A Fluid Future
Why are there inconsistencies between different liquid biopsy tests – and how can the field evolve to maximize accuracy and impact?
Luke Turner | | Longer Read
Advances in disease profiling over recent years have opened new doors in the quest for early diagnosis and personalized treatment, with metastatic prostate cancer being a case in point. As the most widespread malignancy affecting men in the USA – 174,650 new cases and 31,620 deaths are predicted for 2019 – prostate cancer research is crucial (1). One such breakthrough is the use of liquid biopsy tests to detect circulating tumor DNA (ctDNA) in patients’ blood, offering a minimally invasive method of disease profiling.
The field is still young, but already there is a wide range of tests that all claim high performance. And that’s no surprise; the global liquid biopsy market is predicted to grow from US$310 million in 2016 to a staggering $1.2 billion by 2023 (2). With respect to prostate cancer, fierce competition for an early foothold in the market has led to a flurry of commercially available liquid biopsy tests in the US, all with the same ambition: to improve diagnosis and management.
Metastatic prostate cancer is a highly heterogeneous malignancy associated with a wide range of potentially actionable mutations (3). The application of next-generation sequencing (NGS) techniques to primary tumors has begun to identify potential biomarkers for prostate cancer detection and characterization. These targets include circulating tumor cells (CTCs) and ctDNA, which reveal unique and complementary information about the tumor.
Liquid biopsy can be used in a variety of ways with solid malignancies – to detect actionable mutations, to inform treatment decisions, to monitor treatment response or measure efficacy, to detect disease recurrence, and to identify resistance mechanisms (4).
Although ctDNA appears to hold great potential in monitoring and profiling the evolution of tumors, a study by Gonzalo Torga and Kenneth Pienta compared two commercially available liquid biopsy tests – Guardant360 (Guardant Health) and PlasmaSELECT (Personal Genome Diagnostics) – and reported high levels of incongruence (5). We spoke to four experts to unpick the inconsistencies, explore areas for improvement, and ask where liquid biopsy is headed.
Work in Progress
Gonzalo Torga is a postdoctoral fellow at John Hopkins Medicine who conducts translational research while developing new therapies for prostate cancer.
Why did you compare the two tests?
Actually, that wasn’t our original plan; it stemmed from wanting to know which of the two tests worked best. We were considering how to best serve patients in the clinic when my boss told me about two new, commercially available platforms that used targeted NGS of ctDNA to detect actionable mutations. I simply asked, “Which is the most accurate test for our patients?”
What did you find?
First, we only considered genetic alterations that both platforms claimed to cover, which amounted to 42 different genes. Of the 40 patients in our study, 12 showed complete congruence between the two tests, six had partial congruence, and 16 had no congruence at all. The remaining six patients were not evaluable for patient-level congruence (5).
One of the problems we encountered was the high number of reported mutations in the samples. The results showed approximately a 40–50 percent mutant allele fraction, so I contacted the companies and asked them to check whether this was due to germline DNA or from the tumor itself. Although germline DNA mutations can be used to make treatment decisions, the issue is that the PlasmaSELECT test seems to report germline mutations as somatic mutations, giving rise to the divergence between the two tests. We believed that germline contamination was responsible for the relatively high allele fraction; however, neither company was able to confirm for us whether germline DNA was influencing the results.
The concerning part is that we won't be able to base any clinical decisions on these tests until they can either rule out germline contamination or report germline mutations as such. Only then will we be able to trust the consistency of the results.
How could this inconsistency affect patients?
At the time of our study, we were trying to select patients for clinical trials. However, one of the big problems was that the tests have different turnaround times – one takes two-to-three weeks, whereas the other takes four weeks. You can imagine the difficulties this caused. For example, when we were about to take a patient forward for a clinical trial based on the results of one test, the results from the other test would come back with notability different results from the same blood sample. This was obviously controversial and prompted us to put these results into a paper, because you simply can’t make any reliable clinical decisions with this level of uncertainty.
