I Dividing Without Parting
You began as a single cell yourself. One fertilized egg became two cells, two became four. A little over forty rounds of doubling would produce enough cells to make a human being.
The cells in your body, though, did not go their separate ways after dividing. Usually, when one cell divides into two, each goes off on its own. Bacteria separate after division; so do paramecia. When your skin cells divide, their membranes separate, but the occupants stay put, riveted to their neighbors by specialized proteins. Sometimes, then, division happens without a parting. A laboratory can make this happen in only a few dozen days. We will come to that in Section VI.
It looks like a small change. Its consequences are enormous. Once cells cease reproducing separately, the object on which selection acts changes. Previously, one cell competed with another to leave more descendants. Now one group of cells competes with another group. That transition does not take care of itself.
A group of cells stuck together is not necessarily a reproductive unit. They must also reproduce together: the body of the next generation must arise from this body, rather than from a few of its cells wandering off independently to start again. And within that group, each individual cell still carries on making copies of itself.
This is why multicellularity is no small matter. It turns a unit with no internal competition into a unit with competition inside it. The question was never simply how cells could stick together. It was what to do about the competition among them once they had.
Imagine a body made of a thousand genetically identical cells. One cell now acquires a mutation that makes it reproduce faster. The mutation does nothing for the body as a whole; it may even harm it. Within that body, the mutant cell's descendants become more and more numerous.
The body as a whole, meanwhile, may reproduce less successfully as a result. Selection at the two levels pulls in opposite directions. This difficulty has a name everyone knows: cancer.
Cancer is what that problem looks like today. Some have even proposed that cancer cells revert to the way their single-celled ancestors lived: keep dividing, pay no attention to anyone else. The idea is still disputed. But it brings cancer and the origins of multicellularity into the same question.
Cancer cells are precisely such a population. They multiply faster inside a body, and the body that houses them dies as a result. Betrayal is hardly the word. A cancer cell does not know what it is doing. It simply reproduces faster than its neighbors.
And in a body assembled from many parts that can each reproduce, “faster than its neighbors” is enough. The very existence of multicellular organisms therefore tells us something: this conflict must be held below some threshold, or a body cannot keep going.
And it has never been eliminated. The conflict within a multicellular body is no historical problem settled long ago. It is a process that continues every day, held down to a sufficiently low level.
An adult human has a little over thirty trillion cells. Every day, hundreds of billions of old ones die and new ones take their place. With every division, a few letters are miscopied in the DNA. Even now, some cells in your body are probably reproducing against the rules. Most are being cleared away.
Notice the word “most.” The small gap between most and all is where this system still fails to close. For a cell to become cancerous, errors usually have to accumulate in several crucial genes. That takes time, which is why cancer occurs mainly in middle and old age. Over a human lifetime, the body's cells divide more than ten quadrillion times. Each division is an opportunity.
Some cancer cells have even left their original bodies behind. Dogs have a transmissible tumor that spreads from dog to dog through mating. The cells of every one of these tumors descend from a single dog that lived thousands of years ago.
That dog is long dead. A population of its cells is still alive in dogs around the world. Something comparable has happened in humans: cancer cells taken in 1951 from an American patient named Henrietta Lacks were grown in culture flasks. They are still dividing in laboratories around the world today.
The cells were taken without her consent. Tasmanian devils, too, have a transmissible cancer of the face. In a little over twenty years, it reduced their numbers by well over half.
The problem is ancient as well. In 2020, researchers confirmed a form of bone cancer in the leg bone of a horned dinosaur that lived more than seventy million years ago. Plants develop tumors too, but rarely die of them. Their cells are held in place by cell walls. They cannot escape and establish metastases.
Bodies have several ways of suppressing this rule-breaking. Section V will go through them. First, consider something else: cells have assembled into bodies many times in evolutionary history. How many times, exactly, is much harder to count than you might think.
II Twenty-five, Twenty-two, Sixteen, Forty-five
“Multicellularity arose independently more than twenty times” is a much-repeated statement. It is most often traced to two reviews: one from 2007 said “at least twenty-five times”; another, from 2011, said “at least twenty-two times” among eukaryotes. A review published in 2017 gave “at least sixteen eukaryotic lineages.”
In 2023, the Polish scholar Lamża set out to follow this chain of citations. He found that the “at least twenty-five” came with neither explicit inclusion criteria nor an itemized list. A figure in the same article explicitly depicted only about sixteen multicellular eukaryotic lineages.
The “at least twenty-two” likewise came without a complete list. An estimate containing the words “at least” was passed from paper to paper until, in secondary accounts, it became “scientists have counted.” “At least” means “no fewer than.” It does not mean “approximately equal to.”
The size of the number matters. It bears on a larger question: life beyond Earth. If cells arose on another planet, how difficult would it be for them to grow bodies?
If multicellularity appeared here dozens of times, that step is probably not very difficult. The hard steps come later. If it appeared only once or twice, it looks like a rare stroke of luck. If it appeared dozens of times, it looks more like a door that has always stood open.
