I Do the Sums and It Should Lose
In 1971, the British biologist John Maynard Smith, who had designed aircraft as a young engineer and was nearly thirty before he switched to biology, did a very simple calculation. His conclusion: sexual reproduction ought to lose. Here was the setup.
Take a population with separate males and females, mating at random, with roughly equal numbers of each. Now suppose a mutant female appears. She can produce offspring on her own, without mating. In every other respect she is exactly like the other females: the same number of offspring, the same survival rate, the same food.
The sexual female produces two offspring: on average, one daughter and one son. The asexual female produces two daughters. In the next round, the asexual line has two females ready to produce more offspring directly; the sexual line has only one. Twice as many. And it has taken just one generation to open that gap.
The advantage compounds. In the third generation it is four to one; in the fourth, eight to one. After twenty generations, about a million to one. At that rate, sexual reproduction ought to be crowded out within very few generations. It has not been eliminated. That is the question of this essay. One researcher later went so far as to call it “the queen of problems in evolutionary biology.”
Even “has not been eliminated” understates the case. The vast majority of eukaryotes engage in some form of sexual reproduction: animals, plants, fungi, most protists. Something the arithmetic says should have been driven out has become the normal way of life across this whole expanse of living things.
We need to be clear about where that factor of two comes from. Most directly, it comes from “producing males who do not themselves bear young.” A sexual female puts about half her reproductive investment into sons, who do not give birth; an asexual female puts all of it into daughters.
A more precise name, then, is “the cost of producing males.” In organisms without separate males and females, there is no automatic division into “half the offspring unable to bear young,” and no automatic twofold cost. Nor is it a law: change the sex ratio, or have males help rear the young, and the figure changes.
That cost has been measured directly in a natural system. In the lakes of New Zealand lives a freshwater snail only a few millimeters long. In the same lake, some need to mate and others do not. They eat the same food and evade the same predators.
In 2017, researchers raised these snails in large outdoor tanks. A year later, they compared the proportion of asexuals in the parental and offspring generations. The increase in asexuals matched the twofold model. In other organisms, the two forms often live in different places, so ecological differences get mixed into the comparison.
What makes this snail such a rare opportunity is that both forms crowd into the very same water. It later hitched rides on ships to Europe and North America. The ones running rampant there are all of the kind that needs no mate.
Sex has other costs beyond that factor of two. Recombination can dismantle combinations that already work well: several genes may be especially effective together, and recombination pulls them apart.
The reverse is also true: when good genes are tied to bad ones, recombination can separate them. Finding a mate takes time and carries risks. Mating leaves less time to feed and makes it easier to get eaten. Some diseases spread only through mating. Males also have to compete: a peacock’s tail, a deer’s antlers, courtship songs, all are effort spent on mating.
There is another cost, and we bear it in our own bodies. Meiosis, the special division that makes sperm and eggs, does not always sort the chromosomes correctly.
About 10 to 20 percent of recognized human pregnancies end in miscarriage. Of the embryos lost in the first three months, about half have the wrong number of chromosomes. Most of those errors arise during the formation of the egg.
After a woman turns forty, the proportion of eggs with chromosome errors rises to around half. One reason is that a woman’s eggs begin meiosis while she herself is still a fetus. They stop partway through and remain there for decades, resuming the division only shortly before ovulation.
Over those decades, the little protein rings holding the chromosomes together gradually loosen. Down syndrome results from an extra chromosome 21; the great majority of cases originate at the stage when the egg is formed. Add these costs together, and the bill for sexual reproduction comes to more than that factor of two.
II What Exactly Is It?
Before asking why something has not been eliminated, we have to say what it is. The first distinction to make is this: sex and reproduction are two different things.
Reproduction is about numbers: one becomes two. Sex is about inheritance: genetic material from different sources enters the same life cycle and gets mixed.
The two can happen separately or be coupled together. Animal life cycles couple them together, which makes them look like a single event.
Conjugation in ciliates offers the clearest counterexample. Paramecium, the organism in secondary-school biology textbooks, does it. Two cells come together, exchange some genetic material, then part.
That sexual event has not, by itself, turned two cells into four. Numbers increase through ordinary cell divisions afterward. Conversely, fission, budding, and vegetative reproduction can produce vast numbers of individuals without any genetic exchange. So “sex took place” does not entail “reproduction took place at this step.” The reverse does not follow either.
The second distinction: sex in eukaryotes and gene exchange in bacteria are different processes. To explain why, we first need to look at what meiosis actually does.
In each of your body cells, chromosomes come in pairs. One came from your father, the other from your mother. The same set of genes runs along both, though the versions may differ. A gamete, a sperm or an egg, can carry only one set. Otherwise, the chromosome count would double with every generation.
Making gametes therefore requires separating the paired chromosomes so that each gamete gets just one of each. That is meiosis. But it does much more than “separate.” The crucial part lies in a step along the way.
Before separating them, the cell deliberately breaks its chromosomes. As those breaks are repaired, the two homologous chromosomes, one from each parent, find each other, pair up side by side, and use each other as templates to fill the gaps. In the process, the two chromosomes exchange a stretch at the break. This is called crossing over.
