The most instructive error in modern biochemistry was not a wrong answer. It was a wrong category. By the late 1950s everyone accepted that oxidising food released energy and that ATP stored it, and everyone assumed the two were connected by a molecule — a high-energy intermediate, provisionally christened before it was found, the way a planet is named from a wobble. Laboratories across three continents chased it. Purification protocols were refined, candidate compounds announced and retracted, reputations staked. The chase failed for twenty years, and the failure was not a shortage of skill. There was nothing there.
Current Conditions
The Heretic and the Manor House
Peter Mitchell’s 1961 proposal was that the intermediate does not exist because the energy is not carried by a molecule at all. Respiration pumps protons across a membrane; the resulting difference in proton concentration and electrical charge — the proton-motive force — is itself the stored energy; ATP synthesis is driven by letting the protons back through a specific gate. The carrier is a pressure. The store is a disequilibrium. The membrane is not scenery, it is the machine.
It was received as heresy, and it is worth being precise about why. The proposal did not offer a better molecule; it dissolved the question everyone had organised their instruments around. Careers built on isolating a substance do not convert easily into careers measuring a gradient. Mitchell left his university post, restored a manor house in Cornwall, and set up a private research institute funded partly from his own pocket and a herd of dairy cattle — and there, with a small team, produced the measurements that turned the heresy into the textbook. The Nobel came in 1978, by which time the mechanism was no longer controversial and the twenty lost years had quietly become a footnote instead of the lesson.
They are the lesson. A field looked for the wrong kind of object because its instruments could only find that kind of object, and the mistake was invisible from inside precisely because everyone shared it. This blog has recorded the same failure mode in other clothes — most recently in The Night Watch, where an individual defect turned out to be a population feature, diagnosed one sealed bedroom at a time. Here it is at its cleanest: the answer was not a thing, and thing-shaped searching cannot fail informatively. It just keeps not finding.
The Turbine That Had to Be Seen to Be Believed
What sits at the end of the gradient is the most literal machine in biology. ATP synthase has two motors on one shaft. The membrane sector is a ring of subunits, each carrying a single acidic residue that can pick up a proton on the crowded side, carry it around, and release it on the empty side; the geometry of the interface makes the ring turn one way and not the other. The soluble sector is a hexamer of three catalytic heads around a bent, asymmetric axle. As the axle rotates, it deforms each head in turn through three states — open, loose, tight — and the ATP is not squeezed into existence so much as squeezed out of it: the chemistry happens readily on the enzyme, and the rotation’s real work is prying the finished product loose. Paul Boyer proposed this binding-change mechanism from kinetics alone. John Walker’s crystal structure showed the three heads caught in three different states, exactly as required. And in 1997 the rotation was watched directly, by gluing a fluorescent filament to the axle as a flagpole and filming it spin.
The performance numbers still read like a misprint. The complete enzyme from a thermophilic bacterium turns at roughly 350 revolutions per second at body temperature, about 650 at 45 °C. Each proton crossing at a 200-millivolt motive force delivers around 0.2 electron-volts. And the gearing is the detail that gives the whole thing away as evolved rather than designed: three ATP per revolution, but a proton ring whose subunit count varies by species — eight in cattle, ten to fifteen across fungi, bacteria and chloroplasts, up to seventeen at the extreme. The cost of an ATP is therefore n/3 protons, which lands between 2.7 and 5 and is, in most organisms, not a whole number. Generations of textbooks printed a tidy integer because a tidy integer is what an engineer would specify. Evolution specified a ring size, and the ratio is whatever the ring size makes it.
Where the Parts Came From: a Salt Pump and a Hydrogen Enzyme
Complex I is the entry point of the respiratory chain and the largest of the pumps: in mammals, some forty-five subunits, an L-shaped assembly that accepts electrons from food-derived carriers at one end and moves protons across the membrane at the other, coordinating events separated by roughly two hundred ångströms. It is the standing example of a machine too intricate to have been assembled stepwise. The genomes say otherwise, and unusually loudly, because the intermediate steps did not go extinct.