Other problems currently impeding the development of the tests are processing times and methods. Different blood processing methods have a strong influence on the levels of cell-free DNA (cfDNA) and ctDNA and this should be considered when evaluating ctDNA in peripheral circulation. This could mean that if you analyze DNA after it has been processed for 12 hours, the mutations that you find will be different to those that appear after 24 hours. Consequently, the long processing times currently associated with liquid biopsy tests result in inconsistencies when reporting mutations. This disparity could represent an important roadblock for the standardization of these tests (6). I also found that, although these companies are confident in the tests’ ability to detect mutations, the letters that they submitted in response to our paper indirectly acknowledge poor inter-assay performance. This means that, if you submit the same blood sample twice on the same day to the same company, you might still get different results – why? Because a single mutation might be present in one blood sample but not the other. The main issue here is that, if a rare mutation is present in an earlier sample but then isn’t detected again at a later date, you might wrongly think that it has disappeared and conclude that a certain treatment is working. In fact, it may only mean that the second sample does not contain every mutation present in the first.
Alongside germline contamination, we believe there could have been issues with clonal expansions. These appear in the blood – especially in older people – and resemble tumor cells in terms of the mutations they carry, but they are not associated with cancer. A major concern with liquid biopsy at the moment is that there’s no way to distinguish between benign clonal expansions and cancer by looking at ctDNA from the plasma. I think that one way to fix these problems would be to analyze DNA from white blood cells instead; this would reveal the patient’s germline status and whether mutations have originated from clonal hematopoietic expansions. It would also mean that no further sampling is required, because white blood cells will be present in the original blood sample.
How can the issues be resolved?
Ultimately, these tests need to be standardized. For example, when we started testing for prostate cancer, we were advised to use the same lab every time because test location affected the results. We would find differences in prostate-specific antigen (PSA) values that were purely due to technological differences between labs. Over time, those tests became more standardized, and now it doesn’t matter where they are carried out. We need to reach a point where liquid biopsy tests will report the same mutations and the same mutant allele fractions every time, and I think that regulatory agencies need to play their part in achieving this standardization. At the moment, the tests are only approved by the Clinical Laboratory Improvement Amendments (CLIA); however, the ultimate ambition is obtaining approval from the US Food and Drug Administration (FDA), which can only happen if the tests demonstrate clinical utility. The American Association of Pathologists and the American Society of Clinical Oncology have recently published a joint review of these tests, contraindicating their use for monitoring or making treatment decisions at any stage of cancer (7).
What are your hopes for the future?
The technology is exciting for the oncologists and pathologists who will carry out these tests in the future. The prospect of monitoring the evolution of a patient’s treatment frequently and noninvasively is enticing – it means treatments can be changed before something goes wrong.
Alongside the development of plasma-based assays, tests are also being developed that include proteins in their analysis. The cancerSEEK test is one such liquid biopsy intended to detect early-stage cancers with high specificity to minimize false negative results. The test has shown early promise in detecting multiple tumor types and localizing cancers to their anatomic sites of origin (8).
Liquid biopsy is definitely the future; these tests will play a vital role for cancer patients. But it is clear that much more work needs to be done before they are ready for routine clinical use.
Seeking New Standards
John Simmons is Vice President of Translational Medicine at Personal Genome Diagnostics.
Why was there incongruence between the PlasmaSELECT and Guardant360 tests?
When we look at the study carried out by Torga and Pienta, several things come to mind. One of the most important issues is standardization; much of the incongruence is related to where reporting thresholds fall. The problem is that ctDNA may be present in the blood at very low levels that, for currently available tests, are near the limit of detection. Alterations at such low levels comprise very few mutated molecules and might not even be detected by the same assay after repeat testing due to sampling probability (the chance that any given sample may not contain a particular mutation). The PlasmaSELECT test therefore categorizes these mutations as indeterminate, because they are below the threshold of consistent detection. The study that found high levels of incongruence didn’t distinguish between these types of mutations, meaning around half of those labeled discordant fell below PlasmaSELECT thresholds.
There is a need for standardization in this space. We’re working to resolve issues with reporting thresholds by going to the FDA and other regulatory bodies. Another important factor is the clinical evidence; both of these tests are accredited by CLIA, but have yet to go through the FDA. For a diagnostic test to be approved by the FDA, you have to be able to demonstrate clinical validity. Therefore, much of the information about how and when to use the assays would be part and parcel of an FDA application – something on which we are currently working. At the moment, ctDNA tests are relatively new to the field; although we’re experiencing some inevitable growing pains, I think access to this cutting-edge technology can only be a good thing for oncologists and pathologists. Our next step will be to take the test through the regulatory authorities and to hone in on clinical validity and standardization.
How will you work toward standardization?
