Browsing by Author "Gordân, Raluca"

Transcription factor (TF) proteins bind to DNA in a sequence specific manner to regulate gene expression. The binding affinity of TFs for individual sites is well characterized and can be represented using DNA motif models such as position weight matrices. However, there are many factors influencing TF-DNA recognition in the cell, leading to complexities than cannot be captured by motif models alone. Here, we present our studies on two factors: cooperative TF binding and alterations in TF binding due to DNA mutations. Both factors require quantitative and rigorous approaches to distinguish real effects from random noise.

First, we present a new method for characterizing cooperative binding of TFs to DNA. This method addresses the issue that TF binding sites located in close proximity, which occurs frequently across the human genome, are not necessarily bound cooperatively. To distinguish between cooperative and independent binding, we developed a high-throughput on-chip binding assay designed specifically to measure TF binding to neighboring sites. Using the experimental data from our assay, we trained machine learning models to differentiate between cooperative and independent binding of TFs. This method enabled us to reveal molecular mechanisms used by TFs to bind DNA cooperatively.

Second, we introduce QBiC-Pred (Quantitative Predictions of TF Binding Changes Due to Sequence Variants), an ordinary least squares based method to predict the magnitude of the effect of DNA mutations on TF-DNA recognition. We implemented QBiC-Pred as a web service: qbic.genome.duke.edu, which allows users to run our models through a user-friendly web interface. We used this method to identify non-recurring putative regulatory driver mutations in cancer. Our approach is novel because we prioritize mutations based on their effects on transcription factor (TF) binding, instead of relying on the recurrence of the mutations among tumor samples---which is often difficult to perform as individual non-coding mutations are rarely seen in more than one donor. Focusing on the functional effects of non-coding mutations across regulatory regions, we identified dozens of genes whose regulation in tumor cells is likely to be significantly perturbed by non-coding mutations.

Somatic DNA mutations play critical roles in human disease, especially during carcinogenesis and tumor development. Two major sources of mutations are the T-G mismatches resulting from spontaneous deamination of 5-methylcytosine (5meC) and various types of mismatches owing to DNA replication errors. To maintain genome stability, DNA repair enzymes (REs) are responsible for recognizing the mismatches and initiating repair. For spontaneous deamination of 5meC, the Thymine DNA Glycosylase (TDG) is one of the specialized enzymes that recognize the resulting T-G mismatches, excises the thymine to create abasic site (AP site), and initiates base excision repair (BER). For replication errors, various types of mismatches are recognized by the DNA mismatch repair protein MutS, which initiates the downstream mismatch repair (MMR) pathway. The DNA sequence flanking the mismatches (i.e. their context) is known to have an important effect on the specificities of both TDG and MutS, and consequently can influence the specificity of repair. However, the sequence context effects for both REs are poorly understood.To address this gap, we developed high-throughput in-vitro assays to quantitatively measure the repair enzymes’ binding and excision activity for DNA mismatches in tens of thousands of different sequence contexts, in a cell-free system. We found that two base-pairs 5' and three base-pairs 3' of the mismatch have significant effects on TDG binding and activity, whereas four base-pairs 5' and three base-pairs 3' of the mismatch are important for MutS binding specificity. The results are consistent with structural data for both REs. Moreover, we show that predictive modeling can be used to enable high accuracy predictions of RE binding to any new DNA sequence. The sequence context specificity of both REs is highly relevant in living cells. For TDG, DNA sequences that are most methylated in the genome, and thus most prone to deamination and subsequent mutation, are also the ones where T-G mismatches are best recognized by TDG. For MutS, we show that its specificity shapes the mutation landscape by influencing the genetic mutation spectra. More surprisingly, we found that mutations observed in late replicating regions in S-phase are more likely to come from lower-affinity mismatches than mutations observed in early replicating regions. The results indicate that the high mutation rates in late replicating regions are due not only to the insufficient time to repair replications errors in these regions, but also to the relatively long time required for MutS to identify and bind to low affinity mismatches in order to initiate their repair. Therefore, the sequence specificity for both REs is directly relevant to the in vivo genomic landscape of somatic mutations.

