I You Carry Two Sets of DNA
Almost every cell in your body carries two separate sets of DNA. One sits in the nucleus: half comes from your mother, half from your father. The other lives somewhere else, and comes only from your mother.
That somewhere is the mitochondrion, a small structure inside the cell that uses food and oxygen to produce energy. It is often called the cell’s power plant. A cell usually has hundreds of them, sometimes more than a thousand, with especially large numbers in big energy users such as heart muscle and the liver.
And each mitochondrion holds several copies of its own DNA. Within a single cell, then, this second set can outnumber copies of the nuclear set by a thousand to one or more.
Mitochondria also wind around the base of a sperm’s tail: they are the engine that powers its swimming. But after fertilization, the father’s mitochondria are largely cleared away. This set of DNA therefore comes down a single maternal line.
That strictly maternal route makes it a useful tool for tracing ancestry. Follow the lines back, and the maternal lineages of everyone alive today converge on a woman who lived in Africa between a little over a hundred thousand and roughly two hundred thousand years ago. She is known as Mitochondrial Eve.
She was not the only woman alive at the time. She is simply the woman through whom all our surviving maternal lines happen to pass. This DNA can establish family relationships, too. In 1993, Britain’s Prince Philip supplied a blood sample that helped identify the remains of the last tsarina and her daughters. The tsarina’s older sister was his maternal grandmother.
Mitochondrial DNA changes quickly, so it is also often used to distinguish closely related species. The genetic “barcode” used to identify an animal is commonly a stretch of one mitochondrial gene. It can tell you whether the fish sold at a market really is the species on the label. A fault in any of its genes can cause disease, and those faults, too, are inherited only from the mother.
In 2015, Britain passed legislation allowing a procedure that replaces a mother’s defective mitochondria with healthy mitochondria donated by another woman. A child born this way carries DNA from three people. In 2025, doctors in Newcastle, England, reported that eight children had been born using the procedure.
But why should one component of a cell have DNA of its own? Your liver has no separate set. Neither do your bones. The structures inside a cell that handle transport, breakdown, and synthesis have none. Why should this particular component have its own? And DNA is not all it has.
It has its own machinery for reading that DNA and making proteins, and the machinery looks more like a bacterium’s than like the machinery elsewhere in the cell. Some antibiotics designed to kill bacteria can interfere with it while leaving the equivalent process in the cytoplasm alone. It is wrapped in two membranes rather than one. It multiplies by splitting in two, rather than being built anew by the cell.
That last point is especially telling. Most cellular structures are reassembled when a cell divides. Mitochondria are different: they grow and divide themselves, just as bacteria do. Put all this together, and one explanation is remarkably simple. A mitochondrion used to be a bacterium.
Long ago, one cell swallowed another without digesting it. The swallowed cell stayed inside and kept making copies of itself. Gradually, it lost the ability to live independently, until it became a component of its host.
This explanation is called endosymbiosis: one organism moves inside another, and the two never part again. It is in secondary-school textbooks now. Getting from the first proposal to general acceptance took well over half a century.
In 1905, the Russian botanist Mereschkowsky published a paper in a German-language journal arguing that chloroplasts, the structures that carry out photosynthesis in plant cells, descended from once-independent organisms resembling blue-green algae. In the 1920s, the American anatomist Wallin went further. He claimed to have grown mitochondria isolated from animal tissue in culture.
Had that held up, it would have been the most direct support imaginable: if a mitochondrion could still be cultured independently, it was still a bacterium. But contamination by outside bacteria could not be ruled out, and other researchers could not reproduce the experiments. Instead of helping, the cultures gave the idea a bad name.
In 1967, the American biologist Margulis published “On the Origin of Mitosing Cells” under the name Lynn Sagan. The surname came from her first marriage; her former husband was the astronomer Carl Sagan. She brought several cellular structures into a single endosymbiotic framework: mitochondria came from respiring bacteria, chloroplasts from photosynthetic bacteria, and eukaryotic flagella from spirochetes.
The first two claims later received strong molecular support. The third did not. It is widely said that some fifteen journals rejected the paper; that number comes from her own recollection many years later.
The objections raised between the 1920s and the 1970s fall into five main groups, each with a reasonable basis. First, until the early 1960s, no one had directly seen whether mitochondria really contained genetic material of their own.
Second, resemblance alone could not establish the case. Two membranes, bacterial dimensions, multiplication by splitting in two: all were compatible with endosymbiosis, but none could identify the donor. As Part II explained, similarities can arise from shared ancestry, or from separate lineages evolving similar forms.
