The artificial banana flavor in children’s medicine, candy, and budget ice cream tastes more like a variety that stopped being commercially viable in the 1960s than like the banana you can actually buy. The compound is called isoamyl acetate — and the variety it more closely resembles is the Gros Michel, which dominated global export markets until Race 1 of Fusarium wilt moved through Central American plantations across the 1950s and 60s and effectively ended its commercial life. But the resemblance is coincidence, not design. Isoamyl acetate was first synthesized and deployed as a commercial flavoring compound in the 1850s and 60s, decades before the Gros Michel became the world’s leading export banana. Both the ester and the fruit happen to carry high concentrations of the same compound through independent chemistry. The flavor was never derived from the Gros Michel. It just smells more like it than like the Cavendish that replaced it.
The Gros Michel is not extinct, despite what gets repeated online. It still grows on smallholder farms in Central America, Hawaii, Myanmar, and Malaysia — commercially near-absent but not gone. What Race 1 destroyed was its place in supply chains.
This matters because the same pathogen family is now moving against the Cavendish through identical logic. Tropical Race 4 — a distinct Fusarium strain against which the Cavendish has no inherited resistance — has been spreading through Cavendish plantations since the 1990s, reached Latin America in 2019, and shows no sign of being contained. The industry is monitoring, managing, and waiting. The Gros Michel left a chemical vestige that hundreds of millions of people encounter without knowing what it records. The Cavendish is not yet gone. But the loop is running again, and the question is what humanity built to interrupt it.
The mathematics of disappearance
Historians of agriculture estimate that humans have, over millennia, cultivated approximately 7,000 plant species for food. The number grown at meaningful commercial scale today is somewhere under 150. Fewer than a dozen — wheat, rice, maize, soybean, potato among them — account for the majority of the calories humanity actually eats.
The mechanism was the Green Revolution. Between the 1950s and 1970s, agricultural scientists replaced thousands of locally adapted, farmer-selected varieties — landraces, refined by generations of selection in specific soils and microclimates — with centrally bred, high-yield cultivars engineered for uniformity and productivity. The yield gains were real and very large. The genetic narrowing was not an oversight. It was an explicit, understood trade-off, made by people who grasped what they were doing and believed the arithmetic of mass hunger made it necessary. The problem is that trade-offs compound, and the compounding is now the story.
Commercial rice cultivation in several of Asia’s major producing countries now descends overwhelmingly from a handful of parent lines — the product of decades of standardization that pushed thousands of locally adapted landraces out of active farming systems. The same pattern holds across maize, wheat, soybean. Variety after variety moved to the margins or out of cultivation entirely, not through negligence but through rational optimization for yield at scale.
A figure appears everywhere in this literature — that 75 percent of crop genetic diversity has been lost since 1900. Khoury and colleagues, writing in New Phytologist in 2022, found the evidentiary base for the specific percentage unclear; the FAO’s own Third Report on the State of the World’s Plant Genetic Resources for Food and Agriculture, published in 2025, avoids assigning a single global figure. The honest version of the claim is that we don’t know exactly how much has been lost, the most widely cited number is not robustly measurable, and the frequency of its citation is itself part of how this crisis gets discussed — a large estimate whose methodological fragility, once noticed, allows critics to dismiss the underlying reality. The underlying reality, documented crop by crop and region by region, is not in serious dispute: the diversity is largely gone, the loss was largely intentional, and its consequences are maturing.
The Green Revolution's arithmetic Norman Borlaug's wheat varieties are credited with averting famines that could have killed hundreds of millions. He won the Nobel Peace Prize in 1970, and the citation was not wrong. The varieties that narrowed global crop diversity also fed populations that existing agricultural systems could not sustain. The trade-off was accepted as the price of buying time, made in conditions where the alternative was mass death by starvation of a different kind. The genetic cost was never paid down — the narrowing was understood as temporary, and the bill is now coming due in precisely the form the architects could see at the far end but could not themselves resolve.
The mechanism — Ireland, 1845
Ireland in 1845 was not growing potatoes. It was growing one potato: the Lumper, a variety propagated almost entirely through vegetative reproduction — tubers cut from existing plants and replanted, not grown from seed. Vegetative propagation produces no genetic variation. Not reduced variation, not marginal variation — none. Every Lumper in every field shared a single genetic identity.
When Phytophthora infestans arrived, it found that identity uniformly. There was no resistant subset. No natural selection could favor a surviving line, because there was no variation for selection to operate on. The pathogen broke one defense and encountered it everywhere.
