I A Cone, Until You Look from Another Direction
Some of the oldest rocks on Earth lie in southwestern Greenland, around Isua.
There is a rock face there that is 3.7 billion years old. Before going any further, it is worth asking what “3.7 billion years” means to a piece of rock. It has been buried deep underground, baked, squeezed, and eventually heaved back to the surface. That is more than enough punishment to knead most small, delicate shapes out of existence.
To find any orderly structure in such a rock, even one measured in centimeters, is unusual in itself. In 2016, a research team examined a newly exposed stretch of that rock face and found a collection of little cones. They stood one to four centimeters high, side by side, in the same layer. The team believed that living things had left them there. More specifically, living things of the following kind.
In shallow seas today, certain microorganisms spread over the seafloor in a thin mat. The mat traps grains of sand and mud drifting down through the water. Once buried, it grows another layer over the top, catches another layer of sediment, and grows upward again. Year after year, a little mound of stone builds beneath it, layer upon layer.
Such a mound is called a stromatolite. Living examples can still be seen in some Australian bays. If the Greenland cones really grew in this way, microorganisms were laying down their mats there 3.7 billion years ago. Why should that be something to argue about?
The earliest persuasive records of life currently reach back about 3.5 billion years. If the cones hold up, they push that date back by 200 million years in a single move, into a stretch of time for which almost no other evidence is available.
And that stretch is precisely the one that matters. Earth itself is a little more than 4.5 billion years old, and its early history included violent impacts. The possible answers to “How early could life have begun?” are already confined to a narrow window. Every step backward takes another large bite out of the time available for chemistry to do its work.
The team’s case rested on more than shape. They also drew on the rocks’ chemical composition and sedimentary textures, assembling several supporting clues for an origin in shallow water. Shallow water is exactly where microbial mats of this kind are most commonly found today.
Two years later, another team went to the neighboring rock face and examined the same kinds of structures again. This time they did not stop at the exposed surface. They looked at the cones in three dimensions, from three directions. They reached the opposite conclusion.
Cut through one direction, and the structure looks like a triangular cone. Cut through roughly at right angles, and the very same object becomes a long ridge. The ridge runs in the direction in which the surrounding rock was once squeezed and deformed. In other words, it may not have grown at all. It may have been squeezed into shape. There was another, more immediate problem.
Some of the pointed tips faced downward. Microbial mats grow only upward. A downward-pointing tip does not belong to that way of growing.
The original team responded. The two sets of samples, they pointed out, had not come from exactly the same places. Different parts of the rock had experienced different stresses, and fluids had later seeped through and altered them. Those processes, too, could account for irregular shapes.
The dispute is still unresolved. What matters here, though, is not who is right. What matters is the shape of the dispute. Neither side doubts that these rocks are 3.7 billion years old. Neither side doubts that the cones exist. They are studying samples taken less than a meter apart on the same rock face.
Their disagreement begins one step earlier: what sort of shape counts as something life has left behind? To answer that, one must first know what life is. And that has never been settled.
II One Hundred and Twenty-Three Definitions
Someone actually counted them. In 2011, a researcher collected published definitions of life, one by one. He found 123.
Then he did something plodding but useful. He pulled the keywords out of each definition, merged those with similar meanings, and looked at the groups of concepts that remained.
Next he compared which concepts tended to appear together and which encompassed others. He was looking for the smallest common core shared by all those definitions. He arrived at two elements: self-reproduction, and variation.
So he offered an exceedingly spare version: life is self-reproduction with variation. One hundred and twenty-three definitions, pressed into a single sentence.
What happened next is more revealing still. The following year, the same journal organized a whole round of responses. The responses did not converge either.
Some objected to treating every historical definition as an equally weighted piece of evidence. A concept’s frequency does not establish how fundamental it is to a theory of life; counting words is no substitute for explanation. Others challenged the aim of making the definition as short as possible. A minimal definition, they argued, had discarded organization and metabolism, precisely the things that research into life’s origins most needed to understand.
One hundred and twenty-three definitions, then, plus disagreement about what to do with the 123. Lay them out side by side and what emerges? Something much more specific than the observation that there are a lot of definitions.
Aristotle placed the minimum requirements for life at self-nourishment and growth. That boundary fits plants comfortably enough. With a dormant spore, it begins to blur: a spore can go for decades without feeding or growing, then come back to life. With a virus, the definition scarcely knows where to begin.
Around 1800, Bichat described life as the sum of the functions that resist death. It sounds forceful. The difficulty is that death is usually explained as the cessation of life. The definition has walked in a circle and arrived back at itself.
Schrödinger described an organism as avoiding the slide into equilibrium by taking in order from its surroundings. The formulation became immensely influential, but it comes with two difficulties. First, in a note in the very same book, he acknowledged that the strict thermodynamic language would be free energy.
Second, a flame does this too. So does a hurricane. Both maintain local order by continually drawing in matter and energy. The formulation does not mark life off from everything else.
Haldane was more direct. He thought there might be a continuous transition between enzymes, bacteriophages, and life. A bacteriophage, he suggested, was “a step nearer to life” than an enzyme, though calling it fully alive might be an exaggeration. He had not failed to draw the boundary. He was deliberately refusing to draw it.
Another family of definitions changes the question. Instead of asking what a thing is made of, it asks how that thing organizes itself. Imagine a factory whose products happen to be its own machinery and buildings. It even builds the wall around its grounds. When something breaks, it makes a replacement and installs it.
A cell works like that. It makes its own membrane, and it makes the enzymes that make the membrane. Accounts of this kind treat life, first and foremost, as a form of organizational closure.
They are powerful accounts of how an individual keeps itself going. But they do not include heritable variation at the level of a population, or evolution that remains open-ended. They can explain how one cell maintains itself. They cannot explain how a population of cells changes.