The three largest membrane subunits — the ones that do the pumping — are homologous to each other and to subunits of Mrp, a multi-subunit sodium/proton antiporter that bacteria and archaea use for pH homeostasis and salt tolerance. Mrp is not an energy-transducing machine; it is plumbing. A fourth Complex I subunit turns out to be homologous to yet another Mrp subunit, which is why the current reading is not that one gene duplicated its way into a pump, but that an entire multi-subunit antiporter complex was recruited wholesale onto an ancestral redox module. And the recruitment is still reversible in the laboratory: expressed in Bacillus deletion strains, the Complex I subunit NuoL restores the function lost with MrpA but not MrpD, while NuoN does the opposite — the modern respiratory parts drop into an ancient salt-handling role, with their specialisations intact.
The redox half has a comparably ordinary pedigree. Complex I shares a common ancestor with membrane-bound hydrogenases and with a sulfane-sulfur reductase, both of which look like an Mrp-type antiporter fused to a soluble redox module built around a nickel-iron hydrogenase — enzymes for making a living in an anoxic world, where protons and sulfur were the available electron sinks. And the intermediate sizes are populated: an eleven-subunit version of Complex I, lacking the module that accepts electrons from NADH, is widespread across bacteria and archaea, while the classical fourteen-subunit form appears only in certain bacterial phyla. The mammalian forty-five-subunit monster is that fourteen-subunit core wearing three decades’ worth of accessory subunits like barnacles.
Read plainly: nothing here was invented. A pump for getting salt out of a cell was bolted to an enzyme for handling hydrogen, and the combination happened to conserve energy. The reason we can say this with confidence is not the elegance of the story but the survival of the witnesses — Mrp, the hydrogenases, the eleven-subunit version are not reconstructions, they are organisms you can culture.
The Turbine’s Own Ancestry Is Older and Murkier
ATP synthase is a harder case, and honesty requires saying so. It is older: a dated phylogenetic analysis places the split between the F-type synthases of bacteria and the A/V-type of archaea beyond four billion years, plausibly before the last universal common ancestor. The machine predates the division of life into its domains. It was already there when “bacterium” and “archaeon” were not yet different things.
The long-standing account has the two sectors arriving from different directions: the catalytic head descended from an ATP-driven helicase — a hexameric ring motor that pulls nucleic acid strands — and the membrane sector from a passive ion channel, the two co-opted into a single complex. A more radical version notes that the F- and V-type enzymes are homologous in their catalytic and membrane parts but not in the central shaft that connects them, and proposes that both descend from membrane protein translocases, which in turn descend from RNA translocases. In that reading the rotor is a fossil of a threaded polymer: where a strand used to be pulled through, a shaft now turns. Supporting this family resemblance, the rotor domain of the rotary ATPases and that of the bacterial flagellar export apparatus appear to share an origin — a flagellar protein has been shown experimentally to function as a rotor inside a foreign ATPase. Two of biology’s famous rotary motors, distant cousins.
But the co-option story has drawn a serious 2025 rebuttal, and it deserves airtime rather than a footnote. Hexameric helicases in bacteria and archaea are built on different protein folds and translocate DNA in opposite directions, which makes a single pre-LUCA helicase progenitor phylogenetically awkward. The mechanistic objection is sharper: every partially assembled intermediate is a leak. An uncoupled membrane channel drains the gradient the cell is spending its living to maintain; an uncoupled catalytic head burns ATP. Both failure modes are lethal, which narrows the corridor from parts to machine considerably — and modern enzymes carry dedicated inhibitor subunits whose whole job is preventing exactly those failures, implying that some such protection had to arise early. Meanwhile the empirical work has moved from tree-drawing to resurrection: in late 2025, inferred ancestral catalytic subunits were reconstructed, spliced into a thermostable bacterial enzyme, and imaged — suggesting the ancestral mechanism ran a six-step cycle. The ancestors are being rebuilt and switched on. That is the state of the art, and it is a much better place to argue from than a sequence alignment.
The Asymmetry That Points at the Beginning
Now the finding that makes this more than machine appreciation. The ATP synthase is universally conserved — every domain of life, essentially the same turbine. The mechanisms that generate the gradient are not. Respiratory chains, photosynthesis, hydrogenases, light-driven proton pumps: different inventions, some plainly independent, all feeding the same universal machine.
