nonenzymatic interconversions of these analogues in aqueous solutions and organic solvents. Furthermore, the same biosynthetic origin of these compounds was supported by one mined gene cluster. The dominant analogue (1) demonstrated selective cytotoxicity to androgen-sensitive prostate cancer cells and HIF-depleted colorectal cells and mild antiaging activities, linking the bioactivity to oxidative stress. These results provide crucial insight into the formation of fungal epithiodiketopiperazines through chemical interconversions.

Function-Oriented and Modular (+/-)-cis-Pseudoguaianolide Synthesis: Discovery of New Nrf2 Activators and NF-κB Inhibitors.

Described herein is a function-oriented synthesis route and biological evaluation of pseudoguaianolide analogues. The 10-step synthetic route developed retains the topological complexity of the natural product, installs functional handles for late-stage diversification, and forges the key bioactive Michael acceptors early in the synthesis. The analogues were found to be low-micromolar Nrf2 activators and micromolar NF-κB inhibitors and dependent on the local environment of the Michael acceptor moieties.

Dolastatin 15 from a Marine Cyanobacterium Suppresses HIF-1α Mediated Cancer Cell Viability and Vascularization.

Chemical investigation of a benthic marine cyanobacterium yielded the anticancer agent dolastatin 15, originally isolated from a mollusk. Dolastatin 15 is a microtubule-destabilizing agent with analogues undergoing clinical evaluation. Profiling against a panel of isogenic HCT116 colorectal cancer cells showed remarkable differential cytotoxicity against the parental cells over isogenic cells lacking HIF or other key players in the pathway, including oncogenic KRAS and VEGF. Dolastatin 15 displayed an antivascularization effect in human endothelial cells and in zebrafish vhl mutants with activated Hif, thus signifying its clinical potential as a treatment for solid tumors with an angiogenic component. Global transcriptome analysis with RNA sequencing suggested that dolastatin 15 could affect other major cancer pathways that might not directly involve tubulin or HIF. The identification of the true producer of a clinically relevant agent is important for sustainable supply, as is understanding the biosynthesis, and future genetic manipulation of the biosynthetic gene cluster for analogue production.

Yohimbine as a Starting Point to Access Diverse Natural Product-Like Agents with Re-programmed Activities against Cancer-Relevant GPCR Targets.

G protein-coupled receptors (GPCRs) constitute the largest protein superfamily in the human genome. GPCRs play key roles in mediating a wide variety of physiological events including proliferation and cancer metastasis. Given the major roles that GPCRs play in mediating cancer growth, they present promising targets for small molecule therapeutics. One of the principal goals of our lab is to identify complex natural products (NPs) suitable for ring distortion, or the dramatic altering of the inherently complex architectures of NPs, to rapidly generate an array of compounds with diverse molecular skeletal systems. The overarching goal of our ring distortion approach is to re-program the biological activity of select natural products and identify new compounds of importance to the treatment of disease, such as cancer. Described herein are the results from biological screens of diverse small molecules derived from the indole alkaloid yohimbine against a panel of GPCRs involved in various diseases. Several analogues displayed highly differential antagonistic activities across the GPCRs tested. We highlight the re-programmed profile of one analogue, Y7g, which exhibited selective antagonistic activities against AVPR2 (IC50 = 459 nM) and OXTR (IC50 = 1.16 µM). The activity profile of Y7g could correlate its HIF-dependent anti-cancer activity to its GPCR antagonism since these receptors are known to be upregulated in hypoxic cellular environments. Our findings demonstrate that the ring distortion of yohimbine can lead to the identification of new compounds capable of interacting with distinct cancer-relevant targets.

Vicinal difluorination as a C=C surrogate: an analog of piperine with enhanced solubility, photostability, and acetylcholinesterase inhibitory activity.

Piperine, a natural product derived from peppercorns, has a variety of biological activities that make it an attractive lead compound for medicinal chemistry. However, piperine has some problematic physicochemical properties including poor aqueous solubility and a susceptibility to UV-induced degradation. In this work, we designed an analog of piperine in which the central conjugated hydrocarbon chain is replaced with a vicinal difluoroalkane moiety. We show that this fluorinated analog of piperine has superior physicochemical properties, and it also has higher potency and selectivity towards one particular drug target, acetylcholinesterase. This work highlights the potential usefulness of the threo-difluoroalkane motif as a surrogate for E-alkenes in medicinal chemistry.

Development of apratoxin S10 (Apra S10) as an anti-pancreatic cancer agent and its preliminary evaluation in an orthotopic patient-derived xenograft (PDX) model.

