S. aureus est une bactérie pathogène opportuniste de l’homme qui pose un grave problème de santé publique. Le pouvoir pathogène de S. aureus est conféré par un très grand nombre de facteurs de virulence, dont l’expression est finement régulée à de multiples niveaux. Les effecteurs de cette régulation sont à la fois des protéines et des ARN non codants (ARNnc) aussi appelés ARN régulateurs. Je me suis concentré au cours de ma thèse sur la classe majoritaire qui sont les ARNnc qui régulent la traduction d’ARNm. Ils sont impliqués dans de complexes réseaux de régulation qui permettent de contrôler la physiologie de la cellule ainsi que sa virulence. Pour élargir nos connaissances de ces réseaux, j’ai développé deux approches méthodologiques, appelées MAPS et Grad-seq, que j’ai appliquées in vivo chez S. aureus en utilisant RsaA et RsaC comme modèles. L’application du MAPS a permis d’identifier de nouvelles cibles directes pour RsaA et des cibles potentielles pour RsaC. L’approche Grad-Seq est un outil puissant mais demande encore des ajustements. J’ai également pu déterminer un rôle probable pour l’ARNnc RsaC dans la régulation de l’homéostasie oxydo-réductive de S. aureus, en lien avec la résistance au stress oxydatif et avec la persistance lors de l’internalisation par les ostéoblastes. / S. aureus is an opportunistic pathogen of the human species which can express a large array of virulence factors whose expression is under tight regulation at multiple levels. The regulation can be done by proteins and by particular molecules of RNA called non-coding RNA (ncRNA). I focused during my thesis on the main category of ncRNA in S. aureus, which are regulating the translation of mRNA. These ARNs are involved in complex regulatory networks, impacting the physiology of the bacterial cell and its virulence. To understand further these networks, I developped two methodological approaches in vivo in S. aureus, called MAPS and Grad-seq, which were applied using RsaA and RsaC as models of studies. MAPS allowed to find new direct targets of RsaA and plausible targets for RsaC. The Grad-Seq method showed to be a powerful tool but still needs refinements. I also could determine a possible role for RsaC in the regulation of oxydo-reductive homeostasis, in direct link with oxydative stress resistance and persistance during internalisation by osteoblasts.

Introdução: O câncer de bexiga é a segunda neoplasia maligna mais frequente do trato urinário, com 386.000 casos estimados e 150.000 mortes para 2011 no mundo. Noventa e cinco por cento são carcinomas uroteliais (CUB) papilíferos não músculo-invasivos de baixo grau, que apresentam altas taxas de recidiva, mas raramente progridem. Tumores invasivos de alto grau representam 10-20% dos diagnósticos, são altamente agressivos levando à mortalidade elevada. O conhecimento das vias moleculares envolvidas na carcinogênese dessa neoplasia é importante para a identificação de novos marcadores para diagnóstico, acompanhamento, prognóstico e desenvolvimento de novas terapias alvo. Micro RNA (miRNA) são pequenas sequências não codificantes de RNA que regulam a expressão dos genes inibindo a tradução da proteína ou promovendo a degradação do RNA mensageiro, estando atualmente envolvidos em vários processos celulares fisiológicos e patológicos, incluindo o câncer. Objetivos: Caracterizar o perfil de expressão de miRNA no CUB, relacionando-o com os parâmetros prognósticos clássicos para a doença: grau histológico e estadiamento. Além disso, relacionar esse padrão de comportamento dos miRNA com a recidiva tumoral e sobrevida câncer-específica em pacientes tratados cirurgicamente para CUB. Material e Métodos: Catorze miRNA (miR-100, miR-10a, miR-21, miR-205, miR-let7c, miR- 125b, miR-143, miR-145, miR-221, miR-223, miR-15a, miR-16-1, miR-199a e miR- 452) foram isolados de espécimes cirúrgicos de 60 pacientes divididos em 2 grupos: 30 pacientes com CUB não invasivo (pTa) de baixo grau submetidos à RTU de bexiga, 30 com CUB invasivo (pT2-3) de alto grau submetidos à cistectomia radical. O grupo controle é representado por cinco pacientes portadores de bexiga normal sem CUB que realizaram tratamento cirúrgico aberto para tratamento da hiperplasia prostática benigna (HPB). O processamento dos miRNA envolveu três fases: (1) extração do miRNA com kit específico, (2) geração do DNA complementar e (3) amplificação do miRNA por PCR quantitativo em tempo real (qRT-PCR). A expressão de cada miRNA foi obtida através do cálculo 2- CT e os RNU-43 e RNU-48 foram utilizados como controles endógenos. Testes estatísticos foram aplicados para estudar as variáveis envolvidas e curvas de Kaplan-Meyer foram usadas para avaliar a sobrevida livre de recidiva (SLR) e sobrevida câncer-específica (SCE). Resultados: Dos 14 miRNA estudados a maioria apresentou subexpressão nos dois grupos de tumor analisados, com exceção do miR-10a para o grupo pTa de baixo grau e do miR-100, 21 e 205 para os tumores pT2/pT3 de alto grau, onde demonstraram-se superexpressos. Essas diferenças de expressão de miRNA entre os dois grupos foram estatisticamente. Quando estudamos a relação entre expressão de miRNA e a evolução dos pacientes através de curvas de sobrevida, observamos que maiores níveis de expressão do miR-21 relacionou-se com menor SLR para tumores pTa. Ainda, maiores concentrações de miR-10a e miR-145 se associaram com menor SLR e maiores