Evaluating Scalable Supervised Learning for Synthesize-on-Demand Chemical Libraries.

Traditional small-molecule drug discovery is a time-consuming and costly endeavor. High-throughput chemical screening can only assess a tiny fraction of drug-like chemical space. The strong predictive power of modern machine-learning methods for virtual chemical screening enables training models on known active and inactive compounds and extrapolating to much larger chemical libraries. However, there has been limited experimental validation of these methods in practical applications on large commercially available or synthesize-on-demand chemical libraries. Through a prospective evaluation with the bacterial protein-protein interaction PriA-SSB, we demonstrate that ligand-based virtual screening can identify many active compounds in large commercial libraries. We use cross-validation to compare different types of supervised learning models and select a random forest (RF) classifier as the best model for this target. When predicting the activity of more than 8 million compounds from Aldrich Market Select, the RF substantially outperforms a naïve baseline based on chemical structure similarity. 48% of the RF's 701 selected compounds are active. The RF model easily scales to score one billion compounds from the synthesize-on-demand Enamine REAL database. We tested 68 chemically diverse top predictions from Enamine REAL and observed 31 hits (46%), including one with an IC50 value of 1.3 µM.

Predicting kinase inhibitors using bioactivity matrix derived informer sets.

Prediction of compounds that are active against a desired biological target is a common step in drug discovery efforts. Virtual screening methods seek some active-enriched fraction of a library for experimental testing. Where data are too scarce to train supervised learning models for compound prioritization, initial screening must provide the necessary data. Commonly, such an initial library is selected on the basis of chemical diversity by some pseudo-random process (for example, the first few plates of a larger library) or by selecting an entire smaller library. These approaches may not produce a sufficient number or diversity of actives. An alternative approach is to select an informer set of screening compounds on the basis of chemogenomic information from previous testing of compounds against a large number of targets. We compare different ways of using chemogenomic data to choose a small informer set of compounds based on previously measured bioactivity data. We develop this Informer-Based-Ranking (IBR) approach using the Published Kinase Inhibitor Sets (PKIS) as the chemogenomic data to select the informer sets. We test the informer compounds on a target that is not part of the chemogenomic data, then predict the activity of the remaining compounds based on the experimental informer data and the chemogenomic data. Through new chemical screening experiments, we demonstrate the utility of IBR strategies in a prospective test on three kinase targets not included in the PKIS.

Practical Model Selection for Prospective Virtual Screening.

Virtual (computational) high-throughput screening provides a strategy for prioritizing compounds for experimental screens, but the choice of virtual screening algorithm depends on the data set and evaluation strategy. We consider a wide range of ligand-based machine learning and docking-based approaches for virtual screening on two protein-protein interactions, PriA-SSB and RMI-FANCM, and present a strategy for choosing which algorithm is best for prospective compound prioritization. Our workflow identifies a random forest as the best algorithm for these targets over more sophisticated neural network-based models. The top 250 predictions from our selected random forest recover 37 of the 54 active compounds from a library of 22,434 new molecules assayed on PriA-SSB. We show that virtual screening methods that perform well on public data sets and synthetic benchmarks, like multi-task neural networks, may not always translate to prospective screening performance on a specific assay of interest.

Machine Learning Consensus Scoring Improves Performance Across Targets in Structure-Based Virtual Screening.

In structure-based virtual screening, compound ranking through a consensus of scores from a variety of docking programs or scoring functions, rather than ranking by scores from a single program, provides better predictive performance and reduces target performance variability. Here we compare traditional consensus scoring methods with a novel, unsupervised gradient boosting approach. We also observed increased score variation among active ligands and developed a statistical mixture model consensus score based on combining score means and variances. To evaluate performance, we used the common performance metrics ROCAUC and EF1 on 21 benchmark targets from DUD-E. Traditional consensus methods, such as taking the mean of quantile normalized docking scores, outperformed individual docking methods and are more robust to target variation. The mixture model and gradient boosting provided further improvements over the traditional consensus methods. These methods are readily applicable to new targets in academic research and overcome the potentially poor performance of using a single docking method on a new target.

Thalassosamide, a Siderophore Discovered from the Marine-Derived Bacterium Thalassospira profundimaris.