It is a physical cutting and rejoining. Under a microscope, you can see the points where paired chromosomes cross. In 1909, a Belgian cytologist first proposed that these crossings were where the two chromosomes exchanged pieces.
In other words, the chromosome you pass to your child is neither the one your father gave you nor the one your mother gave you. It is a new version, made by cutting the two apart and joining them back together.
Humans have twenty-three pairs of chromosomes. Simply choosing one chromosome at random from each pair gives a single person more than eight million possible combinations in their gametes. Add crossing over, and scarcely any two gametes will be alike. Most of the reason you and your siblings look different lies in this step. And the step is deliberate, not an accident.
The cell produces a dedicated set of enzymes, breaks its own chromosomes, and repairs them using the homologous chromosome as a template. Exchange occurs during the repair. Put another way: to achieve the mixing, the cell deliberately cuts its own DNA.
Afterward, the cell divides twice in succession, with no further DNA copying in between. The first division separates the paired homologous chromosomes. Only the second separates the sister chromatids.
One parent cell thus becomes four gametes, each with a single set. In men, all four become sperm. From the start of division in a spermatogonial cell to the production of sperm takes more than two months; daily output runs into the hundreds of millions.
In women, only one of the four portions becomes an egg. The others are discarded as tiny “polar bodies.” A woman ovulates only around four hundred eggs in a lifetime. Finally, two gametes fuse, restoring the pairs.
If the chromosomes cannot pair, the whole procedure stalls. A horse has thirty-two pairs of chromosomes, a donkey thirty-one. Their hybrid offspring, the mule, is sturdy but almost always infertile: when meiosis comes, its chromosomes cannot find matching partners. Seedless watermelons work on the same principle. They have three sets of chromosomes, which cannot pair off evenly during meiosis, so normal seeds fail to develop.
This describes a common meiotic sexual cycle: pairing, cutting and exchanging, halving, and fusion. Every step can go wrong; every step needs dedicated machinery. But not every lineage follows every item on this list. Male Drosophila melanogaster, for example, normally undergo meiosis without crossing over.
Bacteria do none of this. They can pick up a stretch of DNA straight from the water, receive it through a tube, or have it carried between them by a virus. We covered those three routes in the second essay.
None requires a cycle of “pairing, halving, fusion,” and none belongs to a fixed stage in the life cycle. The exchange is also asymmetrical: one gives, another receives. Bacterial exchange is more like “finding something useful and fitting it on.” Eukaryotic sex is “reshuffling your own cards with your mate’s in every generation.”
Under the broadest definition, sex as genetic exchange, bacteria can be included. But the questions of mechanism in this essay have to be confined to sex involving meiosis.
The third point: meiosis is a thorough reworking of ordinary cell division.
There are at least three sets of major alterations. The first division separates paired homologous chromosomes rather than sister chromatids. One round of copying is followed by two divisions. And the bonds between sister chromatids do not all come loose at once: those along the arms are released first, those near the centromere later. This is precisely what allows the first division to separate the homologous chromosomes while keeping the sister chromatids together for the time being.
These three sets of alterations interlock. Remove any one, and the rest no longer works. “Meiosis evolved by modifying ordinary division” is an inference backward from the machinery we see today. Nobody witnessed that history.
III The Parts Came from Further Back
There is now a way to investigate whether an organism has meiosis: check for the specialized parts.
The usual checklist includes the enzyme that deliberately makes breaks before pairing. A strand-exchange protein that works mainly in meiosis. A checkpoint protein on the chromosome axis.
Pairing proteins that promote exchange between homologous chromosomes. And the structural proteins that fasten two homologous chromosomes side by side, like a zipper.
What this census first changed was not our understanding of sex, but our answer to the question “Which groups really are asexual?” Previously, deciding whether a microorganism had sex meant catching it mating under a microscope. Yet in many microorganisms, the sexual stage is extremely brief, requires very particular conditions, or depends on conditions nobody has yet managed to reproduce in the laboratory.
In 2005, researchers surveyed the genomes then available, paying special attention to Giardia, long described as “primitively asexual.” If dirty drinking water gives you diarrhea, Giardia is one of the usual suspects. They found several core meiosis genes inside it.
In 2008, the toolkit was expanded and the survey extended to several other groups with no generally accepted sexual cycle at the time. Again, the results turned up fairly complete sets.
The most familiar example is yeast. When life is good, it grows by dividing in two. Only when it is hungry does it undergo meiosis, producing four spores enclosed in a small sac.
At least some “asexual eukaryotes” thus became “eukaryotes whose sexual stage has not yet been observed, but whose genomes retain the machinery for meiosis.” The reasoning cannot be reversed. Finding several specialized genes is a strong clue to past or present meiosis. Failing to find a particular gene is not strong proof of asexuality.
This is the same principle we will return to in the eleventh essay: the force of negative evidence depends on “how likely we would have been to find it if it existed.” For a gene that may have changed beyond recognition, that likelihood is low.
