Seven ancestors, and what we conclude from them too quickly

There is a scene I live through several times a year, and it always begins with a sheet of paper.

A Brazilian exporter sends me his offer. Twenty lines, sometimes thirty. Mundo Novo, Catuaí Vermelho, Catuaí Amarelo, Acaiá, Topázio, Obatã, Bourbon Amarelo, Icatu, Arara, Tupi. Names that sound different, sheets that announce different profiles, prices that are tiered. And across from me, often, someone asking which of these coffees has the most character, the way you'd ask a wine merchant whether gamay has more character than syrah.

The question is legitimate. It is also badly framed, and for a long time I answered it without seeing why.

Because these twenty names are not twenty families. They are twenty arrangements of a handful of cards, and the handful is smaller than everyone imagines. If you trace back the genealogy of all the arabica cultivars released in Brazil in the twentieth century, you arrive at thirteen ancestors. And seven of them carry ninety-seven and a half percent of the genetic heritage of the set.

This is not an image, it is a published calculation. But this calculation is regularly taken out of its frame, and the article that follows is as much about the figure as about the misuse made of it.

One hundred twenty-one names, thirteen ancestors

The work comes from Tesfahun Setotaw and his coauthors, published in Crop Science in 2013. They took the one hundred twenty-one arabica cultivars released in Brazil between 1939 and 2009 by the four institutions that do breeding in the country: the Instituto Agronômico de Campinas, known as the IAC, the Federal University of Viçosa, Epamig, which is the agricultural research body of Minas Gerais, and the Fundação Procafé. Seventy years of catalogs, four centers, one hundred twenty-one names officially made available to producers.

Then they calculated, for each pair of cultivars, what is called the coefficient of parentage.

We need to stop for a second on this tool, because everything else depends on it and it is often misunderstood. The coefficient of parentage is the probability that two genes taken at random, one from one variety and the other from a second variety, are in fact copies of the same gene present in a common ancestor. If it equals zero, the two plants have no forebear in common in what we know of their history. If it equals one, they are genetically the same thing. It is an old measure, used in plant breeding for decades, long before anyone knew how to sequence anything, and its raw material is not DNA: it is the crossing records. Who was crossed with whom, and in what year.

What this measure does not say, it has to be stated right away, because that is where misreadings are born. It does not say what a variety is worth in the cup. It does not say whether it is resistant, productive, suited to altitude. It does not even say how much two plants resemble each other in practice, since two brothers receive different halves of the same heritage. And it says nothing at all about the soundness of a supply chain, because that is not its object: it concerns a list of cultivars, not a country.

It says one thing only, and says it well: whom these plants descend from, and how many times the same parents come back in the history.

At the end of the calculation, the authors define the genetic base of this set of cultivars: thirteen ancestors, of which seven carry ninety-seven and a half percent of the whole. The other six share the remaining two and a half percent, which is to say traces.

A single tree accounts for half

Among those seven, the distribution is not balanced either, and that is the figure that stopped me the first time.

Bourbon Vermelho alone accounts for fifty-two and three-quarters percent of the genetic base of these cultivars. More than half. The material called Sumatra contributes nineteen. The Timor Hybrid, eleven and a half. Between them, these three trees carry more than eighty percent of the heritage of seventy years of arabica breeding.

Two dates explain almost everything.

Bourbon Vermelho was introduced to Brazil in 1852, from the island of Réunion. The material called Sumatra arrived in 1896, from the island of the same name. Before them, the first arabica planted in Brazil went by the name Nacional, and it was Typica. That is the starting inventory: Typica, then Bourbon, then an Asian material that itself descends from the same lineage, since Bourbon and Typica are the two branches that came out of the same small Yemeni stock.

Everything selected afterward was selected from what was already there. That is how a breeder works: starting from what grows at home, spotting the tree that stands out from the row, fixing it, crossing it again with another tree that stood out from the row. The Ethiopian material, that is, the real diversity of the species, was only collected and distributed from the missions of the second half of the twentieth century, and it was neither productive nor adapted as it stood. Nobody was going to ask a producer to replace a profitable orchard with wild trees of unpredictable yield.

You see the mechanism by reading the pedigree table, and it is almost geometric.

