Biobanks & human population samples

Every analyte begins to change the moment it leaves the body. Nucleases chew nucleic acids, proteases digest their substrates, cells release and then consume metabolites, and oxidation proceeds at room temperature without asking permission. The preanalytical phase — collection, transport, the hours before a freezer — is therefore not logistics around a measurement; it is part of the measurement. This is why circulating tumour DNA is quoted in half-lives of hours, and why a blood tube for liquid biopsy that sits in a courier van for a day answers a different question than one processed within the stability window its assay assumes. Each analyte has its own clock: RNA degrades fastest, cell-free DNA in hours, many proteins and metabolites in days — and a bank that froze everything identically without recording the pre-freeze biography has banked mixtures of samples and stories.

Long-term preservation is then a choice about which chemistry of decay to arrest. Slow-freezing to −80 °C lets ice crystals form, and ice mechanically and osmotically damages cells even while it halts their chemistry — the reason cell lines go to liquid nitrogen at −196 °C, where vitrification turns the intracellular water glassy instead of crystalline. FFPE, the pathology standard, makes the opposite trade: formalin cross-links proteins and nucleic acids, preserving tissue architecture perfectly while fragmenting DNA and imprinting artefactual C-to-T changes through deamination — a bias modern sequencing pipelines must recognize rather than pretend away. Lyophilization removes water sublimed from the frozen state, buying room-temperature stability for ambient logistics, at the price of rehydration damage to fragile complexes. No format preserves everything; a bank is a portfolio of traded-off chemistries, matched to the analytes its users will interrogate in ten or fifty years.

At population scale the object changes nature: a biobank becomes measurement infrastructure. A million-person cohort’s value is not the freezers but the representativeness — whose alleles define the reference frequencies every future diagnostic calibrates against, which exposures are recorded alongside which genomes, and whether consent structured under GDPR makes the samples legally usable for questions not yet invented. Unevenly recruited cohorts bake their skew into every downstream panel and risk score; this is the same calibration problem the personalization pages carry, with a collection bias instead of a statistical one at its root.

The bank is thus the quiet precondition of the rest of this cluster: stability windows decide whether a liquid biopsy result means the tumour or the courier, and longitudinal cohorts are what turn multi-omics panels from snapshots into trajectories. In diagnostics as in surveying, the instrument you can trust for decades is the one whose zero point was set carefully on day one.