Onc/acc: The revolutionary potential of ctDNA in oncology
ctDNA tests are becoming more sensitive and more information-rich. Are they the next great biomedical measurement technology?
PSA - If you’re a bioinformatician, translational scientist, or engineer with an interest in frontier diagnostics and genomics, drop me a line.
Biomedical revolutions often happen slowly: study by study, use case by use case.
The scientific enterprise is designed to be skeptical of grandiosity. Likewise, clinical medicine operates under the “first, do no harm” principle. Finally, revolutionizing biology is just hard: evolution has had billions of years to generate complexity, and we’ve had less than 100 years to modify it with modern molecular techniques1.
But, while slow, revolutions do happen. I’ve come to believe that we are at the beginning of just such a moment in oncology, brought about by circulating tumor DNA (“ctDNA”) tests. This revolution is already well underway, as evidenced by Natera’s meteoric rise. But we’re still in the very early innings of ctDNA’s impact on both drug development and clinical practice.
Here, I explore these impacts, and briefly assess the investment implications.
MRD: The crown jewel of ctDNA use cases
ctDNA tests look for trace amounts of cancer DNA in the blood. There are several use cases for this capability, which range from firmly established to speculative.
The most well-established use case is the point-in-time “liquid biopsy” - tests for advanced cancer patients that indicate what genetic mutations their cancer has. Liquid biopsy is similar to genetic testing on tissue samples, but since it requires just a blood draw rather than a surgical procedure, it’s more convenient. Guardant trailblazed this market with their Guardant360 test, but many companies now offer competing products, such as Foundation Medicine, Tempus, Caris, and others.
Other ctDNA use cases are nascent, like early cancer detection (which we discussed in a previous post). This involves looking for cancer in asymptomatic, healthy people as a screening test. GRAIL has the most comprehensive evidence base to date, but many diagnostics companies are racing at the opportunity, with none clearly in the lead.
But I expect the major driver of the ctDNA revolution to be the “minimal residual disease” (MRD) use case. MRD tests - such as Natera’s market-leading Signatera product - are longitudinally used in patients who have already had cancer, to determine whether they have trace amounts of cancer remaining in their bodies.
There are two reasons why MRD testing will be critical for the future of oncology.
First, they offer a more sensitive method of detecting cancer recurrence compared to imaging (CT and MRI) - they detect smaller amounts of residual cancer, and detect it earlier. And their ability to do so is getting even better.
Second, MRD tests are becoming more informative for guiding treatment. At present, these tests read-out a binary signal of whether tumor DNA is present, and if so, what the percentage of tumor DNA in the blood is (the “tumor fraction”). In the future, MRD tests will reveal the information content of recurrent cancer - specifically, what genetic changes are driving the recurrence.
As a result, I expect ctDNA tests to both accelerate clinical trials and dramatically change clinical practice.
Accelerating clinical trials
One of the foremost challenges in oncology is how long it takes promising new drugs to progress through clinical trials and reach patients in need.
For example, Moderna’s neoantigen vaccine for melanoma began its Phase II trial in 2019, and reported initial positive results in 2022. The Phase III trial began in 2023, and is set to read out in 2027, at earliest.
That’s an 8 year journey from Phase II initiation to potential approval, to say nothing of preclinical development and Phase I.
There are several reasons for this that span the clinical trial lifecycle2. But one critical determinant of trial length is the endpoint - the way we measure whether drugs work.
The simplest and strictest endpoint is - perhaps intuitively - whether the drug makes you survive longer. This referred to as overall survival endpoint, or “OS” , in clinical trial parlance.
OS is a great endpoint, because it captures both efficacy and safety. If a drug has efficacy but major safety issues that could cause death, OS penalizes it accordingly.
But OS takes a long time to prove - often the better part of a decade. And perversely, as the standard of care improves, it gets harder to prove that new drugs offer survival benefits: improving survival from 20% to 30% requires less participants and less time than improving survival from 70% to 80%. Progress inhibits future progress.
As a result, we’ve developed proxy endpoints that correlate to overall survival, but manifest faster. These include “progression-free survival” (PFS), “disease-free survival” (DFS), and “objective response rate” (ORR). Most modern cancer trials use one of these as their primary endpoint.
However, these endpoints still take years to mature.
New measurement technology needed
To run faster trials, we need endpoints that enable earlier efficacy assessment, but are still predictive of overall survival.
How do we get better endpoints? Better measurement technology is the most critical enabler.