He counted again himself, using the broadest definition: anything that spends most of its life outside the typical single-celled arrangement of “one nucleus inside one membrane” qualified. He examined ninety-one taxa and identified forty-five distinguishable independent cases.
He also divided those forty-five cases into six different kinds of “multicellularity.” Some consisted of multinucleate hyphae with partitions; some were clones that remained in a clump after dividing; some were temporary aggregations; some were networks formed by the partial fusion of many cells. He thought the true number might be substantially higher, perhaps approaching a hundred, because the records for many small protists are so sparse.
The figures differ so widely because nobody has settled what counts as one instance. It is rather like asking how many rivers China has. How long must a watercourse be to count as a river? Do tributaries count? What about channels that carry water only in the rainy season? The number means something only after those questions are settled. Mushrooms make a good example. The mushroom we eat is the “fruit” produced by an extensive underground web of hyphae.
Hyphae are slender tubes. Many nuclei float in a connected stretch of protoplasm, and the partitions between sections are often incomplete or absent. Whether this counts as multicellularity depends on whether you require “multiple nuclei” or “multiple cells separated by complete membranes.”
By the first criterion, it counts. By the second, it does not. A Mediterranean alga, Caulerpa taxifolia, grows parts resembling roots, stems, and leaves. A single specimen can reach around a meter in length, yet its interior is one continuous, unpartitioned tube, with many nuclei suspended in the same mass of protoplasm. Is that one cell, or a multicellular plant? And from mushrooms to molds, the main body of the fungal kingdom is made of hyphae. This one judgment can make whole swaths of fungi appear in the count or vanish from it.
Add more requirements and the number falls. If all you ask is that two or more cells remain joined over time, plenty of filamentous bacteria, yeast clusters, and algal colonies qualify. Require stable mutual dependence between the cells as well, and the set shrinks.
Insist on irreversible differentiation, a division of labor between reproductive and somatic cells, three-dimensional bodies, and developmental programs, and only a very few lineages remain. Revise the evolutionary tree and the count changes again. Some single-celled organisms have actually reverted from multicellular ancestors. Treat them as though they were always single-celled and you will count too many independent origins.
Textbooks often say that “complex multicellularity arose only about six times”: in animals, fungi, red algae, brown algae, green algae, and land plants. A genuine threshold lies behind this claim.
In a one-dimensional filament or a two-dimensional sheet, every cell still touches the water. In a three-dimensional clump, it does not. At the center of a solid ball of cells one millimeter across, a cell lies half a millimeter from the surface. Oxygen takes too long to diffuse that distance to keep pace with the rate at which cells consume it. Beyond a certain thickness, a three-dimensional body has to build a transport system of its own. Otherwise, its interior is dead.
But those six categories are not on the same level. Land plants are already nested within the green plants, while complex fungal fruiting bodies may have arisen independently in eight to eleven branches. Twenty-five, twenty-two, sixteen, forty-five: these are not four measurements of the same quantity.
III Bodies That Grew Separately
These independently evolved bodies can look very different from one another. Begin with the branch we know best: animals. Sponges, jellyfish, worms, fish, and humans all descend from the same multicellular ancestor. The animal body arose only once. Land plants also had a single origin, from one branch of aquatic green algae. The others each took their own route.
Kelp is one example. It belongs to the brown algae and is not closely related to land plants. It evolved a multicellular body of its own. In terms of kinship, kelp and Chinese cabbage are farther apart than humans and mushrooms.
Giant kelp off the California coast can grow half a meter in a day and reach forty or fifty meters in length, holding up an underwater forest. That forest is home to sea urchins that eat the kelp and sea otters that eat the urchins.
When otters become scarce, the urchins devour the kelp and the whole forest becomes a bare seafloor. Giant kelp has specialized cells that carry sugars made by photosynthesis down into its deeper parts. They closely resemble the conducting tissues of plants. The two sides arrived at the solution independently.
Nori and Gelidium belong to the red algae: another origin again. The kelp, wakame, and nori on the dinner table, and the cold jelly made by boiling Gelidium, represent two independent origins of multicellularity. Some red algae coat themselves in a hard shell of calcium carbonate, like a layer of pink lime, cementing pieces of coral reef together.
The fungi behind mushrooms represent another origin. In a forest, hyphae entwine with tree roots and connect one tree to another. Sugars and nutrients can travel some distance along them between trees.
How much this underground network matters is still debated. A mature giant puffball can contain several trillion spores. Step lightly on it and a puff of smoke rises.
Sometimes mushrooms spring up in a complete ring on the grass, a fairy ring in English, commonly called a mushroom circle in Chinese. Underground hyphae start from one point and spread outward year by year. Nutrients are most abundant at the advancing edge, so mushrooms emerge along the circumference. In the mountain forests of Oregon in the United States, a single connected web of honey-fungus hyphae spreads beneath nearly ten square kilometers of ground. Estimated to be thousands of years old, it is often described as one of the largest organisms on Earth.
