Décrypter les données omiques : importance du contrôle qualité. Application au cancer de l’ovaire Au cours des dix dernières années, la taille et la complexité des données biologiques ont littéralement explosé, et une attention particulière doit être portée au contrôle qualité. En effet, certaines données omiques (données génomiques et post-génomiques obtenues à haut débit) sont très incomplètes et/ou contiennent de nombreux biais et erreurs qu’il est facile de confondre avec de l’information biologiquement intéressante. Dans cette thèse, nous montrons que les interactions protéine-protéine issues de curation de la littérature et les interactions identifiées à haut débit sont beaucoup plus corrélées que ce qui est communément admis. Nous examinons l’interactome de la levure d’un point de vue original, en prenant en compte le degré d’étude des protéines par la communauté scientifique et nos résultats indiquent que cette corrélation s’estompe lorsqu’on se restreint aux protéines très étudiées. Ces observations nous permettent de proposer une méthode simple et fiable pour estimer la taille d’un interactome. Notre méthode conduit à une estimation d’au moins 37 600 interactions physiques directes chez S. cerevisiae, et montre que les évaluations précédentes sont trop faibles. Par ailleurs, nous étudions des données de séquençage nouvelle génération de l’ADN. Par une analyse des biais existant entre les short-reads alignés sur un brin ou sur l’autre du génome, nous mettons en évidence de nombreuses erreurs systématiques. De plus, nous observons de multiples positions présentant entre 20 et 40% de short-reads portant l’allèle variant : celles-ci ne peuvent pas être génotypées correctement. Nous proposons une méthode fiable pour appeler les génotypes à partir des données NGS qui permet de s’affranchir de ses difficultés. Enfin, nous appliquons cette méthode sur des données massives de séquençage d’exome de cellules saines et tumorales de 520 patientes atteintes du cancer de l’ovaire, produites par le consortium TCGA. Nous détectons en moyenne 30 632 variants germinaux par patiente. Parmi ces variants, nous identifions ceux les plus enclins à conférer un risque accru de développer la maladie : nous nous restreignons notamment aux variants induisant une perte de fonction de la protéine encodée et significativement plus présents chez les patientes que dans la population générale. Cela conduit à 44 SNVs par patiente en moyenne, répartis sur 334 gènes dans l’ensemble de la cohorte. Parmi ces 334 gènes, 42 ont été reportés comme impliqués dans la cancerogénèse, confirmant que la liste de candidats identifiés est fortement enrichie en gènes de susceptibilité au cancer de l’ovaire. En particulier, nos travaux confirment le rôle de suppresseur de tumeur de la protéine MAP3K8, très récemment proposée comme jouant un rôle clé dans d’autres cancers. / Deciphering omics data : on the importance of quality control. Application to ovarian cancer. Over the past 10 years, the size and complexity of biological data have exploded, and quality control is critical to interpret them correctly. Indeed, omics data (high- hroughput genomic and post-genomic data) are often incomplete and contain bias and errors that can easily be misinterpreted as biologically interesting findings. In this work, we show that literature-curated and high-throughput protein-protein interaction data, usually considered independent, are in fact significantly correlated. We examine the yeast interactome from a new perspective by taking into account how thoroughly proteins have been studied, and our results show that this bias can be corrected for by focusing on well- studied proteins. We thus propose a simple and reliable method to estimate the size of an interactome, combining literature-curated data involving well-studied proteins with high- hroughput data. It yields an estimate of at least 37,600 direct physical protein-protein interactions in S.cerevisiae, a significant increase over previous estimates. We then focus on next-generation DNA sequencing data. An analysis of the bias existing between short- eads aligned on each strand of the genome allows us to highlight numerous systematic errors. Furthermore, we observe many positions that exhibit between 20 and 40% of reads carrying the variant allele : these cannot be genotyped correctly.We then propose a method to overcome these biases and reliably call genotypes from NGS data. Finally, we apply our method to exome-seq data produced by the TCGA for tumor and matched normal samples from 520 ovarian cancer patients. We detect on average 30,632 germline variants per patient. Though an integrative approach, we then identify those which are likely to increase cancer risk : in particular, we focused on variants inducing a loss of function of the encoded protein, and selected those that are significantly more present in the patients than in the general population. We find 44 SNVs per patient on average, impacting 334 genes overall in the cohort. Among these genes, 42 have been previously reported as involved in carcinogenesis, confirming that our list is highly enriched in ovarian cancer susceptibility genes. In particular, our results confirm the tumor suppressor role of the MAP3K8 protein, recently identified in other types of cancer.

