Well-funded startups like Freenome and Grail say they’re making
tests that spot cancer before you know you have it. They have
divulged little about the science behind their vision — and
many scientists are skeptical.
Originally posted on April 12, 2017, 10:31 GMT
Updated on April 12, 2017, 12:51 GMT
BuzzFeed News; Getty Images (2)
Silicon Valley startups are racing to develop a blood test
for cancer that many scientists believe is years, if not
It’s a high-stakes competition, fueled by hundreds of
millions of dollars in venture capital. The winner could help
millions of patients fight off cancer before the disease
shows any outward symptoms — early enough to drastically
improve their odds of survival.
Or at least that’s what these companies envision on their
websites and pitch decks, and at scientific meetings — and
investors are buying it.
Founded in 2014, Freenome raised
$65 million last month from Andreessen Horowitz, Peter
Thiel’s Founders Fund, and Alphabet’s venture arm GV. One of
Freenome’s most prominent rivals, Grail, which spun out of
the DNA-sequencing monolith Illumina in 2016, has raised an
eye-popping $1 billion from the likes of Bill Gates and Jeff
Bezos. Another competitor, Guardant Health, has amassed close
to $200 million.
So far, however, these companies have shared scant few
details about how, exactly, they’re creating a test that
could fundamentally change how we deal with cancer.
Freenome’s CEO and cofounder, Gabriel
Fast Company in April 2016 that its test would hit
the market within nine months, after being published in a
scientific journal. In June, he wrote a blog post claiming
that those results would appear “very
soon.” That hasn’t happened. Otte told BuzzFeed News his
staff is still setting up clinical trials and will publish
results “when we’re ready to publish.” (Otte recently
admitted to BuzzFeed News that he does not have a PhD,
despite multiple references to the contrary in the press,
company materials, and scientific conferences.)
Grail and Guardant have not published any findings, either,
although their executives also say they intend to.
Academic cancer researchers, meanwhile, say producing this
kind of test is an incredibly difficult scientific and
logistical challenge. Nascent tumors sometimes shed telltale
markers in the blood, but often don’t, and these “biomarkers”
can be different from one person to another, or even in one
person from one month to the next. Plus, credible data will
take at least several more years to accumulate in rigorous
clinical trials, if not longer.
“It’s not going to be a Star Trek, ‘let’s take a quick
sample and tell you exactly what disease you have and how to
treat it,’” Jeremy Jones, an assistant professor of cancer
biology at the City of Hope in Los Angeles, told BuzzFeed
There’s no question that such a test would be revolutionary.
There’s no question that such a test would be revolutionary,
giving its inventors enormous social and financial rewards.
“Early detection is incredibly attractive, because if you can
detect cancers early, you can cure them at a much higher
frequency,” Tony Blau, a hematologist who directs the Center
for Cancer Innovation at the University of Washington, told
Just as investors in Theranos, the $9 billion startup now
fighting for its life after very public regulatory missteps,
envisioned a world where doctors could test for all kinds of
conditions from a few drops of blood, Silicon Valley is
embracing the vision of cancer screening for everyone, early
Blood tests for early-stage cancer would be drastically
better than current diagnostic methods like “tissue
biopsies,” in which doctors extract potentially cancerous
tissue with needles and surgeries. Tests on a couple
teaspoons of blood could be much less expensive and invasive,
and performed more often. A highly accurate test would also
have an advantage over today’s non-invasive tests.
Mammograms, for instance, have high rates of both false
negatives (they miss one in five breast cancers) and false
positives (which happen to about half of women who get the
test annually over a decade).
Scientists have long known that cancer cells routinely shed
bits of DNA into the bloodstream. But in the past few years,
thanks to advances in DNA sequencing, these bits have become
far easier to detect.
There are already a handful of tests, led by Guardant, that
use these DNA bits to detect cancer in a cancer patient’s
blood, or to identify certain mutations that might be
treatable with personalized therapies.
But so far, there is no reliably accurate commercial test
that can do this for people who are early in the disease and
have not yet been diagnosed — the lofty goal of Freenome and
The DNA shed by early-stage tumors accounts for less than
one-tenth of a percent of all DNA in a patient’s blood, said
Ash Alizadeh, a Stanford University oncologist who helped
sell a technology to potentially help doctors monitor how
a tumor responds to therapy. For some tumors, such as those
that start in the brain, their DNA is virtually undetectable
even in tens of milliliters of blood, he added.
“The computations are so low,” Alizadeh said. “That’s been a
major challenge for early detection.”
Even if a test was able to detect every single molecule of
DNA in a blood sample, that wouldn’t be enough. Machines
often generate false-positive readings of clumps of cells
that are deceptively cancer-like, such as moles that don’t
turn into melanoma or colorectal polyps that don’t become
colon cancer, Alizadeh says.
“Do we know that those changes are always going to lead to a
cancer that could threaten a patient’s life?” said Blau of
the University of Washington. “The answer to that is no, we
don’t know that.”
Yet another hurdle is simply knowing what DNA sequences to
look for, since, as Blau points out, cancer cells differ
within the same patient, and even within the same tumor.
And even if a test appears to work in a lab, proving it works
in people will take years. “You have to test thousands of
patients, wait long enough that enough of them get cancer and
enough of them ultimately die of the disease, to be able to
really evaluate if a new test is a useful screening test,”
Max Diehn, an assistant professor of radiation oncology at
Stanford, told BuzzFeed News. (He also consults for Roche,
which owns the technology he co-developed with Alizadeh.)
David Silverman / Getty Images
One of Grail’s first steps is a
multi-center clinical trial in the United States with at
least 7,000 patients with untreated cancer, and 3,000
cancer-free people. It began in August 2016 and plans to wrap
up by August 2022.
