In the busy hallway of an urban hospital’s diagnostics wing, Dr. Miles J. Bennell scrambles into a moving crowd of staff and patients. Bennell is frantic, his eyes wide with excitement, struggling to catch his breath.
Excitedly, he grabs one person in the hall.
“They’re here, already,” he bellows, only to be shaken off.
Spinning on his heels, he bumps into someone else.
“You’re next,” he yells, repeating the call as he bounces from person to person.
Warnings of an invasion of alien pods threatening to replace us all?
No, just a melodramatic diagnostician, excited about the opportunities for less-invasive biopsies for his cancer patients.
Pass the tissues
Despite the rapid advances in identifying and understanding the nature of driver mutations and resistance markers in various cancers—information that is going a long way to individualizing patient care—the challenge of fully characterizing a cancer within a single patient remains significant. Even in situations where a solid tumor exists, it is not always possible to obtain a tissue biopsy due to clinical inaccessibility or unacceptable risks to patient safety.
Likewise, with a growing acknowledgement of the cellular diversity inherent within tumors, a single biopsy sample may not reflect the disease’s heterogeneity, whether we are considering just the tumor or including its microenvironment.
Cancer is not a static disease, however, and in its connection to the blood stream and other bodily fluids, it leaves a biological trail. The trail comprises circulating tumor cells (CTCs) that slough from the cancerous mass, as well as cell-free DNA and RNA (cfDNA and cfRNA) that are released as cancer cells turn over or through other processes.
It is to these biological packets that several companies and research groups have turned their attention, looking for ways to characterize cancers more fully and less invasively in tests known as liquid biopsies.
Just over 145 years ago, Thomas Ashworth noticed free-floating cells in the peripheral blood of cancer patients that looked similar in shape, size and morphology to cells within the tumors, which eventually led to speculation that this was the route by which tumors metastasized to other tissues.
It has only been much more recently, however, that technology has allowed researchers to isolate and quantify these CTCs, and it was only in 2004 that FDA clearance was achieved for the first CTC-based assay system: CellSearch from Veridex.
With CellSearch, however, the analysis is purely quantitative. The number of CTCs found in patient blood offers oncologists insights on potential prognosis and the likelihood of progression, as well as helping to monitor the return of disease.
In August 2013, Janssen Diagnostics announced the findings of a study conducted in China of the utility of CellSearch CTC enumeration in metastatic breast cancer patients. They determined that patients with fewer than five CTCs per 7.5 mL of whole peripheral blood at first follow-up had significantly longer median progression-free survival and overall survival versus patients with CTC counts of five or more.
Despite these outcome trends, the test tells clinicians little about the disease itself, such as the mutational baseline or changes in the prevalence of resistance markers. To determine these factors—important to clinical decision makers—CTCs have to be isolated for further analysis.
According to Epic Sciences CEO Murali Prahalad, CTC isolation methods have traditionally fallen into one of two camps.
“One said that all cancer cells must have certain epithelial proteins on their surface and you could use that to isolate them by immunomagnetic beads,” he suggests. This is essentially the way CellSearch functions, as well as Fluxion’s IsoFlux System.
The second perspective, he continues, suggested that all cancer cells are larger than the surrounding white blood cells, and so “you could formulate microfluidic approaches either to enrich for the CTC or to deplete the white blood cells.”
This is effectively the idea behind platforms such as Angle plc’s Parsortix and ApoCell’s ApoStream, which also incorporates a dielectrophoretic component to the microfluidics.
But as Prahalad explains, cancer has proven to be an incredibly heterogeneous condition, involving cells of all morphological and molecular profiles. Thus, methods that select for one characteristic likely come with the cost of missing some facets of the disease.
With this in mind, Epic researchers developed the No Cell Left Behind concept, where they first lyse the red blood cells in a sample and pellet the remaining cells. They then transfer the cell pellet to a series of slides for further analysis or storage in a biorepository.
“Each slide has about three million nucleated cells on it,” Prahalad says. “And we’ve designed a high-speed scanning system that can look at all three million of those cells across all fluorescent channels in less than 15 minutes.”
They then rely on imaging software to identify the cells, whether they be white blood cells or cancer cells, and assign each cell on the slide an X-Y coordinate. Prahalad uses the analogy of Google Earth.
The cells are scanned for a variety of parameters, most tissue morphology-based. And when you overlay immunofluorescent signals, he suggests, “The cancer cells stick out like a sore thumb.”