Standardization has many different interpretations. In our case, there are two parts to what we’re trying to achieve. First, a degree of standardization comes from taking a test through the FDA’s clearly defined intended use requirements; you need transparency in the analytical performance and clear definitions as to what you are reporting. Another layer of standardization that would greatly help the interpretation and comparability of ctDNA tests as a field concerns filtering approaches for germline variants and mutations associated with clonal hematopoiesis of indeterminate potential (CHIP). At the moment, there is no gold standard for applying germline and CHIP filters, and as such, you see a variety of different approaches that can ultimately lead to incongruence in reporting.
Why do germline variants pose such a problem?
These tests use total cfDNA from patient plasma. This means that DNA from both tumor cells and non-tumor tissue is being sequenced, but as the tests are only intended to identify somatic mutations, germline single nucleotide polymorphisms (SNPs) must be removed bioinformatically, which is known as filtering. Currently, as a field there is no standardization of filtering approaches, resulting in variation from test to test. Most filtering approaches rely, at least in part, on information from publicly available databases that catalogue germline polymorphisms along with information on frequency in different populations. Even when using the same databases, different tests apply distinct filtering thresholds that can lead to incongruent reporting of rare germline variants.
There are also algorithmic approaches that can be applied beyond the use of databases, but again, these approaches vary by test and can also be confounded in cases with high levels of ctDNA in the blood. To add to the complexity, there are some drug treatments where efficacy is related to alterations in certain genes regardless of whether the alteration was germline or somatic. There are definitely technological limitations in this space, but it’s also crucial for us as a community to devise some standards for germline filtering in both plasma and tissue-based tests.
Where are you directing future research efforts?
We are using our in vitro diagnostic assays in clinical trials to demonstrate the clinical validity and utility of ctDNA-based tests in a variety of tumor types. This requires collaboration between academic researchers and pharma partners to amass the evidence we need to take the test through regulatory agencies.
I think there is a growing amount of retrospective evidence from a number of solid tumor types that indicates we can use ctDNA to identify treatment response and disease progression more quickly than with imaging alone. The next important step for us is to demonstrate prospectively the clinical utility of changing treatment based on changes in ctDNA levels or detecting minimal residual disease by ctDNA. Many of these studies will be focused on earlier stages of disease than those where we use ctDNA assays today.
What does the future hold for liquid biopsy?
I see liquid biopsy tests entering a more defined role in treatment selection for late-stage patients for whom tissue testing is not appropriate or tissue is not available. I would say that is the lowest-hanging fruit for these assays and this is where we are directing many of our clinical studies. Beyond that, I see additional applications of ctDNA-based assays in treatment response monitoring and detection of minimal residual disease.
NGS tests – both tissue- and ctDNA-based – have reached the marketplace at an incredible speed. Now, we are starting to see strong indications that these tests are developing clinical maturity, meaning that they’re likely to be used in routine settings, rather than just as a last resort. But to realize the great potential of liquid biopsy, I think that we as diagnostic manufacturers need to provide education and outreach to oncologists and pathologists in routine practice. We also need to think about how to incorporate the interpretation of tissue and plasma NGS assays into medical education and residency, so that people coming into the field are comfortable with these tests. That way, oncologists and pathologists will be well-positioned to use the results in their routine clinical practice once the tests are approved by the regulatory authorities.
The Question of Intended Use
Ryan Dittamore is the Chief of Medical Innovation and Head of Translational Research Partnerships at Epic Sciences.
What is the AR-V7 liquid biopsy test?
When a patient with metastatic prostate cancer has failed a first androgen receptor (AR)-directed therapy (either abiraterone or enzalutamide) the oncologist is faced with a difficult question. The patient can either receive the other hormone therapy or start chemotherapy. As you can imagine, patients are often reluctant to go on chemotherapy because of its high toxicity, whereas the AR-directed therapy – an oral pill with low toxicity – is a more attractive prospect. The stakes surrounding this decision are high; if the patient doesn’t react well to the second line of treatment, the disease can progress rapidly. Another factor to consider is the high cost of AR-directed therapies. These drugs bring in over $4 billion globally, so they are extremely expensive for healthcare systems to administer, and there is no way to predict how a patient will react to them.
To address this issue, we tested almost 20 different biomarkers to identify one that could help determine whether AR-directed therapy will work. The outcome of this search was the androgen-receptor splice variant 7 messenger RNA (AR-V7), which encodes a functional protein detectable in clinical specimens. We found that the presence of AR-V7 is associated with resistance to abiraterone and enzalutamide. This is because the protein that AR-V7 encodes has no ligand-binding domain, which is the target of androgens – it is blocked directly by enzalutamide and indirectly by abiraterone. Therefore, neither of these therapies will work if you have an AR-V7 protein that is essentially constitutively signaling the cell to grow.