níveis de miR-10a com menor SCE para tumores pT2-3. Conclusões: Demonstramos um predomínio de subexpressão de miRNA em xv carcinomas de bexiga. Os miR-100, miR-10a, miR-21 e miR-205 demonstraram diferenças no perfil de expressão para grau e estadiamento dentro dos dois grupos de tumor, sendo capazes de diferenciá-los. Maiores níveis de miR-21 se relacionaram com menor SLR para tumores pTa de baixo grau, enquanto maiores concentrações de miR-10a estiveram associadas com menor SLR e SCE para tumores pT2/pT3 de alto grau / Introduction: Bladder cancer (BC) is the second most common malignancy of the urinary tract, with 386,000 cases estimated and 150,000 deaths in 2011. Urothelial carcinomas (UC) represent 95% of BC cases, and knowledge of the molecular pathways associated with BC carcinogenesis is crucial to identify new diagnostic and prognostic biomarkers, and development of new target molecular therapies. MicroRNAs (miRNAs) are short non-coding RNA molecules that play important roles in the regulation of gene expression by acting directly on mRNAs, leading to either mRNA degradation or inhibition of translation, involved in many physiological and pathological processes, including cancer. Objectives: To characterize miRNAs expression profiles in UC, associating with classic prognostic factors: grade and stage. Moreover, correlate miRNA expression with tumor recurrence and survival. Material and Methods: Fourteen miRNAs (miR-100, miR-10a, miR-21, miR-205, miR-let7c, miR-125b, miR-143, miR-145, miR-221, miR-223, miR-15a, miR-16-1, miR- 199a e miR-452) were isolated from surgical specimens from 60 patients classified in two groups: 30 patients with low-grade non-invasive pTa UC that underwent TURB, 30 with high-grade invasive pT2/pT3 UC underwent radical cystectomy. The control group consists in five normal bladder tissue taken from patients that underwent retropubic prostatectomy to treat benign prostatic hyperplasia (BPH). miRNA processing involved three phases: (1) miRNA extraction by specific kits, (2) cDNA generation (3) miRNA amplification through qRT-PCR. Expression profiles were obtained by relative quantification determined by 2-ct method. Endogenous control were RNU-43 and RNU-48. Statistic tests were used to study the prognostic variables and Kaplan-Meyer curves were constructed to analyze disease-free (DFS) and disease-specific (DSS) survivals. Results: All miRNAs were underexpressed in both groups, except miR-10a in pTa and miR-100, 21 and 205 in pT2/pT3 tumors, that where over-expressed. miR-100, miR-21, miR-10a and miR-205 differentialy expressed in both groups and this differences were statistically significant. The Kaplan-Meyer survival curves showed that higher levels of miR-21 were related to shorter DFS for pTa group. Also, higher levels of miR-10a and miR-145 were associated with shorter DFS and higher levels of miR-10a were also related to shorter DSS in pT2/pT3 group. Conclusions: The majority of miRNA were shown to be underexpressed in bladder UC. miR-100, miR-10a, miR-21 and miR-205 were differentially expressed considering tumor grade and stage. The miRNA profile was able to distinguish pTa low grade and pT2-3 high grade tumors. Higher levels of miR- 21 were related to shorter DFS in pTa, while higher levels of miR-10a were associated with shorter DFS and DSS in pT2-3, high grade UC

Terminado o sequenciamento do genoma humano, as atenções se voltaram para a determinação do conjunto completo de transcritos humanos. Diversos trabalhos sugerem que enquanto apenas uma pequena fração de mRNAs codificantes para proteína não é conhecida, existe um grande número de RNAs não-codificantes (ncRNAs) ainda não caracterizados. Nesse contexto, o presente trabalho visou explorar as informações de expressão gênica contidas em ESTs para identificar e caracterizar novos transcritos humanos. A busca genômica por membros de famílias gênicas relacionadas com câncer levou a identificação de novas pequenas GTPases, destacando uma subfamília que deve apresentar função supressora tumoral em próstata. Uma classe de ncRNAs longos, sem splicing, expressos antisenso a partir de regiões intrônicas foi descrita utilizando plataformas de microarrays, construídas pelo grupo, enriquecidas com seqüências sem anotação. O perfil de expressão de 23 ncRNAs intrônicos estava significativamente correlacionado com o grau de diferenciação de tumores de próstata (Gleason Score), e pode ser utilizado como candidato a marcador molecular de prognóstico. Um total de 39 ncRNAs intrônicos responderam à estimulação por andrógeno, apontando para um mecanismo regulatório da expressão intrônica por sinais fisiológicos hormonais. A biogênese da expressão intrônica parece ser complexa, pois uma fração não é transcrita pela RNA Polimerase II. A transcrição intrônica estava correlacionada com uso de exons em células tratadas com andrógeno. Assinaturas de expressão intrônica conservadas em tecidos humanos e de camundongos, e interações de transcritos intrônicos com proteínas regulatórias foram observadas. Este trabalho contribui com novas e originais evidências que dão apoio ao papel postulado para esses ncRNAs no controle fino do programa transcricional humano. / With the completion of the human genome sequence, attention has shifted towards determining the complete set of human transcripts. Multiple lines of evidence suggest that while only a small