Measurement technology undergirds all endpoints. Most of the proxy endpoints cited above use radiographic imaging. Radiologists interpret CT scans or MRIs to evaluate whether tumors are growing or not.
Fundamentally, endpoint quality is constrained by measurement technology quality.
This is why I’m so excited about “ctDNA clearance” as an endpoint - whether patients flip from detectable to undetectable ctDNA. In most cases, it answers the same biologic question (is there any cancer left in the body, and is it increasing or declining) earlier and better than imaging. Many studies and meta-analyses have found that ctDNA tests detect cancer recurrence between 2-12 months earlier than recurrence on imaging.
As such, it will yield faster trials, which will in turn accelerate drug development.
But there’s another, more subtle and more exciting point, and it involves the trajectory of ctDNA tests versus the trajectory of imaging.
I expect ctDNA tests to continue to improve meaningfully, while I don’t necessarily expect the same from imaging. This is because ctDNA tests are riding on an exponential curve: the declining costs of sequencing.
The sequencing exponential isn’t what it used to be, but even in its modern, modest form, it’s formidable3. In 2017, sequencing on Illumina’s flagship sequencing machine cost $5 per GB. In 2023, Illumina reduced the cost to $2 per GB. Ultima Genomics now offers $0.80-1.00 per GB. With an 80% price reduction over the last decade, today’s leading sequencers can do 5x more sequencing for the same price.
I’d bet that by 2030, sequencing will be below $0.50 per GB.
This is important because it enables ctDNA tests to continue to ratchet up precision without increasing costs.
Right now, the most common ctDNA tests like Signatera look for 16 tumor-specific DNA variants - mutations that exist in your cancer but don’t exist in other cells in your body.
Signatera’s limit of detection is about 0.01%, meaning it can reliably detect cancer so long as cancer DNA molecules make up 1 out of 10,000 total DNA molecules. This is an incredible technical achievement.
But, propelled by the sequencing exponential, new “ultra-sensitive” ctDNA assays (led by companies like Personalis) use a lot more sequencing, and look for up to 1,800 cancer-specific variants, rather than just 16. As a result, limit of detection is closer to 0.0003%, or 3 cancer DNA molecules per million total DNA molecules.
Studies have shown that ultra-sensitive tests are better at detecting trace amounts of cancer. For example, the TRACERx study this year showed that ultra-sensitive tests found 57% of Stage I lung cancers, compared with 14% found by a more limited, Signatera-like test.
In the future, I expect that we’ll go beyond Personalis’ 1,800 variants - perhaps tens of thousands of variants, and additional analytes (e.g., cell-free RNA), enabling an even lower limit of detection4.
As a result, ctDNA tests should continue to improve.
Putting ctDNA endpoints into practice
As a result of its better sensitivity and continuous improvement, ctDNA clearance could become an excellent proxy endpoint: highly predictive, but more importantly, fast to read out.
But for ctDNA to get widespread approval, it must correlate highly with overall survival. This is the gauntlet that new endpoints must go through in order to gain regulatory acceptance.
The question is still outstanding, but most studies I’ve seen suggest that ctDNA levels before or after treatment are predictive of survival, and ctDNA clearance correlates with better outcomes. The TRACERx study showed that lung cancer patients with no ctDNA pre-surgery had much better 5-year survival than patients with low ctDNA levels, who in turn had better survival than patients with high ctDNA levels. Likewise, the GALAXY study showed that colorectal cancer patients who cleared their ctDNA during post-surgical chemotherapy had improved survival compared to those who didn’t.
Indeed, in one specific type of cancer - multiple myeloma - regulators have approved MRD clearance as an accelerated approval primary endpoint. The FDA’s “ODAC” committee voted unanimously (12-0) last year to do so5.
However, at present, multiple myeloma is the only cancer where this is the case.
This is an area where I find myself frustrated with regulatory conservatism, and would love a more proactive approach.
While there are workshops studying ctDNA modeling in solid tumors, we’re not exactly sprinting at the opportunity. My baseline expectation is that after 5 years of studies and panels and focus groups, we’ll decide that ctDNA should be an endpoint across most cancer types.
In an age where we’re trying to cut regulatory red tape to accelerate clinical trials and compete with China, are the 5 years really needed?
The revolution in clinical practice
ctDNA’s impact on clinical practice could be even more pronounced than its effects on drug development.
Right now, leading ctDNA tests tell you if there is any tumor DNA detected, and if so, what the “tumor fraction” is (e.g., 0.1% of all DNA molecules are derived from cancer.)
