One Crucible, Two Origins

The origin of life is often pictured as a single miraculous spark in a warm little pond. The reality is far more industrial and dramatic. Recent headlines announcing two origins of life can easily be read as claiming that life began twice, independently. What the work actually describes is narrower and stranger: two distinct emergences of free-living cells from one original crucible — a labyrinth of microscopic pores inside deep-sea serpentinizing alkaline hydrothermal vents.

The transition from dead rock to free-living cells involved a series of compounding explosions in capability. Chemistry came alive through the following stages.

---

1. Abiotic chemosynthesis

References: Martin & Russell (2003), Philosophical Transactions of the Royal Society B; Martin (2020), Frontiers in Microbiology.

Deep under the ocean, alkaline fluids rich in hydrogen met slightly acidic seawater inside towers of iron and sulfur. Martin and Russell argued that these structures — built largely of iron- and nickel-sulfide minerals such as mackinawite and greigite — provided two things at once: semi-enclosed micro-compartments on the scale of a cell, and a natural proton gradient across their mineral walls. Serpentinization makes the fluid inside alkaline, around pH 9–11; the Hadean ocean outside was acidic, perhaps pH 6, because so much carbon dioxide was dissolved in it. That is a gradient of three to four orders of magnitude, with the same polarity that living cells maintain today: alkaline inside, acidic outside.

Those microscopic pores acted as microfluidic continuous-flow reactors. Hydrogen and carbon dioxide flowed over bare metal catalysts. Everything in this section happens without a single enzyme.

1.1 Carbon fixation

References: Fuchs (2011), Annual Review of Microbiology; Martin (2020), Frontiers in Microbiology.

Six routes into organic carbon are known among prokaryotes. Fuchs set out why one of them stands apart. The reductive acetyl-CoA pathway is the only one that is linear rather than cyclic, the only one found in both bacteria and archaea, the only one that proceeds via carbon monoxide as an intermediate — and, critically, the only one that runs downhill. Combining carbon dioxide with hydrogen to make acetate releases free energy rather than consuming it. Martin points out where that asymmetry comes from: the pathway generates carboxyl groups from carbonyls, whereas the other five spend energy reducing carboxyls. It is also the only pathway that yields net ATP rather than demanding it.

That matters enormously at the start, because there is no ATP yet to spend.

Then comes the punchline. In a modern acetogen, this pathway takes roughly ten enzymes, about as many organic cofactors, and over 500 kilodaltons of combined subunit mass. Yet Martin reports that a single naturally occurring hydrothermal alloy — awaruite, Ni₃Fe, both metals in the zero-valent state — converts hydrogen and carbon dioxide into formate, acetate, pyruvate and methane overnight under mild hydrothermal conditions, entirely on its own. Half a megadalton of protein machinery replaced by a lump of metal.

The reactions, in other words, were there first. Enzymes do not shift equilibria; they accelerate reactions that already tend to happen. The acetyl-CoA pathway is older than the genes that encode it.

1.2 Abiotic nucleotide synthesis

Reference: Martin (2020), Frontiers in Microbiology.

From acetyl-CoA and pyruvate, further carbon dioxide incorporations build oxaloacetate, 2-oxoglutarate and sugars — the skeletons from which all twenty amino acids derive. Nucleobases come next, and they come from amino acids: purines and pyrimidines are assembled from glycine, aspartate and glutamine plus one-carbon units and carbon dioxide.

This is worth stating plainly, because it is the strongest structural argument for the whole vent scenario. Real metabolism does not build its bases by condensing formaldehyde, or cyanide, or oxidised methane. It builds them by sequential addition of CO₂. A theory that starts from CO₂ meets central metabolism seamlessly; theories starting from soup ingredients have to graft themselves on.

Honesty requires a caveat here that the carbon-fixation story does not need. Formate, acetate and pyruvate from H₂ and CO₂ over metals is demonstrated experimentally. The full abiotic route onward to ribose, nucleobases and phosphorylated nucleotides is not — it remains the least-supported link in the chain, and anyone telling you otherwise is selling something.

1.3 Fixed nitrogen, taken from the environment

Reference: Boden et al. (2026), Science Advances.

Amino acids and nucleobases need nitrogen, and early networks got theirs the easy way: ammonia, hydrogen cyanide and nitriles produced abiotically in and around the vent.