Slime molds took another route. The yellow slime mold Physarum polycephalum is one gigantic cell containing thousands upon thousands of nuclei. It crawls over rotting wood in the shape of a net.
Japanese researchers put it in a maze with oat flakes at either end. It drew its body into the shortest line connecting the two. In 2010, they arranged oat flakes to match the positions of cities around Tokyo. The network the slime mold grew was strikingly similar to the region's rail system.
It can also “remember the way.” It leaves a layer of slime wherever it has crawled; when it encounters that slime again, it turns aside rather than retracing its steps. Whether this counts as a cell or a body is precisely the question from the previous section.
Cellular slime molds have yet another way of coming together. One, Dictyostelium discoideum, usually lives as separate amoebae eating bacteria in the soil. When food runs out, thousands upon thousands crawl toward the same spot and assemble into a moving “slug” a few millimeters long. It crawls toward light and warmth. When it reaches a suitable place, it raises a slender stalk with a mass of spores at the top. Section IV will look more closely at that stalk.
Then there are bacteria. A group called the myxobacteria also aggregates when food runs out, forming tiny “fruits” just visible to the naked eye. Some look like little trees, with spores inside.
Cyanobacteria form strings of cells, some specialized for fixing nitrogen, as we saw in Part IV. The streptomycetes common in soil grow branching hyphae. Much of the smell of earth after rain comes from a molecule they release. They also raise chains of spores, and more than half the natural antibiotics we use were first discovered in this group of bacteria.
There are also the closest single-celled relatives of animals, the choanoflagellates. Each cell has a flagellum at its tip surrounded by a “collar,” making it look remarkably like the cells that filter water in a sponge. Some choanoflagellates form colonies resembling tiny rosettes. They will return in Section VII.
The ways these bodies form fall into two broad groups. Of the forty-five cases, about thirty-three involve a cell dividing without its descendants separating, and about twelve involve free-living cells gathering together. The difference between these two methods is much greater than it looks.
IV Coming Together and Staying Together
There are two routes to a body, and their difference is not a matter of shape. In clonal formation, daughter cells fail to separate completely after division. Every cell traces back to the same starting point. That is how you were made: all the cells in your body descend from the same fertilized egg.
In aggregative formation, independent, free-living cells exist first. An environmental signal then triggers them to move and assemble. These cells come from different places and may be genetically entirely different.
In 2013, researchers summed up the two routes as “staying together” and “coming together.” The fundamental difference is how closely the members of the group are related, what researchers call within-group relatedness.
This determines whether, when a cell makes a change that benefits itself but harms the whole, the consequences fall on descendants sharing its own genetic lineage.
Think of it this way. In a body that grows from one cell, every cell is descended from that same cell: a large family of siblings. An assembled group may contain cells from entirely different sources: a crowd of strangers brought together for the occasion.
In both cases, someone may shirk. The consequences are quite different. Among siblings, whatever advantage the shirker gains ultimately goes to a set of cells almost identical to itself. Among strangers, the benefit is pure gain.
Bodies that assemble from separate cells can do remarkable things too. In 1907, the American zoologist Wilson squeezed a piece of sponge through fine cloth, breaking it down into loose cells. A few days later, those cells had crawled together and grown into a little sponge again. But not everyone who comes back together necessarily belongs to the original company.
Even the human body has outsiders living in it. During pregnancy, some fetal cells cross the placenta into the mother's body and remain there for decades. Researchers have found a child's cells in a mother's blood, skin, and even brain.
In an aggregative system, then, shirking offers much greater rewards. That directly determines whether a highly developed division of labor can evolve.
A 2013 phylogenetic comparison of 168 species found that nine of twelve lineages with clonal group formation had evolved obligately multicellular life cycles. All five potentially aggregative lineages in the study remained facultatively multicellular. Clonal formation was also associated with more cell types and with the emergence of cells that do not reproduce.
Clones still have their internal conflicts. Mutation, chimerism, and somatic selection all generate differences anew. What clonal formation reduces is the scope for conflict at the starting point.
There is another fact: today's largest and most structurally complex multicellular organisms, giant sequoias, blue whales, and humans, all have life cycles that preserve continuity of descent after cell division. Within-group selection in aggregative systems, meanwhile, has been observed directly.
Dictyostelium discoideum can even “farm.” In 2011, researchers found that some of these amoebae carry a small supply of bacteria with them when they aggregate and migrate. On reaching a new place, they sow the bacteria first, then wait for them to multiply before eating them.
In cellular slime molds, large numbers of cells aggregate after starvation. About one-fifth eventually form a stalk that will never reproduce; the rest become spores. The cells in the stalk leave no descendants at all. They turn themselves into a pole that lifts the spores higher. A little extra height makes the spores easier for the wind or a passing animal to carry away.
So this is not waste. But for that one-fifth, it is the end. When different clones are mixed, some genotypes make up a larger share of the final spores than they did of the initial mixture. They have gained an advantage in the question of who becomes stalk and who becomes spore.
