Phenocopy – A Strategy to Qualify Chemical Compounds during Hit-to-Lead and/or Lead Optimization Von der Fakultät Energie-, Verfahrens- und Biotechnik der Universität Stuttgart zur Erlangung der Würde eines Doktors der Naturwissenschaften (Dr. rer. nat.) genehmigte Abhandlung Vorgelegt von Patrick Baum aus Mutlangen Hauptberichter: Prof. Dr. Roland Kontermann Mitberichter: Prof. Dr. Klaus Pfizenmaier Tag der mündlichen Prüfung: 29.07.2010 Institut für Zellbiologie und Immunologie Universität Stuttgart 2010 3 Table of contents Table of contents .......................................................................................................... 3 Abbreviations ............................................................................................................... 6 Summary ...................................................................................................................... 8 Zusammenfassung ...................................................................................................... 10 1. Introduction ............................................................................................................ 12 1.1 Phenocopy strategy .............................................................................................. 13 1.2.1 RNA Interference (RNAi) ................................................................................. 15 1.2.2 Kinase Inhibitors ............................................................................................. 18 1.3 Drug Development Process ................................................................................... 21 1.4 Microarrays and Gene Expression Analysis ........................................................... 27 1.5 Transforming Growth Factor Beta (TGF-β) ............................................................. 32 1.6 Aim ....................................................................................................................... 36 2. Results .................................................................................................................... 37 2.1 Phenocopy Platform .............................................................................................. 38 2.2 Modulator characterization .................................................................................. 40 2.2.1 siRNA characterization .................................................................................... 40 2.2.1.1 Transfection protocol............................................................................... 40 2.2.1.2 Control siRNA off-target profiling ............................................................ 42 2.2.1.3 Control siRNAs influence expression of different cytokines and MMP1. 50 2.2.1.4 Control siRNAs influence TNF signaling ............................................... 52 2.2.1.5 TGF-βR1 siRNA characterization .............................................................. 53 2.2.2 Kinase inhibitors .............................................................................................. 58 2.3 Phenocopy Experiment ......................................................................................... 66 4 2.3.1 Data normalization ......................................................................................... 66 2.3.2 TGF-β signature ............................................................................................... 71 2.3.3 Off-target signature ........................................................................................ 77 2.3.4 Molecular Function ......................................................................................... 84 2.3.5 Pathway Analysis ............................................................................................ 86 2.3.6 Wet Lab Validation .......................................................................................... 93 2.3.6.1 Cytotoxicity and Cell Death ...................................................................... 93 2.3.6.2 Inflammation ............................................................................................ 99 2.3.7 Kinase Profiling ............................................................................................. 100 3. Discussion ............................................................................................................. 105 3.1 NCE Ranking ........................................................................................................ 106 3.2 Chemical genomic profiling ................................................................................. 111 3.3 TGF-β Biology ...................................................................................................... 113 3.4 siRNAs as modulators ......................................................................................... 117 3.4.1 Control siRNA charaterization ...................................................................... 117 3.4.2 TGF-β siRNAs ................................................................................................. 120 3.5 Conclusion .......................................................................................................... 122 4. Methods ............................................................................................................... 124 4.1 Wet laboratory experiments ............................................................................... 125 4.1.1 Cell culture, NCE treatment and siRNA transfection .................................... 125 4.1.2 RNA extraction .............................................................................................. 126 4.1.3 Quantitative real time polymerase chain reaction (qRT-PCR) ...................... 126 4.1.4 ELISA analysis of PAI-1, phospho Smad2/3, MMP1, IL8 and IL6 ................... 127 4.1.5 LDH release assay .......................................................................................... 129 4.1.6 Amplification, labeling and Beadchip hybridization of RNA samples ........... 129 4.1.7 High content screen Cellomics ...................................................................... 130 4.1.8 Caspase-3 Assay ............................................................................................ 130 4.1.9 In vitro kinase profiling ................................................................................. 131 4.2 Data Analysis ...................................................................................................... 131 5 4.2.1 Data processing ............................................................................................. 131 4.2.2 TGF-β signature (on-target signature) .......................................................... 132 4.2.3 Off-target signature ...................................................................................... 132 4.2.4 Ingenuity Pathway Analysis and Gene Set Enrichment Analysis .................. 135 Reference List ........................................................................................................... 136 Danksagung .............................................................................................................. 148 Erklärung .................................................................................................................. 151 Lebenslauf ................................................................................................................ 152 Abbreviations 6 Abbreviations °C Degrees Celcius ADME Absorption, distribution, metabolism and excretion AGC Containing PKA, PKG , PKC families ATP Adenosine triphosphate au Arbitrary units BI Boehringer Ingelheim CAMK Calcium/calmodulin-dependent protein kinase cDNA complementary/copy DNA CK1 Casein kinase 1 CMGC Containing CDK, MAPK, GSK3, CLK families conc Concentration COPD Chronic obstructive pulmonary disease Cpd Compound cRNA copy RNA CT Cycle treshold Ctrl Control d Day(s) DF DharmaFECT DMSO Dimethylsulfoxid DNA Deoxyribonucleic acid ELISA Enzym-linked immunosorbent assay EMA European Medicines Agency Ex external FCS Fetal calf serum FDA Food and Drug Administration FDR False discovery rate GEO Gene Expression Omnibus GSEA Gene Set Enrichment Analysis h Hour(s) H-t-L Hit to lead HTS High-throughput screening IC50 Half maximal inhibitory concentration IND Investigational New Drug IPF Idiopathic pulmonary fibrosis I-Smad Inhibitory Smad KEGG Encyclopedia of Genes and Genomes l Liter LO lead optimization M Molar m Meter min Minutes Abbreviations 7 miRNA micro RNA mRNA messenger RNA MTD Maximum tolerated dose NCE New chemical entity NME New molecular entity PAI-1 Plasmingon activator inhibitor 1 PAMPS Pathogen-associated molecular patterns PK Pharmacokinetic PMSF Phenylmethylsulfonylfluorid POC Percent of control Prefix m milli = 10-3 Prefix µ mikro = 10-6 Prefix n nano = 10-9 Prefix p piko = 10-12 qRT-PCR Quantitative real time polymerase chain reaction RISC RNA induced silencing complex RNA Ribonucleic acid RNAi RNA interference R-Smad Regulated Smad SARA Smad anchor for receptor activation siRNA small interfering RNA Smad Similar to MAD Smurf Smad ubiquitination regulatory factor STE Homologs of yeast Sterile 7, Sterile 11, Sterile 20 kinases TGF-β Transforming growth factor beta TGF-βR1 Transforming growth factor beta receptor 1 TK Tyrosine kinase TKL Tyrosine kinase-like TLR Toll-like receptor TR Transfection reagent UT Untreated UTP Uridine triphosphate wotgf Without TGF-β stimulation Summary 8 Summary A phenocopy is defined as an environmentally induced phenotype of one individual which is identical to the genotype-determined phenotype of another individual. In the present work, the phenocopy phenomenon has been translated to the drug dis- covery process as phenotypes produced by the treatment of cellular systems with small interfering RNAs (siRNAs) or new chemical entities (NCE) may resemble envi- ronmentally induced phenotypic modifications. Various new chemical entities exert- ing inhibition of the kinase activity of Transforming Growth Factor Beta Receptor I (TGF-βR1) were ranked by high-throughput RNA expression profiling. This chemical genomics approach was able to unravel both on-target effects (effects, caused by the inhibition of the drug target) and off-target effects (effects, caused by the inte- raction of the NCE with additional molecules). It resulted in a precise time- dependent insight into the TGF-β biology (referred to as on-target signature) and allowed furthermore a comprehensive analysis of each NCE’s off-target effects (re- ferred to as off-target signatures). Both signature types can support the drug discov- er process. The on-target signature helps to characterize the mode of action of the drug target (TGF-βR1) and thereby supports the target validation as well as the assay development process. Furthermore, the evaluation of off-target effects by the Phe- nocopy approach allows a more accurate and integrated view on the mode of action of the compounds, supplementing classical biological evaluation parameters such as potency and selectivity. The presented proof of concept study allowed the ranking of NCEs that were before indistinguishable solely based on potency and selectivity. Ac- cording to the newly introduced criteria, several of the tested NCEs revealed liabili- Summary 9 ties at e.g. the induction of off-target effects and of induction of gene regulation in- verse to the desired TGF-β inhibition effect, at the induction of cell death, at acting as pro-inflammatory stimuli and as promoting cellular growth and at induction of cancer pathways. Ultimately, this approach has therefore the potential to become a novel method for ranking compounds during various drug discovery phases. Zusammenfassung 10 Zusammenfassung Eine Phänokopie ist als ein Individuum definiert, dessen Phänotyp durch einen Um- welteinfluss mit dem eines anderen Individuums identisch ist, dessen Phänotyp durch seinen Genotyp bestimmt ist. In der vorliegenden Arbeit wurde das Phänokopiephänomen auf den Wirkstoffentwicklungsprozess übertragen. Hierbei wurden die speziellen Phänotypen in einem Zellsystem durch die Behandlung mit „new chemical entities“ (NCE), potentiellen neuen Wirkstoffkandidaten, oder mittels „small interfering RNAs“ (siRNAs) induziert. Somit können beide Molekülklassen als externe Umweltbedingung angesehen werden. Hierbei wurden diverse der potenti- ellen Wirkstoffkandidaten zur Inhibition der Kinaseaktivität von Transforming Growth Factor Beta Receptor I (TGF-βR1) mittels Hochdurchsatz-Expressions- profilierung klassifiziert. Dieser „chemical genomics“ Ansatz war in der Lage, sowohl die On-target Effekte (Effekte, welche durch die Inhibition des Wirkstofftargets aus- gelöst werden), als auch die Off-target Effekte (Effekte, welche durch die Interaktion des NCEs mit zusätzlichen Molekülen ausgelöst werden) zu identifizieren. Weiterhin ermöglichte er präzise, zeitlich aufgelöste Einblicke in die TGF-β Biologie (im Folgen- den als On-target Signatur bezeichnet) und erlaubte eine umfassende Analyse der jeweiligen Off-target Effekte der Wirkstoffkandidaten (im Folgenden als Off-target Signatur bezeichnet). Beide Signaturtypen können zur Unterstützung des Wirkstoff- entwicklungsprozesses herangezogen werden. Die On-target Signatur charakterisiert die Wirkungsweise des Wirkstoffziels (target) und unterstützt somit sowohl den „target validation“ Prozess, als auch den „assay development“ Prozess. Des Weiteren erlaubt das Ermitteln der Off-target Effekte durch den Phänokopieansatz einen prä- Zusammenfassung 11 zisen und ganzheitlichen Einblick in die Wirkungsweise der Wirkstoffkandidaten und ergänzt klassische, biologische Bewertungsparameter wie Wirksamkeit und Selektivi- tät. Die hier präsentierte Machbarkeitsstudie ermöglicht die Klassifizierung von Wirkstoffkandidaten, die zuvor auf der Basis von Wirksamkeit und Selektivität nicht zu unterscheiden waren. Entsprechend der neu eingeführten Kriterien zeigen ver- schiedene der getesteten Wirkstoffkandidaten unterschiedliche Vorbelastungen, wie z.B. das Auslösen von Off-target Effekten und von Genregulation, welche gegenläufig zu dem gewünschten TGF-β Inhibitionseffekt verlaufen. Des Weiteren konnte gezeigt werden, dass manche NCEs Zelltod induzieren, als proinflammatorische Stimuli fun- gieren, das Zellwachstum steigern könnten und Krebssignalwege induzieren. Letzt- endlich hat dieser Ansatz somit das Potential eine neue Methode zur Klassifizierung von neuen Wirkstoffkandidaten, während verschiedenster Phasen des Wirkstoffent- wicklungsprozesses, zu werden. Ph.D. Thesis Patrick Baum 1. Introduction Phenocopy – A Strategy to Qualify Chemical Compounds during Hit-to-Lead and Lead Optimization Introduction 13 1.1 Phenocopy strategy A phenocopy is defined as an environmental induced, non-heriditary phenotype of one individual which is identical to the genotype-determined phenotype of another individual. In other words, the phenocopy induced by the environmental condition mimics the phenotype produced by a gene. For example, a phenocopy is observed in Himalayan rabbits which have a white colored coat along with a black tail, nose and ears when raised in moderate temperatures. However, when raised in colder cli- mates, they develop phenotypically similar to genetically different black coated rab- bits. The Himalayan rabbits exhibit black coloration of their coats, resembling the genetically encoded black rabbits. Hence, in colder climates the Himalayan rabbit is a phenocopy of the black rabbit1. The phenocopy phenomenon can be translated and used for the drug discovery process through inhibiting a drug target with different functional modulation technologies and thereby mimicking a phenotype of interest. Inhibition can be achieved using RNA interference (RNAi), to knockdown a target, or by small molecule inhibitors (new chemical entities – NCEs), to block or inhibit the activity of the target. These modulators can be used as a particular environmental condition by treating in vitro cultured cells. Effects of the inhibition can be monitored by high-throughput RNA expression profiling and derived gene expression signatures represent either partial or exact phenocopies. Here, induced phenocopies consist of gene expression signatures caused by different pathway modulator treatments (siRNA and NCE). Subsequent analysis of the gene expression signatures will eluci- date two critical issues for drug discovery. First, getting a deeper insight into a tar- get’s biology by identifying genes whose expression are transcriptionally altered af- Introduction 14 ter interfering with the target of interest, referred to as the on-target signature. Second, single observations for each modulator used can identify genes regulated independently of the target inhibition, referred to as the off-target signature. The on-target signature is independent of the used modulator and defines the biological mode of action of the target. In contrast, the off-target signature defines the mode of action for each modulator used, which has to be not necessarily limited to the inhibition of only the target. The obtained off-target signatures can subsequently be used to qualify the NCEs of interest. This strategy was applied to NCEs and siRNAs directed against the target TGF-β receptor 1 (Figure 1). Figure 1 – Phenocopy workflow In vitro cultured HaCaT cells stimulated with TGF-β were treated with NCEs inhibiting the kinase activ- ity of TGF-βR1 or with a siRNA specific against TGF-βR1. After 2, 4 and 12 h total RNA was isolated for hybridization on Illumina Beadchips and expression profiles were generated. The concentration and time-dependent on-target (TGF-β signature) as well as the off-target signatures for every NCE were obtained by bioinformatic analysis. Compounds were qualified according to their off-target signatures by influencing other pathways. Introduction 15 1.2 Modulators – siRNAs & kinase inhibitors There are different ways to modulate a signaling pathway. In the present study two functional modulation technologies were applied to either block the enzymatic activ- ity of the receptor kinase by small molecular compounds (kinase inhibitors) or to perform a siRNA-mediated mRNA knockdown (RNAi) of the entire receptor. Pathway inhibition could also be achieved by the use of other external stimuli such as apta- mers or blocking antibodies. Unfortunately, in the present case both molecule classes were not available for TGF-β receptor 1. 1.2.1 RNA Interference (RNAi) The discovery of small interfering RNA molecules (siRNA) has been a mile stone in the field of molecular biology. RNA interference is not only used in basic research but is also efficiently applied in mammalian cells2 for target identification and validation in pharmaceutical research and even finds its way into the clinic as third generation drug3. During the canonical RNAi pathway, siRNAs target complementary mRNAs for transcript cleavage and degradation in a process known as post-transcriptional gene silencing4. Endogenous siRNAs are generated from a long double-stranded RNA via digestion by the RNase Dicer. Resulting siRNAs are ~21-23 nucleotide duplexes with symmetric 2-3 nucelotides 3’ overhangs and 5’-phosphate groups5. In contrast, syn- thetic siRNAs, initially unphosphorylated, are phosphorylated by hClp1 immediately after transfection into cells6. This phosphorylation is crucial for the subsequent in- corporation of the siRNAs into a multi-component nuclease, the RNA-inducing silenc- Introduction 16 ing complex (RISC)7. During RISC assembly the double stranded siRNAs are unwinded and only one of the strands stably associates with the complex. Rules that govern selectivity of strand loading into RISC are based on differential thermodynamic sta- bilities of the ends of the siRNAs8, 9. The less thermodynamically stable end is favored for binding to RISC. The incorporated strand, called guide strand, directs target rec- ognition by perfect or near-perfect Watson-Crick base pairing10 between the incor- porated siRNA and the mRNA transcript11 whereas the other strand, the passenger strand, of the RNA duplex is discarded. As part of RISC the endonuclease Ago-2 is responsible for the cleavage of the mRNA. The cleavage activity is very precise: the phosphodiester linkage between the nucleotides that are base-paired to siRNA resi- dues 10 and 11 (relative to the 5’ end of the siRNA) is cut to generate 5’- monophosphate and 3’-hydroxyl termini12 leading to subsequent degradation through cellular exonucleases. Upon activation, RISC can undergo multiple rounds of cleavage resulting in a robust gene silencing. However, the specificity of siRNA molecules has been a debate since the first de- scription of RNAi as a technique for the down-regulation of a gene product. In the meantime, many studies on siRNA selectivity provided insight into the origin of siRNA regulations besides the intended reduction of the target sequence. These so called off-target effects are based on two distinct mechanisms: the siRNA sequence dependent and the double stranded RNA related off-target effects. The problem of the sequence dependent off-target effects is explained by either the binding of the passenger strand of the siRNA molecule to the inverted complementary sequence of an off-target transcript or by the low stringency binding of imperfect matches be- tween the guide strand and the off-target transcript. Both events lead to undesired Introduction 17 transcript degradation after incorporation of the respective siRNA derived single stranded RNA into Ago proteins. The double stranded RNA related off-target effects can be summarized as an innate immune system response of the cell recognizing the presence of pathogen-associated molecular patterns (PAMPS) by host pattern rec- ognition receptors13. The recognition is accomplished by cytoplasmatic, double stranded RNA binding proteins like PKR14 leading to the silencing of translation or by RIG-1 and MDA-515, which leads to the activation of interferon regulator factor (IRF) or NFκB signaling. An additional group of membrane bound receptors also recogniz- es siRNA molecules either as double stranded RNA like TLR3 or by the presence of a ribose-backbone in close proximity to multiple uridine residues like TLR716 or more specifically via GU-rich motifs like TLR7 and TLR817, 18. The recognition of PAMPS via these Toll-like receptors again activates the NFκB and interferon signaling cascades19 and markers of this activation have been described20. Several strategies were devel- oped to overcome these off-target effects18 by modifying the ribose backbone of the siRNA molecules. The chemical modifications avoid nuclease degradation, the incor- poration of the passenger strand into RISC as well as the binding and activation of the PAMPS receptors. The strategies of chemical modifications include 2’-O-methyl modification of single21 or multiple sequences positions22, 2’-F-modifications23 or even non-ribose backbones like locked nucleic acids24 or arabinose25 moieties. Introduction 18 1.2.2 Kinase Inhibitors A new chemical entity (NCE), also referred to as new molecular entity (NME), is a novel pharmaceutical agent that does not contain an active chemical moiety pre- viously approved by the United States Food and Drug Administration (FDA). This moiety is further defined as a molecule or ion, responsible for the physiological or pharmacological action of the drug substance. Thereby those appended portions of the molecule are excluded that cause the drug to be an ester, salt or other noncova- lent derivative26. The identification and characterization of these new molecular compounds represents one of the most important areas of pharmaceutical research in the pursuit of the next generation of therapeutic agents. Biological systems con- tain only four types of macromolecules that can be interfered using such a small chemical molecules: proteins, polysaccharides, lipids, and nucleic acids. However, the vast majority of successful drugs achieve their activity by binding to and modify- ing the activity of a protein. According to different estimations mainly based on the identification of druggable protein domains in the human genome there are be- tween 2,000 and 3,000 gene products that can be treated by small chemical mole- cules27, 28. The conservative estimation of Russ et al. assumes that with approximate- ly 520, protein kinases represent with 22 % the biggest group of potential drug tar- gets (Figure 2). Collectively, all these kinases, referred to as the human kinome, phosphorylate approximately 500,000 sites in the human proteome and are there- fore involved in the regulation of virtually every cellular process29. Introduction 19 Figure 2 – Druggable genome Conservative estimation of the druggable genome by Russ et al.28 Abbreviation: GPCR: G-protein coupled receptor; NHR: nuclear hormon receptor; Cyp: cytochrome P450 Thus, kinases have become one of the most intensively investigated drug target classes with at least 30 targets being developed to a level of clinical phase I trial in- cluding 11 inhibitors approved for clinical use so far30. Although kinase inhibition is predominantly discussed for the treatment of cancer, various other diseases, includ- ing immunological-, neurological-, vascular - and metabolic disorders, are linked to deregulation of kinase function30, 31. Kinase activity can be blocked by small chemical molecules in four different ways30: i) Type I inhibitors. Kinases can adopt both active and inactive conformations. These molecules bind to both conformations in an ATP- mimetic manner to the ATP binding pocket of the kinase and thus prevent activity by blocking ATP binding (Figure 3). ii) Type II inhibitors. Kinase activity is blocked by preventing ATP binding and also by stabilizing the inactive conformation. This is achieved by the occupation of the ATP binding site and an adjacent hydophobic pocket involved in the conformational change. iii) Covalent inhibitors. Comparable to 22% 13% 10% 10% 5%2% 3% 35% Protein kinase GPCR Ion channels Proteases Transporters NHR Cyp enzyme Other Introduction 20 Type I inhibitors, they occupy the ATP binding pocket. However, once they are bound they form a covalent bond with an active site residue, usually a cysteine. iv) Alloster- ic inhibitors. These molecules are non-ATP-competitive and bind to a location in which binding modulates the catalytic activity of the kinase. Collectively, due to the mode of action of these molecules and the partial highly pre- served tertiary structure of kinases, it is often challenging to achieve a selective inhi- bition. Therefore, off-target effects of these inhibitors must be thoroughly investi- gated. Standard procedures to annotate kinase selectivity are in vitro enzymatic or binding assay counter screens against a panel of recombinant kinases and are availa- ble from a number of service providers. Discovering the full range of targets is also important for an alternative strategy for the use of kinase inhibitors that block the activity of more than one kinase to achieve a synergetic treatment effect. Such poly- pharmacolgy approaches seem to be a promising strategy for the treatment of dis- orders that tend to result from multiple molecular abnormalities such as cancer. So far, different of these promiscuous drugs have been approved. Among others, the most prominent examples are the anti-cancer drug Imatinib (Gleevec) and the schi- zophrenia drug Clozaril32. Introduction 21 Figure 3 – Type I inhibitor X-Ray structure of the Type I TGFβ-R1 kinase inhibitor BI2 used in the present study. Typically for its chemotype, the indolinones, BI2 binds into the ATP pocket of the kinase and displays the canonical hydrogen bonds between the lactam moiety and Asp281 and His283 of the kinase hinge region. The 6-amido substituent on the Indolinone core points towards the TGFβ-R1 specificity pocket flanked by Phe262 and Lys23233. 