The most widely circulated definition comes from NASA: life is a self-sustaining chemical system capable of Darwinian evolution. It emerged from an agency working group in 1992. It was intended to organize research, rather than to sit in judgment over a sample and pronounce it living or dead.
As soon as the word “evolution” appears, someone will think of a mule. A mule cannot produce offspring. Does this definition still allow it to be alive? Worker bees pose the same question: the great majority never reproduce in their lives.
The definition’s answer is that evolution belongs to a population, not an individual. Mules come from horses and donkeys. So long as horses and donkeys reproduce and change, mules remain on the living side of the boundary. “Self-sustaining” is to be read in the same spirit. It does not mean that every individual must gather all its raw materials unaided. It means that the information sustaining the evolutionary process must belong to the system itself.
The final criticism goes furthest. The philosopher Carol Cleland uses water as an analogy. Before chemistry, people could recognize water only by its observable properties: it was transparent, it flowed, it dissolved things.
By that standard, alchemists included nitric acid among the waters and called it aqua fortis, “strong water.” What eventually resolved the problem was not a better definition. It was a theory: water consists of two hydrogen atoms and one oxygen atom.
Life may present the same kind of problem. Until we have a general theory of life, defining it by the outward properties of life on Earth is rather like identifying water by transparency and flow. The definition may well shut a genuinely new example out of the room while admitting things that have no business being there.
Cleland goes a step further. We cannot yet say whether life even constitutes the same sort of category as water, because Earth is the only example we have. We will return to that in Section VIII. If the definition has never closed, it is not for lack of serious attempts to close it.
It is because each attempt either leaves outside something that ought to be inside, or lets inside something that ought to be out.
III The Line Has Never Been Empty
The things left out and the things let in are not imaginary borderline cases. They have names, dates of discovery, and genomes of measurable sizes. And they are pressing toward the same line from opposite directions.
First, consider the descent from the living side. In 1892, a researcher found that the agent responsible for tobacco mosaic disease could pass through filters that held bacteria back. It was smaller than a bacterium. Six years later, another researcher understood it as a transmissible agent distinct from ordinary bacteria. For the next century, viruses remained caught at the edge of the definition of life.
The argument for counting a virus as alive is not that it has a metabolism. It does not. The argument is that it carries hereditary information, generates heritable variation, undergoes natural selection, and evolves over long periods alongside its host.
The objections are just as concrete. A virus particle has no machinery of its own for making proteins, no independent translation system, and, in the overwhelming majority of cases, no self-sustaining energy metabolism. Without a suitable host, it cannot complete a single round of replication.
Under the stronger requirement of a “self-sustaining chemical system,” a virus falls outside life. Under the weaker requirement of heredity and participation in Darwinian evolution, it falls inside. This is not a disagreement about facts. Both sides possess exactly the same facts. They disagree about where to put the line. And below the virus there are simpler things still.
In 1971, a researcher studying a potato disease found an infectious agent simpler than a virus: a piece of nucleic acid with no protein coat. Later measurements showed it to be a ring, about 359 bases in all. It was called a viroid. It depends entirely on its host’s enzymes to reproduce. Yet its sequence can vary, and those variations can be selected.
If reproduction plus variation is enough, it comes very close to life. If independent metabolism or the ability to make its own proteins is required, it plainly does not qualify. There is another step down.
In 1982, the term “prion” was proposed for the infectious protein particles responsible for diseases such as scrapie in sheep. Early experiments found their infectivity unusually resistant to treatments that destroyed nucleic acids. They might, in other words, contain no nucleic acid at all. For the boundary of life, the prion is a sharper challenge than either of the preceding cases.
It has a form of templated reproduction: a misfolded protein can wrench a normal protein into the same wrong shape. There are distinct strains. There is selection. There is adaptation.
There is no nucleic-acid genome. No membrane. No metabolism. If reproduction with variation is the whole of life, the prion becomes a counterexample that is very hard to accommodate.
So much for descending from the living side. From the other side, things have been growing upward. For a long time, there were three convenient rules of thumb for distinguishing viruses from cells.
Viruses were smaller than cells. Their genomes were exceedingly simple. A conspicuous gulf in complexity separated the two. In twenty years, those three rules were broken one after another.
In 1992, investigators looking into a case of pneumonia in England isolated a microorganism from the water in a cooling tower. It was large enough to be seen through an ordinary microscope, and it stained like a bacterium. So they classified it as a bacterium. It stayed one for more than a decade.
Only in 2003 was it recognized as a virus. Its name tells the story: Mimivirus, a virus that mimics a microbe. The complete genome, published in 2004, contained about 1.18 million base pairs and roughly 911 protein-coding genes. That made it larger than the genomes of many bacteria restricted to life inside other cells. It even carried genes previously regarded as distinctly more “cellular.”
One of the pandoraviruses described in 2013 had a genome of at least 2.5 million base pairs. Many of its predicted genes had no recognizable relatives among known organisms. More intriguing still was Pithovirus, revived from the Siberian permafrost.
Its particle was about 1.5 micrometers long, physically larger than many bacteria. Yet its genome contained only about 610,000 base pairs, smaller than Pandoravirus’s. Size and gene count, it turned out, could go their separate ways.
Meanwhile, on the bacterial side, genomes were shrinking. Long-cited records for the smallest bacterial genomes included one of about 159,000 base pairs, and another of about 112,000 base pairs with 137 predicted protein-coding genes.
By 2026, the record had fallen considerably further. One strain of a symbiotic bacterium living inside planthoppers had a genome of just 50,141 base pairs, retaining 68 protein-coding genes with identifiable functions. Sixty-eight genes. For comparison, E. coli has about four thousand. Human cells have roughly twenty thousand protein-coding genes.