In other words, some of the cells that join the group are less likely to become stalk and more likely to become spores. None of this requires any “intention.” All it takes is for certain cells to be more readily pushed toward the spore side during sorting. A slight difference in their surface properties will do.
Fruiting bodies in the wild, however, usually have very high relatedness. This is interpreted as an important condition for limiting such competition. Under natural conditions, in other words, most of the cells that come together are already kin.
The myxobacteria that make the tiny “fruits” in Section III move in packs even in ordinary life. Like wolves, they surround other bacteria, release digestive enzymes, and feed together. They have furnished an even more clear-cut experiment.
In 2000, researchers evolved a population for about a thousand generations in a nutrient-rich environment, obtaining several strains whose ability to form fruiting bodies had declined. When some of those strains were mixed with their ancestor and induced to develop, they made up a larger share of the resulting spores than of the original inoculum.
Here the point is directly visible: the relative reproductive success of a cell or strain within a mixed fruiting body need not align with the functioning of the developmental structure as a whole. Notice the conditions under which selection took place. Those thousand generations were spent in a nutrient-rich environment. Under those conditions, the capacity to make fruiting bodies was not rewarded.
The experiment therefore does not show that aggregative systems must fall apart. It shows that, so long as a channel for within-group selection remains open, unrewarded group functions deteriorate, while the deteriorated strains gain an advantage in a mixture.
V Conflict Is Suppressed, Not Solved
The precise form of the problem is that selection can act on two levels at once. The overall change in a multicellular population can be separated into two components. One is the difference between groups: which bodies leave more descendants.
The other is the difference between cells within the same body: which cells make more copies of themselves there. If a cellular trait increases the cell's own share of the body while reducing the body's overall reproductive output, those two components point in opposite directions.
A cell lineage that reproduces faster and contributes less to tissue function needs no intention. As long as its relative advantage at the cellular level is greater, it will increase within the body. This separation also makes something else clear: internal conflict is a structural property of the multicellular system. As long as a body consists of parts that can reproduce independently, the conflict is there.
Whenever a reproductive unit consists of multiple parts that can each make copies of themselves, both components are present. Suppressing one of them is additional work that the developmental system has to do. There are several sets of mechanisms for doing it, and each has openings through which something can escape.
The first is the single-cell bottleneck. Each generation of animals, and of many algae and plants, begins as one cell. Why does that matter? If a new generation starts from a large clump of cells, all the cellular differences accumulated within the previous body will be carried into the next one intact. Start from a single cell and those differences are reset to zero.
What this reduces is the internal variation present at the beginning of one new body, not necessarily the long-term variation of the species as a whole. New mutations produce new differences in every generation. A bottleneck also means that the entire large body must repeatedly be rebuilt from very few cells, increasing the importance of chance.
A body that starts from one cell has to build everything the previous generation accumulated all over again. It is a reset. It is also another run through all the risks.
The second mechanism is early separation of reproductive cells from somatic cells. Very early in development, animals set aside a small group of cells specifically to produce the next generation. All the others, skin, muscle, nerve, and blood cells, take no part in heredity.
This cuts off the passage of the vast majority of somatic variants from one generation to the next. Even if a late-arising somatic variant spreads within the body, it usually cannot become an ancestor of the next generation. The consequence is that most cells become hereditary endpoints. Beneficial changes that arise in somatic cells, too, usually cannot be inherited.
Plants do not draw so strict a line. Any meristem in a tree may grow into flowers, and flowers produce gametes. Yet plants maintain complex bodies all the same.
In 2022, researchers used precisely these exceptions to argue that a single-cell bottleneck and an early germline are not logically necessary conditions for complex multicellularity. The counterexample matters. It tells us that the five sets of mechanisms are not a compulsory checklist. They are a range of available methods.
Different lineages use different combinations. All of them have held conflict low enough to keep going.
The third mechanism is cellular recognition and exclusion. A highly polymorphic recognition system in cellular slime molds enables cells with compatible markers to recognize one another during development. Incompatible cells sort apart.
The opening here is equally clear. A recognition system can block only the differences it can recognize. It cannot prevent new mutations that arise later within the same clone. Strong exclusion may also reduce the number of cells that could otherwise join to form a larger structure. What recognition changes, then, is the probability of outsiders getting in.
The fourth mechanism is programmed cell death. Cells carry a self-destruct program. A cell that detects serious damage to its DNA, or finds itself somewhere it ought not to be, can activate that program itself.
But cell death has a cost of its own. Tissue loses cells; regeneration and replacement consume resources. Set the threshold too low, and damaged but repairable cells are lost as well. Set it too high, and abnormal lineages remain.
Selection therefore acts on the trade-off among repair, arrest, and removal. The bowhead whale, which can live for more than two hundred years, makes the point especially well. A 2025 study found that it is exceptionally good at repairing breaks in DNA. That may be one reason for its long life.
The body also has patrols. One class of immune cells specializes in looking for cells with abnormal surfaces. When it finds them, it kills them.