fraction of protein-coding mRNAs remains to be described, there is a huge amount of uncharacterized non-coding RNAs (ncRNAs). In this context, the present work sought to explore the gene expression information provided by ESTs to identify and characterize new human transcripts. A genomic-wide search for cancer related gene family members identified novel small GTPase genes, and highlighted an uncharacterized subfamily that may have a tumor suppressor role in prostate cancer. A class of long unspliced ncRNAs, expressed antisense from introns of protein-coding genes was described using custom-designed microarray platforms enriched with unannotated sequences. The expression profile of 23 intronic ncRNAs was significantly correlated to the degree of prostate tumor differentiation (Gleason Score), and could be used as a candidate prognostic molecular maker. A total of 39 intronic ncRNAs were responsive to androgen stimulation, poiting to a mechanism of intronic expression regulation by physiological hormone signals. Intronic ncRNA biogenesis seems to be complex, since a fraction of them is not transcribed by RNA Polymerase II. Intronic transcription was correlated to exon usage in androgen treated cells. Tissue expression signatures of intronic transcription were conserved in human and mouse, and intronic transcripts were found to interact with regulatory proteins. This work provides new and original contributions that support the postulated role of ncRNAs in the fine tunning of the human transcriptional program.

Até pouco tempo acreditava-se que a maioria das moléculas de RNA estava relacionada à tradução de proteínas. Porém, descobriu-se que outros tipos de moléculas de RNA que não são traduzidas estão presentes em muitos organismos diferentes e afetam uma variedade de processos moleculares, são os chamados RNAs não-codificantes (ncRNAs). Apesar de sua importância funcional, os métodos biológicos e computacionais para a detecção e caracterização de RNAs não-codificantes ainda são imprecisos e incompletos. A identificação de novas espécies de ncRNAs é difícil através de procedimentos experimentais e as técnicas computacionais existentes são lentas. O objetivo deste trabalho foi obter uma ferramenta mais eficiente para a comparação de uma seqüência de RNA não-codificante contra um banco de seqüências. Para isso foi proposto e implementado um modelo para identificação computacional de ncRNAs com apoio dos pacote Viena e Infernal e foram realizados experimentos para avaliá-lo / Until recently it was generally accepted that most RNA molecules were involved in the translation process. However, it was discovered that many types of untranslated RNA molecules are present in many different organisms and they are related to a wide variety of molecular processes. These molecules are called non-coding RNAs (ncRNAs). Despite their functional importance, the biological and computational methods to detect and identify non-coding RNAs are still imprecise and incomplete. The discovery of new ncRNAs species is difficult through experimental procedures and the existing computational techniques are slow. This project aimed at obtaining a more efficient tool that compares a non-coding RNA sequence against a sequence database. In order to achieve this, a computational model for ncRNAs identification using the Vienna and Infernal packages has been proposed and implemented. Experiments were conduced to evaluate the model

Orientador: Guilherme Targino Valente / Resumo: A levedura Saccharomyces cerevisiae é o microrganismo mais utilizado para a produção de etanol devido a sua alta capacidade fermentativa e resistência aos estresses oriundos desse processo. Entretanto, a própria concentração de etanol é um dos fatores mais limitantes no processo de produção desse combustível. Os aspectos da genômica funcional relacionada à tolerância ao etanol são ainda pouco esclarecidos, e nem mesmo se sabe se os lncRNAs tem papel nesse processo. Poucos lncRNAs foram identificados em S. cerevisiae, e nem mesmo se conhece as redes lncRNAs-proteínas nessa espécie e nem se podem codificar micropeptídeos. Nesse contexto, este trabalho visa identificar lncRNAs em linhagens de S. cerevisiae com diferentes níveis de tolerância ao etanol. Para isso, foi realizado a montagem dos lncRNAs, predição de ligações lncRNA-proteínas, buscas de micropepetídeos, análises de conservação genômica, estrutural e funcional dos lncRNAs, avaliação da influência do lncRNAs em regular as expressões de seus vizinhos e comparação dos resultados entre linhagens mais e menos tolerantes ao etanol. As análises de enriquecimento ontológico apontam para uma relação próxima entre os lncRNAs e a tolerância ao etanol e uma conservação funcional, embora os dados não reportem nenhuma conservação nem genômica nem estrutural. Além disso, variados tipos de prováveis regulações foram sugeridas, sendo a regulação em trans majoritariamente inversa entre os lncRNAs e seus genes-alvo, diferentemente da ma... (Resumo completo, clicar acesso eletrônico abaixo) / Abstract: The yeast Saccharomyces cerevisiae is the most used microorganism for ethanol production due to its high fermentative capacity and resistance to different stressors along this process. However, the ethanol concentration is one of the most limiting factors of fuel production. The functional genomics aspects related to the ethanol tolerance