Here we describe the rapid identification and prioritization of novel active marine natural products using an improved dereplication strategy. During the course of our screening of marine natural product libraries, a new cyclic trihydroxamate compound, thalassosamide, was discovered from the α-proteobacterium Thalassospira profundimaris. Its structure was determined by 2D NMR and MS/MS experiments, and the absolute configuration of the lysine-derived units was established by Marfey's analysis, whereas that of C-9, 9', and 9″ was determined via the circular dichroism data of the [Rh2(OCOCF3)4] complex and DFT NMR calculations. Thalassosamide showed moderate in vivo efficacy against Pseudomonas aeruginosa.

A TR-FRET-based functional assay for screening activators of CARM1.

Epigenetics is an emerging field that demands selective cell-permeable chemical probes to perturb, especially in vivo, the activity of specific enzymes involved in modulating the epigenetic codes. Coactivator-associated arginine methyltransferase 1 (CARM1) is a coactivator of estrogen receptor α (ERα), the main target in human breast cancer. We previously showed that twofold overexpression of CARM1 in MCF7 breast cancer cells increased the expression of ERα-target genes involved in differentiation and reduced cell proliferation, thus leading to the hypothesis that activating CARM1 by chemical activators might be therapeutically effective in breast cancer. Selective, potent, cell-permeable CARM1 activators will be essential to test this hypothesis. Here we report the development of a cell-based, time-resolved (TR) FRET assay that uses poly(A) binding protein 1 (PABP1) methylation to monitor cellular activity of CARM1. The LanthaScreen TR-FRET assay uses MCF7 cells expressing GFP-PABP1 fusion protein through BacMam gene delivery system, methyl-PABP1 specific antibody, and terbium-labeled secondary antibody. This assay has been validated as reflecting the expression and/or activity of CARM1 and optimized for high throughput screening to identify CARM1 allosteric activators. This TR-FRET platform serves as a generic tool for functional screening of cell-permeable, chemical modulators of CARM1 for elucidation of its in vivo functions.

Functional characterization of TtnD and TtnF, unveiling new insights into tautomycetin biosynthesis.

The biosynthetic gene cluster for tautomycetin (TTN), a highly potent and selective protein phosphatase (PP) inhibitor isolated from Streptomyces griseochromogenes, has recently been cloned and sequenced. To better understand the transformations responsible for converting the post-polyketide synthase product into the exciting anticancer and immunosuppressive chemotherapeutic candidate TTN, we produced and characterized new analogues resulting from inactivation of two genes, ttnD and ttnF, in S. griseochromogenes. Inactivation of ttnD and ttnF, which encode for putative decarboxylase and dehydratase enzymes, respectively, afforded mutant strains SB13013 and SB13014. The DeltattnD mutant SB13013 accumulated four new TTN analogues, TTN D-1, TTN D-2, TTN D-3, and TTN D-4, whereas the DeltattnF mutant accumulated only one new TTN analogue, TTN F-1. The accumulation of these new TTN analogues defines the function of TtnD and TtnF and the timing of their chemistries in relation to installation of the C5 ketone moiety within TTN. Notably, all new analogues possess a structurally distinguishing carboxylic acid moiety, revealing that TtnD apparently cannot catalyze decarboxylation in the absence of TtnF. Additionally, cytotoxicity and PP inhibition assays reveal the importance of the functional groups installed by TtnDF and, consistent with earlier proposals, the C2''-C5 fragment of TTN to be a critical structural determinant behind the important and unique PP-1 selectivity displayed by TTN.

Complete reversal of epithelial to mesenchymal transition requires inhibition of both ZEB expression and the Rho pathway.