That asymmetry is an argument from distribution, and it points one way. If every lineage had to invent its own pump but nobody had to invent the turbine, the natural reading is that the gradient came first — that the turbine was built to exploit a proton gradient that was already there, and that pumping it yourself was the later, lineage-specific trick. Which is precisely the claim of the alkaline hydrothermal vent hypothesis: in the Hadean, alkaline fluids saturated with hydrogen percolated through labyrinths of micropores with thin inorganic walls containing iron-nickel-sulfide minerals, meeting a mildly acidic, carbon-dioxide-rich ocean. The pH difference across those barriers was around three units — the same magnitude and the same polarity, acidic outside, as the proton-motive force in a living cell. Roughly two hundred millivolts, geologically supplied, for free, continuously, for as long as the vent ran.
The objection is equally concrete and has not been dispatched. There is no direct evidence of thin inorganic membranes holding sharp gradients in modern alkaline vents; the barriers used in the models are on the order of a micrometre thick, while a cell membrane is five nanometres, and a molecular motor a few hundred atoms across cannot span the former. Proposed non-protein machines for tapping such a gradient are unconvincing. Geologists have recently added that the hypothesis’s premises sit awkwardly with vent geochemistry. The advocates answer that the criticisms attack simplified versions of the model, and the bench work — electrochemical flow reactors built to simulate vent conditions — continues. Fair grade: the most coherent story anyone has, resting on a genuine and unexplained universality, with a live and unresolved objection at the scale where it matters most.
Twenty years of searching for a molecule that was never there, because a pressure is not the kind of thing a purification column can find. The instruments determined the ontology — and everyone shared the instruments, so nobody could see the assumption.
What Was Built on Top: Three Claims, Three Grades
An industry now sells interventions premised on this biology, and the claims arrive bundled. Unbundled, they grade very differently.
Mitochondrial decline with age: real, and the least contested item on the list. Mitochondrial function measurably deteriorates in aged tissue — that much is observation, not theory. What follows from it is where the trouble starts.
The free-radical theory of ageing: the once-central hypothesis, now in visible retreat. The claim that reactive oxygen species from respiration accumulate damage that drives ageing organised the field for half a century. It has aged badly. Rodent studies manipulating antioxidant defences repeatedly failed to move lifespan; antioxidant supplementation trials in humans have been largely disappointing; evidence accumulated that mitochondrial DNA mutations arise mostly from replication error rather than oxidative damage. A 2013 review in this literature described the theory as moribund in its original form, and a 2014 review, weighing the whole body of evidence, titled itself a midlife crisis for the theory. The current picture assigns reactive oxygen species a signalling role — at low levels they induce protective responses, which is why indiscriminately mopping them up may do harm rather than good. Grade: a foundational theory that the field has substantially walked back, while the shelf it inspired keeps selling.
NAD-boosting supplements: the effect is real, the endpoint is missing. These raise circulating levels reliably — oral precursors increase NAD in the range of 130 to 150 percent, which is not nothing and distinguishes them from most of the shelf. Preclinical results in mice and worms are genuinely striking. Human trials so far show minimal improvement in cognition, vascular function, or muscle performance, with safety well established across more than a dozen trials and efficacy for age-related disease still limited. The honest formulation: a biomarker moves, the outcome has not yet followed, and the trials that would settle it — large, long, standardised — have not been run. Grade: a plausible mechanism with a demonstrated intermediate and an unproven end, which is the exact profile that supplement marketing is best at obscuring.
Two Honesties
First, the machinery has a hole in it. Complex I’s coupling mechanism — how an electron transfer at one end drives proton pumping two hundred ångströms away at the other — is not fully explained. The conformational-wave picture is a picture; the residue-by-residue transmission chain is not closed. The most photographed machine in cell biology has a gearbox nobody has fully opened, and any account that narrates it smoothly, including this one, is smoothing.
Second, ancestry is inference, at two different strengths. The Complex I story is unusually strong because the intermediate stages are living organisms and the parts remain functionally interchangeable in the laboratory. The ATP synthase story is weaker: homologies across four billion years, a contested co-option scenario, an origin claim resting on a distributional asymmetry that is real but is not a mechanism. Both are told here in the same voice, and they should not be believed at the same strength. Grade the salt-pump ancestry as demonstrated; grade the turbine’s helicase past as the current best guess under live attack; grade the vent as an argument, not a finding.
What Actually Transfers
This blog opens its bioenergetics file here because the mechanism is unusually well documented and the epistemics unusually clean — a rare case where the science, its history, and its commercial afterlife can all be graded in one pass. It also sits directly under my sleep research: the maintenance chemistry those preprints track runs on exactly this machinery.