Genes can be lost or change beyond recognition. Different branches can also use unrelated proteins to perform similar structural tasks. The ancestral state inferred from all this is that the last common ancestor of eukaryotes possessed a set of meiotic machinery.
The reason is not that “one particular gene is very old.” It is that several functionally coordinated components occur in very distant branches, covering different steps: break formation, recombination, pairing, crossing over, and cohesion.
For every major branch to have assembled the same combination independently would require more separate origins than “their common ancestor already had it, and different descendants subsequently lost parts.” Go back one step further, and the parts themselves are older than meiosis.
The clearest example is the strand-exchange enzyme. Bacteria already have it; archaea have a homologous version. In eukaryotes, both the enzyme used for ordinary repair and the one used preferentially in meiosis belong to this older family.
The second example is more intriguing. Experiments in 1997 showed that the enzyme responsible for meiotic breaks participates in making them through a covalent bond. Structural comparisons also linked it to a class of archaeal topoisomerases: enzymes whose particular job is to cut DNA and join it back together to undo tangles.
A protein family originally belonging to this “cut and rejoin” maintenance system took on the job of deliberately making breaks in meiosis. There is therefore substantial evidence from homology for the claim that “meiotic recombination was reorganized from an ancient DNA maintenance system.”
“DNA repair was the sole or primary reason meiosis first arose” is a different claim, at a different level. Homology alone cannot establish it. Modern meiosis actually makes large numbers of breaks on purpose, then calls in repair machinery to deal with them. There is another component whose deep origins are much murkier than those of the recombination enzymes.
The structure that fastens two homologous chromosomes side by side has an overall form under the electron microscope that is astonishingly similar in widely separated eukaryotes: an axis on either side, a region in the middle, filaments spanning the gap.
Yet the proteins making up these structures often show only faint signs of homology in their sequences. Some show no detectable relationship in conventional alignments. A 2021 study of nematodes found that the proteins evolve so quickly that their structure and function are conserved far more strongly than their sequences. A 2025 comparison in fruit flies reached the same conclusion.
So the structural principle is ancient, while the deep origins of many of the particular proteins implementing it remain obscure.
Another asymmetry deserves attention. The members of this toolkit do not all have equal diagnostic value. The break-making enzyme and the specialized strand-exchange enzyme are the most useful because they occur almost exclusively in meiosis. The side-by-side fastening structure is the opposite: its form is highly conserved, but its sequence changes extremely fast.
Within a single machine, then, some components make good markers of deep time, while others are wholly unsuitable. Different parts of the same machine differ by several orders of magnitude in how legible a record they leave through time.
When using this toolkit to survey different groups, you therefore have to know which kind of part you are using. Choose the wrong one, and your result is “not found,” not “not there.”
IV Several Explanations, None Enough
Why, then, has sex not been eliminated? There are five or six current explanations. The real disagreements concern the conditions each one requires.
The first: the ratchet. It takes its name from a mechanism that turns in only one direction. Muller proposed it in 1964; Haigh put it into quantitative form in 1978. The central idea is not that “every generation necessarily gets worse.”
It goes like this. Imagine a population in which every individual carries some harmful mutations. Some carry fewer, some more. The class carrying the fewest often contains only a small number of individuals. If, by chance, none of these “cleanest” individuals leaves offspring in one generation, that class is gone.
Without recombination, it cannot come back. To produce a cleaner individual again, existing mutations would have to be removed, and the chance of a mutation being undone is extremely small. The minimum load therefore rises, notch by notch. Like a ratchet, it can turn in only one direction.
Think of a group of people copying the same book, with nobody allowed to check anyone else’s copy. Once the cleanest copy is lost, it is gone for good. Recombination can turn the mechanism backward: two individuals carrying three mutations each might produce, after recombination, an offspring carrying only one.
The parameter requirements are quite specific. The speed of the ratchet depends strongly on effective population size, the harmful mutation rate, and the size of each mutation’s effect. A vast population, a very low mutation rate, or very strong selection against those mutations can slow the ratchet to a crawl.
What it does not explain is why sex remains frequent in many large populations, or why it occurs every generation rather than occasionally. If a little recombination now and then is enough to recreate low-load individuals, frequent sex still needs an additional explanation.
Empirical support gets us only halfway. Harmful mutations certainly keep arising. The real difficulty is independently measuring a natural lineage’s effective population size, its full harmful mutation rate, and the distribution of those mutations’ effects, then using them to predict the ratchet’s actual speed.
The second: mutations magnify one another.
The models Kondrashov developed in 1982 and 1988 differ from the ratchet. They require something particular: when several harmful mutations occur together, their combined damage must exceed a simple sum. In other words, one mutation costs a little; two cost more than twice as much; five might finish you off altogether.
If that is so, recombination can produce some offspring with exceptionally high loads. Those offspring are eliminated in one stroke, while others with exceptionally low loads survive. It amounts to gathering scattered burdens into one place and clearing them out together. But the necessary condition is demanding: the damage has to accumulate faster than linearly.