BACKGROUND: Epithelial to Mesenchymal Transition (EMT) induced by Transforming Growth Factor-beta (TGF-beta) is an important cellular event in organogenesis, cancer, and organ fibrosis. The process to reverse EMT is not well established. Our purpose is to define signaling pathways and transcription factors that maintain the TGF-beta-induced mesenchymal state. RESULTS: Inhibitors of five kinases implicated in EMT, TGF-beta Type I receptor kinase (TbetaRI), p38 mitogen-activated protein kinase (p38 MAPK), MAP kinase kinase/extracellular signal-regulated kinase activator kinase (MEK1), c-Jun NH-terminal kinase (JNK), and Rho kinase (ROCK), were evaluated for reversal of the mesenchymal state induced in renal tubular epithelial cells. Single agents did not fully reverse EMT as determined by cellular morphology and gene expression. However, exposure to the TbetaRI inhibitor SB431542, combined with the ROCK inhibitor Y27632, eliminated detectable actin stress fibers and mesenchymal gene expression while restoring epithelial E-cadherin and Kidney-specific cadherin (Ksp-cadherin) expression. A second combination, the TbetaRI inhibitor SB431542 together with the p38 MAPK inhibitor SB203580, was partially effective in reversing EMT. Furthermore, JNK inhibitor SP600125 inhibits the effectiveness of the TbetaRI inhibitor SB431542 to reverse EMT. To explore the molecular basis underlying EMT reversal, we also targeted the transcriptional repressors ZEB1 and ZEB2/SIP1. Decreasing ZEB1 and ZEB2 expression in mouse mammary gland cells with shRNAs was sufficient to up-regulate expression of epithelial proteins such as E-cadherin and to re-establish epithelial features. However, complete restoration of cortical F-actin required incubation with the ROCK inhibitor Y27632 in combination with ZEB1/2 knockdown. CONCLUSIONS: We demonstrate that reversal of EMT requires re-establishing both epithelial transcription and structural components by sustained and independent signaling through TbetaRI and ROCK. These findings indicate that combination small molecule therapy targeting multiple kinases may be necessary to reverse disease conditions.

Lactimidomycin, iso-migrastatin and related glutarimide-containing 12-membered macrolides are extremely potent inhibitors of cell migration.

Migrastatin (1), iso-migrastatin (5) and lactimidomycin (7) are all glutarimide-containing polyketides known for their unique structures and cytotoxic activities against human cancer cell lines. Migrastatin, a strong inhibitor of tumor cell migration, has been an important lead in the development of antimetastatic agents. Yet studies of the related 12-membered macrolides iso-migrastatin, lactimidomycin, and related analogues have been hampered by their limited availability. We report here the production, isolation, structural characterization, and biological activities of iso-migrastatin, lactimidomycin, and 23 related congeners. Our studies showed that, as a family, the glutarimide-containing 12-membered macrolides are extremely potent cell migration inhibitors with some members displaying activity on par or superior to that of migrastatin as exemplified by compounds 5, 7, and 9-12. On the basis of these findings, the structures and activity of this family of compounds as cell migration inhibitors are discussed.

Inhibition of transforming growth factor-beta1-induced signaling and epithelial-to-mesenchymal transition by the Smad-binding peptide aptamer Trx-SARA.

Overexpression of the inhibitory Smad, Smad7, is used frequently to implicate the Smad pathway in cellular responses to transforming growth factor beta (TGF-beta) signaling; however, Smad7 regulates several other proteins, including Cdc42, p38MAPK, and beta-catenin. We report an alternative approach for more specifically disrupting Smad-dependent signaling using a peptide aptamer, Trx-SARA, which comprises a rigid scaffold, the Escherichia coli thioredoxin A protein (Trx), displaying a constrained 56-amino acid Smad-binding motif from the Smad anchor for receptor activation (SARA) protein. Trx-SARA bound specifically to Smad2 and Smad3 and inhibited both TGF-beta-induced reporter gene expression and epithelial-to-mesenchymal transition in NMuMG murine mammary epithelial cells. In contrast to Smad7, Trx-SARA had no effect on the Smad2 or 3 phosphorylation levels induced by TGF-beta1. Trx-SARA was primarily localized to the nucleus and perturbed the normal cytoplasmic localization of Smad2 and 3 to a nuclear localization in the absence of TGF-beta1, consistent with reduced Smad nuclear export. The key mode of action of Trx-SARA was to reduce the level of Smad2 and Smad3 in complex with Smad4 after TGF-beta1 stimulation, a mechanism of action consistent with the preferential binding of SARA to monomeric Smad protein and Trx-SARA-mediated disruption of active Smad complexes.

MAGE-A, mMage-b, and MAGE-C proteins form complexes with KAP1 and suppress p53-dependent apoptosis in MAGE-positive cell lines.