The fifth mechanism is centralized control of proliferation. The question “when should a cell divide?” ceases to be governed solely by that cell's local nutritional state and becomes part of a body-level developmental program.
The volvocine algae offer an example. Between single-celled Chlamydomonas and multicellular Gonium, a pathway regulating the cell cycle was repurposed for division in groups. Introducing the relevant regulator from Gonium into Chlamydomonas can induce a colonial form.
Now return to cancer, because cancer is evidence that this problem has never disappeared from the structure itself. The important point is that a body equipped with a single-cell bottleneck, early germline separation, recognition systems, programmed death, and centralized control of proliferation still repeatedly generates cell lineages that violate these constraints from within.
All five sets of mechanisms are operating, and the problem has not been closed. Nor do the mechanisms themselves come free. Bottlenecks discard variation; germline separation makes most cells hereditary dead ends; recognition may reject partners that could have cooperated; death consumes tissue; centralized control requires a whole signaling system.
In the 1970s, the British epidemiologist Peto posed a question that became known as Peto's paradox. Humans have thousands of times as many cells as mice and live dozens of times longer. Why is their risk of cancer no higher? Intuitively, more cells and more years should mean more chances for something to go wrong. A blue whale has thousands of times as many cells as a human. By that reasoning, it should develop cancer very early in life. It does not.
At least, not in proportion to its number of cells. Modern data have carried the question beyond anecdote into cross-species statistics. Roughly ten to twenty-five percent of humans die of cancer over a lifetime; among zoo elephants, the figure is below five percent. In 2022, researchers assembled zoo records for 110,148 individuals from 191 mammal species. They found that cancer mortality risk did not simply increase with adult body mass and adult life expectancy.
Another study that year compared somatic mutations in sixteen mammal species. Annual mutation rates were strongly inversely related to lifespan, while body size alone could not simply predict cancer risk.
Across species, cell number and lifespan fall far short of predicting cancer burden through the straight-line arithmetic of a naive independent-risk model. Large, long-lived lineages must therefore possess additional suppressive mechanisms. And they do not all use the same ones.
Elephants carry around twenty copies of a tumor-suppressor gene that tells a cell to stop or kill itself when DNA damage is detected; humans have only one. Naked mole-rats are only mouse-sized and live in large underground colonies, rather like ants, with just one breeding “queen” per colony. Yet they can live for more than thirty years and rarely develop cancer. Research on them has proposed another mechanism: an exceptionally long form of hyaluronic acid, the molecule familiar from skincare products, contributes to their strong contact inhibition.
Bowhead whales are different again: their strength is efficient repair. Different large bodies have evolved constraints at different levels.
Development and life history reduce the frequency of this conflict, the extent of its spread, and its opportunities for transmission between generations to levels at which the multicellular reproductive unit can persist.
When a problem will not close, the usual way forward is often not to solve it, but to reduce it to a level you can carry with you.
VI Sixty Days in a Laboratory
In 2012, Ratcliff and colleagues in the United States performed an experiment to measure directly how long this would take. They used the yeast familiar from breadmaking and brewing, which normally grows as separate cells.
Their method was very simple. After each day's growth, they let the test tube stand briefly, then transferred only the material that settled to the bottom into a new tube. The things that settled fastest were the larger ones.
The entire apparatus was a test tube and a little patience. Yeast had another convenience: it grows quickly, passing through several generations a day, and it can be frozen and revived.
The researchers froze a sample at intervals as the experiment progressed. Later, they could bring out an “ancestor” and compare it directly with its descendants. There was no genetic engineering, no special environment, no added signaling molecule. Each day, they simply selected the fraction that settled fastest.
Getting bigger thus became the only thing rewarded. The resulting yeast clusters branched and branched again, looking under the microscope like snowflakes. The researchers called them snowflake yeast. Across two independent experiments, all fifteen replicate populations developed a clustered form within sixty days.
The clusters formed not by free cells gathering again, but by mother and daughter cells failing to separate normally after budding. The result was a branching clonal cluster. Each cluster therefore began with extremely high genetic relatedness. Later research linked this form to impaired function of a gene involved in the daughter-cell separation program.
In other words, the step that joins cells into a group can be achieved by switching off a separation program. It need not require the invention of a new adhesion system. This directly weakens the intuition that multicellularity must be difficult to evolve: at least the first step can be easy. The long-term version went much further.
A later experiment continued through six hundred rounds of selection, about three thousand generations, and was published in 2023. The researchers named it the Multicellularity Long-Term Evolution Experiment and intended to keep it running indefinitely. Their model was the American long-term evolution experiment with E. coli that began in 1988 and continues today. Those bacteria have passed through more than seventy thousand generations. In one lineage, after more than thirty thousand generations, they suddenly learned to eat citrate in the growth medium, something their ancestors could not use.
