The Many Origins

Chapter 4 — The Trap

STATUS: v0 (all six parts drafted) Documentary parallel: chapters/04-the-trap.md (Beats 4.1–4.12) Last updated: 2026-05-22

What this chapter does

Chapter 1 left a loose thread. When we walked past the cyanobacterial filaments, one cell in every ten or fifteen along each filament was described as a heterocyst: a permanently committed nitrogen-fixing cell, terminally differentiated, unable to divide, unable to go back. We said this was the oldest known case of multicellularity with that kind of irreversible specialisation. We did not say why a lineage would lock its cells into a one-way commitment.

This chapter answers that. It introduces the framework biologists use for why multicellularity becomes a trap in some lineages and stays part-time in others. The framework is called the ratchet. It walks through the only experimental system where the locking-in process has been watched in real time. It draws out a reading of the empirical pattern that this project takes responsibility for, while flagging which parts are settled in the field and which are this project’s integration of them. And it gives serious time to the lineages where the door swung back the other way.

By the end of this chapter, you should understand:

  1. What an evolutionary ratchet actually is at the level of mechanism, and where the metaphor stops carrying weight and what is doing the work underneath.
  2. The two flavours biologists distinguish (Type 1, which helps the group and hurts the solo cell, and Type 2, which makes a viable revertant unreachable), and why each is favoured by ordinary selection one mutation at a time.
  3. What an experiment at Georgia Tech, running daily for a decade, has now shown about ratchets in real time.
  4. Why the same experimental setup, with the same yeast strain, locks the door for clonal populations and never locks it for aggregative ones.
  5. The project’s reading of why multicellularity has only ever locked in on clonal lineages, and where this is the project’s integration rather than the field’s settled view.
  6. Two worked cases of locked-in multicellularity at very different ages: cyanobacterial heterocysts (~2.5 billion years old) and Volvox (~200 million years young).
  7. The cases where the door swung back: yeasts derived from filamentous fungi, Helicosporidium derived from green algae, Gloeocapsopsis derived from filamentous cyanobacteria, snowflake yeast in the lab under reversed selection.
  8. What ratchets explain (why obligate multicellularity persists) and what they don’t (why it ever starts).

The chapter is in six parts.

PartTitleApprox. length
1What a ratchet is, and the two ways one can form~1100 words
2A ratchet caught on camera — the snowflake-yeast experiment~1600 words
3Clonal versus aggregative, run head to head~1100 words
4The project’s reading — why only clonal lineages have locked in~1400 words
5Two worked cases — heterocysts and Volvox~1400 words
6The door swings both ways, and what ratchets don’t explain~1300 words

Part 1 — What a ratchet is, and the two ways one can form

A ratchet, in a workshop, is the gear inside a wrench or a winch. It has angled teeth and a small spring-loaded catch (a pawl) that rides over the teeth one way and snags them the other. Push the handle one way, the gear turns. Push it the other way, the pawl catches. The motion is biased: easy forward, blocked back.

Biologists borrow the word for any evolutionary situation with the same asymmetry. A lineage changes in some direction, and once it has, the small changes that helped get it there make returning worse than carrying on. Reversion is not forbidden. On the available timescales, on the available evidence, it almost never happens.

The metaphor invites trouble, so keep one thing clear. There is no actual ratchet inside the cell, no mechanical part letting evolution turn one way and not the other. What there is, instead, is a population of cells in which random mutations occur every generation, and natural selection keeps the ones whose carriers leave more descendants. The ratchet word labels that process when its cumulative effect is one-directional: an accumulation of small mutations whose individual benefits make them stick, and whose combined effect over many generations is to make a lineage’s prior state biologically infeasible. No cell decides to commit. No lineage strives toward a one-way door. Mutations occur; some are favoured by selection in the current generation; over time the favoured ones add up; and the cumulative result is a structure, a genome, or a life cycle that a single cell pulled out of the body could no longer become.

Ratchet is a labelled metaphor for the cumulative one-direction effect of ordinary mutation and selection on a multicellular lineage’s ability to revert.

Two flavours

The biologists who developed this framework, most prominently the theoretical-evolution group led by Eric Libby and William Ratcliff, argue that two kinds of mutation can build a ratchet, and that the two kinds reinforce each other [Libby & Ratcliff 2014; Libby et al. 2016].

The first kind they call Type 1. A Type 1 mutation helps the cell’s fitness inside a multicellular body but hurts it outside one. Picture a bacterial cell that has been living in a clonal filament for many generations, with neighbours producing a defensive coat around the whole structure. A mutation that lets this cell stop investing in its own private defence is favoured: the group’s coat protects it anyway, and the cell can spend the energy elsewhere. Now imagine that cell breaks away and tries to live alone. The defence is gone, and the cell no longer makes its own. It is now a poorly defended single cell, slower-growing or more vulnerable than its single-celled cousins outside the lineage. The mutation that helped it inside the group is the same mutation that hurts it outside. A would-be revertant pays a cost. Stack up enough Type 1 mutations and revertants are reliably outcompeted by cells that stayed put.

The second kind is Type 2. A Type 2 mutation reduces the probability that any future mutation could produce a viable revertant. It does this by removing or breaking the molecular machinery a free-living single cell would need, rather than by punishing the cell that leaves. The clearest example is a lineage that loses the regulatory wiring for a single-celled life-history stage. Many bacteria, when isolated and starving, switch on a stress response that lets them survive a long dormant phase; some encyst, some shut down their metabolism almost completely. If a multicellular lineage has been growing happily inside a body for many millions of years, the genes for that single-cell starvation programme are no longer in use. Selection no longer keeps them sharp; mutations accumulate; eventually the programme is broken. A cell that escapes the body and tries to live alone now has neither the defence that group-living provided nor the toolkit a normal single cell would use to survive starvation. The road back is not blocked by a punishment. It is blocked because the destination no longer exists in this lineage’s genome.