Transcription factors (TFs) are proteins that bind DNA based on the sequence and structure to regulate gene expression. They are fundamental components of genomic function, present in all known forms of life. Thus, understanding the conditions required for TF-DNA interactions is a longstanding and active field of study. With the advent of comprehensive k-mer based measurements using protein binding microarrays, the binding profiles of hundreds of TFs have been measured. This dissertation addresses two major problems. First, the information from these comprehensive measurements are used to create simplistic models of binding that capture only the high affinity range. In a biological context, weak binding sites are often the most important in developmental and regulatory processes and can be missed by models targeting high affinity binding sites. Second, that the vast majority of measurements are on structurally unmodified DNA. TF binding occurs in complex and dynamic systems where the DNA structure can be significantly altered due to sources such as DNA damage. First, we look at how DNA shape influences binding through the study of UV induced photoproducts, DNA adducts formed from UV light exposure that distort the shape of pyrimidine dinucleotides. We developed a new k-mer based method for measuring TF binding to UV-irradiated DNA, UV-Bind. Using this technology, we find that the UV-induced changes in DNA structure from pyrimidine dinucleotide photoproducts can change the specificity of TFs. Using high-throughput k-mer measurements, we also found non-canonical sequences that show an increase in binding signal after UV-irradiation. We then introduce a new algorithm for calling TF binding sites using k-mers, CtrlF-TF. CtrlF-TF takes high-throughput k-mer measurements from PBMs and outputs aligned, ranked consensus sites that can be searched in a genome. These sites compare favorably to traditional position weight matrix defined sites via in vivo and in vitro benchmarks.

Cellular reprogramming processes remain poorly characterized at the level of genome- wide chromatin and gene expression changes. Specifically, the extent to which re- programmed cells differ quantitatively from both the starting cells and the target cells is unknown for most reprogramming systems. In addition, direct comparisons between the genome-wide reprogramming efficiencies in systems driven by the over- expression of endogenous versus exogenous master regulator(s) are rarely performed. This thesis presents methods for comparative analyses of genome-wide gene expres- sion and chromatin accessibility data, applied to myogenic reprogramming systems in order to assess reprogramming efficiency and generate testable hypotheses for improving the reprogramming process. First, gene expression and chromatin acces- sibility profiles of MyoD-induced transdifferentiated primary human skin fibroblasts are compared to fibroblasts and myoblasts. Second, similar genome-wide changes are assessed for myogenic conversion of iPS cells driven by overexpression of en- dogenous MyoD versus exogeneous MyoD. Both studies show that (i) while many muscle marker genes are reprogrammed after MyoD overexpression, the genome-wide accessibility and gene expression profiles are still different from those of primary my- oblast or myotube cells; (ii) MyoD induces a continuum of changes in chromatin accessibility, with only a fraction of myogenic chromatin sites gaining a completely reprogrammed accessibility status; and (iii) chromatin-remodeling deficiencies are strongly correlated with incomplete gene expression reprogramming. Classification analyses comparing reprogrammed and non-reprogrammed genes or chromatin sites revealed discriminatory genetic and epigenetic features, suggesting ways to poten- tially improve the reprogramming efficiency. Genomic analysis of transgene MyoD overexpression in iPS cells, compared to endogenous MyoD activation, also showed that MyoD is more “aggressive” in its chromatin opening behavior, showing a large number of off-target chromatin opening events. To further investigate the effects chromatin remodeling events on gene expression in reprogramming studies, a novel cross-cell type gene expression prediction framework (CPGex) is also developed. By integrating and modeling the non-linear combinatorial effects of chromatin accessi- bility as well as the expression levels of regulatory TFs, CPGex is able to weigh the importance of regulatory sites or factors for downstream targeted reprogramming of specific gene(s). The methods described in this thesis can be applied to any cellular reprogramming system in order to quantitatively assess the efficiency of reprogram- ming at the chromatin accessibility and gene expression levels, as well as to generate testable hypothesis for improved genome-wide reprogramming.

The spatial pattern of transcription factor (TF) binding and the level of TF occupancy at individual sites across the genome determine how a TF regulates its targets. Consequently, predicting the location and level of TF binding genome-wide is of great importance and has received much attention recently. Protein-binding microarray (PBM) technology has become the golden standard for studying TF-DNA interactions in vitro, while Chromatin Immunoprecipitation followed by DNA Sequencing (ChIP-seq) is the standard method for inferring TF binding in vivo. However, direct interpretation of in vitro results in an in vivo context is challenging and to-date remains scarce. In this study, we focus on the E2F family of paralogous TFs, whose mode of binding to DNA has been controversial. Previous studies have shown that E2F factors bind to the TTTSSCGCG motif, where S can be a C or a G. Still, only a small fraction of in vivo targets are reported to contain this motif, hinting at indirect recruitment of the protein. We observed that genomic occupancy of E2F factors directly correlates with their in vitro binding affinities. By using data from universal PBM experiments, we show that E2F factors likely bind to DNA through direct sequence recognition and not through cofactor interaction. Furthermore, we developed a kinetic binding model using the PBM data to describe competition between different members of the E2F family and successfully distinguished between their unique targets. Overall, these results demonstrate how the straightforward and simple in vitro PBM experiments can be used for inferring the complex in vivo landscape of TF binding and elucidate the mechanism of E2F-DNA interaction.