And this bacterium, with its 68, still has a cellular structure and cellular machinery inherited from a common ancestor. It has simply outsourced most of its metabolic functions to its host. So this is where matters stand: the viral side grows upward, the bacterial side shrinks downward, and both approach the same line.
There is no vacant space on that line. None at all. Every criterion proposed to separate the two sides, whether size, gene count, autonomy, metabolism, or a translation system, meets examples in the middle that straddle the divide. The line was not drawn and then eroded. It has never been drawn, and the things on both sides have been there all along.
IV The Conditions Were Given, Not Chosen
Take your eyes off the definition for a moment. Turn instead to the time when the thing itself happened.
The solar system began taking shape a little more than 4.56 billion years ago. Earth did not appear all at once. It accumulated through tens of millions of years of collisions, pieces crashing together and joining the growing planet. Along the way came a giant impact: a body the size of Mars struck Earth, and the debris thrown out by the collision gathered into the Moon.
Earth’s first five hundred million years or so are known in geology as the Hadean, after the underworld of Greek myth. Whoever gave the eon its name imagined the surface as a kind of hell. Whether that picture is right remains unsettled. There are still three ways of drawing the scene.
The first is hell itself. After the giant impact that formed the Moon, the surface could briefly have melted into an ocean of magma, weighed down by a high-pressure atmosphere of steam and carbon dioxide. But that is the picture immediately after the collision, not necessarily the one that lasted.
The second is much gentler: Earth acquired oceans very early. The main evidence for this picture comes from a very small mineral. Zircon deserves a moment of introduction.
A grain of zircon is smaller than a grain of fine sand, yet extraordinarily durable. Its parent rock may be heated, squeezed, weathered, and shattered repeatedly while the zircon crystal survives intact. It can carry the chemical information from the moment it crystallized across billions of years. It is therefore one of the few things able to bring us news from rocks that no longer exist.
Researchers measured a collection of these zircons from the Jack Hills in Western Australia, the oldest of them 4.4 billion years old, and found conspicuously elevated oxygen-isotope ratios. We should pause over isotopes, because they will keep returning in what follows.
The same element can come in versions of different weights, with a neutron more or a neutron less. Their chemical properties are almost identical. In many physical and chemical processes, however, the lighter version moves a little faster or reacts a little more readily. After such a process, the ratio of the light and heavy versions may have shifted by an amount that can be measured.
That shift becomes a fingerprint of the process. Now return to the zircons. Their oxygen-isotope ratios are higher than is typical of material from deep within Earth. The explanation is that the magma from which the zircons formed had incorporated crustal material, and that this crustal material had previously exchanged oxygen with water in a low-temperature environment at the surface.
What this supports is the existence, somewhere on Earth’s surface before the zircons crystallized, of a rock cycle involving low-temperature liquid water. What it cannot establish on its own is that a global ocean on anything like today’s scale already existed 4.4 billion years ago.
The objections concern later alteration. These zircons have endured billions of years of radiation damage and fluids moving through cracks. If those processes reset the isotopes, the high readings would not be an original signal. Yet subsequent analyses of tiny regions within the crystals have found that many high-value cores do not correspond to visibly altered zones. So “it is all contamination” does not currently explain all the data either.
The third picture is a world wrapped in a thick blanket of greenhouse gases. The reason lies in the Sun. The young Sun was about thirty percent dimmer than it is today. Under that light alone, Earth ought to have frozen into an iceball.
To keep the seas from freezing, some combination of carbon dioxide, methane, or clouds must have retained the warmth. Three pictures, drawing on the same body of evidence, and all three remain in play. A more immediate disagreement concerns whether the atmosphere was reducing or neutral, because that determines how far its chemistry could go.
In 1953, a graduate student named Miller at the University of Chicago performed the famous experiment. He sealed methane, ammonia, hydrogen, and water vapor inside glassware and used electric sparks to imitate lightning. This was a strongly reducing mixture, rich in substances that readily give up electrons.
On the first day of sparking, the solution turned pink. After a week it was deep red, and amino acids such as glycine and alanine could be detected in it. The building blocks of proteins had made themselves out of a flask of gases. At the time, there were reasons for choosing that particular mixture.
Later, however, researchers used the oxidation state of cerium in Hadean zircons to work backward to the oxidation state of the magma from which they had formed. They found no extreme shift toward reducing conditions compared with the later mantle. That supports a picture of volcanic gases dominated by water vapor, carbon dioxide, and nitrogen: an atmosphere much closer to neutral.
But a neutral atmosphere is not the end of the story either.
Calculations suggested that differences in the composition of incoming meteoritic material could generate strongly reducing atmospheres locally or for a limited time. Another group repeated the spark experiment under more nearly neutral conditions and found that amino-acid yields rose sharply once they removed the oxidizing substances produced by the same electrical discharge. Part of the problem, it turned out, was that the products were being destroyed by their own by-products.
The argument today is therefore not “Miller was wrong; the early atmosphere could never have been reducing.” It is that the long-term global atmosphere may not have been so strongly reducing, while impacts, volcanic plumes, and local hydrothermal activity could create reducing conditions within limited spans of space and time. The consequences extend beyond how much product a reaction yields.
The oxidation–reduction state changes the entire stock of available raw materials. A strongly reducing atmosphere can produce large quantities of compounds such as hydrogen cyanide directly in the air. A more neutral atmosphere depends more heavily on local environments, iron chemistry, and deliveries from meteorites. No one arranged these conditions into the configuration they eventually took.
The mantle’s oxidation state was not chosen. The Sun’s brightness was not negotiated. No one voted on which kinds of meteoritic material would accrete onto the planet.
No intention. No preference. No negotiation. Yet this set of conditions made a cut through the space of chemical possibilities, leaving some available and ruling others out. This series calls that cutting “chiseling.” Ordinarily, when we speak of chiseling, we picture someone holding the chisel. Behind this cut there was no one.