1.3 Drug Development Process The initial state of the drug development process is the identification of a potential therapeutic drug target whose modulation might inhibit or reverse disease progres- sion. A wide variety of different approaches for target identification is used in both academic and pharmaceutical research, including high-throughput technologies like the profiling of changes on transcriptional34 (genomics) or protein35 (proteomics) levels or the use of (genome-wide) RNAi screens36. Furthermore, reverse pharmacol- Introduction 22 ogy approaches are used as a more holistic approach. Here, phenotypes are screened after the treatment of an in vitro system with a compound library, desired phenotypes are selected before the hit drug target is identified37. An alternative strategy that arrived in the “omics era” is data mining where mostly literature or microarray databases are screened for potential hits38. Subsequent to identification, target validation must demonstrate that the target is truly involved in the progress of the disease and that its modulation can result in a therapeutic effect. To unravel this therapeutic potential, extensive in vitro studies in a relevant cellular system as well as in vivo experiments using e.g. genetically mod- ified organisms such as knock-out animal and/or disease related animal models need to be performed. Here, the biological contribution is characterized through activa- tion, inhibition, overexpression or silencing of the target. A robust assay must be developed that can be used in high-throughput screening (HTS). Here, compound libraries are screened to identify substances that antagonize, agonize or modulate the drug target. These compounds are then classified as hits. The hits are further validated in counter screens before the best candidates will result in lead series (hit- to-lead phase) and finally enter the lead optimization (LO) process. Here, the struc- tures of a lead class are chemically modified to first identify a suitable chemotype as lead class that is then further optimized in regards of potency, selectivity and phar- macokinetics of the molecules39. However, very little information may be available that will give insights into the entire mode of action of the different hits that includes triggering of additional unwanted effects like toxicity. Selecting the best lead struc- tures or compounds for optimization that are further promoted to pre-development candidates remains a challenging task. Introduction 23 For an optimal use of any drug it is pivotal to know its fate after administration to the body. Therefore, the pharmacokinetic (PK) and metabolism of the candidates must be determined. Four parameters (absorption, distribution, metabolism and excretion, referred to as ADME scheme) are mainly analyzed to identify what the organism does with the drug40: i) Absorption, which is defined as the movement from the site of administration to the site of measurement (e.g. the bloodstream) in the body. ii) The distribution describes the process during which a drug moves from one site within the body to another. Drugs can either remain in the vascular system, get equally distributed in the body or are concentrated to specific organs or tissues mostly dependent on the molecular weight or solubility41. iii) Metabolism of the drug can be seen as a detoxification function of the body to xenobiotic molecules. Ideally, a drug is eliminated once its effect is no longer required. Although some drugs can be eliminated without structural changes, the vast majority must be metabolized to water soluble components to guarantee their excretion in bile or urine. A first group of enzymes (e.g. cytochrome P450 oxidases) carries out oxidation, reduction or hy- drolytic reactions of the drug molecule (phase I reaction) before a second group (e.g. UDP-glucuronosyltransferases or glutathione S-transferases) introduces hydrophilic residues (phase II reaction)42. iv) Excretion is defined as the process whereby drugs or their metabolites are removed from the body. Hereby the kidney and the liver are major secretory organs. Excretion through salvia, sweat, tears, breast milk, hair and nails is also described40 but only with minor contribution. Different experiments are performed to study ADME profiles, including absorption analysis on cells or artificial membranes43 or protein binding of compounds44, their stability in plasma or serum and their metabolism by e.g. screening their specific cytochrome P450 profile using Introduction 24 recombinant enzymes45. Furthermore, LC-MS/MS based bio-analytical methods are applied to identify and quantify the compounds or their metabolites in blood, urine and other biological fluids or tissues46. Besides potency the drug candidates must also fulfill safety criteria. The therapeutic index compares both parameters: the dose necessary to produce a toxic effect with the dose sufficient to achieve efficacy. It is therefore a numerical measurement for both beneficial effect and relative safety of the drug candidate. The index is calcu- lated upon preclinical in vitro and in vivo safety evaluation. While standard ap- proaches to test in vitro cytotoxicity analyzes cell density, -integrity or -health para- meters (e.g. ATP conc., lysosmal mass, mitochondrial membrane potential, nuclear fragmentation)47, 48, in vivo toxicology involves studies in both rodent and non- rodent species and investigates toxic effects using well-proven markers such as his- topathology, physiology- and blood-chemistry parameters49. Additionally, toxicoge- nomics approaches are performed where RNA profiles of specific organs (mostly liver, kidney and heart) are analyzed after compound application, to mostly rodents, to screen for the induction of known toxicity related genes50. After positive non clinical safety assessment, the drug candidates can enter clinical trial as Investigational New Drug (IND) after acceptance of the regulatory authorities (FDA for US or EMA for Europe). During the clinical trials, the efficacy and safe- ty/tolerability of a drug candidate is investigated in humans. The clinical trails are classified into up to five phases. In 2006, the FDA introduced the Exploratory Investi- gational New Drug guidance26 that allows a so called phase 0 trial for exploratory first-in-man analyses. In this phase, also known as microdosing study, a subtherapeu- tic dose of the drug is administered to a small group of healthy subjects (10-15) to Introduction 25 investigate its pharmacodynamics and -kinetics. This allows to perform preliminary proof-of-concept studies at a very early stage and helps to reduce developmental costs51. In the next stage, phase I trails, the drug is tested in a small group of healthy subjects (20-80) to investigate its safety, and in depth pharmacodynamics and - kinetics. Phase I trials addresses dose escalation studies. Therefore a range of differ- ent drug concentration is administered to indentify the maximum tolerated dose (MTD) and the suitable dose for future therapy. Subsequent phase II trails continue to test drug safety in a larger group of subjects (50-300) both healthy and diseased volunteers. Furthermore, the dose finding process must be completed and ultimately the effectiveness in disease treatment (proof-of-concept) is tested. After accom- plishment of phase II goals the drug can be studied in phase III trial on a large group of patients (200-10,000). This so called pivotal trial needs to demonstrate the signifi- cant therapeutic effect of the drug compared to placebo treatment and ideally supe- rior characteristics compared to current “gold standard” treatment of the disease52. Phase III trails are the most expensive and time consuming studies within the clinical trails and can last for up to several years, since they are often performed with large amounts of cohorts at different sites. After successful phase III trail, an application with all necessary information about the drug out of the data of preclinical and clini- cal studies can be submitted to the regulatory authorities (FDA and/or EMA) to achieve an approval to the market. Even after successful approval, the drug can be further investigate in a phase IV trail either to study rare adverse effects only detect- able in a large collective of patients or for marketing reasons to find superior fea- tures of the treatment compared to competitor drug treatment. Introduction 26 However, the critical issue in drug development is the high attrition rate. The aver- age success rate of a drug development program is approximately 11 %53. However, the success rate differs strongly dependent on the therapeutic indication. While suc- cess rate for drugs treating cardiovascular disease is reasonably high with 20 %, only 5-15 % of all cancer programs succeed53, 54. Additionally, the amount of FDA approv- als decreased from 53 new drugs in 1996 to only 19 in 200955. Possible explanations are that recently diseases of great complexity are attacked but also that the entry bar for new drugs is higher than in the past because they are competing with en- hanced standards of care. Furthermore, the demands of the regulatory authorities have increased over the past years53. Responsible for most failures is the lack of effi- cacy of the treatment in man. Other drugs induce toxic adverse effects, fail due to chemistry and formulation issues, and have poor pharmacokinetic characteristics or their manufacturing process fails. Furthermore, a lack of clinical biomarkers prevents a distinct stratification of the patients or poor design of the clinical trials hampers identification of significant effects56. Ultimately, there is a need for new strategies in all stages of drug development that optimizes the processes and supports the dis- covery of high quality drugs to cope with the challenges in a changing pharmaceuti- cal industry. A summary of all stages of the drug development process is depicted in Figure 4. Introduction 27 Figure 4 – Drug development process Drug development process can be divided in seven phases. It starts with the identification of the drug target whose modulation might inhibit or reverse disease progression. The target validation has to show that the target is causative of the disease symptoms and that modulation of the target ameli- orates these symptoms. The assay development and high-throughput screening process have to find lead structures for the target modulation that are further improved during the lead optimization phase. Pharmacokinetic and metabolism investigate the fate of the drug after administration to an organism and animal safety assessment screens for toxic effects upon treatment. After successful assessment of all these phases the drug candidate can be investigated in clinical trials. 1.4 Microarrays and Gene Expression Analysis Investigation and understanding of complex functional mechanism of a living organ- ism requires a global and parallel analysis of different cellular processes. There are many possibilities to survey these cellular processes including mRNA levels, protein expression, epigenetic modifications or metabolite profiles57. However, state-of-the- art method is the use of microarrays for the detection of transcriptional changes. The analysis of gene expression with microarrays is a well-established procedure58 and various different technical approaches were developed. The most frequently used arrays were developed by Affymetrix59 and Illumina60. Affymetrix arrays are composed of 11 probes per gene that are 25 base pairs long. The probes are directly synthesized on the surface of the chip at a defined position by photolithography. In Target Identification Target Validation Assay Development & HTS Hit-to-Lead & Lead Optimization Drug PK & Metabolism Animal Safety Assessment Clinical Trials Introduction 28 contrast, Illumina arrays have 1-4 probes per gene that are 50 base pairs long. These probes are linked to beads that are then randomly arranged on the surface of the chip. Their location is subsequently decoded by a sequential hybridization proce- dure. Over the past years microarray experiments have resulted in a big variety of gene expression compendia generated for different purposes. The first large-scale gene expression compendium was generated by Hughes et al.61 from 276 yeast deletion mutants, 11 conditional alleles and wild-type cells treated with 13 different drugs. Other efforts were made to generate a Global Cancer Map with samples of 218 hu- man tumors out of 14 histological classes and 90 normal human tissues62. Further- more, a large number of normal human and mouse tissue samples were combined to the Gene Expression Atlas63. Additionally, several commercial databases such as Gene Logic’s BioExpress have arisen, containing thousands of expression profiles from normal, diseased or drug-treated tissues or cell lines from different species. Ultimately, in a broader sense, ArrayExpress64 and Gene Expression Omnibus (GEO)65 can also be considered as huge compendia or repositories containing the stored data of more than 20,000 microarray experiments comprising more than 500,000 sam- ples. However, this almost vast quantity of data can be both boon and bane and the sta- tistical analysis and most notably the interpretation of this data are great challenges. First, normalization of the data has to be performed to minimize systematic effects that are not constant between different samples of an experiment and that are not due to the factors under investigation (e.g. treatment, time). Optimal selection of a normalization method heavily depends on the nature of the experiment. In this re- Introduction 29 gard factors like comparability and quality of single runs play a major role. It has been shown that the normalization method used may influence further downstream analysis to a great extend66, and thus, has to be carefully chosen based on the actual data. Second, the selection criteria for differential expression of a gene have to be deter- mined. Prior to selection an appropriate test statistic must be used to calculate p- values between the sample replicates. Commonly applied statistics are t-test, SAM- test or the U-Mann-Whitney-test67. Afterwards additional statistical tests, using Bon- ferroni or Benjamini Hochberg procedures, are often applied to correct the p-values to reduce the amount of false positives68. Then fold changes between the samples of interest must be determined. Finally, p-value and fold change cut-offs need to be determined, dependent on the conducted experiment and the biological system, to select differentially expressed genes. In order to avoid arbitrary cut-offs, other ap- proaches use related samples to detect e.g. time or dose responses of a regulated gene69, 70. Ultimately, the critical and most challenging step in every expression analysis is to find a meaning in the orchestra of regulated genes. Each expression experiment re- gardless of which technology was used, normally results in a list of genes that are differentially expressed in one sample compared to another. Historically, each of the differentially expressed genes on this list was observed independently, relevant genes were assigned and a biological sense or meaning was deduced from this list, often without the use of prior knowledge. This time-consuming and partial low effi- cient approach can be bypassed through an incorporation of prior knowledge that helps to extract gene groups or gene expression signatures from the list of differen- Introduction 30 tially expressed genes. The use of gene expression signatures originally derived from early work on cancer classification from Golub et al.71. They profiled 38 bone- marrow samples, 27 from patients with acute lymphoblastic leukemia (ALL) and 11 from patients with acute myeloid leukemia (AML) and were able to predict the class of 34 other previously unclassified samples using a set of 50 genes known to be strongly correlated with either ALL or AML. Nowadays, a priori knowledge can be accessed through the use of various public or commercial databases (e.g. Biocarta72, Ingenuity Pathway Analysis73, Kyoto Encyclopedia of Genes and Genomes74, Wiki- Pathways75, Gene Ontology, Reactome76) developed by manual curation from litera- ture, based on expert knowledge or assigned by computational annotation (Gene Ontology). Different algorithms can overlay the differentially expressed genes with biological pathways, networks or processes and are able to identify affected sets of genes with a sometimes small but coordinate change in expression. Additionally, the so called Gene Set Enrichment Analysis (GSEA)77 introduces a rank based scoring me- tric that the magnitude of differential expression can also be included to the analysis. Introduction 31 Figure 5 – Microarray manufacturing process a) Affymetrix microarrays consist of 25mer probes that are chemically synthesized at specific locations on the surface of the silicon wafer. Synthesis is done in parallel, in a series of repetitive steps. Each step appends a particular nucleotide to selected regions of the chip. Selection occurs by exposure to UV light with the help of photolithographic masks. b) Illumina BeadChips are manufactured in four major steps. First, previously synthesized oligonucleotides, containing address and probe sequences, are linked to beads. Probe sequence is gene-specific for the binding of the transcribed and labeled cRNA. Always one species of oligonucleotide is covalently coupled to one bead, resulting in one bead type per surveyed gene. Second, the silicon wafer is coated with a micro-structured resist layer, then plasma etching is performed and subsequently the resist layer is removed. Third, the silicon wafer is flooded with the beads that self-assemble to a randomized microarray. Final, sequential decoding process78, using decoder oligonucleotides complementary to the address sequence of the beads, is performed to map the probe position. a b Introduction 32 1.5 Transforming Growth Factor Beta (TGF-β) TGF-β is a multifunctional cytokine with effects on cell growth, migration, adhesion, differentiation, apoptosis and epithel-to-mesenchymal transition79. The canonical signaling pathway, as described by Shi and Massagué79, consists of TGF-β receptor 1 and 2 (TGF-βR1 & R2), receptor serine/threonine protein kinases, and the family of Smad proteins. The ligands TGF-β1, 2 and 3 but also Activin and Nodal initiate signal- ing through binding and subsequent formation of a receptor complex of TGF-βR1 and TGF-βR2. This allows TGF-βR2 phosphorylation of TGF-βR1, which then propa- gates the signal through phosphorylation of Smad2 and Smad3. Once phosphory- lated, these regulated Smads (R-Smads) form heteromeric complexes with the Co- Smad, Smad4. This activated Smad complex enters the nucleus and in conjunction with other cofactors, such as CBP, TGIF and HDAC, interacts with the DNA and regu- lates gene expression. TGF-β signaling is regulated at several levels. First, the access of the R-Smads is controlled by the Smad anchor for receptor activation, SARA. Second, the E3 ubiquitin ligases, Smad ubiquitination regulatory factor-1 & 2 (Smurf- 1 and 2), control the turnover of the TGF-β receptor and thereby the activation of Smads. Third, the inhibitory Smad (I-Smad), Smad7, acts antagonistically and inhibits receptor mediated activation of R-Smads. Finally, another level of control of the Smad pathway is via the regulation of nuclear accumulation of Smads, by the Ras- Extracellular signal kinase (ERK) pathway. There are also non-Smad mediated signal- ing events80. For instance, TGF-βR1 can mediate JNK signaling by interacting with E3 ubiquitin ligase TRAF6 and subsequent activation of TAK1, MKK4 and 7 to trigger JNK-dependent apoptosis81. Introduction 33 Figure 6 – Canonical TGF-β signaling pathway Pathway illustration was taken from the Ingenuity Pathway Analysis (www.ingenuity.com) library of canonical signaling pathways. Introduction 34 Malfunctions within the TGF-β signaling pathway may result in cancer, fibrosis and diverse hereditary disorders82-84. Therapeutic approaches to inhibit its signaling by targeting TGF-βR1 with antibodies, antisense molecules or kinase inhibitors85 are widely discussed for the treatment of idiopathic pulmonary fibrosis (IPF) and solid tumor growth86, 87. Lung diseases like asthma, chronic obstructive pulmonary disease (COPD) or idiopathic pulmonary fibrosis (IPF) are associated with an abnormal in- flammatory response in combination with airway remodeling through fibrosis, goblet cell hyperplasia and smooth muscle thickening86, 88. Among other growth factors and cytokines, TGF-β is highly expressed in fibrotic tissues and up-regulates the expres- sion of adhesion molecules required for the recruitment of monocytes and neutro- phils which both initiate inflammatory responses. Furthermore, TGF-β plays a pivotal role in the biosynthesis and turnover of extracellular matrix (ECM) proteins like col- lagens, fibronectin and proteoglycans and is thus contributing to fibrosis and stimu- lates smooth muscle cell proliferation89. During cancerogenesis the function of TGF-β is Janus-faced. On the one hand, TGF-β acts as tumor suppressor by inhibiting the proliferation of normal epithelial, endothelial haematopoietic cells and early epi- thelial cancer cells. On the other hand, once tumorigenesis has been initiated tumor cells escape this growth control and produce high levels of TGF-β resulting in pro- found changes of the tumor’s microenvironment in which TGF-β promotes tumor growth. The tumor-promoting effects include extracellular matrix degradation and epithelial mesenchymal transition by increased production of platelet-derived growth factor (PDGF), connective tissue growth factor (CTGF) and matrix metallopro- teinases (MMP). Furthermore, angiogenesis is triggered by an up-regulation of vas- cular endothelial growth factor (VEGF)87, 90. Clearly, inhibition of TGF-βR1 holds Introduction 35 promise as a new modality for the treatment of fibrotic diseases and cancer. Several small molecules targeting TGF-βR1 have been reported in literature85 among them SB-431542, LY-2109761, SD-208, SM-16 and others. Recently, LY-2157299, a deriva- tive of LY-2109761, was advanced into phase I clinical trials for the treatment of can- cer91. Most known TGF-βR1 inhibitors are based on five-membered heterocyclic chemotypes (imidazoles, pyrazoles or thiazoles) occupying similar positions in the ATP pocket of the TGF-βR1 kinase domain. Figure 7 – TGF-β receptor 1 kinase inhibitors Introduction 36 1.6 Aim The goal of the present work was to establish a procedure to support early phases in the drug development process. Therefore, the phenocopy phenomenon should be applied to an in vitro system by inhibiting TGF-β signaling with siRNAs and a novel structural lead class of TGF-βR1 kinase inhibitors and subsequent high-throughput expression profiling. This chemical genomics approach can then be used to elucidate the mode of action of TGF-βR1 as well as for the identification of the inhibitors’ off- targets to allow a more accurate and integrated view on the optimized compounds, supplementing classical biological evaluation parameters such as potency and selec- tivity. The long term goal of such an early compound assessment is to reduce the attrition rate of advance clinical programs by an early discovery of liabilities of the analyzed NCEs. Ph.D. Thesis Patrick Baum 2. Results Phenocopy – A Strategy to Qualify Chemical Compounds during Hit-to-Lead and Lead Optimization Results 38 2.1 Phenocopy Platform To perform the Phenocopy approach HaCaT cells (human keratinocytes) were cul- tured and treated to analyze TGF-βR1 modulators. First, it was important to demon- strate that the TGF-β signaling of the cells is functional and reproducible. Therefore, three readouts were carried out to cover the entire TGF-β signaling process that represent early, intermediate and late responses upon TGF-β stimulation. Direct downstream targets of the activated TGF-βR1 kinase are Smad2 and Smad3 proteins. Their phosphorylation initiates the intracellular signaling cascade. Smad phosphory- lation was investigated by ELISA using an antibody specific for phosphorylated Smad2/3. This assay showed a significant increase of Smad2/3 phosphorylation al- ready 15 minutes after stimulation with TGF-β. Phosphorylation was further en- hanced after 30 and 60 minutes and remained stable for further 60 minutes (Figure 8a). A well characterized downstream target of TGF-β signaling is PAI-1 (SERPINE-1)92. The expression of PAI-1 at mRNA levels (qRT-PCR) and at protein le- vels (ELISA) was measured as surrogate marker for intermediate and late responses of TGF-β signal transduction. PAI-1 mRNA expression was TGF-β concentration- and time-dependent. An up to 70-fold up-regulation was detected 6 h after TGF-β stimu- lation. This expression decreased at later time points (Figure 8b). Subsequently, the supernatants were analyzed for PAI-1 protein expression. The expression of PAI-1 protein was delayed compared to the mRNA expression and can therefore be consi- dered as a late response to TGF-β stimulation. The first significant increase was seen 6 h post stimulation and PAI-1 further accumulated at later time points (Figure 8c). Results 39 After establishment and characterization of the cellular platform, it was possible to start the evaluation of the different pathway modulators. Figure 8 - Phenocopy platform read outs Three readouts representing early (Smad2/3 phosphorylation), intermediate (PAI-1 mRNA) and late (PAI-1 protein) responses to TGF-β stimulation were performed. a: phospho-Smad2/3 ELISA. This assay showed a significant increase of Smad2/3 phosphorylation 15 minutes after stimulation with TGF-β. Phosphorylation was further enhanced after 30 and 60 minutes and remained stable for fur- ther 60 minutes. b: PAI-1 mRNA. Elevated PAI-1 expression was demonstrated by qRT-PCR after TGF-β stimulation in a concentration- and time-dependent manner. c: PAI-protein. The supernatants were analyzed with a PAI-1 ELISA for protein expression. The first significant increase was observed 6 h post stimulation. Subsequently, PAI-1 further accumulated in a concentration-dependent manner. All re- sults are representative of three independent experiments. Student t-test was used to calculate the significance compared to unstimulated cells (*< 0.01 & **<0.001). All error bars indicate the standard deviation of n=3. Results 40 2.2 Modulator characterization Two different functional modulation technologies were used to inhibit TGF-β signal- ing in HaCaT cells: RNAi and enzymatic kinase inhibition. siRNAs and seven NCEs (Table 8) were used to monitor and characterize mRNA transcriptional changes upon knockdown of TGF-βR1 mRNA or inhibition of TGF-βR1 kinase activity. 2.2.1 siRNA characterization In the present study, siRNAs were used as alternative modulation technology and therefore as reference for the NCE treatment. However, the method as well as the treatment with siRNAs had to be adjusted to the cellular system of the Phenocopy platform. First, the protocol for the transfection of HaCaT cells was established, next an appropriate control molecule was identified and finally TGF-βR1 siRNAs were cha- racterized and selected. 