What a Gene's Expression Fingerprint Does Not Show

CFTR, the lung gene, in lung

CFTR is the cystic fibrosis gene, and the disease is defined by what happens in the airway. Its GTEx median expression across 54 tissues puts pancreas first at 65.8 TPM, transverse colon at 17.3, minor salivary gland at 9.9, terminal ileum at 5.1 — and lung fifth, at 2.4 TPM.

Across all 54 tissues the median is 0.13 TPM and 41 of them sit below 0.5. So lung is genuinely above background, and it is 28-fold below pancreas in a gene whose clinical story is a lung disease.

The reason is that CFTR is expressed at high levels in a small subset of airway cells. A bulk measurement of lung is dominated by the cells that are numerous, so a gene concentrated in a rare cell type is diluted into the average. The number is correct and the inference is not.

What TPM can be compared across

TPM is transcripts per million: of every million transcript molecules in that sample, how many were this gene, normalised for the gene's length. That normalisation is what makes it comparable between genes within one sample, which counts per million are not.

Across samples it is a proportion rather than an amount. If one tissue's transcriptome is dominated by a handful of extremely abundant genes — as pancreas and liver are — every other gene's share is pushed down, so a between-tissue TPM comparison is partly a statement about the rest of the transcriptome.

And a median over bulk samples conceals the spread. Two tissues with the same median can differ completely in how consistently the gene is expressed across donors.

The annotations beside the numbers

A tissue-specificity label such as tissue enhanced, which is what CFTR carries, is a classification of the expression profile rather than an independent fact: it says the gene's distribution across tissues is uneven in a particular way, computed from the same kind of data.

Protein class annotations are more useful because they come from a different kind of evidence — CFTR's include transporters, predicted membrane proteins and FDA approved drug targets, which between them tell you more about what to expect from the protein than any single expression number.

An empty subcellular location, as CFTR has in the lookup, means no localisation data in that source for this protein, not that the protein is nowhere.

The gene model is the part with no interpretation in it

Alongside the expression numbers sits something harder: the canonical transcript's structure. CFTR's is ENST00000003084, on the plus strand, 27 exons, a 6,070 nt transcript with a coding sequence from 71 to 4,513 — 4,443 bases, which is 1,481 codons, of which the last is the stop. That leaves 1,480 amino acids, and CFTR's protein is 1,480 residues.

That arithmetic agreeing is a useful habit, and the off-by-one in it is the point: a CDS whose length is not a multiple of three, or whose codon count minus the stop does not match the annotated protein, means you are looking at a different transcript from the one you think.

Exon coordinates are the input to anything practical — designing primers that span a junction, placing a variant in an exon, choosing a region to target — and they belong to a specific transcript, so the transcript has to travel with them.

Reading a dossier without over-reading it

• A high TPM is evidence the gene is expressed there. A low TPM is weak evidence of anything, because the cell type may be rare in the sample.
• Tissue comparisons are proportions, not amounts. A gene can be equally abundant in two tissues and have very different TPMs.
• Disease associations and known drugs are aggregations of other people's evidence, and they inherit those sources' biases: a well-studied gene looks more associated with everything.
• A source that did not respond is a gap in the page, not a zero. Check for that notice before concluding something is absent.

Related tools

Related guides

• Converting Gene IDs: Symbols, Ensembl, Entrez, RefSeq and UniProt →
• Interpret GO Enrichment: FDR & Background →
• Reverse Complement of DNA, Explained →