Scientists hope to understand the molecular differences in
the two groups’ blood samples, as well as how they change as
disease develops. Grail has an advantage in its ties to
Illumina, the world’s dominant supplier of DNA-sequencing
expects Grail to become “one of Illumina’s largest
customers over time.”
“Our hope is over time that we’ll be marching the population
back earlier and earlier to where almost everyone who has
cancer is diagnosed early,” Grail Chief Business Officer Ken
Drazan told BuzzFeed News.
Guardant, on the other hand, believes that the knowledge it
has already gleaned from testing 35,000 cancer samples can
help pinpoint how the disease arises. President AmirAli
Talasaz says that in addition to studying tumor DNA
fragments, the company is looking at other kinds of chemical
changes in tumors, called “epigenomic” variations. Guardant
is running clinical trials of various sizes, including on
high-risk patients and cancer survivors; a spokeswoman said
there is no estimated date of completion.
Meanwhile, Freenome is testing its technology on patient
blood samples from UC San Francisco, UC San Diego,
Massachusetts General Hospital, and other clinics. For some
samples, Freenome tries to predict if the patients received a
cancer diagnosis or not. Other samples are from patients who
are still being tracked for cancer but haven’t yet been
diagnosed. In total, Freenome plans to test “thousands” of
samples, though Otte declined to say how many have been
tested to date. He still contends that these data will be
published in a scientific journal before tests are sold
The CEO reportedly hooked Freenome’s main investor,
Andreessen Horowitz, after acing a blinded test of five blood
samples provided by the venture capital firm (which also
invests in BuzzFeed). As Otte
wrote last year, the company correctly categorized two of
the samples as normal, and the other three as cancerous — and
even accurately labeled what stage of disease. Although two
of the cancer samples were from patients in late stages, he
wrote, the third was stage one, a sign that the test could
detect early cancer in a healthy-looking person.
At a conference in San Francisco in February, Otte shared
some striking numbers with an audience of scientists.
Freenome had an accuracy rate of more than 95% in detecting
the presence or absence of prostate cancer in 351 samples,
according to an
abstract of the presentation that Otte subsequently
confirmed was real.
Supposing that Freenome’s test is as sensitive and specific
as the rate claimed, “that would be a surprising result,”
Diehn, of Stanford, said.
Alizadeh said there doesn’t seem to be enough information
about the studies to know what to make of the technology’s
apparently high performance. “For any test, especially a
clinical one,” he said by email, “interpreting accuracy
requires knowing the error rate.”
In Diehn’s view, too, there isn’t enough information to
evaluate Freenome’s claims. “When you have a surprising
result, the onus is on the scientists to provide evidence —
strong evidence, supportive evidence — for such a claim.”
During his presentation, Otte also said
Freenome had a 97% average accuracy rate in detecting breast,
prostate, lung, and colorectal cancers across four stages of
progression, including stage one. The sample size wasn’t
“If they’re trying to say they can tell what stage a patient
has, based on a blood test, I would find that would be very
surprising,” Diehn said.
A stage is defined by how far a tumor has spread from where
it started, which usually requires imaging and pathological
tests, according to Diehn. Sometimes, he said, the only
difference between a stage-one and stage-two tumor is being
slightly bigger, or spreading to a single lymph node.
According to Jones, who attended the presentation, Otte said
that Freenome could tell when a case of prostate cancer was
aggressive or low-risk. There are “early indications” that
the technology can do this, Otte told BuzzFeed News, but it
is still in development.
Otte acknowledged that all of these results need to be
validated in larger clinical trials. “We’re focused on making
sure that we get the numbers we need to prove to the world
that our tests are safe and function well.”
“What they’re saying could be they’ve developed a great test,
or they don’t really have the data, and they’re trying to
make it sound like they have.”
Otte’s talk left Jones impressed, he said, but wondering,
“How do we know it’s real?” He understands why Freenome
wouldn’t want to reveal its technology’s nitty-gritty to
competitors before the test is on the market. But that
choice, he said, “makes it difficult in academic science to
know how valid it is.”
Diehn put it more bluntly: “What they’re saying could be
they’ve developed a great test, or they don’t really have the
data, and they’re trying to make it sound like they have.”
Freenome’s test picks up not only tumor DNA fragments, Otte
said, but DNA changes that signal “how the immune system is
responding to the presence or absence of the tumor.”
Freenome’s machine-learning platform deduced that a
stronger-than-expected link between these unspecified
immunological signatures and cancer, Otte said.
It makes sense that the immune system would respond to an
abnormal growth early on, Jones says. But “the immune system
obviously responds to things other than cancer all the time,”
he added. “What if a patient is on antibiotics, what if they
have an active viral or bacterial infection? Does that cloud
the ability to detect the cancer pattern?”
Unlike Guardant and Grail, Freenome does not have a clinical
or scientific advisory board. The company is in the process
of building them, Otte
wrote last month.
Vijay Pande, a general partner at Andreessen Horowitz and a
Freenome board member, said the company is rightfully
broadening its analysis beyond just tumor DNA. “There’s a
whole landscape, in principle, of what’s going on in your
body available in blood,” he told BuzzFeed News.
There aren’t existing papers that describe the basis of
Freenome’s approach, Pande acknowledged — but that’s because
its computational methods are so advanced, they’re
effectively creating new knowledge. “This is new territory,”
said the former Stanford computational biologist. “This could
not be done 20 years ago.”
Whether it can be done today remains to be seen.
Stephanie Lee is a senior technology reporter for BuzzFeed
News and is based in San Francisco.
Contact Stephanie M. Lee at email@example.com.
Got a confidential tip? Submit it here.