Because the AR-V7 protein is most specific in the nucleus, we adopted a “no cell left behind” approach to search for it in the nuclei of CTCs. Instead of trying to sort tumor cells from leukocytes and potentially losing CTCs, this platform places every single nucleated cell from the blood on a series of glass slides. From there, we stain those slides – each containing three million nucleated cells – and use digital pathology to identify CTCs from a multiparametric feature. This is essentially the same process that pathologists go through when diagnosing cancer using tissue morphology, architecture, and protein chemistry, but the digital imaging aspect enables us to analyze six million cells for each patient sample. Once we have sorted the cells and identified the abnormal CTCs, the test looks at the AR-V7 protein in the nucleus and analyzes its chemistry to determine whether or not AR-directed therapy will work.
How accurate is the test?
There are two aspects to consider when it comes to the accuracy of the test. In terms of detecting single tumor cells, the limit of the test is essentially one cell per milliliter of blood. From a protein perspective, there’s no way to definitively say how much AR-V7 is present in a patient, because AR-V7 itself is a resistance mechanism that occurs only after a patient has been treated with hormone therapy. It is also only found in a subset of cells, so the only way to measure it would be to take a biopsy from every tumor lesion in a patient. The problem is that this can’t be done – tumors usually spread to inaccessible bones.
Therefore, when talking about accuracy, we have to look at the clinical certainty of the test. We focused on what a positive result actually means for the patient, which is that they experience a primary PSA resistance and immediately resist AR signaling inhibitors. Essentially, they don’t respond to any further AR-directed therapy and their overall survival is very short. For this reason we can’t afford any margin for error in the test results, so it has to be incredibly specific. It’s also predictive – the data suggest that patients who test positive could as much as double their life expectancy if they revert to chemotherapy over AR-directed therapy.
What are the differences between this approach and the PlasmaSELECT and Guardant360 tests?
The main difference is the intended use population. The AR-V7 test is directed at a very specific subgroup of patients with metastatic, castrate-resistant prostate cancer who are considering a second AR signaling inhibitor. The other two tests are aimed at making treatment decisions for many stage III or stage IV metastatic cancer patients, which includes millions of people. The problem is that, because the tests return a long list of mutations, it’s not always clear what action to take.
For a test to have clinical utility it must be fit for purpose, which is achieved by designing it for a specific clinical indication that therapy decisions can be based upon. In addition, the test must impact clinical outcomes, such as patient survival, when its use is compared with non-use. I think a great example of this is the programmed death ligand 1 (PD-L1) question, which pathologists deal with extensively at the moment. PD-L1 expression is a mechanism of immune evasion that various malignancies exploit; it’s usually associated with poorer prognosis. There are a variety of PD-L1 tests that focus on different cell types and different potential treatments – PD-L1 is not the same across different cancer types and decisions points. Just because you find a biomarker and develop a test, that doesn’t necessarily mean it has the same clinical interpretation or clinical value as you switch from one disease to another.
If, for a single clinical question, you have two different tests with contrasting ways to report genomic information that return differing results, how is the physician meant to make a treatment decision? The problem occurs when we don’t know what these biomarkers mean in the context of the clinical question that the physician is trying to solve.
I think that, with liquid biopsies, we need to think very carefully about the intended use population along with validating the tests to clinical endpoints. Understanding – and using – the PlasmaSELECT and Guardant360 tests would have been much easier if the manufacturers had provided some form of clinical interpretation. “If you see this biomarker in biochemically recurrent prostate cancer, we know that the patient will or will not respond to the drug that we’re administering.” But those studies haven’t been done, so the result is a lot of confusion. If we want to avoid this and accelerate precision liquid biopsy tests, we have to be very careful about answering a clinical question.
What are your next steps?
In terms of the AR-V7 test, we are continuing to commercialize it in the US with Genomic Health. We are also evaluating opportunities to globalize the test outside of the US; for instance, in Europe, Canada, Asia, and South America. In addition, we have a deep pipeline of tests in development – those are focused on therapy selection, treatment monitoring, recurrence monitoring, and identifying resistance to therapies. Going forward, I expect a number of tests to come to fruition that have a very focused clinical question. The key, though, is that they must be supported by clinical data and studies that demonstrate their value, so that physicians aren’t left guessing with the results.
