To effectively understand and treat complex diseases such as cancer, mathematical and statistical modeling is essential if one wants to represent and characterize the interactions among the different regulatory components that govern the underlying decision making process. Like in any other complex decision making networks, the regulatory power is not evenly distributed among its individual members, but rather concentrated in a few high power "commanders". In biology, such commanders are usually called masters or canalizing genes. Characterizing and detecting such genes are thus highly valuable for the treatment of cancer. Chapter II is devoted to this task, where we present a Bayesian framework to model pathway interactions and then study the behavior of master genes and canalizing genes. We also propose a hypothesis testing procedure to detect a "cut" in pathways, which is useful for discerning drugs' therapeutic effect. In Chapter III, we shift our focus to the understanding of the mechanisms of action (MOA) of cancer drugs. For a new drug, the correct understanding of its MOA is a key step for its application to cancer treatments. Using the Green Fluorescent Protein technology, researchers have been able to track various reporter genes from the same cell population for an extended period of time. Such dynamic gene expression data forms the basis for drug similarity comparisons. In Chapter III, we design an algorithm that can identify mechanistic similarities in drug responses, which leads to the characterization of their respective MOAs. Finally, in the course of drug MOA study, we observe that cells in a hypothetical homogeneous population do not respond to drug treatments in a uniform and synchronous way. Instead, each cell makes a large shift in its gene expression level independently and asynchronously from the others. Hence, to systematically study such behavior, we propose a mathematical model that describes the gene expression dynamics for a population of cells after drug treatments. The application of this model to dose response data proviodes us new insights of the dosing effects. Furthermore, the model is capable of generating useful hypotheses for future experimental design.

Indiana University-Purdue University Indianapolis (IUPUI) / In eukaryotes, genetic information is encoded by DNA, transcribed to precursor messenger RNA (pre-mRNA), processed into mature messenger RNA (mRNA), and translated into functional proteins. Splicing of pre-mRNA is an important epigenetic process that alters the function of proteins through modifying the exon structure of mature mRNA transcripts and is known to greatly contribute to diversity of the human proteome. The vast majority of human genes are expressed through multiple transcript isoforms. Expression of genes through splicing of pre-mRNA plays crucial roles in cellular development, identity, and processes. Both the identity of genes selected for transcription and the specific transcript isoforms that are expressed are essential for normal cellular function. Deviations in gene expression or isoform proportion can be an indication or the cause of disease. RNA sequencing (RNAseq) is a high-throughput next-generation sequencing technology that allows for the interrogation of gene expression on a massive scale. RNAseq generates short sequences that reflect pieces of mRNAs present in a sample. RNAseq can therefore be used to explore differences in gene expression, reveal transcript isoform identities and compare changes in isoform proportions. In this dissertation, I design and apply advanced analysis techniques to RNAseq, phenotypic and drug response data to investigate disease mechanisms and drug sensitivity. Research Goals: The work described in this dissertation accomplishes 4 aims. Aim 1) Evaluate the gene expression signature of concussion in collegiate athletes and identify potential biomarkers for response and recovery. Aim 2) Implement a machine-learning algorithm to determine if splicing can predict drug response in cancer cell lines. Aim 3) Design a fast, scalable method to identify differentially spliced events related to cancer drug response. Aim 4) Construct a drug-splicing network and use a systems biology approach to search for similarities in underlying splicing events.