The MAGE-A, MAGE-B, and MAGE-C protein families comprise the class-I MAGE/cancer testes antigens, a group of highly homologous proteins whose expression is suppressed in all normal tissues except developing sperm. Aberrant expression of class I MAGE proteins occurs in melanomas and many other malignancies, and MAGE proteins have long been recognized as tumor-specific targets; however, their functions have largely been unknown. Here, we show that suppression of class I MAGE proteins induces apoptosis in the Hs-294T, A375, and S91 MAGE-positive melanoma cell lines and that members of all three families of MAGE class I proteins form complexes with KAP1, a scaffolding protein that is known as a corepressor of p53 expression and function. In addition to inducing apoptosis, MAGE suppression decreases KAP1 complexing with p53, increases immunoreactive and acetylated p53, and activates a p53 responsive reporter gene. Suppression of class I MAGE proteins also induces apoptosis in MAGE-A-positive, p53wt/wt parental HCT 116 colon cancer cells but not in a MAGE-A-positive HCT 116 p53-/- variant, indicating that MAGE suppression of apoptosis requires p53. Finally, treatment with MAGE-specific small interfering RNA suppresses S91 melanoma growth in vivo, in syngenic DBA2 mice. Thus, class I MAGE protein expression may suppress apoptosis by suppressing p53 and may actively contribute to the development of malignancies and by promoting tumor survival. Because the expression of class I MAGE proteins is limited in normal tissues, inhibition of MAGE antigen expression or function represents a novel and specific treatment for melanoma and diverse malignancies.

The antimicrobial potential of Streptomyces from insect microbiomes.

Antimicrobial resistance is a global health crisis and few novel antimicrobials have been discovered in recent decades. Natural products, particularly from Streptomyces, are the source of most antimicrobials, yet discovery campaigns focusing on Streptomyces from the soil largely rediscover known compounds. Investigation of understudied and symbiotic sources has seen some success, yet no studies have systematically explored microbiomes for antimicrobials. Here we assess the distinct evolutionary lineages of Streptomyces from insect microbiomes as a source of new antimicrobials through large-scale isolations, bioactivity assays, genomics, metabolomics, and in vivo infection models. Insect-associated Streptomyces inhibit antimicrobial-resistant pathogens more than soil Streptomyces. Genomics and metabolomics reveal their diverse biosynthetic capabilities. Further, we describe cyphomycin, a new molecule active against multidrug resistant fungal pathogens. The evolutionary trajectories of Streptomyces from the insect microbiome influence their biosynthetic potential and ability to inhibit resistant pathogens, supporting the promise of this source in augmenting future antimicrobial discovery.

Evaluation of new migrastatin and dorrigocin congeners unveils cell migration inhibitors with dramatically improved potency.

Lactimidomycin (LTM, 1), iso-migrastatin (iso-MGS, 2) and migrastatin (MGS, 3) are macrolide antitumor antibiotics differing in macrolide ring size but all bearing a glutarimide side chain. To further develop these natural products and related analogs as drug candidates we have produced and evaluated the biological activities of a small library of iso-MGS and LTM-derived agents; congeners evaluated bear either the MGS scaffold or related acyclic (dorrigocin) scaffolds. Scratch wound-healing (SWH) assays with 4T1 mouse and MDA-MB-231 human mammary tumor cell lines, respectively, reveal structural elements crucial to inhibition of cell migration by these compounds. Moreover, two substances, 14 and 17, with activity far superior to that of MGS are unveiled by SWH assays.

Abl tyrosine kinase and its substrate Ena/VASP have functional interactions with kinesin-1.

Relatively little is known about how microtubule motors are controlled or about how the functions of different cytoskeletal systems are integrated. A yeast two-hybrid screen for proteins that bind to Drosophila Enabled (Ena), an actin polymerization factor that is negatively regulated by Abl tyrosine kinase, identified kinesin heavy chain (Khc), a member of the kinesin-1 subfamily of microtubule motors. Coimmunoprecipitation from Drosophila cytosol confirmed a physical interaction between Khc and Ena. Kinesin-1 motors can carry organelles and other macromolecular cargoes from neuronal cell bodies toward terminals in fast-axonal-transport. Ena distribution in larval axons was not affected by mutations in the Khc gene, suggesting that Ena is not itself a fast transport cargo of Drosophila kinesin-1. Genetic interaction tests showed that in a background sensitized by reduced Khc gene dosage, a reduction in Abl gene dosage caused distal paralysis and axonal swellings. A concomitant reduction in ena dosage rescued those defects. These results suggest that Ena/VASP, when not inhibited by the Abl pathway, can bind Khc and reduce its transport activity in axons.