If five mutations cause exactly five times the loss caused by one, this mechanism does not work at all. Recombination has merely redistributed the same burden. The total has not changed.
The 1988 paper emphasized that if harmful mutations arise at a rate of roughly one or more per genome per generation, the mechanism can generate a sufficiently strong difference in selection. And those parameter requirements are precisely the model’s most contentious point.
A mutation-accumulation experiment in nematodes in 1997 estimated the harmful mutation rate per haploid genome at about 0.0026, far below “one or more.” Later estimates, from different species and different methods, have ranged from well below one to near or above it.
More importantly, “more than additive” has never been established as a universal relationship. Different experiments find effects greater than the sum, less than the sum, or exactly additive.
The third: interference between linked genes. This one is easiest to picture. Suppose two beneficial mutations arise in a population: mutation A in individual A, mutation B in individual B.
Without recombination, A and B can only compete. One eventually wins, and the other beneficial mutation is lost. The loss occurs not because B is bad, but because it was born in the wrong individual. With recombination, A and B can mate and produce offspring carrying both A and B.
Fisher proposed this idea in 1930, and Muller independently in 1932. In 1966, Hill and Robertson generalized it to interference between linked loci. The mechanism is strongest when populations are large enough for several beneficial mutations to coexist.
Empirical support comes from experimental evolution. A 2002 experiment with green algae found that recombination had a more marked effect on the response to selection in large experimental populations; in small populations, the effect was weak. Experiments with yeast, meanwhile, found that one important role of recombination was not to add two good mutations together, but to free a good mutation from a background of bad mutations hitching a ride with it.
The fourth: the Red Queen.
The name comes from the Red Queen’s line in Through the Looking-Glass: “It takes all the running you can do, to keep in the same place.” The mechanism is more specific than the general observation that “environments change.” Here is how it works. To infect a host, parasites usually have to “recognize” it, and their ability to recognize different host genotypes varies.
When a host genotype becomes common, the kinds of parasite good at infecting it also become more common. The common genotype consequently suffers most, while rare genotypes enjoy a temporary advantage. Once the rare ones become common, the parasites catch up again.
No “best genotype” can ever settle into permanent dominance. This may help explain why some of the most variable genes in humans are precisely those that help the immune system recognize pathogens.
Sexual reproduction scrambles combinations every generation, producing new rare types along the way. What the Red Queen requires is therefore not “be better,” but “be different from the previous generation.”
That is its most fundamental difference from the first three explanations. Those concern whether mutations are good or bad; this one concerns whether combinations are old or new. Its conditions include genotype-level matching between host and parasite, parasites harmful enough to matter, and sufficiently rapid change in both as they respond to each other.
Simply “having parasites” does not mean the Red Queen’s conditions are met. The most important natural system for this explanation is, once again, that New Zealand freshwater snail.
In its native populations, sexual and asexual individuals can coexist in the same lake, even along the same stretch of shore. The proportion reproducing sexually ranges from almost none to almost all. One important parasite is a fluke.
That range is itself an experimental opportunity: the same species, the same body of water, yet such enormous variation in the proportion of sexuals.
A study reported in 1998 tracked the system for five consecutive years and found a time lag in the parasites’ response to common host clones. In experimental infections, host clones that had remained rare over the preceding years were less susceptible to local parasites than clones that had once been common. That is evidence of a rare-type advantage, more than merely “there are more parasites where there is more sex.”
Follow-up work in 2009 reported that, in two habitats, the asexual clones that had originally been most common were replaced by formerly rare clones over roughly seven to ten years, while the sexual forms persisted. As time passed, the formerly common clones became more susceptible to local parasites, but showed no equivalent change with parasites from elsewhere.
The reservations are just as specific. Field correlations are hard to disentangle completely from water depth, host density, historical origins, and other ecological differences. So researchers took the question into the laboratory.
A 2011 experiment used nematodes as hosts and a highly virulent bacterium as the pathogen. This system allowed researchers to engineer three mating regimes: exclusive self-fertilization, the wild-type mixture, and compulsory outcrossing. The key lay in three pathogen treatments.
One was a pathogen-free control. In the second, every generation faced the same fixed pathogen strain. The third was a coevolution treatment: bacteria were recovered from the bodies of nematodes killed by infection, then used to infect the next generation, allowing the pathogen to change alongside its host. The experiment ran for about thirty host generations. Its results drew a sharp distinction.
Under coevolution, every population restricted to self-fertilization went extinct within twenty generations. The pathogen-free and fixed-pathogen treatments did not produce the same outcome.
The populations forced to outcross persisted. Wild-type populations began with an outcrossing rate of around 20 to 30 percent. Once exposed to the pathogen, their outcrossing rate rose rapidly, exceeding 70 percent within the first eight generations.
What happened next is the most important comparison. With a fixed pathogen, outcrossing subsequently fell. Only where the pathogen was evolving too did a high outcrossing rate last until the experiment ended. That ruled out an important alternative explanation: “the pathogen simply causes stress, so mating becomes more frequent.”
