1 What Gets Deleted When Copying Speed Is All That Counts
In 1967, the American molecular biologist Spiegelman and two colleagues ran a small experiment whose results are still cited today. What they did takes one sentence to explain: they let a strand of RNA copy itself over and over in a test tube, with speed as the only criterion for selection. The title of their paper began with the words “An Extracellular Darwinian Experiment.”
Here is how it worked. They took a virus that specifically infects E. coli and used its replicase, a protein whose job is to make new RNA by copying an existing strand. Into a test tube went the enzyme, the raw materials, and a little of the virus’s own complete RNA. They let it replicate. Then they transferred a small sample to a fresh tube of raw materials, let it replicate again, and transferred it again.
Every so often, they shortened the time allowed for replication, taking it from twenty minutes per round all the way down to five. There was only one thing to reward: making copies faster within the time available. No cells, no hosts, nothing that needed infecting. The whole world had been reduced to a copying enzyme, raw materials, and time.
The setup was itself a sieve. It measured just one thing, so it kept only what mattered to that one thing. The original viral RNA was more than 4,200 bases long.
As the transfers continued, it grew shorter and shorter. The reason is simple: each time, the replicase has to copy the strand from beginning to end. The shorter the strand, the sooner it finishes one copy and starts the next.
By the seventy-fourth transfer, the RNA that remained was only about one-sixth its original length. Later experiments of this kind selected short variants with barely more than two hundred bases. Other laboratories subsequently obtained versions with only fifty-odd.
From more than 4,200 to a little over two hundred. The product has a nickname: Spiegelman’s monster.
What had been deleted? It was not junk.
For a virus making its living inside cells, those sequences encoded host recognition, the assembly of a protective shell, the completion of infection. Every one of those functions was essential to survival. But in a test tube that measured replication speed alone, they earned nothing. They contributed nothing and still took up time.
What the experiment demonstrates is limited in scope, but hard to dispute: selection optimizes only what its criterion defines. Anything that contributes nothing to that criterion while carrying a cost will be deleted.
Imagine a long-distance race judged only by who crosses the finish line first. Sooner or later, runners carrying luggage will throw it away, even if it contains everything they need to live on after they arrive.
Change the criterion and a different set survives. In 1974, the same laboratory added a dye that interfered with replication. After a number of transfers, the winner was a mutant RNA that could withstand the dye.
Put it another way. The criterion for selection is itself a construct: this series uses that word for something with boundaries and defined terms that decides what is kept. Whatever it rewards, it retains. Everything outside it, however indispensable elsewhere, gets deleted.
There is nothing abstract about this in nature, either. Every environment defines, in some fashion, what counts as success. It simply has no one to write the definition down.
In a test tube, the experimenter sets the criterion. In open water, who sets it? No one. The deletions happen all the same.
One more thing about the test tube needs to be clear: all the molecules share the same water, and everything in it. Selection can see only how fast they copy. The membrane at the center of this essay changes precisely that.
2 Who Gets the Benefit
Picture a perfectly mixed body of water, with many self-copying molecules drifting through it. Such things have a name: replicators. Anything capable of making copies of itself qualifies. Now suppose one of them mutates and acquires a catalytic ability that can diffuse freely through the water: it makes a particular reaction run a little faster.
That molecule bears the cost of producing the ability. And the benefit? The benefit spreads. Every replicator in the pool can use it. It is like one household paying to repair the elevator in an apartment building: everyone gets to ride, and the household that paid gains no special advantage. The connection between “who made the improvement” and “who leaves more descendants” weakens, perhaps to the point of disappearing.
The most important reason to separate molecules, then, is not to protect them. It is to tie the benefit of an improvement back to whoever produced it.
Now consider another problem. Copying errors impose a hard limit.
In 1971, the German chemist Manfred Eigen proposed a threshold relationship between the error rate of replication and the amount of information that can be maintained. Four years earlier, he had won a Nobel Prize for studying extremely fast chemical reactions. The logic of the relationship is straightforward enough to explain with a book being copied by hand.
Suppose you are copying a book and have a one percent chance of getting each character wrong. Copy ten characters and you will probably get them all right. Copy a thousand and an error is all but certain. The longer the text, the faster the chance of a completely accurate copy falls.
Now add a condition: the correct copy is more popular, so people use it to make further copies. The erroneous versions circulate too, but are less popular.
If the book is so long that almost no copy is entirely correct, popularity can no longer keep the “correct version” concentrated in any one copy. The information disperses.
Put in some illustrative numbers. If the correct version has a tenfold advantage and the error rate is one percent per site, the limit is about 230 sites. Reduce the error rate to one in a thousand and the limit rises to roughly 2,300.
Those are orders of magnitude calculated from a model. Today’s cells copy DNA with enzymes and successive rounds of proofreading, making only about one mistake per billion letters. That is how they can preserve genomes billions of letters long.
Today’s RNA viruses offer a living example of the limit. They make many copying errors, and most have genomes only a few thousand to ten-odd thousand letters long. Coronaviruses have about thirty thousand, putting their genomes among the longest found in RNA viruses. They also happen to carry a proofreading enzyme of their own.