are still unclear, and it is not clear if the lncRNAs really have a role in this process. Few lncRNAs were identified in S. cerevisiae, lncRNA-protein networks of this species are still unknown and also if they can code micropeptides. In this context, this thesis aims to identify lncRNAs and evaluate their roles in S. cerevisiae ethanol tolerance. Then, it was performed the assembling of lncRNAs, predictions of lncRNA-protein interactions, searches for potential micropeptides coding-lncRNAs, analysis of genomic, structural and functional conservation of lncRNAs, evaluation of the lncRNAs influence in regulating the expressions of their neighbors, and comparison between strains that are more and less tolerant to the ethanol. Moreover, many putative regulatory pathways were here suggested, being that most trans regulations act on an inversely manner between the expression of the lncRNAs and their target-genes, unlike observed in most of cis regulations. The current literature confirms the lncRNAs functional conservation here observed, and the role of these non-coding molecules as regulators. Finally, here we suggest that lncRNAs are acting to ... (Complete abstract click electronic access below) / Mestre

The study of non-coding RNA genes has received increased attention in recent years fuelled by accumulating evidence that larger portions of genomes than previously acknowledged are transcribed into RNA molecules of mostly unknown function, as well as the discovery of novel non-coding RNA types and functional RNA elements. Here, we demonstrate that specific properties of graphs that represent the predicted RNA secondary structure reflect functional information. We introduce a computational algorithm and an associated web-based tool (GraPPLE) for classifying non-coding RNA molecules as functional and, furthermore, into Rfam families based on their graph properties. Unlike sequence-similarity-based methods and covariance models, GraPPLE is demonstrated to be more robust with regard to increasing sequence divergence, and when combined with existing methods, leads to a significant improvement of prediction accuracy. Furthermore, graph properties identified as most informative are shown to provide an understanding as to what particular structural features render RNA molecules functional. Thus, GraPPLE may offer a valuable computational filtering tool to identify potentially interesting RNA molecules among large candidate datasets.

The detection of non-protein-coding RNA (ncRNA) genes in bacteria and their diverse regulatory mode of action moved the experimental and bio-computational analysis of ncRNAs into the focus of attention. Regulatory ncRNA transcripts are not translated to proteins but function directly on the RNA level. These typically small RNAs have been found to be involved in diverse processes such as (post-)transcriptional regulation and modification, translation, protein translocation, protein degradation and sequestration. Bacterial ncRNAs either arise from independent primary transcripts or their mature sequence is generated via processing from a precursor. Besides these autonomous transcripts, RNA regulators (e.g. riboswitches and RNA thermometers) also form chimera with protein-coding sequences. These structured regulatory elements are encoded within the messenger RNA and directly regulate the expression of their “host” gene. The quality and completeness of genome annotation is essential for all subsequent analyses. In contrast to protein-coding genes ncRNAs lack clear statistical signals on the sequence level. Thus, sophisticated tools have been developed to automatically identify ncRNA genes. Unfortunately, these tools are not part of generic genome annotation pipelines and therefore computational searches for known ncRNA genes are the starting point of each study. Moreover, prokaryotic genome annotation lacks essential features of protein-coding genes. Many known ncRNAs regulate translation via base-pairing to the 5’ UTR (untranslated region) of mRNA transcripts. Eukaryotic 5’ UTRs have been routinely annotated by sequencing of ESTs (expressed sequence tags) for more than a decade. Only recently, experimental setups have been developed to systematically identify these elements on a genome-wide scale in prokaryotes. The first part of this thesis, describes three experimental surveys of exploratory field studies to analyze transcript organization in pathogenic bacteria. To identify ncRNAs in Pseudomonas aeruginosa we used a combination of an experimental RNomics approach and ncRNA prediction. Besides already known ncRNAs we identified and validated the expression of six novel RNA genes. Global detection of transcripts by next generation RNA sequencing techniques unraveled an unexpectedly complex transcript organization in many bacteria. These ultra high-throughput methods give us the appealing opportunity to analyze the complete RNA output of any species at once. The development of the differential RNA sequencing (dRNA-seq) approach enabled us to analyze the primary transcriptome of Helicobacter pylori and Xanthomonas campestris. For the first time we generated a comprehensive and precise transcription start site (TSS) map for both species and provide a general framework for the analysis of dRNA-seq data. Focusing on computer-aided analysis we developed new tools to annotate TSS, detect small protein-coding