2.2.1.1 Transfection protocol First, a protocol had to be chosen, suitable for an optimal transfection of HaCaT cells. Therefore, a well-characterized siRNA directed against GAPDH93 was transfected using either lipofection or electroporation. Subsequently, knockdown efficacy was determined by qRT-PCR as surrogate for transfection efficacy. For lipofection four different transfection reagents from Dharmacon (DF1-DF4) were tested. Electropora- tion was carried out by Amaxa Nucleofector®. Lipofection using DF1 resulted in a knockdown of more than 95 %, but also the use of DF2 and DF4 led to knockdowns Results 41 of approximately 90 % and comparable good results were also achieved by electro- poration (Figure 9a). However, the lipofection protocol using DF1 was better tole- rated by the HaCaT cells. LDH release to the supernatants of the cells was measure to analyze membrane integrity as measure for cell death after transfection. DF1 transfection resulted in less than 5 % cell death compared to the electroporation method with over 15 % dead cells (Figure 9b). Thus, HaCaT cells were transfected using DF1 in all further experiments due to slightly better transfection efficacy and less cell death compared to the other tested reagents or protocols. Figure 9 – Transfection method a: HaCaT cells were transfected with a GAPDH-specific siRNA using either lipofection (DF1-DF4) or electroporation (pulse). Among the different lipofection reagents, highest knockdown was achieved by DF1 (> 95 %). Electroporation resulted in a knockdown of approximately 90 %. b: cell death caused by the transfection was analyzed through LDH activity in the supernatant of the cells. Fewer cells died after DF1 transfection (< 5 %) compared to the eletropration method (> 15 %). Student t-test was used to calculate the significance (*< 0.01). All error bars indicate the standard deviation of n=2. Results 42 2.2.1.2 Control siRNA off-target profiling In general, control siRNA molecules, unable to mediate mRNA knockdown, are used for the normalization of specific siRNA effects. These molecules either consist of se- quences that are not complimentary to any mRNA molecule (non-targeting siRNAs) or are not able to enter RISC (RISC-free siRNAs). In the most optimal situation, the use of these molecules should not introduce any gene regulation or phenotype. In order to find a control siRNA molecule suitable for a wide range of experiments the analysis was not only focused on the Phenocopy cell system. Therefore, the specifici- ty of 13 control siRNA molecules from different vendors (Table 1) was compared by use of microarray expression profiling in two human cell lines HaCaT (keratinocyte phenotype) and HT1080 (fibroblast phenotype), as well as in the murine fibroblast line 3T3-L1. The control siRNA molecules were transfected using optimized condi- tions (high transfection efficacy without increased LDH release) in five independent experiments. All used control molecules are non-targeting siRNAs. The only excep- tion is the siRNA D6ctrl representing the RISC-free control siRNA type. After 48 h the transcriptional changes introduced by these molecules were identified performing Illumina BeadChip based expression profiling. The profile of each treatment was compared with the respective untreated cell cultures. The differences in gene ex- pression of HT1080 cultures are visualized as volcano plots in Figure 10. In the volca- no plots, every point represents a single transcript. The x-axis shows the log2 ratio (LR), representing the fold change, between transfected and untransfected cells. The y-axis is scaled as negative log10 [p-value] as an indicator of significance. P-values were FDR-corrected according to Benjamini-Hochberg94. Since the analysis is based Results 43 on five replicates, it was possible to apply relative low fold change cut-offs and therefore regard all genes which revealed an expression difference of factor of 1.5 or higher with a FDR-corrected p-value of ≤ 0.01 as significantly regulated. Table 1 – List of characterized control siRNA molecules Vendor Description Cat. No. Abbr. Ambion Silencer® Select Negative Control #1 siRNA 4390844 A1ctrl Ambion Silencer® Select Negative Control #2 siRNA 4390847 A2ctrl Dharmacon On-TARGET plus siControl Non-targeting Pool D-001810-10-20 D1ctrl Dharmacon On-TARGET plus siControl Non-targeting siRNA D-001810-01-20 D2ctrl Dharmacon On-TARGET plus siControl Non-targeting siRNA #2 D-001810-02-20 D3ctrl Dharmacon On-TARGET plus siControl Non-targeting siRNA #3 D-001810-03-20 D4ctrl Dharmacon On-TARGET plus siControl Non-targeting siRNA #4 D-001810-04-20 D5ctrl Dharmacon siControl RISC-free siRNA #1 D-001220-01-20 D6ctrl Qiagen Control siRNA_1248 customized Q1ctrl Qiagen Random_2_siRNA customized Q2ctrl Qiagen Control siRNA_2904 customized Q3ctrl Qiagen Random_1_siRNA customized Q4ctrl Qiagen Allstar Negative Control siRNA 1027281 Q5ctrl Results 44 Figure 10 – Control siRNA off-targets in HT1080 cells Shown are Log2 Ratios (plotted on x-axes) and p-values (plotted on y-axes as negative Log10 p-value) of control siRNA transfected versus untreated cells are shown. All genes with a fold change ≥ 1.5 and a p-value ≤ 0.01 were considered as off-target effects of the siRNA. Results represent five indepen- dent experiments. The p-values were FDR-corrected according to Benjamini-Hochberg. Results 45 As quantitative specificity criteria for the control siRNA molecules the deregulated genes of each siRNA are summarized for all three cell lines (Table 2). Considering the number of differentially expressed genes the siRNA D6ctrl (RISC-free) shows the highest specificity among the 13 tested molecules with only four genes being signifi- cantly deregulated. Also only few differences were found for the treatment with the siRNAs D1ctrl (Non-targeting Pool), D3ctrl and D4ctrl. Here, less than 20 genes were altered compared to untransfected cells, whereas the control siRNAs A2ctrl, D5ctrl and Q5ctrl showed only moderate specificity with more than 20 differentially ex- pressed genes. The lowest specificity exerted the siRNAs A1ctrl, Q2ctrl, Q4ctrl, Q1ctrl and Q3ctrl with the latter two showing several hundreds of differentially expressed genes compared to the controls. A summary of all genes that were significantly (p- value < 0.01) deregulated with a fold change larger than 1.5 within the three cell lines is listed in Table 2 and Supplement 1. Table 2 - Number of control siRNA off-target transcripts in three different cell lines Color code: blue indicates a small amount of control siRNA off-target effects, red indicates a high amount. siRNA down up down up down up sum D6ctrl 0 1 0 3 0 0 4 D3ctrl 2 1 0 4 0 0 7 D4ctrl 0 3 0 8 1 1 13 D1ctrl 8 7 0 2 0 0 17 D2ctrl 9 5 1 3 0 0 18 D5ctrl 6 13 0 3 0 0 22 Q5ctrl 12 1 10 2 0 0 25 A2ctrl 18 4 0 0 10 6 38 A1ctrl 45 8 5 1 17 2 78 Q4ctrl 54 13 15 17 6 2 107 Q2ctrl 81 10 47 8 9 0 155 Q3ctrl 112 55 60 54 50 26 357 Q1ctrl 194 58 60 40 110 9 471 HT1080 HaCaT 3T3-L1 Results 46 Unspecific gene regulations induced by siRNA molecules can often be explained by partial sequence homologies with the respective mRNA off-targets. An important observation in this regard was that several of the gene regulations were recapitu- lated with other control siRNAs. Detailed analysis of the two human cell lines (HT1080 and HaCaT) identified 26 genes that are commonly deregulated by Q1ctrl and also 26 by Q2ctrl, 21 by Q3ctrl, 10 by Q4ctrl, 4 and 1 by A1ctrl and Q5ctrl (Sup- plement 1). However, only a small intersection is observed with the murine cell line 3T3-L1. Here, only the siRNA Q1ctrl down-regulated 3 genes (SPSB1, TOB1 and PPP1CC) in all three cell lines. Most deregulated genes were found after transfection of HT1080 cells. Exemplary, for the effects siRNAs can have on cell lines, the analysis was focused on the effects of the control siRNAs in HT1080 cells. Hierarchical clustering of all 595 identified genes altered after treatment with any of the control molecules clearly illustrates different degrees of similarity among the used control molecules (Figure 11). All Dharmacon siRNAs have similar effects on gene expression and are all arranged in one sub cluster structure. The closest similarity of gene regulation is observed for the Non-targeting Pool D1ctrl and the siRNA D2ctrl. Treatment with one of the other control siRNAs resulted in more heterogeneous expression patterns. However, the control siRNA molecules A1ctrl and Q1ctrl are grouped in one subcluster due to their overlapping, multiple and strong effects on gene expression. Results 47 Figure 11 – Hierarchical clustering of control siRNA off-targets in HT1080 cells Hierarchical clustering was performed based on all off-targets identified after expression profiling in HT1080 cells. The expression patterns of the different control siRNAs reveal several intersections in gene regulation. Blue indicates decreased expression relative to untreated cells, red indicates in- creased expression. A detailed analysis on commonly deregulated genes revealed that 79 genes are diffe- rentially expressed (FC > 1.5, p-value < 0.01) after the treatment with two or more siRNAs. Interestingly, it was also possible to identify a group of genes that were commonly influenced by up to seven control siRNA molecules (Table 3). Results 48 Table 3 - List of Genes Altered by 2 or More siRNAs Symbol Ensembl Gene Reg. by x siRNAs siRNA Regulation IL24 ENSG00000181856, ENSG00000162892 7 A1, A2, Q1, Q2, Q3, Q4, Q5 down TFRC ENSG00000072274, ENSG00000163975 6 D2, D3, D5, Q2, Q4, Q5 down FST ENSG00000134363, ENSG00000125744 5 D1, D2, D5, Q1, Q4 down IL1B ENSG00000117480, ENSG00000125538 5 A1, A2, Q1, Q2, Q5 down RGS4 ENSG00000115598, ENSG00000117152 5 A1, D1, D2, Q1, Q5 down AMMECR1 ENSG00000101935, ENSG00000160957 3 D1, D3, Q2 down CTGF ENSG00000198898, ENSG00000118523 3 A1, Q1, Q2 down ESM1 ENSG00000164283 3 A1, D2, Q1 down NOX4 ENSG00000086991 3 A1, Q1, Q2 down NUPR1_HUMAN ENSG00000176046, ENSG00000180035 3 Q1, Q3, Q4 down PNMA2 ENSG00000171362 3 Q1, Q3, Q4 down RRM2 ENSG00000176076, ENSG00000171848 3 D1, D2, Q2 down ABCA1 ENSG00000165029, ENSG00000129673 2 Q1, Q2 down ADI1 ENSG00000182551 2 Q1, Q3 down ATG2A ENSG00000156802, ENSG00000110046 2 Q2, Q4 down CBX6 ENSG00000183741, ENSG00000204149 2 D1, D5 down CLIC3 ENSG00000169583, ENSG00000160199 2 Q1, Q3 down DDAH1 ENSG00000153904, ENSG00000168172 2 A2, Q1 down DPYSL2 ENSG00000131264, ENSG00000092964 2 Q1, Q4 down DPYSL3 ENSG00000172352, ENSG00000113657 2 A1, Q3 down DUSP4 ENSG00000120875 2 A1, Q1 down E2F1 ENSG00000087237, ENSG00000101412 2 D2, Q4 down ENPP1 ENSG00000197594 2 Q1, Q3 down FLJ42986 ENSG00000196460 2 Q1, Q2 down GALNT4 ENSG00000120322 2 A1, Q1 down GDF15 ENSG00000130513 2 Q1, Q3 down GFPT2 ENSG00000131459 2 Q1, Q3 down H1FX ENSG00000184897, ENSG00000011405 2 Q2, Q3 down HAS2 ENSG00000134762, ENSG00000170961 2 A1, Q1 down IL11 ENSG00000095752 2 Q1, Q2 down IL6 ENSG00000151726, ENSG00000136244 2 Q1, Q3 down INSIG1 ENSG00000137714, ENSG00000186480 2 Q2, Q4 down JAG1 ENSG00000101384 2 A1, Q1 down KLC1 ENSG00000126214 2 A1, Q2 down MSRB3 ENSG00000174099 2 A1, Q3 down NDRG1 ENSG00000104419, ENSG00000184779 2 Q1, Q3 down PAPPA ENSG00000182752 2 Q1, Q5 down PDGFRB ENSG00000113721 2 Q1, Q3 down PRRT2 ENSG00000167371 2 Q1, Q3 down PTPN1 ENSG00000196396 2 A1, Q1 down PYGB ENSG00000211598, ENSG00000100994 2 A1, Q1 down RBPMS2 ENSG00000166831 2 Q1, Q3 down Results 49 Table 3 – continued Symbol Ensembl Gene Reg. by x siRNAs siRNA Regulation RDX ENSG00000136634, ENSG00000137710 2 Q2, Q4 down RNF144 ENSG00000151692, ENSG00000168118 2 A1, Q1 down SBSN ENSG00000189001 2 Q1, Q3 down SCN2A ENSG00000140859, ENSG00000136531 2 Q3, Q4 down SLC2A3 ENSG00000059804, ENSG00000150457 2 Q2, Q4 down SLC39A10 ENSG00000196950 2 Q2, Q3 down SNAI2 ENSG00000019549 2 A1, Q1 down SORBS2 ENSG00000154556 2 A1, Q1 down SPIRE1 ENSG00000134278 2 A1, Q4 down SPOCD1 ENSG00000134668 2 A1, Q1 down SSFA2 ENSG00000138434 2 A1, Q1 down STAG2 ENSG00000095539, ENSG00000101972 2 A1, Q3 down THBS2 ENSG00000186340 2 Q1, Q3 down UCN2 ENSG00000145040 2 Q1, Q2 down UNG ENSG00000076248 2 D2, Q4 down VGF ENSG00000128564 2 Q1, Q3 down ZMAT3 ENSG00000172667 2 Q2, Q3 down ZNF467 ENSG00000181444 2 Q1, Q3 down DICER1 ENSG00000148655, ENSG00000100697 7 A1, A2, D1, D2, D5, Q1, Q4 up PAEP ENSG00000125944, ENSG00000122133 6 D1, D2, D3, D4, D5, Q4 up COL22A1 ENSG00000169436 3 A2, D1, D5 up HES4 ENSG00000188290 3 D1, D5, Q3 up IL1A ENSG00000115008 3 D2, Q3, Q4 up KRT80 ENSG00000167767 3 A1, Q1, Q2 up LIN28B ENSG00000187772 3 A1, Q4, Q1 up TGFBR3 ENSG00000152127, ENSG00000069702 3 D1, D5, Q2 up ABCC3 ENSG00000108846, ENSG00000214570 2 A1, Q1 up C2CD2 ENSG00000157617, ENSG00000104413 2 D2, D5 up CALB2 ENSG00000172137, ENSG00000138363 2 Q1, Q4 up CDC25A ENSG00000137875, ENSG00000164045 2 Q1, Q3 up EGR1 ENSG00000101204, ENSG00000120738 2 D5, Q4 up GPR116 ENSG00000069122 2 D5, Q1 up KIAA1324 ENSG00000116299 2 A1, Q1 up MAFB ENSG00000204103, ENSG00000117906 2 D1, D5 up PHF17 ENSG00000077684 2 Q1, Q3 up SKP2 ENSG00000168028, ENSG00000145604 2 Q1, Q3 up STX3 ENSG00000099399, ENSG00000166900 2 Q1, Q3 up Results 50 2.2.1.3 Control siRNAs influence expression of different cytokines and MMP1 The gene expression analysis clearly indicated that the mRNA levels of several well- characterized genes are altered after the transfection of the control siRNAs. Especial- ly siRNAs A1ctrl, Q1ctrl but also A2ctrl, Q2ctrl, Q3ctrl, Q4ctrl and Q5ctrl showed sig- nificant deregulations of cytokines like IL1 or IL24. To further understand whether the observed expression differences correlate with differences in the respective sig- naling pathways the well-known influence of IL1 on the gene expression of MMP1 as a marker for the status of the IL1 signaling pathway95 was observed. MMP1 le- vels in supernatants of control siRNA transfected cells were analyzed by ELISA and a strong MMP1 down-regulation was observed after A1ctrl (3.7-fold) and Q1ctrl (2.5- fold) transfection. Less significant effects on MMP1 levels were measured for the siRNAs A2ctrl, Q2ctrl, Q3ctrl and Q5ctrl. Again, the moderate effects on MMP1 re- lease correlates with IL1 and IL24 expression (Figure 12). Results 51 Figure 12 - Control siRNA-dependent MMP1 protein secretion HT1080 cells were transfected with 13 different control siRNA molecules (A1-D5) a: 72 hours after the cells were transfected MMP1 levels in the supernatants were determined by ELISA. siRNAs A1ctrl, A2ctrl, Q1ctrl, Q2ctrl, Q3ctrl and Q5ctrl significantly decrease MMP1 levels. b: the protein secretion of MMP1 measured by ELISA correlates with the mRNA expression of IL1β and IL24 measured by micro- array. The results represent two independent MMP1 measurements performed in triplicates which are correlated to the expression profiling data of IL1β and IL24. Student’s t-test was used to calculate the significance compared to untreated cells (**< 0.0001, *< 0.01). All error bars indicate the standard deviation of n=3. (TR: transfection reagent, UT: untreated cells). Results 52 2.2.1.4 Control siRNAs influence TNF signaling Next, it was addressed whether the lack of specificity of the control siRNAs does not only reduce the basal cytokine expression levels or MMP1 secretion but also inter- feres with major cell signaling pathways. Since IL1 and IL24 can both lead to NF B activation96, 97, cells were stimulated with TNF for activation of the IKK/NF B signal- ing cascade. IL8, a known target gene of NF B, showed a strong response upon TNF stimulation (Fig. 4). Again ELISA analysis of supernatants derived from TNF treated HT1080 cultures demonstrated a reduced sensitivity of IL8 release in the presence of the control siRNA molecules A1ctrl and Q1ctrl (Figure 13). Figure 13 - Control siRNA-dependent IL8 release HT1080 cells were transfected with 13 different control siRNA molecules (A1ctrl-D5ctrl). 72 hours after transfection cells were stimulated with 30 ng/ml TNFα for 8 hours. IL8 release was determined by ELISA of the cell supernatants. IL8 levels were 7.3 fold decreased after A1ctrl and 2.2 fold de- creased after Q1ctrl treatment compared to untreated cells. The results are representative of two independent experiments performed in triplicates. Student’s t-test was used to calculate the signific- ance compared to untreated cells (**< 0.0001, *< 0.01). All error bars indicate the standard deviation of n=3. (TR: transfection reagent, UT+: untreated cells with TNFα stimulation, UT-: untreated cells without TNFα stimulation) Results 53 In summary, the control siRNA D5ctrl was selected for further use in the Phenocopy experiment. D5ctrl was chosen due to its highest selectivity among all siRNAs in HaCaT cells. Hereby excluded were the siRNAs D1ctrl and D6ctrl although both siRNAs resulted in less off-target effects. However, both controls differ from the oth- ers. Within the Phenocopy experiment the goal was to use single siRNAs and not siRNA pools. Therefore, the use of siRNA D1ctrl, composed of four different siRNAs (SMARTpool), is not an appropriate control for this approach. Furthermore, a control molecule unable to enter RISC, such as D6ctrl, can also not be used to test the entire spectrum of a siRNA experiment since the RNAi machinery will not be affected through such a molecule. 2.2.1.5 TGF-βR1 siRNA characterization Subsequent to the selection of a proper transfection protocol and the identification of a good control siRNA molecule, assessment of specific TGF-βR1 siRNAs was carried out. Criteria for a good candidate were a strong and stable mRNA knockdown, sub- sequent inhibition of the TGF-β signaling pathway and low off-target effects. To guarantee optimal siRNA-mediated TGF-βR1 knockdown, ten commercially available siRNAs were qualified (Table 4). First, knockdown efficacy was determined on mRNA level using qRT-RCR. Only five siRNAs (A1tgf, D1tgf, D2tgf, Q3tgf and Q4tgf) which let to a knockdown of more than 90 % were selected for off-target profiling (Figure 14a). Second, inhibition of downstream signaling of each selected siRNA was determined by phospho-Smad2/3 (Figure 14b) and PAI-1 ELISA (Figure 14c). Interestingly, al- though transfection of siRNA D1 resulted in the best mRNA knockdown (98 %), this Results 54 finding was not represented in the functional readouts. The strongest functional knockdown was observed for siRNA A1tgf. Figure 14 – TGF-β siRNA characterization a: siRNA knockdown efficiency was measured by qRT-PCR 48 h post transfection. Ten different com- mercially available siRNAs (A - Ambion, D - Dharmacon & Q - Qiagen) were used. b and c: siRNAs with the best knockdown efficacy (A1tgf, D1tgf, D2tgf, Q3tgf & Q4tgf), as well as the untreated control (UT) were analyzed for functional blockade of TGF-β signaling determined by inhibition of p-Smad2/3 (b, p-Smad2/3 ELISA) or PAI-1 protein (c: PAI-1 ELISA). All error bars indicate the standard deviation of n=3. Results 55 Finally, the off-target effects of all siRNAs were determined by microarray analysis using Illumina Beadchip technology. All deregulated genes (p-value < 0.01 and |LR| ≥ 1) were identified for the selected five siRNAs (Figure 15). To exclude genes from the off-target list that are relevant for the mechanism of the procedure or relevant for the TGF-βR1 biology, only those genes were selected that were uniquely deregulated by the respective siRNA. The profiling was performed without TGF-β stimulation to focus the analysis on off-target effects. Due to its superior functional knockdown abilities (Figure 14b) and little off-target effects siRNA A1tgf (Figure 15b) was used in all further experiments. Table 4 – List of TGF-βR1 siRNAs Vendor Description Cat. No. Abbr. Ambion Silencer® Select TGF beta receptor 1 AM51331-556 A1tgf Ambion Silencer® Select TGF beta receptor 1 AM51331-557 A2tgf Dharmacon On-TARGET plus Duplex TGFBR1 J-003929-09-05 D1tgf Dharmacon On-TARGET plus Duplex TGFBR1 J-003929-10-05 D2tgf Dharmacon On-TARGET plus Duplex TGFBR1 J-003929-11-05 D3tgf Dharmacon On-TARGET plus Duplex TGFBR1 J-003929-12-05 D4tgf Qiagen Hs_TGFBR1_5_HP Validated siRNA SI00301903 Q1tgf Qiagen Hs_TGFBR1_6_HP Validated siRNA SI02223627 Q2tgf Qiagen Hs_TGFBR1_7_HP Validated siRNA SI02223634 Q3tgf Qiagen Hs_TGFBR1_9_HP Validated siRNA SI02664158 Q4tgf Results 56 Figure 15 – TGF-β siRNA off-target effects Volcano plots for siRNAs A1tgf, D1tgf, D2tgf, Q3tgf & Q4tgf. Total RNAs of biological triplicates were isolated post siRNA transfection and were hybridized to Illumina Beadchips. The off-target effects were analyzed by volcano plots. Each circle represents a single gene of the human genome. The x-axis depicts the log2 ratio (LR) between each siRNA and untreated cells. The y-axis is scaled as -log10 [p- value] (Student t-test) as a indicator of significance. An off-target is defined to have a |LR|≥1 and a -log10 [p-value] > 2. D5ctrl siRNA vs. untreated D5ctrl revealed no off-target effects (a). The siRNAs A1tgf revealed 22 genes to be deregulated (b), the siRNA D1tgf - 8 genes (c), the siRNA D2tgf – 25 genes (d), the siRNA Q3tgf – 58 genes (e) & the siRNA Q4tgf – 42 genes (f). Results 57 In addition, the stability of the receptor knockdown using siRNA A1tgf was deter- mined. Therefore, HaCaT cells were transfected and TGF-βR1 mRNA was analyzed 1, 2, 3, 4 and 7 days after transfection. Transfection of siRNA A1tgf resulted in a stable knockdown of at least 70 % even 7 days after transfection (Figure 16). Figure 16 – Knockdown stability HaCaT cells were transfected with siRNA A1tgf and knockdown efficacy was determined by qRT-PCR 1, 2, 3, 4 and 7 days post transfection. Even after 7 days knockdown was stable with approximately 70 %. All error bars indicate the standard deviation of n=2. Results 58 2.2.2 Kinase inhibitors The kinase inhibitors used in the Phenocopy experiment derived from a Boehringer Ingelheim (BI) lead optimization project. Additionally, two previously described com- petitor compounds98 were also incorporated in the analysis. The BI inhibitors were identified in a high-throughput screening of the in-house compound collection using a biochemical TGF-βR1 substrate phosphorylation as- say99. Among the hits, compound 5 (Figure 17) and several other indolinone deri- vates were identified, displaying down to double-digit nanomolar potency. Since it was known from earlier projects that the indolinone chemotype can show low cross- reactivity with the human kinome100, it was chosen as starting point for further op- timization. Structure-activity relationship studies were performed by substitution of the indolinone core in position R1, R2 and R3 of the scaffold (Figure 17) in order to increase the potency. Besides the inhibitory potency of the compounds in the prima- ry biochemical kinase assay it needed to be shown that they also inhibit TGF-β signal transduction in a cellular setting. Therefore the pSmad2/3 ELISA was used to further characterize the inhibitors’ potency. All compounds were analyzed in seven different concentrations: 0.0032, 0.016, 0.08, 0.4, 2, 10 and 50 µM. In the following the compounds were named with numbers adopted from Roth et al.33 where their synthesis and design is described in more detail. The poten- cy data derived from the TGF-βR1 substrate phosphorylation assay was also taken from this work. Results 59 N H O N O N H N N H O N H R 3 R 2 R 15 6 Figure 17 –TGF-βR1 hit compound 5 (BIBF0775) and general structure 6 The first subset of compounds contained indolinones substituted in position R1 (Table 5). Hereby, the compounds substituted with smaller secondary and tertiary amides (5, 35-38) in position 6 were favorable for a good TGF-βR1 potency. The un- substituted indolinone 45 was significantly less active confirming that the 6-amido substituents contribute to the overall binding energy. In contrast, the same amides in position 5 resulted in less potency of the compound (compare compounds 46 and 37). In general, the trend of inhibition in the cellular setting correlated with the data from the biochemical assay. However, the compounds were less potent when com- pared to the biochemical TGF-βR1 inhibition (factor 3-12 for potent compounds). Taken together, due to their good potency on a biochemical as well as on cellular level, these compounds looked promising as starting points for further optimization. All potencies of the modified indolinones are listed in Table 5. Results 60 N H O N H N R 1 5 6 Table 5 – TGF-βR1/pSMAD inhibition of substituted indolinones Cpd. R1 TGF-βR1 IC50 (nM) a pSMAD IC50 (nM) a 35 6-CONH(CH2CH3) 24 ± 17 75 ± 53 36 6-CONHCH3 32 ± 26 NT b 5 6-CONEtMe 34 ± 30 105 ± 70 37 6-CONMe2 35 ± 22 246 ± 130 38 6-CONHnBu 91 ± 49 1065 ± 588 39 6-(pyrrolidine-1-carbonyl) 245 ± 170 696 ± 542 40 6-CONHiPr 318 ± 188 708 ± 453 41 6-CONH2 369 ± 213 >50000 42 6-CONH(CH2CH2OH) 430 ± 227 >50000 43 6-CONEt2 625 ± 396 NT b 44 6-CONHBn 1532 ± 908 NT b 45 H 3462 ± 2621 NT b 46 5-CONMe2 186 ± 104 198 ± 111 a Values are averages ± SD of at least three independent determinations. Values “greater than” indicate that half-maximum inhibition was not achieved at the highest concentration tested. b Not tested. To further explore structure-activity relationships, the R2 side chain pointing towards the water phase was chosen for the next round of modifications (Table 6). Here, a large degree of freedom for structural variation was observed with various linkers between the aniline and the basic moiety were tolerated (47a-u). The distance be- tween the basic moiety and the core seemed to play a minor role demonstrating that structure-activity relationships in this respect were shallow. Neutral compounds were clearly inferior (compare especially compound 47b with 47q). Anilines with smaller substituents (compounds 47r and 47u) or without any substitution (com- pound 47v) were less active. Shifting the R6 substituent on the aniline (see Table 6) from position 4 into position 3 was also detrimental to activity (compound 47s). Re- placing the aniline by saturated cyclic systems such as 48 and 49 led to complete loss Results 61 of potency. Comparable to hit compounds 5 and 35, indolinones substituted in posi- tion 6 with either an ethylamido or an ethylmethylamido-substituent usually showed comparable activities with a tendency for higher potency for the ethylamido- substituted compounds (compare e.g. couples 47a/b, 47c/d, 47e/f etc.). Although, structure-activity relationships were shallow, improved compounds with single-digit nanomolar TGF-βR1 inhibition could be identified by variation of position R2, with compound 47a showing an optimized IC50 of 1 nM. Cellular efficacy of the com- pounds in Table 6 was again evaluated by pSmad2/3 ELISA assay. For many com- pounds, IC50 values followed the same trend as mentioned before, i.e. cellular effica- cy differed by a certain factor from biochemical IC50. However, many compounds of this series displayed increased shifts between cellular and biochemical inhibition. Amongst them, potent compounds such as 47c, 47e, 47g, and 47h failed to show convincing cellular potency at all. This could be rationalized by the increased polarity of the compounds probably preventing them from permeating into the cells. How- ever, despite their cellular shifts, several optimized compounds with attractive cellu- lar activities such as 47a, 47d, 47i or 47l were identified by variation of R2. Results 62 N H O R 1 N H R 6 N H O N O N H N N H O N O N H 48 4947a-v Table 6 – Inhibitory profile of 6-amido-substituted indolinones Cpd. R1 R6 TGFβ-R1 IC50 [nM] a pSmad IC50 [nM] a 47a CONHEt 4-(NSO2CH3)(CH2)2NMe2 1 ± 1 108 ± 65 47b CONEtMe 4-(NSO2CH3)(CH2)2NMe2 7 ± 5 209 ± 108 47c CONHEt 4-(NCOCH3)(CH2)2NMe2 3 ± 3 1066 ± 659 