Cette thèse est consacrée à la modélisation qualitative des réseaux biologiques pour l'innovation thérapeutique. Elle étudie comment utiliser les réseaux Booléens, et comment les améliorer, afin d'identifier des cibles thérapeutiques au moyen d'approches in silico. Elle se compose de deux travaux : i) un algorithme exploitant les attracteurs des réseaux Booléens pour l'identification in silico de cibles dans des modèles Booléens de réseaux biologiques pathologiquement perturbés, et ii) une amélioration des réseaux Booléens dans leur capacité à modéliser la dynamique des réseaux biologiques grâce à l'utilisation des opérateurs de la logique floue et grâce au réglage des arrêtes. L'identification de cibles constitue l'une des étapes de la découverte de nouveaux médicaments et a pour but d'identifier des biomolécules dont la fonction devrait être thérapeutiquement modifiée afin de lutter contre la pathologie considérée. Le premier travail de cette thèse propose un algorithme pour l'identification in silico de cibles par l'exploitation des attracteurs des réseaux Booléens. Il suppose que les attracteurs des systèmes dynamiques, tel que les réseaux Booléens, correspondent aux phénotypes produits par le système biologique modélisé. Sous cette hypothèse, et étant donné un réseau Booléen modélisant une physiopathologie, l'algorithme identifie des combinaisons de cibles capables de supprimer les attracteurs associés aux phénotypes pathologiques. L'algorithme est testé sur un modèle Booléen du cycle cellulaire arborant une inactivation constitutive de la protéine du rétinoblastome, tel que constaté dans de nombreux cancers, tandis que ses applications sont illustrées sur un modèle Booléen de l'anémie de Fanconi. Les résultats montrent que l'algorithme est à même de retourner des combinaisons de cibles capables de supprimer les attracteurs associés aux phénotypes pathologiques, et donc qu'il réussit l'identification in silico de cibles proposée. En revanche, comme tout résultat in silico, il y a un pont à franchir entre théorie et pratique, requérant ainsi une utilisation conjointe d'approches expérimentales. Toutefois, il est escompté que l'algorithme présente un intérêt pour l'identification de cibles, notamment par l'exploitation du faible coût des approches computationnelles, ainsi que de leur pouvoir prédictif, afin d'optimiser l'efficience d'expérimentations coûteuses. La modélisation quantitative en biologie systémique peut s'avérer difficile en raison de la rareté des détails quantitatifs concernant les phénomènes biologiques, particulièrement à l'échelle subcellulaire, l'échelle où les médicaments interagissent avec leurs cibles. Une alternative permettant de contourner cette difficulté est la modélisation qualitative étant donné que celle-ci ne requiert que peu ou pas d'informations quantitatives. Parmi les méthodes de modélisation qualitative, les réseaux Booléens en sont l'une des plus populaires. Cependant, les modèles Booléens autorisent leurs variables à n'être évaluées qu'à vrai ou faux, ce qui peut apparaître trop simpliste lorsque des processus biologiques sont modélisés. En conséquence, le second travail de cette thèse propose une méthode de modélisation dérivée des réseaux Booléens où les opérateurs de la logique floue sont utilisés et où les arrêtes peuvent être réglées. Les opérateurs de la logique floue permettent aux variables d'être continues, et ainsi d'être plus finement évaluées qu'avec des méthodes de modélisation discrètes tel que les réseaux Booléens, tout en demeurant qualitatives. De plus, dans le but de considérer le fait que certaines interactions peuvent être plus lentes et/ou plus faibles que d'autres, l'état des arrêtes est calculé afin de moduler en vitesse et en force le signal qu'elles véhiculent. La méthode proposée est illustrée par son implémentation sur un petit échantillon de la signalisation du récepteur au facteur de croissance épidermique... [etc] / This thesis is devoted to the qualitative modeling of biological networks for therapeutic innovation. It investigates how to use the Boolean network formalism, and how to enhance it, for identifying therapeutic targets through in silico approaches. It is composed of two works: i) an algorithm using Boolean network attractors for in silico target identification in Boolean models of pathologically disturbed biological networks, and ii) an enhancement of the Boolean network formalism in modeling the dynamics of biological networks through the incorporation of fuzzy operators and edge tuning. Target identification, one of the steps of drug discovery, aims at identifying biomolecules whose function should be therapeutically altered in order to cure the considered pathology. The first work of this thesis proposes an algorithm for in silico target identification using Boolean network attractors. It assumes that attractors of dynamical