The direction of the problem is clear: replication without enzymes can maintain only a limited length of sequence. Yet making the enzymes that improve accuracy requires longer sequences. This is the famous Eigen paradox.
Eigen had already suggested a way out in his 1971 paper. Around 1977, he and Schuster developed it in a systematic series of papers. They called it the hypercycle.
The idea is to avoid asking one long chain to hold all the information on its own. Instead, several shorter chains help one another. The product of A promotes the copying of B, B promotes C, and the last member loops back to promote A, completing the circle. Each chain can remain short while the whole carries a great deal of information.
The weakness was spotted early. What if one chain mutates so that it still accepts help from the previous link but no longer helps the next? It uses replication resources without contributing, and the effort it saves lets it copy faster. In a well-stirred soup, a member that takes without giving can eat the whole circle out of existence.
In 1979, John Maynard Smith directly challenged the scheme’s stability. The following year, three other researchers published a paper whose title placed three words side by side: hypercycles, parasites, and packages. The third word offered a way out.
In 1987, two researchers built the framework later known as the stochastic corrector model. Several chains with complementary functions are distributed among many small compartments. The chains replicate inside them; when a compartment reaches a certain size, it divides, and its chains are randomly apportioned between the two daughter compartments.
The clever part is this: random assortment keeps creating differences between compartments, and success or failure at the compartment level selects among those differences. Some compartments happen to receive a better-balanced combination and multiply faster as a whole. Compartments heavily overrun by members that take without giving perform worse as a whole and leave fewer daughter compartments.
It is rather like breaking a large class into small groups for an assignment. In the large class, the shirkers disappear into the crowd and collect their marks with everyone else. Once there are small groups, the one that happens to get too many shirkers fails to hand in its work, while the group with fewer does well. Shirking has not changed. Its consequences have: they now fall on the group that contains the shirker.
Selection now pulls in opposing directions at two levels. At the molecular level, it favors the faster copiers. At the compartment level, it favors those that maintain complementary functions, pulling the other way. Selection at the higher level constrains selection at the lower. We will encounter this pattern repeatedly later in the series.
When units at a lower level can each replicate, but are enclosed in a higher-level unit that reproduces as a whole, selection at the two levels can run in opposite directions. Whether the higher level can restrain the lower depends on one thing: whether its success or failure is genuinely tied to what it contains.
An experiment in 2016 supported this mechanism. Researchers periodically separated an RNA replication system into tiny droplets, then mixed it again. In a continuously uniform system, RNA that took without giving could commandeer the replication resources. Periodic separation created differences between groups, allowing droplets with a lighter burden to contribute more. The entire system avoided being eaten out of existence.
This is a demonstration in a modern laboratory, not a record of what actually happened on the early Earth. It works only under certain conditions: how fast a compartment grows must genuinely depend on what is inside it. If compartments merge too quickly, their differences are erased.
Nor is a membrane the only way. Molecules attached to mineral surfaces, or unable to diffuse very far, can keep the benefits nearby without one. What selection actually needs is proximity between the benefit and its producer. Section 7 will turn to those other possibilities.
3 Thirteen Orders of Magnitude
First, what is a membrane made of? Its constituent molecules have a particular trait: one end likes water, the other avoids it. These are called amphiphilic molecules. The most familiar everyday example is soap. One end can catch hold of oil while the other dissolves in water, which is how soap carries grease away.
Drop a collection of these molecules into water and something interesting happens. Exposing the water-avoiding end to water is energetically costly, so the molecules gather on their own. Their water-avoiding ends press together, hidden inside, while the water-loving ends face outward.
At a high enough concentration, they can arrange themselves in two layers: two rows of molecules lie tail to tail, with their water-loving heads facing the water on either side. The exposed edges of such a bilayer are still energetically costly. So the sheet curls up by itself, joins its edges, and closes into a little bubble: a vesicle.
A soap bubble also has two layers of soap molecules, but with water sandwiched between them and the tails pointing outward into the air. A cell membrane is the reverse: two molecular layers with an oily layer in between. Such a membrane is only about five nanometers thick. Stack twenty thousand of them and you would have the thickness of a sheet of paper.
No protein catalyst is needed. It happens on its own. In 1985, researchers extracted organic molecules from the Murchison meteorite, which had fallen in Australia in 1969. Put into water, they too assembled themselves into little vesicles. Even rocks that fall from the sky contain the raw materials for membranes.
Other things can help. In 2003, researchers found that a very common clay made fatty-acid vesicles form much faster. Particles of the clay were enclosed inside the vesicles as well.
Years earlier, researchers had already found that the surface of this same clay could help RNA units join into chains several dozen units long. One grain of clay could help with both jobs.
Self-assembly has its own window, though, and it is narrower than one might imagine. Before the molecules start gathering, their concentration in water must reach a threshold. That threshold is no constant: it varies with temperature, salt concentration, and acidity, and differs enormously from one molecule to another.
The phospholipids in today’s cell membranes each trail two water-avoiding tails. The simplest fatty acids have only one. Oleic acid, found in olive oil, is an example.
Chemists often express concentration in moles per liter. To put it in terms of salt, dissolve nearly sixty grams in a liter of water and you have one mole per liter. Seawater is roughly half that concentration.