47d CONEtMe 4-(NCOCH3)(CH2)2NMe2 8 ± 5 102 ± 61 47e CONHEt 4-(NCOCH3)(CH2)3NMe2 3 ± 3 1249 ± 664 47f CONEtMe 4-(NCOCH3)(CH2)3NMe2 9 ± 6 411 ± 309 47g CONHEt 4-(NCH3)COCH2-(4-methyl-piperazin- 1-yl) 9 ± 11 1370 ± 735 47h CONEtMe 4-(NCH3)COCH2-(4-methyl-piperazin- 1-yl) 7 ± 6 833 ± 549 47i CONHEt 4-CH2NMe2 19 ± 13 185 ± 99 35 CONHEt 4-CH2(piperidin-1-yl) 24 ± 17 75 ± 53 5 CONEtMe 4-CH2(piperidin-1-yl) 34 ± 30 105 ± 70 47j CONHEt 4-(NCH3)COCH2NMe2 35 ± 22 3383 ± 3291 47k CONEtMe 4-(NCH3)COCH2NMe2 29 ± 16 382 ± 207 47l CONEtMe 4-CH2NHEt 32 ± 18 160 ± 83 47m CONHEt 4-CONH(CH2)2NEt2 33 ± 23 505 ± 264 47n CONEtMe 4-CONH(CH2)2NEt2 64 ± 53 542 ± 317 47o CONHEt 4-(CH2CH2)NMe2 47 ± 27 135 ± 70 47p CONEtMe 4-(CH2CH2)NMe2 64 ± 40 259 ± 190 47q CONEtMe 4-(NSO2CH3)CH2CONMe2 86 ± 53 > 3000 47r CONEtMe 4-(NCH3)SO2Me 231 ± 129 > 10000 47s CONEtMe 3-CH2NEt2 616 ± 353 NT b 47t CONEtMe 4-COOH 782 ± 435 NTb 47u CONEtMe 4-COOCH3 1970 ± 1085 NT b 47v CONEtMe H 2586 ± 1844 NTb 48 > 50000 > 10000 49 > 50000 > 10000 a Values are averages ± SD of at least three independent determinations. Values “greater than” indi- cate that half-maximum inhibition was not achieved at the highest concentration tested. b Not tested. Results 63 Finally, the influence of substitution at R3 (Figure 17) was explored (see Table 7). When comparing compounds in which R3 = phenyl with the corresponding com- pounds with R3 = H (47w-z), a slight drop of potency could be observed for TGF-βR1 inhibition. Since compounds 47w-z (R3 = H) are more polar than their phenyl coun- terparts, poor cell permeability is probably again the reason for the low activity of the compounds in the pSmad2/3 ELISA assay. Only the more lipophilic compound 47y showed improved cellular activity in this series. N H O R 3 N H R 6 NH O Table 7 – Inhibitory profile of various R4-substituted indolinones Cpd. R3 R6 TGFβ-R1 IC50 [nM] a pSmad IC50 [nM] a 47e Ph 4-(NCOCH3)(CH2)3NMe2 3 ± 3 1249 ± 664 47w H 4-(NCOCH3)(CH2)3NMe2 17 ± 10 597 ± 332 47g Ph 4-(NCH3)COCH2-(4-methyl-piperazin-1-yl) 9 ± 11 1370 ± 735 47x H 4-(NCH3)COCH2-(4-methyl-piperazin-1-yl) 15 ± 9 2175 ± 845 35 Ph 4-CH2(piperidin-1-yl) 24 ± 17 75 ± 53 47y H 4-CH2(piperidin-1-yl) 69 ± 44 376 ± 246 47j Ph 4-(NCH3)COCH2NMe2 35 ± 22 3383 ± 3291 47z H 4-(NCH3)COCH2NMe2 52 ± 30 4355 ± 2298 a Values are averages ± SD of at least three independent determinations. Results 64 As mentioned above, five kinase inhibitors out of this lead optimization program were selected to further characterize them in the Phenocopy approach and to dem- onstrate that this approach is able to unravel new insights into the mode of action of these compounds beyond those that are currently investigated during a lead optimi- zation phase. Therefore five BI compounds, in the following referred to as BI1-5 and two competitor substances (Ex1 and Ex2) with a wide range in inhibitory potency of TGF-βR1, from 19 nM (BI3) to 1537 nM (Ex2), were selected. A detailed list of the compounds’ features, including biochemical and cellular IC50 values, is provided in Table 8. The potencies (IC50) for the inhibition of TGF-βR1 kinase, Smad2/3 phospho- rylation (pSmad) and PAI-1 protein are indicated for compounds BI1 to BI5 (indoli- nones [INDO]) and Ex1 and Ex2 (pyridopyrimidinones [PyPy]). The PubChem CIDs are indicated. According to the chemical synthesis of the compounds (Roth et al.33), the corresponding compounds identification numbers are indicated in brackets. Results 65 Table 8 - List of profiled TGF-βR1 kinase inhibitors in the Phenocopy project Formula C30H32N4O2 MW [g/mol] 480.609 Structural Class INDO IC50 TGFβR1 [nM] 186 IC50 pSmad [nM] 198 IC50 PAI-1 [nM] 438 PubChem CID Formula C31H34N4O2 MW [g/mol] 494.636 Structural Class INDO IC50 TGFβR1 [nM] 34 IC50 pSmad [nM] 105 IC50 PAI-1 [nM] 165 PubChem CID Formula C27H28N4O2 MW [g/mol] 440.5442 Structural Class INDO IC50 TGFβR1 [nM] 19 IC50 pSmad [nM] 185 IC50 PAI-1 [nM] 227 PubChem CID Formula C28H30N4O2 MW [g/mol] 454.571 Structural Class INDO IC50 TGFβR1 [nM] 32 IC50 pSmad [nM] 160 IC50 PAI-1 [nM] 223 PubChem CID Formula C29H32N4O2 MW [g/mol] 468.598 Structural Class INDO IC50 TGFβR1 [nM] 64 IC50 pSmad [nM] 259 IC50 PAI-1 [nM] 1550 PubChem CID Formula C26H27Cl2N5O2 MW [g/mol] 512.438 Structural Class PYPY IC50 TGFβR1 [nM] 25 IC50 pSmad [nM] 211 IC50 PAI-1 [nM] 220 PubChem CID 5311382 Formula C20H14Cl2N4O MW [g/mol] 397.264 Structural Class PYPY IC50 TGFβR1 [nM] 1537 IC50 pSmad [nM] 400 IC50 PAI-1 [nM] 855 PubChem CID 5327885 Ex2 [PD164199] Ex1 [PD166285] INDO - Indolinone | PYPY - Pyridopyrimidinone BI4 (Cpd #47l) BI5 (Cpd #47p) BI1 (Cpd #46) BI2 (Cpd #5 / BIBF0775) BI3 (Cpd #47i / BI34659) N O Cl Cl CH 3 N N N N O N N O CH 3 CH 3 N CH3 CH 3 N O Cl ClCH 3 N N N O N CH 3 CH 3 N N O N O CH 3 CH 3 N N N O N O CH 3 CH 3 N N O N N O CH 3 CH 3 N CH 3 N O N N O CH 3 N CH 3 CH 3 - - - - - Results 66 2.3 Phenocopy Experiment The profiling of seven NCEs (BI1-5 & Ex1, Ex2) at seven different concentrations (0.0032, 0.016, 0.08, 0.4, 2, 10, 50 µM) and three time points (2, 4, 12 h), including siRNA A1tgf, all appropriate controls and in biological triplicates for each condition resulted in an overall experimental setup of 651 samples to be submitted to array profiling. Subsequently, an optimal normalization method was selected for the ex- pression data, the TGF-β signature as well as the NCE’s off-target effects were de- termined and the obtained in silico results were validated in wet laboratory experi- ments. 2.3.1 Data normalization Expression data was pre-processed in 24 different ways (Table 10). In this part, the analysis was focused on analyzing the TGF-β stimulated and control samples meas- ured at three time points (2, 4, and 12 h). Generally speaking, first either background normalization from BeadStudio101 (bg_*) or no background modification (noBg_*) was applied. In a next step, the data was transformed using either log2- transformation (log) or vst102. Since background normalization can lead to negative values, the data had to be transformed to contain only positive values by using ei- ther the background correction of rma103 or forcePos104 to be able to apply log2- transformation. In a last step, the data was normalized using quantile, loess or rsn normalization. Alternatively, the transformation steps were skipped and vsn105 or normalization methods supplied by BeadStudio (average, rankInvariant, cubicSpline) were used for normalization. Different pre-processing methods were evaluated by Results 67 analyzing the resulting gene expression intensities via various statistical measures. These methods were scored from -2 to 2 based on how well they matched the re- quired criteria for the different analyses (Table 10). Their impact on the quality for normalization of the presented expression study was analyzed in a separate study and was submitted for publication at BMC Genomics106. In addition, to the investiga- tion of the actual expression intensities, fold changes derived from resulting gene expression intensities were compared to fold changes based on quantitative mea- surements of RNA abundance as determined by qRT-PCR. Therefore, the expression of eight genes that are known to be regulated by TGF-β signaling (CDKN1A, CDKN2B, HAND1, JUNB, LINCR, RPTN, PAI-1 and TSC22D1) was analyzed. By this means, it is possible to compare the results of the normalization methods to values that reflect the real abundance of the respective mRNA in the cells. To guarantee that the com- parisons of the normalization methods are not biased towards certain intensities, the mRNAs used in qRT-PCR experiments were chosen such that the respective sig- nals on the chips cover a broad range of expression intensities (Table 9). Table 9 – qRT-PCR fold changes of TGF-β dependent genes Fold change of TGF-β stimulated vs. unsti- mulated HaCaT cells Gene 2 h 4 h 12 h PAI-1 70.07 98.32 124.85 CDKN1A 6.22 2.36 4.17 LINCR 5.06 0.74 0.05 CDKN2B 10.26 6.36 4.12 HAND1 3.65 2.68 0.54 TSC22D1 2.42 1.73 1.32 RPTN 9.03 4.63 0.03 JUNB 18.41 8.63 10.19 Results 68 Based on the different normalization procedures for the gene expression experiment and based on the qRT-PCR measurements, the Pearson correlation of the respective fold changes measured for TGF-β stimulated versus untreated cells at 2, 4 and 12 h were calculated. Figure 18 displays the ranked correlation coefficients describing the relation be- tween the different normalization methods and the qRT-PCR results. Quality values were assigned based on correlation cut-offs. A value of 2 is assigned to correlation coefficients ≥ 0.96, a value of 1 to coefficients between 0.94 and 0.96, a value of 0 to coefficients between 0.92 and 0.94, a value of -1 to coefficients between 0.9 and 0.92 and a value of -2 to correlation coefficients ≤ 0.9 (Table 10). Values derived from most of the methods not utilizing background correction (noBg_*) show a low- er correlation to the qRT-PCR results than expression intensities that are background corrected (bg_*). An exception in this regard are methods that are based on vst transformation (bg_vst_*). These three methods are amongst the six methods re- sulting in the lowest correlation coefficient values. Correlation coefficients exhibiting high values are delivered by methods introducing BeadStudio’s background correc- tion combined with either rma background correction and log2-transformation (bg_rma_log_*), cubic spline normalization (bg_cubicSpline) or variance stabilizing normalization (bg_vsn). Results 69 Figure 18 - Pearson correlation of log2 ratios for different normalization methods Correlations of log2 ratios determined based on different normalization methods for gene expression data obtained from the Illumina array measurement to qRT-PCR results. On the x-axis, pre-processing methods are ranked according to their correlation. The dashed red lines indicate the cut-offs used for assigning quality score between -2 (< 0.9) and 2 (> 0.96). As it is difficult to clearly categorize the methods based on the examined measures, the final decision of which score to assign to some extent stays subjective. However, it is unambiguously possible to separate better pre-processing methods from worse. A summary of all analyzed parameters, statistical measures and the fold change cor- relation is given in Table 10. Results 70 Table 10 –Quality scores given for the different pre-processing methods Displayed are the quality scores for the different pre-processing methods given for the analyses con- ducted. Quality scores range from -2 (bad/red) to 2 (good/blue). The last column displays the sum over the quality scores assigned. Based on this sum, the preprocessing method finally used to normal- ize the Phenocopy data has been chosen. An integration of all scores for the analyzed parameters revealed that best results for the present expression data were obtained when no background modification is used in combination with log2-transformation and either quantile or rsn normaliza- tion. It was therefore decided to use rsn normalization in the Phenocopy study. lo g1 0( p- va lu e) ve rs us M SQ be tw ee n Bo xp lo ts M SQ De ns ity fu nc tio ns o f M SQ s Vo lca no p lo ts Re sid ua l s d ve rs us e xp re ss io n le ve ls Sc at te rp lo ts AU C Sl op e of re gr es sio n Co rr el at io n to q RT -P CR Su m bg_average 0 -1 0 -1 -2 -1 0 0 1 -4 bg_cubicSpline 0 -1 1 -1 -2 -1 0 1 2 -1 bg_forcePos_log_loess 1 -1 1 0 -1 0 0 1 1 2 bg_forcePos_log_quantile 1 -1 1 0 -2 -1 0 1 1 0 bg_forcePos_log_rsn 1 -1 1 0 -2 -1 0 0 1 -1 bg_noNorm 1 -1 1 -1 -2 -1 0 0 0 -3 bg_rankInvariant 0 -2 -1 -1 -2 -1 1 0 1 -5 bg_rma_log_loess -1 -1 -1 -2 -2 -1 -1 2 2 -5 bg_rma_log_quantile -1 -1 0 -2 -2 -2 0 2 2 -4 bg_rma_log_rsn -1 -1 0 -2 -2 -2 0 2 2 -4 bg_vsn -1 -1 0 -1 -2 -2 0 1 2 -4 bg_vst_loess 2 0 -2 -1 2 1 -1 -1 -2 -2 bg_vst_quantile 2 2 1 1 2 2 0 -1 -1 8 bg_vst_rsn 2 2 1 1 2 2 0 -1 -1 8 noBg_average 1 -2 -2 1 1 1 -1 -1 -1 -3 noBg_cubicSpline 2 1 1 1 0 2 0 0 0 7 noBg_log_loess 2 0 -2 0 0 1 0 0 0 1 noBg_log_quantile 2 2 2 2 1 1 0 0 0 10 noBg_log_rsn 2 2 2 2 1 1 0 0 0 10 noBg_noNorm 1 2 1 -1 0 1 -1 -1 -1 1 noBg_rankInvariant 0 -2 -1 -1 0 2 1 -2 -1 -4 noBg_vsn 2 2 2 2 0 1 0 0 0 9 noBg_vst_loess 2 0 -2 0 2 2 -1 -1 -2 0 noBg_vst_quantile 2 2 1 1 2 2 0 -1 -1 8 noBg_vst_rsn 2 2 1 1 2 2 0 -1 -1 8 Results 71 2.3.2 TGF-β signature To gain a deeper insight into the TGF-β biology first genes were identified that are regulated due to TGF-β stimulation (5 ng/ml). To unravel the time-dependent effects of TGF-β treatment, HaCaT cells were stimulated with TGF-β for 2, 4 and 12 h. While immediate early genes that are directly regulated by the TGF-β pathway are de- tected at 2 h post stimulation, more and more secondary effects linked to TGF-β sig- naling are found after 4 h and/or 12 h. To avoid arbitrary log ratio cut-offs a modula- tor-based approach was used to identify TGF-β dependent gene regulation. There- fore, two criteria were applied: only genes that were significantly deregulated (p- value < 0.01) in a basic comparison of TGF-β stimulation versus unstimulated cells were further analyzed. We found 1,046, 1,949 and 5,725 genes (6,525 non- redundant genes) to be regulated 2, 4 and 12 h after stimulation (Figure 19). Next, these genes were proven to be affected dose-dependently by kinase inhibitor treat- ment after TGF-β stimulation. The approach allowed separating these genes from potential compound related off-target effects. All transcripts identified for each NCE were merged to a common signature of TGF-β dependent genes. This strategy al- lowed the identification of a common on-target signature minimized for the amount of false positive and false negative genes. The Venn diagram (Figure 19b) depicts the number of genes that were identified after 2, 4, and 12 h of stimulation: 446 genes (2 h), 772 genes (4 h) and 1,932 genes (12 h). All gene identifier annotations and regulations are listed in Supplement 2. Beyond the inhibition of the kinase activity by chemical compounds, the TGF-β pathway was also silenced by siRNA knockdown. All previously selected genes (TGF-β Results 72 stim. vs. unstimulated cells, Figure 19a) were tested to be regulated by siRNA- mediated knockdown of TGF-βR1. To exclude mechanistical effects, genes were only selected when they were regulated by siRNA A1tgf and not by a control siRNA (p- value < 0.01). By siRNA knockdown of TGF-βR1, 303 (2 h), 419 (4 h) and 1,112 (12 h) genes are identified as TGF-β dependent (Figure 19c). Although fewer genes were identified compared to the NCE approach, the vast majority of genes were identified by both approaches. According to the siRNA transfection procedure, a slightly differ- ent experimental setup was performed regarding cell seeding and culture conditions. This variation resulted in procedure specific changes in gene regulation, which had to be separated from the TGF-β signature. Addressing the level of TGF-βR1 activity upon siRNA transfection, the expression of PAI-1 as a surrogate marker for the TGF-β signaling activity was analyzed. Using NCEs it was possible to inhibit the PAI-1 ex- pression by more than 95 % at all time points. In contrast, the use of siRNA A1tgf reduced PAI-1 levels only to 62 % after 2 h of TGF-β stimulation (Figure 19d). Despite the high efficiency of the siRNAs (A1 showed a mRNA knockdown efficiency of great- er than 90 %; Figure 14a) treatment resulted only in a partial reduction of the TGF-β signaling. Results 73 Figure 19 – TGF-β signature The TGF-β signature was generated based on gene regulation upon treatment with TGF-β, TGF-βR1 kinase inhibitors (NCEs) or a siRNA. a: volcano plots of the comparison between TGF-β stimulated and non stimulated cells at 2, 4 and 12 h. Every circle represents a single transcript. The x-axis shows the log2 ratio (LR) between TGF-β stimulated vs. untreated HaCaT cells. The y-axis is scaled as negative log10 [p-value] as an indicator of significance. P-values were FDR-corrected according to Benjamini- Hochberg. Blue circled genes are significantly regulated by the stimulation with TGF-β (p-value < 0.01). b: the list of non-redundant genes was filtered for a concentration-dependent regu- lation upon NCE treatment and TGF-β stimulation.: 446, 772 and 1,932 genes were identified as the NCE-dependent on target TGF-β signature after 2, 4 and 12 h. c: the siRNA dependent TGF-β signature identified 307, 419 and 1,112 genes which were classified as siRNA-dependent on-target TGF-β signa- ture genes after 2, 4 and 12 h. d: expression level of PAI-1 mRNA as a surrogate marker for TGF-β signaling pathway activity after treatment with NCE B1 or siRNA. Treatment with NCE BI1 resulted in a complete knockdown of PAI-1 expression (>95 %) for all time points. In contrast the siRNA A1 me- diated knockdown of the TGF-β signaling only reduced PAI-1 levels partially to 62 %, 65 % and 78 % after 2, 4 and 12 h of TGF-β stimulation. Results 74 Subsequently, the genes of the TGF-β signature were used to perform gene set enrichment analysis (GSEA)107,108. The annotation of the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways delivered gene sets corresponding to 201 different pathways74, 109, 110. The GSEA resulted in 16 different signaling pathways which were significantly influenced upon TGF-β stimulation of HaCaT cells. The signaling path- ways were clustered in four groups (Figure 20). Not surprisingly, the TGF-β signaling pathway itself, as well as directly affected pathways like WNT and p53 signaling, were significantly regulated by the treatment of TGF-β (Cluster 1). In Cluster 2 MAPK, cytokine, ErbB, Hedgehog, as well as apoptosis signaling pathways are strongly af- fected immediately early upon TGF-β stimulation. Their modulation is reduced at later time points (4 h and 12 h), when more secondary effects, such as DNA polyme- rase, actin cytoskeleton, amino acid metabolism, gap junction and tight junction sig- naling become apparent (Cluster 3). The activation of these pathways in combination with the modulation of the cell cycle and cell communication activity (Cluster 4) seems to be the phenotypic consequences of TGF-β stimulation of HaCaT cells. To proof the findings obtained from the KEGG analysis, additionally Ingenuity Pathway Analysis (Ingenuity Systems®, www.ingenuity.com) was used to link and group genes from the TGF-β signature. In line with the KEGG results, the analysis identified the same connections and networks containing signaling but also WNT and ERK/MAPK signaling (Figure 21). In addition, diverse networks of genes were identified that play a role in embryonic development of different organs, but also in cellular proliferation and growth. Results 75 Figure 20 – TGF-β signature gene set enrichment analysis I Gene set enrichment analysis (GSEA) using KEGG gene annotation resulted in 16 significantly affected genesets/signaling pathways. Clustering of -log10 [p-values] using complete linkage and manhattan distance resulted in four major clusters: immediate early affected pathways (cluster 2), permanently affected pathways with emphases at early (cluster 1) and late time points (cluster 4) or late estab- lished events (cluster 3). The color code defines the significance determined by Fisher’s exact test: blue < 2 – not significant; white = 2 – significant & red > 2 – highly significant). Results 76 Figure 21 – TGF-β signa- ture gene set enrich- ment analysis II Networks of interact- ing and regulated mo- lecules from the on- target signature gen- erated by Ingenuity Pathway Analysis. All depicted molecules are represented as nodes and the biological relationship between two nodes is re- presented as an edge (line). All edges are supported by at least one literature refer- ence. The intensity of the node color indi- cates the degree of up- (red) or down- (green) regulation. Nodes are displayed using various shapes that represent the functional class of the gene product. a: a network of molecules directly related to the canonical TGF-β signal- ing pathway containing genes involved in cell signaling, connective tissue development and function and in skeletal tissue devel- opment and function. b: a network of mole- cules of the WNT and the ERK/MAPK signal- ing pathways contain- ing genes responsible for organ-, tissue and cellular development. Results 77 2.3.3 Off-target signature After the identification of the TGF-β signature (on-target signature) as well as the affected pathways by GSEA using KEGG and IPA, the NCEs’ off-target effects were identified and mapped to their molecular function and signaling pathways. Each compound treatment resulted in a unique gene expression signature (pheno- copy) of regulated genes. These signatures are composed of the cellular response to two different stimuli (TGF-β and NCE) and are integrated to the corresponding treatment signature. Thereby, elucidating the effects based on NCE treatment is more demanding since both TGF-β and off-target effects occur. Minor effects can also be observed for the interaction of the vehicle (DMSO) with the NCEs. The effects of the different stimuli overlap and also interfere with each other impeding a clear signature dissection. The profile of a given gene may therefore be dependent on which effect prevails and thus, dose-dependency might no longer be observed. In general, all regulated genes can be grouped into six classes, including single and in- tegrated effects. Single effects derive either from the treatment with TGF-β (on- target effect) (Figure 22a) or from the treatment with NCEs (off-target effect). Single off-target effects can be further grouped in pure effects, where the genes are dose- dependently regulated (Figure 22b) and in so called common effects, where genes are dose-independently regulated by each of the seven compounds (Figure 22c). Results 78 Figure 22 – Case profile definition (single effects) NCE treatment and TGF-β stimulation resulted in different case profiles of gene regulation: represent- ative examples for an on-target effect triggered by TGF- β (a), an off-target effect triggered by a NCE (b); common off-target effects induced by all seven NCEs in a dose-independent manner (c). Results 79 Additionally to the single-derived effects, an integration of both TGF-β and off-target effects can be detected: an NCE effect can be additive (Figure 23a) or inverse (Figure 23b) to the effect of TGF-β. Furthermore, opposed bipolar effects for high and low dosage of the NCE mostly linked with toxicity (Figure 23c), are observed. Figure 23 – Case profile definition (integrated effects) NCE treatment and TGF-β stimulation resulted in different cases profiles of gene regulation: repre- sentative examples for integrated effects of on- and off-targets in an additive (c), inverse (d) and bipo- lar (c) manner. Results 80 In a first approximation, the NCE treatment phenotypes (phenocopies) were deter- mined as the total of all regulated genes (p-value < 0.01 and |LR| ≥ 1) comparing NCE treated and TGF-β stimulated cells to DMSO control treated TGF-β stimulated cells. This analysis was done separately for each of the tested compounds at each concentration. Subsequently, the different phenotypes obtained after 2 h NCE treatment were clustered to unravel similarities between the different signatures (Figure 24). The early time point allowed focusing on primary affected genes that were altered as direct response to the treatment. Hierarchical clustering clearly re- vealed two major clusters separating the group of indolinones (BI1 to BI5) from the pyridopyrimidinones (Ex1 & Ex2). The fact that most obviously the specific chemo- type has a major impact on differences in gene expression confirms that the classical notion of chemotypes determining biological profiles of NCEs holds true in this case. However, not only the scaffold itself, but also the specific decoration of each chemo- type affected gene expression. The hierarchical cluster analysis demonstrates that treatment signatures can be used to differentiate even between analogs of the same chemotype. It was possible to clearly distinguish between signatures of BI1 treated cells from the other indolinones. These can be further subdivided into two clusters for either BI2 and BI3 and BI4 and BI5 that resulted in similar treatment effects. In general, treatment with each NCE resulted in clusters for high and low dose of the compound. Results 81 Figure 24 – Treatment signature A hierarchical clustering of all 4,314 significant regulated genes (|LR| ≥ 1 & p-value < 0.01) after NCE treatment and TGF-β stimulation for 2 h in HaCaT cells. The expression patterns of the different NCE- treated cells revealed several intersections in their effects on regulation: the five indolinones (BI1-BI5) are grouped and separated from the two pyridopyrimidinones (Ex1 & Ex2). Expression patterns are grouped in high vs. low dose fractions. The indolinone BI1 separates from the other class members, which can be further divided into two subgroups containing BI2 and BI3 and BI4 and BI5, respectively. Blue indicates decreased expression relative to untreated cells, red indicates increased expression. Results 82 However, the identification of a particular off-target based on this approach is diffi- cult. Further analyses were therefore performed to extract the compounds’ off- target effects from the treatment signatures. As above mentioned not all off-target effects can be identified through dose dependence correlation due to overlapping, inverse and additive effects (Figure 23). Hence, off-targets can only be identified based on NCE treated samples in presence and absence of the TGF-β stimulus. Therefore, all regulated genes (p-value < 0.01 and |LR| ≥ 1) comparing compound treated cells (either 0.08 µM or 2 µM) to DMSO treated controls were selected. Genes were considered once the regulation was observed during compound treat- ment upon TGF-β stimulation as well as without TGF-β stimulation. Thus, it was en- sured to select only drug target and TGF-β independent alterations. All genes that matched the criteria were allocated to the off-target signature of the NCE after 2, 4 and 12 h and are listed in Supplement 3. Based on this analysis, huge differences in the amount of off-target genes were observed. While treatment with BI1 deregu- lated 2,752 genes at all time points, BI3 deregulated “only” 973 genes. Slightly more off-target genes were identified for the indolinones BI2, BI4 and BI5 (1,050, 1,064 and 1,100). Both pyridopyrimidinones regulated 1,347 (Ex1) and 1,306 (Ex2) genes. The largest off-target increase over time was seen for Ex1 and Ex2 with almost four times more genes being regulated comparing the 12 h to the 2 h time point. In con- trast, the amount of off-targets for the five indolinones was at a maximum doubled within this period (Figure 25). In summary, looking at the off-target signatures in general, the indolinones appeared more favorable compared to the pyridopyrimidi- nones at later points in time. Among the indolinones, BI2 to BI5 deregulated fewer genes than BI1 at all points in time which was confirmed by the different kinome Results 83 specificities (see Chapter ‘Kinase Profiling’). It also confirmed the structure-activity relationships described in Roth et al.33 demonstrating that indolinones substituted in position 5 (such as BI1) showed a less favorable selectivity profile compared to indo- linones substituted in position 6 (such as BI2-5). Among the indolinones, BI3 ap- peared to be the most attractive compound when merely looking at the off-target analysis. Figure 25 – Off-target effects (numbers) Every circle represents one of the seven profiled compounds. The size of each circle corresponds to the number of