A complex set of factors and interactions underlie cancer development, progression and response to treatment. Despite an unprecedented number of new targeted therapies, relatively small effects on patient survival have been achieved due to therapeutic resistance. My research is dedicated to the development of systems biology and computational approaches to discover integrated genomic blueprints that enable us to understand and predict how cancer will evolve, and respond to treatment.
Ultimately, we need a complete understanding of how a normal cell rewires its circuits and interacts with the surrounding microenvironment to turn into a cancer cell, and how this affects the progression of the disease and response to treatment. I study how and where the blueprint of the genome is functional, how this function is impaired or changed in cancer, what is the role of the tumour microenvironment in driving these changes, and how this information can be used to improve clinical practice. My research has been addressing these questions using combined machine learning and mathematical modelling alongside integrative genomic approaches. This has provided new insight into the function of several previously uncharacterized genes, and has generated robust gene signatures which are being evaluated as biomarkers.
Network of miRNA associated with gene signatures of cancer hallmarks. miRNAs are small non-coding RNAs regulating the expression of target genes post-transcriptionally. The figure shows a network ‘map’ of signatures (coloured nodes) and miRNAs (grey nodes) connected by edges when an association was found by using a penalized regression approach. Integration of transcriptomic, methylation, copy number, and mutation data across thousands of clinical samples from multiple cancer types unveiled a complex map of interactions, characterized by a strong interconnectivity between distinct hallmarks of cancer, and a central core set of miRNAs (Dhawan et al., Nature Communications, 2018). This highlighted the pervasive nature of miRNA-mediated repression of tumour suppressor genes such as PTEN, FAT4, and CDK12, uncovering a potential alternative mechanism for tumourigenesis in the absence of mutation, methylation or copy number changes.
The tumour microenvironment encompasses a heterogeneous, dynamic and highly interactive system of cancer and stroma cells which impact on how the cancer will progress and respond to treatment. One of the key physiological and microenvironmental differences between tumour and normal tissues is the presence of hypoxia, which drives selective growth of more aggressive cancer cells, and thus enables tumour progression. It also causes therapeutic resistance, including resistance to chemotherapy and radiotherapy. To identify hypoxic tumours and tailor treatment, we have identified gene signatures that are being validated in prospective studies and translated into biomarkers (see for example Masiero et al, Cancer Cell, 2013).
Systematic evaluation of gene signatures of cellular processes. A summary illustration of protocol and package we developed, SigQC (Dhawan et al, Nature Protocols, 2019). SigQC enables a streamlined methodological and standardised approach for the quality control validation of gene signatures. The figure shows some of the critical steps involved in the generation of a clinically and biologically useful, transportable gene signature, including ensuring sufficient expression, variability, and compactness (e.g. gene-gene correlation).
Hypoxia along with acidosis increases clonal selection resulting in aggressive cancer phenotypes, but the drivers of this selection are not fully understood. In order to elucidate mechanistic insights into the dysregulated pathways in hypoxic tumours, it is crucial to study genomic aberrations alongside transcriptional and post-transcriptional changes. Our recent studies demonstrate that by studying the aberrant cancer genomic landscape together with transcriptional deregulation in hypoxic tumours, we can identify factors driving adaptation to hypoxia in cancer. Following this, we are using powerful integrative approaches combining genomic data with transcriptional, post-transcriptional, and functional assay data, together with clinical knowledge to improve our understanding of the selection pressure generated by hypoxia (see for example Haider et al, Genome Biology, 2016).
The graph above shows the interplay between genes regulated in a hypoxic tumour microenvironment. Genes are shown as circles, while their relationship with co-partners genes which help fuel progression of cancer is shown by green edges. The network demonstrates complex interplay which emerges when cancer cells undergo hypoxia. Our network reconstruction algorithm aims to unravel such interplay by starting with known key-player genes in cancer hypoxia. It then extends the network by finding other genes in human genome exhibiting similar patterns of expression. As a result, highly active network is identified which is shown to contain primary culprits of driving aggressive cancers.
While our principal aim is to build integrated models of cancer which improve our understanding of the disease, we also develop new approaches and computational tools to aid the translation of this data-generated knowledge into beneficial use in the clinic for treatment of patients. Specifically, current high-throughput technology enables us to obtain multiple levels of information about the entire genome of individual cancers. Integrating these data has the potential to improve dramatically our understanding of the differences between cancer types, and of the molecular mechanisms underlying response and resistance to treatment.
We developed, MicroC, a novel computational framework for conducting in-silico biological experiments and generate or test new hypotheses. MicroC may be used to study the effects of mutations and cell-cell or cell-microenvironment interactions on the dynamics of cell growth. This allows biological hypotheses to be tested in a controlled stepwise fashion, and lends itself naturally to model a heterogeneous population of cells acting and evolving in a dynamic microenvironment, which is needed to predict the evolution of complex multi-cellular dynamics. Importantly, this enables modelling co-occurring intrinsic perturbations, such as mutations, and extrinsic perturbations, such as nutrients availability, and their interactions (see Voukantsis, Gigascience, 2019 and https://process.innovation.ox.ac.uk/software/p/13519/microc/1).
In several cases, we have a good understanding of the frequent mutations and alterations which can drive oncogenesis and maintain tumour viability, and drugs exist to target these. However, cancer cells depend on multiple such alterations, requiring treatment with a combination of drugs. Furthermore, different tumours can have many different alterations. In 2018 the European Research Council has awarded Francesca Buffa a programme to model this extreme diversity, and to develop ‘virtual’ cells based on information learnt from genomic studies.
These models enable us to study cancer initiation, competition between cells carrying different mutations, interactions of cancer cells with the host microenvironment, and their response to different drug combinations. Ultimately, we want to facilitate in-silico experimentation to identify likely resistance mechanisms, and predict the most appropriate treatment for each patient.