Cell Identity, Cell State and Disease

In middle-aged and elderly organ donors with healthy oesophagus, cells carrying a NOTCH1 mutation made up an estimated 12 to 80% of the lining.

NOTCH1 mutations also turn up in oesophageal cancers. Yet this tissue looked normal under the microscope, and in that 2018 study the mutations were more common in normal tissue than in tumours. So a cancer-associated mutation did not make the tissue malignant. To see why, you need two skills. First, telling what kinds of cells a tissue holds, and what condition each is in. Second, telling what a cell does from what it merely looks like. This lesson builds both, then returns to the oesophagus.

From a tissue to a count matrix

Every cell in a tissue carries the same genome. What makes a keratinocyte differ from a macrophage is which genes it transcribes, a pattern held in place by transcription factors and chromatin (see eukaryotic gene regulation). Single-cell RNA sequencing (scRNA-seq) reads that pattern one cell at a time.

The steps are simple to state. You dissociate the tissue into single cells, usually with enzymes. Each cell’s RNA is captured and copied into DNA tagged with a cell barcode, a short sequence shared by every molecule from that cell. In many methods each molecule also gets a unique molecular identifier, so copies made later can be collapsed back into one original molecule. The pooled DNA is sequenced (see sequencing and genomes), and the reads are sorted by barcode and counted.

The output is a count matrix: one row per gene, one column per cell, each entry the number of molecules (or reads) counted. Analysis then runs through quality control, normalization, choosing informative genes and reducing dimensions. Cells are linked to their most similar neighbours in a graph, and groups of tightly linked cells become clusters. Finally each cluster is annotated: given a name by comparing its genes with what is already known about expected expression profiles.

Think of it as a census taken by reading everyone’s shopping list. You can sort people into groups by what they buy, but a list does not tell you who the person is.

Type, state, or artifact?

Two words matter here. A cell type is a stable identity, such as a basal keratinocyte or a T cell. A cell state is variation within that identity: cycling, stressed, activated, early or late in differentiation. A single cell type can vary a great deal in its global gene expression, and cells of one type in different states can land in separate clusters.

So a cluster is an observation to explain, not a type by definition. Six things can create or erase one.

• Sampling zeros. Only some of a cell’s RNA molecules are captured, so a count of zero does not mean the gene is off.
• Doublets. One barcode can label two cells, giving a hybrid profile that looks like a new type.
• Batch effects. Samples handled on different days or by different hands can differ for technical reasons and cluster apart.
• Dissociation stress. Breaking up a tissue can perturb cell state before the RNA is captured.
• Resolution. A clustering parameter sets the scale. Turn it up and one type splits into states; turn it down and distinct types merge.
• Circular marker testing. If you cluster on the expression data and then test the same data for genes that differ between clusters, the P values are inflated, because the clusters were drawn to differ. Markers also depend on what else is in the data set.

The 2016 melanoma study of Tirosh and colleagues shows the payoff when this is done with care. Among 4,645 cells from 19 patients’ tumours, malignant cells within one tumour differed in programmes associated with the cell cycle, a state rather than a separate cancer type.

Bulk averages hide composition

Before single-cell methods, you ground up the tissue and measured RNA in bulk. That measurement is an average over every cell, weighted by how many of each kind there are. A change in the average can come from a change in composition with no change in regulation in any cell.

Work it through with made-up numbers. A tissue holds cell type T, which expresses gene X at 50 units per cell, and other cells that express it at 5 units per cell. Before a treatment, T is 20% of cells. After it, T is 40%. Expression per cell does not change in either group.

bulk signal = (fraction T x level in T) + (fraction other x level in other)

1. Bulk before = 0.20 x 50 + 0.80 x 5 = 10 + 4 = 14 units per average cell.
2. Bulk after = 0.40 x 50 + 0.60 x 5 = 20 + 3 = 23 units per average cell.
3. Fold change = 23 / 14 = 1.64. A bulk experiment would report gene X as up 1.6-fold, although no cell regulated it.
4. Single-cell data would show the same expression inside each cluster and a changed proportion of clusters.

One caution cuts the other way. If dissociation loses fragile cells preferentially, the proportions in your single-cell data are themselves biased. A shift in cluster sizes between two conditions can come from a difference in how well each sample survived the enzymes.

From description to cause: lineage and perturbation

Software such as Monocle can arrange cells along a path and give each a pseudotime, its distance along that inferred path from the start. It is tempting to read the path as a movie of one cell maturing. It is not. Sequencing destroys each cell, so pseudotime orders different cells that were caught at different points, and its direction is a guess.

Three kinds of evidence turn such a picture into a cause.

Lineage. Mark a cell and every descendant. Weinreb, Klein, Camargo and colleagues gave mouse haematopoietic progenitors inherited DNA barcodes that are expressed as RNA, sequenced some cells early and followed their sisters’ progeny. Sister cells shared fate biases that the RNA profiles had not detected. In tissues, the same logic uses inducible Cre recombinase to switch on a fluorescent reporter in a few cells (see stem cells and tissue renewal).

Perturbation. Remove or block the candidate and see whether the outcome changes, then restore it (see experiments that distinguish mechanisms).

Protein. RNA does not measure protein. CITE-seq, described in 2017, reads DNA-tagged antibodies alongside RNA, so a surface marker can be measured in the same cell rather than inferred.

Homeostasis lost: clones, competition and cancer

A renewing epithelium stays the same size only if cell gain matches cell loss. In mouse oesophagus, only basal progenitor cells divide. Each division can give two progenitors (PP), two differentiating daughters (DD) or one of each (PD), and in wild-type tissue these outcomes are balanced. Differentiating cells move upward and are shed.

Crowding adds a second balance. In crowded MDCK sheets and in zebrafish epidermis, Eisenhoffer, Rosenblatt and colleagues saw epithelia squeeze out living cells (see epithelia); most extruding cells lacked the apoptosis marker active caspase-3 (see cell death). Knocking down the stretch-activated channel Piezo1 in zebrafish blocked this extrusion and produced masses of epidermal cells.

A mutant clone wins by tilting these balances. It need not divide faster (see cell-cycle control). Notch signalling is linked to differentiation of squamous epithelial cells (see Notch signalling). In mouse oesophagus, Abby and colleagues found in 2023 that Notch1-mutant clones expanded without a higher division rate. EdU tracking was consistent with more progenitor and fewer differentiating daughters per division, and wild-type neighbours at the clone edge differentiated faster.

Neighbours also matter for tumours. In a mouse model of oesophageal carcinogenesis, Colom and colleagues found that early lesions are removed more often when the surrounding clones are highly competitive, and less often when competition is blocked with a drug. In a mouse carcinogenesis model, Abby and colleagues also found that deleting Notch1 reduced tumour growth and that a NOTCH1-blocking antibody reduced the size of established lesions. These are mouse results about mechanism, not human prevention data. Which genes push a clone toward cancer is covered in cancer genetics and oncogenes.

Design problem

Notch1-mutant clones take over mouse oesophageal epithelium. Did the mutant cells divide faster, or did their daughters differentiate less? Design the experiment, and say what each answer predicts.