Thiopurine methyltransferase (TPMT) risk alleles (mainly *2,*3B, *3C and *3A) are the major determinants of interindividual differences in the severe toxicity or efficacy of 6-mercaptopurine (6MP) during the treatment of childhood acute lymphoblastic leukemia (ALL). The frequencies of these risk alleles, known to functionally impair TPMT activity, were investigated among 167children with ALL and 206 healthy adult controls in Turkish population by using allele specific PCR and PCRRFLP methods. TPMT*3A and TPMT*3C were the only deficiency alleles detected in Turkish population with an allele frequency of 0.5% for both. The total frequency of mutant TPMT alleles in Turkish population (1.0%) was found to be significantly lower than those of other Caucasian populations (5.3-7.0%), but it was found to be very similar to Kazak population (1.2%) which is also Caucasian in ethnic origin. v In the patient group, two individuals were found to be heterozygote for *3C and *3A allele. One individual was homozygous mutant (*3B/*3C). In this study, the clinical histories of the patients with TPMT defects were examined retrospectively from hospital records. The patients with heterozygous or homozygous mutant genotypes had systematically developed severe neutropenia, infection and some other specific conditions (like lesions around mouth, oral herpes and high fever) when they were administered with 6MP during the therapy. This study provides the first data on the frequency of common TPMT risk alleles in the Turkish population, based on analysis of pediatric patients with ALL. The results would contribute valuable information to the public health, as more clinicians and patients become aware of the importance of TPMT polymorphisms, less patients will suffer from 6MP related adverse effects. In addition, in this study two genes EPHX1-microsomal epoxide hydrolase (exon 3 and exon 4 polymorphisms) and SULT1A1*2 variant &ndash / sulfotransferase 1A1, either alone or in combination were investigated as risk modifiers in the development of childhood acute lymphoblastic leukemia due to their dual role (activation/detoxification) in the metabolism of various carcinogens. Also interactions of these polymorphisms with non-genetic risk factors (parental smoking exposure and parental age at conception) were investigated. The conclusion inferred from results was that only genetically reduced EPHX1 activity (homozygous mutant genotype for EPHX1 exon 3 polymorphism and some specific genotype combinations with exon 4 polymorphism) was found to be significantly associated with the risk of childhood ALL.

Final publication is available at http://www.sciencedirect.com/science/article/pii/S1873506115001191 / 京都大学 / 0048 / 新制・課程博士 / 博士(医学) / 甲第19579号 / 医博第4086号 / 新制||医||1013(附属図書館) / 32615 / 京都大学大学院医学研究科医学専攻 / (主査)教授 井上 治久, 教授 髙橋 良輔, 教授 岩田 想 / 学位規則第4条第1項該当 / Doctor of Medical Science / Kyoto University / DFAM

The last decade has witnessed a tremendous increase in the amount of available biological data. Different technologies for measuring the genome, epigenome, transcriptome, proteome, metabolome, and microbiome in different organisms are producing large amounts of high-dimensional data every day. High-dimensional data provides unprecedented challenges and opportunities to gain a better understanding of biological systems. Unlike other data types, biological data imposes more constraints on researchers. Biologists are not only interested in accurate predictive models that capture complex input-output relationships, but they also seek a deep understanding of these models. In the last few years, deep models have achieved better performance in computational prediction tasks compared to other approaches. Deep models have been extensively used in processing natural data, such as images, text, and recently sound. However, application of deep models in biology is limited. Here, I propose to use deep models for output prediction, dimension reduction, and feature selection of biological data to get better interpretation and understanding of biological systems. I demonstrate the applicability of deep models in a domain that has a high and direct impact on health care. In this research, novel deep learning models have been introduced to solve pressing biological problems. The research shows that deep models can be used to automatically extract features from raw inputs without the need to manually craft features. Deep models are used to reduce the dimensionality of the input space, which resulted in faster training. Deep models are shown to have better performance and less variant output when compared to other shallow models even when an ensemble of shallow models is used. Deep models are shown to be able to process non-classical inputs such as sequences. Deep models are shown to be able to naturally process input sequences to automatically extract useful features. / Ph. D.