They did not fix it themselves, and this is where an intuitive ordering goes badly wrong. Breaking the triple bond of atmospheric N₂ is one of the hardest reactions biology has ever mastered, and it was mastered late. Boden and colleagues make the timing explicit: LUCA carried genes for importing ammonium, but the first organisms with the full catalytic nitrogenase gene set appear at least a billion years later. Abiotic nitrogen sustained the biosphere for several hundred million years after life began.

(One live disagreement to flag: Weiss and colleagues' 2016 reconstruction, which this post leans on heavily in section 3.2, does place nitrogenase in LUCA's gene set. The more recent phylogenetic and molecular-clock work runs the other way.)

1.4 Abiotic phosphorylation

References: Mrnjavac et al. (2026), Science Advances; Whicher et al. (2018), Origins of Life and Evolution of Biospheres; Pinna et al. (2022), PLOS Biology; Martin (2020), Frontiers in Microbiology.

The acetyl-CoA pathway runs without ATP. Almost nothing else does. But ATP is a complicated molecule built by enzymes — and its own synthesis consumes six ATP-driven phosphorylation steps. Something simpler had to come first.

Mrnjavac and colleagues found the mineral half of the answer in phosphite, a reduced form of phosphorus that occurs in serpentinizing systems. In their experiments nickel oxidised phosphite to phosphate and hydrogen at high yield, while palladium proved the better catalyst for attaching phosphate groups to organic molecules — converting AMP to ADP, and serine to phosphoserine. Phosphite plus a native metal substitutes for ATP plus an enzyme.

The organic half comes from acetyl phosphate. Whicher and colleagues showed it forms readily in water under mild alkaline-hydrothermal conditions; Pinna and colleagues then found it converts ADP to ATP at roughly 20% yield in water, with startling selectivity — it requires ferric iron specifically, and it will not phosphorylate any other nucleoside diphosphate. ATP may be universal because its formation is chemically favoured, not because anything selected it.

Martin adds a corrective that is easy to miss and changes how you should picture all of this. In substrate-level phosphorylation, the energy in the high-energy bond does not come from phosphorus. It comes from carbon. Inorganic phosphate is an unreactive bystander that performs nucleophilic attack on a reactive carbonyl generated during CO₂ reduction. This cuts against a long tradition — going back to Lipmann — of looking for energy-rich phosphorus minerals in the environment, an idea Martin notes has no counterpart in any living cell: no organism grows chemotrophically on pyrophosphate or polyphosphate.

So the picture is not "reactive phosphorus attacks inert carbon." It is the reverse. The vent's continuous, exergonic reduction of CO₂ throws off reactive carbonyls, and phosphate simply latches on.

---

2. Chemical evolution: templates, catalysts, and the retreat from the rock

Calling what follows "autocatalytic" needs a caveat, because the closure was never complete. These networks made more and more of their own catalysts, but they never stopped depending on the mineral surfaces around them — not through the ribozyme world, not at LUCA, not until the first free-living cells. What begins in this section is not a handover but a slow displacement, and it runs for the rest of the story.

2.1 When the products become the catalysts

References: Martin (2020), Frontiers in Microbiology; Mrnjavac et al. (2026), Science Advances; Koonin & Martin (2005), Trends in Genetics.

Everything so far has been driven from outside: metal surfaces, a geological proton gradient, a supply of gases. The system becomes interesting the moment some of its own products start catalysing the reactions that made them.

Cofactors first. Look at what a modern acetyl-CoA pathway actually requires — tetrahydrofolate, cobamide, thiamine, molybdopterin, coenzyme A, NAD, and in methanogens methanofuran, coenzyme M, coenzyme B and F420. Each of those is the endpoint of a long biosynthetic pathway of comparable enzymatic demand to the pathway it serves. And here is the trap: several of those pathways require the very cofactors they produce. Cofactor biosynthesis is autocatalytically closed. You cannot start it from nothing.

The only way out is that the cofactors were not needed at the start. Bare metal did the job — awaruite, greigite, magnetite — and cofactors were added afterwards, not to make the reactions possible but to make them portable. Each cofactor taken up was one more reaction that no longer required physical contact with a mineral surface.

Two steps are easy to blur here, and the difference matters. Simple cofactors could be picked up ready-made from the environment, produced by the same abiotic chemistry as everything else; that costs nothing. What is expensive — and what actually severs the dependence — is evolving the enzymes to synthesise those cofactors reliably from central metabolites, because only then is the supply internal. That is a campaign, not an event. It starts here, with the first ribozymes, and it is still unfinished at LUCA, which is precisely why Mrnjavac and colleagues describe LUCA's catalysis as a metal-enzyme hybrid rather than a completed metabolism. It closes in section 4.4 — twice, separately.