Caturra Vermelho is a mutation of Bourbon Vermelho, released by the IAC in 1951. It is Bourbon that stopped growing tall, which makes it a shorter, denser tree, easier to harvest and more demanding of fertilizer. Mundo Novo is a natural cross between Sumatra and Bourbon Vermelho, spotted in an orchard in the State of São Paulo and released in 1977. Catuaí is a cross of Caturra Amarelo with Mundo Novo, released in 1972: so, mutated Bourbon crossed with Bourbon crossed with Sumatra. Acaiá comes from Mundo Novo. Topázio comes from Catuaí and Mundo Novo. Ibairi crosses Bourbon Vermelho with Mokka, which is itself an arabica of the same origin.

Each generation reshuffles the same cards. And Setotaw sums the matter up in a way that is, I find, the most telling of the whole study: Mundo Novo and Icatu Vermelho, between them, carry eighty-seven and a half percent of the genetic base as first descendants. Two names that hold the vast majority of a catalog of one hundred twenty-one.

It wasn't broad to begin with

Here we have to add a layer, otherwise we put Brazil on trial for something that is not its doing.

Arabica is a young species, born of an accident. It is a hybrid between the ancestors of two other species, canephora and eugenioides, followed by a doubling of the number of chromosomes that made this cross fertile. The genomic work of Jarkko Salojärvi and his coauthors, published in 2024, dates this founding event between three hundred fifty thousand and six hundred ten thousand years ago. On the scale of cultivated species, that is yesterday.

And above all, this study shows that the species went through several genetic bottlenecks before any domestication. Before anyone planted a single tree, arabica had already seen its numbers collapse several times, probably in step with climate oscillations, and each collapse carried off variability that does not come back. The split between the wild populations and those that gave rise to the cultivars goes back some thirty thousand years. In other words, the plant we cultivate was already a narrow branch of an already thin tree.

Then comes human history, which narrows things further, and it comes down to two episodes.

Around 1600, a few seeds leave Yemen clandestinely and found the Indian lineages. A century later, the Dutch plant in Southeast Asia, and a single plant shipped to Amsterdam ends up at the origin of a large part of the American Typica group. One plant. The coffee that made the fortune of Latin America descends essentially from a handful of seeds and one tree in a greenhouse.

Let us put the two narrowings one behind the other. Arabica arrives in Brazil already poor. Brazilian breeding did not create this narrowness, it inherited it, and the interesting question is what was done with it afterward.

What the authors write, and what we make them say

Setotaw and his coauthors conclude that the reduction of diversity in this set of cultivars calls for a broadening of the base, and that diversifying is useful to limit losses due to diseases, pests and abiotic stresses, that is, drought, frost and heat. It is a breeder's conclusion, addressed to breeders, and it is perfectly reasonable: their job is precisely to have material in front of them.

The slide happens afterward. From this sentence, people regularly draw a conclusion the study does not carry: that Brazilian coffee would be fragile. That is a leap of reasoning, and it contains two.

The first leap is an error of perimeter. The study measures a list of arabica cultivars. A producing country is not a list of cultivars. It is also a research apparatus, a network of nurseries, a geography, a fleet of machines, agronomic practices, and in Brazil's case a second species cultivated on a large scale that is not mentioned once in this calculation.

The second leap is a confusion between exposure and capacity to respond. A narrow genetic base tells you what you are exposed to. It does not tell you what happens when the event occurs, and that is nevertheless the only thing that matters to a buyer. Two farms with the same orchard do not come out of it the same way depending on whether they have a technician, water, cash and a replanting plan.

We have the historical precedent. Coffee leaf rust, a fungus that attacks the underside of leaves and makes them fall, thus weakening the tree and collapsing the following harvest, emptied Ceylon at the end of the nineteenth century. It pushed planters toward tea, which incidentally explains why Sri Lanka is today a country of tea and not of coffee. It crossed Java, then Africa, then the Atlantic: it was reported in Brazil in 1970.

And Brazil did not empty out. In the decades that followed, it became the world's leading producer by a margin nobody approaches. The exposure was real and the feared scenario did not take place, because something stood between the two.

What the country sets against it

That something has a rather prosaic name: public research, continuous, for seventy years.