Third, control was heavily concentrated in the nucleus, which encoded a great many mitochondrial proteins. Before anyone understood how proteins were delivered into mitochondria, this could be read as evidence that they were “not like independent bacteria.”
Fourth came the disrepute left by Wallin’s culture experiments. Fifth, the part of Margulis’s theory that was not eventually accepted cast doubt on the whole account.
What changed people’s minds was a succession of specific discoveries. The first two predated Margulis’s paper; indeed, she used them as evidence.
In 1962, researchers saw DNA-like regions inside the chloroplasts of a green alga. The following year, a husband-and-wife research team observed fine strands with the characteristics of DNA inside mitochondria from chick embryos.
That dismantled the first objection outright. But removing one objection does not establish a case. The third, in particular, would not turn from evidence against the theory into evidence for it until the import of proteins into mitochondria was understood.
Next came the protein-making machinery. The machinery inside mitochondria and chloroplasts was confirmed to be bacterial in type, and certain antibiotics that target bacteria could disrupt it too. Some antibiotics owe their side effects to collateral damage to this bacterial-style machinery inside our mitochondria.
Yet all this evidence ran up against the same ceiling. It could say “bacterial in character,” but not “most closely related to this group of bacteria.” Sequences broke through that ceiling.
In 1975 and 1976, researchers compared ribosomal RNA from red-algal chloroplasts with that of prokaryotes. This was the molecule used to build trees in Part II, and it placed the chloroplasts on the cyanobacterial side. In 1978, the American researchers Schwartz and Dayhoff compared protein and nucleic-acid sequences together to map the evolutionary relationships among prokaryotes, eukaryotes, mitochondria, and chloroplasts.
Here, then, was a use for the tree from Part II. Once you can build a tree, you can ask which branch something belongs on, a question that resemblance in form cannot answer. Nor did agreement on the two organelles arrive at the same time.
The relationship between chloroplasts and cyanobacteria was established earlier, and was more straightforward. Establishing the relationship between mitochondria and their bacterial relatives required more sequences and more elaborate corrections. That is the subject of the next section.
II Which Branch Supplied the Donor, and When?
There is strong evidence that mitochondria came from bacteria. Exactly which branch of bacteria remained unsettled for a long time, for very specific reasons.
Mitochondria have been living inside cells for roughly two billion years. Throughout that time, their genomes have changed rapidly, far faster than those of free-living bacteria, because many functions once held in check by selection were no longer needed. On a tree of relationships, that leaves them at the end of an exceptionally long branch. And long branches are precisely the ones most likely to be put in the wrong place.
Start with the broad group. Alphaproteobacteria are a major group of bacteria that includes many inhabitants of other cells, many free-living marine bacteria, and the nitrogen-fixing rhizobia in the root nodules of legumes. Sequence evidence strongly supports a deep relationship between mitochondria and this group. The difficulty has been pinning down their exact position within it.
The order Rickettsiales, which includes the pathogen responsible for typhus, was once the leading candidate. Its members all live inside other cells, just as mitochondria do. In 1998, Swedish researchers sequenced the typhus pathogen’s genome and found many similarities to mitochondria. For a time, it was regarded as their closest relative.
But living inside another cell can itself make sequences change rapidly and skew their composition. These are exactly the conditions that produce long-branch attraction, the illusion discussed in Part II: two especially fast-evolving branches can be pulled together on a tree even without a close relationship.
A 2018 study broadened the sampling and specifically addressed these biases. Its best model placed mitochondria outside all the alphaproteobacteria then known. In 2022, more refined models produced a result in which “mitochondria are the sister group to most known alphaproteobacteria.” What we can say today, then, is that mitochondria came from a branch sharing an origin with alphaproteobacteria. Their closest relatives may have died out, or may simply not have been found.
The timing is trickier. It runs into a circle, something like the chicken-and-egg problem. For a modern cell to engulf a bacterium, it needs an entire apparatus: the cytoskeleton deforms the membrane, which wraps around the target, forms a vesicle, and then undergoes fusion. The apparatus is complicated, and it costs energy to run.
Assume that the full engulfment apparatus was already there, and the bacterium’s entry is easy to describe. But then you have to explain how a host without an internal “power plant” first built this energy-hungry apparatus. Put the arrival of mitochondria earlier, and you face the opposite question: how did a host that could not yet swallow things form a sufficiently close relationship with a bacterium to bring it inside? Modern amoebas and white blood cells, after all, feed and capture bacteria by engulfing them.