Transcription factors (TFs) control the temporal and spatial expression of target genes by interacting with DNA in a sequence-specific manner. Recent advances in high throughput experiments that measure TF-DNA interactions in vitro and in vivo have facilitated the identification of DNA binding sites for thousands of TFs. However, it remains unclear how each individual TF achieves its specificity, especially in the case of paralogous TFs that recognize distinct target genomic sites despite sharing very similar DNA binding motifs. In my work, I used a combination of high throughput in vitro protein-DNA binding assays and machine-learning algorithms to characterize and model the binding specificity of 11 paralogous TFs from 4 distinct structural families. My work proves that even very closely related paralogous TFs, with indistinguishable DNA binding motifs, oftentimes exhibit differential binding specificity for their genomic target sites, especially for sites with moderate binding affinity. Importantly, the differences I identify in vitro and through computational modeling help explain, at least in part, the differential in vivo genomic targeting by paralogous TFs. Future work will focus on in vivo factors that might also be important for specificity differences between paralogous TFs, such as DNA methylation, interactions with protein cofactors, or the chromatin environment. In this larger context, my work emphasizes the importance of intrinsic DNA binding specificity in targeting of paralogous TFs to the genome.

Most previous efforts to identify cancer driver mutations have focused on protein-coding genes. In recent years, the decreasing costs of DNA sequencing have enabled whole-genome sequencing (WGS) studies of thousands of tumor samples, making it possible to systematically survey non-coding regions for potential driver events. From these studies, millions of somatic mutations in cancer have been identified, the majority of which are non-coding. However, driver identification remains a far greater challenge in non-coding regions than in coding genes, primarily due to the incomplete annotation of the non-coding genome and the unknown functional impact of non-coding mutations.

In this work, we present new approaches to identify putative regulatory driver mutations in cancer, based on new methodology for predicting the quantitative effects of single nucleotide variants on transcription factor (TF) binding. Unlike most of the previous work on driver identification, our method does not require the driver mutations to be highly recurrent; instead, we assess the mutations’ significance by testing if they cause larger TF binding changes than expected in the case of completely random mutations. Since gene regulation relies on the cooperation of multiple regulatory elements, we have devised a way to combine the effects of all regulatory mutations of a gene in order to identify genes whose regulation is likely to be significantly perturbed by the mutations observed in their regulatory elements, through changes in TF binding.

We have applied our TF-centric approaches to analyze single nucleotide variants identified in a liver cancer data set from the International Cancer Genome Consortium (ICGC), and identified potentially dysregulated genes whose regulatory mutations could trigger significant TF binding changes. Notably, the genes identified by us are different from the ones prioritized by recurrence-based approaches. However, most of the potentially dysregulated genes we have identified have large changes in gene expression and/or are cancer prognostic genes. Our results suggest that regulatory mutations should be investigated further, not just by their recurrence, but also by their functional effects such as TF binding changes, to uncover dysregulated genes that may drive tumorigenesis.

Transcription factors (TFs) bind genomic DNA to regulate gene expression. In the cell, the genome is decorated with numerous proteins, including nucleosomes and proteins involved in processes such as DNA repair and replication, which could compete with TFs. While the competition with nucleosomes is well studied, TFs can also compete with other DNA-binding proteins (e.g. other TFs, DNA repair enzymes, polymerases). The rules and the impact of such competition remain largely unknown. Here, we investigate how TFs compete with each other and with repair enzymes, and we reveal the significant role TFs play as competitors in multiple pathways.To capture the binding profiles of competing TFs, we designed a quantitative cell-free assay that we applied to study Cbf1-Pho4 competition in yeast and MYC-MAD competition in human. We found that TFs greatly influence each other’s occupancy, in a way that is dictated by the proteins’ divergence in DNA-binding specificity. Analyses of ChIP-seq data confirmed that the patterns of TF-TF competition, as observed in vitro, are preserved in the nuclear environment. Furthermore, gene expression data suggests that Cbf1-Pho4 competition plays a critical role in the specific activation of target genes in the cell. In the MYC-MAD system, we found that quantitative in vitro knowledge facilitates the interpretation of in vivo ChIP-seq data and reveals subtle signals in gene regulatory networks, demonstrating the advantage of combining in vitro quantification with in vivo detection. Next, we adapted our assay to study the competition between TFs and DNA repair enzymes. Recently (Afek et al. 2020) we showed that TFs bind with high affinity to mismatches, which can result from replication errors. We thus hypothesized that TFs can compete with TDG, the glycosylase that recognize T-G mismatch and initiates base excision repair, and MutS, the mismatch-binding enzyme that initiates mismatch repair. Our high-throughput competition assay showed that, as predicted, the binding of both repair enzymes to DNA decreases significantly in the presence of TFs. In addition, the magnitude of the decreases in repair enzyme binding correlates well with the TF binding levels, indicating specific competition. This suggests that, in the cell, TFs bound to mismatches may affect repair and lead to increased mutagenesis at regulatory sites. Overall, our study proposes an approach for studying competition between DNA-binding proteins in a quantitative and high-throughput manner, and highlights the significance of this competition not only for gene regulation (where TFs are known to play an important role), but also in DNA repair.