The mechanism does the work, not the names. A Type 1 mutation is favoured now, in the current generation, because the cell currently lives in a group. A Type 2 mutation is strictly neutral now, because the broken machinery isn’t being used anyway. Both are ordinary outcomes of ordinary genetics. Neither requires evolution to “know” anything about future revertants.

The two flavours feed each other. Once Type 1 mutations have made a single-celled revertant unfit, the regulatory wiring for the single-celled lifestyle stops being maintained: there are no revertants for selection to act on, and the next Type 2 mutation to hit those genes is no longer purged. Each round of one type creates the conditions for the next round of the other. This is the synergy Libby and colleagues build into their models [Libby et al. 2016]. The picture is a slow, mutually reinforcing accumulation of small changes whose combined effect is the asymmetry the metaphor labels.

Why this matters

The shift to keep from this part is from barrier to accumulation. The popular story sometimes describes irreversibility as if there were a wall between multicellular and single-celled life: once a lineage crosses, it cannot get back. There is no wall. There is a long, gentle slope made of small mutations, each of which made sense in the moment it happened, none of which was about preventing reversion, and whose cumulative effect is that a lineage multicellular for hundreds of millions of years is no longer the kind of lineage a viable single-celled descendant could plausibly emerge from.

That is the framework. The rest of this chapter puts the framework on data: first a laboratory experiment that has now watched ratchets accumulate for the better part of a decade, then nature.

→ Continue with Part 2 — A ratchet caught on camera — the snowflake-yeast experiment.

What this part draws on:

  • The Type 1 / Type 2 ratchet framework and the synergy claim: content/00-framework/ratchet-mechanisms.md; [Libby & Ratcliff 2014]; [Libby et al. 2016].
  • Why “ratchet” must be labelled as metaphor: content/00-framework/ratchet-mechanisms.md §“Is ‘ratchet’ a metaphor or a mechanism?”; voice principle 4 from explainer/README.md.
  • The slope-not-wall framing as project-internal synthesis of the literature.

Part 2 — A ratchet caught on camera — the snowflake-yeast experiment

Until the 2010s, almost everything in Part 1 was theoretical. The mathematics worked. The conceptual story was clean. Ratchet mutations had not yet been watched in the sense of a researcher selecting a population of cells, running it for years, and catching the cells in the act of becoming dependent on each other in ways that locked the multicellular state in.

That changed at a laboratory at Georgia Tech, in a programme known by an acronym: the MuLTEE, the Multicellularity Long-Term Evolution Experiment. The setup is the simplest thing you could imagine.

The setup

Take ordinary baker’s yeast. The same single-celled fungus people have used to make bread and beer for thousands of years. Put a flask of yeast cells in liquid, let them grow, and then at the end of the day let the flask sit still for a moment. Whatever sinks to the bottom fastest, you keep. Whatever stayed suspended, you throw away. The next day, you grow the survivors back up, and at the end of the day you do it again. Daily transfer. The selection is for sinking fast.

A single yeast cell sinks slowly; it is small and light. A clump of yeast cells sinks faster; it is heavier and bigger. The daily settle-and-keep regime is, indirectly, selection for clumpiness.

Ratcliff’s lab first ran this experiment for around sixty daily transfers, roughly three hundred and fifty generations of yeast, and got a striking result. The yeast evolved into branched, snowflake-shaped clusters. A single mutation in a gene called ACE2, whose normal job is to help mother and daughter cells separate after division, was enough to make daughter cells stay attached after they budded off. The cells were dividing and staying stuck where they were already attached. Clonal multicellularity in a few months [Ratcliff et al. 2012].

The lab kept going.

Six hundred transfers

By transfer 600, roughly three thousand generations in, the experiment had produced something that almost no theorist had predicted. The clusters were no longer microscopic curiosities. They were macroscopic. Visible to the naked eye. Millimetres across. Around twenty thousand times the volume of the original ancestor cell; around ten times tougher to pull apart [Bozdag et al. 2021]. You could pick one up with tweezers.

How? The cells inside the clusters had done two specific things. They had elongated, becoming tube-shaped instead of egg-shaped. And the branches had begun to tangle physically through each other, threading like roots so that even when individual cell-to-cell bonds broke, the cluster as a whole did not fall apart. You could break thousands of individual connections and the cluster would still hold, because the geometry of the entanglement carried the load.

This is a Type 1 ratchet, observed in real time. A single cell that loses the elongated shape, or fails to tangle properly, is now badly adapted to being a single cell (its growth rate is reduced compared to the wild ancestor) but very well adapted to being part of a cluster. A cell breaking out of one of these tangled balls and trying to live solo is not just leaving its home; it has spent three thousand generations evolving in a direction that makes solo life worse. The accumulation of small changes in branch geometry, cell shape, and division pattern adds up to a body that is much better at being a body than its component cells are at being cells.

The cluster did not “decide” to stay together, and the cells did not “agree” to commit. The selection regime, sink fast and get kept, favoured cells whose lineages stuck together over cells whose lineages dispersed, generation by generation, mutation by mutation. Nothing in the experiment enforced commitment except the daily centrifuge. The ratchet built itself.

A second ratchet, of a different kind

In the same experiment, with the same cell lines, something else happened that is equally important for our story and that ratchet theorists had predicted but never confirmed.

Across all ten replicate populations the lab was running, the yeast doubled the entire genome. Every chromosome, present in two copies in the ancestor, became present in four. The cells went tetraploid.

