A gloved hand holds a blood sample up to the camera above a centrifuge

Liquid biopsies involve looking for fragments of tumour DNA in blood samples.Credit: Nicola Tree/Getty

C2i Genomics is a spin-off from the New York Genome Center, New York City, and one of the final eight for The Spinoff Prize 2021.

Asaf Zviran’s early career was far removed from the world of biomedical research. After training as an engineer and mathematician, he spent years developing decision-support software for the Israeli navy. But then, multiple members of his family developed cancer during the late 2000s — including Zviran himself.

The experience of grappling with consultations, treatments and complications led to a marked change in his aspirations. “I started looking for ways to integrate the systems I had developed in the defence sector into oncology,” says Zviran, who now has a PhD in molecular biology from the Weizmann Institute of Science in Rehovot, Israel.

Zviran asked himself: how can we develop a ‘radar system’ that can inform physicians in real time if a treatment is working? This idea led to the launch of C2i Genomics, a company built around an ultra-sensitive liquid-biopsy approach developed by Zviran (who is chief executive) and his postdoctoral mentor, medical oncologist Dan Landau at the New York Genome Center in New York City, who is on the start-up’s scientific advisory board.

A liquid biopsy involves analysing a blood sample for stray cancer cells or fragments of circulating tumour DNA (ctDNA) released from those cells. This is less invasive than a biopsy taken directly from the tumour, and can potentially offer a sensitive indicator of how much cancer tissue is present in a person’s body. Clinicians are especially enthusiastic about potentially detecting scant traces of cancer that remain after treatment, known as minimal residual disease (MRD).

Landau says that clinical studies have offered robust evidence that MRD detection can provide an early warning of recurrence of blood-based cancers such as leukaemia, informing clinicians that more-aggressive treatment might be warranted. A few companies have also developed assays for detecting blood signatures of MRD from certain solid tumours — targeted ctDNA sequencing is used to screen for panels of mutations that are present in only the tumour genome. Such analyses generally require deep-sequencing strategies that involve generating thousands of overlapping sequencing reads for each location of interest on a chromosome, known as a locus. This maximizes the sensitivity and accuracy of the detection of very rare ctDNA molecules against a background of more abundant healthy genomic DNA.

But this approach has a crucial limitation: the inherent scarcity of tumour DNA. A sequencing-based assay that targets a few dozen specific mutations will succeed only if the tiny population of ctDNA fragments present in a given blood sample also happens to contain those particular genomic features. This scarcity imposes a strict constraint on the sensitivity of deep-sequencing approaches. “That’s it, mathematically,” says Zviran. “It doesn’t matter what you do — even if you have perfect error suppression and the perfect detection method.”

To solve this problem, Landau and Zviran are taking advantage of the steadily falling cost of whole-genome sequencing to profile ctDNA in its entirety (see ‘Stacking the cancer-detection odds’), rather than just selected parts. The idea is to reach more targets by looking for thousands of tumour-related signatures rather than just dozens. “The coverage at each locus is going to be fairly small, but if you aggregate the signals statistically, then in principle you will have a much higher sensitivity,” says Landau. But this shift to a shallower but broader analytical approach has a drawback: it creates more opportunities for interference from sequencing errors. To overcome this, Zviran and Landau trained a machine-learning algorithm to distinguish normal and tumour-specific genomic data, greatly improving the accurate classification of individual ctDNA fragments.

Stacking the cancer-detection odds

Last year, Landau, Zviran and their colleagues applied their technique, which they term MRDetect, to serum samples from 60 people with cancer who had undergone surgery or immunotherapy1. The researchers could routinely detect extremely scarce ctDNA, even when tumour-derived sequences represented less than 1 in every 100,000 DNA molecules present in a serum sample; by contrast, detection using conventional deep sequencing began faltering at a ratio of 1 in 1,000. Crucially, Landau notes that this detection was achieved in samples collected soon after treatment, when clinicians must decide whether additional rounds of adjuvant therapy are needed. “This is a juncture where we believe there is still curative potential,” he says.