But Ireland did not grow only Lumpers because the country made a poor agricultural choice. The colonial land tenure system — absentee landlordism, rack-renting, subdivision of holdings through generations to sizes where only the highest-caloric-yield crop per acre could sustain a family — made monoculture the economically imposed condition. Subsistence farmers on one-acre plots did not choose the Lumper from among alternatives. They were structurally coerced into it by a system designed to extract maximum rent from minimum land, leaving no margin for agricultural variety.
The blight struck across Europe in 1845 and 1846. Germany, Belgium, and the Netherlands all experienced crop failures. Only Ireland starved at this scale.
Food exports from Ireland continued through the famine years — grain, livestock, dairy — to meet landlord obligations and British market demand. Whig government relief was constrained by laissez-faire ideology; Charles Trevelyan, the Treasury official who administered much of the response, believed government intervention would undermine the natural correction the market was performing. One million people died. Two million emigrated. Ireland’s population fell by roughly a quarter between 1845 and 1852.
The causal chain runs three links deep, each requiring the one before it: biological uniformity made total crop failure certain when the pathogen arrived; colonial land tenure made monoculture the only economically viable option for subsistence farmers; British policy made mass death the outcome rather than a recoverable crisis. The genetic argument is real — diversity would have provided some resistance, would have let selection operate, would have meant the pathogen had more than one lock to pick. But it is one factor in a three-factor system.
What the International Potato Center in Lima holds today is the range Ireland did not have. The CIP genebank conserves 4,797 traditional landrace cultivars and more than 2,500 accessions representing 140 wild potato species — genetic breadth assembled from across the Andes and beyond, representing the diversity that vegetative monoculture inherently forecloses.
The Andean collection The CIP assembled its potato collection through decades of fieldwork across the Andean highlands, where potato originated and where wild relatives continue to evolve alongside pathogens, altitude variation, and seasonal extremes. The collection has been drawn on in breeding programs in over 100 countries. Its existence is an acknowledgment — made retrospectively — of what Ireland didn't have: genetic variation that gives a pathogen more than one lock to pick. Assembling the collection hedges against the biological factor in the three-part chain. The other two remain political problems.
The archive
In 2008, Norway bored a tunnel into a sandstone mountain on the Arctic archipelago of Svalbard and opened the Global Seed Vault. The facility sits 130 meters above sea level on Spitsbergen, the main island, at 78 degrees north — designed to remain frozen without active refrigeration should power fail, relying on surrounding permafrost to hold the temperature. The operating logic is that even if the institution running it collapses, the seeds survive.
As of early 2026, the vault holds approximately 1.387 million seed samples from 132 depositors — roughly one-third of its 4.5 million sample capacity. The remaining space is room for material that could still be collected, if it still exists to collect.
The governance structure requires precise description, because imprecision here is where public reassurance becomes misleading. Norway owns the facility. NordGen, the Nordic Genetic Resource Center, manages operations. The Crop Trust provides financial support to depositors and co-funds running costs. Seeds deposited remain the legal property of the depositing institution. NordGen staff can move boxes within the facility but cannot open them without the depositor’s authorization. Researchers and plant breeders cannot request seeds from the vault directly — they must go to the original depositing genebank.
This is not a global commons. It is a backup drive, and the files belong to whoever uploaded them.
The active working layer sits one level below Svalbard in institutional terms: the eleven CGIAR genebanks, which together hold approximately 730,000 accessions covering more than 3,000 plant species — the collections that researchers actually draw from, that breeders actually request. Svalbard is the backup of the backup, the copy of last resort if something goes wrong with an active collection.
In 2015, something went wrong with an active collection. The Syrian civil war had made it impossible to operate the ICARDA genebank in Aleppo, which held the world’s most significant barley germplasm collection and approximately 141,000 accessions of wheat, fava bean, lentil, and chickpea. ICARDA became the first institution in history to withdraw seeds from the Svalbard vault — 38,073 samples in total, split across two destinations: 8,840 accessions of forages, faba beans, grasspea, and wild relatives of cereals and pulses sent to Lebanon, and 29,233 accessions of cultivated wheat, barley, lentil, and chickpea sent to Morocco. The seeds were regenerated in active collections over the following years, and in 2020 ICARDA re-deposited the refreshed material.
The system worked as designed.
But it was also a specific, bounded proof. The vault protected a threatened genebank from political instability — allowed material to be moved, regenerated, and returned to use. It has not been tested as a recovery mechanism for an active crop failure at agricultural scale, for a blight that has already swept through farms in multiple countries while a breeding pipeline races to catch up. A backup that can protect a building from a war is a different instrument from a backup that can restore a crop system after a pathogen has destroyed it. The distinction is what the Cavendish situation is now beginning to expose.