In a normal single-celled yeast, this is an unstable condition. Tetraploid yeast cells in standard lab populations almost always lose the extra chromosomes within a small number of generations. Carrying a doubled genome is costly to a free-living cell: the extra DNA has to be copied at every division, the chromosomes have to be sorted properly into daughters, the load on the cellular machinery is heavier. Cells that drop back to two copies grow faster, and they take over.

In the MuLTEE snowflake yeast, that didn’t happen. The tetraploid state was maintained for around five thousand generations, across all ten replicate populations. Not one of them reverted [Tong et al. 2025].

That is a Type 2 ratchet, observed in real time. The mutations that maintained tetraploidy (and there were specific genetic changes the lab was able to identify) would have been costly for a free-living single cell. In a snowflake cluster, the selective regime had changed: the unit selection was working on was the cluster, not the cell, and at the level of the cluster the doubled genome was neutral, possibly slightly beneficial. The reversion pathway that would have purged the tetraploidy in normal single-celled populations was no longer favoured. The cells had stopped being single cells, in the relevant evolutionary sense, and the selection that maintains single-cell appropriate genome architecture had been switched off.

Try the picture this way. In normal yeast, going tetraploid is like loading two suitcases onto a bicycle: it slows you down and you drop them. In the snowflake cluster, the bicycle has become a small wagon pulled by many others; the extra suitcase is no longer the limiting factor on anyone’s progress, and the wagon keeps it. The fitness consequences of the same genetic change depend on what kind of thing you are. The MuLTEE yeast had become a different kind of thing.

What the experiment proves, and what it doesn’t

The story is visually compelling enough that it would be easy to overclaim. What the experiment proves is that ratchet mutations of both kinds can accumulate in a laboratory population, on a timescale of a few years of daily transfers, under a selection regime that is not particularly exotic. In this experimental system, in this lineage, ratchets are no longer hypothetical. They have been watched. The Bozdag et al. 2021 paper on the macroscopic clusters and the Tong et al. 2025 paper on the persistent tetraploidy are the two halves of that demonstration.

What the experiment does not prove is that ratchets of the same strength operate in natural lineages on the timescales of hundreds of millions of years, or that animals and plants and brown algae locked in by the same exact mechanism. The MuLTEE is one experimental system in a Petri dish; the lock-in it has produced is not the same as the lock-in animals have had six or seven hundred million years to develop. The honest line: the experiment converts ratchet theory from “plausible mechanism in models” to “watched mechanism in the lab.” A substantial promotion. Not the same as saying ratchets explain everything that needs explaining.

The MuLTEE has shown one more thing, the lever for the rest of the chapter. The lab ran a second experiment, alongside the snowflake one, that varied just one thing. It is the cleanest piece of evidence in the field for why clonal lineages can lock in and aggregative ones cannot. That is Part 3.

→ Continue with Part 3 — Clonal versus aggregative, run head to head.

What this part draws on:

Part 3 — Clonal versus aggregative, run head to head

Chapter 1 introduced two ways multicellularity can be put together and promised this chapter would come back to the distinction. Now it does.

A clonal multicellular body is built from a single founder cell whose descendants stay stuck together after each division. Every cell is a near-copy of every other cell, the same genome give or take a handful of mutations that have appeared along the way. The cells are, in the relevant genetic sense, the same individual. The body is a fan of clones.

An aggregative multicellular body is built by coming together. The cells live their normal lives as single cells, often spread out across a patch of soil or a film of water. When some trigger fires, typically starvation, they migrate toward each other, pile up, and build a shared structure: a stalk, a fruiting body, a slug. When the structure has done its job, the cells either disperse as spores or melt back into the population. The next time the trigger fires, the cells that come together will be a different set, often from different parents, with no necessary genetic relationship between them.

Both strategies are common in nature. The bacterial cases from Chapter 1 covered each: cyanobacterial filaments and Streptomyces are clonal; myxobacteria are aggregative. On the eukaryotic side, animals, plants, fungi, Volvox, and the brown algae are clonal; the dictyostelid slime moulds and at least five other independently invented eukaryotic lineages are aggregative.

That is the natural distribution. What does it do under controlled selection?

Twenty against twenty

The Ratcliff lab ran a second experiment alongside the snowflake one, with the same yeast strain and the same selection regime (daily settling, cells that sink fastest kept, the rest thrown out), but with one variable changed. They ran twenty replicate populations of clonal yeast alongside twenty replicate populations of aggregative yeast. The clonal populations were the snowflake type, where mother and daughter cells stay attached and the cluster grows from a single founder. The aggregative populations used yeast strains where cells clumped physically but did not stay attached after division; each generation the lab mixed the cells thoroughly, breaking up the clonal relationships, so that any cluster forming the next round contained cells from many different parents [Pentz et al. 2022].

Same yeast. Same daily transfer. Same selection. Same number of generations. The one thing that differed was the mode of inheritance: whether cluster-mates were each other’s relatives or strangers.

The outcomes diverged sharply. The clonal populations did what the snowflake experiment had done: clusters got larger over time, the cells inside them changed, the lineage locked in. The aggregative populations did not. Clusters formed; they sank; the lab kept them. But the cluster-level adaptation never accumulated. The aggregative populations stayed roughly where they started, clumping under the pressure, falling apart when the pressure was off, with no progressive build-up of the body-level traits that turned the snowflake populations into macroscopic objects.

Same molecular toolkit, same selection, same number of generations, different outcome. The variable that mattered was mode of inheritance.

Why this should be unsurprising, in retrospect

The reason for the difference is not new biology; biologists have been working it out since the 1960s, when a young theorist named William Hamilton spelled out the conditions under which costly cooperative behaviours can spread.

