Cell-free DNA: from blood draw to error-corrected call

Cell-free DNA is a corpse report. Cells everywhere in the body die constantly, and their DNA enters the bloodstream as short fragments — about 140 to 170 base pairs when the source is programmed cell death, longer when the source is necrosis or a leukocyte lysing in the tube. Pregnancy adds a second voice: the placenta sheds cfDNA that carries the fetal genome at a few percent of the total, which is how counting chromosomes from a blood draw became the largest genomic test by volume in history. Tumors add a third, rarer voice, and the entire cfDNA industry is organized around hearing it against the other two.

The physics of the assay is a noise budget. At its foundation sits a ratio problem: a patient’s tumor-derived fragments may sit at one part in a thousand after surgery, or one part in a million during surveillance, dissolved in a background of healthy DNA. No sequencer reads accurately enough to see that directly, so the pipeline manufactures its own precision. Stabilizing tubes stop blood cells from rupturing and diluting the signal. Size selection exploits the fact that tumor cfDNA is shorter than leukocyte DNA. Molecular barcodes tag each original fragment before amplification, so that dozens of reads of the same starting molecule collapse into one consensus call — and a sequencing error, which appears once, stays an error. What reaches the report is not a readout but an arithmetic argument: this many independent molecules carried the variant, and chance cannot explain that count.

The three clinical markets are really three settings of the same instrument. Prenatal screening counts chromosome-scale doses and needs the least depth. Therapy selection in oncology sequences a panel of hundreds of genes from plasma to pick an approved drug, at depths that tolerate one-in-a-hundred variants. Minimal-residual-disease surveillance is the extreme case: panels built from a patient’s own tumor genotype track dozens of private mutations, hunting recurrence months before imaging could see it. Early-detection screening pushes hardest of all, hunting many cancer types in asymptomatic blood, where every gain in sensitivity is paid for in specificity — a false positive in a healthy population is a colonoscopy, a false negative is a missed cancer.

The chemistry is stable; the economics and the regulation move. Laboratory licensure, unique reimbursement codes and companion-diagnostic approvals determine which assays reach patients, and each new approval re-prices the installed base of sequencers, robots and freezers behind them. The cfDNA laboratory of 2026 is a factory of assumptions made auditable: every step from tube to report is a controlled process with a documented error model, and the field’s real product is not a variant list but a calibrated statement about what was not seen. The scale is measurable: the field’s two tabled vendors publish 2025 revenues of roughly $2.3 billion and $1.0 billion, and their flagship assays were the first in their classes through FDA review and IVDR Class C certification.