The replay
The Cavendish banana is not a crop in the conventional sense. It is a clone. Every commercial Cavendish on every supermarket shelf is genetically identical to every other. The variety propagates through vegetative offshoots from existing plants, not from seed — no sexual recombination, no genetic shuffling, no variation. This is precisely why it dominates approximately 95 percent of global banana export trade: uniformity produces consistent sizing, consistent ripening, consistent shelf life. Logistically optimal. Biologically, the Lumper.
The banana market was valued at approximately $128 billion in 2024. Latin America supplies the majority of the world’s exported bananas by volume. More than 400 million people depend on bananas — as food, as income, as both — in the world’s lowest-income producing regions. The scale of what is planted in monoculture is the scale of what is exposed.
Tropical Race 4 is a strain of Fusarium oxysporum f. sp. cubense, the same fungal species as Race 1, but genetically distinct. The Cavendish’s resistance to Race 1 — the property that made it the replacement for the Gros Michel — offers nothing against TR4. There is no commercial treatment. Once TR4 establishes in soil, it persists for decades; it cannot be rotated out, cannot be fumigated away. It is a permanent contamination of whatever ground it reaches.
TR4 was detected in Southeast Asia in the 1990s, spread through Australia and East Asia, reached Africa in 2013, and arrived in Latin America in 2019. By early 2024, Peru’s national plant protection authority, SENASA, had eradicated more than 400 separate outbreaks. The confirmed infected area covered roughly 330 hectares of Peru’s approximately 171,000 banana-growing hectares — a foothold that flooding and cyclone activity in recent years have been widening.
The industry’s response tracks its response to Race 1: monitor, contain, manage, wait for the situation to become commercially unworkable.
The lesson the Gros Michel offered was not complicated. Race 1 killed its commercial dominance through the 1950s and 60s. The industry replaced it with the Cavendish. It then had sixty years to diversify away from dependence on a single export clone. It did not, because the Cavendish was profitable, logistically convenient, and exactly what consumers recognized. The lesson was: when a pathogen forces a switch, switch. The industry appears ready to learn that lesson again.
The seeds for a different outcome exist. The International Transit Centre genebank in Leuven, Belgium — operated by the Alliance of Bioversity International and CIAT — holds more than 1,700 accessions of edible and wild banana species, including germplasm showing resistance to Fusarium wilt. In January 2026, researchers at the University of Queensland published findings in Horticulture Research identifying a genomic region on chromosome 5 that confers resistance to Subtropical Race 4 in a wild banana subspecies. But STR4 is not TR4. They are genetically distinct races with different virulence profiles: STR4 is less aggressive and geographically restricted; TR4 causes disease under tropical and subtropical conditions and is spreading through Latin American Cavendish plantations right now. Resistance to STR4 does not confer resistance to TR4. TR4 resistance in a commercially scalable banana variety remains an unsolved breeding challenge.
Developing a commercially viable, genetically stable TR4-resistant banana variety is a pipeline measured in decades. TR4 spreads through soil that stays contaminated for decades. The two timelines are running simultaneously, and neither is under anyone’s control.
The Gros Michel The Gros Michel is richer and sweeter than the Cavendish, with higher concentrations of isoamyl acetate — which is why the artificial banana ester resembles it more than the variety in your kitchen, even though the ester predates the Gros Michel's commercial dominance and was never derived from it. The variety was commercially displaced, not eliminated: specialty growers in Hawaii sell it; smallholder farmers in Central America, Myanmar, and Malaysia still grow it. Race 1 moved through the region's large commercial plantations in the 1950s and 60s, destroying monoculture operations while smallholder cultivation largely survived. The structural parallel to TR4 is exact. A pathogen optimized for a monoculture; a monoculture allowed to become global before the pathogen caught up.
The fine print
The Crop Trust was built on a specific financial logic: a permanent endowment of $850 million, generating approximately $34 million annually at a 4 percent return, would fund the global genebank system indefinitely without dependence on annual government appropriations. The endowment model was designed for permanence — removing the system’s survival from the budget cycles of governments with other priorities.
As of the end of 2023, the endowment stood at $305 million. Thirty-six percent of the target. New contributions that year totaled $5.9 million — $5.4 million from the United States government, the balance from two other donors. At that rate, the $545 million gap is not a temporary shortfall that institutional momentum will close. It is a structural deficit. The Crop Trust is, in practice, still dependent on the annual goodwill of states whose attention to crop diversity has been episodic at best.