Hamilton’s rule says, in plain terms, that a gene which costs the carrier something but benefits another organism can still spread in a population if the benefit, weighted by the genetic relatedness between the carrier and the recipient, outweighs the cost. The shorthand is rb > c: relatedness times benefit must exceed cost [Hamilton 1964]. The number that does the work is r. Between full siblings sharing a mother and father, r is one half. Between cousins, one eighth. Between two unrelated members of the same species, zero.

Between cells produced by simple division from a single founder, with respect to the genes they carry in their nucleus, r is essentially one. The cells are the same gene-copies. A “cooperative” trait that helps the group at the carrier’s expense is, in this regime, helping copies of itself. There is no cost in the relevant sense; the gene propagates through the group whether the individual cell carrying it reproduces or not.

In the aggregative case, r is whatever it happens to be when the cells come together that round. If the local patch is dense with cells from one parent, r is high; in a mix from many parents, low. The value of r resets every cycle. A Type 1 mutation that helped this generation’s group doesn’t keep paying off in the next round; next round’s group is a different cast of cells. The mutation has no way to accumulate.

In the clonal populations, every cycle reinforces the same fan of relatives, the unit of selection effectively becomes the cluster, and ratchet mutations can build up. In the aggregative populations, the unit of selection stays at the cell level, because every cycle the group is dismantled and reassembled from scratch.

A reintroduction worth making explicit: aggregative is not “weaker” or “worse.” Aggregative multicellularity is an enormously successful strategy. The dictyostelid slime moulds have been doing it for somewhere between three hundred and nine hundred million years; the myxobacteria for at least one and a half billion. These are ancient, ecologically successful lineages. The point is structural: aggregative occupies a different attractor in evolutionary space, one in which ratchets cannot lock the lineage in, and so the multicellular state remains facultative.

One more hedge

The Pentz experiment is one experiment, in one lab, in one yeast strain; the findings stay attached to the system they came from. The lab’s framing (clonal life cycles shift the unit of selection to the group, aggregative life cycles continue to be selected at the cell level) is consistent with everything ratchet theory predicts, and is the cleanest direct empirical test the field has. The next part draws together the rest of the evidence.

What the experiment establishes is that under controlled, identical conditions, clonal populations evolve toward lock-in and aggregative populations do not. Whether that pattern shows up in nature, across the lineages that have done multicellularity over the last two and a half billion years, is Part 4.

→ Continue with Part 4 — The project’s reading — why only clonal lineages have locked in.

What this part draws on:

  • The clonal-vs-aggregative head-to-head experiment: content/00-framework/ratchet-mechanisms.md §“Type 1 vs Type 2 in real systems”; [Pentz et al. 2022].
  • Hamilton’s rule and the inheritance-mode argument: content/00-framework/kin-selection-and-cooperation.md; [Hamilton 1964]; [Queller & Strassmann 2009]; [West et al. 2015].
  • Aggregative-lineage ages and the “not worse, just different” framing: content/02-model-systems/dictyostelium.md; [Sheikh et al. 2018]; Chapter 1 Part 2 of this companion (myxobacteria).

Part 4 — The project’s reading — why only clonal lineages have locked in

Part 3 finished with the question this part takes up. If clonal populations lock in and aggregative ones don’t, in the lab, does that pattern survive transplant to the actual polyphyly bush, to the twenty, forty-five, possibly hundred lineages that have invented multicellularity across Earth’s history?

Take every lineage the field counts as having locked in, every lineage where the multicellular state is obligate, where the organism cannot complete its life cycle as single cells. The list is short and famous. Animals. Land plants. Brown algae (the kelps and their relatives). The complex Dikarya fungi (mushrooms, cup fungi, lichens). The filamentous cyanobacteria with heterocysts. Volvox. Every one of those lineages is clonal. Without exception. The body, in each case, develops from a single founder cell: a zygote in animals and plants, a spore in fungi, a gonidium in Volvox, the single cell at the start of a filament in cyanobacteria. Each new generation passes through a one-cell stage and the body that grows from it is a fan of clones.

Now take every lineage that does multicellularity aggregatively. The dictyostelid slime moulds. Myxobacteria. The other five or six independent eukaryotic aggregators: Sorogena in the alveolates, Guttulinopsis in the cercozoans, Sorodiplophrys in the stramenopiles, Fonticula in the holomycotans, the acrasid slime moulds in heterolobosea, Copromyxa in amoebozoa. Every one of those lineages is facultative. The multicellular stage is real, evolutionarily meaningful, often elaborate, but each lineage spends part of its life as single cells, and each generation reassembles the multicellular body from scratch. None has locked in.

That pattern is the empirical claim this part hangs on: every locked-in lineage is clonal; no aggregative lineage has locked in.

What the integration is, and whose it is

This is the place to hedge carefully. The pieces of what comes next are not the project’s. The ratchet framework is Libby and Ratcliff’s [Libby & Ratcliff 2014; Libby et al. 2016]. Kin selection is Hamilton’s [Hamilton 1964]. The clonal-vs-aggregative distinction is in the literature in many places [Queller & Strassmann 2009; West et al. 2015]. The Pentz et al. 2022 head-to-head experiment is the empirical demonstration that the modes of inheritance lead to different selection regimes. Every one of these is settled science that any researcher in the field would recognise.

The proposition this project reads as the integration of those pieces is this: clonal inheritance is the structural precondition for multicellular lock-in, across the polyphyly map, on geological timescales. The pattern that every obligate-multicellular lineage is clonal is not an accident, and not a property of any particular molecule or molecular toolkit. It is a consequence of the population-genetic structure that clonality enforces. The integration is this project’s contribution, not a discovery the field has consolidated.