The one hundred twenty-one cultivars of the study can be read in both directions. As a narrow base, which is what they are. Or as the trace of a breeding apparatus capable of putting out new material every seven months on average, of testing it in several environments and over several harvests, of registering it and putting it into nurseries. The same table that shows the concentration also shows the cadence, and the cadence is a form of response. It is, in practice, the time a country needs to change what is planted at home.

The response to rust shows it. It came from the Timor Hybrid, a natural cross between arabica and canephora spotted on the island of Timor in the middle of the twentieth century, in an orchard where the two species grew side by side. A cross like that is normally sterile, because the two species do not have the same number of chromosomes, and that is what makes this case remarkable: that hybrid was fertile, and crossable with arabica. It brought resistance genes that arabica did not possess, for the simple reason that canephora, for its part, coevolved with the fungus.

Brazil exploited this route to the end, and fast. Icatu Vermelho, which comes from a canephora whose chromosome number was doubled so it could be crossed with Bourbon Vermelho, was released in 1992. Obatã and Tupi, derived from Villa Sarchi through a Timor Hybrid, in 1996. Canário, which crosses Catuaí Amarelo with a Timor Hybrid, and Siriema, which goes looking for a completely different species, Coffea racemosa, were released by the Fundação Procafé around 2000. Some twenty years between the arrival of the fungus and the availability of cultivars that resist it: that is the real lead time. That is what capacity to respond means, and it appears in no coefficient of parentage.

There is also a geography, and it does work that nobody credits to genetics. Brazilian coffee is planted in states that have neither the same climate, nor the same altitude, nor the same calendar: Minas Gerais, São Paulo, Espírito Santo, Bahia, Paraná, Rondônia. A frost that hits the south does not hit Rondônia. A drought on the cerrado does not empty Espírito Santo. A genetically homogeneous orchard spread over two thousand kilometers and six climate regimes is not exposed like the same orchard concentrated in one valley, and that is exactly what Ceylon lacked.

Finally, there is everything that is not the plant: irrigation, managed fertilization, the pruning and stumping cycles that make it possible to renew an orchard without uprooting it, mechanization of the harvest on the cerrado, and an agricultural extension network that carries new material out to the farms. All of that is financed and decided, so all of it can be mobilized within a few years. A genetic heritage cannot.

And the quarter the calculation doesn't speak of

There remains the point that, in my view, settles the question, and it had escaped me.

Canephora represents about twenty-five percent of Brazil's coffee production, and forty percent of world production. In Brazil, it comprises two sets of material, the Conilon group and the Robusta group, which are two forms of the same species, and it is cultivated on a large scale in Espírito Santo, Rondônia, Bahia, Minas Gerais and Mato Grosso. None of this enters Setotaw's table, which covers arabica only. When we say “Brazilian coffee,” we are therefore talking about a whole of which a quarter is outside the calculation.

And that quarter is bred according to a strictly opposite logic.

Incaper, the agricultural research institute of Espírito Santo, and its partners have registered thirty canephora cultivars with the ministry of agriculture. Most are clonal, and we need to explain what that means because it is not at all the equivalent of an arabica variety. A clonal cultivar is not a plant, it is an assemblage: superior mother plants are selected, multiplied by propagation by cuttings, and several clones are grouped under a single cultivar name. What the producer plants is not a line, it is a composite population.

The number of clones is written in each sheet, and it is never small. EMCAPA 8111, released in 1993, groups nine clones. EMCAPA 8121, the same year, fourteen. Robustão Capixaba, in 1999, ten. Vitória Incaper 8142, in 2004, thirteen. The three cultivars of 2013, Diamante, Jequitibá and Centenária, nine each.

And the reason for these numbers is explicitly written in Incaper's documentation: it is recommended that a clonal cultivar be formed by grouping at least eight superior clones that are compatible with each other, in order to offer the producer a better guarantee of pollination and not to abruptly reduce the genetic base of the plant population. This is not a researcher's precaution added in an appendix. It is the construction rule of the cultivar, and its motive is the genetic base.

The most striking case is EMCAPER 8151, called Robusta Tropical, released in 2000. That one is propagated by seed, and comes from the recombination of fifty-three elite clones of the program. Its sheet puts forward, as a selling point to the producer, a broad genetic base, and presents production stability as flowing from it. Fifty-three.

