A Drop in the Ocean

Molecular biologists have recognized for decades that when both diseased and healthy cells die, they slough their contents – including DNA – into the blood stream, and that the genetic variation present in these cells might be discernible in circulating cell-free DNA (cfDNA) found in blood plasma. The community theorized that it might be possible to monitor the status of solid tissues through a minimally invasive blood draw – a “liquid biopsy.”

And indeed, liquid biopsy can detect many types of genetic variations, ranging from single nucleotide mutations to amplifications (or deletions) of entire genes. But, in a typical blood sample, such mutations are usually present in a tiny fraction of the detected cfDNA (the majority of cfDNA being derived from the patient’s healthy cells), presenting a classic “needle in a haystack” problem.

Real-time limitations

Real-time PCR (rtPCR) – the basis of the oldest commonly used qualitative or semi-quantitative liquid biopsy tests – is too blunt a method for some applications; for example, detecting biomarkers of fetal trisomy in a mother’s blood. Why? Because it is limited in its ability to precisely measure the slight change in fetal chromosome copy number when compared with the background of healthy maternal cfDNA. Similarly, rtPCR struggles to assess whether a single base mutation from a tumor, which often occurs at very low frequencies, is present in a patient’s plasma either before or after treatment.

Facing very low concentrations of target species (perhaps only a few copies per milliliter of blood) creates a need for highly sensitive, specific, and precise tools. It is critical to have a platform that can discriminate single-nucleotide mutations, whether they’re found in cfDNA fragments, miRNAs or other RNA types, from an abundant background of highly similar wildtype sequences. In other cases, it is crucial to have the capability to detect small changes in the copy number of a key gene target. Laboratories need robust technology with high-reproducibility. And to be adopted widely, the technology must enable rapid turnaround, be cost-effective, and be compatible with low and high throughput needs. To meet these needs, a new solution is emerging.

Digitizing nanoscale reactions

Droplet digital PCR (ddPCR) can measure samples with much higher precision and sensitivity than conventional rtPCR. Reactions can be “digitized” by subdividing a PCR reaction into nanoliter-sized compartments, or partitions, where the sample in each partition is separately amplified. The “positive” partitions containing the specific target(s) being assayed generate a strong fluorescent signal, and the “negative” partitions emit only weak fluorescence. The platform then counts the number and fraction of positive fluorescent partitions to determine the concentration of the target sequence in the sample.

The concept behind ddPCR was developed in the early 1990s but, back then, it was not easy to perform nor cost-effective. Researchers initially used the wells of PCR plates as individual partitions, and later, they used pre-formed nanofluidic chambers in microfluidic arrays. The technique was then further developed to overcome these limitations and provide a high throughput, affordable and scalable platform that delivers absolute concentrations of DNA or RNA targets with high precision and sensitivity, without the need for a standard curve.

The key to modern ddPCR technology is the reliable generation of uniform, nano-sized droplets from the input sample, which required an inexpensive, disposable microfluidic chip. The droplets need to be stable enough to be transferred from the chip to a PCR plate for thermocycling, and then to be autosampled and microfluidically transported within the dedicated droplet reader.

Instrument control and analysis software was also needed to perform quality control on the droplet fluorescence signals in two wavelengths (corresponding to the emission of the two different dyes used). The resulting software, QuantaSoft, can reliably identify droplets and assign them to a particular category: double negative (droplets with a signal from neither of the dyes), single positive droplets, or double positive droplets (those containing a strong signal from both dyes).

Additionally, the software was developed to calculate the concentration of detected target species, and to display the results in various types of plots and charts showing concentrations, ratios and numbers of accepted droplets per sample. Lastly, it was necessary to develop a variety of ddPCR reaction master mixes for different applications and assay design strategies for producing either Taqman probe assays or EvaGreen probeless assays that function well in droplets.

The road to adoption

Although clinical adoption of ddPCR for liquid biopsy of solid tumors is still in its early stages, about half a dozen molecular diagnostics labs are offering such tests for both liquid biopsy and tissue genotyping. So far, it has garnered a positive reception (1)(2).

