The moment of truth for cancer liquid biopsy

I previously asked whether the Cancer Liquid Biopsy was ready for primetime (QH). We will find out soon enough as various methodologies implementing this potential diagnostic breakthrough will be rigorously tested over the next few years. More specifically, the National Cancer Institute will sponsor prospective trials to evaluate multi-cancer early detection (MCED) tests in 24,000 people over four years (Nature Biotechnology News). The vast majority of MCED tests rely on sampling the blood for diagnostic signals, and hence are considered liquid biopsy approaches. Seventeen companies have stated that they would like to participate in the trials.

The idea of the liquid biopsy is to try to find evidence of disease from a sample of blood rather than sampling the diseased tissue directly. In the case of cancer, tumor cells and debris from these cells slough off into the blood where they can be captured by a blood draw and then tested. One advantage is the convenience of easier access to blood rather than internal tissue, but the main hope is the the possibility of detecting the cancer (disease) at an earlier stage before it is visible on imaging or manifests any symptoms. For all cancers, survival rate is higher the earlier the cancer is detected. The question is whether the MCED liquid biopsy tests will be accurate enough diagnosing the cancer with sufficiently few false positive and false negatives to not cause harm compared to current standard diagnostic procedures.

The signal for these MCED diagnostics is primarily DNA or protein. Traditionally cancer diagnostic methods relied on detecting some specific antigen (protein) associated with a cancer such as prostrate-specific antigen (PSA) for prostate cancer or CA125 for ovarian cancer. These could be sensitively and quantitatively measured using an immunoassay like ELISA. But thanks to the remarkable advances in DNA amplification from a tiny sample (PCR), high-throughput DNA sequencing and more generally, achieving a better understanding of the wide variety of DNA signatures (both in the genome as well as the epigenome) associated with cancer had led to a proliferation of DNA-based techniques.

One of the most talked about MCED companies is Grail. They employ a technique called whole genome bisulfite sequencing (WGBS) which is a modification of whole genome sequencing to map areas of methylated DNA (QH). Methylated DNA is associated with bundling of the DNA into compact chromatin that prevents expression of genes. Scientists have identified patterns of methylated genes that are associated with specific types of cancers such as methylation of tumor suppressor genes that turn off genes that can inhibit tumor formation.

Although WGBS is quite promising, the field has not settled on a single technique. To the contrary, a whole plethora of methods are being pursued by the “at least 27 companies” developing MCED tests (Figure 1). These alternatives to WGBS include the following:

1. Whole genome sequencing (WGS) of specific cancer-associated genetic mutations such as SNPs or copy number variations (CNVs).
2. Fragmentomics focuses on characteristics of cell-free DNA fragments such as size and sequence at fragment ends. These fragments reflect chromatin structure (i.e. nucleosomes that package DNA and protect it from nucleases) which vary between tumor versus normal cells.
3. Proteomics techniques such as mass spectrometry can assess a panel of protein biomarkers instead of just one or two.
4. Circulating extracellular vesicles (EVs) are small particles (vesicles) secreted by cells including tumor cells containing proteins from the cells (tumor). EVs can be purified from a blood sample and then the proteins analyzed by a proteomics technique.
5. Machine learning can combine the information from more than one of the above approaches (i.e. multiple biomarkers).

According to the article, seventeen companies are interested in participating in the NCI clinical trials. But we can expect data even before then. Grail in particular has been aggressively running its own trials. I have written previously about a retrospective study published in 2021 with approximately 15,000 enrolled subjects who were screened for over 50 cancer types (e.g. lung, pancreatic, etc.). They estimated the specificity of their method to be 99.5% (true negative rate = percent negative that are predicted negative) and the sensitivity to be 51.5% (true positive rate = percent positive predicted to be positive) which increased with cancer stage as one might expect. Because the trial was retrospective, the cancer diagnosis of all the patients were already known before the analysis began, but half of the dataset was hidden from the machine learning algorithm until it had been finalized. The concern is that there was indirect or unintentional information leak that allowed the algorithm to gain insight into the test data before the final prediction. A prospective study in which the predictions are made before the final cancer diagnosis is known would resolve this concern.

To address these criticisms, Grail is currently running the NHS-Galleri trial in the UK which is a randomized controlled prospective trial enrolling 140,000 subjects aged 50-77 who have not had a cancer diagnosis over the past three years. The Grail MCED test will be tested as an adjunct to current screening standard of care for the various types of cancers. In other words the control group will receive standard cancer screening, whereas the experimental group will receive both standard and Grail cancer screening. The primary outcome of the trial is the accuracy of stage 3 and 4 cancers diagnosed over the trial period of three years. The Grail test will be administered each of the three years.

By comparison the NCI trials will have a smaller number of subjects but will test multiple MCED methods. For example, in the prospective trial, a number of companies can submit predictions over the same pool of patients, and then be evaluated in parallel. The only limitation is how much blood each patient will submit (which has to be divided up among the different tests). In particular, it will be informative to see how the other liquid biopsy approaches perform head-to-head with Grail especially with Grail accumulating separate data on its diagnostic accuracy and thus being able to serve as a type of reference.

The Grail NHS-Galleri trial finished enrolling in 2022, and so we can look forward to the results in three years around 2025. It will be a landmark moment for the cancer liquid biopsy. Then a few years after that we can expect data from the NCI trials to start to roll in giving a chance for Grail's competitors to make a splash especially in comparison to the acknowledged leader in the field.

By the end of the decade we should have a much clearer picture of whether the cancer liquid biopsy can become a reality that transforms cancer medicine and becomes part of standard cancer screening.

Figure 1. Schematic diagram of cancer liquid biopsy. First, a blood sample is taken from the patient. It can be harvested for proteins, extracellular vesicles (EVs), circulating tumor cells (CTCs), and cell-free DNA (cfDNA). The DNA can be analyzed for genomic (sequence) patterns and epigenomic patterns (i.e. methylation) that are associated with cancers. The information from multiple types of biomarkers can be combined by computer analysis. Finally, predictions are made on the likelihood of various cancers making possible multi-cancer early detection (MCED) (link).