Most known drug targets and metabolizing enzymes are located inside cells. Interactions with these proteins are determined by intracellular unbound drug concentrations. Assessing intracellular drug exposure is technically challenging, but essential for predicting pharmacokinetic, pharmacological, and toxicological profiles of new drugs. This thesis aims at establishing and applying a straightforward methodology to measure intracellular unbound drug concentrations. This was achieved by separately measuring cellular drug binding (fu,cell), and total intracellular drug accumulation (Kp). This allowed the calculation of intracellular drug bioavailability (Fic), which represents the fraction of the concentration added to the cells that is unbound in the cell interior. The methodology was initially developed in HEK293 cells, where the Fic of 189 drug-like compounds was measured. Binding to HEK293 cells was governed by compound lipophilicity and was correlated with binding to more complex systems, such as hepatocytes and brain. Due to negligible expression of drug transporters, Fic in this cell line was consistent with pH-dependent subcellular sequestration of lipophilic cations in low pH compartments. The methodology was then applied to study the effects of drug transporters on Fic. The uptake transporter OATP1B1 increased the Fic of its substrates in a concentration-dependent manner. In contrast, the Fic of P-gp substrates was decreased when P-gp was present. In human hepatocytes, the methodology allowed the determination of Fic without prior knowledge of transporter mechanisms or metabolic activity. Finally, the methodology was applied to measure the impact of Fic on target binding and cellular drug response. Intracellular concentrations of active metabolites of pro-drugs targeting the intracellular target thymidylate synthase were in agreement with the level of binding to this target. Further, high Fic was generally required for kinase and protease inhibitors to be active in cellular assays. In conclusion, the methodology can be used to predict if new drug candidates reach their intracellular targets in sufficient amounts. Furthermore, the methodology can improve in vitro predictions of drug clearance and drug-drug interactions, by measuring the drug available for intracellular enzymes. Finally, this work can be expanded to other xenobiotics, e.g., to predict their intracellular toxicity.

The ras genes, including Harvey ras (H-ras) and Kirsten ras (K-ras), were among the first oncogenes discovered, and are the most commonly mutated oncogenes in human cancer. The H-ras and K-ras proteins are 85% identical and share considerable functional overlap. However, there is increasing evidence for functional differences between the two proteins that may impart different properties to tumors arising from mutations in these two genes. To study the functional differences between H-ras and K-ras in an in vivo setting, we used two different transgenic mouse tumor models, MMTV-H-ras and MMTV-K-ras mice. The MMTV-H-ras mice were originally developed in Dr. Leder's lab and have been well characterized with regard to tumor properties. We created a similar line of transgenic mice expressing mutant K-ras (G12V) under the control of the MMTV promoter. Female mice of both lines develop primarily mammary tumors. We compared differences between the H-ras and K-ras lines with regard to age of tumor onset, rate of tumor growth, and rates of tumor proliferation and apoptosis. The tumors were also characterized by microarray analysis to look for genes that are differentially expressed in the two tumor types. Finally, the response of tumors to two common chemotherapeutic agents, doxorubicin and taxol, was also measured. We found that tumors in the MMTV-H-ras and MMTV-K-ras mice were similar with respect to several tumor properties, including age of onset, histopathology, and proliferation and apoptotic indices. While tumors from mice of these two genotypes clustered separately in an unsupervised analysis of gene expression profiles, the differentially expressed genes did not fall within any well-defined signaling pathways. However, drug studies indicated differences in response to doxorubicin between the two isoforms, with H-ras tumors responding better than K-ras tumors. In conclusion, our studies point to specific differences between H-ras and K-ras that may represent novel signaling pathways not currently known to be regulated by Ras. In spite of the few differences in properties of tumors arising from H-ras and K-ras mutation, there might be differences in response to chemotherapeutic agents that could have clinical significance.