The most influential proposal was the hydrogen hypothesis, put forward by Martin and Müller in 1998. The host was an archaeal-type cell that lived on hydrogen. Its bacterial partner could respire or produce hydrogen, depending on the conditions.
The two were initially tied together by an exchange of substances, with no pre-existing engulfment system required. Imagine two shops opening next door to each other, one shop’s waste happening to be the other’s raw material. After long enough as neighbors, they merge into one business. The Asgard archaea discussed in Part II are the closest known relatives of this host.
One strength of the hypothesis is that it explains why many anaerobic eukaryotes still possess structures derived from mitochondria. Later genome studies showed that many organisms once classified as “primitively without mitochondria” actually carried reduced versions of them. What the hypothesis does not explain is equally specific: it does not directly lay out the steps by which the nucleus, internal membrane system, and complex cytoskeleton arose.
In 2014, two paternal cousins named Baum, one a researcher of cell shape and the other of plant evolution, proposed a different geometry. In their inside-out model, the host’s membrane grew numerous outward projections that enclosed the ancestors of mitochondria clinging to its outer surface.
The projections expanded and fused to form the cytoplasm and internal membranes. The modern nucleus corresponds to the original main body of the ancestral cell. This model can account both for the continuity between the nuclear envelope and the internal membranes, and for the absence of any need to engulf things at the outset. Its difficulty is that there is no direct record of the crucial intermediate forms.
Other models propose that, when mitochondria arrived, the host was already quite large, had dynamic membranes and a cytoskeleton, and might even have been able to engulf things. The old idea that some eukaryotes never had mitochondria has been disproved. That does not eliminate every model in which mitochondria arrived late.
What we can be confident of today is that the last eukaryotic common ancestor, from which yeast, kelp, and human beings all descend, already had mitochondria and a complex set of eukaryotic cellular machinery. The order in which these systems appeared has remained unresolved. A study published in late 2025 used the relative timing of gene duplications to argue that the nucleus, cytoskeleton, membrane trafficking, and even the engulfment apparatus all preceded mitochondria, which were the last to arrive. It is not the final word, but it adds substantial weight to the late-arrival side.
When did this common ancestor live? Molecular-clock estimates range from about one billion to more than 2.3 billion years ago, when there was not a plant on land or an animal in the sea. The differences come from the choice of fossil calibration points, the models, the sampling, and how the common ancestor itself is defined.
III Incorporation in Progress
Endosymbiosis did not happen only two billion years ago. In 2024, American and Japanese researchers announced in Science that a cyanobacterial lineage living inside a single-celled marine alga had become one of its organelles, specializing in nitrogen fixation.
They named it the nitroplast. Its host is a coccolithophore clothed in small, pentagonal plates of calcium carbonate. The researchers estimate that this partnership began about a hundred million years ago.
Nitrogen fixation converts the nitrogen gas in the air into nitrogen that organisms can use. Plants and animals cannot do it themselves. Legumes rely on bacteria in their root nodules; crops rely on chemical fertilizers.
Until then, no known eukaryotic organelle could fix nitrogen. The researchers suggested that this system might offer clues toward an enticing possibility: one day installing such organelles in crops.
The cyanobacterium had first been detected in Pacific seawater back in 1998, but no one could keep its host alive in culture. One Japanese researcher made more than three hundred sampling trips to sea and spent well over a decade on the effort before managing to grow the alga in the laboratory. Only with living material could researchers see how the two partners lived together.
There were three lines of evidence. Proteomic analysis showed that more than three hundred proteins encoded in the host nucleus were being imported into the symbiont. Its division was synchronized with the host’s cell cycle: when the host split in two, so did the symbiont, with an equal share going to each daughter cell. Its size also remained proportional to the host’s, as in mitochondria and chloroplasts. Its genome had shrunk dramatically to around 1.4 million base pairs. It lacks a complete photosynthetic apparatus and several central parts of its metabolism.
So does it count as an organelle? The main dispute concerns not these observations but where to draw the boundary around the word “organelle.” A small genome does not settle it: many bacteria that can live only inside other cells have tiny genomes and are still called symbiotic bacteria.
Synchronized division and reliable inheritance are also strong indicators, but some symbiotic bacteria manage those too. Some diatoms harbor nitrogen-fixing cyanobacteria whose degree of integration varies. Calling all of them organelles would not be appropriate.
The strongest evidence is the import of host proteins. It shows that the resident can no longer do without deliveries from its host. Biology has no rule saying, “Meet this many conditions and you automatically qualify as an organelle.”
A comparison with something that does not qualify makes the point clearer. When one kind of sea slug eats algae, it saves their chloroplasts and stashes them inside its own cells, living for months on what they make in the sunlight.