In the same year that Peru’s plant protection authority was eradicating over 400 TR4 outbreaks, the global insurance system designed to respond to exactly this kind of crisis received less than $6 million in new capital.
The access problem is separate and runs deeper. Seeds at Svalbard belong to their depositors. The ITC collection in Leuven — which holds the germplasm from which a TR4-resistant banana variety might eventually be bred — belongs to the Alliance of Bioversity International and CIAT. Whether that material can be rapidly deployed into a breeding program, and under what intellectual property conditions the resulting varieties would be distributed, involves institutional negotiation that the existence of the seeds does not resolve. Seeds in storage and seeds in circulation are different things. The vault solves for storage.
Peer-reviewed literature, including Khoury and colleagues in New Phytologist in 2022, has made a point that seed banking advocates sometimes understate: genebanks preserve a genetic snapshot frozen at the date of collection. They do not preserve the living, co-evolving diversity of crops grown by farmers in their original landscapes across centuries of ongoing selection. An Andean farmer selecting potato tubers annually across altitude gradients and seasonal variation produces a different kind of genetic resource than a genebank accession collected thirty years ago and held at minus 20 degrees Celsius. Both are valuable. Neither is sufficient without the other. In situ conservation — supporting the farming communities and agricultural practices that maintain living diversity — has received a fraction of the institutional attention that the concrete tunnel in the Arctic attracts. The smallholder farmers in the Andes, the Caucasus, and the Ethiopian highlands who maintain crop diversity through daily agricultural practice are not a story anyone is adequately funding to protect.
The commercial seed market is the final layer. Bayer, Corteva, and Syngenta dominate the global market for high-yield commercial seeds. When genebank material is used in breeding programs — as it regularly is — the resulting commercial varieties frequently carry intellectual property restrictions: patents, plant breeders’ rights, licensing conditions that affect who can grow them and at what cost. Seeds released from vaults in a crisis may end up in varieties owned by companies that sell them back to the farmers in the most vulnerable countries. The architecture of commercial seed markets creates the conditions for that outcome. The seed banks are not causing it, and they cannot fix it. But the vault unlocks into that architecture.
The CGIAR network is also undergoing structural change. The One CGIAR reform is consolidating formerly independent research centers under a common board and leadership structure. The stated rationale is efficiency and coordination. The structural risk is that consolidating governance may reduce the distributed redundancy that made the system resilient: eleven independent genebanks with separate management chains are more failure-resistant than a single hierarchical structure, for the same reason that a diverse field is more resilient than a monoculture.
The Plant Treaty's fine print The International Treaty on Plant Genetic Resources for Food and Agriculture, in force since 2004, governs access to genebank material for research, breeding, and training. Roughly 64 crops and forages are covered; material transfers operate under a standard agreement with benefit-sharing obligations attached when commercialization occurs. The treaty's Benefit Sharing Fund — designed to channel some of those commercial returns back to developing country farmers and conservation programs — has received substantially less than was envisioned at negotiation. A legal framework exists. It is structurally underfunded and its benefit-sharing provisions remain contested in practice.
The Svalbard vault is not the story. It is a bunker built for a problem whose solution requires something the bunker cannot provide: a functioning political and financial apparatus that is ready to deploy when the claim comes in.
What the current moment looks like: monoculture expansion continues while the endowment meant to fund its genetic insurance is 64 percent short of its operational target, in situ conservation — the one form of crop diversity preservation that cannot be replicated by a freezer — draws a fraction of the institutional attention the vault attracts, and the IP architecture governing commercial seeds means that vaulted germplasm deployed in a crisis travels back to vulnerable farmers through the market structures of the same industry whose varieties displaced the diversity that created the vulnerability. Each element of this was foreseeable. None of it has been resolved.
The banana makes this concrete. The ITC collection in Leuven holds the germplasm that could theoretically underpin a TR4-resistant Cavendish successor. The breeding pipeline for a commercially scalable, genetically stable variety runs in decades. A genomic region conferring resistance to a related but distinct race was identified in early 2026. TR4 is establishing in Latin American soil that will stay contaminated for decades. The funding to accelerate what comes next is a political decision.
That decision has not been made. TR4 does not wait.
Avis de non-responsabilité de Gen AI
Certains contenus de cette page ont été générés et/ou édités à l'aide d'une IA générative.