A High-Throughput Screening Strategy to Identify Inhibitors of SSB Protein-Protein Interactions in an Academic Screening Facility.

Antibiotic-resistant bacterial infections are increasingly prevalent worldwide, and there is an urgent need for novel classes of antibiotics capable of overcoming existing resistance mechanisms. One potential antibiotic target is the bacterial single-stranded DNA binding protein (SSB), which serves as a hub for DNA repair, recombination, and replication. Eight highly conserved residues at the C-terminus of SSB use direct protein-protein interactions (PPIs) to recruit more than a dozen important genome maintenance proteins to single-stranded DNA. Mutations that disrupt PPIs with the C-terminal tail of SSB are lethal, suggesting that small-molecule inhibitors of these critical SSB PPIs could be effective antibacterial agents. As a first step toward implementing this strategy, we have developed orthogonal high-throughput screening assays to identify small-molecule inhibitors of the Klebsiella pneumonia SSB-PriA interaction. Hits were identified from an initial screen of 72,474 compounds using an AlphaScreen (AS) primary screen, and their activity was subsequently confirmed in an orthogonal fluorescence polarization (FP) assay. As an additional control, an FP assay targeted against an unrelated eukaryotic PPI was used to confirm specificity for the SSB-PriA interaction. Nine potent and selective inhibitors produced concentration-response curves with IC50 values of <40 µM, and two compounds were observed to directly bind to PriA, demonstrating the success of this screen strategy.

Biemamides A-E, Inhibitors of the TGF-ß Pathway That Block the Epithelial to Mesenchymal Transition.

Screening of a marine natural products library for inhibitors of TGF-ß revealed five pyrimidinedione derivatives, biemamides A-E (1-5). The structures were determined by 2D NMR and HRMS experiments; absolute configurations were established by advanced Marfey's analysis and ECD calculations. Biemamides A-E specifically inhibited in vitro TGF-ß induced epithelial to mesenchymal transition in NMuMG cells. Additionally, using Caenorhabditis elegans, selected biemmamides were found to influence in vivo developmental processes related to body size regulation in a dose-dependent manner.

Nox2 and Cyclosporine-Induced Renal Hypoxia.

BACKGROUND: We hypothesized that nicotinamide adenosine diphosphate oxidase 2 (Nox2) plays an important role in cyclosporine A (CsA)-induced chronic hypoxia. METHODS: We tested this hypothesis in Fisher 344 rats, C57BL/6 J wild type and Nox2-/- mice, and in liver transplant recipients with chronic CsA nephrotoxicity. We used noninvasive molecular imaging (blood oxygen level-dependent magnetic resonance imaging and dynamic contrast-enhanced magnetic resonance imaging) and molecular diagnostic tools to assess intrarenal oxygenation and perfusion, and the molecular phenotype of CsA nephrotoxicity. RESULTS: We observed that chemical and genetic inhibition of Nox2 in rats and mice resulted in the prevention of CsA-induced hypoxia independent of regional perfusion (blood oxygen level-dependent magnetic resonance imaging and dynamic contrast-enhanced magnetic resonance imaging, pimonidazole, HIF-1α). Nicotinamide adenosine diphosphate oxidase 2 knockout was also associated with decreased oxidative stress (Nox2, HIF-1α, hydrogen peroxide, hydroxynonenal), and fibrogenesis (α-smooth muscle actin, picrosirius red, trichrome, vimentin). The molecular signature of chronic CsA nephrotoxicity using transcriptomic analyses demonstrated significant changes in 40 genes involved in injury repair, metabolism, and oxidative stress in Nox2-/- mice. Immunohistochemical analyses of kidney biopsies from liver transplant recipients with chronic CsA nephrotoxicity showed significantly greater Nox2, α-smooth muscle actin and picrosirius levels compared with controls. CONCLUSIONS: These studies suggest that Nox2 is a modulator of CsA-induced hypoxia upstream of HIF-1α and define the molecular characteristics that could be used for the diagnosis and monitoring of chronic calcineurin inhibitor nephrotoxicity.