For two-tailed phospholipids, the strong water-avoiding effect typically brings the threshold below one millionth of a mole per liter. For structurally simpler, single-tailed fatty acids, it can be as high as one thousandth or even a few hundredths of a mole per liter. Several orders of magnitude separate them.
This has a very practical consequence. Simple fatty-acid membranes and modern membranes are entirely different propositions when it comes to how many molecules are loose in the water and how quickly material passes between membranes.
Simple fatty acids are also particular about acidity. Oleic acid, for example, forms stable vesicles only within the narrow range of about pH 8 to 9. Make the water more acidic and it gathers into oil droplets; make it more alkaline and it tends to break up into smaller clusters.
Then there is permeability. A membrane separates inside from outside, but it does so according to a ranking that spans an enormous range. Molecules such as oxygen pass very easily through its water-avoiding region, with permeabilities that can reach the order of a hundred or more centimeters per second. Water is around one-hundredth of a centimeter per second.
Glycerol is small and uncharged, but strongly attracted to water: its permeability drops to roughly five millionths of a centimeter per second. For potassium ions, the figure can be as low as one trillionth of a centimeter per second.
From the fastest, oxygen, to the slowest, potassium ions, the range is about thirteen to fourteen orders of magnitude. The latter passes through more than ten trillion times more slowly than the former.
What do thirteen orders of magnitude mean? Roughly the difference between one second and three hundred thousand years. The same membrane is almost wide open to one molecule and almost completely closed to another.
These numbers come from different membranes, temperatures, and methods of measurement. They are not a single set of precise parameters. But the orders of magnitude make one point with remarkable consistency: this is not a wall. It is a sieve with an extraordinarily steep scale of permeability. The shape of that sieve determines the problems a cell must solve.
It holds potassium ions back, allowing a cell to maintain different ionic compositions inside and out. Your own cells contain about thirty times more potassium than their surroundings and about ten times less sodium. It does not hold water back, so osmotic pressure needs a separate solution.
It does not hold oxygen back, so closing the door offers no escape from oxygen’s toxicity. The other side of that fact is that oxygen enters your blood from your lungs by passing straight through membranes in just this way. No doors required.
There is one anomaly in the ranking: the proton, the nucleus left when a hydrogen atom loses its electron. Judging by other ions that carry a single charge, it ought to fall into the same range as potassium.
In fact, its permeability is of the order of one ten-thousandth to one-hundredth of a centimeter per second. It leaks through far more easily than potassium does: seven or eight orders of magnitude more easily in the same membrane. Explanations differ as to why.
One is that water molecules occasionally line up inside a membrane, allowing a proton to be passed along the row, one molecule at a time, rather than having to squeeze through on its own. Keep that in mind. We will need it in section 5.
Selectivity immediately creates a transport problem. Some things a membrane cannot keep out; others it keeps out all too well. Many of the raw materials a cell needs happen to belong to the second group.
Modern cells solve this with three kinds of equipment. Channels provide a continuous water-friendly passage through which substances move down a concentration gradient, usually without consuming energy themselves. Carriers bind their targets, change shape, and deliver them to the other side. Pumps use energy directly to move things against a concentration gradient.
The earliest membranes may not have needed this equipment, however. An experiment in 2008 used membranes made from simple molecules such as fatty acids. When raw materials were added outside, small molecules could enter while longer nucleic-acid chains remained within. Materials could come in without the internal templates escaping, allowing new chains to be extended along those templates inside the vesicles. What was being assembled in that experiment was not yet RNA. The first successful assembly of RNA along an RNA template inside such vesicles came in 2013.
No transport proteins were involved at any stage. These experiments also exposed a trade-off. The looser the membrane, the more easily nutrients enter without transport proteins, but the more easily small molecules and ions escape from within. The tighter the membrane, the better it retains its contents, and the more it depends on specialized equipment to acquire charged raw materials.
There is no simple progression here in which a more stable membrane is always better. Longer chains strengthen the effect of the water-avoiding ends and stabilize the membrane, while slowing the exchange of molecules in and out. Shorter chains do the reverse.
Doubly charged metal ions such as magnesium and calcium are especially troublesome. They bind to the heads of these molecules and break the membrane apart. Just a few millimoles per liter of magnesium ions can destroy vesicles made of pure oleic acid. Yet magnesium is precisely what many RNA-catalyzed reactions need.
4 A Boundary Solves Dilution and Creates Pressure
A membrane keeps certain molecules together within a local volume. The problem of dilution eases. A mechanical problem takes its place. Start with osmosis.
Take a bag whose walls have exceedingly fine pores: water molecules pass through, sugar molecules do not. Fill it with sugar water and immerse it in fresh water. What happens?
Water enters the bag. The bag swells until the pressure of its swelling prevents further water from entering. That pressure is called osmotic pressure. A raisin left in fresh water slowly plumps up for the same reason.
Its magnitude depends on just one thing: how many more particles that cannot cross are inside than outside. Whether the particles are large or small does not matter; what counts is their number. Now replace the bag with a cell membrane and the sugar with RNA, proteins, and other large molecules. The same thing happens.