In the treatment where cells could not use oxygen for respiration, and therefore escaped the internal penalty of oxygen diffusion, all five replicate populations became visible to the naked eye. Mean cluster radius increased from about sixteen micrometers to about 434 micrometers, overall size increased roughly twenty-thousandfold, and physical toughness increased roughly ten-thousandfold.
From needing a microscope to being visible without one. All within three thousand generations, with no foreign genes. The crucial change was in the shape of the cells.
The cells became more elongated, making branches more likely to entangle. A fracture now required more cellular connections to break at once. Large clusters could therefore withstand greater mechanical stress.
A change in shape at the cellular level had produced a new property at the group level. The oxygen conditions themselves also showed why the results could not simply be extrapolated.
A 2021 experiment found that intermediate oxygen levels actually restricted large clusters, because the cells inside could not obtain enough oxygen. By about eight hundred generations, the different oxygen treatments had produced markedly different upper limits on size.
There have been at least two versions of the experiment that pairs green algae with a predator.
In 2019, researchers used a ciliate to prey on populations of Chlamydomonas produced by crossing strains. The logic was straightforward. A predator's mouth is only so large. A few cells in a clump can become too big to swallow. In two of the five populations exposed to predation, a heritable multicellular life cycle absent from the controls emerged within about 750 asexual generations.
In 2021, another team used rotifers, relatives of the tiny animals in Part VI that can dry out and drift away, to prey on ten genetically identical lines of green algae. After about five hundred generations, they obtained a stable, heritable colonial growth form. They also detected parallel genetic changes shared by all the colonial isolates.
“Reversible” needs to be separated into two meanings. Some evolved strains produce smaller, even single-celled reproductive stages during their life cycle. Multicellular and single-celled stages can therefore alternate within one inherited life cycle. That does not mean that removing the predator immediately reverses the genotype.
These experiments provide strong evidence for one conclusion: simple groups need not take a long time to form.
What they did not simultaneously produce were stable lineages with multiple cell types, complex spatial patterns, long-distance transport systems, finely tuned position-dependent signaling, and complete developmental programs for reproductive and somatic cells.
After roughly three thousand generations and a twenty-thousandfold change in size, researchers still explicitly use snowflake yeast as a system of “undifferentiated multicellularity.” That is experimental evidence in itself that “millimeter scale” and “complex multicellularity” are not synonyms. Selection itself imposes a further limit on what the experiments can tell us.
The criterion in these experiments is set by the researcher, and it usually has only one dimension: settle quickly, or avoid being eaten. It has the same shape as the test tube at the beginning of Part III, which measured only the speed of reproduction. Whatever the selection criterion rewards is what remains.
Selection in nature operates along many dimensions at once: resources, diffusion, predation, disease, space, reproduction, seasons. So “cell clusters obtained within tens to hundreds of generations” supports the claim that the threshold for group formation can be low. It cannot tell us how long a complex developmental system would take to evolve.
Have the experiments discussed above, then, demonstrated the new evolution of stable, heritable cell-type differentiation from a single-celled ancestor? The results considered here do not yet establish that.
Something else interesting appeared in snowflake yeast. As clusters grew larger, some of their cells died first. The dead cells became breaking points. A cluster snapped there, releasing a smaller new cluster. Death was being put to work in reproduction.
Snowflake yeast has evolved position-dependent death and clear functional differences. These can be called a rudimentary division of labor. It still has not produced two genetically identical cell types with fixed fates, repeatedly generated by a stable developmental program, as in Volvox.
VII A Ladder Is a Teaching Aid, Not a Lineage
Volvox is a tiny green ball found in ponds, just visible to the naked eye. In 1700, the Dutch microscopist Leeuwenhoek first watched it turn slowly through water under a microscope he had made himself. Inside the ball were smaller balls of the next generation. Volvox and its relatives do offer a striking range of forms.
Single-celled Chlamydomonas. Gonium, usually eight or sixteen cells arranged in a little flat plate, without stable differentiation. Pandorina, sixteen cells packed into a small ball. Eudorina, thirty-two cells arranged as a hollow sphere.
Another genus has colonies usually numbering from a few dozen to more than a hundred cells; some species already divide labor between front and back. Volvox itself has around two thousand small, terminally differentiated somatic cells and about sixteen large reproductive cells.
The two thousand small cells bear flagella and do the paddling. They spend their lives at it, then die. Only the dozen or so large cells can produce the next generation. This is one of the simplest forms of setting reproductive cells apart. A Volvox sphere often holds a dozen or so daughter spheres. The parent eventually breaks open, and the daughters swim out and grow on their own. During development, a Volvox embryo also turns its entire body inside out, like a sock, so that its flagella face outward.
Arrange these forms in a row and they look like a ladder from simple to complex. Textbooks often draw them that way. But they cannot be arranged as a ladder of ancestors and descendants.
Every one of them is a living branch tip. Each continued to evolve after diverging from a shared ancestor. A molecular-clock analysis in 2009 placed the early divergence of multicellular volvocine algae at least around two hundred million years ago. The simple colonies alive today are not ancient ancestors frozen in place.
