Measuring Life: From Single Cells to Liquid Biopsies

# Cell-free RNA: can a blood sample tell us what tissues are doing? Watch the video first. Use this companion to revisit the reasoning and its evidence limits. Cell-free DNA mostly tells us whose genome is present. But nearly every cell carries the same DNA, so DNA alone says little about what cells are doing. RNA is different: it reflects which genes are being expressed. Cell-free RNA, or cfRNA, is fragments of RNA circulating in plasma, released from cells around the body. Why was that hard to measure? Plasma RNA is scarce and fragmented, and much of it comes from blood cells themselves. Signals from a placenta, a liver or a brain are faint traces in the mix. In 2014, Koh, Quake and colleagues characterized circulating RNA broadly, and used tissue-specific genes to follow changes across pregnancy. Pregnancy offered a natural test. In 2018, Ngo, Moufarrej and colleagues measured cfRNA in maternal blood. In a pilot study of 31 women, nine placenta-specific transcripts predicted gestational age with accuracy comparable to ultrasound, at substantially lower cost. A related study of 38 women at elevated risk of preterm delivery, 23 who delivered at term and 15 early, found seven transcripts that classified preterm delivery up to two months before labor. The authors called for validation in larger, blinded clinical trials. These were pilot results, not a validated clinical test. Use known signatures. Think back to the orchestra. Single-cell atlases recorded each instrument separately, so we know what a violin, a piano and a trumpet sound like. Blood is the mixed recording. If you know each instrument's sound, you can estimate how much of each is in the mix. That's deconvolution: inferring the parts of a mixture from reference signatures. Back to molecules. For each gene, the measured blood signal is approximately a sum: each cell type's fraction, times that cell type's reference expression signature. The computer searches for the fractions that best explain the data. How do you know the unmixing works? Test it where the answer is known. Before turning to plasma, the team applied their reference to bulk samples of known tissues, and checked that it recovered sensible cell type fractions. Then they turned to blood. In 2022, Vorperian, Moufarrej, the Tabula Sapiens Consortium and Quake did this with Tabula Sapiens as the reference. In healthy plasma, most of the cell type signal came from platelets, red blood cell lineages and white blood cells, but they also detected contributions from cell types in the liver, intestine, lung, pancreas, heart and kidney. Limits matter. Similar cell types have overlapping signatures, so some had to be grouped, and origin isn't always uniquely identifiable. Cell types missing from the reference can't be assigned at all; brain was absent from that first atlas. And these were research cohorts, not routine diagnostics. ## Evidence guide PILOT / EARLY STUDY: Ngo 2018 studied 31 women for gestational age and 38 elevated-risk women for preterm delivery; larger blinded trials were required. COLLABORATIVE PRIMARY RESULT · CONSORTIUM: reference-based cfRNA deconvolution estimates contributions, with overlapping and missing signatures. TEACHING ANALOGY: the orchestra explains mixtures, not a biological mechanism. These studies are not universally validated clinical tests. Sources: [koh2014], [ngo2018], [vorperian2022], [moufarrej2022]. See the course bibliography and claim audit.