That hedge is the point of this part being labelled the project’s reading. The pieces are settled. The synthesis is offered as a coherent reading of the pieces, not as a result the field has put a stamp on.

Why the synthesis hangs together

The mechanism, restated cleanly. In a clonal life cycle, every new body starts from a single cell and develops by simple division. The cells of that body are near-identical copies; relatedness with respect to nuclear genes is essentially one. Hamilton’s rule says that under this condition, a mutation that helps the group at any cost to the individual cell is favoured as if the gene were helping itself, because in the relevant gene-level sense, it is. Type 1 ratchet mutations can accumulate generation after generation, because the conditions for their selection are present every generation. Type 2 mutations accumulate alongside: once the single-celled lifestyle is no longer something the lineage’s cells ever practise, the genes for it drift, break, and are not repaired. The slope of Part 1 keeps tilting.

In an aggregative life cycle, every new multicellular body starts from a fresh assembly of cells, often with low and variable relatedness, and disperses back to single cells at the end of the cycle. A Type 1 mutation that helped one cycle’s group does not necessarily help the next cycle’s group; the cell carrying it may not even be in next cycle’s group, or may be next to non-relatives whose genes do not gain from its behaviour. The selection regime cannot stabilise the way it does in clonal lineages. Type 2 mutations have an even harder time. In an aggregative lineage, the single-celled lifestyle is not obsolete; every generation, the cells go back to it. The genes for single-cell living are under continuous purifying selection. They don’t drift; they don’t break. There is nothing for a Type 2 ratchet to attach to.

On a polyphyly map, a clean asymmetry. Lineages with clonal inheritance accumulate ratchets, eventually become obligate, and elaborate complex bodies behind the locked door. Lineages with aggregative inheritance stay facultative, often for spectacular lengths of evolutionary time, never crossing the threshold.

How long is “spectacular”? Long enough to take the no-counter-example claim seriously. Dictyostelium has been doing aggregative multicellularity for somewhere between three hundred and nine hundred million years, with a central estimate around five hundred million [Sheikh et al. 2018]. Myxobacteria have been at it for at least one and a half billion years. The other aggregative eukaryotic origins range across hundreds of millions of years each. None of them has locked in. We do not have a single counter-example. A long enough silence is itself a kind of evidence.

What the claim does and doesn’t say

A few notes on what this synthesis is not claiming.

It is not claiming that clonal multicellularity is more advanced or more impressive. Aggregative is a different attractor in evolutionary space, not a failure mode. Dictyostelids are extraordinary organisms. They sit in a regime that doesn’t lock in. Different, not worse.

It is not claiming that no aggregative lineage could ever lock in. The empirical statement is “we do not have a single example of one that has,” not “we know none ever will.” Voice principle five: ranges, not absolutes; biased, not certain.

It is not claiming the molecular details are the same across the locked-in lineages. They are not. Animals, plants, brown algae, fungi, Volvox, and cyanobacterial filaments each invented different adhesion molecules, different ways of holding cells together, different developmental programmes. Chapter 3 was about exactly that variability. What the synthesis claims is that underneath the molecular variety, the abstract condition for lock-in is not a property of the molecules but of the inheritance structure across generations.

And it is not claiming ratchets explain everything. They explain why obligate multicellularity, once present, persists. They do not explain why each lineage went multicellular in the first place. Part 6 is honest about how badly we answer that.

A wrinkle

The picture is messier than the synthesis in at least one place. Wild Dictyostelium fruiting bodies, when sampled from natural soil cores at small spatial scales, are often substantially clonal: most cells in a fruiting body do come from the same founder, simply because cells in a small patch of soil are usually descended from local division. The lab-experimental picture of Dictyostelium as wildly chimeric, with rampant cheating between unrelated clones, may overstate what happens in nature most of the time. The species-level question is open. That doesn’t undo the asymmetry; Dictyostelium has still spent hundreds of millions of years aggregative and still has not locked in. It is a reminder that clonal and aggregative sit on a continuum, not on either side of a wall.

Part 5 turns to two worked cases, on opposite ends of the time axis, of what locked-in multicellularity looks like up close. The chapter then closes by acknowledging that even the locked-in cases can, in rare conditions, slowly, in specific lineages, give the door a push the other way.

→ Continue with Part 5 — Two worked cases — heterocysts and Volvox.

What this part draws on:

  • Kin selection backbone: content/00-framework/kin-selection-and-cooperation.md; [Hamilton 1964]; [Queller & Strassmann 2009]; [West et al. 2015]; [Fisher et al. 2013].
  • The obligate/facultative table that anchors the empirical pattern: content/00-framework/ratchet-mechanisms.md §“Facultative vs. obligate multicellularity”.
  • The Pentz 2022 head-to-head as empirical anchor: [Pentz et al. 2022]; [Libby et al. 2016].
  • Dictyostelium age, wild-relatedness caveat, contested edges: content/02-model-systems/dictyostelium.md; [Sheikh et al. 2018].
  • The clonal-only-ratchet synthesis is flagged repeatedly as project-internal integration, not field-consolidated finding (BIG-PICTURE.md §“What is actively contested”; the brief for this chapter; TREATMENT.md Part 3).

Part 5 — Two worked cases — heterocysts and Volvox

The two locked-in lineages this part walks through could not be more different in age. One is older than breathable air; the other is younger than the dinosaurs. Neither is on the way to becoming the other. They are two independent lineages, on opposite branches of the tree, that arrived at irreversible cell-type specialisation by completely separate routes. The same abstract condition (clonality) and the same abstract mechanism (accumulated Type 1 and Type 2 mutations) made the lock-in possible in each.

