The biomarker to improve cancer management.
The gold standard of cancer diagnosis, tissue biopsy, is fraught with many disadvantages that must be overcome to better manage cancer as a disease. Tissue biopsies are often invasive and can be traumatic to patients. Tissue biopsies might also be impossible to collect depending on the location of the suspected tumour. Another major issue with tissue biopsies is tumoral heterogeneity. Cancer masses can have different genetic make-ups, within a single mass, and between different masses within a single patient. This phenomenon can ultimately hinder the personalised medical care utility of tissue biopsies. Liquid biopsies have recently emerged as a promising new tool for diagnosing cancer, bypassing the drawbacks of tissue biopsy.
Liquid biopsies are, as the name suggests, an assay of tumour tissue from body fluids, typically the blood. The term encompasses a few different biomolecules, but one of the most promising is circulating tumour DNA (ctDNA). All cells naturally shed DNA into the bloodstream, and tumour cells also contribute to plasma-borne DNA in cancer patients. The utility of ctDNA in cancer management comes from the fact that ctDNA represents the genomes of all cancer subclones. All the mutations that drive the tumour and the mutations that confer resistance to chemotherapies are all represented in ctDNA, in theory. If one can collect ctDNA data, this information can be used to diagnose patients earlier, inform treatments, and predict treatment responses & prognosis.
It is well understood diagnosing cancer in the early stages of disease is critical in the efficacy of treatment options. However, a lot of cancers do not have good screening assays, or screening compliance for at risk individuals is low, or some tests may just be too expensive to roll-out on a large scale. Circulating tumour DNA screening assays can address these shortcomings, potentially improving early detection rates and driving down the mortality rates of cancer.
Let’s take a look at colorectal cancer (CRC), as one the diseases that can benefit from a ctDNA based screening assay. The established non-invasive test for CRC (serum carcinoembryonic antigen quantification) shows low sensitivity meaning the test throws up a lot of false negative results. Colonoscopy is uncomfortable for patients and too expensive to scale up. The ultimate result is low screening compliance for at risk individuals. A ctDNA test can potentially bypass all these drawbacks. Firstly, all that would be required is a blood sample so the test would be non-invasive. Some research projects on ctDNA tests for CRC involve sequencing ctDNA to call mutations indicative of cancer. Sequencing technologies have been around for long enough now that most hospitals have efficient workflows for sequencing samples, which could reduce the costs of a ctDNA CRC screening assay. Currently, the biggest drawback of using ctDNA in early detection of cancer is the low ctDNA fraction in plasma. Circulating tumour DNA makes up as little as >0.1% of DNA in the blood during the early stages of disease.
The non-invasive nature of ctDNA also presents the opportunity of monitoring disease progression over time. As discussed before, tissue biopsies are uncomfortable and can result in other complications. Imaging masses less than 5 cm in diameter is also difficult. Collectively, this means monitoring the progress of cancer in patients is difficult using these tools. Pancreatic cancer has one of the higher mortality rates among all cancer types. Mutations in a gene called KRAS are associated with over 90% of all pancreatic ductal adenocarcinoma. Mutations in KRAS detected in ctDNA have been shown to be predictive of prognosis and response to treatment. Absence of ctDNA carrying KRAS mutations after surgical treatment has been associated with better prognosis for patients.
Cancers can also develop resistance to therapeutics over time. Resistance can be predicted by the appearance of specific mutations in the cancer genome. A classic example of this is mutations in EGFR predict resistance to tyrosine kinase inhibitors in non-small cell lung cancer. ctDNA has been used to detect this mutation to predict when treatment will become ineffective and change therapeutics.
Disease monitoring requires repeated sampling of tumour DNA, a possibility ctDNA affords and tumour biopsy does not. The limitations of ctDNA for monitoring disease are again rooted in low abundance. It is difficult to accurately detect SNPs in a ctDNA fraction of >0.1%, but the technology will likely catch up. Digital PCR, for example is able to detect low abundance targets exploiting Poisson statistics.
Circulating tumour DNA has a range of biological features that can be leveraged in machine learning models for a range of applications. The problem we want to solve is distinguishing ctDNA fragments from healthy DNA originating from normal cells (cfDNA). This problem is perfect for machine learning models because we can train an algorithm on ctDNA features against healthy cfDNA. We know ctDNA fragments are generally shorter than cfDNA fragments. The methylation profile of ctDNA is also markedly different from cfDNA. And of cause, mutations in ctDNA including single nucleotide polymorphisms and copy number alterations carry distinguishing features for a model to learn from. Machine learning (ML) models have been presented in research that can accurately predict patients with cancer with high sensitivity and positivity. As a screening assay, combining ML methods with ctDNA features is advantageous because it would be cheaper than some screening programs, and collecting ctDNA is a non-invasive procedure.