You’re not getting information content from the DNA itself, which is unfortunate, because the information is right there.
Cancer is an adaptive, evolutionary disease. When we try to eliminate it with drugs, we often select for cancer cells with new genetic abnormalities that are resistant to those drugs.
But what if we had a diagnostic “Eye of Sauron” that was watching for every possible resistance mutation, enabling physicians to snuff them out early?
ctDNA is becoming just such an all-seeing eye.
This year’s ASCO conference provided a tantalizing glimpse of the future, via a trial called SERENA-6.
In SERENA-6, investigators followed breast cancer patients after surgery, performing ctDNA tests every 2-3 months. Critically, not only did they look for ctDNA levels, but they looked for the presence of actual mutations - specifically, the ESR1 mutation. ESR1 mutations indicate that breast cancer is escaping from standard treatments.
When ESR1 mutations emerged, investigators immediately put the patient on a new drug that targets ESR1.
The study was successful - there was significantly higher progression-free survival in the group that got the drug immediately upon ESR1 mutation emergence, versus those that didn’t.
SERENA-6 is the first step in a much larger and more exciting direction: ctDNA-guided adaptive cancer treatment6.
Here’s the paradigm that I see emerging:
After initial therapy (e.g., surgery, or first-line treatment), cancer patients get ultra-sensitive ctDNA tests every month or two7.
These tests not only indicate the presence or absence of cancer, but also report on the specific mutations.
As a result, we can understand new drivers of cancer resistance and recurrence at their earliest stages
Armed with this information, oncologists adaptively change therapies to kill emerging cells before they’re too numerous.
I also expect cell-free RNA and epigenetic profiling to gain importance over the coming years, since they offer complementary information around expression-based resistance mechanisms that aren’t readily visible in DNA itself.
All of this may take a long time to become routine in clinical practice. But the direction of travel is clear.
Investment implications
If this is the future for oncology, what are the right companies to invest in? This is a difficult and fascinating question.
Natera is the most obvious. Signatera owns the MRD market, with 90-95% market share and the most reimbursed indications. But, investors realize this, and Natera is expensive, trading at 17.1x consensus 2025 gross profit.
There’s another reason to be somewhat guarded. Currently, Signatera carries 65-70% gross margins, due to strong ASPs and minimal COGS.
But the market seems to be moving towards ultra-sensitive tests. Natera itself recently launched “Signatera Genome”, which looks for 64 variants rather than 16. Personalis has shown excellent results with its 1,800 variant test.
More variants means more COGS: more library prep reagents, more PCR primers, more sequencing consumables.
If ASP doesn’t increase concomitantly, it could augur lower gross margins, which would put pressure on future cash flows and valuation.
Personalis should be considered. Personalis’ NeXT Personal MRD is the most sensitive MRD test currently on the market, and the one that I would want as a cancer patient.
Further, Personalis is still a tiny company. At $520M enterprise value, it’s 40x smaller than Natera. If you believe Personalis will take share in the MRD market (and do so profitably), it has a lot of room to run.
Therein lies the rub. At the moment, the Personalis test has high COGS and poor reimbursement. The company’s economics are grim: Personalis produced only $7M gross profit in the first quarter of 2025, at 35% gross margins and -85% EBIT margins.
Personalis is actually trading at a more expensive multiple than Natera, at 26.5x 2025 gross profit8.
While Personalis has upside, I can’t help but think they are in the position of Anakin Skywalker in the famous duel against Obi-Wan in Revenge of the Sith.
Anakin (Personalis) is more powerful, but Obi-Wan (Natera) has the high ground, and thus wins the duel. Natera is a commercial behemoth with sales reps in every oncologist’s office. All they have to do is improve Signatera just enough to dissuade patients and physicians from using Personalis. Signatera Genome is a great example of this strategy.
Is Tempus a play? They have a commercialization agreement with Personalis, and own a significant share of the company. Tempus stands to benefit if Personalis takes share. As a scaled company nearing breakeven, Tempus also carries less existential risk than Personalis. But, while Tempus is modestly cheaper than Natera (14.5x 2025 gross profit), it’s still a relatively high multiple. Tempus’ heavy AI marketing and web of complex deals has made it more “meme-y” than many investors are comfortable with. But it’s an impressive business, and it warrants a deeper look than I’ve given it.
Guardant is the most interesting, with a strategy that is sharply coming into focus. A year ago, Guardant seemed to be a laggard in MRD testing - their tumor-naive “Reveal” test just isn’t very performant9.