The deep prototype: cyanobacterial heterocysts

Chapter 1 introduced filamentous cyanobacteria: chains of photosynthesising cells, hundreds long, that solve the photosynthesis-and-nitrogen-fixation conflict by partitioning the two jobs into different cells along the same chain. One cell in every ten to fifteen transforms into a thick-walled, non-photosynthesising, nitrogen-fixing specialist: a heterocyst. What Chapter 1 did not unpack was why the heterocyst is the deepest known example of a ratchet locked-in cell type.

A heterocyst is terminally committed. Once a vegetative cell begins the differentiation programme, commitment becomes irreversible roughly eight to fourteen hours after nitrogen step-down [Wolk et al. 1994; Kumar et al. 2010]. After that window closes, adding fixed nitrogen back (the conditions that would have made differentiation unnecessary in the first place) cannot rescue the cell. It will not revert to vegetative growth. It will fix nitrogen until it stops working, then it will die. It cannot divide. A heterocyst that detaches from the filament dies.

In some species the irreversibility is enforced at the genome level. During the late stages of heterocyst commitment, the cell physically rearranges chunks of its DNA, excising entire developmental loci the heterocyst no longer needs [Rossetti et al. 2010]. The deletions are permanent. This is a Type 2 ratchet at the molecular level you can actually see: the genetic machinery that would have let the cell reverse course has been cut out of its DNA.

The dates should make you stop. The gene that controls heterocyst commitment, called hetR, can be dated by comparing how much it has changed across different cyanobacterial species. Molecular-clock work places its appearance at around 2.6 to 2.7 billion years ago. Fossil akinetes (dormant spore-like cells diagnostic of cyanobacterial multicellularity) are documented at at least 2.0 billion years old [Boden et al. 2025; Schirrmeister et al. 2013; Tomitani et al. 2006]. All of this is before the Great Oxidation Event, before atmospheric oxygen, before almost everything else in this companion’s story.

Popular accounts of cell-type differentiation sometimes frame the trick as a eukaryotic invention, something that emerged with animals, or at the earliest with the green algae. That framing is wrong by about two billion years. The deepest known case of heritable, irreversible cell-type commitment is in a bacterium, and it preceded essentially every eukaryotic case by a margin that is hard to overstate. In many of the metrics that matter, cyanobacteria are the prototype rather than a curiosity preceding the real multicellular story.

The clonal condition matters here. A heterocyst is one cell in a chain whose members are all descendants of the same starting cell; the relatedness between the heterocyst and the photosynthesising cells it feeds is essentially one. The heterocyst’s death without reproduction costs its genes nothing; copies of those genes propagate, identically, through the dividing cells around it. Hamilton’s rule, on a filament that began with one cell, is satisfied trivially. The lineage could lock in because the inheritance structure was clonal. The lineage did lock in because, given that structure, the ratchet mutations had room to accumulate, and over hundreds of millions of years they did.

The young, legible case: Volvox

Now slide forward two and a half billion years.

Volvox carteri is a green alga: a spherical green ball about half a millimetre across, visible to the naked eye, made of around two thousand cells in two types. Most of them, around two thousand small biflagellate cells, line the outer surface and beat their flagella to propel the colony through the water. They will never reproduce. The remaining sixteen or so are large, non-flagellated, sequestered toward the interior; they are the reproductive cells, the only ones whose descendants will form the next generation. The two types have the same genome; they are different cell types in the same body, making different choices about what to do with it.

This is the cleanest cell-type differentiation outside animals, plants, and the more complex fungi. It is also, on geological timescales, very recent. The colonial volvocines as a whole probably arose somewhere between 250 and 140 million years ago, with the germ-soma division in Volvox itself within that range [Herron et al. 2009; Ma et al. 2023; Lindsey et al. 2024]. Volvox is younger than the dinosaurs. The germ-soma lock-in evolved within a geologically short window: a few tens of millions of years to go from facultative coloniality to obligate division of reproductive labour.

Part of the answer to how it happened so fast is one gene.

Inside each somatic cell sits a regulator whose job is to switch off the genes the cell would need for reproductive growth. The somatic cell already has, in its genome, everything it would need to become a reproductive cell. What it has in addition is this regulator, active in somatic cells, that shuts the reproductive programme down. The cell swims; it does not grow into a new colony; it dies in a few days, and a new generation starts from the reproductive cells it has been propelling around.

The regulator is called regA. Loss-of-function mutations in regA produce somatic cells that escape the somatic programme: they start growing, they start dividing, they start trying to form colonies of their own [Kirk 2005]. The lock-in is local: one gene is sufficient to enforce the commitment, and breaking that one gene is sufficient to release it.

Where did regA come from? Not from a new evolutionary invention. It was co-opted from a regulator that, in single-celled green-algal ancestors, was part of the daily life cycle. In a Chlamydomonas-like ancestor (Chlamydomonas reinhardtii being the unicellular green alga closest to the volvocine root), a closely related gene mediated a trade-off between motility and reproduction across the day-night cycle: during the swimming phase the gene’s products kept the cell focused on swimming rather than dividing; at the right phase they relaxed and let division proceed [Nedelcu & Michod 2006]. The ancestor had a temporal programme, “swim now, reproduce later,” wired up by this regulator.

In Volvox, that same regulator was rewired into a spatial programme. Instead of “swim now, reproduce later” at the level of one cell across one day, it says “swim here, reproduce there” at the level of two cell types across one body. The variable changed: time in the ancestor, position in Volvox. A small genetic change in regulation produced a large change in cellular behaviour, because the machinery the change acted on was already there.