Several academic centers have adopted the technology, including the Dana Farber Cancer Institute (DFCI), MD Anderson Cancer Center, and the Olivia Newton-John Cancer Research Institute. Some independent molecular testing labs have adopted the technology as well, including Biodesix Inc., which largely serves physicians at community care centers. Biodesix developed and validated a ddPCR-based test panel for non-small cell lung cancer (NSCLC) mutations, which turns around results in about 33 hours of receiving patient blood samples.

Pathologists at the DFCI/Brigham and Women’s Hospital Pathology Lab are using ddPCR liquid biopsy testing to measure EGFR-sensitizing and resistance mutations in lung cancer patients to rapidly identify those who are eligible for tyrosine kinase inhibitor therapy. Nowadays, patients who do not have access to ddPCR technology must wait an average of two weeks (if the patient is newly diagnosed) or four weeks (if the patient has relapsed) while their tissue samples undergo next-generation sequencing, if the samples are available at all. 

Currently in the labs that have adopted it, ddPCR is mainly serving NSCLC, melanoma and thyroid cancer patients. And importantly, the institutions using ddPCR have found success obtaining reimbursement for these tests. The approach to testing is ideal in two scenarios: i) when the physician needs to identify the potential presence of a recurrent hotspot mutation in a cancer patient quickly and inexpensively to make a treatment decision and ii) when a physician wants to monitor a patient’s disease progression or response to treatment over time.

These types of monitoring investigations are informing the evaluation of patient responses to immunotherapies such as anti-PD1 (for example, in melanoma and lung cancer). ddPCR could also potentially help physicians make “real-time” adjustments to a patient’s therapy based on changes in tumor DNA levels in the plasma; however, we must demonstrate ddPCR’s clinical utility for this purpose. The approach is less effective for genotyping a cancer where there are numerous potential disease drivers, and when these mutations occur at low frequency in the population.

Future growth of ddPCR in the clinic appears likely, as scientists have discovered a number of actionable markers for new indications (resistance to anti-hormone therapy, for example) in breast and prostate cancers that can be readily tested. Further, numerous groups are running clinical trials that incorporate ddPCR, demonstrating its potential use in clinical decision-making (3)(4)(5).

A word of caution

Despite the positive results we’ve seen so far, I would alert laboratory professionals to the risks of “magical thinking.” Even though digital PCR is a marvelous tool that can deliver absolute counts of target molecules without reliance on a standard curve, it is still necessary to validate your assays to verify that they are giving accurate answers. Of course, a poorly-validated assay will not perform well, even in ddPCR. But even a well-designed assay occasionally doesn’t reach its target efficiently, particularly for large DNA fragments that are multiple kilobases long. In these cases, assays may significantly undercount the number of target molecules present. Here, a pathologist should reduce the size of the template fragment – such as by restriction enzyme digestion – to alleviate this. However, it should be noted that this is only seen in a small minority of cases, and is therefore an infrequent problem.

Similarly, digital PCR does not remove the need for reference standards or controls that can confirm that the testing run was successful and that the results can be trusted. In this regard, it may even be prudent to use spike-in controls to assess pre-analytical sources of noise, such as extraction efficiency, and unusual inhibitors present in a sample, such as another drug in the patient’s blood.

In addition, pathologists should be careful about using the terms “precision” and “accuracy.”  Even the best scientists are prone to conflating these two terms and assuming if a result is reproducible, that it is accurate. But even if the results are reproducible, the assay needs to be validated before its results can be trusted in terms of accuracy.

Ask the right questions

It is very important for a physician or scientist to understand exactly what question he or she is trying to answer, and to choose a technological approach that will let them collect the right information. If a researcher is looking to discover new cancer biomarkers, he or she may wish to broadly profile a patient’s DNA variations using NGS. But if a physician is trying to determine the optimal treatment for his or her patient, particularly after the patient’s tumor has been profiled, it may be more expeditious and economically sustainable to use a more focused approach – in which case, ddPCR is an excellent tool.

George Karlin-Neumann is the Director of Scientific Affairs at Bio-Rad Laboratories’ Digital Biology Center.