Finally there is a structural reason, and it is biological. Canephora is self-incompatible: a gene, called S, with at least three alleles, prevents a plant from fertilizing itself with its own pollen or with a relative that is too close. In other words, this species cannot be inbred like arabica, which is self-pollinating and whose lines are fixed precisely by self-pollination. The narrative of narrowing that holds for arabica cannot be transposed as is: on a quarter of Brazilian production, the plant itself forbids it.

What this changes on an offer sheet

Let's go back to my twenty lines, because that is where the calculation becomes a working tool again.

If Mundo Novo, Catuaí, Acaiá and Topázio are all recombined Bourbon, then the variety name, on a Brazilian arabica offer, is weak information. Not false, weak. It tells you something about agronomy, that is, the habit of the tree, its fertilizer needs, its resistance, the ease of harvest. It tells you almost nothing about taste, because the genetic distance between these lots is too small to produce stable sensory differences.

And that is good news, not a disappointment, because it means that what makes the difference in the cup is elsewhere, and that this elsewhere is precisely what we can act on. The altitude and climate of the plot. Ripeness at picking, which is by far the first factor and the worst paid. The processing: dry process, wet process, pulped natural, fermentations. Drying, its speed, its homogeneity. Storage and the time elapsed since harvest. Orchard management, shade, fertilization, pruning.

When a seller offers me a Catuaí and a Mundo Novo from the same farm explaining that one is fruitier, I know that what he is describing to me is not a difference of variety. It is a difference of plot, of lot, of drying pass or sorting, and that is perfectly fine: I still need to find out which, because that is the one that is reproducible the following year.

For canephora, it is the reverse, and it is a reflex few buyers have. The name of the cultivar carries real information there, because it designates a documented assemblage: a number of clones, a ripening season, a drought tolerance, a bean size, and in recent sheets an evaluation of the cup. Asking “which cultivar” of a conilon producer is a useful question. Asking it of a Catuaí producer is much less so.

What this does not prove

I don't want to turn a study into a certainty, in either direction.

The coefficient of parentage is calculated on declared pedigrees, not on sequenced genomes. It assumes two things: that the published parentages are accurate, and that the founding ancestors are independent of one another. The first assumption is generous, because a natural cross spotted in an orchard in 1940 does not always have two well-identified parents. The second is false by construction, since Bourbon and Typica descend from the same Yemeni stock, and this error goes in a precise direction: it makes the base of arabica look broader than it is. A study on molecular markers would give other numbers, probably of the same order.

The study stops in 2009. Everything released since does not appear in it, and that is precisely where what comes next is at stake. F1 hybrids, which cross an Ethiopian material with an American cultivar and exploit the vigor of the first generation, are a serious attempt to broaden the stock. They have a practical drawback that explains how slow they have been to establish themselves: an F1 hybrid does not reproduce faithfully from seed, so it has to be multiplied by cuttings or in vitro culture, which changes the economic model of a nursery and the producer's relationship to their own seed. It is worth noting that this is exactly the mode of propagation the conilon sector has practiced for thirty years.

In the other direction, nothing I wrote about the offsetting factors guarantees anything. A research apparatus is funded, so it can be defunded. A twenty-year response to a fungus does not mean we will respond in twenty years to a climate. And the fact that a country held up once is not a law.

What can be said fits in two sentences, and I will stop there. The genetic base of Brazil's arabica cultivars is narrow, and that is measured. To deduce from that measure alone a judgment on the soundness of the supply chain is to confuse a list of plants with a country, and to forget the quarter of its production that is bred on the opposite principle.

The question I ask now

For years, faced with a Brazilian offer sheet, I asked for the variety. I still ask for it, but I know what the answer is worth, and I no longer expect from it what it cannot give.

What I ask for in addition today is what has been planted in the last five years. Not what is in production: what is coming in. Because that is the only question whose answer commits the future of a farm, and because it tells me about the thing that really matters, which is not the genetic heritage of the orchard but the speed at which its owner knows how to change it.

And when a producer tells me they are trying two rows of something I have never heard of, on a plot they don't yet dare to convert, I know the conversation will be interesting. Those two rows are the agricultural version of what a breeder does with a tree that stands out from the row. That is precisely how Mundo Novo, Caturra and the Timor Hybrid were born: someone looked at an odd tree and decided not to pull it out.