off-target genes (in red). On-target genes numbers are shown in blue. Results 84 2.3.4 Molecular Function Different in silico strategies can be applied to analyze the off-target signatures of the compounds in order to find a conclusion about their effects on the cells. Initially, the genes from the 12 h off-target signatures were assigned to their molecular function using Ingenuity Pathway Analysis (Figure 26). Hierarchical clustering resulted in one major cluster for both pyridopyrimidinones and in one for the indolinones (Figure 26). The genes regulated by the indolinones are distributed in more classes and genes involved in Vitamin and Mineral Metabolism, Cellular Compromise, Nucleic Acid- and Amino Acid Metabolism are exclusively regulated by the indolinones. In accordance with the structure-activity findings mentioned before, within the indoli- none subcluster, BI1 stands apart from the four other indolinones and BI2 and BI3 as well as BI4 and BI5 are grouped in one cluster, respectively. The additionally by BI1 regulated genes are involved in RNA Post-Transcriptional Modification, Energy Pro- duction, Cellular Response to Therapeutics and in RNA Trafficking. Half of the mole- cular functional classes identified are regulated by all compounds. However, this does not necessarily mean that the same genes are regulated since the categories are rather spaciously defined, such as Cell Cycle or Cellular Growth and Proliferation. Furthermore, different NCEs reach far higher significance scores for some categories than others caused by the higher amount of regulated genes in the respective bio- logical process e.g. both pyridopyrimidinones regulate a huge amount of genes in- volved in Cell Death and Cellular Growth and Proliferation. Results 85 Figure 26 – Off-target effects (molecular function) Molecular function of the off-target genes of all seven NCEs after 12 h treatment: clustering of the -log10 [p-values] using complete linkage and manhattan distance depicts the 30 significantly ranked categories. The color code defines the significance determined by Fisher’s exact test as –log10 [p- value]: blue < 2 – not significant; white = 2 – significant & red > 2 – highly significant. Results 86 2.3.5 Pathway Analysis Subsequent to the annotation of the molecular function, Ingenuity Pathway Analysis was used to analyze the off-target signatures in order to enrich influenced signaling pathways to get a more precise idea about the compound effects. We found 39 (2 h), 38 (4 h) and 51 (12 h) canonical signaling pathways scored with a significant –log10 [p-value] > 2 (Fisher’s exact test) for at least one of the NCEs (Figure 27 - Figure 29). While immediate early affected processes can be found 2 h after compound treat- ment, their effect, the treatment phenotype, however manifests during late phases. Caused NCE effects can thus be best observed 12 h after treatment. Hierarchical clustering of the pathway analysis results, again separated the indolinones from the pyridopyrimidinones, indicating that both series share not only a common mode of action like TGF-β inhibition, but also generate a distinct affection of other pathways by their specific off-target function. Again, BI1 stands apart from the four other indo- linones with 5 significantly ranked pathways and the smallest overlap with the other indolinones. BI3 affects 15 signaling pathways and almost exclusively regulates genes involved in different cancer pathways. The indolinones BI2 and BI4 regulated genes that are significantly enriched in only 4 (BI2) and 2 (BI4) signaling pathways, respec- tively. However, pathways such as the Aryl Hydrocarbon Receptor Signaling and the LPS/IL-1 mediated inhibition of RXR function are also significantly ranked high for up to six compounds at all three time points, indicating a more general effect like a xe- nobiotic response to NCE treatment rather than a true compound specific effect. The highest numbers of significantly affected pathways are found for the two pyridopy- rimidinones with 29 (Ex1) and 24 (Ex2). Additionally, genes involved in 30 out of the Results 87 51 signaling pathways are exclusively regulated by Ex1 or Ex2 treatment. In line with the annotation of the molecular function of the off-targets (Figure 26), 13 out of the 51 identified pathways are known mediators of toxicity and cell death. These 13 pathways reach highest significance scores for either Ex1 or Ex2 with 8 being solely affected by the two pyridopyrimidinones indicating a cytotoxic mode of action for both of them. Interestingly, the amount of these pathways enrich over time. Path- ways, such as p53 signaling and VDR/RXR Activation are significantly ranked at all time points and others, such as Death Receptor Signaling, are only found at the later time points (4 h and 12 h). Besides cytotoxicity, these two NCEs deregulate genes involved in inflammatory processes like IL6 signaling, ERK/MAPK signaling and p38 MAPK signaling. In contrast to the findings for the pathways involved in cytotoxicity and cell death, all of these pathways can already be found at the 2 h and 4 h time point, although their signifi- cant scores slightly increase over time. These findings indicate a constant activation of pro-inflammatory processes right from the start of the treatment that are again increased upon later time points (Figure 27 -Figure 30). Results 88 Figure 27 – Ingenuity signaling pathways (2h) All 39 significantly ranked (Fisher’s exact test: –log10 [p-value] > 2; highlighted in red) signaling path- ways from IPA analysis. BI1 4.17 BI1 1.71 BI1 0.96 BI1 0.50 BI2 1.15 BI2 1.61 BI2 1.63 BI2 1.31 BI3 1.05 BI3 2.88 BI3 0.82 BI3 2.18 BI4 1.22 BI4 4.10 BI4 0.94 BI4 1.41 BI5 1.66 BI5 2.31 BI5 0.82 BI5 1.62 Ex1 0.00 Ex1 0.31 Ex1 0.00 Ex1 0.00 Ex2 0.00 Ex2 0.39 Ex2 2.46 Ex2 1.13 BI1 3.83 BI1 1.68 BI1 0.95 BI1 0.47 BI2 4.61 BI2 2.02 BI2 0.75 BI2 1.62 BI3 4.19 BI3 1.30 BI3 0.94 BI3 2.13 BI4 5.65 BI4 2.15 BI4 0.83 BI4 0.56 BI5 4.86 BI5 1.28 BI5 0.61 BI5 0.00 Ex1 1.32 Ex1 2.39 Ex1 2.09 Ex1 2.42 Ex2 3.78 Ex2 4.52 Ex2 1.79 Ex2 1.78 BI1 3.03 BI1 1.66 BI1 0.78 BI1 0.45 BI2 1.82 BI2 2.24 BI2 1.58 BI2 0.46 BI3 2.79 BI3 2.96 BI3 2.08 BI3 0.39 BI4 3.47 BI4 2.42 BI4 1.67 BI4 0.51 BI5 4.29 BI5 4.69 BI5 1.42 BI5 0.00 Ex1 1.32 Ex1 1.91 Ex1 0.85 Ex1 2.13 Ex2 1.15 Ex2 1.64 Ex2 0.97 Ex2 1.71 BI1 2.68 BI1 1.58 BI1 0.75 BI1 0.44 BI2 1.76 BI2 1.72 BI2 1.14 BI2 0.54 BI3 2.85 BI3 0.61 BI3 1.01 BI3 0.68 BI4 3.28 BI4 1.25 BI4 1.22 BI4 0.00 BI5 2.16 BI5 1.53 BI5 1.00 BI5 0.00 Ex1 1.94 Ex1 1.22 Ex1 3.51 Ex1 2.29 Ex2 1.42 Ex2 2.16 Ex2 2.97 Ex2 2.06 BI1 2.51 BI1 1.56 BI1 0.70 BI1 0.00 BI2 1.38 BI2 0.34 BI2 0.91 BI2 0.53 BI3 2.02 BI3 0.60 BI3 1.33 BI3 1.30 BI4 3.05 BI4 1.81 BI4 2.24 BI4 0.36 BI5 2.47 BI5 2.09 BI5 1.32 BI5 0.42 Ex1 1.41 Ex1 0.64 Ex1 1.47 Ex1 2.27 Ex2 0.68 Ex2 2.14 Ex2 1.65 Ex2 0.58 BI1 2.10 BI1 1.47 BI1 0.62 BI1 0.00 BI2 3.85 BI2 1.15 BI2 0.64 BI2 0.00 BI3 1.57 BI3 0.54 BI3 2.50 BI3 0.00 BI4 2.52 BI4 2.71 BI4 2.00 BI4 0.00 BI5 2.14 BI5 1.04 BI5 1.10 BI5 0.62 Ex1 1.23 Ex1 0.00 Ex1 1.02 Ex1 2.71 Ex2 2.17 Ex2 1.94 Ex2 1.14 Ex2 2.31 BI1 2.08 BI1 1.47 BI1 0.59 BI1 0.00 BI2 1.46 BI2 2.40 BI2 1.93 BI2 0.34 BI3 3.03 BI3 0.28 BI3 2.39 BI3 0.61 BI4 3.34 BI4 0.90 BI4 2.76 BI4 0.00 BI5 2.99 BI5 1.44 BI5 1.74 BI5 0.00 Ex1 1.25 Ex1 1.26 Ex1 2.93 Ex1 1.22 Ex2 1.15 Ex2 0.56 Ex2 2.34 Ex2 3.01 BI1 2.04 BI1 1.19 BI1 0.50 BI1 0.00 BI2 2.32 BI2 1.95 BI2 0.78 BI2 0.30 BI3 1.39 BI3 1.74 BI3 0.97 BI3 0.54 BI4 1.58 BI4 2.08 BI4 0.86 BI4 0.33 BI5 0.74 BI5 1.72 BI5 0.38 BI5 0.25 Ex1 1.22 Ex1 0.53 Ex1 2.14 Ex1 1.78 Ex2 1.35 Ex2 0.22 Ex2 0.76 Ex2 2.02 BI1 1.82 BI1 1.00 BI1 0.50 BI1 0.00 BI2 2.59 BI2 1.15 BI2 0.92 BI2 0.00 BI3 2.46 BI3 0.98 BI3 0.84 BI3 0.34 BI4 0.63 BI4 0.86 BI4 0.36 BI4 0.00 BI5 1.41 BI5 1.34 BI5 0.31 BI5 0.90 Ex1 0.82 Ex1 1.56 Ex1 2.30 Ex1 2.41 Ex2 0.89 Ex2 2.50 Ex2 0.59 Ex2 0.63 BI1 1.80 BI1 0.98 BI1 0.50 BI2 1.03 BI2 2.78 BI2 2.55 BI3 1.91 BI3 3.19 BI3 0.96 BI4 2.94 BI4 1.65 BI4 0.53 BI5 3.16 BI5 0.90 BI5 0.63 Ex1 0.20 Ex1 1.75 Ex1 0.61 Ex2 0.69 Ex2 2.80 Ex2 0.76 Hepatic Fibrosis / Hepatic Stellate Cell Activation SAPK/JNK Signaling p38 MAPK Signaling Circadian Rhythm Signaling ATM Signaling VDR/RXR Activation Acute Phase Response Signaling Coagulation System Androgen Signaling ERK5 Signaling Systemic Lupus Erythematosus Signaling G-Protein Coupled Receptor Signaling Huntington's Disease Signaling Notch Signaling IL-12 Signaling and Production in Macrophages IL-8 Signaling Cholecystokinin/Gastrin-mediated Signaling Cell Cycle Regulation by BTG Family Proteins ILK Signaling Hypoxia Signaling in the Cardiovascular Sys. IL-6 Signaling Renal Cell Carcinoma Signaling Endothelin-1 Signaling Molecular Mechanisms of Cancer MIF Regulation of Innate Immunity RAN Signaling HIF1α Signaling RAR Activation Pancreatic Adenocarcinoma Signaling LPS/IL-1 Mediated Inhibition of RXR Function p53 Signaling Bladder Cancer Signaling Cell Cycle: G1/S Checkpoint Regulation Semaphorin Signaling in Neurons Aryl Hydrocarbon Receptor Signaling Xenobiotic Metabolism Signaling PPAR Signaling Leukocyte Extravasation Signaling Neuregulin Signaling Results 89 Figure 28 – Ingenuity signaling pathways (4h) All 38 significantly ranked (Fisher’s exact test: –log10 [p-value] > 2; highlighted in red) signaling path- ways from IPA analysis. BI1 3.96 BI1 0.63 BI1 0.28 BI1 0.00 BI2 0.32 BI2 0.77 BI2 1.66 BI2 0.33 BI3 0.00 BI3 0.00 BI3 2.47 BI3 0.52 BI4 0.00 BI4 0.00 BI4 1.56 BI4 0.43 BI5 0.00 BI5 0.00 BI5 1.93 BI5 0.00 Ex1 0.35 Ex1 4.12 Ex1 0.61 Ex1 2.51 Ex2 0.87 Ex2 2.09 Ex2 0.54 Ex2 1.72 BI1 2.69 BI1 0.57 BI1 0.26 BI1 0.00 BI2 5.46 BI2 1.69 BI2 0.00 BI2 2.39 BI3 5.31 BI3 0.47 BI3 0.28 BI3 2.96 BI4 5.51 BI4 1.92 BI4 0.00 BI4 2.70 BI5 4.56 BI5 2.14 BI5 0.00 BI5 2.01 Ex1 2.92 Ex1 1.00 Ex1 2.54 Ex1 0.92 Ex2 2.72 Ex2 0.95 Ex2 1.06 Ex2 0.00 BI1 2.34 BI1 0.51 BI1 0.22 BI1 0.00 BI2 2.38 BI2 4.03 BI2 0.38 BI2 0.00 BI3 2.37 BI3 5.58 BI3 0.50 BI3 0.55 BI4 2.81 BI4 5.61 BI4 0.45 BI4 0.00 BI5 2.43 BI5 4.17 BI5 0.00 BI5 0.00 Ex1 1.90 Ex1 1.14 Ex1 1.91 Ex1 2.61 Ex2 1.78 Ex2 1.42 Ex2 2.76 Ex2 1.22 BI1 2.22 BI1 0.49 BI1 0.00 BI1 0.00 BI2 0.00 BI2 1.52 BI2 1.21 BI2 0.36 BI3 0.00 BI3 0.71 BI3 0.56 BI3 1.08 BI4 0.00 BI4 1.80 BI4 0.46 BI4 0.92 BI5 0.00 BI5 2.08 BI5 0.57 BI5 1.11 Ex1 0.00 Ex1 1.66 Ex1 2.63 Ex1 2.66 Ex2 0.00 Ex2 3.00 Ex2 1.82 Ex2 1.25 BI1 1.35 BI1 0.47 BI1 0.00 BI1 0.00 BI2 1.19 BI2 2.31 BI2 0.47 BI2 0.72 BI3 0.55 BI3 5.57 BI3 1.75 BI3 3.19 BI4 0.46 BI4 4.77 BI4 0.65 BI4 0.90 BI5 0.57 BI5 3.67 BI5 0.00 BI5 1.09 Ex1 1.32 Ex1 0.97 Ex1 0.88 Ex1 3.36 Ex2 2.45 Ex2 1.17 Ex2 2.06 Ex2 1.80 BI1 1.20 BI1 0.00 BI1 0.00 BI1 0.00 BI2 0.00 BI2 0.00 BI2 0.74 BI2 1.08 BI3 0.34 BI3 0.24 BI3 1.16 BI3 0.97 BI4 0.00 BI4 0.00 BI4 2.01 BI4 1.34 BI5 0.00 BI5 0.00 BI5 1.20 BI5 1.00 Ex1 2.08 Ex1 3.06 Ex1 1.81 Ex1 3.13 Ex2 1.26 Ex2 2.16 Ex2 1.66 Ex2 1.65 BI1 1.15 BI1 0.46 BI1 0.00 BI1 0.00 BI2 0.66 BI2 1.79 BI2 0.76 BI2 0.34 BI3 0.25 BI3 2.23 BI3 0.71 BI3 1.03 BI4 0.00 BI4 1.00 BI4 2.04 BI4 1.42 BI5 0.00 BI5 1.78 BI5 1.78 BI5 1.06 Ex1 2.19 Ex1 3.06 Ex1 3.07 Ex1 3.30 Ex2 2.35 Ex2 1.43 Ex2 3.53 Ex2 2.40 BI1 0.85 BI1 0.40 BI1 0.00 BI1 0.00 BI2 0.95 BI2 0.00 BI2 0.42 BI2 0.00 BI3 0.25 BI3 0.00 BI3 0.22 BI3 0.79 BI4 0.60 BI4 0.00 BI4 0.53 BI4 0.68 BI5 0.72 BI5 0.00 BI5 0.64 BI5 0.30 Ex1 3.16 Ex1 2.66 Ex1 0.92 Ex1 2.60 Ex2 2.23 Ex2 0.60 Ex2 2.01 Ex2 2.47 BI1 0.73 BI1 0.35 BI1 0.00 BI2 0.59 BI2 2.91 BI2 0.27 BI3 0.90 BI3 2.79 BI3 0.23 BI4 0.76 BI4 2.36 BI4 0.69 BI5 0.93 BI5 2.26 BI5 0.00 Ex1 2.95 Ex1 0.00 Ex1 1.82 Ex2 4.24 Ex2 0.00 Ex2 2.16 BI1 0.64 BI1 0.29 BI1 0.00 BI2 1.84 BI2 1.10 BI2 0.23 BI3 1.10 BI3 1.48 BI3 0.90 BI4 1.52 BI4 1.31 BI4 0.29 BI5 1.79 BI5 1.52 BI5 0.35 Ex1 2.01 Ex1 2.67 Ex1 2.08 Ex2 1.28 Ex2 0.59 Ex2 0.67 Pancreatic Adenocarcinoma Signaling TR/RXR Activation Cell Cycle: G2/M DNA Damage Checkpoint Reg. Aryl Hydrocarbon Receptor Signaling Bladder Cancer Signaling Mitotic Roles of Polo-Like Kinase p53 Signaling ATM Signaling Colorectal Cancer Metastasis Signaling Endoplasmic Reticulum Stress Pathway Molecular Mechanisms of Cancer Small Cell Lung Cancer Signaling IL-8 Signaling CD27 Signaling in Lymphocytes Leukocyte Extravasation Signaling Death Receptor Signaling RAR Activation ERK5 Signaling Circadian Rhythm Signaling Neuregulin Signaling LPS/IL-1 Mediated Inhibition of RXR Function LXR/RXR Activation Xenobiotic Metabolism Signaling IL-6 Signaling IL-10 Signaling Prolactin Signaling Coagulation System ILK Signaling HMGB1 Signaling Retinoic acid Mediated Apoptosis Signaling SAPK/JNK Signaling PPAR Signaling IGF-1 Signaling p38 MAPK Signaling NRF2-mediated Oxidative Stress Response Hepatic Cholestasis Hepatic Fibrosis / Hepatic Stellate Cell Activation VDR/RXR Activation Results 90 Figure 29 – Ingenuity signaling pathways (12h) – Part I All 52 significantly ranked (Fisher’s exact test: –log10 [p-value] > 2; highlighted in red) signaling path- ways from IPA analysis. BI1 2.36 BI1 0.86 BI1 0.46 BI1 0.00 BI2 0.56 BI2 0.92 BI2 0.00 BI2 3.43 BI3 1.70 BI3 2.86 BI3 0.78 BI3 1.15 BI4 1.29 BI4 1.09 BI4 0.49 BI4 3.08 BI5 3.10 BI5 2.06 BI5 0.42 BI5 3.39 Ex1 2.55 Ex1 1.64 Ex1 2.91 Ex1 2.77 Ex2 3.24 Ex2 2.18 Ex2 1.80 Ex2 4.37 BI1 2.25 BI1 0.86 BI1 0.44 BI1 0.00 BI2 0.26 BI2 0.00 BI2 5.37 BI2 1.20 BI3 0.25 BI3 3.19 BI3 7.18 BI3 1.87 BI4 0.00 BI4 0.78 BI4 5.80 BI4 0.90 BI5 0.00 BI5 0.34 BI5 7.95 BI5 1.35 Ex1 0.00 Ex1 1.99 Ex1 4.92 Ex1 3.56 Ex2 0.00 Ex2 1.58 Ex2 6.41 Ex2 2.42 BI1 2.25 BI1 0.80 BI1 0.43 BI1 0.00 BI2 0.44 BI2 1.00 BI2 0.71 BI2 0.59 BI3 1.13 BI3 2.18 BI3 0.25 BI3 0.59 BI4 1.64 BI4 0.78 BI4 0.00 BI4 0.42 BI5 2.38 BI5 3.13 BI5 0.47 BI5 0.37 Ex1 0.86 Ex1 2.30 Ex1 1.07 Ex1 2.70 Ex2 2.11 Ex2 1.84 Ex2 2.72 Ex2 1.71 BI1 2.07 BI1 0.79 BI1 0.40 BI1 0.00 BI2 0.34 BI2 0.93 BI2 0.00 BI2 0.53 BI3 0.34 BI3 3.50 BI3 0.43 BI3 0.90 BI4 0.71 BI4 0.26 BI4 0.89 BI4 0.32 BI5 0.00 BI5 2.77 BI5 1.52 BI5 0.83 Ex1 0.00 Ex1 0.29 Ex1 2.72 Ex1 1.00 Ex2 0.30 Ex2 1.09 Ex2 3.60 Ex2 2.14 BI1 2.04 BI1 0.78 BI1 0.38 BI1 0.00 BI2 0.28 BI2 1.18 BI2 0.00 BI2 0.59 BI3 1.42 BI3 2.62 BI3 0.00 BI3 0.28 BI4 0.58 BI4 0.89 BI4 0.00 BI4 0.66 BI5 1.03 BI5 1.94 BI5 0.34 BI5 0.32 Ex1 1.73 Ex1 1.30 Ex1 3.06 Ex1 1.88 Ex2 1.27 Ex2 0.59 Ex2 3.79 Ex2 3.37 BI1 1.35 BI1 0.71 BI1 0.35 BI1 0.00 BI2 0.38 BI2 0.73 BI2 0.40 BI2 0.72 BI3 1.00 BI3 1.20 BI3 2.81 BI3 2.89 BI4 1.45 BI4 0.84 BI4 1.53 BI4 1.63 BI5 2.97 BI5 0.75 BI5 1.44 BI5 2.18 Ex1 1.18 Ex1 1.96 Ex1 0.77 Ex1 3.32 Ex2 1.22 Ex2 3.57 Ex2 0.80 Ex2 1.61 BI1 1.21 BI1 0.70 BI1 0.35 BI1 0.00 BI2 1.08 BI2 0.00 BI2 0.00 BI2 0.34 BI3 1.08 BI3 0.30 BI3 2.31 BI3 2.47 BI4 0.40 BI4 0.00 BI4 0.74 BI4 0.69 BI5 0.73 BI5 0.00 BI5 1.12 BI5 0.64 Ex1 2.05 Ex1 2.47 Ex1 0.69 Ex1 1.00 Ex2 3.92 Ex2 1.71 Ex2 0.44 Ex2 0.29 BI1 1.19 BI1 0.57 BI1 0.33 BI1 0.00 BI2 1.07 BI2 0.88 BI2 0.35 BI2 0.51 BI3 1.07 BI3 0.88 BI3 0.00 BI3 2.25 BI4 1.29 BI4 1.55 BI4 0.00 BI4 1.18 BI5 1.75 BI5 1.98 BI5 0.42 BI5 1.08 Ex1 1.54 Ex1 0.31 Ex1 1.92 Ex1 1.36 Ex2 2.10 Ex2 3.71 Ex2 2.01 Ex2 1.03 BI1 1.14 BI1 0.57 BI1 0.32 BI1 0.00 BI2 0.00 BI2 0.00 BI2 0.00 BI2 1.49 BI3 0.00 BI3 0.00 BI3 0.51 BI3 1.08 BI4 0.00 BI4 0.00 BI4 0.35 BI4 1.66 BI5 0.00 BI5 0.00 BI5 0.31 BI5 1.47 Ex1 2.65 Ex1 0.82 Ex1 2.84 Ex1 4.21 Ex2 0.97 Ex2 2.08 Ex2 2.39 Ex2 3.51 BI1 1.13 BI1 0.56 BI1 0.25 BI1 0.00 BI2 1.65 BI2 0.00 BI2 0.72 BI2 0.57 BI3 2.50 BI3 0.49 BI3 0.25 BI3 0.57 BI4 0.71 BI4 0.00 BI4 0.53 BI4 0.40 BI5 2.73 BI5 0.25 BI5 0.48 BI5 0.36 Ex1 0.60 Ex1 4.72 Ex1 3.32 Ex1 2.08 Ex2 1.06 Ex2 2.84 Ex2 1.61 Ex2 3.32 p38 MAPK Signaling LXR/RXR Activation Endometrial Cancer Signaling Glioma Signaling Xenobiotic Metabolism Signaling PPAR Signaling Chronic Myeloid Leukemia Signaling Hepatic Cholestasis IL-6 Signaling Germ Cell-Sertoli Cell Junction Signaling IL-12 Signaling and Production in Macrophages ILK Signaling LPS/IL-1 Mediated Inhibition of RXR Function VDR/RXR Activation FXR/RXR Activation RAR Activation ERK5 Signaling Prostate Cancer Signaling Neuregulin Signaling IGF-1 Signaling Endoplasmic Reticulum Stress Pathway ATM Signaling B Cell Receptor Signaling 14-3-3-mediated Signaling Role of Oct4 in Mam. Embr.Stem Cell Pluripotency Hepatic Fibrosis / Hepatic Stellate Cell Activation Melanoma Signaling Small Cell Lung Cancer Signaling p53 Signaling Pancreatic Adenocarcinoma Signaling ERK/MAPK Signaling FLT3 Signaling in Hematopoietic Progenitor Cells Bladder Cancer Signaling Aryl Hydrocarbon Receptor Signaling Cell Cycle: G2/M DNA Damage Checkpoint Regulation Molecular Mechanisms of Cancer Agrin Interactions at Neuromuscular Junction Amyloid Processing Role of BRCA1 in DNA Damage Response MIF Regulation of Innate Immunity Results 91 Continued: Figure 29 – Ingenuity signaling pathways (12) – Part II All 52 significantly ranked (Fisher’s exact test: –log10 [p-value] > 2; highlighted in red) signaling path- ways from IPA analysis. BI1 1.04 BI1 0.50 BI1 0.00 BI1 0.89 BI2 2.42 BI2 0.48 BI2 0.29 BI2 0.47 BI3 6.63 BI3 1.21 BI3 0.29 BI3 0.47 BI4 1.96 BI4 0.37 BI4 0.21 BI4 0.28 BI5 4.40 BI5 1.68 BI5 0.00 BI5 0.00 Ex1 2.65 Ex1 2.26 Ex1 2.38 Ex1 1.81 Ex2 2.72 Ex2 1.61 Ex2 3.12 Ex2 2.74 BI1 0.93 BI1 0.48 BI1 0.00 BI1 0.47 BI2 0.30 BI2 1.19 BI2 1.68 BI2 0.34 BI3 0.00 BI3 2.09 BI3 1.18 BI3 0.34 BI4 1.18 BI4 0.97 BI4 0.25 BI4 0.25 BI5 1.10 BI5 0.91 BI5 0.00 BI5 0.23 Ex1 3.13 Ex1 0.94 Ex1 2.44 Ex1 2.79 Ex2 1.38 Ex2 0.49 Ex2 2.15 Ex2 2.19 BI1 0.91 BI1 0.48 BI1 0.00 BI1 0.00 BI2 2.16 BI2 0.00 BI2 0.55 BI2 0.00 BI3 2.16 BI3 0.00 BI3 0.00 BI3 0.00 BI4 1.01 BI4 0.00 BI4 1.26 BI4 0.00 BI5 5.89 BI5 0.62 BI5 0.34 BI5 0.00 Ex1 0.51 Ex1 1.32 Ex1 1.97 Ex1 1.49 Ex2 1.65 Ex2 2.33 Ex2 2.04 Ex2 2.05 IL-10 Signaling Leukocyte Extravasation Signaling Amyotrophic Lateral Sclerosis Signaling NRF2-mediated Oxidative Stress Response Role of CHK Proteins in Cell Cycle Checkpoint Ctrl Ephrin Receptor Signaling Cell Cycle: G1/S Checkpoint Regulation Death Receptor Signaling Coagulation System Docosahexaenoic Acid (DHA) Signaling Cholecystokinin/Gastrin-mediated Signaling CD27 Signaling in Lymphocytes Results 92 Figure 30 – Off-target effects (pathway analysis) Ingenuity pathway analysis for the off-target genes of all seven NCEs after 12 hours: clustering of the -log10 [p-values] using complete linkage and manhattan distance depicts the 51 significantly ranked canonical signaling pathways. Off-target genes deregulated by BI3 treatment affect almost exclusively 10 cancer signaling pathways (red arrows). Ex1 & Ex2 off-target genes play a role in 12 pathways in- volved in cytotoxicity or cell death (grey arrows) and in 5 pathways involved in inflammation (green arrows). The color code defines the significance determined by Fisher’s exact test as –log10 [p-value]: blue < 2 – not significant; white = 2 – significant & red > 2 – highly significant. Results 93 2.3.6 Wet Lab Validation Results from the pathway analysis strongly implied different induced phenotypes after treatment with specific NCEs. However, pathway analysis tools only generate hypotheses and their proof of biological relevance must be verified. To address the accuracy of the pathway analysis, it was aimed to confirm the in silico generated hy- potheses by experimental laboratory data. 2.3.6.1 Cytotoxicity and Cell Death According to the expression data, both pyridopyrimidinones (Ex1 & Ex2) are involved in processes such as cell death and inflammation. To investigate several cytotoxicity parameters high content screen analysis using the high-capacity automated fluores- cence imaging platform from Cellomics was performed. HaCaT cells were incubated with increasing compound concentrations (3.2 nM – 50 µM) for 24 h. Subsequently, the cells were stained with cytotoxicity cocktails and images were acquired and ana- lyzed on the Cellomics ArrayScan II. Cells were stained using i) Hoechst DNA dye to count cell density and investigate nuclear condensation and fragmentation, ii) Lyso- Tracker Red to analyze the amount of lysosomes per cell as an early marker for cyto- toxicity, iii) Sytox Green as a membrane impermeable dye to detect loss of mem- brane integrity as late event for cytotoxicity. Example pictures of healthy and dying cells regarding nuclear fragmentation, membrane permeability and lysosomal mass per cell are given in Figure 31. Results 94 Figure 31 – Cytotoxicity parameters Shown are examples for healthy in contrast to dying cells after NCE treatment regarding the meas- ured parameters for nuclear fragmentation, membrane permeability and lysosomal mass per cell. All four analyzed parameters were normalized to DMSO treated (vehicle) cells as negative control (0 %) and Valinomycin and Chloroquine treated cells as positive controls (100 %). Exemplary, the images of Ex1, BI5 and control treated cells are de- picted in Figure 32. According to pathway analysis the highest toxicity is observed by treatment with Ex1 and Ex2. Cell density is strongly decreased to less than 10 % of control. Nuclear fragmentation, lysosomal mass per cell and membrane permeability are increased by 100 % for both pyridopyrimidinones and even lower concentrations of Ex2 were sufficient to raise the membrane permeability of more than 50 % of the control. Treatment with the indolinones resulted in mild toxicity effects for the treatment with BI1, BI2 and BI3 at high concentrations and almost no toxicity for BI4 and BI5 (Figure 33). Nuclear fragmentation Membrane permeability Lysosomal mass per cell C yt o to xi ci ty N o c yt o to xi ci ty Results 95 Figure 32 – NCE induced cytotoxicity I Cellomics high content screens analyzing cytotoxic parameters were performed to test for NCE-induced cytotoxicity. High content screen images of HaCaT cells treated with increasing concentrations (3.2 nM – 50 µM) of Ex1 and BI5 including DMSO (negative control) or 100 µM Valinomycin and 10 µM Chloroquine (posi- tive control) for 24 h. Subsequent staining with Hoechst dye, Sytox Green and Lysotracker Red deter- mined cell density, nuclear fragmentation, permeabili- ty and lysosomal mass per cell illustrated a strong de- crease of cell density and an increase of nuclear frag- mentation and lysosomal mass for Ex1 but not for BI5 treated cells. NCE concentration [µM] 0.0032 0.016 0.08 0.4 2 10 50 Vehicle Val. Chloro. C o n tr o ls Ly so tr ac ke r S yt o x G re e n H o e ch st Ex 1 Ly so tr ac ke r S yt o x G re e n H o e ch st B I5 Ly so tr ac ke r S yt o x G re e n H o e ch st Results 96 Figure 33 – NCE induced cytotoxicity II Results of dose response experiments for all seven NCEs for cell density (blue), nuclear fragmentation (grey), permeability (green) and lysosomal mass (red) obtained from Cellomics high content screen analysis. Values of NCE treated cells are compared to DMSO treated cells and shown as percent of control (POC). Outlier data points are shown as filled red circles. Nuclear fragmentation Membrane permeability Lysosomal mass per cell Cell density Results 97 To analyze the mode of cell death, Caspase-3 activation assays were performed to distinguish between apoptosis and necrosis, since activation of this executioner cas- pase is a clear marker for apoptotic cell death. Again, HaCaT cells were treated with 2 µM of each compound for 24 h and subsequently Caspase-3 activation was de- tected using a Caspase-3 Detection Kit (Calbiochem) and quantified by flow cytome- try. Among all NCEs tested, only treatment with Ex1 resulted in an activation of Cas- pase-3 with approximately 30 % positive cells (Figure 34b). This is in line with the results of the pathway analysis, in which only the off-target signature of Ex1 ex- ceeded the significance threshold for Death Receptor Signaling after 4 h (Figure 28, Figure 34a) and was further increased after 12 h (Figure 29, Figure 34a). Results 98 Figure 34 – Death Receptor Signaling I a: Ingenuity pathway analysis results for Death Receptor Signaling obtained for the NCE off-targets. Only Ex1 treatment for 4 h and 12 h achieves the significance threshold of –log10 [p-value] > 2 (Fish- er’s exact test; red). b: HaCaT cells were treated with 2 µM of each NCE and incubated for 24h. Cas- pase-3 activation was analyzed. A significant signal was identified only after treatment with Ex1 (30 %) or with Valinomycin (15 %) and Chloroquine (100 %). An overlay of the off-targets of Ex1 with the canonical Death Receptor Signaling pathway revealed that six genes are deregulated after 12 h of treatment (Figure 35). NFκB1a, a subunit of the Iκκ-complex, is down-regulated, while five pro-apoptotic genes, such as Apaf1 as part of the apoptosome, the death receptors DR3 and 6, the death receptor ligand APO2L and the initiator Caspase-9, are up-regulated. Taken together, these findings clearly demonstrate that it was not only possible to predict the compound’s cytotoxicity based on mRNA profiles but also its apoptotic mode of action. Results 99 Figure 35 – Death Receptor Signaling II Canonical Death Receptor Signaling pathway (Ingenuity Pathway Analysis) overlaid with the off-target of Ex1. Color code: red – up-regulated gene; green – down-regulated gene. 2.3.6.2 Inflammation Pathway analysis not only labeled the pyridopyrimidinones for cell death induction, but also for affection of inflammatory mechanisms. To proof this prediction, HaCaT cells treated with each compound were analyzed for the induction of pro- inflammatory cytokines (IL1β, TNFα, IL8 and IL6). While no significant alteration in release of IL1β, TNFα and IL8 was observed, the IL6 levels were dose-dependently increased after treatment with the pyridopyrimidinones Ex1 and Ex2. Compared to DMSO treated control cells, 10 µM of Ex2 increased IL6 secretion by factor 5 and Ex1 treatment at the same concentration even resulted in a 25-fold increase (Figure 36). Results 100 Figure 36 – IL6 release HaCaT cells were treated with increasing NCE concentrations and DMSO for 12 h. IL6 levels were app. 25-fold increased after Ex1 treatment and 5-fold increased after Ex2 treatment compared to DMSO treated cells. Student t-test was used to calculate the significance compared to DMSO treated cells (*< 0.01 & **<0.001). All error bars indicate the standard deviation of n=3. 