Martin calls this pattern patterned evolution of pathways: naturally occurring chemistry sketches the route, and catalysts are added later to trace over it more boldly. It runs outward from CO₂, which is precisely the opposite of the classical retrograde model, in which pathways grow backwards from a soup.

Then templates. The tipping point is a molecule that does both jobs at once. Folded, a short RNA can catalyse a bottleneck step and raise the local supply of nucleotides; unfolded, the same strand serves as a template as free nucleotides pair along it and link up. Catalyst and copy in one molecule. From here, variation in sequence produces variation in catalytic performance, and the whole system acquires something it never had before: heredity, and therefore selection.

Koonin and Martin argue that inorganic compartments are what make this survivable — pores hold concentrations high enough for the chemistry to work, while diffusion between pores gives a crude form of dispersal and competition between neighbouring networks.

Crossing that threshold triggered a ribozyme explosion of three-dimensional shapes acting as molecular copiers and scissors.

2.2 The templating trick and the proteome explosion

Ribozymes are fragile and needed structural support. Some acquired a mechanical solution: template RNAs — the precursors of messenger RNA — lined up specific amino acids held on transfer RNAs and linked them into short peptides that wrapped around and stabilised the ribozyme.

The consequences ran far past the original purpose. Some of those peptides turned out to be useful catalysts in their own right, and once amino acid sequence became specifiable, a vastly larger space of catalysts opened up than RNA folding alone could reach. This was the proteome explosion. The core of that ancient machinery survives today as the peptidyl transferase centre of the ribosome — which is still, tellingly, made of RNA.

2.3 The DNA trick and the information explosion

Reference: Forterre (2006), PNAS.

RNA degrades easily. Remove one oxygen atom from the sugar and the molecule becomes far less reactive — that is the whole chemical difference between ribose and deoxyribose, and it is worth orders of magnitude in stability.

Forterre's provocative suggestion is that DNA was invented not by cells but by viruses, as a way of disguising their genomes from host nucleases, and was subsequently captured by cells. Either way, double-stranded DNA permitted an information explosion: with a complementary strand as a built-in backup copy and a chemically inert backbone, genetic information could be stored and replicated with far higher fidelity, and genomes could grow. Large unified genomes came much later.

---

3. Cellular boundaries and LUCA

3.1 Organo-mineral boundaries

Reference: Lane & Martin (2012), Cell.

So far the networks had lived in rock: mineral catalysts, mineral walls, a geological gradient. The next step was building their own boundaries. Proteins began synthesising amphiphilic lipids that coated the inner surfaces of the mineral pores.

Lane and Martin point out that this creates a problem as much as it solves one. Early lipid membranes would have been leaky, and protons slipping through them do very little work on the way. Their answer is the sodium-proton antiporter, a transmembrane pump that harnesses the inward rush of protons to push sodium ions out. Because a leaky membrane blocks sodium far better than it blocks protons, the protocell converts a weak, leaky proton gradient into a stable sodium gradient it can actually use.

The result is a hybrid object: a lipid-lined compartment, still sitting inside a mineral pore, still drawing its power from the rock.

3.2 What LUCA actually was

References: Weiss et al. (2016), Nature Microbiology; Mrnjavac et al. (2026), Science Advances.

A common misconception portrays LUCA as a standalone cell — the first bacterium, more or less. The evidence points somewhere stranger.

Weiss and colleagues reconstructed LUCA from 355 protein families that survive a strict phylogenetic filter, and the portrait that emerged is anaerobic, thermophilic, hydrogen-dependent, running the Wood–Ljungdahl pathway, and thick with iron-sulfur clusters, radical mechanisms and transition-metal cofactors. That is a biochemistry that only makes sense inside a vent.

Mrnjavac and colleagues sharpen this into a specific claim about catalysis. LUCA was a hybrid — part enzyme, part rock — still leaning on mineral catalysts for reactions its descendants would later hand over to proteins. It possessed a leaky membrane, a genetic code, ribosomes and the ATP synthase, but it could not have survived outside the pore that made it. Half-alive by modern standards: a membrane-lined community of biochemistry, anchored in stone.