genes and to infer homology of newly detected transcripts. We discovered hundreds of TSS in intergenic regions, upstream of protein-coding genes, within operons and antisense to annotated genes. Analysis of 5’ UTRs (spanning from the TSS to the start codon of the adjacent protein-coding gene) revealed an unexpected size diversity ranging from zero to several hundred nucleotides. We identified and validated the expression of about 60 and about 20 ncRNA candidates in Helicobacter and Xanthomonas, respectively. Among these ncRNA candidates we found several small protein-coding genes that have previously evaded annotation in both species. We showed that the combination of dRNA-seq and computational analysis is a powerful method to examine prokaryotic transcriptomes. Experimental setups are time consuming and often combined with huge costs. Another limitation of experimental approaches is that genes which are expressed in specific developmental stages or stress conditions are likely to be missed. Bioinformatic tools build an alternative to overcome such restraints. General approaches usually depend on comparative genomic data and evolutionary signatures are used to analyze the (non-)coding potential of multiple sequence alignments. In the second part of my thesis we present our major update of the widely used ncRNA gene finder RNAz and introduce RNAcode, an efficient tool to asses local protein-coding potential of genomic regions. RNAz has been successfully used to identify structured RNA elements in all domains of life. However, our own experience and the user feedback not only demonstrated the applicability of the RNAz approach, but also helped us to identify limitations of the current implementation. Using a much larger training set and a new classification model we significantly improved the prediction accuracy of RNAz. During transcriptome analysis we repeatedly identified small protein-coding genes that have not been annotated so far. Only a few of those genes are known to date and standard proteincoding gene finding tools suffer from the lack of training data. To avoid an excess of false positive predictions, gene finding software is usually run with an arbitrary cutoff of 40-50 amino acids and therefore misses the small sized protein-coding genes. We have implemented RNAcode which is optimized for emerging applications not covered by standard protein-coding gene annotation software. In addition to complementing classical protein gene annotation, a major field of application of RNAcode is the functional classification of transcribed regions. RNA sequencing analyses are likely to falsely report transcript fragments (e.g. mRNA degradation products) as non-coding. Hence, an evaluation of the protein-coding potential of these fragments is an essential task. RNAcode reports local regions of high coding potential instead of complete protein-coding genes. A training on known protein-coding sequences is not necessary and RNAcode can therefore be applied to any species. We showed this with our analysis of the Escherichia coli genome where the current annotation could be accurately reproduced. We furthermore identified novel small protein-coding genes with RNAcode in this extensively studied genome. Using transcriptome and proteome data we found compelling evidence that several of the identified candidates are bona fide proteins. In summary, this thesis clearly demonstrates that bioinformatic methods are mandatory to analyze the huge amount of transcriptome data and to identify novel (non-)coding RNA genes. With the major update of RNAz and the implementation of RNAcode we contributed to complete the repertoire of gene finding software which will help to unearth hidden treasures of the RNA World.

Dissertação (mestrado)—Universidade de Brasília, Instituto de Ciências Exatas, Departamento de Ciência da Computação, Programa de Pós-Graduação em Informática, 2016. / Submitted by Albânia Cézar de Melo (albania@bce.unb.br) on 2016-08-03T13:45:05Z No. of bitstreams: 1 2016_JoaoVictorAraujoOliveira.pdf: 3385598 bytes, checksum: 87023d9eae07bd39a3d1cb8613c3d33f (MD5) / Approved for entry into archive by Patrícia Nunes da Silva(patricia@bce.unb.br) on 2016-12-06T13:01:15Z (GMT) No. of bitstreams: 1 2016_JoaoVictorAraujoOliveira.pdf: 3385598 bytes, checksum: 87023d9eae07bd39a3d1cb8613c3d33f (MD5) / Made available in DSpace on 2016-12-06T13:01:15Z (GMT). No. of bitstreams: 1 2016_JoaoVictorAraujoOliveira.pdf: 3385598 bytes, checksum: 87023d9eae07bd39a3d1cb8613c3d33f (MD5) / Métodos de aprendizagem de máquina vêm sendo amplamente usados na identificação e classificação de diferentes famílias de RNAs não-codificadores (ncRNAs). Muitos desses métodos são baseados na aprendizagem supervisionada, onde atributos anteriormente conhecidos, chamados features, são extraídos de uma sequência e usados em um classificador. Nesta dissertação, apresentamos dois métodos para a identificação das duas classes principais de snoRNAs, C/D box e H/ACA box snoRNAs: snoReport 2.0, uma melhoria significativa da primeira versão do snoReport; e o snoRNA-EDeN, um novo método baseado no EDeN, que é um kernel decomposicional de grafos. O snoReport 2.0 é um método que, usando features extraídas de sequências candidatas em genomas, combina predição de estrutura secundária de ncRNAs com Máquina de Vetores de Suporte (Support Vector Machine - SVM), para identificar