systems, such as Boolean networks, correspond to phenotypes produced by the modeled biological system. Under this assumption, and given a Boolean network modeling a pathophysiology, the algorithm identifies target combinations able to remove attractors associated with pathological phenotypes. It is tested on a Boolean model of the mammalian cell cycle bearing a constitutive inactivation of the retinoblastoma protein, as seen in cancers, and its applications are illustrated on a Boolean model of Fanconi anemia. The results show that the algorithm returns target combinations able to remove attractors associated with pathological phenotypes and then succeeds in performing the proposed in silico target identification. However, as with any in silico evidence, there is a bridge to cross between theory and practice, thus requiring it to be used in combination with wet lab experiments. Nevertheless, it is expected that the algorithm is of interest for target identification, notably by exploiting the inexpensiveness and predictive power of computational approaches to optimize the efficiency of costly wet lab experiments. Quantitative modeling in systems biology can be difficult due to the scarcity of quantitative details about biological phenomenons, especially at the subcellular scale, the scale where drugs interact with there targets. An alternative to escape this difficulty is qualitative modeling since it requires few to no quantitative information. Among the qualitative modeling approaches, the Boolean network formalism is one of the most popular. However, Boolean models allow variables to be valued at only true or false, which can appear too simplistic when modeling biological processes. Consequently, the second work of this thesis proposes a modeling approach derived from Boolean networks where fuzzy operators are used and where edges are tuned. Fuzzy operators allow variables to be continuous and then to be more finely valued than with discrete modeling approaches, such as Boolean networks, while remaining qualitative. Moreover, to consider that some interactions are slower and/or weaker relative to other ones, edge states are computed in order to modulate in speed and strength the signal they convey. The proposed formalism is illustrated through its implementation on a tiny sample of the epidermal growth factor receptor signaling pathway. The obtained simulations show that continuous results are produced, thus allowing finer analysis, and that modulating the signal conveyed by the edges allows their tuning according to knowledge about the modeled interactions, thus incorporating more knowledge. The proposed modeling approach is expected to bring enhancements in the ability of qualitative models to simulate the dynamics of biological networks while not requiring quantitative information. The main prospect of this thesis is to use the proposed enhancement of Boolean networks to build a version of the algorithm based on continuous dynamical systems...[etc]

Mutations in the hereditary breast/ovarian cancer genes BRCA1/2 were shown to be synthetic lethal with poly(ADP-ribose) polymerase inhibitors (PARPi). This toxicity is assumed to derive from PARPi-induced DNA double strand breaks (DSBs) that necessitate BRCA function in homologous recombination (HR) and/or fork protection (FP). However, PARPi accelerates replication forks. While high-speed replication could cause DSBs, the finding that PARPi leads to single-stranded DNA (ssDNA) gaps/nicks suggests replication gaps could also or alone be the cause of synthetic lethality. Here, we demonstrate that PARPi toxicity derives from replication gaps. Isogenic cells deficient in BRCA1 or the BRCA1-associated FANCJ, with common DNA repair defects in HR and FP, exhibit opposite responses to PARPi. Deficiency in FANCJ, a helicase also mutated in hereditary breast/ovarian cancer and Fanconi anemia, causes aberrant accumulation of fork remodeling factor HLTF and limits unrestrained DNA synthesis with ssDNA gaps. Thus, we predict replication gaps as a distinguishing factor and further uncouple HR, FP and fork speed from PARPi response. BRCA-deficient cells display excessive gaps that are diminished upon resistance, restored upon re-sensitization and when targeted augment synthetic lethality with PARPi. Furthermore, we define the source of gaps to defects in Okazaki fragment processing (OFP). Unchallenged BRCA1-deficient cells have elevated poly(ADP-ribose) and chromatin-associated PARP1 but aberrantly low XRCC1 indicating a defective backup OFP pathway. Remarkably, 53BP1 loss resuscitates OFP by restoring XRCC1-LIG3 that suppresses the sensitivity of BRCA1-deficient cells to drugs targeting OFP or generating gaps. Collectively, our study highlights unprotected lagging strand gaps as a determinant of synthetic lethality, providing a new paradigm and biomarker for PARPi toxicity.