Human red blood cells have no cell wall. Put them in fresh water and they soon swell and burst. That is why hospitals use saline for intravenous fluids rather than plain water.
A simple relationship lets us estimate osmotic pressure in dilute solutions. Around room temperature, every mole per liter of an impermeant substance corresponds to about 2.5 megapascals of pressure. What does 2.5 megapascals amount to? More than ten times the pressure in a car tire.
A difference of just ten millimoles per liter between inside and outside therefore gives about twenty-five kilopascals. A hundred millimoles per liter gives about 250 kilopascals. A very small difference in concentration can stretch a lipid membrane until it tears. Enclosing large amounts of impermeant macromolecules in a vesicle does more than raise the concentration. It also creates an ongoing burden.
Modern bacteria spread that burden across at least three mechanisms.
The cell wall bears most of the mechanical load from swelling. In salty surroundings, cells can actively accumulate or release small molecules to adjust the difference between inside and outside. And when a sudden influx of fresh water strikes, devices in the membrane act like safety valves on a pressure cooker: rising tension opens them automatically, letting the contents out quickly before the membrane bursts.
That last kind of device is not a normal feeding channel. It is an emergency pressure-release valve.
How much pressure is there inside a modern bacterium? Different methods give estimates ranging from tens of kilopascals to around two megapascals, with the highest estimates for thick-walled bacteria. The relation between the orders of magnitude is clear: a lipid bilayer can withstand only limited tension. Modern cells transfer the sustained load to the wall and use pressure-release valves to handle sudden changes.
The cell wall does not come free, either. As bacteria grow, the wall has to be cut open and rebuilt at the same time. Penicillin blocks the rebuilding step. The cutting continues but the repairs cannot keep up, and the bacterium bursts under its own internal pressure.
A large group of bacteria adds a second membrane outside the first, blocking many harmful substances and some nutrients along with them. Bacteria of the kind that cause tuberculosis go furthest, wrapping themselves in a waxy lipid layer that many drugs cannot penetrate. That is one reason tuberculosis is hard to treat. Every extra layer of protection brings another bill. These are the bills faced by present-day cells, though; we cannot charge them retrospectively to the earliest ones.
Experiments outside living cells use three approaches to evade or exploit this problem. Keep the solutions inside and outside the vesicle as close as possible in concentration. Or use substances that cross the membrane quickly, allowing volume to grow along with surface area. Or reverse the approach: deliberately add impermeant macromolecules inside and use the pressure difference to drive competition for membrane material.
Next comes division. To pass its contents on, a vesicle must be able to grow and then split in two. The earliest method was fairly brutal. The same researchers working in 2003 let fatty-acid vesicles grow by absorbing added fatty acids, then forced them through tiny pores to split them into smaller ones.
That route can complete the cycle, but the squeezing is imposed by laboratory equipment and causes some of the contents to be lost. This was precisely the problem a study in 2009 set out to address. Could a vesicle bring itself to the verge of division instead of being forced apart?
First, a geometric fact. A sphere’s surface area is proportional to the square of its radius; its volume is proportional to the cube. The larger the sphere, the less surface it has for its volume. This is one reason cells tend to be small: grow too large and the membrane cannot bring supplies in fast enough.
This constraint has nothing to do with what a membrane is made of. It is pure geometry. What happens if membrane molecules keep arriving and adding surface area, but the volume inside cannot keep pace?
A sphere is no longer the only shape that can accommodate all that area. The excess surface takes the form of an ellipsoid, a tube, or a long filament. Reverse the situation, with water or internal material increasing faster than membrane area, and tension rises until the membrane leaks or breaks.
Early in his career, Szostak studied the ends of chromosomes, work for which he shared the 2009 Nobel Prize in Physiology or Medicine. From the 1990s onward, he redirected his laboratory toward the question of how life began.
In 2009, Ting Zhu and Szostak turned this geometric constraint into a route to growth and division that required no proteins. They placed large oleic-acid vesicles in a solution whose solute crossed the membrane slowly, then added roughly five times as much oleic acid as the vesicles already contained in membrane lipids. The area of the outer membrane grew first. The volume inside could not keep up.
Within tens of minutes, the spherical vesicles elongated into long filaments. The shear force from gentle agitation was then enough to break the filaments into multiple daughter vesicles, with far less leakage of the RNA inside than in the earlier mechanical-extrusion method.
The same paper also supplied a control. Substitute a substance that can cross the membrane quickly, and water follows the solute inside. Volume growth catches up with surface-area growth, the vesicles tend to remain spherical, and filament formation is greatly reduced. The point of the control is that adding fatty acids does not automatically cause division.
What matters is a temporary mismatch between the growth of area and the growth of volume, followed by external shear to complete the break. Here, growth and division are not some capability. They are a mismatch at work. That matters, because it changes the shape of the question.
If division is a “capability,” we have to explain where that capability came from. If it is a mismatch, all we need to explain is why area and volume failed to grow in step. The answer to that question is geometry and permeability, not some new piece of equipment.
5 The Membrane Is a Generator, Not a Container
Now for the most counterintuitive point in the essay: a membrane can itself be an energy device. Imagine a dam. Water is raised, creating a difference in level between the two sides. That difference is itself energy. Open a gate, let the water flow back downhill, and it can turn a waterwheel.