C/D box e H/ACA box snoRNAs. Seu classificador de H/ACA box snoRNA mostrou um F-score de 93% (uma melhoria de 10% em relação à primeira versão do snoReport), enquanto o classificador de C/D box snoRNA obteve F-score de 94% (melhoria de 14%). Alem disso, ambos os classificadores tiveram todas as medidas de performances acima de 90%. Na fase de validação, o snoReport 2.0 identificou 67,43% dos snoRNAs de vertebrados de ambas as classes. Em Nematóides, o snoReport 2.0 identificou 29,6% dos C/D box snoRNAs e 69% dos H/ACA box snoRNAs. Para as Drosofilídeas, foram identificados 3,2% dos C/D box snoRNAs e 76,7% dos H/ACA box snoRNAs. Esses resultados mostram que o snoReport 2.0 é eficiente na identificação de snoRNAs em organismos vertebrados, e também para H/ACA box snoRNAs de organismos invertebrados. Por outro lado, em vez de usar features de uma sequência (em geral, difíceis de identificar), uma abordagem recente de aprendizagem de máquina é descrita a seguir. Dada uma região de interesse de uma sequencia, o objetivo é gerar um vetor esparso que pode ser usado como micro-features em algum algoritmo de aprendizado de máquina, ou pode ser usado para a criação de features poderosas. Essa abordagem é usada no EDeN (Explicit Decomposition with Neighbourhoods), um kernel decomposicional de grafos baseado na técnica Neighborhood Subgraph Pairwise Distance Kernel (NSPDK). O EDeN transforma um grafo em um vetor esparso, decompondo-o em todos os pares de subgrafos vizinhos de raios pequenos, a distâncias crescentes. Baseado no EDeN, foi desenvolvido um método chamado snoRNA-EDeN. Na fase de testes, para C/D box snoRNAs, o snoRNA-EDeN obteve um F-score de 93,4%, enquanto que para H/ACA box snoRNAs o F-score foi de 85.12%. Na fase de validação, para C/D box snoRNA, o snoRNA-EDeN mostrou uma grande capacidade de generalização, identificando 94,61% de snoRNAs de vertebrados e 63,52% de invertebrados, um resultado significantemente melhor em comparação ao snoReport 2.0, que identificou apenas 52,92% dos vertebrados e 14,6% dos invertebrados. Para o H/ACA box, o snoReport 2.0 identificou 79,9% dos snoRNAs de vertebrados e 73,3% dos snoRNAs de Nematóides e Drosofilídeos, enquanto que o snoRNA-EDeN identificou 95,4% dos vertebrados e 57.8% dos nematóides e drosofilas. Ambos os métodos estão disponíveis em: http://www.biomol.unb.br/snoreport e http://www.biomol.unb.br/snorna_eden. ___________________________________________________________________________ ABSTRACT / Machine learning methods have been widely used to identify and classify different families of non-coding RNAs. Many of these methods are based on supervised learning, where some previous known attributes, called features, are extracted from a sequence, and then used in a classifier. In this work, we present two methods to identify the two main classes of snoRNAs, C/D box and H/ACA box: snoReport 2.0, a significant improvement of the original snoReport version; and snoRNA-EDeN, a new method based on EDeN, a decompositional graph kernel. On one hand, snoReport 2.0 is a method that, using features extracted from candidate sequences in genomes, combines secondary structure prediction with Support Vector Machine (SVM) to identify C/D box and H/ACA box snoRNAs. H/ACA box snoRNA classifier showed a F-score of 93% (an improvement of 10% regarding to the previous version), while C/D box snoRNA classifier a F-Score of 94% (improvement of 14%). Besides, both classifiers exhibited performance measures above 90%. In the validation phase, snoReport 2.0 predicted 67.43% of vertebrate organisms for both classes. SnoReport 2.0 predicted: for Nematodes, 29.6% of C/D box and 69% of H/ACA box snoRNAs; and for Drosophilids, 3.2% of C/D box and 76.7% of H/ACA box snoRNAs. These results show that snoReport 2.0 is efficient to identify snoRNAs in vertebrates, and also H/ACA box snoRNAs in invertebrates organisms. On the other hand, instead of using known features from a sequence (difficult to find in general), a recent approach in machine learning is described as follows. Given a region of interest of a sequence, the objective is to generate a sparse vector that can be used as micro-features in a specific machine learning algorithm, or it can be used to create powerful features. This approach is used in EDeN (Explicit Decomposition with Neighbourhoods), a decompositional graph kernel based on Neighborhood Subgraph Pairwise Distance Kernel (NSPDK). EDeN transforms one graph in a sparse vector, decomposing it in all pairs of neighborhood subgraphs of small radius at increasing distances. Based on EDeN, we developed a method called snoRNA-EDeN. On the test phase, for C/D box snoRNAs, snoRNA-EDeN showed a F-score of 93.4%, while for H/ACA box snoRNAs, the F-score was 72%. On the validation phase, for C/D box snoRNAs, snoRNA-EDeN showed a better capacity of generalization, predicting 94.61% of vertebrate C/D box snoRNAs and 63.52% of invertebrates, a significantly better result compared to snoReport 2.0, which predicted only 52.92% of vertebrates and 14.6% of invertebrates. For H/ACA box snoRNAs, snoReport 2.0 predicted 79.9% of vertebrate snoRNAs and 73.3% of Nematode and Drosophilid sequences, while snoRNA-EDeN predicted 95.4% of vertebrate snoRNAs and 57.8% of Nematode and Drosophilid sequences. Both methods are available at http://www.biomol.unb.br/snoreport and http://www.biomol.unb.br/snorna_eden.