There is one more matter, not directly about life, that determines how we should read all the ancient records still to come. In 1974, three researchers noticed that many isotope ages in the lunar rocks brought back by Apollo crowded around 3.9 billion years ago.
They proposed that the Moon had experienced a brief, violent peak in impacts at that time. During the latter part of the twentieth century, this late lunar cataclysm became the standard story. In 2005, it was linked to a model in which the giant planets rearranged their orbits.
Over the past decade or more, the case has weakened considerably. The Apollo samples may contain material thrown out by a single impact basin, while lunar meteorites and new dating work produce a much broader spread of ages.
Apollo visited only the Moon’s near side. Analyses reported in 2026 of the samples Chang’e-6 brought back from the far side in 2024 likewise found no clustering of ages at 3.9 billion years. But another piece of modeling work matters more.
It points out that ancient impact melts are continually destroyed by later collisions, while the isotope clocks inside them are repeatedly reset to zero. Under those conditions, even a smoothly declining impact rate can leave surviving samples whose age distribution appears to contain a peak.
The peak, in other words, may not have been made by the events. It may have been made by the odds of preservation. That sentence is worth setting apart in the mind. It is not a minor point about impact history. It is a general rule for every ancient record:
The shape of the distribution we can see contains a mixture of two things: what happened, and what could survive. And those two things cannot always be disentangled.
For the origin of life, this opens up some room. If the peak was not real, we need not assume a bombardment 3.9 billion years ago that wiped the slate clean.
Add the zircons’ evidence for liquid water 4.4 billion years ago, and life may have had a much longer interval in which to emerge. Impacts could have interrupted the process without forcing it to start from nothing every time. Of course, that tells us only that there may have been enough time. It does not tell us that life was already there.
V A, B, C: Three Levels of Evidence
Return to the question raised by the cones: what kind of record counts? Researchers approach the earliest evidence of life by sorting it into three levels. The ordering concerns how firmly the evidence stands, not how far back it reaches.
Level A is the least disputed. On an ancient block of crust in Western Australia lies the Dresser Formation, a sequence of rocks about 3.48 billion years old. Its strength does not rest on any single “shape that looks like a bacterium.”
It comes from several independent kinds of evidence coming together: stromatolite-like layering, carbon-bearing material, sulfur- and carbon-isotope anomalies, and microscopic structures. Work published in 2017 also identified rock deposited by geysers, suggesting that at least part of the setting may have consisted of volcanic hot springs on land.
On the same ancient block lies another sequence, the Strelley Pool Formation, about 3.43 billion years old. It preserves a stromatolite reef extending over a large area. Here, too, the central argument is not that an isolated cone happens to resemble a modern stromatolite.
It is that cones, domes, and layered forms are distributed in regular patterns within the same sedimentary system, recurring through many strata. A single mechanism of inorganic precipitation has great difficulty explaining all those forms and their spatial relationships at once.
A reasonable statement of Level A is this: by about 3.5 to 3.48 billion years ago, microbial communities were leaving sedimentary structures and carbon-bearing records. The consensus is strong. The questions concern the particular environments and metabolisms involved, rather than whether Earth had life at all.
At Level B, the record itself exists, but its biological interpretation is substantially disputed. In seafloor sedimentary rocks at Isua more than 3.7 billion years old, researchers have reported particles of graphitic carbon about two to five micrometers across. Before examining those, we need to be clear about why carbon isotopes can point toward life. Much of what follows depends on it.
When living things take up carbon, they favor the lighter version. The organic matter they produce therefore has a different balance of light and heavy carbon from the inorganic carbon around it. It contains more of the light kind.
That shift can be measured in rock. On one condition: the carbon really must be what organisms left behind at the time. It must not have seeped in later, and it must not have acquired the same shift through a nonbiological process.
Back to the graphite. Its carbon isotopes are weighted toward the light side, which has been interpreted as the residue of original organic matter altered by high temperatures. In 2025, measurements made in place on graphite enclosed and protected within mineral grains found somewhat larger shifts than before. The objections are equally specific.
Critics point out that the high temperatures and pressures of those early environments could themselves produce isotopically light graphite through the breakdown and reduction of carbonates, precipitation from fluids, and metamorphic reactions. The same fingerprint, in other words, can be left by processes that have nothing to do with life.
A more famous Level B case concerns a set of microscopic filaments, and its history deserves to be followed all the way through. In 1993, researchers reported carbon-rich filamentous structures from the Apex chert of Western Australia, about 3.465 billion years old. They divided them into eleven morphological groups and interpreted some as prokaryotes resembling filamentous cyanobacteria.
For years, these structures were consequently described as the oldest microfossils then known. Nine years later, two papers appeared in the same issue of Nature, arguing directly against one another.
One side illuminated the samples with a laser and read their molecular vibrations, confirming that the structures did contain highly mature carbonaceous material. The other reexamined the geological setting and argued that the specimens came from veins formed by later infusions of hot fluid, rather than undisturbed original sediment. In addition, the outlines and size distributions of many supposed “cells” did not meet the standards for reliable microfossils.
The dispute continued for more than another decade. In 2018, the first side used a technique capable of measuring isotopes within exceedingly small regions to examine eleven individual structures. Five morphological groups yielded distinct carbon-isotope values that matched the groups identified by shape. The researchers suggested that different forms of metabolism might be represented. That did not end the dispute. The reason fits into two sentences.
A laser can establish that something is carbon; it cannot establish that the carbon used to be a cell. Light carbon-isotope values can establish that fractionation occurred; they cannot rule out chemical reactions in hot water that also produce isotopically light carbon.
Level C comprises claims supported by only one line of research, still lacking independent confirmation, or already facing strong rebuttals. The Isua cones from Section I belong here.