Cells do the same thing, except that they raise protons rather than water. They pump protons from one side of the membrane to the other, creating both a difference in proton concentration and a difference in electrical potential across it.
Those differences provide the drop. How large is it? The voltage across the membrane is only a hundred-odd millivolts, but the membrane is just a few nanometers thick. That works out to thirty million volts per meter. Some have compared it to a bolt of lightning.
The membrane carries a device like a waterwheel. Protons flowing back down the gradient make it rotate, and the rotation is used to manufacture the cell’s energy currency. In a day, a person makes and spends a total amount of this currency roughly equal to their own body weight. The membrane here is no container. It is the dam itself.
This explanation met with sixteen years of rejection. In July 1961, Peter Mitchell published a paper in Nature that proposed precisely the picture above. The prevailing framework at the time was entirely different.
The prevailing view held that the respiratory chain first produced a diffusible “high-energy intermediate molecule,” which then passed the energy to the next step. This fitted another form of energy production already known at the time, so many researchers expected to isolate, trap, or demonstrate the existence of a particular intermediate molecule.
Mitchell said there was no need for one. Energy could first exist as an imbalance across a membrane, without corresponding to any high-energy molecule in the conventional sense. The hypothesis faced prolonged, fierce resistance, for reasons that went beyond conservatism.
It asked researchers to accept something deeply counterintuitive at the time: energy need not be stored in a molecule. There were technical reasons, too. Early experimental preparations tended to leak protons, and membrane integrity was difficult to control. Many experiments consequently could not distinguish cleanly between the two models.
From the 1950s to the early 1970s, numerous candidates for the high-energy intermediate were proposed. Almost every component of the respiratory chain came under suspicion. None received decisive confirmation that could be widely reproduced. In 1962, Boyer, who would later win a Nobel Prize for his work on the rotary device, thought he had caught one. The molecule turned out to be an intermediate in the work of a different enzyme, unrelated to this process.
In 1963, Mitchell left the university for health reasons. He restored a derelict old house in southwestern England, paid to establish a small research institute of his own, and carried out experiments there with his collaborator Moyle. In the mid-1960s, they did something else.
They showed that the onset of respiration in mitochondria was accompanied by measurable proton expulsion, and studied the numerical relationships between stages of the respiratory chain and the movement of protons. This directly supported the idea that electron transport was indeed moving protons across a membrane.
An elegant experiment appeared in 1966. Two researchers first put a plant membrane structure into an acidic environment, filling its interior with protons, then rapidly transferred it to a more alkaline medium containing the raw materials. They had created an artificial proton gradient between inside and outside.
In darkness, with no light-driven electron transport at all, that gradient briefly powered the synthesis of energy currency. The result strongly favored Mitchell. It still did not end the argument at once: advocates of an intermediate molecule could revise their account to say that the gradient indirectly produced some unknown intermediate.
A reconstruction in 1974 was much harder to evade. Two researchers took a light-driven proton pump from a salt-loving archaeon, one of the kinds that turn the water in salt ponds pink, and put it into artificial vesicles alongside the rotary device from mitochondria. Once illuminated, the pump established a gradient. The rotary device used that gradient to make energy currency.
Every other component of the electron transport chain had been removed. Only three things remained: a pump, a sealed membrane, and a waterwheel. Energy conversion happened all the same.
About sixteen years passed between the proposal in 1961 and the winding down of the main dispute around 1977. The 1978 Nobel Prize in Chemistry went to Mitchell alone.
Resolution did not mean one side conceding defeat. By the late 1970s, the main argument over whether the transmembrane gradient was central to energy conversion was largely settled. Major disagreements remained about exactly how the coupling worked. In 1977, several leading figures from the formerly opposing camps coauthored a review. That fact alone showed that they had begun to share a larger framework.
The scale of the mechanism is worth a look, too. In a typical mammalian mitochondrial respiratory chain, every pair of electrons received from NADH, a carrier of electrons released during the breakdown of food, results in about ten protons being transferred to the outer side of the membrane across several stages. The membrane-embedded rotor of the rotary device consists of a ring of repeating components. One full turn produces three molecules of energy currency.
If the rotor ring has ten components, a full turn takes roughly ten protons, or about 3.3 protons per molecule produced. If it has fourteen, the figure is about 4.7.
Known rings in nature range from eight components to seventeen. Mammalian mitochondria have eight, giving about 2.7 protons per molecule. The proton cost of manufacturing one molecule of energy currency therefore differs between organisms.
The rotation is no metaphor. In 1997, Japanese researchers attached a fluorescent filament to the device’s shaft and watched the filament turn under a microscope.
They used only the half of the device that projects beyond the membrane. Feed it energy currency and it runs backward, consuming the currency as it turns. A waterwheel can also work as a water pump. That same year, Boyer, who had long predicted that it would rotate, and the British scientist Walker, who had solved its structure, shared the Nobel Prize in Chemistry.
One further point. This mechanism does not require the membrane to be absolutely impermeable to protons in a mathematical sense. Leakage merely has to be sufficiently lower than the flows being pumped out and returning through the device.