This is the textbook example of how a major developmental commitment can lock in through the rewiring of a pre-existing toolkit rather than the invention of new genes [Prochnik et al. 2010; Matt & Umen 2016]. The lock-in is a Type 1 ratchet (a somatic cell with regA active cannot reproduce solo; a hypothetical revertant would have to break regA) with Type 2 components (the regulatory wiring for solo Chlamydomonas-style life-history switching has been repurposed and would have to be re-evolved for a revertant to function as a normal single-celled green alga).

Extant cousins, not stages

Both worked cases come with a temptation the chapter cannot let slip. The volvocines, in particular, are widely presented in popular biology as a ladder of complexity within a single algal family. Chlamydomonas (one cell), then Gonium (eight cells, no specialisation), then Eudorina (thirty-two cells, still no specialisation), then Pleodorina (partial germ-soma), then Volvox (full germ-soma). Look how nicely evolution works its way up, says the popular version.

That is not what is happening. Every one of those genera is alive right now. They are not stages. Chlamydomonas reinhardtii is not the ancestor of Volvox carteri. They are extant cousins, descendants of a common ancestor that is neither of them, and recent phylogenomic work argues that even within the volvocine clade, multicellularity arose at least twice and cellular differentiation four to six times [Hanschen et al. 2018; Lindsey et al. 2021; Ma et al. 2023]. Even the textbook ladder is, on close inspection, a small bush of independent inventions inside one algal family.

The lesson is the same as in Chapter 1, applied to the chapter’s own examples. Heterocysts and Volvox are independent. They are not steps in a sequence of increasing complexity; one did not lead to the other. They are two of the at least twenty lineages on the polyphyly map where multicellularity locked in, separately, using different molecular tools, on different time-scales. What they share is not ancestry but a structural condition: in each, the body that locks in is a clonal body, and on top of that condition, ratchet mutations had room to accumulate.

→ Continue with Part 6 — The door swings both ways, and what ratchets don’t explain.

What this part draws on:

Part 6 — The door swings both ways, and what ratchets don’t explain

Most of this chapter has been about lock-in: why, in some lineages, the multicellular state holds for hundreds of millions of years without coming apart. That is half the truth. The other half is that multicellularity has been lost, repeatedly, in lineages whose ancestors were unambiguously multicellular and whose modern descendants are unambiguously single-celled. The door swings both ways. Not often. Not easily. But it does.

Reversion is real

Start with an example in your kitchen.

Baker’s yeast (Saccharomyces cerevisiae, the cells in your bread and your beer and your wine) is a single-celled fungus. It lives as separate cells, dividing by budding, with no developmental programme of body-building. It is also derived from a multicellular ancestor. The Saccharomycotina, the fungal subgroup that contains baker’s yeast and its relatives, descend from filamentous ancestors that made elaborate hyphal bodies, the kind of mycelium you would see if you broke open a mushroom or scraped mould off bread. Their modern descendants do not [Nagy et al. 2014; Kiss et al. 2019; Naranjo-Ortiz & Gabaldón 2020].

The genetic signature is still there. Yeasts retain a disproportionate fraction of the genes their filamentous ancestors used to build bodies; the toolkit is mostly intact, even though the cell is no longer using it. Under certain lab conditions, yeast will partially re-acquire a filamentous growth pattern called pseudohyphae. The capacity is dormant rather than absent. The lineage went unicellular because, in the niches it exploited (sugar-rich liquid environments, rapid colonisation of fresh food sources, fast asexual division), the multicellular lifestyle was a worse strategy than the single-celled one.

A second case, less familiar. Helicosporidium is an obligate intracellular parasite of insects, a single-celled organism that lives inside the cells of beetles and other invertebrates, having lost its photosynthetic capacity. It is also, by ancestry, a green alga, descended from multicellular photosynthetic ancestors whose closest extant relatives are still multicellular [de Koning & Keeling 2006; Pombert et al. 2014]. The lineage went from a free-living, photosynthesising, multicellular organism to a single-celled parasite hiding inside an animal cell.

A third case, in bacteria. Gloeocapsopsis sp. UTEX B3054 is a unicellular cyanobacterium that lives in extreme desert environments. By ancestry, it is filamentous, descended from cyanobacteria of exactly the kind we walked through in Part 5. Its genome still carries the signatures of the filamentous ancestor; the sepJ, fraE, fraH genes that built the cell-to-cell channels are still there, in a cell that no longer forms a filament [Urrejola et al. 2020]. The lineage went unicellular and kept the machinery as a kind of evolutionary souvenir.

The same Ratcliff group that built snowflake yeast over six hundred transfers has also shown the reversal in the lab. Under reversed selection, keeping the cells in the suspension fraction rather than the settled fraction so that single cells are favoured over clusters, snowflake clusters can be made to dissolve. Cells stop sticking after division; the macroscopic body falls apart; the lineage goes back to something close to its single-celled ancestor over a few hundred generations [Kuzdzal-Fick et al. 2019; Rebolleda-Gómez & Travisano 2019]. The reversion does not retrace the same molecular path the forward evolution took. Phenotypically reversible, molecularly not.

Four cases. Four different lineages: fungi, green algae, cyanobacteria, lab-evolved yeast. The door swings.

Probabilistic, not absolute

Go back to Part 1’s mechanical ratchet for a moment. The tooth-and-pawl metaphor implies a hard catch: once the gear has turned, the pawl absolutely prevents reverse motion. That is the metaphor’s failure mode. The biology is gentler. The ratchet biases; it does not abolish. Under most conditions, on most timescales, lineages that have crossed into obligate multicellularity stay there. Under unusual conditions, some lineages go back: strong selection for unicellularity (yeast in sugar-rich liquid), niches favouring small cells (Helicosporidium’s intracellular parasitism), extreme physical environments (Gloeocapsopsis’s desert), experimentally reversed selection (snowflake yeast in suspension). Not most. But some.