Despite the significant progress in the field of cancer therapeutics, the incidence of pancreatic cancer (PC) has continuously increased. One possible mechanism for this increasing burden is impaired drug delivery and drug resistance resulting from a unique tumor microenvironment and genetic mutations. Apratoxins are potent anticancer agents and cotranslational translocation inhibitors with potential therapeutic applications to treat cancers with active secretory pathways. Here, we developed apratoxin S10 (Apra S10) as an anti-pancreatic cancer agent which potently inhibited the growth of both established and patient-derived primary pancreatic cancer cells. We validated its mechanism of action on pancreatic cancer cells by demonstrating the downregulation of multiple receptor tyrosine kinases and inhibition of growth factor and cytokine secretion. Apra S10 also inhibited a number of cytokines secreted by stromal cells, suggesting that Apra S10 not only inhibited pancreatic cancer cell secretion, but also reduced the level of factors secreted by other cell types active within the tumor microenvironment. As Apra S10 tissue distribution indicated its high enrichment in pancreas tissue, an orthotopic pancreatic patient-derived xenograft mouse model that closely mimics the human pancreatic tumor microenvironment was for the first time used in apratoxin studies. Apra S10 showed promising antitumor effect in this pancreatic cancer model and this effect was mediated through anti-proliferation properties.

Englerin A Inhibits EWS-FLI1 DNA Binding in Ewing Sarcoma Cells.

High-throughput screening of extracts from plants, marine, and micro-organisms led to the identification of the extract from the plant Phyllanthus engleri as the most potent inhibitor of EWS-FLI1 induced luciferase reporter expression. Testing of compounds isolated from this extract in turn led to the identification of Englerin A (EA) as the active constituent of the extract. EA induced both necrosis and apoptosis in Ewing cells subsequent to a G2M accumulation of cells in the cell cycle. It also impacted clonogenic survival and anchorage-independent proliferation while also decreasing the proportion of chemotherapy-resistant cells identified by high ALDH activity. EA also caused a sustained increase in cytosolic calcium levels. EA appears to exert its effect on Ewing cells through a decrease in phosphorylation of EWS-FLI1 and its ability to bind DNA. This effect is mediated, at least in part, through a decrease in the levels of the calcium-dependent protein kinase PKC-ßI after a transient up-regulation.

A Tryptoline Ring-Distortion Strategy Leads to Complex and Diverse Biologically Active Molecules from the Indole Alkaloid Yohimbine.

High-throughput screening (HTS) is the primary driver to current drug-discovery efforts. New therapeutic agents that enter the market are a direct reflection of the structurally simple compounds that make up screening libraries. Unlike medically relevant natural products (e.g., morphine), small molecules currently being screened have a low fraction of sp3 character and few, if any, stereogenic centers. Although simple compounds have been useful in drugging certain biological targets (e.g., protein kinases), more sophisticated targets (e.g., transcription factors) have largely evaded the discovery of new clinical agents from screening collections. Herein, a tryptoline ring-distortion strategy is described that enables the rapid synthesis of 70 complex and diverse compounds from yohimbine (1); an indole alkaloid. The compounds that were synthesized had architecturally complex and unique scaffolds, unlike 1 and other scaffolds. These compounds were subjected to phenotypic screens and reporter gene assays, leading to the identification of new compounds that possessed various biological activities, including antiproliferative activities against cancer cells with functional hypoxia-inducible factors, nitric oxide inhibition, and inhibition and activation of the antioxidant response element. This tryptoline ring-distortion strategy can begin to address diversity problems in screening libraries, while occupying biologically relevant chemical space in areas critical to human health.

Synthesis and biological evaluation of largazole zinc-binding group analogs.

Histone acetylation is an extensively investigated post-translational modification that plays an important role as an epigenetic regulator. It is controlled by histone acetyl transferases (HATs) and histone deacetylases (HDACs). The overexpression of HDACs and consequent hypoacetylation of histones have been observed in a variety of different diseases, leading to a recent focus of HDACs as attractive drug targets. The natural product largazole is one of the most potent natural HDAC inhibitors discovered so far and a number of largazole analogs have been prepared to define structural requirements for its HDAC inhibitory activity. However, previous structure-activity relationship studies have heavily investigated the macrocycle region of largazole, while there have been only limited efforts to probe the effect of various zinc-binding groups (ZBGs) on HDAC inhibition. Herein, we prepared a series of largazole analogs with various ZBGs and evaluated their HDAC inhibition and cytotoxicity. While none of the analogs tested were as potent or selective as largazole, the Zn2+-binding affinity of each ZBG correlated with HDAC inhibition and cytotoxicity. We expect that our findings will aid in building a deeper understanding of the role of ZBGs in HDAC inhibition as well as provide an important basis for the future development of new largazole analogs with non-thiol ZBGs as novel therapeutics for cancer.

Discovery, Total Synthesis and Key Structural Elements for the Immunosuppressive Activity of Cocosolide, a Symmetrical Glycosylated Macrolide Dimer from Marine Cyanobacteria.