It is a condition on a ratio, not an absolute condition. Whether the ion driving the waterwheel must be a proton is itself disputed. One camp thinks sodium came first. One reason is physical: if early membranes leaked protons badly but let far fewer sodium ions through, a sodium gradient would indeed have been easier to sustain.
The other camp thinks protons came first. Naturally alkaline hydrothermal fluids meeting more acidic seawater can supply a ready-made proton gradient, so geochemistry could have provided the first energy to be harnessed. The evidence does not give an undisputed answer as to which ion first turned the wheel.
One thing does not depend on that answer. This mechanism is not preserving “proton concentration” in a uniform solution. It defines energy as an imbalance between the two sides of a membrane. Let those sides mix freely and the gradient disappears.
The 1974 reconstruction made this into a controlled comparison. Without closed vesicles, the proton displacement produced by the pump could not create a gradient across a membrane. With intact vesicles and the waterwheel installed alongside the pump, light produced energy currency.
So boundary and energy device are not two separate functions of a membrane. Creating the distinction between inside and outside is itself the physical condition required to store this kind of energy.
Now we can return to the anomaly in section 3. The energy systems of the vast majority of organisms today are staked on protons; only a minority of bacteria and archaea use sodium. Yet among ions carrying a single charge, protons are the ones that leak through membranes most easily.
This is not a choice that can be casually swapped for another. To preserve a gradient across a membrane, the slower the leakage, the better. Yet the system uses the ion that leaks fastest. As to why, the constraint at the end of section 3 points in one direction: early membranes were vulnerable to doubly charged ions such as magnesium and calcium, while magnesium was indispensable to catalysis.
There were never many usable ions to choose from.
6 Incomplete but Still Competing
A membrane encloses some molecules and, in the same act, shuts others out. That sounds like a truism. Its consequences, however, can be calculated.
If a vesicle forms by randomly enclosing a volume of a uniform solution, the number of molecules of any given kind that end up inside is like the number of sesame seeds that land in each of a field of little bowls. You can calculate the average number per bowl. How many a particular bowl gets is random.
Take a vesicle a hundred nanometers in diameter. If a key molecule has an environmental concentration of one nanomole per liter, the mean is only about 0.0003 molecules per vesicle: roughly three vesicles in ten thousand get at least one. Raise the concentration to one micromole per liter and the mean becomes about 0.32, with a roughly twenty-seven percent chance of getting at least one.
Raise it to ten micromoles per liter and the mean reaches about 3.2; only then does the probability rise to roughly ninety-six percent. Now suppose five different components are needed, each present at one micromole per liter, and at least one of each must be enclosed.
Under the simplifying assumption that these events are independent, the chance of one vesicle getting all five is about 1.4 in a thousand. For ten components, it is about two in a million. Imagine a dish requiring five seasonings, each of which happens to be stocked by only about one kitchen in four. Among a thousand kitchens, only one or two will have the complete set.
This calculation is a common experimental baseline, and real systems can depart from it. Molecules may adhere to the membrane, cluster together, or have an affinity for particular membranes. But the direction and size of the departure have to be measured for each system. We cannot simply assume the deviation will always help.
The mechanism that creates differences between compartments is the very same mechanism that leaves each compartment missing pieces. You cannot have just the first half.
Small volumes mean that the number of each low-concentration molecule may be a discrete zero, one, or two. “Which one is missing” becomes a substantive difference between compartments, rather than a small fluctuation in concentration. The stochastic corrector model exploits precisely these differences, and also bears their cost.
Today’s cells cannot escape this either. In 2002, researchers put two fluorescent genes into E. coli, one making the cells glow cyan, the other yellow. Identical switches were placed in front of the genes. In principle, each cell should have produced equal amounts of both colors.
Under the microscope, though, some cells in the same population leaned toward cyan, others toward yellow. Same switches, same environment. The difference came from molecules present in very small numbers inside the cells, each taking its own chances.
Beyond missing pieces, there is something more direct: compartments compete. In 2004, three researchers, Szostak among them again, enclosed impermeant RNA inside fatty-acid vesicles, raising their internal osmotic pressure. A fatty-acid membrane’s material does not stay put. Individual fatty-acid molecules can travel through the water and exchange between vesicles.
When taut vesicles coexist with slacker ones, the former relieve their tension by taking up fatty acids, while the latter lose some membrane. One grows; the other shrinks. It is like two balloons, the inflated one slowly peeling skin off the deflated one to use for itself. Depending on what was enclosed, the former gained between twenty and fifty percent in surface area. No catalysis was involved in the mechanism.
RNA in this experiment was simply an impermeant macromolecule: more RNA inside, higher pressure, a net inflow of fatty acids, more membrane area. An internal chemical difference altered the boundary’s rate of growth through physical mechanics alone.
This is a laboratory demonstration, but it makes something visible: what a pouch contains can determine how fast the pouch grows, without a single enzyme at any stage. Look back at the test tube in section 1, where selection could see only copying speed. Here, what is inside the pouch has become tied to the pouch’s own fate.
