Crop gene banks

Why store thousands of seed varieties that no farmer alive will ever plant?

In cold vaults around the world sit millions of small packets of seed. Most of them will never be sown in a field. Many are landraces no farmer has grown commercially for a century, wild relatives that yield badly and shatter their grain on the ground, or duplicates of duplicates. Keeping them alive is not cheap: they must be dried, chilled, monitored, and every few decades grown out and re-harvested before viability fails.

So a reasonable person asks the obvious thing. If a variety is worse than what we plant, and nobody wants it, why is it worth money to keep? What exactly are we paying for?

Let us test the premise that a variety is "worse". Worse under what conditions? A landrace that yields half as much under irrigation and fertiliser may be the only one that sets seed in a drought year, or the only one carrying resistance to a disease that does not currently exist in your country. The judgement "worse" is always made against today's environment — and the environment is exactly the thing we cannot hold fixed.

That reframes the question. We are not storing plants; we are storing answers to questions not yet asked. Which invites the sharpest form of the problem: could we not simply keep the ones we will need? Ask yourself how that selection would be made. To know which accession carries resistance to a pathogen, you must have the pathogen to test against. Before it arrives, the useful accession is indistinguishable from the useless one. So the selection cannot be made in advance, and keeping "the useful subset" is not an available option.

Now weigh what happens if we are wrong in each direction. Keep something never needed and you have wasted its storage cost — small, and payable in cash. Discard something later needed and there is no second chance: an extinct landrace cannot be re-derived, because the specific combination of alleles it carried was assembled by centuries of selection in one place. That asymmetry — cheap to keep, irreversible to lose — is the entire logic of redundancy. It is the same reason aircraft carry duplicate hydraulic systems that are, on any ordinary flight, dead weight.

Does the need actually arrive? It does, and the historical cases are specific. In 1970 the United States maize crop lost roughly fifteen percent of its yield to southern corn leaf blight, because the great majority of hybrids shared a single source of male-sterile cytoplasm — uniformity that the pathogen exploited across a continent at once. In the early 1970s the search for resistance to rice grassy stunt virus ran through thousands of accessions at the International Rice Research Institute before finding it in a single wild sample of Oryza nivara; that one accession went on to underpin resistant varieties across Asia.

There is also a case for redundancy of the bank itself, not just of the varieties. The Svalbard Global Seed Vault opened in 2008 as a backup to other collections — a copy of a copy. In 2015 it made its first withdrawal, when ICARDA, whose genebank in Aleppo had become inaccessible in the Syrian war, drew on its deposits to rebuild its collections elsewhere and later returned duplicates. Notice what that shows: the failure that struck was not the one the vault was popularly described as guarding against, and the redundancy still paid.

Two clarifications keep this honest. The first is that gene banks are not passive archives. Seed longevity is finite even at minus eighteen degrees Celsius, and viability must be monitored and the accession regenerated — grown out and re-harvested — before it falls too far. Regeneration is where genetic drift and accidental selection creep in, and where under-resourced banks accumulate backlogs. Some crops cannot be seed-banked at all: potato, cassava, banana and most fruit trees are held as living plants, in tissue culture, or cryopreserved, all considerably more expensive.

The second is that redundancy is not the same as usefulness. A large fraction of stored accessions are poorly characterised — passport data thin, traits unscreened — so breeders cannot find what they need even when it is present. That is a real limitation of the argument, and it is why genomic characterisation of collections matters as much as the collections themselves; a resource that cannot be searched is only weakly insurance.

So the defence of gene banks is not that every accession is valuable. It is that value is conditional on futures we cannot forecast, the cost of retention is low and payable, and the cost of loss is unbounded and unrecoverable. That combination is precisely when redundancy is rational rather than wasteful — and it is why the argument does not generalise to keeping everything about everything.

• How do plant breeders actually search a collection — what do core collections and genomic screening do to the needle-in-a-haystack problem?
• Why do some crops resist seed banking entirely, and what does cryopreservation cost by comparison?
• Who owns a landrace collected from a farmer's field, and how do access and benefit-sharing agreements try to answer that?

Every term the collection defines is gathered in the glossary.