p53 is an important tumor suppressor gene which is mutated in ~50% of all human cancers. Some of the p53 mutants appear to have acquired novel functions beyond merely losing wild-type functions. To investigate these gain-of-function effects in vivo, we interbred MMTV-v-Ha-ras transgenic mice to either p53-/- knock-out mice or p53R172H/+ knock-in mice to generate mice of three different genotypes: MMTV-ras, MMTV-ras/p53-/-, and MMTV-ras/p53R172H/R172H. Male mice of each of these genotypes were characterized with regard to age of salivary tumor onset and the tumors were characterized with regard to mean growth rates, proliferation fraction, apoptotic levels, and tumor histopathology, as well as responses to doxorubicin treatment. Microarray analysis was also performed to profile gene expression.The MMTV-ras/p53-/- and MMTV-ras/p53R172H/R172H mice display similar properties in age of tumor onset, tumor growth rates, and tumor histopathology, as well as response to doxorubicin. However, a subset of genes show differential expression between the two groups of tumor , and do not appear to be regulated by wild-type p53. At the same time, the MMTV-ras/p53R172H/R172H and MMTV-ras/p53+/+ tumors share similar expression levels of a group of genes that are differentially expressed in the MMTV-ras/p53-/- tumors. Thus, the gain-of-function effects may be caused in part by perturbed regulation of genes not normally regulated by wild-type p53, in addition to imbalances in the regulation of normal p53 target genes.

Les cancers de la prostate sont classés en deux catégories. Les cancers de haut grade se distinguent des cancers de bas grade par une plus forte agressivité et un pronostic plus mauvais. Lorsqu’ils deviennent résistants à l’hormonothérapie, les cancers de haut grade sont traités par une chimiothérapie basée sur les taxanes. Néanmoins, les taux de réponse restent faibles. Il existe donc un réel besoin quant à l'identification d'alternatives thérapeutiques qui soient spécifiques de ce type de tumeur. Dans cette optique, notre travail a été de proposer une telle alternative par une approche qui prenne en compte la génétique spécifique des cancers de haut grade. Nous avons exploité une signature de 86 gènes dont le niveau d’expression permet de discriminer entre les tumeurs de haut et de bas grade. Par une approche in silico originale utilisant la banque de données du NCI, nous avons identifié 382 corrélations entre le niveau d’expression de 50 gènes et la sensibilité à 139 agents antiprolifératifs. Parmi ces corrélations, nous avons identifié une signature de 9 gènes qui est spécifique de la réponse à l’oxaliplatine. Cette signature a été confirmée sur le plan fonctionnel dans les lignées cancéreuses prostatiques DU145 et LNCaP. Nous avons donc fourni la preuve de concept que notre approche permet d’identifier de nouvelles molécules pouvant être utilisées en alternative aux taxanes pour traiter spécifiquement les cancers de haut grade. Cette stratégie permet aussi d’identifier de nouveaux marqueurs (gènes) régulant la sensibilité à certains médicaments. Nos résultats démontrent par exemple le rôle des gènes SHMT, impliqués dans la régulation du métabolisme monocarboné, dans la sensibilité spécifique à l’oxaliplatine par un mécanisme qui fait intervenir, du moins en partie, une dérégulation du niveau de méthylation global de l’ADN. / Prostate cancers are classified in two categories. High grade cancers are distinguished from low grade cancers by their higher agressivity and worse prognostic. When they become refractory to hormone therapy, high grade cancers are treated with a taxane-based chemotherapy. However, response rates remain low. Therefore, there is a real need for the discovery of new therapeutic alternatives which are specific for this type of tumors. For that purpose, our work aimed at proposing such an alternative with a strategy that took into account the high grade genetic background. We exploited a signature of 86 genes for which expression level could distinguish between low grade and high grade tumours. With an original in silico approach, we searched the NCI databases and identified 382 correlations between 50 genes and the sensitivity to 139 antiproliferative agents. Among these, a signature of 9 genes was able to specifically predict cell response to oxaliplatin. This signature was validated at the functional level in two prostate cancer cell lines, DU145 and LNCaP. We have thus provided the proof-of-concept that our approach allows the identification of new drugs that can be used alternatively to taxanes in order to specifically treat high grade prostate cancers. This strategy also allows the identification of new markers (genes) regulating the sensitivity to various drugs. Our results demonstrate for example the implication of SHMT genes, which are involved in the regulation of the one-carbon metabolism, in the specific sensitivity to oxaliplatin, by a mechanism which involves, at least in part, the deregulation of the global level of DNA methylation.