Notes and sources

1. On the genetic base of Brazilian arabica cultivars and the coefficients of parentage: T. M. Setotaw, E. T. Caixeta, A. A. Pereira, A. C. B. de Oliveira, C. D. Cruz, E. T. Zambolim, L. Zambolim and N. S. Sakiyama, “Coefficient of Parentage in Coffea arabica L. Cultivars Grown in Brazil,” Crop Science, 2013. One hundred twenty-one cultivars released between 1939 and 2009 by the IAC, the UFV, Epamig and the Fundação Procafé; genetic base of thirteen ancestors, seven carrying 97.55% of the heritage, including Bourbon Vermelho 52.76%, Sumatra 19.05% and Timor Hybrid 11.59%; Mundo Novo and Icatu Vermelho carrying 87.65% of the base as first descendants; pedigree table with parents and release years; the authors' recommendation to broaden the base. The study covers arabica only.
2. On the introductions to Brazil: the dates of 1852 for Bourbon Vermelho, from Réunion, and of 1896 for the Sumatra material are those taken up by Setotaw from the historical works of Carvalho (1957) and Bettencourt (1968, 1973). The first arabica planted in Brazil is designated there under the name Nacional, that is, Typica.
3. On the origin and genomic history of the species: J. Salojärvi et al., “The genome and population genomics of allopolyploid Coffea arabica reveal the diversification history of modern coffee cultivars,” Nature Genetics, 2024. Founding hybridization between the ancestors of C. canephora and C. eugenioides dated between 350,000 and 610,000 years, bottlenecks predating domestication, separation of wild accessions and cultivar progenitors around 30,500 years, clandestine exit from Yemen around 1600, founding of the Typica group by the Dutch in Southeast Asia and the Amsterdam plant.
4. On Brazilian canephora, clonal cultivars and their genetic base: M. A. G. Ferrão et al., “Cultivares de cafés Conilon e Robusta,” Informe Agropecuário, Cafés Conilon e Robusta : potencialidades e desafios, Belo Horizonte, vol. 41, no. 309, p. 17-25, 2020, Incaper. C. canephora, Conilon and Robusta groups, about 25% of Brazilian production and 40% of world production; producing states: Espírito Santo, Rondônia, Bahia, Minas Gerais, Mato Grosso; thirty cultivars registered with the ministry of agriculture, most of them clonal; recommendation of at least eight superior and compatible clones per clonal cultivar, motivated by pollination and by the refusal of an abrupt reduction of the population's genetic base (after Fonseca et al. 2008 and Ferrão et al. 2017a, 2019); EMCAPA 8111 nine clones and EMCAPA 8121 fourteen clones in 1993, Robustão Capixaba EMCAPA 8141 ten clones in 1999, Vitória Incaper 8142 thirteen clones in 2004, Diamante ES8112, Jequitibá ES8122 and Centenária ES8132 nine clones each in 2013; EMCAPER 8151 Robusta Tropical, seed-propagated, released in 2000, derived from the recombination of fifty-three elite clones, presented on a broad genetic base and the production stability that follows from it; gametophytic self-incompatibility governed by an S gene with at least three alleles (after Lewis 1949; Ramalho, Santos and Pinto 1990).
5. On coffee leaf rust: Hemileia vastatrix, described in Ceylon in 1869, reported in Brazil in 1970, then spread across all of Latin America.
6. On the Timor Hybrid and its Brazilian descendants: natural interspecific cross between C. arabica and C. canephora, source of the Catimor and Sarchimor groups. For Brazil, Icatu Vermelho, C. canephora tetraploidized by Bourbon Vermelho, 1992; Obatã IAC 1669-20 and Tupi IAC 1669-33, Villa Sarchi by Timor Hybrid CIFC 832/2, 1996; Canário, Catuaí Amarelo by Timor Hybrid, and Siriema, C. racemosa by C. arabica Blue Mountain, Fundação Procafé. Years and parentages after the table of Setotaw and coauthors.
7. On F1 hybrids and vegetative multiplication: the constraints of propagation by cuttings or somatic embryogenesis, and their consequences for the seed sector, are documented in the breeding literature and in the technical documents of World Coffee Research.