Stored energy is usually a disequilibrium, not a substance. The gradient, and ATP itself — whose usable energy comes from being held far from equilibrium rather than from any special bond — are the same idea at two scales. Whenever you are told something “contains” energy, ask what it is out of balance with. The answer is more often a ratio than a material, and the ratio is where the leverage lives. Readers of The Weight of Nothing will recognise the move: the substance-shaped answer loses again.
Complexity with living intermediates is a solved case, not a mystery. Complex I looks unbuildable until you notice its ancestors are culturable and its parts still swap into their old jobs. Before treating any intricate system as sui generis, look for the surviving simpler versions — in biology they are species, in institutions they are the older organisations nobody bothered to shut down.
A moved marker is not a moved outcome. NAD boosters raise NAD; that is measured. Whether raising it does anything you would notice is not. This is the standard architecture of a supplement claim — deliver the intermediate, imply the endpoint — and it is worth reading the same way in every field that sells a proxy.
The wrong category is invisible from inside. Twenty years of excellent chemistry could not find a compound that did not exist, and no individual experiment revealed the error, because the error was in what everyone agreed to look for. Where a field’s failures are all the same shape, suspect the shape.
Instrument Check — Worth Your Attention
Study — Steiner & Sazanov, structure of the Mrp antiporter, eLife, 2020. The ancestor, photographed: a seven-subunit cation/proton exchanger whose architecture is visibly the membrane arm of Complex I with the redox module missing. The clearest single image of a modern machine’s previous job still being done by something alive.
Read — the chemiosmosis origin debate: Lane’s vent case against Jackson’s rebuttal. Read them adjacently rather than sequentially. Lane’s case rests on the universality of the turbine against the diversity of the pumps; Jackson’s on the fact that vent walls are a thousand times thicker than a membrane and no sharp natural gradient has actually been measured. Neither side is careless, which is what makes the exchange worth the evening.
Follow — ancestral sequence reconstruction of rotary ATPases, the resurrection literature, 2025 onward. The method that turns evolutionary storytelling into benchwork: infer an ancestral subunit from the phylogeny, synthesise it, build it into a working enzyme, and watch what it does. Recent work suggests the ancestral catalytic cycle ran in six steps. Expect this technique to settle arguments in the next decade that alignments could not settle in the last four.
Flight Log — Dispatch From Altitude
Aviation independently discovered Mitchell’s insight and called it bleed air. An engine does not produce cabin pressurisation, and it does not produce wing anti-ice, and it does not produce air conditioning. It produces hot compressed air tapped from the compressor stages — one intermediate, in one form, distributed to a set of consumers with nothing else in common. The energy that keeps a cabin breathable at cruise and the energy that keeps a leading edge free of ice are the same energy, converted at the point of use rather than delivered as a purpose-built substance. Anyone raised on that architecture finds the proton gradient immediately intuitive: of course the carrier is a pressure. What else would you route through a whole system?
Which makes the industry’s own architectural fight the exact analogue of the question this piece could not settle. The 787 abolished bleed air and went electric — generators instead of tapped compressed air, electric compressors for the cabin, resistive heating in the wing. Two philosophies, both flying: one universal pneumatic intermediate, or a common electrical bus with conversion at each end. Cells argued this out too, and settled on the pneumatic answer, with a second currency layered on top. Nobody would call either aircraft architecture irreducible; both are assemblies of parts that existed for other reasons, arranged by requirements and constrained by what the previous generation left behind. That is also, on the current evidence, what a mitochondrion is.
The part that stays with me is the load hierarchy in reverse. An airliner’s electrical system is built around the certainty that supply will one day falter, and every load is ranked in advance for the moment it does. A cell has no such ranking and no such moment — the gradient it runs on has, in an unbroken line through every ancestor, never once been allowed to collapse. Not for a single generation in something like four billion years. Every cell you are made of is the surviving end of a chain of maintained voltage that has never gone dark. We build machines that fail gracefully because we expect them to fail. Life built a machine that has simply never been switched off, and left the evidence in the parts list: a salt pump, a hydrogen enzyme, a threaded shaft — nothing new, nothing spare, and a manor house in Cornwall where somebody finally worked out what all of it was for.