Introdução: O câncer de bexiga é a segunda neoplasia maligna mais frequente do trato urinário, com 386.000 casos estimados e 150.000 mortes para 2011 no mundo. Noventa e cinco por cento são carcinomas uroteliais (CUB) papilíferos não músculo-invasivos de baixo grau, que apresentam altas taxas de recidiva, mas raramente progridem. Tumores invasivos de alto grau representam 10-20% dos diagnósticos, são altamente agressivos levando à mortalidade elevada. O conhecimento das vias moleculares envolvidas na carcinogênese dessa neoplasia é importante para a identificação de novos marcadores para diagnóstico, acompanhamento, prognóstico e desenvolvimento de novas terapias alvo. Micro RNA (miRNA) são pequenas sequências não codificantes de RNA que regulam a expressão dos genes inibindo a tradução da proteína ou promovendo a degradação do RNA mensageiro, estando atualmente envolvidos em vários processos celulares fisiológicos e patológicos, incluindo o câncer. Objetivos: Caracterizar o perfil de expressão de miRNA no CUB, relacionando-o com os parâmetros prognósticos clássicos para a doença: grau histológico e estadiamento. Além disso, relacionar esse padrão de comportamento dos miRNA com a recidiva tumoral e sobrevida câncer-específica em pacientes tratados cirurgicamente para CUB. Material e Métodos: Catorze miRNA (miR-100, miR-10a, miR-21, miR-205, miR-let7c, miR- 125b, miR-143, miR-145, miR-221, miR-223, miR-15a, miR-16-1, miR-199a e miR- 452) foram isolados de espécimes cirúrgicos de 60 pacientes divididos em 2 grupos: 30 pacientes com CUB não invasivo (pTa) de baixo grau submetidos à RTU de bexiga, 30 com CUB invasivo (pT2-3) de alto grau submetidos à cistectomia radical. O grupo controle é representado por cinco pacientes portadores de bexiga normal sem CUB que realizaram tratamento cirúrgico aberto para tratamento da hiperplasia prostática benigna (HPB). O processamento dos miRNA envolveu três fases: (1) extração do miRNA com kit específico, (2) geração do DNA complementar e (3) amplificação do miRNA por PCR quantitativo em tempo real (qRT-PCR). A expressão de cada miRNA foi obtida através do cálculo 2- CT e os RNU-43 e RNU-48 foram utilizados como controles endógenos. Testes estatísticos foram aplicados para estudar as variáveis envolvidas e curvas de Kaplan-Meyer foram usadas para avaliar a sobrevida livre de recidiva (SLR) e sobrevida câncer-específica (SCE). Resultados: Dos 14 miRNA estudados a maioria apresentou subexpressão nos dois grupos de tumor analisados, com exceção do miR-10a para o grupo pTa de baixo grau e do miR-100, 21 e 205 para os tumores pT2/pT3 de alto grau, onde demonstraram-se superexpressos. Essas diferenças de expressão de miRNA entre os dois grupos foram estatisticamente. Quando estudamos a relação entre expressão de miRNA e a evolução dos pacientes através de curvas de sobrevida, observamos que maiores níveis de expressão do miR-21 relacionou-se com menor SLR para tumores pTa. Ainda, maiores concentrações de miR-10a e miR-145 se associaram com menor SLR e maiores níveis de miR-10a com menor SCE para tumores pT2-3. Conclusões: Demonstramos um predomínio de subexpressão de miRNA em xv carcinomas de bexiga. Os miR-100, miR-10a, miR-21 e miR-205 demonstraram diferenças no perfil de expressão para grau e estadiamento dentro dos dois grupos de tumor, sendo capazes de diferenciá-los. Maiores níveis de miR-21 se relacionaram com menor SLR para tumores pTa de baixo grau, enquanto maiores concentrações de miR-10a estiveram associadas com menor SLR e SCE para tumores pT2/pT3 de alto grau / Introduction: Bladder cancer (BC) is the second most common malignancy of the urinary tract, with 386,000 cases estimated and 150,000 deaths in 2011. Urothelial carcinomas (UC) represent 95% of BC cases, and knowledge of the molecular pathways associated with BC carcinogenesis is crucial to identify new diagnostic and prognostic biomarkers, and development of new target molecular therapies. MicroRNAs (miRNAs) are short non-coding RNA molecules that play important roles in the regulation of gene expression by acting directly on mRNAs, leading to either mRNA degradation or inhibition of translation, involved in many physiological and pathological processes, including cancer. Objectives: To characterize miRNAs expression profiles in UC, associating with classic prognostic factors: grade and stage. Moreover, correlate miRNA expression with tumor recurrence and survival. Material and Methods: Fourteen miRNAs (miR-100, miR-10a, miR-21, miR-205, miR-let7c, miR-125b, miR-143, miR-145, miR-221, miR-223, miR-15a, miR-16-1, miR- 199a e miR-452) were isolated from surgical specimens from 60 patients classified in two groups: 30 patients with low-grade non-invasive pTa UC that underwent TURB, 30 with high-grade