Another example is a set of tubes and filaments from Nuvvuagittuq in northern Quebec, Canada. A 2017 paper reported them in iron-rich rocks interpreted as deposits from seafloor hydrothermal activity. The authors suggested that they resembled traces left by modern iron-oxidizing bacteria.
The proposed age range is enormous. The authors themselves gave a minimum of about 3.77 billion years and a possible age of about 4.28 billion. There are five hundred million years between those two numbers.
The objections include whether the older date actually records the formation of the rock. Systems of iron and silica can also produce tiny tubes and branching structures through inorganic mineral growth. Moreover, several of the reported indicators come from the same set of samples. They are not independent of one another.
Once the three levels are laid out, a prior question becomes available: what are the standards of judgment themselves?
To decide whether a carbon-isotope signal was left by life, several matters have to be considered together. Is the size of the shift compatible with known biological fractionation? Did the carbon form at the same time as the original sedimentary structures, or did it arrive in later veins? Could alteration at high temperature have reset the isotopes? Can a nonbiological model generate similar values under the same conditions?
One carbon-isotope value on its own cannot exclude a nonbiological origin. Deciding whether a filament was left by life asks still more. Its dimensions must show some regularity. Cells grow within characteristic size ranges, so the diameters of a group of cells ought to cluster within a range, rather than form an unbroken spectrum from the finest to the thickest.
There should be regular partitions and branching. If a filament consists of cells joined end to end, one should be able to see the crosswalls between them. The carbon must be original, rather than something that later seeped in from elsewhere. The structure must be in its original setting, rather than washed there by a current.
Cut it open from another direction, and its shape must still make sense. We have already met this requirement in Section I: it was where the Isua cones came to grief. Nor should pure chemistry in a laboratory readily reproduce the same object. Yet not one of these criteria, taken alone, can completely exclude a nonbiological structure.
Experiments have already shown that systems of silicates and carbonates can grow astonishingly convincing imitations of life. They branch. They form filaments. They look exactly like cells. And at no stage in their formation have they had anything to do with life. At the heart of both the Apex chert and Isua disputes lies this simple point: resemblance of form does not establish sameness of origin.
A set of criteria is a construct. This series uses “construct” for things of this kind: things people make so that they can judge and compare. A construct has boundaries, specified requirements, and a checklist that can actually be used.
And it does work. Those criteria are precisely what have kept the 3.7-billion-year-old cones out of Level A. But what the criteria cannot do is just as clear. They can rule out. They cannot confirm.
VI Seven Roads, and the Final Row Is Empty
So far, the question has been how to recognize life. Now turn it around: how did life begin? Before considering the proposed routes, we need to look at a famous deadlock.
In today’s cells, two kinds of molecule are crucial. One keeps the record, passing information from generation to generation: DNA and RNA. The other does the work, catalyzing chemical reactions so that processes which would otherwise be hopelessly slow run fast enough to be useful: proteins, and particularly enzymes. The two kinds depend on one another.
Copying DNA requires enzymes. But enzymes are proteins, and the order of their amino acids is recorded in DNA and conveyed through RNA before they can be made. Without enzymes, no accurate copies; without accurate copies, no enzymes. That is the famous circle. Any account of life’s origins must first explain how it proposes to break it.
Now consider the routes. They are not different versions of one theory, differing only in details still to be worked out. Their answers to “What came first?” exclude one another. Was it a molecule that could copy itself, a network of reactions that could sustain itself, or an energy gradient? Choose one, and the others have to be rewritten.
The first route begins with a molecule that can copy itself. It later became known as the RNA world. The way the idea arose was unusual: it was inferred backward.
In the late 1960s, several researchers independently noticed peculiar traces preserved in the genetic code and protein-making machinery of modern cells. RNA’s role was far larger than one would expect of a mere messenger.
They proposed that an early stage might have existed in which RNA held the central role and the machinery for making proteins had not yet fully developed. That hypothetical stage was later named the RNA world. The order matters. When they proposed it, no one had yet seen an RNA molecule catalyze a reaction.
More than a decade later, two laboratories made separate discoveries: some RNA could splice itself, and in a certain enzyme the component actually doing the catalytic work was RNA. The two researchers shared the 1989 Nobel Prize in Chemistry. RNA molecules capable of catalysis are called ribozymes.
The conjecture, reached by reasoning backward, had found experimental support. And the discovery broke half of the deadlock: in principle, the same kind of molecule could both store information and catalyze reactions. Proteins did not have to come first. What it did not explain was where the first sufficiently long, replicable strand of RNA came from.
In 2009, researchers made a substantial advance. They constructed a pair of ribozymes that catalyzed one another’s formation. A joined two shorter strands to make B; B, in turn, joined two shorter strands to make A. With a continuing supply of raw materials, the system could amplify exponentially, reaching an increase on the order of a hundred millionfold in about thirty hours. In mixtures of different variants, selection and recombination could also occur.
This was one of the earlier laboratory systems to approach Darwinian evolution at the molecular level. But there was a limitation. It did not begin with individual building blocks and copy its own full sequence one base at a time. It joined longer fragments that the researchers had synthesized in advance.
Another family of ribozymes took the harder route: copying other RNA strands base by base. These grew steadily more capable. A 2011 version could copy functional RNA about 95 bases long. A 2020 version copied its own evolutionary ancestor in three pieces. Yet the ribozymes themselves were nearly 200 bases long. They could copy other strands, in some cases strands longer than themselves, but they still could not copy themselves.
Only in 2026 did the first crack appear. A laboratory in Cambridge found a ribozyme just 45 bases long that assembled a copy of itself in ice, and also assembled the complementary strand that pairs with it. Seventy-two days, for a yield of only about two parts in a thousand. The two steps had still not been joined into a complete replication cycle.
