290 ‒ Liquid biopsies for early cancer detection, the role of epigenetics in aging, and the more

KEYPOINTS

• 01

  Genome sequencing costs have fallen faster than Moore's Law.

  Alex Aravanis, Peter's medical school classmate, is an electrical engineer turned MD-PhD. He later became Illumina's chief technology officer and co-founded GRAIL. Sequencing one more genome right after the Human Genome Project would have cost a few hundred million dollars. By about 2012 it was in the low tens of thousands, and today it's a couple hundred dollars. The leap came from sequencing-by-synthesis chemistry developed by Cambridge chemist Shankar Balasubramanian and brought to Illumina through its acquisition of Solexa. Billions of DNA fragments are attached to a glass slide and read one base at a time, all in parallel.

• 02

  Tumors constantly shed DNA into the blood, which makes liquid biopsies possible.

  Cell-free DNA was first seen in blood in the 1940s, and in the 1970s researchers noticed cancer patients had much more of it. Tumors grow chaotically, so their cells die at high rates and release DNA. Fragments are about 160 base pairs long because DNA wound twice around a nucleosome is protected from enzymes. Alex describes three uses: choosing targeted therapy without risky biopsies (lung biopsies can cause a collapsed lung), monitoring for residual disease after treatment, and screening people with no known cancer.

• 03

  Prenatal tests accidentally proved blood DNA could reveal hidden cancers.

  Dennis Lo, and Stanford's Steve Quake, developed blood tests that detect fetal trisomies from placental DNA. Alex says these cut amniocentesis by about 80% in the US. At Verinata, an Illumina company, Meredith Halks-Miller and Daria Chudova noticed rare results with chaotic abnormalities across many chromosomes, which isn't compatible with a viable fetus. The mothers turned out to have undiagnosed invasive cancers. All of the roughly 10 women flagged, described in a 2015 JAMA paper, had cancer. That was no false positives, though most cancers in that population were surely missed.

• 04

  The key challenge for cancer screening is specificity, not sensitivity.

  Alex notes that in 50 years nobody had built a broad blood test for cancer, and PSA's high false-positive rate cost it its USPSTF A or B grade. Proteins and metabolites rise in benign conditions too, giving false-positive rates of 5–10%. Population screening needs about 1% or less. GRAIL's CCGA study took nearly three years and hundreds of millions of dollars, using about 15,000 samples split into blinded training and test sets. It compared mutations, chromosomal changes, cell-free RNA and DNA methylation.

• 05

  Methylation, the 'dark horse,' won decisively.

  Methylation adds a methyl group to cytosines at about 28 million CpG sites and controls which genes a cell uses. Alex calls it 'the software of the genome.' It's read by chemically converting unmethylated cytosines (bisulfite sequencing). Methylation detected the most cancers, reached a false-positive rate under 1%, and uniquely predicted tissue of origin. Adding other signals didn't help, because they weren't independent. His analogy: once you spot Ford's blue oval, also looking for 'F-150' adds nothing. Two tubes (about 10 mL) of blood suffice, since background biology, not blood volume, limits screening sensitivity.

• 06

  In PATHFINDER, about 40% of positive Galleri results turned out to be cancer.

  In this interventional study of about 6,500 people at average or slightly elevated risk, with cancer prevalence just under 1%, about 40% of 'cancer signal detected' results were confirmed as cancer. By comparison, about 4–5% of abnormal mammograms lead to a cancer diagnosis. Specificity was above 99%, and the test caught about half of the cancers that emerged. Added to guideline screening, it roughly doubled the number of cancers found, mostly types with no existing screen. Alex notes that about 70% of cancer deaths come from unscreened cancers. The test predicts a top-two site of origin about 90% of the time, using a methylation atlas of healthy cell types built with Yuval Dor in Jerusalem.

• 07

  It's a functional test: it preferentially finds the dangerous cancers.

  Sensitivity varies widely by type. Peter's review found roughly 20% detection of early-stage hormone-positive breast cancer versus about 75% for triple-negative. To shed detectable DNA, a tumor needs enough mass, cell turnover and blood access, traits linked to aggressiveness. In CCGA survival curves, people whose cancers the test missed did very well, while detected cancers carried substantial mortality. Low sensitivity for indolent prostate or hormone-positive breast cancer may therefore partly avoid overdiagnosis.

• 08

  Does earlier detection save lives? The two make their case, and NHS data are coming.

  Peter states his bias that it does, noting that breast and colorectal cancers treated with the same adjuvant therapy fare far better when localized. Alex cites five-year survival of 70–80% for localized solid tumors versus under 20% for stage 4, and asks what surgeon would decline to treat a stage 1 pancreatic cancer found this way. The NHS is testing Galleri in about 140,000 people, using a reduction in stage 4 diagnoses as an earlier endpoint than mortality. Peter adds that single-cancer trials are rarely powered to show changes in all-cause mortality.