invasive pT2/pT3 UC underwent radical cystectomy. The control group consists in five normal bladder tissue taken from patients that underwent retropubic prostatectomy to treat benign prostatic hyperplasia (BPH). miRNA processing involved three phases: (1) miRNA extraction by specific kits, (2) cDNA generation (3) miRNA amplification through qRT-PCR. Expression profiles were obtained by relative quantification determined by 2-ct method. Endogenous control were RNU-43 and RNU-48. Statistic tests were used to study the prognostic variables and Kaplan-Meyer curves were constructed to analyze disease-free (DFS) and disease-specific (DSS) survivals. Results: All miRNAs were underexpressed in both groups, except miR-10a in pTa and miR-100, 21 and 205 in pT2/pT3 tumors, that where over-expressed. miR-100, miR-21, miR-10a and miR-205 differentialy expressed in both groups and this differences were statistically significant. The Kaplan-Meyer survival curves showed that higher levels of miR-21 were related to shorter DFS for pTa group. Also, higher levels of miR-10a and miR-145 were associated with shorter DFS and higher levels of miR-10a were also related to shorter DSS in pT2/pT3 group. Conclusions: The majority of miRNA were shown to be underexpressed in bladder UC. miR-100, miR-10a, miR-21 and miR-205 were differentially expressed considering tumor grade and stage. The miRNA profile was able to distinguish pTa low grade and pT2-3 high grade tumors. Higher levels of miR- 21 were related to shorter DFS in pTa, while higher levels of miR-10a were associated with shorter DFS and DSS in pT2-3, high grade UC

Les éléments transposables d’ADN sont presque ubiquitaires dans le monde vivant et leur mobilité peut être délétère pour le génome. Leur régulation dans les tissus germinaux animaux passe par la voie de silencing des piRNAs (PIWI-interacting RNAs). Les piRNAs sont produits à partir de loci contenant des fragments d’éléments transposables insérés en clusters. Nous étudions l’émergence de ces clusters de piRNAs chez la drosophile.Nous avons activé de novo un cluster de transgènes par héritage maternel de piRNAs homologues. Il s’agit d’un cas de paramutation, ou conversion épigénétique stable et récurrente.Nous avons montré que ce cluster paramuté produit de novo des piRNAs, et étonnement dessiRNAs.J’ai caractérisé de façon fonctionnelle et moléculaire ce phénomène de paramutation par l’utilisation de mutants. J’ai montré que les propriétés de silencing, ainsi que la production depiRNAs et de siRNAs, sont abolies en contexte mutant pour tous les gènes testés de la voie despiRNAs (voies primaire et secondaire). Parallèlement, j’ai étudié un cas de paramutation« partiellement homologue » dans laquelle le cluster reçoit des piRNAs homologues seulement àune partie de sa séquence. J’ai montré qu’il y a production de piRNAs par la totalité du cluster dès la 3e génération.J’ai montré, enfin, que des clusters activés de novo par la chaleur, présentent des propriétés fonctionnelles et moléculaires semblables aux clusters activés par les piRNAsmaternels.Ces travaux apportent des éléments clés pour la compréhension de la mise en place de l’épigénome, tant d’un point de vue mécanistique qu’évolutif. / DNA transposable elements are almost ubiquitous in the living world and their mobility can be deleterious for the genome. Their regulation in germaria is mediated by the piRNAsilencing pathway (PIWI-interacting RNAs). piRNAs are produced by loci formed by clusters of fragments of transposable elements. We are studying the emergence of these piRNA-producing clusters in Drosophila.We have de novo activated a cluster of transgenes via maternal inheritance of homologous piRNAs. This is a case of paramutation i.e. a stable and recurrent epigenetic conversion process.We have shown that this paramutated cluster produces de novo piRNAs and, surprisingly, also siRNAs. I have characterized this paramutation functionally and molecularly, by a mutant approach. I have shown that its silencing properties, as well as piRNA and siRNA production are abolished in mutant contexts for all the genes from the primary and secondary piRNA pathways I have tested. At the same time, I have studied the case of a partially homologous paramutation, in which piRNAs maternally inherited by the cluster are homologous to only a part of its sequence. I have shown that piRNA are produced all along the cluster as early as the 3rdgeneration.Finally, I have shown that a cluster activated de novo by an environmental stress shows the same functional and molecular properties as a cluster paramutated via maternal piRNA inheritance.These studies provide key elements for understanding the emergence of the epigenome from a mechanistic and an evolutionary perspective.