Making a ladder out of them is like lining up a mouse, a dog, and a human alive today and calling that the course of mammalian evolution. Even the number of origins has been revised as the evolutionary tree has changed.
Older frameworks tended to treat undifferentiated multicellularity as having one principal origin. A phylogenomic tree built from transcriptomes in 2021 gave a different result: multicellularity arose independently at least twice, and one traditionally recognized family was not monophyletic. Even “volvocine multicellularity had only one origin” is not a settled fact. The genomic findings are more counterintuitive still.
A 2010 comparison of the Volvox and Chlamydomonas genomes had already found them far more similar overall than their differences in form might suggest.
When the genome of undifferentiated Gonium was added in 2016, researchers detected a net gain of only about 110 gene families and nine new protein domains at the node where multicellularity appeared. The changes were far smaller than at earlier nodes. One hundred and ten gene families is a surprisingly small number for a transition “from one cell to many.”
The clearest example is that cell-cycle regulatory pathway. It already existed in single-celled Chlamydomonas. What Gonium changed was its regulation. The parts were there first. Their use changed.
Put these findings together: when a new level of organization appears, the number of newly added gene families may be small, while the regulatory relationships, copy numbers, and uses of existing families change. The intuition that “multicellularity requires a whole new set of genes” is therefore not supported in this group.
On the animal side, the picture is similar and points toward the same conclusion.
A choanoflagellate sequenced in 2008, one of the closest known single-celled relatives of animals, had about 9,200 genes. Among them were genes for many cell-adhesion and signaling proteins previously known mainly from animals. Another single-celled relative within the broader animal kinship group supplied further parts of the toolkit.
But not every part of the toolkit was already present. The same body of work also emphasized that many animal cell-to-cell signaling pathways and extensive developmental regulatory networks expanded on the animal branch itself.
Sponges are among the simplest animals. They have no nerves, no muscles, no true organs; their bodies are full of little pores. Water enters through the pores. Inside, the flagella of a layer of collar cells beat, filtering bacteria out of the water to eat. Choanoflagellates look remarkably like these cells. That resemblance was the original reason for considering them close relatives of animals.
One particularly clear case comes from the King laboratory at Berkeley in the United States.
In the absence of a particular bacterial signal, one species of choanoflagellate can live as single cells or linear chains. Stimulate it with molecules produced by a bacterium it eats, and successive cell divisions form stable, spherical colonies.
The discovery was accidental. At first, the researchers could not get the little rosettes to grow. They eventually found that the key was a bacterium that had entered the culture medium. The bacterium supplied only a signal. The capacities for cell adhesion, polarity, repeated division, and the making of form belonged to the choanoflagellate itself. So the claim that “most of the molecular toolkit for multicellularity predates multicellularity” can stand, but it must be separated into two levels.
The first concerns the age of the protein components. Adhesion proteins, signaling proteins, and many classes of transcription factors all have homologs in single-celled relatives of animals.
The second concerns the age of the regulatory combinations. These ancient proteins were already used in combinations during different stages of single-celled life. One important change in animal multicellularity was the reconnection of those existing modules into regulatory networks specific to place, time, and cell type. But the existence of a toolkit is not the existence of a system.
Single-celled organisms, too, need to attach to a surface, recognize food, detect neighbors, and switch between ways of living.
An ancestral version of a protein later used to bind cell to cell might therefore once have bound a cell to a surface. The adhesion system of that single-celled relative already helps it attach to a surface in its single-celled state. That is the most direct functional example.
So “animal adhesion proteins were already present in single-celled organisms” can sound as though multicellularity was all prepared and waiting. What it actually says is that the protein was doing something else at the time.
What is new at the level of the system is this: a unified life cycle must continually call on these parts, so that one cell's state reliably affects another's and different positions repeatedly produce different fates.
VIII Those That Did Not Go Far
First, consider how much fossils let us see. The strongest grounds for identifying a fossil as a multicellular individual are visible, genuine cell boundaries; regular repetition of neighboring cells; repeatable differences in cell type at different positions; a consistent direction of growth; and specialized attachment structures.
Yet every one of these features can be partly imitated. Large prokaryotic filaments may have cross-walls. A single multinucleate cell can branch. Great size may belong to a giant single cell, like the nearly two-centimeter sulfur bacterium in Part V. Several criteria must therefore hold together and agree with one another before the case is secure.
The North China fossils mentioned in Part VI carry a different weight here. Reported in 2024 and dated to about 1.635 billion years ago, these filaments contain directly preserved cells up to about 190 micrometers across. The evidence for early multicellularity moves from “this object is large and regular” to “we can see the individual cells that make it up.”
Seeing cells depends on how they were preserved. A carbonaceous compression is a film of carbon left after an organism has been flattened. The cells' three-dimensional structure is lost in the flattening.
Phosphatization and silicification, by contrast, “cast” the structure before it decays. The record of early multicellularity therefore depends partly on what organisms existed and partly on which windows of preservation happened to be open.