Where the raw materials came from is another difficulty. The traditional approach was to make the sugar and the base separately, then join them. That joining step proved extremely difficult in experiments. Researchers found a way around it, starting with simpler precursors and passing through shared intermediates directly to usable building blocks. They later proposed an entire chemical network in which the same starting materials could yield precursors for nucleic acids, amino acids, and lipids together.
The difficulty is geological. In the laboratory, a researcher changes the flasks step by step: do this reaction, concentrate the product, move on to the next. Nature would need the corresponding pools, flows of water, and alternations of wet and dry to change the flasks for them. This route has established that a chemical pathway exists. It has not yet established that any particular place in the Hadean actually ran that pathway from beginning to end.
The second route begins with a network of reactions able to keep itself running. Its opening sentence already contradicts the first route. The earliest thing, it proposes, was not a molecule capable of replication, but a set of chemical reactions that catalyzed one another and sustained their own operation. Information was added later.
The concrete proposal is that, on mineral surfaces such as iron sulfide, nickel sulfide, and pyrite, energy from the reduction of carbon dioxide and the oxidation of sulfides first supported self-maintaining reaction networks fixed to those surfaces. Hereditary molecules were tenants who moved in afterward.
The experimental progress has been substantial. Under conditions involving iron and nickel sulfides, researchers have produced activated acetate and thioester compounds from carbon monoxide. A series of studies from 2018 onward used metallic iron and iron–nickel minerals as catalysts to make acetate, pyruvate, and citramalate from carbon dioxide. These are all compounds found in the central metabolism of cells today.
Together, the experiments establish something of real weight. Several central metabolic transformations that seem to demand modern enzymes do not absolutely require protein enzymes under suitable mineral and reducing conditions.
What they have not established is a closed, self-sustaining, enzyme-free metabolic system able to regenerate its own catalysts. And the question this route still has not answered is how the state of a metabolic network is copied into the next generation.
The third route borrows an energy difference already there. It can sound like the claim that life began in hot water, but it is much more specific than that, and rather ingenious. The first two routes have to find some way of securing energy. This one says there is no need to make your own supply. The seafloor has one ready.
When certain kinds of seafloor rock react with water, they produce alkaline hydrothermal fluids rich in hydrogen. A porous mineral wall separates these fluids from the more acidic water of the early ocean. A chemical gradient therefore arises naturally across the wall. At the same time, the walls of the mineral pores provide both compartments and catalytic surfaces.
The earliest energy use would have borrowed this ready-made gradient; only later would life have evolved pumps of its own. A hydrothermal field discovered in 2000 near the Mid-Atlantic Ridge offers a modern analogue. Its fluids are alkaline and rich in hydrogen and methane, and its chimneys are built mainly of carbonates. It is called Lost City.
This is quite different from the familiar “black smokers.” Those are driven by magma and usually discharge acidic fluids at around 350 degrees Celsius, a temperature that would rapidly break RNA apart. The gradient required by this proposed route comes instead from the interface between alkaline fluid and relatively acidic seawater.
The strongest objections concern dilution and phosphorus. The seafloor is an open environment, and small molecules formed there can easily disperse into the enormous reservoir of the ocean. Alkaline, calcium-rich conditions also tend to precipitate phosphate out of solution, while nucleic-acid and membrane chemistry ultimately depend heavily on phosphorus.
Then there is a particularly difficult threshold: how does a system leave that ready-made natural gradient behind and replace it with one maintained across its own membrane?
The fourth route begins with a pool that repeatedly dries and fills. Here the setting changes. We leave the deep sea for the edges of shallow volcanic hot-spring pools on land. What the route exploits is not a particular substance, but a rhythm.
The pool fills, evaporates, and fills again. During the dry phase, dilute solutions become concentrated and molecules are driven to join into chains. When water returns, oily molecules spontaneously form little bubbles, randomly enclosing some of the newly made long chains. The most direct disagreement with the deep-sea route is not whether the setting is hot. It is how much water there is.
Salt lies at the center of the dispute. Simple fatty-acid bubbles readily fall apart in high concentrations of magnesium and calcium, and at higher salinities. Yet magnesium is also required by many RNA catalysts. This route therefore favors low-salt freshwater hot springs over seawater as a setting for primitive membranes.
Its unanswered questions are just as specific. How much land was actually exposed in the Hadean? Could suitable volcanic pools persist through enough cycles of wetting and drying? Intense ultraviolet light and complete desiccation could also destroy the long chains that had already formed.
There are several subsidiary routes, each strong at one point and empty at another. One proposes that assemblies of oily molecules can themselves carry a kind of “compositional heredity.” Compartments and growth come naturally to this picture. Its weakness is that the information capacity is far lower than that of molecules with a sequence.
Another proposes peptides first. Amino-acid chemistry and catalysis through metal binding offer abundant possibilities, but there is no clear rule for templated replication comparable to base pairing. Then there is the clay-template route. It has shown that minerals can promote the formation of long chains. It has not shown that a crystal lattice can provide heredity with sufficient information capacity.
Panspermia deserves a separate word, because it is often mistaken for an answer. The part backed by evidence is real: organic compounds can arrive from space. Certain meteorites contain abundant amino acids, and a small excess of left-handed molecules has been measured in the Murchison meteorite.
But the logical limit has always been clear. If life arrived from somewhere else, the question becomes how it first arose there. Panspermia changes the location. It does not explain the transition.
Place these routes side by side and a neat pattern emerges. Each is strongest at a particular threshold, and that threshold is often exactly where another route has a gap.
For compartments, alkaline vents offer mineral micropores, while the hot-spring route offers bubbles that form spontaneously when water returns. For energy, metabolism-first and alkaline-vent accounts are strongest. For hereditary information, only the RNA world has a firm footing, and that is precisely the central weakness of the preceding two. As for why biological molecules all have the same handedness, none provides a complete explanation.