The right word is probabilistic irreversibility. The probability per generation per cell of a viable revertant arising is very low. Over enough lineages and enough time, very low probabilities do produce events, and a small number of well-documented cases show up. The ratchet does its job by making reversion unlikely, not by making it impossible. The popular version of the story treats it as impossible. The literature does not.

None of the four reversion cases is a failure or a degraded state. Baker’s yeast is one of the most successful organisms on Earth, more or less responsible for human civilisation’s bread, beer, and wine, and one of the most thoroughly understood model organisms in biology. Helicosporidium is a successful parasite. Gloeocapsopsis is a successful extremophile. Each of these lineages reverted because, in the specific environment it ended up in, single-celled was a better strategy than multicellular. Calling reversion a regression inverts the relationship between the organism and its environment.

What ratchets explain, and what they don’t

If the chapter has done its job, you now have an account of why obligate multicellularity persists across lineages and across timescales: the accumulation of Type 1 and Type 2 ratchet mutations, on a clonal substrate, slowly closing the door behind a lineage. That is a real answer to a real question, and it is what the field of experimental evolution has been developing for the last twenty years.

What it is not is an answer to why multicellularity starts. Ratchets explain persistence, not initiation.

Why did the ancestor of animals stop releasing its daughter cells after division? Why did the cyanobacterial ancestor of Anabaena start producing one heterocyst per ten cells rather than relying on temporal separation? Why did the Chlamydomonas-like ancestor of Volvox begin growing colonies rather than dispersing as single cells? Why did the ancestor of land plants step out of the water?

We don’t have unified answers. Different lineages probably went multicellular for different reasons. In some, the trigger may have been predator pressure: being a larger thing is harder to eat. In others, resource availability: a clump of cells holds onto local nutrients longer than dispersed cells. In others, a developmental accident: a single mutation that disabled mother-daughter separation, in a lineage where the resulting clump happened to be slightly favoured. In each case, local conditions and local genetics likely interacted in ways no single model captures, and the molecular triggers, where we can identify them at all, look different across lineages.

There may not be a single answer at the molecular level. Ratchet theory does not require one; its claim is structural, about what happens after the multicellular state exists, not about how the state came to exist. The framework you have just walked through is necessary but not sufficient. The first question has good answers. The second, on current evidence, does not.

What this chapter doesn’t take up

The four reversion cases this part walked through (yeasts, Helicosporidium, Gloeocapsopsis, snowflake yeast in the lab) share something. They are all lineage-level reversions. A whole lineage of multicellular ancestors gave rise to a whole lineage of unicellular descendants. The organism that exists today is unicellular. The lineage went back.

There is a different kind of reversion with a categorically different character. It happens when, inside a multicellular body whose lineage is still locked in, a cell line escapes the body’s developmental rules and starts behaving like an independent cell: dividing without permission, ignoring the signals that should hold it in place, ignoring the death programmes that should retire it. The lineage stays multicellular. The body stays multicellular. The cell line, inside the body, reverts. It is not free-living, and it does not give rise to a new species; it kills the body it lives in.

That is the next chapter. The trap is real. It holds most of the time. Some lineages slip back across millions of years, and some lineages do so, most consequentially, from inside a body that is otherwise still locked in. Sometimes by accident. Sometimes by force. Sometimes, most dangerously of all, from the inside.

That is Chapter 5.

→ Continue with Chapter 5 — The Door Swings Both Ways. (Not yet written.)

What this part draws on:

  • Reversion cases: content/05-breakdown-and-fragility/reversion-to-unicellularity.md; [Nagy et al. 2014]; [Kiss et al. 2019]; [Naranjo-Ortiz & Gabaldón 2020] (yeasts); [Pombert et al. 2014]; [de Koning & Keeling 2006] (Helicosporidium); [Urrejola et al. 2020]; [Schirrmeister et al. 2011] (Gloeocapsopsis and cyanobacterial reversion); [Kuzdzal-Fick et al. 2019]; [Rebolleda-Gómez & Travisano 2019] (experimental reversion in snowflake yeast).
  • “Probabilistic, not absolute” and the explicit warning against treating reversion as failure: content/00-framework/ratchet-mechanisms.md §“Reversion”; content/05-breakdown-and-fragility/reversion-to-unicellularity.md misconceptions 1 and 5.
  • “Ratchets explain persistence, not initiation” — the open-questions framing: BIG-PICTURE.md §“What we don’t know”; content/00-framework/ratchet-mechanisms.md §“Open questions”; content/06-open-questions/open-questions.md (referenced from the documentary script).
  • The setup to Chapter 5 (cell-line reversion inside a still-multicellular body) without naming cancer: [Aktipis et al. 2015] cited lightly here as the placeholder; the full treatment is in content/05-breakdown-and-fragility/cancer-as-cooperation-breakdown.md and Chapter 5 of this companion.

End of Chapter 4 (draft state).

The chapter is around 9,100 words across six parts. It can be read in one sitting by a reader with no prior biology, on the assumption that Chapters 1 through 3 have already done their setting-up work — in particular, the polyphyly map (Chapter 1), the cooperation-and-cheating problem (Chapter 2), and the molecular toolkit reuse pattern (Chapter 3 — regA sits inside Chapter 3’s larger argument). The next chapter — 05-the-door.md — picks up from Part 6’s closing image: the cells that revert inside a multicellular body whose lineage is otherwise locked in. That category of reversion has its own name in the popular literature, but Chapter 4 has not yet spoken it.