Leveraging mutation biases can improve prognostic biomarkers in the study of cancer genomics, writes Constance Li.
The genetic origins of cancer were first suspected over 100 years ago. Long before the discovery of the first cancer gene, David von Hansemann and Theodor Boveri each proposed a link between carcinogenesis and alterations in the hereditary material of a cell. Subsequent studies into cancer-causing agents, from exogenous sources such as tobacco and UV radiation, to endogenous sources including oxidative damage, have consistently shown how changes in the genome can give rise to cancer. The study of cancer genomics aims to characterise the genomic alterations underlying tumour initiation, promotion and progression. A tumour’s genomic composition is also constantly changing in response to selective pressure, giving rise to tumour heterogeneity and the study of tumour evolution. By leveraging this genomic information and machine learning techniques, we can create biomarkers that stratify patients into subgroups with specific sets of defining features and prognostic outlook. By gaining a better understanding of the disease, we can improve biomarker accuracy and progress towards a personalised treatment approach specific to each patient’s unique disease.
Somatic mutations accumulate in a cell from a combination of stochastic, environmental and hereditary factors. In normal cells, there is a balance where genomic mutations are corrected by DNA repair machinery before cell division. In this way, errors are not passed on to daughter cells and the cells function normally. However, DNA repair is imperfect and mutations can sometimes propagate. Stochastic mutations are a natural by-product of errors made during DNA replication. At the same time, exposure to environmental mutagens leads to genomic lesions that introduce more genomic alterations. Consequently, somatic mutations naturally accumulate in an individual’s cells over time. If these mutations do not affect the function of a set of oncogenes and tumour suppressors known as cancer driver genes, an individual may never develop a malignant tumour. Carcinogenesis occurs when the mutation of cancer driver genes gives a cell a selective advantage and the ability to uncontrollably proliferate, invade and metastasize.
Some individuals also have a higher risk of developing a particular cancer due to inherited cancer driver gene variants. This leads to a genetic predisposition to developing cancer, such as in the well-known example of BRCA mutations increasing breast cancer risk.
While we often talk about cancer as a single disease, it is actually a collection of several heterogeneous diseases, each with its own biology and standard of care. Cellular pathways important in the function of one tissue may not be as critical or even active in another. The types of mutations also differ: some tumour types accumulate more somatic single nucleotide variants, while others have more structural changes such as copy number aberrations. The efforts of large-scale cancer genomics studies in the last 20 years, such as those of The Cancer Genome Atlas (TCGA) and the International Cancer Genome Consortium (ICGC), have leveraged advances in sequencing technology and data analysis to reveal the genomic profiles of multiple tumour types. While some genes, such as the ‘guardian of the genome’ TP53, are mutated in almost every type of cancer, other events are tumour type-specific. Some genes found recurrently mutated across multiple tumour types have different biological roles, are regulated in different ways or are mutated differently depending on the tissue. The Catalogue of Somatic Mutations in Cancer maintains a comprehensive resource for mutations in cancer. Every tumour type has its own genomic landscape and it is important to keep in mind the genomic heterogeneity between different cancers.
Genomic heterogeneity is not limited to different tumour types or even different tumours. Intra-tumour heterogeneity, or spatial and temporal differences within one tumour, remains a major unresolved challenge in cancer genomics. Over the course of tumour development, tumour cells continue to accumulate mutations. Some tumour cells acquire mutations, conferring an additional selective advantage and multiply to form subpopulations of cells termed subclones. These subclones may give rise to further subclones so that one tumour is often a collection of multiple subclones, each with a defining set of mutations while still sharing common events from the original ancestor clone. Frequently, sequencing samples are taken at one point in time from one tumour region. It is easy to imagine that a genomic profile obtained by this limited sampling can miss rare subclones and does not capture changes over time in response to selective pressures such as a shifting microenvironment. As genomic information is used more frequently to guide patient treatment, it is increasingly important to account for genomic heterogeneity, as a resistant subclone can limit treatment effectiveness, seed metastatic lesions and lead to relapse.
To address problems posed by genomic tumour heterogeneity, there have been recent landmark studies in tumour evolution. By studying how tumours developed and evolved in the past, we can predict how they will respond and behave in the future. The ongoing TRACERx (TRAcking Cancer Evolution through therapy [Rx]) projects use longitudinal sampling to track treatment response over time in hundreds of melanoma, prostate, lung and renal patients. Other studies focus on the effects of the tumour microenvironment, of immune editing, on varying models of tumour evolution and on incorporating tumour subclone information to make ‘evolution-aware’ prognostic biomarkers. Still others investigate the non-invasive use of circulating tumour DNA in blood as a means of monitoring tumour evolution. Understanding how cancer adapts to selection pressures allows us to be better prepared to combat the challenges of tumour evolution in the clinic.
The end goal of cancer genomics is to improve and inform clinical care. One way of doing this is to leverage genomic information to generate biomarkers. By combining a priori knowledge with statistical analysis and machine learning, we can stratify patients into smaller groups defined by shared common features like a recurring mutation or relatively high mutation burden. If patients in one group are more likely to develop cancer, the set of defining features is a diagnostic biomarker. If the patient group is more likely to experience aggressive disease, it is a prognostic biomarker. And if the patient group has a greater response to a specific treatment, it is a predictive biomarker. There are myriad biomarker objectives with some genomic biomarker tests already used in the diagnosis of breast cancer (Prosigna and Oncotype DX, among others) and to guide targeted therapies (F1CDx, MSK-IMPACT). With ever-growing datasets and improving technology, biomarkers are also not limited to single-data types: integrative studies combine genomics with complementary and orthogonal data such as epigenomic, proteomic and imaging data to improve biomarker accuracy.
Cancer genomics is an extraordinarily expansive field centred on the study of mutations in the cancer genome landscape. By continuing to characterise molecular aberrations in the genesis and evolution of a tumour, we aim to not only better understand the disease, but to predict its future trajectory. As technological improvements helped advance the field from single-gene studies to genome-wide projects, technology continues to play a critical role in generating, storing and analysing the vast amounts of data needed in today’s multidisciplinary cancer genomics study.
The past 30 years of cancer genomics have largely focused on describing genomic alterations and their impact on cancer biology. Moving forward, there is increased focus on translating genomic findings into the clinic such as in the upcoming ICGC ARGO (Accelerating Research in Genomic Oncology) project. Through combining our knowledge, machine learning and integrating other data, we can transition into an era of truly personalised medicine.