A Bright Future
Jacqui Shaw is Professor of Translational Cancer Genetics at the Leicester Cancer Research Center and leads the cfDNA Advisory Group for Genomics England.
What are liquid biopsy’s main advantages and limitations?
A liquid biopsy test that is being used to detect a particular mutation – or other tumor specific change – needs to have specified limits for sensitivity and specificity. The current limitations to liquid biopsy have already been explored in detail but, put simply, you can’t accurately report back on a test that hasn’t been validated. The most sensitive tests at the moment are those that will detect few molecules of tumor DNA with high sensitivity and specificity; either using high-depth sequencing with a molecular barcoding or target enrichment strategy, or droplet digital PCR.
Each of these approaches will have a reported minimum detection threshold; for example, in 20 nanograms of DNA, the minimum might be five molecules of tumor DNA. In this case, if you detect two mutant molecules, then the test result should be reported negative because it’s below the minimum accepted threshold for the test. The advantage of liquid biopsy, though, is that you can easily carry out repeated sampling – so if you repeat the test two months down the line and there are six tumor molecules, then the result becomes positive. With a tissue biopsy, there is always the possibility that results are affected by sampling bias or intratumoral heterogeneity, alongside the fact that biopsy might not be possible anyway. Despite the current caveats associated with different liquid biopsy technologies each with slightly different sensitivity and specificity readouts, the offer of real-time, repeated sampling and monitoring potentially gives them a big advantage.
Do you think the genomic complexity of the tests hinders the interpretation of results?
I think you could make a case for both sides of this argument. Some people think there is too much information involved with genomic tests and that a lot of the data may not be relevant to the clinical question. But I think the point of a genomic test is to produce a baseline of information to store and use as a reference point over time. The pathologist or oncologist can, of course, choose to filter these data, allowing them to see only particular genes of interest or actionable mutations. One of the big breakthroughs with liquid biopsy is that we are now looking at a genome-wide analysis. Although there are challenges associated with the large amount of data that needs to be stored, mined, and analyzed, it’s crucial to have this information to build cancer genome profiles so we can better understand how cancers change over time.
How does the commercial side of liquid biopsy affect the field’s development?
We’re living in a world where there’s an increasing amount of engagement between industry, research institutions, and healthcare providers – and I don’t think that’s going to change any time soon. In my opinion, having a number of competitors developing various technologies may lead to the delivery of high-quality tests in a shorter timescale. Generally, in the world of science, it’s great to have collaborators and competitors alike; they drive the field forward, and I don’t think that’s any different in liquid biopsy.
There are a number of large, ongoing clinical studies and trials in the field that will inform the future role of liquid biopsy in clinical decision-making. If they perform well, then I believe that liquid biopsy tests will lead to more patient-tailored interventions or changes in treatment at earlier stages. Obviously, the ultimate success is improving outcomes – which is why one of the most enticing possibilities of the technology is profiling cancers in patients where it’s not possible to physically access the tumor. The future is bright because there are so many potential opportunities for liquid biopsy!
- RL Siegel et al., “Cancer statistics, 2019”, CA Cancer J Clin, 69, 7–34 (2019). PMID: 30620402.
- Stratistics Market Research Consulting, “Liquid biopsy - global market outlook (2017-2023)” 2017. Available at: bit.ly/2SmkGW2. Accessed January 14, 2019.
- GJ Morrison, A Goldkorn, “Development and application of liquid biopsies in metastatic prostate cancer”, Curr Oncol Rep, 20, 35 (2018). PMID: 29572775.
- SI Ou et al., “Liquid biopsy to identify actionable genomic alterations”, Am Soc Clin Oncol Educ Book, 23, 978–997 (2018). PMID: 30231331.
- G Torga, KJ Pienta, “Patient-paired sample congruence between 2 commercial liquid biopsy tests”, JAMA Oncol, 4, 868–870 (2018). PMID: 29242909.
- S Parpart-Li et al., “The effect of preservative and temperature on the analysis of circulating tumor DNA”, Clin Cancer Res, 23, 2471–2477 (2017). PMID: 27827317.
- JD Merker et al., “Circulating tumor DNA analysis in patients with cancer: American Society of Clinical Oncology and College of American Pathologists joint review”, J Clin Oncol, 36, 1631–1641 (2018). PMID: 29504847.
- JD Cohen et al., “Detection and localization of surgically resectable cancers with a multi-analyte blood test”, Science, 359, 926–930 (2018). PMID: 29348365.