The fixed pathogen caused stress too, but did not sustain high outcrossing over the longer term. Only a continually changing pathogen population produced the result the model predicted. Of all the evidence in this essay, this comes closest to a direct test of causation.
It did more than observe a correlation. It set up a comparison differing in a single variable: the same pressure, changing in one treatment, unchanging in the other. Its scope must be stated just as clearly. It lasted only a few dozen generations. The host–pathogen combination was selected by the researchers.
Taken together, modern reviews commonly adopt a position that does not seek one answer for every organism. Small populations may feel the ratchet more strongly. Large, rapidly adapting populations may be more affected by linkage interference. Systems under heavy parasite pressure may show Red Queen dynamics.
Existing theories provide several mechanisms that can offset the asexual advantage within different parameter ranges. None has been shown to be the universal reason that all eukaryotic lineages retain sex involving meiosis. The substance of the disputes lies not in the mechanisms themselves, but in the parameters.
Nobody doubts that the ratchet can turn, or that recombination can break up linked genes. The dispute is whether, in real populations, these effects are large enough.
Answering that requires independent measurements of effective population size, the full mutation rate, the distribution of effects, and parasite pressure. Even today, these quantities can rarely all be measured accurately in the same natural system.
V Asexuality Never Went Away
Asexual reproduction has always been here, right beside us. Many of the dandelions by the roadside set seed without fertilization. Their seeds are clones of the mother plant; one plant is almost identical to the next. In summer, female aphids give birth to more females without mating, one after another. A newborn daughter often already carries the next generation, her mother’s granddaughters, inside her. Only in autumn do they produce males, mate, and lay eggs to survive the winter. Within a single year, asexual and sexual reproduction take turns.
Zoos have seen it too. In 2006, Komodo dragons at two British zoos laid eggs that could hatch, despite being kept apart from males.
All the hatchlings were male. Sex in these lizards is determined by sex chromosomes, and offspring produced by a female without mating can only be male. In 2001, at an aquarium in an American zoo, a female hammerhead shark that had never encountered a male also gave birth to a young shark.
In 2021, researchers discovered that two chicks in the captive-bred California condor population had hatched without their mothers mating, even though males were available. There is also a mushroom-farming ant in South America whose queens perpetuate themselves entirely by cloning. Not a single male can be found in most populations; only a few populations in the Brazilian Amazon still have them.
Then there is a crayfish, an asexual species whose birth humans have all but witnessed. In the 1990s, a marbled crayfish appeared in ornamental aquarium shops in Germany. There were only females, all belonging to the same clone.
The pet trade carried it around the world. It now runs rampant in the wild in places including Madagascar. Honeybees use another method: drones develop from unfertilized eggs. They have mothers, but no fathers.
Several species of whiptail lizard in the southwestern United States are also entirely female. They descend from hybrids between two sexual lizard species. Females still go through mating motions with one another, which stimulates ovulation. No males; the ritual remains.
“Asexuality” actually contains several different things. Some forms are close to pure cloning. Others undergo meiosis first, then find a way to restore the pairs, for instance by fusing two nuclei that were supposed to go their separate ways. The genes still get shuffled.
When plants produce clonal seeds without fertilization, it is called apomixis. Dandelions do this. Self-pollination, by contrast, is not asexual: it still involves meiosis and the fusion of two gametes. It simply does not mix with anyone else.
Where do these lineages come from? Most asexual lineages have branched off repeatedly from sexual ancestors, and most are young. In 2009, researchers compiled the ages of numerous asexual animal and plant lineages. Very few were genuinely ancient.
But that distribution must be read carefully. If transitions from sexual to asexual reproduction keep happening, then even if both kinds go extinct at the same rate, the tips of the tree will still be full of young asexual lineages. “Most asexual lineages are young” holds up. “Asexuality necessarily leads to rapid extinction” cannot be established from that pattern alone.
There are examples in the other direction too. Lichens are organisms in which fungi and algae live together. Estimates for lichens suggest twenty-six to forty-four transitions from sexual to asexual reproduction, alongside about fourteen to twenty-five transitions back. In a group of stick insects found around California in the United States, asexuality has arisen independently several times. A 2011 study found that some of these branches had been reproducing asexually for more than a million generations.
Asexuality also favors certain places. Compared with closely related sexual forms, asexuals often live farther north, higher in the mountains, and more often in areas covered by ice during glacial periods. Their geographical ranges are also larger.
One explanation is that, after the ice retreats, a single parthenogenetic female can establish a new population without waiting for a mate. Yet many asexual forms are also polyploid, or the products of hybridization. Their tolerance of cold might come from extra chromosome sets rather than from asexuality. Several explanations have examples to support them. None covers everything.
There is an example of asexuality’s weakness that everyone has seen: the banana. Almost all supermarket bananas belong to the same variety. They are seedless, propagated generation after generation by separating off shoots, every plant a member of the same clone.