The oncogenic protein BCL-3, a member of the IκB family, was originally identified in a subset of human B-cell chronic lymphocytic leukemias that carry a translocation t(14,19), which results in BCL-3 overexpression. BCL-3 is also overexpressed in many solid tumors, such as in breast cancers and in cylindromas. This IκB protein activates or represses gene transcription through binding with the NF-κB proteins p50 and p52. Furthermore, BCL-3 is K63-linked polyubiquitinated, which leads to its translocation into the nucleus and to its target genes expression. BCL-3 is also K48-linked polyubiquitinated after GSK3 phosphorylation, which leads to its subsequent proteasomal degradation. However, the mechanisms underlying both its polyubiquitination and its ability to repress gene transcription remain poorly understood. In order to gain more insight into these BCL-3 functions, parallel screenings involving both yeast-two-hybrid experiments and biochemical purifications led to the identification of BCL-3-interacting partners. Those screenings identified CtBP as a molecule required for the ability of BCL-3 to repress gene transcription. CtBP is also required for the stability, for the oncogenic potential and for the ability of BCL-3 to inhibit UV-mediated cell apoptosis in keratinocytes. We also defined the E3 ligase TBLR1 as a key element involved in BCL-3 polyubiquitination and degradation through a GSK3-independent pathway and the proteasome subunit PSMB1 as a protein required for the GSK3-dependent and -independent proteasomal degradation of polyubiquitinated BCL-3. Importantly, all interactions require unique motifs within the amino-terminal domain of BCL-3. In conclusion, our data define multiple BCL-3-associated proteins that differentially and specifically regulate its function and stability and indicate that a better understanding of the mechanisms underlying the oncogenic properties of this IκB protein could be achieved through similar interactomic studies.

A VDAC é a proteína mais abundante da membrana mitocondrial externa. Possui diversas funções, tais como o controle da troca de metabólitos, através da membrana, e a participação no maquinário apoptótico. Estudamos o interactoma da VDAC com as proteínas mitocondriais neuronais do cérebro bovino e murino, a fim de compreender se a expressão diferenciada da VDAC1 e VDAC2 verificada entre essas células estão associadas às diferenças nas interações da VDAC. Os complexos proteicos foram analisados por 2D Blue Native SDS-PAGE e identificados via MALDI-TOF TOF usando o software Mascot e o banco de dados NCBInr. Foram identificados 27 e 46 spots em murino e bovino, respectivamente. Nós identificamos proteínas solúveis e incorporadas na membrana que não são participantes da fosforilação oxidativa, dentre elas a aldeido deidrogenase e muitas outras constituintes de complexos mitocondriais já conhecidos tão bem como novos, tais como a putative stomatin-like protein 2 complex e a switch-associated protein 70. Nossos resultados mostraram que os neurônios bovinos possuem mais complexos (5) contendo a VDAC do que em ratos (1), os quais indicam uma cinética diferencial de acoplamento e desacoplamento. Interessantemente, a lista contendo as proteínas identificadas inclui algumas proteínas conhecidas ou supostamente localizadas em compartimentos não-mitocondriais, por exemplo, a myc-induced nuclear antigen. O interactoma diferencial da VDAC entre as espécies bovina e murina, evidencia a presença de uma base comum, porém com diferentes ambientes estruturais, as quais podem ser a base da diferença entre os sítios de ligação A e B observados nas diferentes espécies. / The voltage dependent anion channel (VDAC) is the most abundant protein of outer mitochondrial membrane. VDAC controls metabolite exchange through this membrane and the apoptosis machinery. We studied the interactome of VDAC with mitochondrial proteins of neuronal cells from rat and bovine brain. We wished to understand if the differential expression of VDAC1 and VDAC2 verified between these cells was linked to differences in the VDAC interactions. Protein complexes were analyzed by 2D Blue Native SDS-PAGE and were identified by MALDI-TOF TOF using Mascot software against the NCBInr database. Number of 27 e 46 spots were identified from rat and bovine brain, respectively. We identified soluble and membrane-embedded non-OXPHOS proteins, among them aldehyde dehydrogenase, and many as constituents of known mitochondrial complexes as well as novel ones such as putative stomatin-like protein 2 complex and switchassociated protein 70. Our results showed that bovine neurons had more protein complexes (5) containing VDAC than rat cells (1), which indicates a differential kinetics of assembly or disassembly. Interestingly, the identification list included some proteins known or presumed to be localized to nonmitochondrial compartments, for example, myc-induced nuclear antigen. Our results support evidences of differential apoptotic and energetic mechanisms verified in these brains. The differential VDAC interactome between bovine and murine, support evidences of a common base, but whith different structural environment, which may be the basis of the difference between the binding sites A and B observed in these brains.