---

4. The two origins of living cells

To leave the vents, LUCA's descendants needed four things they did not have: a membrane tight enough to hold a gradient, a way to divide reliably, doors to let nutrients through a sealed boundary, and a metabolism that no longer touched the rock. Each of those was solved twice.

4.1 Two membranes

References: Koga et al. (1998), Journal of Molecular Evolution; Sojo et al. (2014), PLOS Biology.

Koga and colleagues first drew attention to the lipid divide, and it is stark. The ancestors of Bacteria build membranes from ester-linked straight-chain fatty acids; the ancestors of Archaea build them from ether-linked branched isoprenoids — attached to a glycerol backbone of the opposite stereochemistry. These are not variations on a theme. They are two solutions to the same problem, and the enzymes that make them are unrelated.

Sojo and colleagues supply the bioenergetic reason it had to happen this way. A leaky membrane is fine while you are borrowing a geological gradient, but useless for generating your own: to hold a large proton gradient you need a tight membrane, and a tight membrane cuts you off from the vent. So LUCA's bioenergetics were sodium-based, and the switch to proton-based chemiosmosis — the version powering ATP synthase in every cell today — happened after LUCA, separately in each lineage.

4.2 Two ways to divide

References: Erickson (2007), BioEssays; Samson et al. (2008), Science.

As lipid linings expanded they bulged out of the mineral pores like balloons. Sometimes a functional cell budded off; often the vesicle popped, or carried away an incomplete set of molecules. Reliable reproduction requires mechanical control — a molecular drawstring that pinches the membrane closed once genetics and metabolism have been divided evenly.

Bacteria use FtsZ, a tubulin relative, which Erickson traces back through the prokaryotes alongside the actin-like MreB. Many archaea instead divide using an ESCRT-III system, which Samson and colleagues showed is homologous to the machinery eukaryotes use for membrane scission. Two domains, two unrelated drawstrings. The most economical reading is that LUCA had no dedicated division machine at all, because LUCA did not need one: budding out of a pore is not the same problem as dividing a free cell.

4.3 Two ways to open a door

Reference: Pohorille et al. (2005), Astrobiology.

Tight membranes solved the leakage problem and created a starvation problem. A sealed membrane also blocks the passive entry of small organic nutrients — and by now these cells were making rather than scavenging, so they needed controlled traffic in both directions.

Pohorille and colleagues argue the barrier was lower than it looks. Transmembrane segments are structurally simple and fold without highly specific sequences, so even short peptides can assemble into functional ion channels. Those simple architectures can then be tuned into selective, gated, energy-transducing transporters by local sequence changes rather than wholesale redesign — which is exactly what you would expect of something invented more than once.

4.4 Two completions of metabolism

References: Mrnjavac et al. (2026), Science Advances; Poudel et al. (2018), Frontiers in Microbiology.

The retreat from the rock that began in section 2.1 finishes here. Every cofactor whose synthesis had been internalised, every enzyme that took over a reaction a mineral surface used to catalyse, had chipped away at the dependence — but it was still not gone at LUCA. Completing it is what made a cell free-living, and this is where Mrnjavac and colleagues found the signature of a doubled history written into metabolism itself.

In at least five essential reactions — in alanine synthesis, the shikimate pathway, riboflavin synthesis — bacteria and archaea use structurally unrelated proteins to catalyse the same chemistry. Not divergent versions of a common ancestor: separate inventions. Both lineages finished building their metabolism after they had already diverged.

The same pattern shows up in bioenergetics. Flavin-based electron bifurcation — the seesaw trick in which one exergonic electron transfer pays for a second, endergonic one, driving ferredoxin to a very low potential — looks like deep ancestral machinery. It is not. Poudel and colleagues found that the bifurcating enzymes were not a property of LUCA at all: non-bifurcating versions of each oxidoreductase predate the bifurcating ones, and flavoproteins were recruited to them repeatedly and independently. LUCA did not need the trick, because vent effluent already reaches potentials low enough to reduce ferredoxin directly. Its descendants needed it the moment they left.

Equipped with tight membranes, their own ion pumps, division machinery, transporters and a complete enzymatic metabolism, the ancestors of Bacteria and Archaea finally detached from the hydrothermal vents and entered the open ocean.

One crucible. One genetic code. One common ancestor. Two separate escapes into independent life.

---

Written by providing an outline and paper references to Gemini 3.1 Pro and iterating, then iterating Claude Opus 4.6 for fact-checking, order-checking, citation and reorganization.