2.3.7 Kinase Profiling One pivotal issue of kinase inhibitors is cross-reactivity with other kinases which may contribute one source of off-targets. In vitro kinase profiling is the state-of-the-art method to examine selectivity of kinase inhibitors. The Ingenuity Pathway Analysis database was used to extract the literature described downstream targets for 239 in HaCaT cells expressed kinases. For 147 kinases 807 non-redundant downstream tar- gets had been described and annotated. The downstream targets were used as sur- rogate markers and overlaid with the NCE’s off-target lists to assign off-target genes to off-target kinases (Figure 38). To proof the predictivity of our model all com- pounds were tested against 239 kinases available in the biochemical SelectScreenTM Kinase Profiling (Invitrogen) at concentrations of 2 µM and 200 nM. All kinases inhi- bited by at least 90 % at 2 µM conc. and additionally by at least 50 % at 200 nM were selected as off-target kinases (Table 11). Most off-target kinases were identified for Results 101 BI1 (75) and Ex1 (60). All other NCEs showed a higher selectivity with only 21 (BI2), 17 (BI3), 12 (BI4), 15 (BI5) and 14 (Ex2) kinases inhibited by the respective compound (Figure 37b, Table 11). For BI1 84 of the 366 known surrogate marker genes were found to be regulated. A comparably good ratio was also identified for the two pyri- dopyrimidinones with 84 out of 361 (Ex1) and 55 out of 223 (Ex2). A summary of the data is shown in Figure 37. Integrating all criteria, the identification of kinase selec- tivities was limited, as shown for BI2 in Figure 38. Although, various annotated sur- rogate marker genes were identified as off-targets, fortifying the used identification procedure, a clear association to a specifically upstream-acting kinase was often not possible since too many surrogate markers were redundantly identified to act down- stream of several receptor kinases. The cellular system might be optimized e.g. in regard to the addition of receptor ligands, but it will not replace testing NCEs in bio- chemical assays for kinase selectivity. Results 102 Figure 37 – Kinase Selectivity I a: each compound was profiled against a panel of 239 protein kinases and the number of kinases inhibited by each compound is shown (Table 11). No surrogate marker (e.g. literature known down- stream target of a given kinase) were identified for 92 enzymes, whereas 807 surrogate markers were identified for 147 of the enzymes. Based on the kinase inhibition profile of each compound, the ex- pression of these kinases in HaCaT cells and the availability of surrogate markers the number of po- tentially effected surrogate marker genes was predicted. b: human kinome dendrograms showing the NCEs’ kinase specificity profiles. Circle size corresponds to the precentaged inhibition of the kinase at 200 nM concentration. AGC – Containing PKA, PKG, PKC families; CAMK – Calcium/calmodulin- dependent protein kinase; CK1 – Casein kinase 1; CMGC – Containing CDK, MAPK, GSK3, CLK families; STE – Homologs of yeast Sterile 7, Sterile 11, Sterile 20 kinases; TK – Tyrosine kinase; TKL – Tyrosine kinase-like. Kinome dendrograms used for visualization were shown with permission from Cell Signal- ing Technology, Inc. (http://www.cellsignal.com). Results 103 Figure 38 – Kinase Selectivity II Projections of BI2-specifically regulated kinase surrogate marker genes. Indicated are all 21 inhibited kinases: 3 kinases with no affection of known surrogate marker genes (c’), 4 kinases of which no sur- rogate markers are described (c’’) and 14 kinases with deregulated surrogate markers genes (blue line = transcriptional down-regulation, red line = transcriptional up-regulation, red box = in silico predicted BI2-specific kinase hits). Results 104 Table 11 – Kinase Selectivity Ph.D. Thesis Patrick Baum 3. Discussion Phenocopy – A Strategy to Qualify Chemical Compounds during Hit-to-Lead and Lead Optimization Discussion 106 Since the approval of Imatinib (Gleevec) in 2001, the first marketed kinase inhibitor, many additional kinase inhibitors have been advanced into clinical development. The most advanced kinase programs in research and development are aimed at the treatment of various cancers. However, additional therapeutic applications like im- munological, metabolic- or infectious diseases and also the treatment of central nervous system disorders by kinase inhibitors are under investigation31, 111, 112. Dur- ing the optimization of kinase inhibitors one often has to cope with challenges like the improvement of kinase selectivity113, 114. In combination with the overall high attrition rates of new drug candidates53 there is a need for new strategies that sup- port and optimize the drug discovery process. In the present study, an approach is introduced to support the drug development process by enabling a highly flexible and highly resolved ranking of the NCEs and by providing additional information of the drug target biology. 3.1 NCE Ranking So far, the in vitro biological evaluation of NCEs has often been based on biochemical and cellular potencies as well as on the selectivity of the respective NCE. This limited view may result in wrong decisions for further time and cost consuming processes, such as in vivo experiments. In the present study, a workflow was established to alle- viate the lead identification and optimization of NCEs in general and kinase inhibitors in particular by elucidating the mechanism of action of both the target and the NCE. Thereby, more knowledge about drug candidates is obtained at an early stage of drug discovery and several new categories for their qualification are available (Figure Discussion 107 39). Although the rating of each category remains subjective, one can clearly distin- guish between good and bad results within a group of tested compounds. Especially the exclusion criteria (shown in red; Figure 39) are important breakpoints for the ranking since they are potential sources for failures of the respective NCE later on in the process. Figure 39 – NCE ranking Quality parameters used to gauge the seven NCEs. The Phenocopy strategy introduces ten additional parameters dealing with potency, off‐target numbers and affected pathways. NCEs are ranked from blue (good) to red (bad). Integration of all parameter scores identifies BI4 and BI2 as superior to all other NCEs. In case of the tested inhibitors the decision based on classical parameters such as potency or kinase specificity can be misleading or fall far short. Besides Ex2 all other NCEs are potent inhibitors with comparable IC50 values in cellular assays and only minor distinction in their biochemical activity. Also the kinome specificity profiling is not sufficient to arrive at a final decision, since only BI1 and Ex1 inhibit a wide panel BI4 BI5 BI1 BI2 BI3 Ex1 Ex2 + - Discussion 108 of other kinases. Furthermore, the Phenocopy data strongly indicates that off-target effects do not only derive from additionally inhibited kinases in line with the fact that within the human genome over 2,000 other nucleotide-dependent enzymes can be found115 which may be potentially affected by NCEs blocking an ATP binding site. In addition, identification of bioactive compounds revealed a high degree of promiscui- ty for kinases inhibitors with GPCRs and phosphodiesterases116. Hence, an approach using the Phenocopy strategy will deliver a wider view on the NCEs’ selectivity. How- ever, it can not replace a kinase selectivity screen, since mapping of the off-targets to published kinase surrogate markers were not sufficient to indentify all off-target kinases or did not result in unique predictions (Figure 38). This is potentially due to the experimental design and the current state of knowledge. The correlation of the in vitro predicted kinase inhibition with the off-target effects requires some criteria to be fulfilled: i) the kinase has to be expressed in the cellular system, ii) the signaling pathway must be functional, which iii) depends on the availability of appropriate ligands in the in vitro system. As the cells in our study had been starved for com- pound and/or TGF-β profiling, these criteria might have been only partially met. Fi- nally, iv) surrogate markers have to be described. Nevertheless, the comparison of the compounds’ off-target effects with the kinase profiling fortifies the Phenocopy findings and supports the chosen analysis procedure since a good overlap between surrogate markers and off-targets is found in many cases, where sufficient surrogate markers are published. By evaluating of TGF-βR1 inhibitors, it was possible to clearly differentiate the indoli- none chemotype from the pyridopyrimidinones in several parameters. Furthermore, even within the indolinone cluster differences between compounds with various Discussion 109 decorations were identified. This can help at different steps during the drug devel- opment process. The differentiation between chemotypes helps to qualify the dif- ferent hit classes after high-throughput screening that are promoted to the hit-to- lead phases. Later on, the differentiation between the compounds of a particular chemotype is suitable to support lead optimization programs. As above mentioned, the presented approach enables a wider view on the com- pounds selectivity in a cellular context by the identification of the total amount of target-independent regulated genes and also sheds light on the chronological se- quence when these regulations occurs. In earlier studies, compound-dependent gene regulation is either identified by treatment of the cells with only one concen- tration in comparison to untreated cells followed by the application of fold change and/or p-value cut-offs57 or by the treatment with a series of concentrations, fol- lowed by the selection of dose-dependently regulated genes70. The Phenocopy data however demonstrates that on the one hand, the use of only one concentration is not sufficient to identify all off-targets and might be too less stringent dependent on the used concentration. On the other hand, an off-target identification based on dose-dependency will only select that subgroup of regulated genes that are not sub- jected to integrated effects. Thus, the introduction of a selection procedure where different NCE concentrations are used in presents as well as in absence of TGF-β guarantees a sharp classification and enables a high resolution of detected genes. The strategy can also help to interpret the identified off-target effects and offers the possibility to assign the regulated genes to relevant biological processes and net- works. Good results that were reproduced by wet lab experiments were thereby achieved using Ingenuity Pathway Analysis (Ingenuity Systems®). However, the anno- Discussion 110 tations of canonical pathways are superior to the analysis of the molecular function. Although the annotation of the molecular function of the regulated genes can help to understand the compound effects, e.g. in case of the predicted cytotoxicity of both pyridopyrimidinones, its informative value is limited since some of the catego- ries are too general. Thus, it can not be consulted for the NCE ranking. In contrast, the pathway analysis can be used for the ranking in a flexible format by the respec- tive scientist by defining context relevant processes or just by prioritizing the com- pounds in terms of the absolute number of affected processes or pathways. In terms of TGF-β inhibition for instance one goal is to reduce inflammation processes trig- gered by this cytokine89 as one potential strategy to treat fibrotic diseases or cancer. Thus, the predicted and confirmed pro-inflammatory properties of both pyridopyri- midinones (Figure 30, Figure 36) are the opposite of the desired effect making both NCEs inferior to the indolinones. Furthermore, relevant and unwanted processes are regulation of growth and proliferation and obviously induction of cell death (Figure 30). Finally, the combination of TGF-β and off-target signatures revealed that some com- pounds regulate genes inverse to the desired therapeutic effect (Figure 23). This can potentially affect the efficacy of the treatment with this NCE. This is of particular interest since three times more substances undergo attrition due to a lack of efficacy than due to clinical safety reasons53. Especially both pyridopyrimidinones regulated 217 (Ex1) and 317 (Ex2) TGF-β dependent genes in the opposite direction to the de- sirable treatment effect. In contrast, the indolinones only affected a lower number (BI3: 42; BI2: 57, BI1: 70; BI5: 81; BI4: 101) of these genes. Discussion 111 According to the ten introduced Phenocopy criteria, a couple of compounds revealed liabilities through downstream inhibition of PAI-1 transcription (BI1, BI5 & Ex2), at the regulation of off-target genes per se (BI1, Ex1 & Ex2), at the affection of inverse TGF-β signaling (Ex1 & Ex2), at the induction of cell death (Ex1), at acting as pro- inflammatory stimuli (Ex1 & Ex2) and as promoting cellular growth and induction of cancer pathways (BI3 & BI5) (Fig.6). An integration of all obtained information re- commends the use of BI4 and BI2 for further optimization due to superior overall performance of these two drug candidates (Figure 39). 3.2 Chemical genomic profiling So far, microarray technology has been successfully applied during the drug devel- opment process for target discovery by profiling disease models, for target validation by profiling alterations caused by disease-related genes, for elucidating drug meta- bolism by measuring transcriptional changes of known drug metabolizing genes in rat livers or human hepatocytes and to address drug safety in toxicogenomics ap- proaches. However, only few approaches have been tempted to fill the gap between target validation and drug metabolism and aimed to support the hit-to-lead or lead optimization processes. In fact, gene expression signatures have been used to func- tionally annotate and characterize small molecules in yeast and in mammalian cells57, 117, 118. However, these approaches mainly focused on the identification of new NCEs directed against a given target, or to build novel connections to a disease, but not to obtain an in depth analysis of the off-target effects. In the present study, several optimized parameters are introduced to achieve a comprehensive qualification of Discussion 112 the investigated compounds. First, the screening platform was chosen by the use of a relevant cellular system functionally expressing the drug target and its downstream signaling. Second, various time points and concentrations were monitored. Third, siRNAs against TGF-βR1 were used as an additional target modulation technology to confirm the results obtained with the NCEs. By combining those data it was demon- strated that the off-target signatures can be used to identify the most selective NCE among the tested compounds and to detect unwanted off-target effects such as in- duction of pro-inflammatory processes or of death receptor signaling. The data also allows identifying the target promiscuity of the NCE. These polypharmacological ap- proaches, most notably discussed in fields of cancer treatment32, can not be faced with conventional single target-based assays but need approaches containing multi- parallel readouts for NCE characterization. Such a strategy will also help to accumu- late an iterative knowledge about both the drug candidates itself and the structural classes. The drug candidates’ off-target signatures can be overlaid with other data- bases containing drug-dependent gene signatures like the Connectivity Map57. Inte- gration of additional data sources will further characterize the NCEs by flagging them for potential side effects and the identification of desirable pharmacology profiles or even find a repositioning idea for other indications. Nevertheless, there are some limitations inherently comprised in each chemical ge- nomics approach. Most obviously, cells that grow on plastic in a laboratory are very different from tissues in a whole organism. Therefore, effects modulated by a specif- ic microenvironment or based on the cooperation of multiple cell types can simply not be accessed by such approaches. Some of these limitations could be overcome by using more complex cell culture systems such as primary cells or mixed co-culture Discussion 113 systems, containing more than one cell type. The downside of such approaches however is the increased variability and the reduced throughput in combination with the rising costs of such cell systems. Another challenge of chemical genomic strate- gies derives from the determination of the appropriate incubation time or concen- tration for the treatment in order to obtain a strong signal response for the respec- tive process. It was shown that being off by one log of concentration has turned a strong signal into a barely detectable one119. It is therefore mandatory to profile a wide range of concentrations and several time points to obtain an integrated view of the NCE effects. Furthermore, there is an uncertainty to which extent gene expres- sion signatures of cells from one tissue differ in terms of NCE treatment from cells of another. Surprisingly, earlier chemical genomics studies analyzing histone modifica- tion, molecular chaperones and mTOR signaling have shown that gene expression signatures derived from cells of differing origin are remarkably similar57. This might differ for other cellular processes and only empirical evidence can resolve this issue. However, the Phenocopy project does not claim to unravel the entire spectrum of NCE effects but to provide guidance along the decisions of early drug discovery projects, where such topics cannot be addressed. 3.3 TGF-β Biology Gene set enrichment analysis resulted in 16 highly affected processes with known links to the TGF-β biology (Figure 20). TGF-β is involved in a plethora of biological processes and its cross-talk with other pathways can be detected in literally every stage of metazoan life from birth to death. During embryogenesis, complex interac- Discussion 114 tions of TGF-β with BMP, Wnt Hedgehog, Notch and mitogen-activated protein ki- nase (MAPK) signaling are crucial for stem cell maintenance, body patterning, cell fate determination, organogenesis and further retain their role in regulation of cellu- lar growth and functioning in adult tissues to control homeostasis. Often a deregula- tion of this pathway cross-talk is found in aged or diseased animals, e.g. in case of cancerogenesis84, 120-123. Most of the above mentioned pathway interactions are also represented in the on-target signature of the Phenocopy data emphasizing their ubi- quitous role in various cellular backgrounds. However, the respective cross-talks are complex and therefore hard to resolve. They result in the activation of various linked pathways and processes. MAPK signaling, for instance can either be regulated by TGF-β stimulation, which represents an important mechanism for non-Smad signal- ing80, or vice versa modulate the function of TGF-β signaling, primarily triggered through regulation of Smad functions124. Furthermore, TGF-β can potentially affect three different MAPK signaling pathways125 and result in oppositional effects. While Erk1/2 phosphorylation of Smad3 sites is supposed to prevent the transcriptional activity126 of TGF-β, p38 MAPK phosphorylation rather supports it127. It is however demanding to dissect which of these processes are triggered in HaCaT cells simply by looking at the regulated genes, since the KEGG category contains genes from all MAPK signaling pathways. Also not unexpected is the observed TGF-β effect on p53 signaling since the tumor suppressor protein physically interacts with Smad2/3 and jointly regulates the tran- scription of several TGF-β target genes128. For instance, one of the best known TGF-β effects is triggered by this interaction: the induction of cell cycle arrest though the up-regulation of the CDK inhibitors CDKN1A and CDKN2B129. Discussion 115 Another example for complex cross-talks, also represented in the Phenocopy data, is the interaction of TGF-β with ErbB2 signaling. This finding is backed up by different studies showing that ErbB2 signaling, which activates both, MAPK and PI3K Akt pathways, strongly communicates with TGF-β signaling in the control of mammary epithelial cell biology and breast cancer development130-133. In fact, the synergy of TGF-β and ErbB2/Ras/MAPK signaling often leads to the secretion of different cyto- kines and growth factors, including TGF-β itself and thereby promote epithel-to- mesenchymal transition (EMT) and cell invasion134-136. Also not surprisingly, TGF-β stimulation resulted in the regulation of genes involved in apoptosis. At this, TGF-β signaling can lead to opposing effects and provides sig- nals for either cell survival or apoptosis. The outcome of TGF-β stimulation is thereby determined by the interaction and the balance of different stimuli as well as by the given cell type125. Comparably to the findings of the MAPK signaling category, the KEGG apoptosis annotation is also ambiguously grouped regarding the different out- comes of apoptosis. TGF-β is involved in the remodeling of the extracellular matrix resulting in EMT. An early event in EMT involves the dissolution of tight junctions triggered by an interac- tion of Par6, a regulator of cell polarity and tight junction assembly with TGF-β sig- naling components137. Not only the tight junctions but also the gap junctions are affected by TGF-β through regulation of the gap junction subunit Cx43 expression which alters gap junction-mediated intercellular communication138-140. The outcome varies between positive and inverse relationships, depending on the cell type, the type of membrane receptors employed and the initial activation/phosphorylation state of the cells141. It has been further shown that this down-regulation at wound Discussion 116 sites leads to a reduced inflammatory response, enhanced keratinocyte proliferation and wound fibroblast migration142. Apart from the alteration of cell connection components, EMT is accompanied by a dramatic reorganization of the actin cytoskeleton also induced by TGF-β stimulation via Rho GTPase-dependent pathways143, 144. Interestingly, no influence or interaction of TGF-β signaling to the metabolisms of glycine, serine and threonine has been published so far. Therefore, the findings pro- vided by the Phenocopy experiment are a good starting point for further experi- ments to confirm and unravel the influence of TGF-β on the metabolism of these amino acids. The last three categories contained in Cluster 4 are wide-ranged and highly ambi- guous. They most likely represent the integration of the above mentioned pathways and processes. As mentioned for the off-target pathway prediction, in silico results based on path- way analysis tools merely draws attention to putatively affected processes and only wet laboratory validation can provide insight into their contribution to the targets biology. Furthermore, results obtained from only one cell line cannot cover the en- tire biology of the cytokine TGF-β and only provide initial insights and approaches for further experiments. Nevertheless, not only the off-target signatures but also an on- target signature can help to support the drug discovery process. On the one hand, these signatures can be overlaid with known disease signatures in order to annotate the targets contribution to the state of disease. On the other hand, it can be used to identify potent biomarkers for efficacy of the treatment and to support the clinical biomarker assay development process. This is especially important if the target’s Discussion 117 biology is not as well characterized as for TGF-βR1. Despite the good characterization of the receptor biology, using the Phenocopy strategy it was possible to significantly increase the amount of known TGF-β regulated genes by several hundred compared to earlier studies145-149. 3.4 siRNAs as modulators 3.4.1 Control siRNA charaterization The aim of this experimental section was to shed light on how different negative control siRNA molecules can influence RNAi experiments. The obtained knowledge could then be directly incorporated into the siRNA part of the Phenocopy strategy. In addition, a broader approach was chosen to identify the influence of the control mo- lecules in order to deduce information for siRNA experiments in general, beyond the use in the presented strategy. Dependent on the used cell line a wide range of diffe- rentially expressed genes was observed after the transfection with different siRNA molecules. Overall almost three times more genes were altered in HT1080 compared to 3T3-L1 cells indicating that the later are less sensitive to the treatment with siR- NAs. The transfection of siRNA Q1ctrl resulted in a small overlap of three genes that are differentially expressed in all tested cell lines, arguing for a potential sequence homology of this siRNA to the respective off-target transcripts. Due to commercial reasons at the providers, the sequences of the control siRNA molecules cannot be provided but the alignment of the siRNA sequences with the respective off-targets showed a low stringency match to nearly all the identified transcripts. Discussion 118 A larger overlap was found within the two human cell lines HT1080 and HaCaT. Here, the control siRNAs Q1ctrl, Q2ctrl, Q3ctrl and Q4ctrl share a common cell line inde- pendent off-target signature, suggesting that these genes are altered due to se- quence similarities to the siRNA sequence. However, no perfect or seed matches to the control siRNA sequences were identified. Subsequently, the focus was set on the analysis of the specific off-targets observed in HT1080 cells. Since most of these off- targets do not fall into the classical group of siRNA induced interferon response genes150, it can be anticipated that these regulations are sequence specific for the respective siRNA molecule151, 152. However, among the 595 identified off-targets, 79 genes were altered by the treatment with more than one single siRNA. The hierar- chical clustering revealed similar expression patterns for all Dharmacon siRNAs indi- cating that some off-target effects are sequence independent and one can speculate that these are related to chemical modification strategies of a specific manufacturer. The cluster analysis also demonstrated strong correlation between the cells treated with the molecules A1ctrl and Q1ctrl. In these experiments expression of the two cytokines IL1 and IL24 is significantly altered compared to the untreated control samples. The low expression of both cytokines argued already for a reduced re- sponse of the siRNA treated cells towards an inflammation signaling through NF B153. This hypothesis was further strengthened by the reduced level of MMP1 expression in the supernatants of these cultures97, which correlates with the reduced expression of both cytokines. However, not only the basal NF B signaling cascade, but also the stimulated pathway, as shown by the activation of IL8 secretion in HT1080 cells after TNF treatment, argues for a reduction of siRNA A1ctrl and Q1ctrl treated cells towards the inflammatory signaling. Discussion 119 Furthermore, the unwanted side effects are the opposite of what is described as an inflammation stimulation effect via the double stranded RNA recognition system. The specificity in terms of number of identified off-targets did not correlate with the observed reduction of TNFα sensitivity, but was probably dependent on specific se- quence motifs or chemical modifications. Variations in immune stimulation activity have already been described for a GFP control siRNA19 but the effects were re- stricted to the context of primary peripheral blood monocytes. Only very low expres- sion levels of the toll-like-receptors 3, 7 and 8 were found in the HT1080 fibroblast cell line by the gene chip expression study. However, the intracellular, cytoplasmatic double stranded RNA binding domain (DRBD) proteins like RIG-1, PKR and MDA-5 were constitutively expressed with strong detection signals. It remains questionable whether the control siRNA-induced resistance of the NF B pathway implicates bind- ing to DRBD-proteins since the interaction with these cytoplasmatic components would activate rather than silence this signaling cascade. The underlying mechanism for this observation remains obscure and will need a detailed understanding of the binding and signaling mechanisms of DRBD-proteins in general. The findings have some major implications on the design of target identification and validation experiments with siRNA molecules. The reduced inflammatory response described is clearly cell type specific, since only the HT1080 cell line did show the IL1 /IL24 effects. However, the data exemplifies that during the establishment of an assay protocol, the phenotypic analysis of the control siRNA treatment is essential. This is especially important since the identified off-target regulations are cell type specific and not a common feature of the molecules but related to the cellular back- ground of the experiment. The cell type specific effects of siRNAs can derive from Discussion 120 different sources. First, double stranded RNA related effects can be different depen- dent on the expression of double stranded RNA binding domain (DRBD) proteins, discussed above, or in general by a different expression of the RNAi machinery com- ponents. Second, sequence specific effects can differ due to different expression profiles and the biology of the hit target. For instance, the knockdown of an essential key player in cell homeostasis or of an important transcription factor will later on influence the expression of far more genes than the knockdown of a more distal tar- get. Finally, siRNAs can compete with endogenous miRNA pathways due to satura- tion effects that are specific for the respective cell line154. Although it is not mandatory to perform expression analysis upfront of each siRNA experiment, the impact of the control molecule on the assay system must be careful- ly analyzed. Therefore, only a close comparison of untreated to control siRNA treated cell cultures will help to identify the right control molecules, which show a controlled or even no impact on the assay system and will be appropriate for the normalization of the experimental observations. 