Not only does this method allow the company to see CTCs that they never knew existed before, Prahalad continues, but the presence of white blood cells provides insights into the patient’s immune status.
“As we enter this new world of immuno-oncology, we can not only interrogate the cancer cells that are present, we can see markers like PD-L1 on white blood cells, as well,” he enthuses. “So we can get an interesting view about not just the cancer, but also how the immune system may be reacting to the different cancers.
“I think it’s an ironic twist that the very cell that everyone thought didn’t matter, the cell that you wanted to get rid of, we actually preserve and now we’re in a position to start to analyze in a very different way.”
A key advantage of looking at whole, individual cells is that it allows researchers to not only appreciate the cellular heterogeneity of a patient but to actually use that information in a clinically meaningful way.
“We see cancer not as a disease of averages but a disease of multiple clonal species that exist in a patient, each of which has a unique genomic and proteomic profile,” he offers, but adds that this is only the first step.
“It’s not only the presence or absence of the mutation you’re targeting, but the presence or absence of other resisting clones that may actually make the patient refractory to what you’re doing,” he presses. “In a clinical trial context, I have to have a grasp of the cellular diversity; otherwise, I don’t really have a sense of what’s happening in my patient.”
With some collaborations, he continues, they were able to see changes in the number and composition of the cells as quickly as one week after initial administration of a drug. And you can then chart the evolution and changes in the cell population over time, especially in response to therapeutic selective pressure.
While acknowledging the slow transition in the CTC field from mere enumeration to analysis, Trovagene CEO Antonius Schuh is less convinced of their utility beyond being biologically interesting.
“These CTCs are cells that, for whatever biological reason, have decided to leave the tumor, so they are behaviorally not representative of those cells that you’re concerned about,” he says, feeling that the speculation on their role in metastasis is oversimplified. “If you now give a patient a drug, are you expecting that the number of CTCs should go up or down? There is really no basis for that.”
“And the second significant challenge for CTCs is that they simply will not give you information about quantitative tumor dynamics,” he presses.
For these reasons, he and many others have turned their attention from cells to cfDNA and ctDNA.
Back in circulation
First noted in the late 1940s, cfDNA arises when cells of any stripe die, whether through apoptotic or necrotic processes, and slough their contents into the surrounding biofluids. In many cases, this is blood, but in the case of brain cells, it might also be cerebrospinal fluid.
And because tumor cells replicate rapidly as the tumor matures, more and more cells die with each passing generation, releasing greater quantities of ctDNA into circulation.
When compared to CTCs, says Guardant Health CEO Helmy Eltoukhy, cfDNA has a distinct advantage from a sheer numbers perspective.
He points to a 2013 study published in the New England Journal of Medicine that quantified how many copies of specific mutations occurred in blood samples taken from 30 breast cancer patients. When the blood samples were divided into cellular and cell-free fractions and analyzed with digital PCR and sequencing, the researchers found that the mutational signal was more than 100 times stronger in the ctDNA than in the CTCs.
“It’s a much richer source of that information, and that richer signal allows you to do things like capture the heterogeneity that’s there in the individual or potentially get to earlier stages of cancer, Stage II and Stage I,” Eltoukhy says.
As a sign of how quickly this research is translating into the clinical setting, QIAGEN announced in January receipt of European registration for its therascreen EGFR RGQ Plasma PCR kit for use as a companion diagnostic with AstraZeneca’s Iressa in the treatment of non-small cell lung cancer (NSCLC).
Similarly, in June, both Personal Genome Diagnostics and NeoGenomics announced the launch of cfDNA-based liquid biopsy services. In the case of PGD, the LungSelect platform identifies somatic sequence mutations and translocations in NSCLC patients that can be treated with drugs that are FDA-approved or in clinical trial. The NeoLAB assays from NeoGenomics, meanwhile, cover 12 different mutational profiles in different hematological disease.
Also working in NSCLC and in the same month, Guardant Health announced a partnership with the National Cancer Institute to apply its Guardant360 cfDNA testing platform for genomic profiling of 600 lung cancer patients participating in SWOG-1403. The goal is to test patients upon enrollment and upon disease progression, and then use that information to adjust patient therapy.
But even with this early excitement and richer signal compared to CTCs, there are still significant technical challenges for ctDNA-based liquid biopsy, says Trovagene’s Schuh. His top two are signal dilution and analyte size.