The previously dominant variety was devastated around the 1950s by a fungus in the soil. Today’s variety is being pursued by a new form of that same fungus. A whole field of bananas looks alike, and when disease arrives, they fall together. Agriculture and horticulture are full of clones anyway: grafted apples and grapes grown from cuttings, to name two.
Humans would like to put asexuality to work too. Hybrid rice yields well, but the hybrid advantage lasts only one generation. Sow its seeds again, and the traits segregate, so farmers have to buy new seed each year. For all the decades China has grown hybrid rice, this has remained an obstacle. If hybrid rice could clone itself through its seeds, that hybrid vigor could be preserved generation after generation.
In 2019, two teams independently replaced meiosis in rice with ordinary division, then induced egg cells to begin developing without fertilization. For the first time, rice produced clonal seeds. At the time, these made up at most 10 to 30 percent of the seeds. Later improvements applied to hybrid rice raised the proportion above 90 percent. Switching this institution off means changing several genes at once. The adjustments are still under way.
VI The Scandal
Bdelloid rotifers are animals so small that you need a magnifying glass to see them clearly. Less than a millimeter long, they have a fringe of cilia at the front that seems to turn like two wheels. They live in moss, roof gutters, and temporary puddles.
When the water dries up, they dehydrate into tiny lumps. A breeze can carry them elsewhere; water revives them. Dried rotifers can remain that way for years. In 2021, researchers recovered one from Siberian permafrost that had been frozen for about twenty-four thousand years. Once thawed, it came back to life and began reproducing.
They are all female. Among the hundreds of known species, nobody has ever seen a male. The traditional account says they have gone tens of millions of years without sexual reproduction. In 1986, Maynard Smith, the man doing the sums in the first section, called large, ancient groups with no observable sex an evolutionary scandal. According to the theories in Section IV, they ought to have collapsed long ago.
They have another oddity. Their genomes contain large numbers of genes “picked up” from bacteria, fungi, even plants, an astonishingly high proportion for animals. During dehydration, DNA breaks into fragments. When water returns, the cells have to put it back together. Some researchers suspect that foreign DNA slips in at this point.
How do they escape parasites? In 2010, researchers found that when a lethal fungus attacks rotifers, drying out and blowing away offers an escape. The fungus cannot keep up. This led some to suggest that they shake off parasites by drifting, rather than by having sex.
In 2000, the American geneticist Meselson and his colleagues analyzed genes from several species. This was the Meselson who, with Stahl, performed the famous 1958 experiment showing how DNA replicates. They found that the two copies of a given gene were remarkably different, unlike the similar pairs found in sexual organisms, and proposed that the rotifers had gone without sex for a very long time.
For a while, this was the strongest molecular evidence of ancient asexuality. A genome published in 2013 strengthened the case, although its authors also noted that extremely rare, unconventional genetic exchange could not be ruled out entirely.
That qualification later became important. From 2015 onward, researchers found stretches of shared sequence between individuals in the genes of another rotifer genus. They interpreted this as evidence of some sexual process, perhaps a form of meiosis using a different pairing arrangement from the usual one. In 2016, another team analyzed 576 rotifers and likewise found sharing between individuals, but thought it looked more like horizontal transfer, the “picked-up genes” approach.
Both sides accepted that exchange occurred. They disagreed about how, though in 2018 other researchers pointed out that some of the signal might come from contamination between samples. Already, the scandal had changed.
The old question was “How can a large group go tens of millions of years with no exchange at all?” The new question became “How rarely does it exchange, and by what means?” If the premise of the first question held, it would challenge the whole body of theory. The second does not.
In 2020, a whole-genome analysis of eleven wild rotifers found genotype proportions close to those expected under random mating, which the researchers judged incompatible with strict, long-term cloning. In 2022, another study reported signs of sexual recombination in a different genus.
What can be said today is this: the claim that they have gone tens of millions of years with no exchange at all is now very hard to sustain. Yet nobody has seen them produce males and undergo normal fertilization like ordinary animals. The dispute has narrowed to how much exchange there is, whether every branch has it, and how it works.
There is more than one candidate for scandal. One family of tiny crustaceans, the darwinulid ostracods, was named in Darwin’s honor. Each has two shell valves and looks like a little crawling bean. Only females had been observed, leading to speculation that they had been asexual for a hundred million, or even two hundred million, years.
In 2006, researchers found living males, the first reliable record in a century. Some species in the family, however, still have no known males.
Oribatid mites are common in soil, and about a tenth of their species consist only of females. Some may have lived this way for several million years or more. A 2007 analysis also found sexual branches nested among asexual ancestors: asexuality had turned back into sex. A 2017 study found that the machinery for removing harmful mutations was still working in these ancient asexual mites.
There is also a group of fungi that lives inside the roots of most land plants and helps them absorb phosphorus: arbuscular mycorrhizal fungi. With no observable sexual structures, they too were called ancient asexuals. In 2009, researchers detected recombination in a field population. Later genomes revealed large numbers of meiosis genes.