Look only at the upper half of this comparison, and the routes seem between them to cover almost the whole problem. The trouble lies in the final row. That row is the end-to-end experiment: begin with geologically plausible, simple raw materials and demonstrate the whole passage to autonomous replication, compartments, energy coupling, complete division with inheritance, and sustained, open-ended evolution.
RNA world: none. Metabolism first: none. Alkaline vents: none. Terrestrial hot springs: none. Lipid world: none. Peptides first: none. Clay templates: none. Not one. The longest chains stop at different places, and the places where they stop do not join up.
VII After Cutting Back to 473 Genes
If we cannot work our way backward to it, let us try the other direction. Build one. Craig Venter’s team has gone furthest along that road. The first step was to assemble an entire bacterial chromosome from chemical reagents.
Not copy it. Synthesize it. Short pieces of artificially made DNA were joined in stages, like building blocks, until the researchers had the complete genome of a mycoplasma: about 583,000 base pairs. In 2008, they succeeded. And then? Then nothing happened.
The DNA lay in a test tube. It was a correct, complete chromosome, every base open to inspection and verification. It did not move, metabolize, or divide. It was a set of instructions. And instructions do not build their own house.
Two years later, the team tried a different approach. They synthesized a genome and transplanted it into a cell that was already alive. After the artificial genome went in, the cell divided through several rounds, followed by selection, until the recipient’s original chromosome was lost and only the artificial one remained. This time it worked. The synthetic genome took over the cell. Its descendants displayed the proteins and physical form of the donor species.
This was widely reported as “artificial life.” It was not, for a very specific reason. Only the genome had been synthesized. The entire apparatus that started it up came from the cell receiving the transplant: the membrane, the protein-making machinery, the transport proteins, the metabolic enzymes, the ion gradients, every last piece of equipment in the cytoplasm.
And that cell was alive. Its ancestry ran in an unbroken line of living cells all the way to the laboratory bench. The artificial genome took the wheel of a car already running. Without the recipient cell, the DNA would not have turned itself into a cell.
If building from nothing is so difficult, try the reverse. Take an existing cell, delete its genes one at a time, and see how far it can be slimmed down while remaining alive.
The team took that route too. In 2016, the minimal cell appeared: roughly 530,000 base pairs, 473 genes, able to grow and divide on its own under nutrient-rich conditions.
For comparison, the planthopper bacterium discussed earlier has just 68 recognizable protein-coding genes, whereas the minimal cell has more than four hundred. But the bacterium is supported by its host. At the time, 473 was the smallest known gene count in an autonomously replicating cell. That autonomy still depended on a rich laboratory medium; it was not a universal minimum for life. Then came the most counterintuitive sentence in the whole study.
Of those 473 genes, about thirty percent had no specific function anyone could identify. This was not the polite uncertainty conveyed by “the mechanism remains unclear.” It meant that after extensive deletions under those culture conditions and that experimental design, roughly a third of what remained was still of unknown function.
The researchers had pared the cell down as far as their design then allowed. And looking at the small handful left over, they could not say what more than a hundred of those genes were for. Ten years later, in 2026, some fifty or sixty still had no known purpose. There is an implication here that is easy to slide past.
The lesson is not “we are almost there.” It is that even knowing every essential function identified at the time would not have allowed us to design this minimal genome. It was arrived at by deletion, not by calculation.
Later derivatives had to have a few genes put back to improve their abnormal size and patterns of division. The fewest genes needed to reproduce and the fewest genes needed to divide normally turned out to be different questions.
Another team approached from the opposite direction. Instead of taking things away from a cell, they built upward from the simplest chemistry. What they managed to build was considerable.
In suitable water, simple oily molecules spontaneously assemble into little bubbles that enclose larger molecules. Add more of the oily molecules, and the bubbles take them up and grow. Bubbles drawn out into long filaments can be broken by gentle stirring into many daughter bubbles, without spilling what is inside.
Their membranes are selective, too. Small raw-material molecules can pass through, while the larger nucleic-acid templates remain inside. Supplies can enter without the templates escaping. In 2013, the researchers directly demonstrated enzyme-free, template-directed RNA synthesis inside such a model protocell.
They even produced competition. If the long chains inside a bubble become more abundant, its internal pressure rises. It then takes more membrane material from its surroundings, stealing membrane from its neighbors.
A little pouch that grows, divides, permits copying inside, and competes with its own kind for resources. It sounds as though only one step remains. The missing step is to join those achievements into a circle.
A complete genome is copied inside the bubble. The two strands separate. The membrane grows at the same time. The bubble reliably divides in two, each daughter receiving a working hereditary system. Then the daughters repeat the entire process. And the whole arrangement must use simple raw materials from outside to regenerate for itself the catalysts that copying requires.
That circle has still not closed. The gap is not a vague “we have a long way to go.” Its missing pieces can be named one by one. Autonomy over supplies. In the strongest current replication systems, the experimenter puts in the raw materials and membrane components directly. Even the energy required to join the molecules is fed to the system.
Strand separation. Copying a template produces a complementary strand, and the two stick together. Modern cells use specialized proteins to pull them apart. Without enzymes, separating them gently enough not to break the RNA is a hurdle in its own right. The error threshold. As sequences grow longer, copying errors accumulate. Beyond a certain point, all functional information is lost.
Regeneration of catalysts. A truly self-sustaining system must reproduce the very catalysts needed for reproduction, rather than depend on the experimenter to replenish them in every generation.
Now place several of the strongest systems beside the working definition and check them against it. Cross-catalytic ribozymes have exponential replication, heritable variation, and selection.
They have no metabolism and no membrane, and their raw materials are precisely prepared by the experimenter. Fatty-acid bubbles have compartments, uptake, and some template copying. They do not have a complete, closed cycle of replication and division.