O câncer de ovário é uma importante causa de mortalidade. O subtipo seroso de alto grau é o mais frequente e caracteriza-se por comportamento agressivo, com crescimento rápido e metástase precoce. A falta de ferramentas para diagnóstico em estádios iniciais e a insuficiência de resposta à quimioterapia convencional são dois principais obstáculos para o manejo do câncer seroso de ovário. A detecção precoce neste tumor é complicada por sintomas inespecíficos e ausência de biomarcadores confiáveis. Além disso, o desenvolvimento de resistência à quimioterapia é um desafio para o tratamento, que é geralmente baseado na combinação de platina e paclitaxel. A influência do microambiente tumoral na resposta terapêutica ainda é pouco conhecida. No entanto, há evidências crescentes de que a resposta imunológica pré-existente pode estar relacionada com a variação da sensibilidade à quimioterapia. O microambiente imunorreativo foi associado à melhor prognóstico no câncer do ovário seroso de alto grau em recente estudo canadense. Proteínas da família de Transdutores de Sinal e Ativadores da Transcrição (STATs) participam da regulação de citocinas e são determinantes nas respostas imunes no microambiente tumoral, podendo promover ou inibir o crescimento tumoral. Dados recentes mostram que a expressão elevada de um dos reguladores de STAT, STAT1 atua na tumorigênese do câncer de ovário, facilitando a resposta imune e, potencialmente, alterando a resposta à quimioterapia. Para avaliar o papel da expressão de STAT1 como biomarcador preditivo em 65 pacientes brasileiras com câncer seroso de ovário, examinamos os níveis de STAT1 por imunoistoquímica, e analisamos se houve correlação entre expressão dessa proteína e resposta clínica. Alta expressão de STAT1 foi significativamente associada maior intervalo livre de doença (P=0,0256) e maior sobrevida global (P=0,0193). Estes achados da coorte brasileira, com tempo de seguimento maior que cinco anos, confirmam a associação entre alta expressão de STAT1 e melhor resposta à quimioterapia, e fornecem validação adicional desta proteína como um biomarcador preditivo. Além disso, estes resultados chamam atenção para a possibilidade de utilizar a via de STAT1 para o desenvolvimento de novos medicamentos imunomoduladores, que poderiam melhorar a resposta ao tratamento / Ovarian cancer is a major cause of mortality worldwide. The most frequent subtype is high grade serous, which is characterized by aggressive behavior with rapid growth and early metastasis. Lack of early diagnostic tools and failure of response to conventional chemotherapies are two major impediments to serous ovarian cancer management. Early detection in initial stages is complicated by non-specific symptoms and lack of reliable biomarkers. In addition, development of chemotherapy resistance is a challenge for treatment, which is generally based on combination of platinum and paclitaxel. The influence of the microenvironment of the tumor on therapeutic response is still unknown. However, there is increasing evidence that a pre-existing immunological response may be related to variation in chemotherapy sensitivity. The immunoreactive microenvironment has been shown to be associated with better prognosis in high grade serous ovarian cancer in a recent Canadian study. The Signal Transducer and Activator of Transcription (STAT) proteins regulate cytokines and are central in determining whether immune responses in the tumor microenvironment promote or inhibit cancer. Recent data show that high expression of one of the STAT regulators, STAT1, operates in ovarian cancer tumorigenesis, facilitating immune response and potentially altering response to chemotherapy. To evaluate the role of STAT1 expression as a predictive biomarker in 65 Brazilian serous ovarian cancer patients, we examined STAT1 levels by immunohistochemistry to determine if there was correlation between expression of this protein and clinical response. High expression of STAT1 was significantly associated with both improved disease-free survival (P=0,0256) and overall survival (P=0,0193). These findings from a Brazilian cohort after more than five years of follow up confirm the association of high STAT1 expression with better response to chemotherapy, and provide additional validation of this protein as a predictive biomarker. Moreover these results draw attention to the possibility of utilizing the STAT1 pathway for the development new immunomodulator drugs, that could enhance response to treatment