Casting a wider net

The key innovations of the MRDetect platform involve the software and the analytics. C2i Genomics was founded with the goal of developing the robust computational infrastructure required to support deployment of this ctDNA testing approach worldwide. “We want to allow any clinical lab in any location in the world to have access to our technology and quick turnaround without the logistical issues of sending samples to one centralized lab,” says Zviran.

The company has already partnered with laboratories in Denmark, Switzerland, Singapore and the United States to further validate the strategy’s accuracy and sensitivity. Each collaborator is trained so that everyone follows a standardized sample-collection and sequencing workflow, and the resulting data are uploaded to a cloud-based analytical platform. The company is headquartered in New York City, but also has a facility in Haifa, Israel, that maintains the computational infrastructure for the testing process. Because the company is handling patient data, its platform is compliant with the General Data Protection Regulation (GDPR) in Europe, and the Health Insurance Portability and Accountability Act (HIPAA) in the United States. Analyses are carried out on servers in the same region as the testing lab. This decentralized design enables fast turnaround, with labs receiving results from a one-millilitre serum sample in as little as one week.

The first-generation MRDetect platform was tested against three solid tumour types — colorectal, lung and skin — and designed to recognize multiple types of genetic aberrations, including single-nucleotide variants and abnormalities in gene copy number. It performed well in this initial demonstration, achieving up to 90% sensitivity for detecting some tumour types. Even so, the team revamped the underlying algorithm in an effort to improve its performance further. Zviran notes that the influx of data from clinical laboratories has been invaluable in improving the analytical process, and predicts that more-advanced machine learning will lead to improved sensitivity and specificity. The company is now looking at additional classes of mutation, and training its system to detect MRD for glioblastoma — a deadly form of brain cancer that typically releases only minuscule amounts of ctDNA.

Early commercial applications will focus on assessing people’s response to experimental drugs during clinical trials. But the start-up’s founders are also excited about the opportunity to help clinicians to optimize the safety and efficacy of cancer treatment. For example, in the case of an early-stage solid tumour with a 40% likelihood of recurrence, clinicians would typically recommend chemotherapy in addition to the main treatment. But that leaves 60% of people receiving a very toxic therapy that they arguably don’t need, Landau says. A sensitive assay for MRD could help clinicians to focus such treatments on people with confirmed trace amounts of disease, and merely monitor those who do not. MRD detection could also prove useful for neoadjuvant care — presurgical pharmacological treatment that can, in some cases, eradicate the disease without the need for surgery. A negative ctDNA test could spare some people the protracted recovery and reduced quality of life associated with surgery.

Michail Ignatiadis, a medical oncologist at the Jules Bordet Institute in Brussels, thinks that the start-up’s platform might represent an important step forward for ctDNA-based liquid biopsies. “Compared to other technologies that are out there, this seems to be more sensitive,” he says. “And conceptually, I like it — you’re taking advantage of both single-nucleotide variants and copy-number variants.” But he points out that it remains unclear how meaningfully the company’s test — or any liquid biopsy for MRD in solid tumours — can affect patient care in the real world. “These technologies need to be validated in 1,000 or 5,000 patients,” says Ignatiadis. “It will take a few years to enrol these patients and see whether an intervention based on this test can improve their outcome.”

C2i Genomic’s founders agree that this is a top priority, and it is conducting multiple clinical studies in Europe, Asia and the United States to address it. “We are moving to larger and larger prospective trials,” says Zviran. He adds that unpublished preliminary data are encouraging, with at least one study demonstrating improved sensitivity for the company’s test relative to a competing commercial ctDNA test.

From Landau’s perspective, the company’s technology has already made an impact simply by demonstrating a new approach to a difficult problem. He hopes that its work will spur greater interest in expanding and extending the use of whole-genome analysis as a routine tool for cancer profiling and treatment planning. “We want this to bring solid-tumour oncology to where diabetes management is today,” says Landau. By equipping clinicians with a clearer real-time view of tumour activity in the body, the company could one day make the quest to conquer cancer a little bit simpler and shorter for families like Zviran’s.