Now look at the present. The title says most did not go far. If “far” means more cell types, thicker bodies, and deeper levels of development, that is true: of the forty-five independent cases, only a few became three-dimensional, organized bodies with many cell types. Most remained filaments, hyphae, sheets, clonal groups, or temporary aggregates.
Change the ruler, however, and the ranking changes. In 2010, researchers in Aarhus, Denmark, noticed something strange. Oxygen was being consumed at the surface of seafloor mud while hydrogen sulfide was being consumed one or two centimeters below. The two processes were astonishingly synchronized, as though a wire connected them. In 2012, they found the “wire”: a cable bacterium, a slender filament of many cells joined end to end.
One end is buried deep in oxygen-poor mud; the other reaches the oxygenated surface. Electrons travel along the filament from one end to the other, picked up in one place and handed over in another, like a living electrical wire. The conductors lie beneath the cells' outer wall: bundles of parallel, electrically conductive fibers.
A single cell cannot span that distance, so it has to be a filament. One filament can reach several centimeters in length and contain more than ten thousand cells. Beneath a single square centimeter of seafloor, the combined cables can stretch as far as two kilometers.
They have been found in marine salt marshes, mangroves, lakes, streams, and groundwater, connecting the chemistry of deep sediments to that of the surface. They have no differentiation of cell types, no tissues, no organs. Ranked by number of cell types, they sit at the bottom. Ranked by how much seafloor chemistry they alter, they do not.
A paramecium has only one cell, yet it has an oral groove, food vacuoles, contractile vacuoles, two kinds of nuclei, and a whole apparatus of cilia beating in coordination. Count its cell types and the answer is one. Look at its internal structure and it is anything but simple.
The limestone of the Egyptian pyramids is full of fossil foraminifera called Nummulites. Each is a shell built by a single-celled organism; the larger ones measure several centimeters. In the Mediterranean, an alga called Acetabularia grows a few centimeters tall, like a little open parasol. The entire thing is one cell. In the 1930s, the German researcher Hämmerling cut the cap off one species and substituted a nucleus from another. The new cap took its form from the nucleus, demonstrating that the nucleus directs shape.
Measurable aspects of complexity include, at the very least, the number of cell types, body size, levels of organization, the scale of gene-regulatory networks, and ecological connections. These quantities do not all rise together. Humans have about twenty thousand protein-coding genes, roughly as many as a nematode one millimeter long.
An onion's genome is about five times the size of ours. In 1991, the American paleontologist McShea pointed out that the widely shared intuition that complexity generally increases over the course of evolution had scarcely been tested quantitatively.
Nor does going far mean traveling a one-way road. Among fungi, the single-celled way of life seen in yeasts has reappeared independently in several branches. In 2014, researchers found parallel changes to an ancient regulatory network behind this return. There was no need to reinvent a set of genes for single-celled life.
The same can be seen in experiments. In 2019, researchers repeatedly selected the fraction that stayed suspended from populations of clumping yeast, producing descendants that reproduced as single cells again. A follow-up in 2025 found that, when settling selection was applied once more, all of them grew back into multicellular clumps, although at very different speeds.
Complex multicellularity is certainly distinctive. Three-dimensional tissues, long-distance transport, signals keyed to position, and stable cell types really do occur in only a few lineages. But “a few” is not, in itself, a judgment on all the others. Rarity and superiority are two different things.
Now return to this essay's construct. In this series, a construct is a line drawn: what lies inside counts; what lies outside does not. Here, the line runs around the edge of the body. The cells inside reproduce together and undergo selection together. Those outside go their separate ways.
The unit of selection has shifted up another level, from one cell to a collection of cells. The “once” in Part V concerned the basic organization shared by living eukaryotes, not a claim that every kind of cellular incorporation occurred only once. Multicellularity, by contrast, arose dozens of times across different lineages.
Snowflake yeast did it in sixty days. Green algae did it in 500 to 750 generations. In nature, by the broadest definition, it has happened at least forty-five times, perhaps close to a hundred. Raising the unit of selection from the cell to the cell group is, in itself, a low-threshold step. What is rare is further organization after that shift: fixing cell types, establishing developmental programs, and suppressing internal conflict enough for the whole to persist over the long term.
And that conflict has still not closed. As Section I said, even now some cells in your body are reproducing against the rules. Most are cleared away. The small gap between most and all is this system's first remainder. What the construct cannot contain and cannot eliminate is what this series calls a remainder.
The second remainder is fully visible. All those dozens of origins that did not lead to further organization are still there. They have names, distributions, biomass, ecological uses, and places on the evolutionary tree.
We call them things that “did not go far” not because they are invisible, but because the ruler we use to line them up, complexity, has no natural unit. For those said to have “not gone far,” that judgment applies only along the particular axis we chose. Choose another axis, such as “how many kinds of sedimentary environment they inhabit” or “how much material they move through an ecosystem,” and the ranking changes completely.
The origins are still being recounted, and every recount must first redefine what counts as one. The ruler is still unsettled.