Reciprocal Regulation of ERα and ERß Stability and Activity by Diptoindonesin G.

ERß is regarded as a "tumor suppressor" in breast cancer due to its anti-proliferative effects. However, unlike ERα, ERß has not been developed as a therapeutic target in breast cancer due to loss of ERß in aggressive cancers. In a small-molecule library screen for ERß stabilizers, we identified Diptoindonesin G (Dip G), which significantly increases ERß protein stability while decreasing ERα protein levels. Dip G enhances the transcription and anti-proliferative activities of ERß, while attenuating the transcription and proliferative effects of ERα. Further investigation revealed that instead of targeting ER, Dip G targets the CHIP E3 ubiquitin ligase shared by ERα and ERß. Thus, Dip G is a dual-functional moiety that reciprocally controls ERα and ERß protein stability and activities via an indirect mechanism. The ERß stabilization effects of Dip G may enable the development of ERß-targeted therapies for human breast cancers.

High-throughput screening identifies idarubicin as a preferential inhibitor of smooth muscle versus endothelial cell proliferation.

Intimal hyperplasia is the cause of the recurrent occlusive vascular disease (restenosis). Drugs currently used to treat restenosis effectively inhibit smooth muscle cell (SMC) proliferation, but also inhibit the growth of the protective luminal endothelial cell (EC) lining, leading to thrombosis. To identify compounds that selectively inhibit SMC versus EC proliferation, we have developed a high-throughput screening (HTS) format using human cells and have employed this to screen a multiple compound collection (NIH Clinical Collection). We developed an automated, accurate proliferation assay in 96-well plates using human aortic SMCs and ECs. Using this HTS format we screened a 447-drug NIH Clinical Library. We identified 11 compounds that inhibited SMC proliferation greater than 50%, among which idarubicin exhibited a unique feature of preferentially inhibiting SMC versus EC proliferation. Concentration-response analysis revealed this differential effect most evident over an ∼10 nM-5 µM window. In vivo testing of idarubicin in a rat carotid injury model at 14 days revealed an 80% reduction of intimal hyperplasia and a 45% increase of lumen size with no significant effect on re-endothelialization. Taken together, we have established a HTS assay of human vascular cell proliferation, and identified idarubicin as a selective inhibitor of SMC versus EC proliferation both in vitro and in vivo. Screening of larger and more diverse compound libraries may lead to the discovery of next-generation therapeutics that can inhibit intima hyperplasia without impairing re-endothelialization.

Smad3 induces atrogin-1, inhibits mTOR and protein synthesis, and promotes muscle atrophy in vivo.

Myostatin, a member of the TGF superfamily, is sufficient to induce skeletal muscle atrophy. Myostatin-induced atrophy is associated with increases in E3-ligase atrogin-1 expression and protein degradation and decreases in Akt/mechanistic target of rapamycin (mTOR) signaling and protein synthesis. Myostatin signaling activates the transcription factor Smad3 (Small Mothers Against Decapentaplegic), which has been shown to be necessary for myostatin-induced atrogin-1 expression and atrophy; however, it is not known whether Smad3 is sufficient to induce these events or whether Smad3 simply plays a permissive role. Thus, the aim of this study was to address these questions with an in vivo model. To accomplish this goal, in vivo transfection of plasmid DNA was used to create transient transgenic mouse skeletal muscles, and our results show for the first time that Smad3 expression is sufficient to stimulate atrogin-1 promoter activity, inhibit Akt/mTOR signaling and protein synthesis, and induce muscle fiber atrophy. Moreover, we propose that Akt/mTOR signaling is inhibited by a Smad3-induced decrease in microRNA-29 (miR-29) expression and a subsequent increase in the translation of phosphatase and tensin homolog (PTEN) mRNA. Smad3 is also sufficient to inhibit peroxisome proliferator-activated receptor-γ coactivator-1α (PGC1α) promoter activity and to increase FoxO (Forkhead Box Protein, Subclass O)-mediated signaling and the promoter activity of plasminogen activator inhibitor 1 (PAI-1). Combined, this study provides the first evidence that Smad3 is sufficient to regulate many of the events associated with myostatin-induced atrophy and therefore suggests that Smad3 signaling may be a viable target for therapies aimed at preventing myostatin-induced muscle atrophy.