Set these cases beside one another, and the same pattern keeps appearing. A group is classified as asexual because no males or recognizable sexual structures can be seen. Later evidence loosens that judgment. Rotifer and mycorrhizal fungal genomes carry traces of exchange or the machinery of meiosis; in ostracods, living males simply turned up. Oribatid mites are the exception. They still look like genuine ancient asexuals.
The criterion was negative all along: using “not seen” to decide “did not happen.” And most of the evidence that overturned these judgments came from genomes, not microscopes. To see a sexual event, you have to look at just the moment it occurs. To see the traces it left, you only have to sequence.
VII Two Sexes Are Not Inevitable
Mating types are not the same thing as males and females. In many organisms, the two gametes are identical in size and appearance. But they cannot fuse indiscriminately: fusion is possible only between different “compatibility classes.” These classes are called mating types.
This already constitutes a sexual process. There is meiosis, gamete fusion, and recombination. But there are no males and females defined by gamete size. So “two mating types” cannot be translated as “two sexes.” Tetrahymena, a single-celled ciliate found in pond water, has seven mating types. Any two different types can conjugate when they meet.
The split-gill fungus takes this to an extreme. It grows on dead wood, fan-shaped, with gills that look as though they have split down the middle. In Yunnan, people often stir-fry it.
Its compatibility is jointly determined by two regions, each with many versions. The classic estimate lists about 288 specificities at one and 81 at the other. Combined, they yield 23,328 potential mating types.
The number sounds absurd, but the logic is simple. Each region has many versions, and compatibility requires only that “both be different.” Two individuals usually have to differ at both regions for sexual development to proceed normally.
The proper name for that figure is “the classic estimate of the number of potential mating types,” not “23,328 sexes.” The advantage of such an arrangement is straightforward: on meeting another individual at random, the probability of compatibility is much higher than in a system with only two mutually exclusive mating types. So where did males and females come from?
The classic model developed by Britain’s Parker and colleagues in 1972 begins with gametes of similar size. An individual has finite resources for reproduction, creating a trade-off between “a few large gametes” and “many small gametes.” Larger zygotes might survive better, while smaller gametes might improve their odds of an encounter through sheer numbers.
Under particular conditions, the intermediate-size strategy becomes unstable, and the two extremes coexist. We need only look at ourselves to see how far that divergence can go. A human egg is about a tenth of a millimeter across, just barely visible to the naked eye. A sperm head measures only a few micrometers. Their volumes differ by more than ten thousandfold.
The biological definition is this: the type producing larger gametes is called female, the type producing smaller gametes male. The male–female division is thus a classification that follows the differentiation of gamete sizes.
There is also an elegant explanation for why males and females tend to be roughly equal in number. In 1930, the British statistician Fisher pointed out that if males become scarce in one generation, each male can leave more offspring on average. Parents producing more sons gain an advantage, and sons become more numerous. If females become scarce, the reverse happens. Again and again, the ratio is pulled back toward one to one.
Meiosis, recombination, and gamete fusion can all exist in systems with no male–female division at all. Even where males and females exist, the roles need not belong to separate individuals forever. In most flowering plants, a single flower contains both stamens and pistils; one plant acts as both father and mother. In a group of clownfish, the largest fish is female. When she dies, the largest male becomes female.
Volvox is a little green ball found in ponds: several thousand cells arranged into a hollow sphere, slowly rolling through the water. The green algae in the Volvox group provide an empirical system for investigating this, because they include closely related species with equal-sized gametes, unequal-sized gametes, and egg-and-sperm reproduction. A 2018 comparison found that the emergence of unequal gametes did not require a vast, entirely new sex-determining region. The relevant regions could actually be smaller.
Other work has found that a regulatory factor determining mating type in species with equal-sized gametes is associated with male determination in related species with unequal gametes. This is direct comparative evidence of genetic continuity between mating-type systems and the male–female systems that came later. There is another, more basic point: sexual and asexual reproduction are not an on–off switch.
Facultative sex is widespread. A common pattern is rapid asexual multiplication when resources are plentiful, with a greater chance of entering a sexual cycle under nutrient limitation, changes in density, or seasonal signals. The precise triggers, however, are highly specific to each group. “Worsening conditions trigger sex” cannot be taken as a universal rule.
Genetic analyses of wild yeast offer a reminder: in the wild, yeast undergoes meiosis roughly once every thousand generations. A sexual cycle can occur at an extraordinarily low frequency and still substantially alter long-term genetic structure. So “we keep observing asexual division under natural conditions” does not entail “this lineage never has sex.”
VIII You Cannot See Meiosis in a Fossil
When, then, was this institution established? Start with the genes. Those specialized parts are distributed across widely separated branches of eukaryotes, from animals, plants, and fungi to all sorts of single-celled protists.
The most economical explanation is that their common ancestor already possessed fairly complete meiotic machinery. What the genes can tell us is that the machinery existed no later than that common ancestor.
Fossils supply a different “no later than”: by a certain date, sex was already occurring in a particular organism. Both kinds of evidence say “at least this early.” The oldest possible limit can only be placed loosely after the eukaryotic branch split off. Neither can give us “the very first time.”