A VDAC é a proteína mais abundante da membrana mitocondrial externa. Possui diversas funções, tais como o controle da troca de metabólitos, através da membrana, e a participação no maquinário apoptótico. Estudamos o interactoma da VDAC com as proteínas mitocondriais neuronais do cérebro bovino e murino, a fim de compreender se a expressão diferenciada da VDAC1 e VDAC2 verificada entre essas células estão associadas às diferenças nas interações da VDAC. Os complexos proteicos foram analisados por 2D Blue Native SDS-PAGE e identificados via MALDI-TOF TOF usando o software Mascot e o banco de dados NCBInr. Foram identificados 27 e 46 spots em murino e bovino, respectivamente. Nós identificamos proteínas solúveis e incorporadas na membrana que não são participantes da fosforilação oxidativa, dentre elas a aldeido deidrogenase e muitas outras constituintes de complexos mitocondriais já conhecidos tão bem como novos, tais como a putative stomatin-like protein 2 complex e a switch-associated protein 70. Nossos resultados mostraram que os neurônios bovinos possuem mais complexos (5) contendo a VDAC do que em ratos (1), os quais indicam uma cinética diferencial de acoplamento e desacoplamento. Interessantemente, a lista contendo as proteínas identificadas inclui algumas proteínas conhecidas ou supostamente localizadas em compartimentos não-mitocondriais, por exemplo, a myc-induced nuclear antigen. O interactoma diferencial da VDAC entre as espécies bovina e murina, evidencia a presença de uma base comum, porém com diferentes ambientes estruturais, as quais podem ser a base da diferença entre os sítios de ligação A e B observados nas diferentes espécies. / The voltage dependent anion channel (VDAC) is the most abundant protein of outer mitochondrial membrane. VDAC controls metabolite exchange through this membrane and the apoptosis machinery. We studied the interactome of VDAC with mitochondrial proteins of neuronal cells from rat and bovine brain. We wished to understand if the differential expression of VDAC1 and VDAC2 verified between these cells was linked to differences in the VDAC interactions. Protein complexes were analyzed by 2D Blue Native SDS-PAGE and were identified by MALDI-TOF TOF using Mascot software against the NCBInr database. Number of 27 e 46 spots were identified from rat and bovine brain, respectively. We identified soluble and membrane-embedded non-OXPHOS proteins, among them aldehyde dehydrogenase, and many as constituents of known mitochondrial complexes as well as novel ones such as putative stomatin-like protein 2 complex and switchassociated protein 70. Our results showed that bovine neurons had more protein complexes (5) containing VDAC than rat cells (1), which indicates a differential kinetics of assembly or disassembly. Interestingly, the identification list included some proteins known or presumed to be localized to nonmitochondrial compartments, for example, myc-induced nuclear antigen. Our results support evidences of differential apoptotic and energetic mechanisms verified in these brains. The differential VDAC interactome between bovine and murine, support evidences of a common base, but whith different structural environment, which may be the basis of the difference between the binding sites A and B observed in these brains.