Measuring Life: From Single Cells to Liquid Biopsies

# Single-cell genomics: why one cell at a time matters Watch the video first. Use this companion to revisit the reasoning and its evidence limits. It can't tell you. The same average comes from every cell expressing both genes at half level, or from two separate populations. That's the problem of bulk averaging. It's like recording an orchestra with one microphone: you hear the sum, not which instrument played which note. Single-cell measurement puts a microphone on every instrument. Back to molecules. Genes in DNA are transcribed into RNA, and many RNAs are translated into protein. The genome is the DNA sequence. The transcriptome is the collection of RNA transcripts a cell is expressing at a particular time. Because they use that genome differently. A liver cell and a neuron transcribe different sets of genes, so their transcriptomes differ even though their genomes are nearly identical. Nearly, because some cells carry mutations, and immune cells rearrange their receptor genes. But mostly, identity lives in which genes are switched on. Single-cell RNA sequencing reads transcriptomes one cell at a time. Conceptually: separate a tissue into individual cells, isolate each one, capture its RNA, and copy that RNA into DNA, which sequencers read. Every molecule is tagged with a short DNA barcode recording which cell it came from, and often a unique molecular identifier, so later copies can be traced back to one original molecule. Then sequence and count. Each sequenced fragment is called a read. Reads are sorted by barcode and matched to genes, giving a table: one row per cell, one column per gene. The entry X i j is the count for gene j in cell i. It's a sample, not a perfect census: many RNA molecules are never captured, so plenty of entries are zero by chance. And it measures RNA, not protein. With that table, we can ask what kind of cell each one is. A cell type is a relatively stable identity: a T cell, an endothelial cell lining a blood vessel, a fibroblast. A cell state is a temporary or contextual program within a type: activated, dividing, stressed, or infected. One T cell can pass through several states. Each cell has thousands of gene counts, far too many dimensions to see. Researchers group cells with similar profiles by clustering, squeeze the data into two dimensions for viewing with methods like UMAP, and use known marker genes to name the groups. But a cluster isn't automatically a new cell type. It might be a state, a technical artifact, or a slice of a continuum. Naming it takes biological evidence. Quake and collaborators used early single-cell RNA sequencing to reconstruct how lung cells mature in developing mice, from 198 cells, and to survey the adult and fetal human brain, from 466 cells. Early studies like these counted hundreds of cells. Atlases now count hundreds of thousands. ## Evidence guide ESTABLISHED MOLECULAR BIOLOGY: DNA sequence and expressed RNA differ, including genome exceptions. MEASUREMENT TECHNOLOGY: scRNA-seq samples RNA, not protein or every molecule. COLLABORATIVE PRIMARY RESULT: historical lung, brain and endometrial studies involved other labs. A cluster requires biological annotation; cell type and cell state differ. Sources: [treutlein2014], [darmanis2015], [wang2020]. See the course bibliography and claim audit.