“ctDNA in the patient is highly diluted, and you may not be able to observe a signal simply because it’s not present in your limited sample,” he opens, suggesting that while the blood of a late-stage cancer patient might be flooded with ctDNA, patients with earlier-stage cancers or who have received treatment may simply not present a lot of signal.
“If I have a Stage III colon cancer patient with a robust KRAS signal and then I perform surgery on that patient, there’s a 50-percent chance that this patient will transition from Stage III to Stage 0,” he explains. “If I now want to confirm that there is no KRAS signal left in that person, then I will not even remotely achieve the necessary sensitivity from a blood sample, because there is just nothing in there anymore.”
Thus, rather than looking exclusively at blood samples, Trovagene has decided to focus much of its resources on isolating and identifying ctDNA signals in urine, which clinicians can collect in much larger volumes and more frequently.
But regardless of their starting point, the second challenge remains that ctDNA and cfDNA is highly fragmented—on the order of 150 base pairs—which presents a problem when you’re trying to identify mutations using next-generation sequencing (NGS), according to Schuh.
“The system noise for next-gen sequencers is relatively high, the floor is about 1 percent,” he says. “But we need to be able to look at 0.001 percent because the signals are so diluted. So we need to enrich those signals by a factor of 100 to 1,000 so that we can reliably see it with NGS.”
“The way that we do this at Trovagene is we make extremely small primers, primers that are so short that by themselves they do not uniquely map to the genome,” he continues. “These primers have a synthetic tail, and they anneal in the immediate proximity of the mutation you are interested in.”
By aggressively optimizing the enrichment assay, Trovagene increases the likelihood of a mutation-bearing fragment being amplified and therefore detected by NGS.
“Within days to weeks with very high reliability, we can determine whether a patient is benefiting from treatment by observing on the one side, the initially highly accelerated cell death, which is measured by spikes of mutational signal, and the subsequent depletion of that cellular reservoir, which is measured by a sustained vastly reduced mutational signal,” Schuh enthuses, quoting examples in NSCLC, pancreatic cancer and histocytic disease.
“In near real-time—earlier than with a biopsy, earlier than imaging—you can observe emerging resistance,” he says. “You can plan for a different therapeutic agent.”
“Then, rather than waiting many weeks or months before you know whether your treatment choice was a good idea or not, you can accelerate that feedback dramatically,” he continues. “That’s why we call this Precision Cancer Monitoring.”
Pressing the point further, Schuh sees ctDNA as a reflection of disease dynamics, a sign that something is actually happening.
“When we look at oncogene mutations in tissue, we are looking at something in a static manner,” he explains. “If I obtain this tissue from a late-stage cancer patient, it’s reasonably okay to assume that an oncogene mutation observed in the cancer tissue is involved in disease progression and development.”
“But I can also observe a BRAF mutation in a mole, but that mole will probably never turn into skin cancer,” he postulates. “Or I can even be born with a BRAF mutation—there are people who have that as germline mutations—and my risk to develop cancer that is associated with that mutation is actually only moderate.”
“In contrast, if you observe that mutation in ctDNA, you have one very critical additional piece of information,” he continues. “You are observing this mutation and it is originating from cells that are undergoing accelerated turnover. Otherwise, you wouldn’t even see it.
“The mere fact that I see a KRAS mutation in a person’s ctDNA tells me not only that it is there, it also tells me that it is doing something, because the cells that released that signal into circulation grow and die at an accelerated rate.”
Although Prahalad is quick to note that companies like Trovagene have done a lot to amp up analytical specificity and sensitivity, he raises a cautionary flag about the averaging out of a mutational signal.
Once a cell dies and releases its contents into the body’s circulation, you may pick up individual mutations, but you can’t reconstruct the cell of origin, he warns, and that is vital to understanding what is happening in these patients.
“There is a big difference biologically if you have one mutation in five cells or five mutations in one cell,” he explains. “They will radically change the biology and likely change the response to specific therapies.”
Dying cells aren’t the only source of circulating DNA, however, as live cells release small spheres of biomolecules into circulation, encapsulated in a lipid bilayer known as microvesicles. One form of microvesicle is the exosome.
Message in a bottle
Thought to be part of a natural signalling mechanism between living cells, exosomes are small bubbles of cellular cytoplasm wrapped in protective lipid coats. Once thought to be essentially garbage bags for cells, there is growing evidence that these small particles—often less than 100 nm in diameter—are a way for cells to securely transfer biomolecules such as DNA, RNA and proteins across large distances within the body.