Abstract Mutations in the currently known breast cancer susceptibility genes account for only 25–30% of all familial cases. Novel susceptibility genes can be identified by several methods, including candidate gene re-sequencing and genome-wide microarrays. We have applied microarrays for the detection of a new genomic variation class, copy number variants (CNVs), which potentially could disrupt genes in multiple pathways related to breast cancer susceptibility. The aim of the current study was to evaluate the role of the RNF8, UBC13, MMS2 and RAD51C DNA damage response genes in breast cancer susceptibility as well as to study if rare CNVs are associated with the predisposition to this disease. The analysis of 123 familial breast cancer cases revealed altogether nine different changes in the RNF8 and UBC13 candidate genes. However, none of the observed alterations were considered pathogenic. No alterations were observed in MMS2. The obtained results suggest that breast cancer predisposing alterations in RNF8, UBC13 and MMS2 are rare, or even absent. The RAD51C mutation screening of 147 familial breast cancer cases and 232 unselected ovarian cancer cases revealed two deleterious mutations: c.-13_14del27 was observed in a breast cancer case with familial history of ovarian cancer and c.774delT in an ovarian cancer case. Both mutations were absent in the control cohort. The results of the study support the hypothesis that rare variants of RAD51C predispose predominantly to ovarian cancer. A genome-wide scan of CNVs was performed for 103 familial breast cancer cases and 128 controls. The biological networks of the genes disrupted by CNVs were different between the two groups. In familial breast cancer cases, the observed mutations disrupted genes, which were significantly overrepresented in cellular functions related to maintenance of genomic integrity (P=0.0211). Biological network analysis showed that the disrupted genes were closely related to estrogen signaling and TP53-centered tumor suppressor network, and this result was confirmed by the analysis of an independent young breast cancer cohort of 75 cases. These results suggest that rare CNVs represent an alternative source of genetic variation contributing to hereditary risk for breast cancer. / Tiivistelmä Tunnetut rintasyöpäalttiusgeenien mutaatiot selittävät vain 25–30 prosenttia kaikista perinnöllisistä rintasyöpätapauksista. Uusia alttiusgeenejä voidaan tunnistaa useilla eri menetelmillä, kuten kandidaattigeenien mutaatiokartoituksella ja genomin-laajuisilla mikrosirutekniikoilla. Tässä tutkimuksessa sovelsimme mikrosirutekniikkaa uuden geneettisen variaatioluokan, kopiolukuvariaation (CNV), tutkimiseen. CNV:t voivat vaurioittaa lukuisia rintasyöpäalttiuteen liittyviä biokemiallisia reittejä. Tämän tutkimuksen tarkoitus oli arvioida RNF8-, UBC13-, MMS2- ja RAD51C -DNA- vauriovastegeenien sekä harvinaisten CNV:iden yhteyttä rintasyöpä-alttiuteen. 123 familiaalisen rintasyöpätapauksen analyysissä löytyi yhteensä yhdeksän muutosta RNF8- ja UBC13-geeneistä, joista yksikään ei osoittautunut patogeeniseksi. MMS2-geenissä ei havaittu muutoksia. Tulosten perusteella rintasyövälle altistavat muutokset RNF8-, UBC13- ja MMS2- geeneissä ovat joko erittäin harvinaisia tai niitä ei esiinny lainkaan. RAD51C-geenin mutaatiokartoitus 147 familiaalisesta rintasyöpätapauksesta sekä 232 valikoimattomasta munasarjasyöpätapauksesta paljasti kaksi haitallista mutaatiota. c.