3.4.2 TGF-β siRNAs To guarantee an optimal siRNA-mediated knockdown it is mandatory to pre-select a siRNA, validated for the respective biological background and used cellular system. Additionally, different chemical modifications can lead to an altered activity, stability and tolerability of the respective siRNA155. Since each siRNA provider follows its own modification strategy, it is beneficial to validate a panel of siRNAs from different pro- viders. Therefore, a total of ten siRNAs from three providers was assessed by the Discussion 121 analysis of three quality parameters: i) mRNA knockdown, ii) functional inhibition and iii) amount of off-target effects. Only the use of the siRNAs A2tgf, D4tgf and Q1tgf led to knockdowns of less than 80 %. In contrast, reasonable good knockdowns were achieved by more than 90 % for the siRNAs A1tgf, D3tgf, Q2tgf, Q3tgf and Q4tgf and superior knockdowns with more than 95 % for D1 and D2. Functional pathway inhibition, assayed by Smad phosphorylation and PAI-1 expression, led to good performances of the siRNAs A1tgf, D2tgf, Q3tgf and Q4tgf. Surprisingly, siRNA D1tgf that resulted in the best mRNA knockdown failed to inhibit TGF-β signaling. This finding remains obscure and can not be rationalized by any of its off-target ef- fects. One can only speculate that the siRNA mimics a miRNA effect that can poten- tial promote TGF-β signaling. Moreover, the fact that a siRNA mediating a good or perfect knockdown does not result in the desired phenotype is not unique and can also be experienced in nearly each large scale siRNA screen. Vice versa, each hit in a siRNA screen is normally backed up by at least one counter experiment using a dis- tinct siRNA to rule out single siRNA derived phenotypes. In parallel, the off-target analysis of these siRNAs favored both Dharmacon siRNAs D1tgf and D2tgf and the Ambion siRNA A1tgf and excludes the use of both Qiagen siRNAs with approximately twice as much off-target effects than the others. Collectively, siRNA A1tgf resulted in the best performance and was chosen for the use in the Phenocopy experiment. Although treatment with the validated siRNA virtually erased TGF-βR1 mRNA, the use of siRNAs as pathway modulators came of badly in contrast to the kinase inhibi- tors and was not able to totally block TGF-β signaling according to PAI-1 mRNA levels (Figure 19d). One possible explanation is the internalization of the activated TGF-β receptor complex. TGF-β receptors are internalized through two distinct endocytic Discussion 122 pathways. On the one hand, TGF-β receptors can be internalized in caveolin positive vesicles for receptor degradation. On the other hand, Clathrin-dependent internali- zation into early endosomes is important for propagating signals. This was first eluci- dated by Di Guglielmo et al.156 who showed that the internalized TGF-β receptor complex is first co-localized with EEA1 (Early Endosome Antigen 1) and then with Rab11 suggesting that the EEA1 compartment has subsequently entered Rab11- positive recycling endosomes156, 157. Since a recycling of TGF-β receptors is possible, one can rational that due to the long half-life and turnover of the protein, even a complete blockage of its de novo synthesis using siRNAs will rather fade out TGF-β- mediated signaling then abruptly stop it. 3.5 Conclusion Neither the complexity of a living organism nor a disease state can be entirely represented by profiling of a single cell line. Nevertheless, the Phenocopy strategy demonstrates one possibility to significantly alleviate the drug discovery process at an early stage. Comparing such an approach to classical toxicology testing or toxico- genomics studies, the Phenocopy strategy offers a couple of advantages: it addresses on- and off-target effects and is able to differentiate between target-related vs. compound-related events. This differentiation is only possible when a couple of compounds of different compound classes will be investigated. Due to costs and ca- pacities, the analysis of a certain number of compounds can only be run in vitro. Al- though cellular systems cannot replace in vivo studies, they show less variability, Discussion 123 guarantee the expression and signaling of the target protein and are less cost and time consuming. The Phenocopy approach offers an opportunity to qualify and rank compound classes and single compounds early during hit-to-lead and lead optimization processes, which will subsequently reduce the attrition rates later on, e.g. during toxicological assessment of the development candidates. However, the addition of new technologies and checkpoints like Phenocopy is contributing to the ever-rising costs of getting innovative medicine to the market. But nevertheless, the assembly of workflows of successfully used tools during early lead generation processes will become crucial for the discovery of novel quality of entities in a changing pharma- ceutical industry. One useful tool is the phenocopy principle, where external stimuli like the climate of the environment for the Himalayan rabbit or like NCEs, siRNAs, antibodies or aptameres for the inhibition of a cellular process are committing a cer- tain phenotype. By investing in the qualification of NCEs during the early drug dis- covery process, later on the attrition rate during development phases will be re- duced. Indirectly, this investment will reduce the overall cost for developing innova- tive medicine. Ph.D. Thesis Patrick Baum 4. Methods Phenocopy – A Strategy to Qualify Chemical Compounds during Hit-to-Lead and Lead Optimization Methods 125 4.1 Wet laboratory experiments 4.1.1 Cell culture, NCE treatment and siRNA transfection 3T3-L1, HT1080 and HaCaT cells were maintained in DMEM (3T3-L1, HaCaT; Gibco; Cat. No. 31966) or RPMI (HT1080; RPMI; Cat. No. 61870 ) supplemented with 10 % (3T3-L1, HT1080) or 5 % (HaCaT) FCS, respectively. All cell lines were exponentially grown at 37 °C in a 5 % CO2 atmosphere. Cells were seeded in 96-well (ELISA) or in 24-well (RNA expression profiling) plates and grown overnight to a confluence of approximately 70 %. For TGF-β expression experiments, cells were first starved for 3 h in DMEM containing no FCS. Cells were then pre-incubated with increasing NCE concentrations (0.0032, 0.016, 0.08, 0.4, 2, 10, 50 µM) for 15 min and subsequently stimulated with 5 ng/ml of TGF-β (R&D Systems; Cat. No. 240-B) and incubated for the indicated time points. In parallel, control sample groups were analyzed that were either only stimulated with TGF-β, treated with NCEs but not TGF-β stimulated, treated with DMSO in presence and absence of TGF-β or left untreated. Finally, the medium was collected or discarded cells were washed with PBS and lysed using RLT buffer (Qiagen; Cat. No. 79216). All siRNAs were purchased from Ambion, Dharmacon or Qiagen and prepared ac- cording to manufacturer’s instructions (Table 1, Table 4). For transfection experi- ments, cells were seeded in 24-well plates and grown overnight to a confluency of 50-70 %. siRNAs were transfected at a final medium concentration of 20 nM. Cells were either transfected using Lipofectamine RNAiMAX (Invitrogen; Cat. No. 13778- 150) for 3T3-L1, HT1080 or DharmaFECT1 reagent (Dharmacon; Cat. No. T-2001-03) for HaCaT cells. 24 h post transfection, the medium was replaced. 48 h after trans- Methods 126 fection cells were then also treated starved and TGF-β stimulated according to the above mentioned conditions, only treated with the transfection reagent or left un- treated dependent on the performed experiment. Subsequently, cells were washed with PBS and lysed using RLT buffer (Qiagen Cat. No. 79216). For electroporation siRNAs were transfected using an Amaxa Nucleofector® (Lonza) and the Amaxa Cell line Nucleofector® Kit L (Lonza, Cat. No. VCA-1005) according to the manufacturer’s protocol. 4.1.2 RNA extraction RNA isolation was carried out using a MagMAX™ Express-96 Magnetic Particle Pro- cessor (Ambion) and the MagMAX™-96 Total RNA Isolation Kit (Ambion; Cat. No. AM1830) according to the manufacturer’s protocol. Total RNA concentration was quantified by fluorescence measurement using SYBR Green II (Invitrogen; Cat. No. S- 7568) and a Synergy HT reader (BioTek) as previously described158. The RNA quality was characterized by the quotient of the 28S to 18S ribosomal RNA electrophero- gram peak using an Agilent 2100 bioanalyzer and the RNA Nano Chip (Agilent; Cat. No. 1511). 4.1.3 Quantitative real time polymerase chain reaction (qRT-PCR) mRNA expression levels of TGF-βR1, PAI-1, GAPDH, CDKN1A, CDKN2B, LINCR, HAND1, RPTN and JUNB were determined by qRT-PCR analysis using a 7900HT Fast Real-Time PCR System (Applied Biosystems) and the Universal ProbeLibrary System (Roche; Cat. No. 04 683 633 001). Gene specific forward- and reverse primer se- Methods 127 quences were designed using the Universal Probe Library Assay Design Center (Roche) and are shown in Table 12. Total RNA was transcribed into cDNA using the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems; Cat. No. 4368814) according to the manufacture’s instructions. PCRs were performed accord- ing to the manufacturer’s protocol using TaqMan reagents for 40 cycles. The thre- shold cycles (CT) for each cDNA were obtained from triplicate samples. The ΔCT was the calculated difference between the average CT for the target gene and the aver- age CT for the house keeper gene POLR2A (RNA polymerase 2) as a control for total starting RNA quantity. ΔΔCT method was then used to relatively quantify mRNA le- vels of treated or stimulated samples compared to untreated, unstimulated or un- transfected controls. Table 12 – qRT-PCR primer Gene Forward primer Reverse primer TGF-βR1 aaattgctcgacgatgttcc cataataaggcagttggtaatcttca PAI-1 aaggcacctctgagaacttca cccaggactaggcaggtg GAPDH gctctctgctcctcctgttc acgaccaaatccgttgactc CDKN1A tgcgttcacaggtgtttctg agctgctcgctgtccact CDKN2B ttgttagaaaccaggctgcac ttctctctttctgtggtttctcaat LINCR ccttcgagagcatttgccta tgttggcagcgtgatagaag HAND1 aactcaagaaggcggatgg ggaggaaaaccttcgtgct TSC22D1 tttcccgttgaaggtgctac ttgtcaatagctaccacacttgc RPTN gctcttggctgagtttggag aggttcaagatggtttccaca JUNB atacacagctacgggatacgg gctcggtttcaggagtttgt POLR2A ttgtgcaggacacactcaca caggaggttcatcacttcacc 4.1.4 ELISA analysis of PAI-1, phospho Smad2/3, MMP1, IL8 and IL6 To analyze Smad2/3 phosphorylation after TGF-β stimulation and/or NCE treatment cells were lysed in 100 µl Cell Lysis Buffer (Cell Signaling; Cat. No. 9803) supple- mented with 1 mM PMSF (Sigma; Cat. No. P7626). 96-well plates (Nunc MaxiSorp™; Methods 128 Cat. No. 44-2404-21) were coated with anti-Smad2/3 monoclonal antibody (1 µg/ml; BD Bioscience; Cat. No. 610843) for 24 h at 4 °C. To reduce unspecific binding the wells were blocked with PBS (Gibco; Cat. No. 14190) + 2 % BSA (Sigma; Cat. No. A- 7030) for 2 h at RT. After washing three times with PBS + 0.1 % Tween20 (Sigma; Cat. No. P-7949), the protein lysate was added and incubated for 2 h at RT. Wells were washed three times with wash buffer and incubated with an anti-phoshpo-Smad2/3 specific rabbit antisera (Eurogentec) diluted in PBS + 0.2 % BSA (Sigma; Cat. No. A- 7030) + 0.02 % Tween20 (Sigma; Cat. No. P-7949) and incubated for 2 h at RT. An AP- conjugated monoclonal antibody mouse anti-rabbit IgG (Sigma; Cat. No. A2556) was added and incubated for 2 h at RT. pNPP Liquid Substrate System (Sigma; Cat. No. N7653) was added and developed in the dark at 37 °C for 2 h before the absorbance was measured at 405 nm in a Synergy HT plate reader (BioTek). To analyze protein expression of PAI-1 supernatants of TGF-β stimulated cells were collected 1, 2, 4, 12 and 24 h after treatment and analyzed using the PAI-1 Antigen Kit (Haemochrom Diagnostica; Cat. No. HD44006), according to the manufacturer’s protocol. To analyze protein expression of MMP1 and IL8 supernatants of trans- fected or transfection reagent treated cells were collected 72 h after control siRNA transfection. In case of IL8 cells were previously stimulated with 30 ng/ml TNFα (Alexis; Cat. No. ALX-520-002) for 8 h. Subsequently, protein expression was deter- mined using the MMP1 ELISA kit (Calbiochem; Cat. No. QIA55) and the human IL8 ELISA kit (BD Bioscience; Cat. No. 555244) according to the manufacturer’s protocol. To analyze the expression of the four pro-inflammatory cytokines IL1β, TNFα, IL8 and IL6 cells were treated with NCEs at the indicated concentrations and incubated at 37 Methods 129 °C for 12 h. Supernatants were analyzed using a Mesoscale Discovery muliplex ELISA System (MSD) and the Hu ProInflammatory-4 II Tissue Culture Kit (MSD; Cat. No. K15025B-1) for detection according to the manufacturer’s instruction. 4.1.5 LDH release assay LDH release was measured to analyze cell toxicity after siRNA transfection using ei- ther lipofection or electroporation protocols. Therefore, the CytoTox-ONE™ Homo- geneous Membrane Integrity Assay Kit (Promega; Cat. No. G7890) was used accord- ing to the manufacturer’s instruction. As positive control, cells were treated with Triton X-100 (Sigma; Cat. No. T9284) and incubated for 3 h at 37 °C. The signal ob- tained for positive control treated cells was then set as 100 %. 4.1.6 Amplification, labeling and Beadchip hybridization of RNA samples Illumina TotalPrep RNA Amplification Kit (Ambion; Cat. No. 4393543) was used to transcribe 350 ng of the isolated total RNA. Briefly, total RNA was first converted into single-stranded cDNA with reverse transcriptase using an oligo-dT primer containing the T7 RNA polymerase promoter site and then copied to produce double-stranded cDNA molecules. An overnight (14 h) in vitro transcription was performed using a T7 polymerase to generate single-stranded RNA molecules (cRNA). cRNA molecules were labeled by incorporation of biotin-UTP. A total of 700 ng of cRNA was hybri- dized at 58°C for 16 h to either the Illumina HumanRefseq-8 v2 Expression BeadChip (HaCaT, HT1080; Illumina; Cat. No. BD-102-0203) or the MouseRef-8 v1.1, (3T3-L1; Illumina; Cat. No. BD-202-0202), respectively for the identification of the control Methods 130 siRNA off-target effects. All sample of the Phenocopy experiment were hybridized to the next generation of Illumina chips the HumanHT-12 Expression BeadChips (Illumi- na; Cat. No. BD-103-0203). After hybridization, the arrays were washed, blocked and the labeled cRNA was detected by staining with streptavidin-Cy3. BeadChips were scanned using an Illumina Bead Array Reader and the Bead Scan Software (Illumina). 4.1.7 High content screen Cellomics The high-content cytotoxicity assay kit 2 was performed according to the manufac- turer’s instructions (ThermoFisher Cellomics; Cat. No. 8400102). Briefly, HaCaT cells were cultured overnight in black 96-well plates, incubated for 24 h with each NCE at the indicated concentrations and stained with cytotoxicity cocktail. Cells were fixed, washed and scanned on the Cellomics ArrayScan II platform. Images were analyzed with the Cell Health image analysis algorithm. Cytotoxicity indices were calculated for each of the four parameters to indicate the percentage of cells outside of the normal range which was defined using a vehicle-treated reference cell population. 4.1.8 Caspase-3 Assay Cells were seeded in 6-well plates and grown overnight to a confluence of approx- imately 70 % before they were treated with 2 µM of each NCE and incubated for 24 h. Caspase-3 activity was quantified using Facs Canto (BD Biosciences) and the Caspase-3 Detection Kit (Calbiochem; Cat. No. QIA91) according to the manufactur- er’s instruction. Methods 131 4.1.9 In vitro kinase profiling The SelectScreenTM kinase Profiling Service was performed (Invitrogen) to indentify the compound selectivity against 239 kinases. Single-point kinase inhibitory activities of each compound at 2 µM and 0.2 µM were measured at 100 µM or Km ATP con- centrations. Downstream targets of the identified off-target kinases were manually extracted from Ingenuity’s Knowledge Base and overlaid with the NCE off-targets for comparison. 4.2 Data Analysis 4.2.1 Data processing Phenocopy data was processed with BeadStudio version 3.0 and the R Language and Environment for Statistical Computing (R) 2.7.0 in combination with Bioconductor 2.2159. The Bioconductor lumi package104 was used for quality control and normaliza- tion. The data was log2 transformed and normalized using robust spline normaliza- tion (rsn). Linear models (Bioconductor package limma)160 were used to calculate log2 ratios, the resulting p-values were FDR-corrected 68. Other pre-processing me- thods tested prior to the Phenocopy experiment are summarized in the work of Schmid et al.106. Differential expression for all siRNA experiments were analyzed for siRNA treated samples versus transfection reagent treated samples. Identification of differential expression after TGF-β or NCE treatment is described below. The differentially ex- pressed transcripts after control siRNA transfection in HT1080 cells (Figure 11) as well as the transcripts of the treatment signature (Figure 24) were clustered using Methods 132 Spotfire® DescisionSite® 9.1.1. (Spotfire). Hierarchical clustering was performed us- ing manhattan distance and complete linkage as similarity measure. 4.2.2 TGF-β signature (on-target signature) To define genes deregulated by TGF-β signaling, three sequential filtering steps were applied to the log2 transformed expression values of each time point separately. i) The first filtering is based on the comparison of TGF-β stimulated cells against un- treated cells by linear models160 (FDR corrected68 p-value < 0.01 and |log2 ratio| ≥ 0.5). ii) A linear model was applied to the dose groups of each compound to extract all probes which are significantly deregulated (FDR-corrected p-value < 0.01) by at least one concentration compared to the respective control (cells treated with TGF-β and DMSO but no compound). iii) To detect probes with a dose dependent deregula- tion the likelihood ratio test statistic for monotonicity (R package IsoGene161) was used. IsoGene performs an isotonic regression based on the replicates for each con- centration resulting in regression values μ1, μ2, …, μ6 for each probe and each com- pound treatment. Only probes that are significantly regulated by at least one com- pound with |μ1-μ6| ≥ 1 and an FDR corrected p-value < 0.01 are included in the fur- ther analysis. For each time point the probes that passed all three filters are pooled to the final TGF-β signature. 4.2.3 Off-target signature To detect transcripts that are deregulated due to off-target effects of the com- pounds unstimulated cells (wotgf class) as well as TGF-β stimulated cells (tgf class) Methods 133 were considered. For the wotgf class, the NCE samples 0.08 μM just as 2 μM were compared to untreated cells (d11 and d12, respectively). Additionally, the 2 μM sam- ple was compared to the 0.08 μM sample (δ1) for each time point using linear mod- els160 (FDR-corrected68 p-value < 0.01 and |log2 ratio| ≥ 1). The same comparisons were made based on the tgf class (d21, d22 and δ2). Transcripts that are up- or down- regulated by either compound treatment (wotgfup and wotgfdown, respectively) or by TGF-β stimulation together with compound treatment (tgfup and tgfdown, respective- ly) were detected based on the described comparisons as follows: a transcript be- longs to the class wotgfup if either δ1 is significantly up-regulated (δ1_up) or if δ1 is not significantly down-regulated (≦δ1_down) but d11, d12, and δ1 indicate an increasing course of expression intensity for higher compound concentrations. That is, if ≦δ1_down holds true, five different trends render up-regulation: 1) d11 and d12 are both significantly up-regulated; 2) d11 but not d12 is significantly down-regulated and log2ratio(δ1) ≥ 1, thereby showing an increasing trend of expression for increasing compound concentrations; 3) d11 but not d12 is significantly up-regulated and log2ratio(δ1) > -1, allowing for a small but not significant decreasing trend for increas- ing compound concentration; 4) d12 but not d11 is significantly down-regulated and log2ratio(δ1) ≥ 1; 5) d12 but not d11 is significantly up-regulated and log2ratio(δ1) > -1. On the one hand, as soon as one of d11 or d12 is significantly up-regulated (cases 3 and 5), a small amount of noise was allowed by claiming log2ratio(δ1) > -1. On the other hand, as soon as one of d11 or d12 is significantly down-regulated (cases 2 and 4), one are more strict by claiming log2ratio(δ1) ≥ 1 to call a transcript as being up- regulated. Methods 134 Stated in a more mathematical fashion, transcripts up-regulated within the wotgf class are defined as follows: The mirrored method was used to detect wotgfdown and the analogous methods are used to detect tgfup and tgfdown based on the cells stimulated with TGF-β. 1))]- ratiolog dd( 1)ratiolog dd( 1)- ratiolog d (d 1)ratiolog d (d )d ((d )[( wotgf 1212_up11_up 1212_down11_down 1212_up11_up 1212_down11_down 12_up11_up1_down 1_up up 1))]- ratiolog dd( 1)ratiolog dd( 1)- ratiolog d (d 1)ratiolog d (d )d ((d )[( wotgf 1212_up11_up 1212_down11_down 1212_up11_up 1212_down11_down 12_down11_down1_up 1_down down 1))]- ratiolog d d( 1) ratiolog d d( 1)- ratiolog d (d 1) ratiolog d(d )d ((d )[( tgf 2222_up21_up 2222_down21_down 2222_up21_up 2222_down21_down 22_up21_up2_down 2_up up 1))]- ratiolog dd( 1) ratiolog dd( 1)- ratiolog d (d 1)ratiolog d (d )d ((d )[( | tgf 2222_up21_up 2222_down21_down 2222_up21_up 2222_down21_down 22_down21_down2_up 2_down down Methods 135 The final off-target signature was defined based on the following transcripts: (tgfup wotgfup) (tgfdown wotgfdown) (tgfup wotgfdown) (tgfdown wotgfup) The profiles of the respective transcripts can be assigned to different categories as described in Figure 22 and Figure 23. tgfup wotgfup: additive or bipolar on- and off-target effect or common off-target effect tgfdown wotgfdown: additive or bipolar on- and off-target effect or common off-target effect tgfup wotgfdown: inverse of bipolar on- and off-target effect tgfdown wotgfup: inverse or bipolar on- and off-target effect 4.2.4 Ingenuity Pathway Analysis and Gene Set Enrichment Analysis Based on the on- and off-target signatures, standard IPAs were used to generate networks and perform GSEA using Fisher’s exact test for molecular function and ca- nonical pathways defined by the Ingenuity Knowledge Base73. Additionally, GSEA for the on-target signatures was conducted using Fisher’s exact test based on gene sets defined by KEGG pathways as annotated by the Bioconduc- tor package KEGG.db version 2.2.074. The p-values calculated based on Fisher’s exact test were clustered using manhattan distance and complete linkage. Reference List 136 Reference List 1. Sawin,P.B. Hereditary Variation of the Chin-Chilla Rabbit: In Coat and Eye Color. J. Hered. 23, 39-46 (1932). 2. Tuschl,T., Zamore,P.D., Lehmann,R., Bartel,D.P., & Sharp,P.A. Targeted mRNA degradation by double-stranded RNA in vitro. Genes Dev. 13, 3191- 3197 (1999). 3. Whitehead,K.A., Langer,R., & Anderson,D.G. Knocking down barriers: advances in siRNA delivery. Nat. Rev. Drug Discov. 8, 129-138 (2009). 4. Zamore,P.D., Tuschl,T., Sharp,P.A., & Bartel,D.P. RNAi: double- stranded RNA directs the ATP-dependent cleavage of mRNA at 21 to 23 nucleotide intervals. Cell 101, 25-33 (2000). 5. Elbashir,S.M., Lendeckel,W., & Tuschl,T. RNA interference is mediated by 21- and 22-nucleotide RNAs. Genes Dev. 15, 188-200 (2001). 6. Weitzer,S. & Martinez,J. hClp1: a novel kinase revitalizes RNA meta- bolism. Cell Cycle 6, 2133-2137 (2007). 7. Nykanen,A., Haley,B., & Zamore,P.D. ATP requirements and small in- terfering RNA structure in the RNA interference pathway. Cell 107, 309-321 (2001). 8. Khvorova,A., Reynolds,A., & Jayasena,S.D. Functional siRNAs and miR- NAs exhibit strand bias. Cell 115, 209-216 (2003). 9. Schwarz,D.S. et al. Asymmetry in the assembly of the RNAi enzyme complex. Cell 115, 199-208 (2003). 10. Watson,J.D. & Crick,F.H. The structure of DNA. Cold Spring Harb. Symp. Quant. Biol. 18, 123-131 (1953). 11. Martinez,J., Patkaniowska,A., Urlaub,H., Luhrmann,R., & Tuschl,T. Sin- gle-stranded antisense siRNAs guide target RNA cleavage in RNAi. Cell 110, 563-574 (2002). 12. Tomari,Y. & Zamore,P.D. Perspective: machines for RNAi. Genes Dev. 19, 517-529 (2005). 13. Meylan,E., Tschopp,J., & Karin,M. Intracellular pattern recognition receptors in the host response. Nature 442, 39-44 (2006). 14. Sledz,C.A. & Williams,B.R. RNA interference and double-stranded- RNA-activated pathways. Biochem. Soc. Trans. 32, 952-956 (2004). Reference List 137 15. Pippig,D.A. et al. The regulatory domain of the RIG-I family ATPase LGP2 senses double-stranded RNA. Nucleic Acids Res. 37, 2014-2025 (2009). 16. Diebold,S.S. et al. Nucleic acid agonists for Toll-like receptor 7 are de- fined by the presence of uridine ribonucleotides. Eur. J. Immunol. 36, 3256- 3267 (2006). 17. Hornung,V. et al. Sequence-specific potent induction of IFN-[alpha] by short interfering RNA in plasmacytoid dendritic cells through TLR7. Nat. Med 11, 263-270 (2005). 18. Judge,A. & MacLachlan,I. Overcoming the innate immune response to small interfering RNA. Hum. Gene Ther. 19, 111-124 (2008). 19. Robbins,M. et al. Misinterpreting the therapeutic effects of small in- terfering RNA caused by immune stimulation. Hum. Gene Ther. 19, 991-999 (2008). 20. Sledz,C.A., Holko,M., de Veer,M.J., Silverman,R.H., & Williams,B.R. Activation of the interferon system by short-interfering RNAs. Nat. Cell Biol. 5, 834-839 (2003). 21. Jackson,A.L. et al. Position-specific chemical modification of siRNAs reduces "off-target" transcript silencing. RNA 12, 1197-1205 (2006). 22. Czauderna,F. et al. Structural variations and stabilising modifications of synthetic siRNAs in mammalian cells. Nucleic Acids Res. 31, 2705-2716 (2003). 23. Cekaite,L., Furset,G., Hovig,E., & Sioud,M. Gene expression analysis in blood cells in response to unmodified and 2'-modified siRNAs reveals TLR- dependent and independent effects. J. Mol. Biol. 365, 90-108 (2007). 24. Puri,N. et al. LNA incorporated siRNAs exhibit lower off-target effects compared to 2'-OMethoxy in cell phenotypic assays and microarray analy- sis. Nucleic Acids Symp. Ser. (Oxf)25-26 (2008). 25. Dowler,T. et al. Improvements in siRNA properties mediated by 2'- deoxy-2'-fluoro-beta-D-arabinonucleic acid (FANA). Nucleic Acids Res. 34, 1669-1675 (2006). 26. U.S. Food and Drug Administration. www.fda.gov . 2010. 27. Hopkins,A.L. & Groom,C.R. The druggable genome. Nat. Rev. Drug Discov. 1, 727-730 (2002). 