Like CTCs, exosomes provide a window on active physiological processes within the body, but tend to occur in vastly larger quantities than the circulating cells. Thus, there is speculation that these structures may more accurately reflect what is happening with a cancerous tumor than ctDNA or CTCs.
A much more recent discovery than the other two analytes, early efforts to isolate exosomes focused on ultracentrifugation methods, but this technique requires specialized equipment and sample preparation. More recently, however, the aptly named Exosome Diagnostics has developed a spin-column approach that greatly accelerates exosome harvesting from plasma.
In September, as part of its partnership with QIAGEN, the company published a study in PLoS One that demonstrated its platform could yield high-quality exosomal RNA (exoRNA) of equal or higher quantity in a faster timeframe than traditional ultracentrifugation methods, and with more consistent results. The study is part of the collaborator’s efforts to commercialize, via QIAGEN, a series of research kits under the exoRNeasy and exoEasy banners to serve the biomarker discovery and liquid biopsy markets.
While one part of Exosome Diagnostics pursues the spin-column technology, however, another part is firmly pursuing the development of clinical assays to identify cancer-driving and resistance-building mutations in patient samples. And to do so, the company isn’t playing favorites with the analytes it chooses.
“When you’re talking about non-invasive genotyping—and this is one big application within the liquid biopsy space—you clearly wouldn’t want to exclude or ignore one big source of molecular signal,” says Vince O’Neill, chief medical officer for diagnostics. “If you combine captured cfDNA and exoRNA, you essentially get two to four times the yield than you would just by looking at cfDNA.”
“Why is that important?” he continues. “In the non-invasive genotyping space, you want to maximize sensitivity, and if you look at the mutation signal from both of these sources, you can up that sensitivity. It’s as simple as that, really.”
Thus, the company has a series of tests that characterize exoRNA plus cfDNA or exoRNA alone in blood or urine to look for mutations associated with prostate, lung and solid tumor cancers.
“There are two big applications here for our technology,” O’Neill presses. “Molecular response, where you follow the genetic markers post therapy. What you want to see is a collapse in the molecular signal. And then, obviously, molecular relapse, which as the data suggests will precede clinical or radiographic relapse by weeks to months.”
This lead time is critical to optimizing patient treatment, according to Exosome collaborator Keith Flaherty, director of the Henri and Belinda Termeer Center for Targeted Therapies at the Massachusetts General Hospital Cancer Center.
“In patients with an aggressive cancer, such as melanoma, insights about treatment response and disease progression are immensely time-critical in order to help guide and adjust treatment strategy,” he said in an announcement of melanoma data presented at the annual meeting of the American Society of Clinical Oncology in June.
As part of a longitudinal study that monitored BRAF-mutant melanoma, the company and its collaborators noted they could detect early disease progression several months before any clinical signs of progression appeared.
“We’re very encouraged by these data as they validate the utility of the combined capture of exoRNA and cfDNA to detect BRAF,” continued Flaherty. “Moreover, they demonstrate the ability of this plasma-based approach to detect disease progression much sooner than clinical or radiographic evidence, which would represent a potentially landmark advance in the diagnostic paradigm for melanoma and a host of other cancers.”
On the back end
Simply isolating the analyte is only the first step, however, and just as much development has gone into finding ways to identify rarer and rarer events, as well as to expanding for what signals within the analytes researchers can look.
“Ultimately, the choice of platforms and required detection limit will depend on the clinical sample being analyzed, as the most sensitive methods are reported to detect allelic frequencies of as little as 0.01 percent, providing a theoretical lower limit to detect one mutated copy in a background of 10,000 wild-type alleles,” wrote Graham Brock and colleagues at Exosome Diagnostics in a recent review in Translational Cancer Research. “Thus, this level of sensitivity requires samples/patients where at least 10,000 target alleles enter the downstream analytical assay.”
As one of the key steps in identifying the needle in the haystack is enlarging the needle, PCR-based methods of signal amplification have seen significant improvements over the last few years.
In June, for example, Transgenomic announced its plans to launch up to six new cancer tests within the year based on its Multiplexed ICE COLD-PCR (MX-ICP) platform. Originally licensing the platform from the Dana-Farber Cancer Institute, the company suggests the system can amplify mutant signal more than 500-fold over background signal, offering detection rates down to 0.01 percent.