RNA replication systems with translation machinery can undergo Darwinian evolution inside cell-like compartments. But they consume large supplies of prefabricated protein-making machinery. Those mature biological components are precisely what an attempt to rebuild life from nothing is supposed to explain. Each system occupies part of the definition. Not one occupies all of it at once.
To this day, no experimental system starting from nonliving raw materials has simultaneously achieved self-sustaining chemistry, fully autonomous reproduction, and sustained Darwinian evolution.
VIII A Sample of One, and Tools Made in Its Image
How easily, then, does this happen? A familiar intuition runs as follows: life seems to have appeared on Earth quite soon after the planet became habitable, so its origin probably cannot have been very difficult. There is a specific flaw in that reasoning.
We are not observing a planet drawn at random from a collection of planets. We are observing one that has already produced observers capable of asking the question. The conditions of the sample already include the requirement that life must have appeared here early enough.
In other words, any planet on which someone can stand and ask the question will make the origin of life look quick.
Researchers have treated the origin of life approximately as a random process with a constant rate and subjected the argument to statistical analysis. Once the condition that observers require life to have existed for billions of years is included, even a very low underlying rate of origin remains compatible with finding ourselves in a successful case where life began early.
Numbers have nevertheless been offered. The most optimistic estimate put the probability of life arising on an Earth-like planet within a billion years at a minimum of a little over ten percent. But it treated Earth as though it were a randomly selected sample, committing exactly the mistake just described. Another estimate used a more careful method and reached a much weaker conclusion: the data favor a rapid origin by odds of at least roughly three to one. That is some evidence. It is nowhere near a verdict.
So there is still no defensible number that can be called “the probability of life arising.” And the phrase “a sample size of one” has another layer to it.
All known cellular life shares the same core apparatus: protein-making machinery, an almost universal genetic code, the same core amino acids, the same handedness in its proteins, and the same energy currency. The millions of species living on Earth cannot therefore be counted as millions of independent events in which life began.
For the question of origins, they most likely represent a single sample left by a single origin. One.
In 2005, Cleland, the philosopher who used the water analogy in Section II, and a microbiologist proposed a possibility. In 2009, she and several other researchers developed it more fully in a joint paper.
If life could arise once under the conditions of early Earth, perhaps it arose more than once. Descendants of a second origin, using a different biochemistry, might still live alongside the life we know. Yet microbiological methods aimed at DNA and ribosomes could miss them altogether. This is the possibility of a shadow biosphere.
The possibility has never been systematically tested. And not because no one wants to test it. The detection tools themselves were made to fit the one kind of life already known.
The methods used to amplify DNA require nucleic acids that the corresponding enzymes can recognize. Universal primers assume that their targets have ribosomes. A culture medium is itself a selector: the very phrase “unculturable microorganisms” tells us that existing culture conditions cover only a small portion of ecological space. Extracting genetic material directly from the environment and sequencing it still assumes that the material consists of standard nucleic acids that can be extracted and sequenced.
Every tool decides, before it begins to look, what the thing it seeks must look like. If a second kind of life really exists, it will not appear on the screen as an unfamiliar sequence. The screen will be blank. Nothing detected. Failing to find it will look exactly like its not being there.
Conversely, finding something strange is not enough. There was such a case in 2010. One of the authors of the 2009 paper cultured a bacterium from a salt lake in California and announced that it could substitute arsenic for phosphorus, incorporating arsenic into its DNA.
NASA held a press conference. Two years later, two laboratories independently repeated the work. Neither could confirm arsenic in the bacterium’s DNA. In 2025, the journal that had published the original paper retracted it.
And even if the bacterium really had used arsenic, that would still not have been enough. It could still have represented an extreme adaptation within known life, sharing an ancestor with us.
To establish a second origin, we would have to rule out common ancestry and also rule out an exchange of genes with known life. That is a far harder task than finding something strange.
Beyond Earth, the closest approach to finding life came in September 2025. The previous year, a Mars rover had photographed a scatter of little spots on a rock. Each was about a millimeter across, dark at the edge and pale at the center, like the markings of a leopard.
They were described as the clearest potential signs of life yet found. The word “potential” cannot be removed. Such spots might form without life, though that would usually require temperatures of one or two hundred degrees Celsius, and the researchers found no evidence that this rock had been heated. The position is the same as with the filaments in Section V: nonbiological explanations have not been fully excluded, and the biological explanation cannot yet be confirmed.
As of August 2026, there was still no confirmed sample of life from Mars, Europa, or Enceladus. There is strong evidence for a salty ocean beneath Europa’s ice. A spacecraft launched on October 14, 2024, is expected to reach the Jupiter system in 2030. Its question is whether Europa is suitable for life, rather than whether life is there. The mission is Europa Clipper.
The real value of such exploration does not lie in locating one more habitable place. Only a second sample shown to have originated independently of life on Earth would fundamentally change both the statistical problem and the problem of definition. Finding water, organic matter, or an energy source cannot turn a sample of one into a sample of two.
Return to the cones in Section I. Both sides of the dispute use the same set of criteria. Those criteria are a construct that genuinely works: it has specified requirements, it can be applied, and it really does keep some things out.
And it does shift. The same cones were described as stromatolites in 2016, cut open from another direction in 2018, and defended again in 2019. This construct has not closed, and the reason is not that the people involved have been insufficiently careful.
It is that the thing the construct is trying to enclose has no edge of its own. After 123 definitions, the viruses are still there, the viroids are still there, the prions are still there. Mimivirus grows upward. The 68-gene bacterium inside the planthopper shrinks downward. And the line has been full all along.
The criteria keep sifting. The rocks keep being cut open again. In the laboratories, the chains keep being joined, piece by piece. The line of descent has not yet been joined up. It is still dividing.