-13_14del27 havaittiin rintasyöpäpotilaalla, jonka suvussa esiintyi munasarjasyöpää, ja c.774delT todettiin munasarjasyöpäpotilaalta. Kumpaakaan mutaatiota ei havaittu verrokkiaineistossa. Tulokset vahvistavat hypoteesia RAD51C-geenin harvinaisten varianttien yhteydestä pääasiassa munasarjasyöpäriskiin. CNV:iden genomin-laajuinen skannaaminen suoritettiin 103 familiaaliselle rintasyöpätapaukselle ja 128 verrokille. CNV:iden häiritsemien geenien muodostamat biologiset verkostot olivat erilaiset näiden kahden ryhmän välillä. Familiaalisilla rintasyöpätapauksilla havaitut CNV:t vaikuttivat geeneihin, jotka olivat voimakkaasti korostuneita genomin eheyttä ylläpitävissä tehtävissä (P=0.0211). Biologisten verkostojen analyysi paljasti, että CNV:iden vahingoittamat geenit liittyivät läheisesti estrogeenisignalointiin sekä TP53-tuumorisupressoriverkostoon, ja tämä tulos vahvistettiin analysoimalla riippumatonta nuorista rintasyöpäpotilaista koostuvaa kohorttia (N=75). Tutkimuksen tulosten mukaan harvinaiset CNV:t ovat vaihtoehtoinen geneettisen variaation lähde perinnölliseen rintasyöpäalttiuteen.

Indiana University-Purdue University Indianapolis (IUPUI) / Fanconi anemia (FA) is a heterogenous genetic syndrome characterized by progressive bone marrow failure, aneuploidy, and cancer predisposition. It is incompletely understood why FA-deficient cells develop gross aneuploidy leading to cancer. Since the mitotic spindle assembly checkpoint (SAC) prevents aneuploidy by ensuring proper chromosome segregation during mitosis, we hypothesized that the FA signaling network regulates the mitotic SAC. A genome-wide RNAi screen and studies in primary cells were performed to systematically evaluate SAC activity in FA-deficient cells. In these experiments, taxol was used to activate the mitotic SAC. Following taxol challenge, negative control siRNA-transfected cells appropriately arrested at the SAC. However, knockdown of fourteen FA gene products resulted in a weakened SAC, evidenced by increased formation of multinucleated, aneuploid cells. The screen was independently validated utilizing primary fibroblasts from patients with characterized mutations in twelve different FA genes. When treated with taxol, fibroblasts from healthy controls arrested at the mitotic SAC, while all FA patient fibroblasts tested exhibited weakened SAC activity, evidenced by increased multinucleated cells. Rescue of the SAC was achieved in FANCA patient fibroblasts by genetic correction. Importantly, SAC activity of FANCA was confirmed in primary CD34+ hematopoietic cells. Furthermore, analysis of untreated primary fibroblasts from FA patients revealed micronuclei and multinuclei, reflecting abnormal chromosome segregation. Next, microscopy-based studies revealed that many FA proteins localize to the mitotic spindle and centrosomes, and that disruption of the FA pathway results in supernumerary centrosomes, establishing a role for the FA signaling network in centrosome maintenance. A mass spectrometry-based screen quantifying the proteome and phospho-proteome was performed to identify candidates which may functionally interact with FANCA in the regulation of mitosis. Finally, video microscopy-based experiments were performed to further characterize the mitotic defects in FANCA-deficient cells, confirming weakened SAC activity in FANCA-deficient cells and revealing accelerated mitosis and abnormal spindle orientation in the absence of FANCA. These findings conclusively demonstrate that the FA signaling network regulates the mitotic SAC, providing a mechanistic explanation for the development of aneuploidy and cancer in FA patients. Thus, our study establishes a novel role for the FA signaling network as a guardian of genomic integrity.