Les médias
Principales sources et références
Svalbard Global Seed Vault, official website, seedvault.no. Statistics on deposits, total seed samples, and capacity; 2026 deposit record confirming approximately 1.387 million samples from 132 depositors. URL: https://www.seedvault.no/
CGIAR Genebanks Platform, “About,” genebanks.cgiar.org. Figures on the eleven CGIAR genebanks, 730,000 accessions, and more than 3,000 plant species. URL: https://genebanks.cgiar.org/about/
Crop Trust, Finance and Investments, 2023 Annual Report, report.croptrust.org. Endowment fund value ($305 million as of end 2023, up from $277 million), new contributions ($5.9 million, of which $5.4 million from the United States), and the $850 million endowment target. URL: https://report.croptrust.org/2023/our-work/finance-and-investments/
Munhoz, C.F., Vargas, L., Teixeira, T., Staver, C., and Dita, M. “Fusarium Tropical Race 4 in Latin America and the Caribbean: status and global research advances towards disease management.” Frontiers in Plant Science, Volume 15, 2024. TR4 infection data for Peru: 330 hectares affected of approximately 171,000 total banana hectares; more than 400 outbreaks eradicated by early 2024 per SENASA. URL: https://www.frontiersin.org/journals/plant-science/articles/10.3389/fpls.2024.1397617/full
Chen, A. et al. “Identification of a QTL conferring resistance to the Subtropical Race 4 of Fusarium oxysporum f. sp. cubense in Calcutta 4 (Musa acuminata ssp. burmannica).” Horticulture Research, January 2026, University of Queensland. STR4 resistance mapped to a genomic region on chromosome 5. DOI: 10.1093/hr/uhag001
Khoury, C.K. et al. “Crop genetic erosion: understanding and responding to loss of crop diversity.” New Phytologist, 2022, vol. 233(1), pp. 84–118. Discussion of the methodological basis for the “75 percent crop diversity lost” claim, and the case for both ex situ and in situ conservation. DOI: 10.1111/nph.17733
International Potato Center (CIP), Genebank documentation, cipotato.org. Collection figures: 4,797 traditional landrace cultivars and more than 2,500 accessions representing 140 wild potato species. URL: https://cipotato.org/genebankcip/potato-wild/
ICARDA, press releases on the Svalbard withdrawal (2015) and re-deposit (2020). 38,073 samples total: 8,840 accessions to Lebanon and 29,233 accessions to Morocco, following the closure of the Aleppo genebank. URL: https://icarda.org/media/press-release/icarda-deposits-vital-seeds-svalbard-vault-securing-food-generations-come; withdrawal confirmation: https://www.seedvault.no/2019/09/11/withdrawal-of-icarda-aleppo-seeds-accomplished/
Food and Agriculture Organization of the United Nations, Third Report on the State of the World’s Plant Genetic Resources for Food and Agriculture, 2025. Global figures on the number of plant species in commercial cultivation; avoidance of a single global diversity-loss percentage. URL: https://openknowledge.fao.org/handle/20.500.14283/cd4711en
Food and Agriculture Organization of the United Nations, Markets and Trade: Bananas, fao.org. Latin American share of global banana export origin; Cavendish production volume as just under half of global production. URL: https://www.fao.org/markets-and-trade/commodities-overview/bananas-tropical-fruits/bananas/en
Alliance of Bioversity International and CIAT, International Musa Germplasm Transit Centre (ITC), Leuven, Belgium. More than 1,700 accessions of edible and wild banana species. URL: https://alliancebioversityciat.org/services/genebanks/international-musa-germplasm-transit-centre
Food and Agriculture Organization of the United Nations. “Fighting the deadly disease that is killing the world’s most exported fruit.” FAO Transboundary Plant Pests and Diseases, June 13, 2025. Bananas critical to food security and livelihoods of around 400 million people; TR4 described as spreading through Asia, Africa, the Middle East, and Latin America with no cure available absent replacement with resistant varieties. URL: https://www.fao.org/transboundary-plant-pests-diseases/news/detail/fighting-the-deadly-disease-that-is-killing-the-world-s-most-exported-fruit/en
Arias, P., Dankers, C., Liu, P. and Pilkauskas, P. The World Banana Economy, 1985–2002. FAO Commodity Studies No. 1. Rome: FAO, 2003. Cavendish described as accounting for the vast majority of international banana trade; approximately 99 percent of exports to developed countries during the study period; the manuscript’s “approximately 95 percent of global banana export trade” is conservative within this range. URL: https://www.fao.org/4/y5102e/y5102e04.htm
Market Data Forecast. “Banana Market Size, Share, Trends and Analysis.” Global banana market valued at USD 128.75 billion in 2024. URL: https://www.marketdataforecast.com/market-reports/banana-market
English 
Español 
Français 
Deutsch 
日本語 
Polski