In 2011, the same laboratory added a small amount of the phospholipids used by today’s cells to fatty-acid vesicles. The vesicles containing phospholipids held on to fatty acids, taking membrane material from those without them and growing larger and larger.
This may help explain why cell membranes later switched to phospholipids. Just a little at first could confer an advantage in the contest for membrane material. But as phospholipids increased, the membrane tightened and outside raw materials could no longer get in. The cell then had to build a set of transport devices. One advantage brought the next problem along with it.
7 Inside and Outside Without a Membrane
Your cells contain many membrane-free “rooms” right now. The nucleolus inside a cell’s nucleus is a droplet with no membrane around it. When a cell encounters high temperatures or other stresses, new droplets condense within minutes, then disperse again after the stress has passed.
This section considers three ways to create an inside and an outside without a membrane. Droplets are the first. Mix certain charged macromolecules together and they spontaneously gather into a highly concentrated region that coexists with the surrounding dilute solution without mixing evenly into it.
It is a little like oil and water separating, though not quite. These droplets contain a great deal of water, and molecules can move in and out rapidly; they simply tend to partition into the interior. What forms is a boundary in concentration, not a continuous barrier.
The phenomenon was noticed long ago. In 1929, two Dutch colloid chemists described it systematically. In the 1930s, the Soviet biochemist Oparin used it to imagine the origin of life, calling the droplets coacervates.
Before that, Oparin and the British scientist Haldane had independently imagined organic material slowly accumulating in the early seas, like a thin soup. Coacervates were the first containers Oparin found for the molecules in that soup.
Around 2009, modern cell biology returned to studying droplets of this kind as a way cells organize themselves. Researchers had observed certain granules in the germ cells of nematode worms merging, dripping, and exchanging material rapidly, all in the manner of liquids.
In a 2018 experiment, researchers put an RNA-cleaving ribozyme into model droplets. The ribozyme and the RNA it was to cut became concentrated more than a thousandfold inside, and cleavage still occurred. Concentration is not necessarily an improvement, though: at equal concentrations, cleavage in these droplets was about sixty times slower than in an ordinary solution.
The limitations of droplets are equally concrete. With no continuous barrier, they cannot hold ions back and therefore cannot preserve the kind of gradient discussed earlier. Droplets also merge with one another, and once they merge, their differences disappear.
Researchers have tried to work around that last problem. A study in 2022 built systems in heated microscopic rock pores and on the surfaces of gas bubbles, observing droplets becoming concentrated, growing, merging, and breaking apart under temperature differences and flow.
This shows that environmental flow can substitute for some of the machinery that divides membranes. It does not prove that a particular rock pore in the Archean ever sustained such a population of droplets.
The second method uses protein shells. Bacteria have a kind of tiny protein cage with an icosahedral shell, like a minute soccer ball, typically a hundred to several hundred nanometers across. Inside are enzymes that fix carbon dioxide.
The same enzyme occurs in plant leaves and is often called the most abundant protein on Earth. The cage keeps carbon dioxide more concentrated inside than outside, letting the enzyme work much faster.
These cages occur in all known cyanobacteria and in several autotrophic bacteria. Structural evidence suggests that their two main types are very likely two independently constructed systems.
Another kind of shell is smaller, about twenty to forty nanometers across. It self-assembles from proteins resembling those in viral coats and encloses cargo by recognizing specific tags. Yet another kind contains nothing but gas, helping many aquatic bacteria and archaea regulate their buoyancy, much like a fish’s swim bladder.
Many of these compartments were invented independently. There is no total we can give for how often evolution has arrived at separation.
The third kind of compartment is not made by organisms at all. Since the 1980s and 1990s, some researchers have argued that the first local reaction spaces need not have been freely floating vesicles.
In parts of the seafloor, seawater seeps into rock, reacts with it, and emerges as a warm, alkaline fluid. Mixing with the surrounding seawater, it precipitates minerals into small, porous honeycomb structures with thin mineral walls.
The first essay mentioned Lost City on the Atlantic seafloor, where white carbonate chimneys rise as high as sixty meters, the height of a twenty-story building. Water in each pore exchanges only to a limited extent with neighboring pores and the outside. The mineral structure begins by providing geometric separation.
Physical experiments and simulations support this proposal. In 2002, two researchers studied how DNA behaved in water subjected to temperature differences over very small distances. A temperature difference alone drew DNA toward the colder side. Add convection, however, and the flow repeatedly carried molecules back into the temperature-gradient region, trapping them. Experiments and models could produce local enrichments exceeding roughly a thousandfold.
In 2007, computer simulations took this combined effect into hydrothermal channels. In long, narrow pores a millimeter wide and closed at one end, even individual nucleotides could be concentrated more than a hundred millionfold.
Salinity imposes a specific limit, however. At a salt concentration of about half a mole per liter, roughly the salinity of seawater, the effect weakens substantially or becomes difficult to detect. Findings from low-salt experiments cannot be extrapolated directly to the ionic composition of natural seawater.
These experiments and simulations show that heat-driven flow in microscopic pores really can concentrate molecules. There is still no evidence of which hydrothermal system in the Archean actually did so.
Mineral pores have a problem they cannot solve for themselves. A pore does not replicate. It can concentrate, catalyze, and provide a compartment, but it cannot make a second copy of itself.