A new dimeric macrolide xylopyranoside, cocosolide (1), was isolated from the marine cyanobacterium preliminarily identified as Symploca sp. from Guam. The structure was determined by a combination of NMR spectroscopy, HRMS, X-ray diffraction studies and Mosher's analysis of the base hydrolysis product. Its carbon skeleton closely resembles that of clavosolides A-D isolated from the sponge Myriastra clavosa, for which no bioactivity is known. We performed the first total synthesis of cocosolide (1) along with its [α,α]-anomer (26) and macrocyclic core (28), thus leading to the confirmation of the structure of natural 1. The convergent synthesis featured Wadsworth-Emmons cyclopropanation, Sakurai annulation, Yamaguchi macrocyclization/dimerization reaction, α-selective glycosidation and ß-selective glycosidation. Compounds 1 and 26 potently inhibited IL-2 production in both T-cell receptor dependent and independent manners. Full activity requires the presence of the sugar moiety as well as the intact dimeric structure. Cocosolide also suppressed the proliferation of anti-CD3-stimulated T-cells in a dose-dependent manner.

Tasiamide F, a potent inhibitor of cathepsins D and E from a marine cyanobacterium.

In search of novel protease inhibitors with therapeutic potential, our efforts exploring the marine cyanobacterium Lyngbya sp. have led to the discovery of tasiamide F (1), which is an analogue of tasiamide B (2). The structure was elucidated using a combination of NMR spectroscopy and mass spectrometry. The key structural feature in 1 is the presence of the Phe-derived statine core, which contributes to its aspartic protease inhibitory activity. The antiproteolytic activity of 1 and 2 was evaluated in vitro against cathepsins D and E, and BACE1. Tasiamide F (1) displayed IC50 values of 57nM, 23nM, and 0.69µM, respectively, indicating greater selectivity for cathepsins over BACE1 compared with tasiamide B (2). Molecular docking experiments were carried out for compounds 1 and 2 against cathepsins D and E to rationalize their activity towards these proteases. The dysregulated activities of cathepsins D and E have been implicated in cancer and modulation of immune responses, respectively, and these proteases represent potential therapeutic targets.

C3 and 2D C3 Marfey's Methods for Amino Acid Analysis in Natural Products.

We validate the improved resolution and sensitivity of the C3 Marfey's method, including an ability to resolve all Ile isomers, against an array of amino acids commonly encountered in natural products and by comparison to an existing Marfey's method. We also describe an innovative 2D C3 Marfey's method as an analytical approach for determining the regiochemistry of enantiomeric amino acid residues in natural products. The C3 and 2D C3 Marfey's methods represent valuable tools for probing and defining the stereocomplexity of hydrolytically accessible amino acid residues in natural products.

Caldoramide, a Modified Pentapeptide from the Marine Cyanobacterium Caldora penicillata.

The isolation, structure determination, and biological activities of a new linear pentapeptide, caldoramide (5), from the marine cyanobacterium Caldora penicillata from Florida are described. Caldoramide (5) has structural similarities to belamide A (4), dolastatin 10 (1), and dolastatin 15 (2). We profiled caldoramide against parental HCT116 colorectal cancer cells and isogenic cells lacking oncogenic KRAS or hypoxia-inducible factors 1α (HIF-1α) and 2α (HIF-2α). Caldoramide (5) showed differential cytotoxicity for cells containing both oncogenic KRAS and HIF over the corresponding knockout cells. LCMS dereplication indicated the presence of caldoramide (5) in a subset of C. penicillata samples.

Grassypeptolides as natural inhibitors of dipeptidyl peptidase 8 and T-cell activation.

Natural products made by marine cyanobacteria are often highly modified peptides and depsipeptides that have the potential to act as inhibitors for proteases. In the interests of finding new protease inhibition activity and selectivity, grassypeptolide A (1) was screened against a panel of proteases and found to inhibit DPP8 selectively over DPP4. Grassypeptolides were also found to inhibit IL-2 production and proliferation in activated T-cells, consistent with a putative role of DPP8 in the immune system. These effects were also observed in Jurkat cells, and DPP activity in Jurkat cell cytosol was shown to be inhibited by grassypeptolides. In silico docking suggests two possible binding modes of grassypeptolides-at the active site of DPP8 and at one of the entrances to the internal cavity. Collectively these results suggest that grassypeptolides might be useful tool compounds in the study of DPP8 function.

Morphological Profiling for Drug Discovery in the Era of Deep Learning.

Morphological profiling is a valuable tool in phenotypic drug discovery. The advent of high-throughput automated imaging has enabled the capturing of a wide range of morphological features of cells or organisms in response to perturbations at the single-cell resolution. Concurrently, significant advances in machine learning and deep learning, especially in computer vision, have led to substantial improvements in analyzing large-scale high-content images at high-throughput. These efforts have facilitated understanding of compound mechanism-of-action (MOA), drug repurposing, characterization of cell morphodynamics under perturbation, and ultimately contributing to the development of novel therapeutics. In this review, we provide a comprehensive overview of the recent advances in the field of morphological profiling. We summarize the image profiling analysis workflow, survey a broad spectrum of analysis strategies encompassing feature engineering- and deep learning-based approaches, and introduce publicly available benchmark datasets. We place a particular emphasis on the application of deep learning in this pipeline, covering cell segmentation, image representation learning, and multimodal learning. Additionally, we illuminate the application of morphological profiling in phenotypic drug discovery and highlight potential challenges and opportunities in this field.