Building a map of the human body, one cell at a time
Use cell-atlas reference profiles while accounting for sampling and capture bias.
Loading video…
# Building a map of the human body, one cell at a time
Watch the video first. Use this companion to revisit the reasoning and its evidence limits.
First, a definition. A cell atlas is a reference map connecting cell identity, gene expression, and the tissue a cell came from. Tabula Sapiens is one such atlas, built by a large consortium that includes the Quake group. Its first main paper, in Science in 2022, reported nearly five hundred thousand cells from twenty-four tissues and organs, with more than four hundred cell types annotated by tissue experts.
And the answer to the design question: the same person. Tabula Sapiens collected many tissues from individual donors, fifteen donors in all, with seventeen tissues from one of them. Comparing lung and heart vessels from one person holds genetic background, age and environmental exposure fixed, so the differences that remain are more likely due to the tissue itself.
The cells were sequenced two complementary ways: one captures many cells in tiny droplets; another sorts cells into plates and reads each one more deeply. Experts in each tissue then named cell types using a shared vocabulary, the cell ontology. Clusters were a starting point, not the final answer.
What can an atlas reveal? Because many organs came from one donor, the team could follow immune cells around the body. T cells carrying identical receptor sequences, members of one clone, turned up in several different organs of the same person.
It also revealed RNA splicing, the cutting and joining that lets one gene produce several versions of its RNA. Some genes, such as M Y L 6 and C D 47, were spliced in cell type specific ways: the same gene, processed differently in different cells.
And broad cell types weren't identical everywhere. Endothelial cells, which line blood vessels, and macrophages, immune cells that patrol tissues, showed tissue specific differences in their gene expression programs: the same broad type, with local accents.
Another caution. To read single cells, a tissue must be broken apart, and not every cell type survives that equally well. So the number of cells captured isn't a direct census of the tissue. The team paired sequencing with pathologists' examination of tissue sections under the microscope to estimate the true proportions.
An atlas is a reference, not a complete model of every human. Fifteen donors can't capture every ancestry, age, disease or cell state, and some tissues were missing. A later preprint, Tabula Sapiens two point oh, added nine donors and four tissues, doubling the cell count. As a preprint, it had not yet been peer reviewed.
The real value of a reference atlas is that it's a coordinate system. A new sample, from a patient, an organoid, or a drop of blood, can be compared against it: whose gene expression pattern does this look like?
## Evidence guide
COLLABORATIVE PRIMARY RESULT · CONSORTIUM: the 2022 atlas reports 483,152 cells, 24 tissues/organs, 475 annotated cell types and 15 donors. It is a sampled reference, not all human populations or states. CURRENT RESEARCH DIRECTION · PREPRINT: the supplied audit identifies Tabula Sapiens 2.0 as not peer reviewed. Preserve consortium credit and dissociation bias.
Sources: [ts2022], [ts2025]. See the course bibliography and claim audit.