But, Guardant has re-established themselves as a major contender. And surprisingly, it doesn’t involve Reveal.
In the SERENA-6 trial (referenced above), investigators used Guardant360 every 3 months to identify the ESR1 mutation. This is provocative, because Guardant360 isn’t a classical MRD test. It has historically been marketed as a test used once, when a patient is first diagnosed with advanced cancer.
Is it possible that Guardant360 - or something like it - could become an important MRD test?
I believe so, because it’s the test with the most information content.
Guardant360 is designed to report on the mutation status of 740 critical cancer-related genes. Signatera, on the other hand, with its mere 16 variants, was originally designed to indicate whether cancer exists in the bloodstream, not its genetic drivers. Even Personalis, with its 1,800 variants, likely isn’t covering important cancer-associated genes as comprehensively as Guardant360.
Thus, Guardant360 takes a different approach, and in doing so, can longitudinally track the genetic drivers of cancer10.
Guardant still isn’t an obvious investment. On the positive side, Guardant360 is an excellent franchise, the MRD opportunity is becoming clearer, and Helmy and AmirAli have shown a consistent ability to skate where the puck is going. On the negative side, Guardant’s balance sheet is ugly, and they’re burning money hand over fist in pursuit of early detection. The stock has rallied 78% over the last year, and now trades at a 12.3x gross profit multiple.
Table: Key ctDNA diagnostics companies (as of July 5, 2025)
All of the companies referenced above are growing rapidly, with an enormous opportunity in front of them. They’re also all either unprofitable or just barely breakeven.
Given these unstable financial profiles and the shifting competitive dynamics, the ctDNA diagnostics space is perhaps the most lively of all healthcare investment segments today.
For more sober investors, there are other value chain enablers that should benefit from the rise of ctDNA, especially those that provide COGS components for the diagnostics themselves.
Illumina is the most obvious, since ctDNA tests require next-generation sequencing. Ultra-sensitive tests use much more sequencing, and Illumina thus stands to benefit. Ultra-sensitive tests also use many more polymerase chain reaction (PCR) primers compared to standard tests. This should benefit companies like Twist and Integrated DNA Technologies (IDT, a subsidiary of Danaher).
Conclusion
Cancer is an incredibly complex disease. Even the synthesized list of cancer’s “hallmark” characteristics has increased from 6 to 14 since the turn of the millennium.
Any genuine treatment of oncology must engage with the underlying genetic complexity of the disease.
Future ctDNA tests will do exactly this. They will incorporate more genetic information content, and new analytes like RNA that provide complementary information about gene expression. This will reveal the ever-shifting cellular mechanisms of recurrence and resistance. With such an “all-seeing eye”, we’ll be able to fight resistance in its infancy, before such cells have multiplied and strengthened.
ctDNA won’t be the final chapter in the battle against cancer - but as a foundational measurement technology, it will be a pivotal chapter indeed.
Compare the original notion of DNA —> RNA —> protein to the incredible complexity of how we now understand genes to be expressed and regulated.
One key reason is that modern cancer trials enroll patients with specific, genetically-defined tumors. Narrow enrollment criteria typically means slower enrollment.
My movie analogy for the state of the declining sequencing cost/GB curve: Wolverine at the end of Logan. Running on fumes, but alive for at least one last go.
Based on a quick scan, it looks like 0.5 parts per million is approaching the lowest theoretical sensitivity for ctDNA alone, based on how much blood could reasonably be drawn at a blood draw. Perhaps the addition of cfRNA and other analytes can decrease it further.
One clarification - ODAC voted to accept MRD clearance via bone marrow aspirate rather than blood draw.
To be clear, SERENA-6 was controversial for complicated statistical reasons. But this is above my pay grade.
Or maybe every 3-4 months, since payers are annoying.
That said, for a high growth, high potential micro-cap business, I weight TAM-to-market cap and balance sheet health more highly than multiples.
Tumor naive tests like Guardant Reveal look for minimal residual disease without knowing anything about a patient’s specific cancer mutations. They can thus be used on anyone, regardless of whether or not they ever had their cancer tissue sequenced. On the other hand, tumor informed tests like Signatera and Personalis NeXT Personal require upfront sequencing of the patient’s tumor to find the mutations that are specific to that patient. While tumor naive tests thus have faster turnaround time (no upfront sequencing), they often have worse sensitivity, because they know less about the patient’s cancer.
To be clear, Guardant360 probably does have a worse limit of detection than Signatera and Personalis. But, information content is a key advantage.