A droplet can at least split in two. Protein shells are built by cells and reproduce along with them. A mineral pore is merely a place, staying where it is.
Suppose a favorable combination of molecules appears in one pore. If the pore itself does not grow, divide, and pass its contents down a lineage to daughter pores, then the “compartment reproduction” required by the stochastic corrector model has not occurred.
Flow can carry the contents downstream to new pores, but that is not the same as an entire compartment moving with all its constituents aboard. Several hurdles separate mineral compartments from free cells: how a boundary grows inside a pore; how it leaves the mineral with its contents without falling apart; how, once free, it grows and divides on its own; and how it generates a gradient of its own after losing the ready-made one.
Every step is a concrete engineering problem. Wherever the earliest chemistry happened, what eventually emerged was, in every case, a cell carrying its own membrane.
8 The Boundary Is Visible. Its Material Is Not.
How small can a self-maintaining compartment be? Starting from the need to fit genetic material, protein-making machinery, basic metabolic proteins, and enough water and ions inside, classic estimates put the lower limit for an autonomous spherical cell at about 250 nanometers in diameter.
That figure has a history. In 1996, researchers reported tiny bacteria-like structures in a Martian meteorite, only a few tens to a hundred nanometers long.
Two years later, the US National Academy of Sciences convened a meeting specifically to ask how small a cell could be. The lower limit estimated there was around the 250-nanometer mark. Those “Martian bacteria” were too small to hold all the equipment a cell needs.
Among cultured marine planktonic bacteria, Pelagibacter ubique is one of the smallest free-living, autonomously replicating cells, and one of the most abundant bacteria in the ocean. Its volume is about 0.015 to 0.058 cubic micrometers. It is only a hundred-odd nanometers wide: line up some five hundred side by side and they would span the width of a human hair.
Smaller candidates can have volumes as low as around 0.01 cubic micrometers, but these extremely small groups often have drastically reduced genomes and may depend on other members of their communities for metabolites. Whether they count as “independently self-maintaining” is therefore a matter of disputed definition.
Mycoplasmas lack cell walls. One of them causes mycoplasma pneumonia, and their cells are often only about three hundred to eight hundred nanometers across. They have lost many biosynthetic abilities, however, and rely on hosts to supply various substances. Some giant viruses are bigger than these smallest cells.
“The smallest self-replicating cell” and “the smallest genuinely free-living cell” are thus different records. Both, though, are at the limits of ordinary light microscopy. Even magnified a thousand times, they appear as little more than a blurred dot.
What, then, did the earliest membranes leave behind? Membranes themselves do not fossilize. Their lipid molecules break down after death. What can remain are the relatively stable carbon skeletons left after that degradation.
Different groups of organisms make slightly different skeletons, so these residues are called biomarkers. Searching for the earliest membranes really means searching for the earliest molecular remains that can be attributed to cell membranes.
This is much harder than it sounds. Concentrations of these molecules in rock are extraordinarily low, often below one part per billion. Drilling mud, plastics, laboratory solvents, and oil that seeped into the rock later can all introduce younger organic material.
Late in the twentieth century, researchers reported such molecules in Australian rocks about 2.7 billion years old. In 2012, new cores were drilled using ultraclean methods. Results published in 2015 showed amounts scarcely different from blank controls containing no sample at all. The fifth essay will tell that story in full.
After scrutiny of this kind, membrane-lipid molecules that can be confirmed to be as old as their host rock reliably reach back about 1.64 billion years; examples about 1.73 billion years old have also been reported. Cell-like microfossils, meanwhile, go back about 3.4 billion years. Throughout the intervening 1.7 billion years, cells clearly had membranes. Those membranes simply left no molecules we can trace.
The first essay examined the three layers of evidence for those microfossils. What they preserve is carbonaceous shape, mineralized outline, filamentous structure. They do not preserve a lipid bilayer with its molecular structure intact. Seeing a cell-like boundary and determining what its membrane was made of are a long way apart.
The second essay described the two different sets of membrane lipids used by today’s bacteria and archaea: their backbones are mirror images, and their tails and the bonds joining them differ too. Which set made up the first membrane, or whether it was neither, remains unanswered.
Now we can return to this essay’s construct. A membrane defines “outside” by defining “inside.” That act solves dilution, ties the benefit of an improvement back to its producer, and becomes, along the way, the physical prerequisite for storing energy. From the moment it is established, it also begins creating its own remainder: what the construct can neither accommodate nor eliminate. In this series, that is what “remainder” means.
The components left out by random assortment are a remainder. Missing is missing; small volumes inherently mean discrete numbers. Osmotic pressure is a remainder. It has not been solved but passed on to cell walls and pressure-release valves. Whatever has to be pumped out is a remainder, and the pumps must keep consuming energy. Everything shut outside is a remainder too.
Not one of these remainders has been resolved. They are all still there. Only their location has changed, or who bears them. In the rocks, there is another remainder: we can see that the boundary existed, but not what it was made of.
The criteria are still sifting. The rocks are still being cut open anew. Vesicles in laboratories are still drawing out into filaments, breaking apart, stealing membrane lipids from one another.