28. Russ,A.P. & Lampel,S. The druggable genome: an update. Drug Discov. Today 10, 1607-1610 (2005). Reference List 138 29. Kinexus. www.kinexus.ca . 2010. 30. Zhang,J., Yang,P.L., & Gray,N.S. Targeting cancer with small molecule kinase inhibitors. Nat. Rev. Cancer 9, 28-39 (2009). 31. Chico,L.K., Van Eldik,L.J., & Watterson,D.M. Targeting protein kinases in central nervous system disorders. Nat. Rev. Drug Discov. 8, 892-909 (2009). 32. Frantz,S. Drug discovery: playing dirty. Nature 437, 942-943 (2005). 33. Roth,G.J. et al. Design, Synthesis and Evaluation of Indolinones as In- hibitors of the Transforming Growth Factor Beta Receptor I. (submitted). 34. Zou,J. et al. Microarray profile of differentially expressed genes in a monkey model of allergic asthma. Genome Biol. 3, research0020 (2002). 35. Rajcevic,U., Niclou,S.P., & Jimenez,C.R. Proteomics strategies for tar- get identification and biomarker discovery in cancer. Front. Biosci. 14, 3292- 3303 (2009). 36. Rodrigues,C.D. et al. Host scavenger receptor SR-BI plays a dual role in the establishment of malaria parasite liver infection. Cell Host. Microbe 4, 271-282 (2008). 37. Terstappen,G.C., Schlupen,C., Raggiaschi,R., & Gaviraghi,G. Target deconvolution strategies in drug discovery. Nat. Rev. Drug. Discov. 6, 891- 903 (2007). 38. Yang,Y., Adelstein,S.J., & Kassis,A.I. Target discovery from data mining approaches. Drug Discov. Today 14, 147-154 (2009). 39. Cheng,K.C., Korfmacher,W.A., White,R.E., & Njoroge,F.G. Lead Optimi- zation in Discovery Drug Metabolism and Pharmacokinetics/Case study: The Hepatitis C Virus (HCV) Protease Inhibitor SCH 503034. Perspect. Medicin. Chem. 1, 1-9 (2008). 40. Benedetti,M.S. et al. Drug metabolism and pharmacokinetics. Drug Metab. Rev. 41, 344-390 (2009). 41. Tillement,J.P. A low distribution volume as a determinant of efficacy and safety for histamine (H1) antagonists. Allergy 50, 12-16 (1995). 42. Zamek-Gliszczynski,M.J., Hoffmaster,K.A., Nezasa,K., Tallman,M.N., & Brouwer,K.L. Integration of hepatic drug transporters and phase II metabo- lizing enzymes: mechanisms of hepatic excretion of sulfate, glucuronide, and glutathione metabolites. Eur. J. Pharm. Sci. 27, 447-486 (2006). Reference List 139 43. Wexler,D.S. et al. Linking solubility and permeability assays for maxi- mum throughput and reproducibility. J. Biomol. Screen. 10, 383-390 (2005). 44. Cheng,Y., Ho,E., Subramanyam,B., & Tseng,J.L. Measurements of drug- protein binding by using immobilized human serum albumin liquid chroma- tography-mass spectrometry. J. Chromatogr. B Analyt. Technol. Biomed. Life Sci. 809, 67-73 (2004). 45. Li,A.P. Screening for human ADME/Tox drug properties in drug discov- ery. Drug Discov. Today 6, 357-366 (2001). 46. Ragu Ramanathan Mass spectrometry in drug metabolism and phar- macokinetics (2009). 47. Barile,F.A., Dierickx,P.J., & Kristen,U. In vitro cytotoxicity testing for prediction of acute human toxicity. Cell Biol. Toxicol. 10, 155-162 (1994). 48. O'Brien,P.J. et al. High concordance of drug-induced human hepato- toxicity with in vitro cytotoxicity measured in a novel cell-based model us- ing high content screening. Arch. Toxicol. 80, 580-604 (2006). 49. Ulrich,R. & Friend,S.H. Toxicogenomics and drug discovery: will new technologies help us produce better drugs? Nat. Rev. Drug Discov. 1, 84-88 (2002). 50. Khor,T.O., Ibrahim,S., & Kong,A.N. Toxicogenomics in drug discovery and drug development: potential applications and future challenges. Pharm. Res. 23, 1659-1664 (2006). 51. Kinders,R. et al. Phase 0 clinical trials in cancer drug development: from FDA guidance to clinical practice. Mol. Interv. 7, 325-334 (2007). 52. U.S. National Institue of Health - Clinical Trails. clinicaltrails.gov . (2010). 53. Kola,I. & Landis,J. Can the pharmaceutical industry reduce attrition rates? Nat. Rev. Drug. Discov. 3, 711-715 (2004). 54. DiMasi,J.A., Hansen,R.W., & Grabowski,H.G. The price of innovation: new estimates of drug development costs. J. Health Econ. 22, 151-185 (2003). 55. Hughes,B. 2009 FDA drug approvals. Nat. Rev. Drug Discov. 9, 89-92 (2010). 56. Walker,I. & Newell,H. Do molecularly targeted agents in oncology have reduced attrition rates? Nat. Rev. Drug Discov. 8, 15-16 (2009). Reference List 140 57. Lamb,J. et al. The Connectivity Map: using gene-expression signatures to connect small molecules, genes, and disease. Science 313, 1929-1935 (2006). 58. Schena,M., Shalon,D., Davis,R.W., & Brown,P.O. Quantitative monitor- ing of gene expression patterns with a complementary DNA microarray. Science 270, 467-470 (1995). 59. Lockhart,D.J. et al. Expression monitoring by hybridization to high- density oligonucleotide arrays. Nat. Biotechnol. 14, 1675-1680 (1996). 60. Kuhn,K. et al. A novel, high-performance random array platform for quantitative gene expression profiling. Genome Res. 14, 2347-2356 (2004). 61. Hughes,T.R. et al. Functional discovery via a compendium of expres- sion profiles. Cell 102, 109-126 (2000). 62. Ramaswamy,S. et al. Multiclass cancer diagnosis using tumor gene expression signatures. Proc. Natl. Acad. Sci. U. S. A 98, 15149-15154 (2001). 63. Su,A.I. et al. A gene atlas of the mouse and human protein-encoding transcriptomes. Proc. Natl. Acad. Sci. U. S. A 101, 6062-6067 (2004). 64. ArrayExpress. www.ebi.ac.uk/microarray-as/ae . (2010). 65. Gene Expression Omnibus. www.ncbi.nlm.nih.gov/geo . (2010). 66. Lim,W.K., Wang,K., Lefebvre,C., & Califano,A. Comparative analysis of microarray normalization procedures: effects on reverse engineering gene networks. Bioinformatics. 23, i282-i288 (2007). 67. Chen,J.J., Wang,S.J., Tsai,C.A., & Lin,C.J. Selection of differentially ex- pressed genes in microarray data analysis. Pharmacogenomics. J. 7, 212- 220 (2007). 68. Benjamini,Y. & Hochberg,Y. Controlling the false discovery rate: A practiacal and powerful approach to multiple testing. J. Roy. Statist. Soc. Ser. B 57, 289-300 (1995). 69. Lin,D. et al. Testing for trends in dose-response microarray experi- ments: a comparison of several testing procedures, multiplicity and resam- pling-based inference. Stat. Appl. Genet. Mol. Biol. 6, Article26 (2007). 70. Ji,R.R. et al. Transcriptional profiling of the dose response: a more po- werful approach for characterizing drug activities. PLoS. Comput. Biol. 5, e1000512 (2009). Reference List 141 71. Golub,T.R. et al. Molecular classification of cancer: class discovery and class prediction by gene expression monitoring. Science 286, 531-537 (1999). 72. BioCarta. www.biocarta.com . (2010). 73. Ingenuity Pathway Analysis. www.ingenuity.com . (2010). 74. Kanehisa,M. & Goto,S. KEGG: kyoto encyclopedia of genes and ge- nomes. Nucleic Acids Res. 28, 27-30 (2000). 75. WikiPathways. www.wikipathways.org . (2010). 76. Matthews,L. et al. Reactome knowledgebase of human biological pathways and processes. Nucleic Acids Res. 37, D619-D622 (2009). 77. Subramanian,A. et al. Gene set enrichment analysis: a knowledge- based approach for interpreting genome-wide expression profiles. Proc. Natl. Acad. Sci. U. S. A 102, 15545-15550 (2005). 78. Gunderson,K.L. et al. Decoding randomly ordered DNA arrays. Ge- nome Res. 14, 870-877 (2004). 79. Shi,Y. & Massague,J. Mechanisms of TGF-beta signaling from cell membrane to the nucleus. Cell 113, 685-700 (2003). 80. Zhang,Y.E. Non-Smad pathways in TGF-beta signaling. Cell Res. 19, 128-139 (2009). 81. Sorrentino,A. et al. The type I TGF-beta receptor engages TRAF6 to activate TAK1 in a receptor kinase-independent manner. Nat. Cell Biol. 10, 1199-1207 (2008). 82. Blobe,G.C., Schiemann,W.P., & Lodish,H.F. Role of transforming growth factor beta in human disease. N. Engl. J. Med. 342, 1350-1358 (2000). 83. Massague,J., Blain,S.W., & Lo,R.S. TGFbeta signaling in growth control, cancer, and heritable disorders. Cell 103, 295-309 (2000). 84. Massague,J. TGFbeta in Cancer. Cell 134, 215-230 (2008). 85. Lahn,M., Kloeker,S., & Berry,B.S. TGF-beta inhibitors for the treatment of cancer. Expert. Opin. Investig. Drugs 14, 629-643 (2005). 86. Kang,H.R., Cho,S.J., Lee,C.G., Homer,R.J., & Elias,J.A. Transforming growth factor (TGF)-beta1 stimulates pulmonary fibrosis and inflammation via a Bax-dependent, bid-activated pathway that involves matrix metallo- proteinase-12. J. Biol. Chem. 282, 7723-7732 (2007). Reference List 142 87. Yingling,J.M., Blanchard,K.L., & Sawyer,J.S. Development of TGF-beta signalling inhibitors for cancer therapy. Nat. Rev. Drug Discov. 3, 1011-1022 (2004). 88. Zanini,A., Chetta,A., & Olivieri,D. Therapeutic perspectives in bronchial vascular remodeling in COPD. Ther. Adv. Respir. Dis. 2, 179-187 (2008). 89. Rosendahl,A. et al. Activation of the TGF-beta/activin-Smad2 pathway during allergic airway inflammation. Am. J. Respir. Cell Mol. Biol. 25, 60-68 (2001). 90. Derynck,R., Akhurst,R.J., & Balmain,A. TGF-beta signaling in tumor suppression and cancer progression. Nat. Genet. 29, 117-129 (2001). 91. Jachimczak,P. TGFbeta in cancer and other disease - AACR special con- ference in cancer research. IDrugs. 9, 239-241 (2006). 92. Wilkins-Port,C.E. & Higgins,P.J. Regulation of extracellular matrix re- modeling following transforming growth factor-beta1/epidermal growth factor-stimulated epithelial-mesenchymal transition in human premalig- nant keratinocytes. Cells Tissues. Organs. 185, 116-122 (2007). 93. Ovcharenko,D., Jarvis,R., Hunicke-Smith,S., Kelnar,K., & Brown,D. High-throughput RNAi screening in vitro: from cell lines to primary cells. RNA. 11, 985-993 (2005). 94. Benjamini,Y., Drai,D., Elmer,G., Kafkafi,N., & Golani,I. Controlling the false discovery rate in behavior genetics research. Behav. Brain Res. 125, 279-284 (2001). 95. Vincenti,M.P. & Brinckerhoff,C.E. Transcriptional regulation of colla- genase (MMP-1, MMP-13) genes in arthritis: integration of complex signal- ing pathways for the recruitment of gene-specific transcription factors. Arthritis Res. 4, 157-164 (2002). 96. Borden,P. & Heller,R.A. Transcriptional control of matrix metallopro- teinases and the tissue inhibitors of matrix metalloproteinases. Crit Rev. Eukaryot. Gene Expr. 7, 159-178 (1997). 97. Klatt,A.R. et al. TAK1 downregulation reduces IL-1beta induced ex- pression of MMP13, MMP1 and TNF-alpha. Biomedecine & Pharmacothe- rapy 60, 55-61 (2006). 98. Klutchko,S.R. et al. 2-Substituted aminopyrido[2,3-d]pyrimidin-7(8H)- ones. structure-activity relationships against selected tyrosine kinases and in vitro and in vivo anticancer activity. J. Med. Chem. 41, 3276-3292 (1998). 99. Chaudhary,N.I. et al. Inhibition of PDGF, VEGF and FGF signalling atte- nuates fibrosis. Eur. Respir. J. 29, 976-985 (2007). Reference List 143 100. Roth,G.J. et al. Design, synthesis, and evaluation of indolinones as triple angiokinase inhibitors and the discovery of a highly specific 6- methoxycarbonyl-substituted indolinone (BIBF 1120). J. Med. Chem. 52, 4466-4480 (2009). 101. Illumina® BeadStudio. http://www.illumina.com . (2010). 102. Lin,S.M., Du,P., Huber,W., & Kibbe,W.A. Model-based variance- stabilizing transformation for Illumina microarray data. Nucleic Acids Res. 36, e11 (2008). 103. Irizarry,R.A. et al. Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics. 4, 249-264 (2003). 104. Du,P., Kibbe,W.A., & Lin,S.M. lumi: a pipeline for processing Illumina microarray. Bioinformatics. 24, 1547-1548 (2008). 105. Huber,W., von,H.A., Sultmann,H., Poustka,A., & Vingron,M. Variance stabilization applied to microarray data calibration and to the quantification of differential expression. Bioinformatics. 18 Suppl 1, S96-104 (2002). 106. Schmid,R. et al. Comparison of Normalization Methods for Illumina BeadChip® HumanHT-12 v3. (submitted). 107. Subramanian,A., Kuehn,H., Gould,J., Tamayo,P., & Mesirov,J.P. GSEA- P: a desktop application for Gene Set Enrichment Analysis. Bioinformatics. 23, 3251-3253 (2007). 108. Mootha,V.K. et al. PGC-1alpha-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes. Nat Genet 34, 267-273 (2003). 109. Kanehisa,M., Goto,S., Furumichi,M., Tanabe,M., & Hirakawa,M. KEGG for representation and analysis of molecular networks involving diseases and drugs. Nucleic Acids Res. (2009). 110. Kanehisa,M., Goto,S., Furumichi,M., Tanabe,M., & Hirakawa,M. KEGG for representation and analysis of molecular networks involving diseases and drugs. Nucleic Acids Res. (2009). 111. Cohen,P. Protein kinases-the major drug targets of the twenty-first century? Nat. Rev. Drug Discov. 1, 309-315 (2002). 112. Zhang,J., Yang,P.L., & Gray,N.S. Targeting cancer with small molecule kinase inhibitors. Nat. Rev. Cancer 9, 28-39 (2009). 113. Bain,J. et al. The selectivity of protein kinase inhibitors: a further up- date. Biochem J. 408, 297-315 (2007). Reference List 144 114. Cohen,P. Protein kinases--the major drug targets of the twenty-first century? Nat. Rev. Drug Discov. 1, 309-315 (2002). 115. Venter,J.C. et al. The sequence of the human genome. Science 291, 1304-1351 (2001). 116. Paolini,G.V., Shapland,R.H., van Hoorn,W.P., Mason,J.S., & Hopkins,A.L. Global mapping of pharmacological space. Nat. Biotechnol. 24, 805-815 (2006). 117. Hieronymus,H. et al. Gene expression signature-based chemical ge- nomic prediction identifies a novel class of HSP90 pathway modulators. Cancer Cell 10, 321-330 (2006). 118. Wei,G. et al. Gene expression-based chemical genomics identifies ra- pamycin as a modulator of MCL1 and glucocorticoid resistance. Cancer Cell 10, 331-342 (2006). 119. Cornwell,P.D., De Souza,A.T., & Ulrich,R.G. Profiling of hepatic gene expression in rats treated with fibric acid analogs. Mutat. Res. 549, 131-145 (2004). 120. Attisano,L. & Labbe,E. TGFbeta and Wnt pathway cross-talk. Cancer Metastasis Rev. 23, 53-61 (2004). 121. Mishra,L., Shetty,K., Tang,Y., Stuart,A., & Byers,S.W. The role of TGF- beta and Wnt signaling in gastrointestinal stem cells and cancer. Oncogene 24, 5775-5789 (2005). 122. Pires-daSilva,A. & Sommer,R.J. The evolution of signalling pathways in animal development. Nat. Rev. Genet. 4, 39-49 (2003). 123. Rao,M. Conserved and divergent paths that regulate self-renewal in mouse and human embryonic stem cells. Dev. Biol. 275, 269-286 (2004). 124. Guo,X. & Wang,X.F. Signaling cross-talk between TGF-beta/BMP and other pathways. Cell Res. 19, 71-88 (2009). 125. Sanchez-Capelo,A. Dual role for TGF-beta1 in apoptosis. Cytokine Growth Factor Rev. 16, 15-34 (2005). 126. Matsuura,I., Wang,G., He,D., & Liu,F. Identification and characteriza- tion of ERK MAP kinase phosphorylation sites in Smad3. Biochemistry 44, 12546-12553 (2005). 127. Kamaraju,A.K. & Roberts,A.B. Role of Rho/ROCK and p38 MAP kinase pathways in transforming growth factor-beta-mediated Smad-dependent growth inhibition of human breast carcinoma cells in vivo. J. Biol. Chem. 280, 1024-1036 (2005). Reference List 145 128. Cordenonsi,M. et al. Links between tumor suppressors: p53 is re- quired for TGF-beta gene responses by cooperating with Smads. Cell 113, 301-314 (2003). 129. Cordenonsi,M. et al. Integration of TGF-beta and Ras/MAPK signaling through p53 phosphorylation. Science 315, 840-843 (2007). 130. Janda,E. et al. Ras and TGF[beta] cooperatively regulate epithelial cell plasticity and metastasis: dissection of Ras signaling pathways. J. Cell Biol. 156, 299-313 (2002). 131. Muraoka,R.S. et al. Increased malignancy of Neu-induced mammary tumors overexpressing active transforming growth factor beta1. Mol. Cell Biol. 23, 8691-8703 (2003). 132. Seton-Rogers,S.E. & Brugge,J.S. ErbB2 and TGF-beta: a cooperative role in mammary tumor progression? Cell Cycle 3, 597-600 (2004). 133. Ueda,Y. et al. Overexpression of HER2 (erbB2) in human breast epi- thelial cells unmasks transforming growth factor beta-induced cell motility. J. Biol. Chem. 279, 24505-24513 (2004). 134. Lehmann,K. et al. Raf induces TGFbeta production while blocking its apoptotic but not invasive responses: a mechanism leading to increased malignancy in epithelial cells. Genes Dev. 14, 2610-2622 (2000). 135. Oft,M. et al. TGF-beta1 and Ha-Ras collaborate in modulating the phenotypic plasticity and invasiveness of epithelial tumor cells. Genes Dev. 10, 2462-2477 (1996). 136. Yue,J. & Mulder,K.M. Requirement of Ras/MAPK pathway activation by transforming growth factor beta for transforming growth factor beta 1 production in a smad-dependent pathway. J. Biol. Chem. 275, 35656 (2000). 137. Ozdamar,B. et al. Regulation of the polarity protein Par6 by TGFbeta receptors controls epithelial cell plasticity. Science 307, 1603-1609 (2005). 138. Hurst,V., IV, Goldberg,P.L., Minnear,F.L., Heimark,R.L., & Vincent,P.A. Rearrangement of adherens junctions by transforming growth factor-beta1: role of contraction. Am. J. Physiol 276, L582-L595 (1999). 139. Rudkin,G.H. et al. Transforming growth factor-beta, osteogenin, and bone morphogenetic protein-2 inhibit intercellular communication and al- ter cell proliferation in MC3T3-E1 cells. J. Cell Physiol 168, 433-441 (1996). 140. Dai,P., Nakagami,T., Tanaka,H., Hitomi,T., & Takamatsu,T. Cx43 me- diates TGF-beta signaling through competitive Smads binding to microtu- bules. Mol. Biol. Cell 18, 2264-2273 (2007). Reference List 146 141. Chanson,M. et al. Gap junctional communication in tissue inflamma- tion and repair. Biochim. Biophys. Acta 1711, 197-207 (2005). 142. Mori,R., Power,K.T., Wang,C.M., Martin,P., & Becker,D.L. Acute down- regulation of connexin43 at wound sites leads to a reduced inflammatory response, enhanced keratinocyte proliferation and wound fibroblast migra- tion. J. Cell Sci. 119, 5193-5203 (2006). 143. Boland,S. et al. TGF beta 1 promotes actin cytoskeleton reorganization and migratory phenotype in epithelial tracheal cells in primary culture. J. Cell Sci. 109 ( Pt 9), 2207-2219 (1996). 144. Lee,J., Ko,M., & Joo,C.K. Rho plays a key role in TGF-beta1-induced cytoskeletal rearrangement in human retinal pigment epithelium. J. Cell Physiol. 216, 520-526 (2008). 145. Kretschmer,A. et al. Differential regulation of TGF-beta signaling through Smad2, Smad3 and Smad4. Oncogene 22, 6748-6763 (2003). 146. Ranganathan,P. et al. Expression profiling of genes regulated by TGF- beta: differential regulation in normal and tumour cells. BMC Genomics 8, 98 (2007). 147. Dawes,L.J., Elliott,R.M., Reddan,J.R., Wormstone,Y.M., & Worm- stone,I.M. Oligonucleotide microarray analysis of human lens epithelial cells: TGFbeta regulated gene expression. Mol. Vis. 13, 1181-1197 (2007). 148. Padua,D. et al. TGFbeta primes breast tumors for lung metastasis seeding through angiopoietin-like 4. Cell 133, 66-77 (2008). 149. Wu,X., Ma,J., Han,J.D., Wang,N., & Chen,Y.G. Distinct regulation of gene expression in human endothelial cells by TGF-beta and its receptors. Microvasc. Res. 71, 12-19 (2006). 150. Smyth,G.K., Yang,Y.H., & Speed,T. Statistical issues in cDNA microarray data analysis. Methods Mol. Biol. 224, 111-136 (2003). 151. Anderson,E.M. et al. Experimental validation of the importance of seed complement frequency to siRNA specificity. RNA (2008). 152. Birmingham,A. et al. 3' UTR seed matches, but not overall identity, are associated with RNAi off-targets. Nat. Methods 3, 199-204 (2006). 153. Aggarwal,S. et al. Melanoma differentiation-associated gene-7/IL-24 gene enhances NF-kappa B activation and suppresses apoptosis induced by TNF. J Immunol. 173, 4368-4376 (2004). 154. Svoboda,P. Off-targeting and other non-specific effects of RNAi expe- riments in mammalian cells. Curr. Opin. Mol. Ther. 9, 248-257 (2007). Reference List 147 155. Bramsen,J.B. et al. A large-scale chemical modification screen identi- fies design rules to generate siRNAs with high activity, high stability and low toxicity. Nucleic Acids Res. 37, 2867-2881 (2009). 156. Di Guglielmo,G.M., Le,R.C., Goodfellow,A.F., & Wrana,J.L. Distinct en- docytic pathways regulate TGF-beta receptor signalling and turnover. Nat. Cell Biol. 5, 410-421 (2003). 157. De Renzis,S., Sonnichsen,B., & Zerial,M. Divalent Rab effectors regu- late the sub-compartmental organization and sorting of early endosomes. Nat. Cell Biol. 4, 124-133 (2002). 158. Schmidt,D.M. & Ernst,J.D. A fluorometric assay for the quantification of RNA in solution with nanogram sensitivity. Anal. Biochem. 232, 144-146 (1995). 159. Gentleman,R.C. et al. Bioconductor: open software development for computational biology and bioinformatics. Genome Biol. 5, R80 (2004). 160. Smyth,G.K. Linear models and empirical bayes methods for assessing differential expression in microarray experiments. Stat. Appl. Genet. Mol. Biol. 3, (2004). 161. Lin,D. et al. Testing for trends in dose-response microarray experi- ments: a comparison of several testing procedures, multiplicity and resam- pling-based inference. Stat. Appl. Genet. Mol. Biol. 6, Article26 (2007). Danksagung 148 Danksagung Knowledge is in the end based on acknowledgement. Ludwig Wittgenstein Die vorliegende Arbeit wäre ohne die Unterstützung zahlreicher Personen nicht mög- lich gewesen. Der größte Dank gebührt meinem Betreuer, Dr. Detlev Mennerich, nicht nur für die vorbildhafte, wissenschaftliche Anleitung und fachliche Diskussion, sondern auch dafür, immer ein offenes Ohr für meine Probleme gehabt zu haben und seinen un- ermüdlichen Einsatz diese dann auch zu lösen. Ferner, möchte ich mich bei ihm für die unkomplizierte und freundschaftliche Art der Betreuung bedanken, die mich auch in schwierigen Phasen motiviert hat und mich nie den Spaß an meiner Arbeit verlieren lies. Zu guter Letzt möchte ich ihm auf diesem Wege noch für die erteilten Lehrstunden in Doppelkopf danken. Durch seine Hilfe war es mir möglich, mich auch auf diesem Gebiet signifikant weiterzuentwickeln. Des Weiteren bin ich meinen Laborkollegen Karoline Schwarz, Dagmar Knebel, Su- sanne Siewert, Dr. Tanja Fauti, Dr. Svenja Weikert, Martin Baur und Werner Rust zu großem Dank verpflichtet. Ich möchte mich bei Ihnen sowohl für die umfassende Einführung in die Nukleinsäureanalytik als auch für ihre tatkräftige Unterstützung bedanken. Darüber hinaus, war die gute Arbeitsatmosphäre in Labor mit Sicherheit auch einer der Eckpfeiler für das gelingen dieser Arbeit. Mein besonderer Dank gilt auch Dr. Tobias Hildebrandt, Dr. Jörg Rippmann und Dr. Sebastian Kreuz. Jedem einzelnen habe ich eine Erweiterung meines wissenschaftli- chen Erfahrungsschatzes zu verdanken. Zusätzlich zu den zahlreichen aufschlussrei- Danksagung 149 chen Diskussionen bekam ich hier auch immer gute Ratschläge und tatkräftige Un- terstützung zur Bewältigung der jeweiligen Fragestellung. Auch nicht vergessen will ich meine Bioinformatikerkollegen Dr. Carina Ittrich, Dr. Katrin Fundel-Clemens, Dr. Eric Simon, Dr. Fabian Birzele und Dr. Karsten Quast. Ich möchte mich zum einen für Rat und Tat bei den Analysen aber auch für so profane Dinge wie das Lösen von diversen Computerproblemen bedanken. Besonders her- vorheben möchte ich hierbei meine Doktorandenkollegin und „Leidensgefährtin“ Ramona Schmid, da ihr Beitrag maßgeblich zum Gelingen des „Phenocopy Projects“ beigetragen hat. Durch ihren Einsatz war es möglich, die wirren Gedankengänge und Ideen eines Biologen in mathematische Bahnen zu lenken. Ich möchte ebenfalls ausnahmslos allen Kollegen aus der Genomics Gruppe danken. Die äußerst angenehme Atmosphäre und der gute Zusammenhalt, auch über die Arbeit hinaus, schufen ideale Arbeitsbedingungen. Mein besonderer Dank gilt hier natürlich auch dem „Chefgenomiker“ Dr. Andreas Weith. Zum einen dafür mir die Möglichkeit gegeben zu haben, diese Arbeit in der Genomics Gruppe anzufertigen und zum anderen für Diskussionen, Ratschläge und die erhaltenen Freiräume bei der Realisierung des Projekts. In diesem Zusammenhang möchte ich mich auch bei PD Dr. Florian Gantner bedan- ken, der dem Projekt immer wohlwollend gegenüber stand und ohne dessen Zu- stimmung eine derartige Studie nie möglich gewesen wäre. Mein spezieller Dank gilt an dieser Stelle natürlich auch meinem Doktorvater Prof. Dr. Roland Kontermann, der ohne zu Zögern meine Betreuung seitens der Universität Stuttgart übernahm und die Arbeit immer sehr interessiert begleitet und unterstützt hat. Danksagung 150 Schließlich, möchte ich mich noch bei all meinen Freunden und meiner Familie be- danken. Besonders hervorheben möchte ich hier die stetige Unterstützung und den großen Rückhalt durch meine Eltern Marlies und Manfred, meinen Bruder Philipp und meine Freundin Katja, die mir auch in schwierigen Zeiten immer die Kraft gege- ben haben, mich allen Herausforderungen zu stellen. Erklärung 151 Erklärung Ich erkläre hiermit, dass ich die vorliegende Arbeit ohne unzulässige Hilfe Dritter und ohne Benutzung anderer als der angegebenen Hilfsmittel angefertigt habe; die aus fremden Quellen direkt oder indirekt übernommenen Gedanken sind als solche kenntlich gemacht. Patrick Baum Biberach, den 15.04.2010 Lebenslauf 152 Lebenslauf Persönliche Angaben Geburtsdatum/-ort: 17.06.1978, Mutlangen Familienstand: Ledig Staatsangehörigkeit: Deutsch Schulische Ausbildung 09/89 bis 08/98 Allgemeine Hochschulreife am Limes-Gymnasium in Welzheim Zivildienst 08/98 bis 09/99 Anstellung beim Bund für Umwelt und Naturschutz in Deutsch- land (BUND) in Welzheim Studium 10/99 bis 08/00 Studium der Pharmazie an der Universität Würzburg 10/00 bis 10/06 Studium der technischen Biologie an der Universität Stuttgart in der Fakultät für Geo- und Biowissenschaften 02/05 bis 08/05 Studienarbeit am Tyndall National Institute, Cork, Irland mit dem Thema „The optimisation of the hybridisation of a PDITC modified silicon surface for use in biochip development” 09/05 bis 10/06 Diplomarbeit am Institut für Zellbiologie und Immunologie in der Arbeitsgruppe Biomedical Engineering mit dem Thema „Herstellung und Charakterisierung von single-chain Fv- Immunliposomen selektiv für das Fibroblast Activation Protein“ Doktorarbeit 02/07 bis 05/10 Doktorarbeit bei Boehringer Ingelheim Pharma GmbH & Co KG, Abteilung Atemwegsforschung, Group Genomics mit dem Thema „Phenocopy – A Strategy to Qualify Chemical Compounds during Hit-to-Lead and/or Lead Optimization” Lebenslauf 153 Publikationen Baum P., Müller D., Rüger R. & Kontermann R.E. Single-chain Fv immunoliposomes for the targeting of fibroblast activation protein- expressing tumor stromal cells (Journal of Drug Targeting, 2007, Vol. 15, No. 6, Pages 399-406) Baum P., Fundel-Clemens K., Kreuz S., Kontermann R.E., Weith A., Mennerich D. & Rippmann J.F. Off-Target Analysis of Control siRNA Molecules Reveals Important Differences in the Cytokine Profile and Inflammation Response of Human Fibroblasts (Oligonucleotides, 2010,Vol. 20, No.1, Pages 17-26) Birzele F., Schaub J., Rust W., Clemens C., Baum P., Kaufmann H., Weith A. Schulz T.W. & Hildebrandt T. Into the unknown: Expression profiling without genome sequence information in CHO by next generation sequencing (Nucleic Acids Research, doi:10.1093/nar/gkq116) Roth G.J., Heckel A., Brandl T., Grauert M., Hoerer S., Kley J., Schnapp G., Baum P., Mennerich D., Schnapp A. & Park J.E. Design, Synthesis and Evaluation of Indolinones as Inhibitors of the Transforming Growth Factor Beta Receptor I (TGF-βR1) (submitted) Schmid R., Baum P., Ittrich C., Fundel-Clemens K., Huber W., Brors B., Eils R., Weith A., Mennerich D. & Quast K. Comparison of Normalization Methods for Illumina BeadChip® HumanHT-12 v3 (submitted) Baum P., Schmid R., Ittrich C., Rust W., Fundel-Clemens K., Siewert S., Baur M., Mara L., Gruenbaum L., Heckel A., Eils R., Kontermann R.E., Roth G.J., Ganter F., Schnapp A., Park J.E., Weith A., Quast K. & Mennerich D. Phenocopy – A Strategy to Qualify Chemical Compounds during Hit-to-Lead and/or Lead Optimization (submitted) Schmid R., Baum P., Ittrich C., Fundel-Clemens K., Lämmle B., Birzele F., Weith A., Brors B., Eils R., Mennerich D. & Quast, K. Detecting Meaningful Protein Interactions by Accumulating Evidence (submitted)