In August, the company announced the launch of a pilot project to validate MX-ICP to guide and monitor live clinical trials being run by several unnamed pharma companies.
“As we have already shown in our own studies, we expect to validate for our partners that MX-ICP liquid biopsies are comparable or superior to DNA analyses obtained from conventional FFPE tissue samples,” said Transgenomic President and CEO Paul Kinnon in announcing the project. “We believe liquid biopsies will ultimately improve a clinician’s ability to diagnose and treat cancer, and they should also help to improve patient stratification in clinical trials and accelerate FDA submissions, providing a near-term advantage to our collaborators and customers.”
For Bio-Rad, however, another way to find the needle in the haystack is simply to make the haystack smaller.
Rather than try to spot the lone needle in all that hay, where it might be but one of a billion or more items, the company’s droplet digital PCR (ddPCR) platform starts by breaking the haystack up into significantly smaller stacks. Thus, while most mini-stacks will only contain hay, a few will contain needles, and those needles will now be a much higher starting percentage of the overall mass of the mini-stacks. The needles within the mini-stacks are then amplified, and the resulting product is a much higher signal in a significantly reduced noise.
“I would say that where people are using ddPCR extensively is in driver mutation and rare mutation detection, because you can get such a nice robust reliable measurement over what they could see prior using perhaps real-time PCR,” says Paula Stonemetz, director of Diagnostics Business Development at Bio-Rad.
Stonemetz describes the recent work of collaborator David Polsky of New York University in melanoma. In a retrospective ddPCR analysis of blood samples from former melanoma patients, Polsky was able to identify changes in the appearance of BRAF and NRAS mutations anywhere from four to 16 months before any changes were detected clinically. Without the ability to break the samples into smaller aliquots and enrich the signals, the very slight differences in DNA sequence would never have been evident, Stonemetz suggests.
The technology is not only being used to identify cancer-related mutations, however, but also changes in methylation patterns within the analytes. And she says that there has also been a growing literature of its use in the identification and analysis of microRNAs, very short molecules (approximately 22 nucleotides) that have significant impact on gene expression at the post-transcriptional level.
The other predominant method for mutation identification is NGS, and Guardant Health is not to be outdone when it comes to the digital sphere. Taking several cues from the digital communications world, they developed a platform they call digital sequencing that they believe addresses three significant issues with NGS.
The first problem was search space, which Eltoukhy explains by continuing the haystack analogy.
“If you only look in 5 percent of the haystack, you’re probably not going to find the needle,” he says. “It turns out that traditional NGS technologies, off-the-shelf kits, only convert about 3 to 5 percent of the DNA that you start out with into something that makes its way into the sequencer.”
Thus, the first step they undertook was to optimize sample processing to ramp that up to about 90 percent, which he suggests translates directly as increased sensitivity.
The second issue was specificity. As he explains, although most NGS offer 99.9-percent sequencing accuracy, if you’re looking at a 10,000-base gene like BRCA1, that translates into 10 false positives per sequencing experiment. This is where they leveraged the digital communications angle.
“Just like DSL is able to send bits 1,000-times faster over the same copper phone lines than dial-up modem, you can trade speed for error rate,” he says. Thus, rather than getting 1,000-times faster service, you get 1,000-fold more accurate service. This they were able to accomplish by precoding the DNA and recovering all of the molecules to correct their sequences afterward.
“And then finally, the third piece we had to get right is to see every type of genomic alteration in a patient sample,” Eltoukhy says, including single-nucleotide variants, copy number variants, gene fusions and insertion-deletions. “Each class is important because there are drugs associated with each one that will help patients quite dramatically.
“The cost of sequencing is not quite economical yet, but as that continues to come down, we can imagine doing whole-exome at some point.”
Digital PCR or NGS, the key for companies like Exosome Diagnostics is to remain open and flexible to downstream partners and platforms.
“Essentially, we’re downstream agnostic,” offers O’Neill. “What we do care about is that those analytics are optimized for our platform. And again, we’ve done a lot of work there using things like NGS, digital PCR, standard qPCR, etc.”
The non-invasion is on
Regardless of the analyte or the analytical platform, a degree of humility and open-mindedness is required going forward, as no single system will address all scenarios. As Epic Sciences’ Prahalad reminds us: “None of us can say yet that we understand the full range of mutations that cells can acquire under therapeutic selective pressure that allow them to remain viable and drive progression.”