The fossil most often cited is Bangiomorpha, mentioned in the fifth essay. It comes from an island in the Canadian Arctic and dates to about 1.047 billion years ago. It consists of strings of cells attached to rock, looking very much like the red algae used to make nori today.
Its name means “shaped like Bangia,” a seaweed still eaten along the Fujian coast. When Cambridge University’s Butterfield described it in 2000, he found two different ways of producing spores, two forms of reproductive stage, in the same fossil assemblage. The whole arrangement matched the sexual life cycle of living red algae.
Red algal life cycles are difficult to recognize even today. Nori has a filamentous stage that grows by boring into seashells, and for a long time it was taken for a different alga. In 1949, the British phycologist Drew-Baker discovered the connection. Building on her work, Japanese growers learned to raise the young algae artificially, transforming nori cultivation. Some Japanese fishing villages still commemorate her every year.
This is much stronger than “seeing a round object and calling it an egg,” because it rests on correspondence across a whole life cycle. But the fossils show neither chromosome pairing nor crossing over nor two gametes fusing. They therefore provide some of the earliest and most widely accepted evidence of sexual reproduction, yet no photograph of meiosis itself.
A Chinese fossil makes the point from the other direction. In 2024, Chinese researchers redescribed a filamentous multicellular organism about 1.6 billion years old from the Yanshan region of northern China.
Some cells contain small round bodies ranging from a little over ten to twenty micrometers in diameter. The researchers compared them with the asexual spores of living algae and judged that the organism probably reproduced by spores. A reproductive structure is not automatically a sexual structure. To count this as sex 1.6 billion years ago merely because “there are reproductive bodies inside the cells” is to claim something the original paper does not support.
Later come spores in groups of four. From about 470 million years ago onward, fossil spores of early land plants often occur in orderly groups of four, closely matched in size. Meiosis divides one into exactly four. The mechanism predicts four; the fossils give us four. It is a beautiful correspondence.
But groups of four are not exclusive to meiosis. Other forms of division can also produce tetrads. Only when wall structure, the geometry of the arrangement, the relationship to sporangia, and the developmental patterns of living relatives all agree does the meiotic interpretation stand firm.
By about 400 million years ago, a chert deposit at a place in Scotland called Rhynie, discovered by a local doctor in 1912, had sealed a hot-spring wetland into stone just as it was. Plant stems and sporangia were preserved, along with the other generation in their life cycles, a tiny generation devoted to producing sperm and eggs. Here, at last, an entire sexual plant life cycle can be pieced together from fossils.
Living ferns still retain this alternation of generations. The rows of little brown dots beneath a fern frond are sacs holding spores. When a spore lands, it does not grow into a new fern. It produces a thin, green, heart-shaped sheet about the size of a fingernail. Sperm must swim across a film of water on it to reach an egg. Only after fertilization does the fern we recognize begin to grow.
Now return to this essay’s construct. In this series, a construct means drawing a line: what falls inside counts; what falls outside does not. Here, the line takes the form of a procedure. When the typical sexual cycle described here is entered, pairing, cutting, exchanging, halving, and fusion must be coordinated in a regulated sequence. This is not a change of level. The unit of selection has not changed; the individual remains the same individual.
It does something else: it turns genetic exchange, which could once happen by chance, into a compulsory, fixed procedure, with dedicated machinery, checkpoints, and bonds released in stages. Chance, institutionalized.
This typical procedure has requirements to meet and ways to fail. Homologue pairing, the making of breaks, the formation of crossovers, and the two-stage release of cohesion must work together. Errors at critical points can cause infertility or produce gametes with the wrong number of chromosomes.
Some of the miscarriages discussed in the first section involve chromosome abnormalities caused by meiotic errors; not every miscarriage can be laid at this procedure’s door. What institutionalization buys is not flexibility, but repeatability. Each time this kind of cycle is entered, the same rules shuffle the cards, with a different result. Yet repeatability itself opens a new way to fail: an asexual individual that bypasses this chromosome-pairing step cannot be made sterile by its failure; a sexual individual can.
What the construct cannot accommodate and cannot eliminate is what this series calls the remainder. In this essay, the remainder lies in two places.
One is that twofold deficit itself. It has not vanished. Something else merely makes up for it, and what that something is still depends on which group you ask about, how large its population is, and how much parasite pressure it faces. Every explanation can balance the books within its own parameter range. None can balance them across all ranges.
The other is the asexual lineages. They have never disappeared. They arise frequently; most are very young; a few have lasted more than a million generations. A very small number of candidates remain unresolved.
On the most famous scandal, both sides are now moving toward the middle. Most researchers think rotifers exchange genetic material. The dispute is whether their way of exchanging it is sex, while some also warn that part of the signal may come from sample contamination. Narrowed to the question of how the exchange happens, the disagreement has become a technical problem rather than a crisis of theory.
As for when this institution was established, fossils cannot show us meiosis. What they can show is reproductive structures. This point in history can therefore only be inferred backward from the parts shared by organisms alive today. Like the common ancestor in the second essay, its position is clear, its contents are not.