Computational Approaches Linking Chromatin Organization and Genome Regulation
EMBARGOED UNTIL 2026-10-14
Loading...
Author / Producer
Date
2024
Publication Type
Doctoral Thesis
ETH Bibliography
yes
Citations
Altmetric
EMBARGOED UNTIL 2026-10-14
Data
Rights / License
Abstract
In cells, the often meter-long DNA is packed into nuclei that are only a few micrometer in size. The packing of the DNA results in a highly structured organization of the genome, i.e. the 3D chromatin organization. This organization varies between cell states and is closely related to the gene regulatory networks that control cellular function in cells. The connection arises from the optimized packing, which for instance is expressed by the spatial colocalization of genes for their efficient co-regulation. However, the depth of the connection between the chromatin organization and genome regulation, along with its regulating mechanisms, is not yet fully understood. It also remains unclear whether this connection is profound enough to use the chromatin organization of cells as an unified readout of functional cell states and study cell state transitions. This would have wide-ranging implications for disease diagnostics and therapeutic design.
In this thesis, we address these important questions. Specifically, we first demonstrate that deep representation learning can be used to effectively model the link between chromatin organization and gene regulation. To achieve this, we developed a computational pipeline that learns how the gene regulatory programs of cells are reflected in their chromatin organization. This enables our pipeline to accurately identify alterations in these regulatory programs simply by using cost-efficient chromatin images as a readout of their chromatin organization. Our work provides novel avenues for studying gene regulation and alterations thereof using image-based readouts of the nuclear chromatin organization
and thus emphasizes the deep connection between chromatin organization and genome regulation.
While it has become evident that the functional optimization of the 3D chromatin packing gives rise to this connection, the mechanisms responsible for the optimized packing are not well understood. This particularly applies to the establishment of contacts between genomic regions from different chromosomes, which are crucial for cell function. In this thesis, we propose that the binding of long noncoding RNAs (lncRNAs) to the DNA may be a mechanism facilitating the formation and stability of these contacts, thereby enabling the colocalization and co-regulation of important gene clusters. To this end, we analyze multimodal sequencing data to identify spatial gene clusters formed by interchromosomal contacts. We show that the formation or loss of these gene clusters during aging could be explained by the changing abundance of specific lncRNAs. These lncRNAs might act as “glues” binding the corresponding genomic regions together. Although additional validation experiments are ongoing, our findings provide convincing evidence that the transcriptional control of lncRNAs and their binding to the DNA plays an important role for the functionally optimized 3D genome organization.
Finally, we demonstrate how computational methods can be used to exploit the link between chromatin organization and genome regulation, enabling the development of cost-effective image-based chromatin biomarkers with applications in disease diagnostics and therapeutic design. We first showcase this using chromatin images of immune cells obtained from liquid biopsies of tumor patients. These images are input to a computational pipeline which we developed and which uses simple image analysis and machine learning methods. We show that this pipeline identifies chromatin biomarkers of immune cells that enable accurate tumor diagnosis. We further illustrate the potential of such chromatin biomarkers by applying a similar pipeline to characterize differences in the chromatin organization of distinct B cell populations in lymphoid tissue. Our corresponding analyses reveal chromatin condensation differences in these B cell populations, which correlate with the presence or absence of other immune cell populations in their local microenvironments. This could potentially provide novel insights into the mechanisms that control T cell infiltration in B cell lymphomas, also implying that inhibition of chromatin condensation may prevent immune cell evasion in B cell lymphomas. Thus, this study also highlights the potential of chromatin biomarkers to guide therapeutic design.
In summary, in this thesis we present novel computational methods to study, describe, model, and leverage the relationship between chromatin organization and genome regulation. The results presented in this thesis improve our understanding of this connection and its regulating mechanisms as well as demonstrate its large potential for applications such as disease diagnostics and therapy evaluation.
Permanent link
Publication status
published
External links
Editor
Contributors
Book title
Journal / series
Volume
Pages / Article No.
Publisher
ETH Zurich
Event
Edition / version
Methods
Software
Geographic location
Date collected
Date created
Subject
Computational Biology; Data analysis; Machine Learning; Chromatin biology
Organisational unit
09691 - Shivashankar, G. V. / Shivashankar, G. V.
Notes
Funding
Related publications and datasets
Has part: https://doi.org/10.3929/ethz-b-000649288
Has part: https://doi.org/10.3929/ethz-b-000685510