Yet no import of host proteins into these chloroplasts has been found, and they do not divide along with the sea slug’s cells. Stolen chloroplasts cannot make copies of themselves. Once they wear out, it is time to eat again.
Another living example has been known for longer. Paulinella is a group of shelled amoebas, mostly found in freshwater, with a photosynthetic structure called a chromatophore. Each cell has just two, shaped like a pair of sausages.
The chromatophore came from a different branch of cyanobacteria in an independent incorporation, entirely separate from the origin of chloroplasts. Early estimates placed it at least sixty million years ago. A molecular clock in 2016 put the upper age limit in the range of ninety to a hundred and forty million years.
That makes it more than an order of magnitude younger than the chloroplast event, so many intermediate states are still visible. Its chromatophore genome contains about 1.02 million base pairs and encodes roughly 867 proteins. Closely related free-living cyanobacteria have about 2,917.
It has kept only about a quarter to thirty percent of the genes, while chloroplasts have shrunk to less than five percent. They come from separate incorporations, but let us glimpse different stages of comparable integration processes. The younger lineage is not guaranteed the same destination.
Dozens of genes clearly derived from this cyanobacterial lineage have already been found in the host’s nucleus. In 2012, experiments directly showed that proteins encoded in the host nucleus entered the chromatophore.
Proteomic work in 2017 then showed that the imported proteins went far beyond the products of those few dozen transferred genes. This is already a system that can function only if its host supplies proteins.
Incorporation can also proceed one layer inside another. Some algae engulfed a eukaryotic cell that already had a chloroplast. The result is a nesting doll: the host on the outside, the remains of the swallowed cell in the middle, and the descendant of the original cyanobacterium innermost. The chloroplasts of some groups consequently have three or four surrounding membranes. Two groups of algae even retain a small remnant of the swallowed cell’s nucleus.
Marine diatoms, responsible for about a fifth of all photosynthesis on Earth, went through this step in their ancestry, as did many of the dinoflagellates that cause red tides. So did the malaria parasite, Plasmodium. It still carries a reduced chloroplast that lost photosynthesis long ago, and some antimalarial drugs target this very structure. It was not recognized as a reduced chloroplast until the 1990s.
How many times has this secondary incorporation happened? There is no agreed whole-number answer. Complex chloroplasts derived from red algae and those derived from green algae cannot have come from the same event, so the answer is at least two. Allowing for several independently originating groups brings even a conservative count to at least three.
IV Where Did the Genome Go?
In 1981, a team in Cambridge, England, read the entire sequence of human mitochondrial DNA. It was the first genome in the human body to be fully sequenced. The human mitochondrial genome has 16,569 base pairs and thirty-seven genes. Thirteen encode proteins; the rest encode the RNA needed for translation. For comparison, one free-living alphaproteobacterium has a genome of about four million base pairs and more than 3,700 genes.
The human mitochondrial genome is only about one two-hundred-and-fortieth as long. Genome length, however, is not the same thing as the number of ancestral genes retained.
Some campions, wildflowers in the carnation family, have mitochondrial genomes of around 11.3 million base pairs, divided among more than a hundred small circular chromosomes. The Siberian larch’s is about 11.7 million base pairs long.
Those figures exceed the total length of many bacterial genomes. Their gene counts have not grown in proportion. The expansion consists of noncoding sequences, repeated sequences, and sequences acquired from elsewhere.
So when we ask where the genome went, what matters is the genes that still do a job, not the length.
A typical chloroplast genome has about 120,000 to 160,000 base pairs and usually around a hundred to a hundred and twenty identifiable genes, far more than a mitochondrion. Many core components of photosynthesis remain on site. A free-living Synechococcus, by comparison, has about 2.7 million base pairs.
The reduction on the chloroplast side therefore has the same shape as the reduction on the mitochondrial side. Both have shrunk to a tiny fraction of their free-living relatives’ genomes, though the degree varies from lineage to lineage.
Where did the missing genes go? There are three possibilities. They moved to the nucleus and kept working. They were lost outright, because the host already had a system doing the same job or because the job was no longer needed. Or they were replaced by genes from elsewhere.
Paulinella makes the distinction directly visible. Its chromatophore retains only about a quarter to thirty percent of the genes present in closely related free-living cyanobacteria. Yet the nuclear genes identified as clearly coming from that source number only in the dozens. Most of the remaining gap represents gene loss, not wholesale transfer to the nucleus.
“Moving house,” then, tells only a small part of the story. Most genes did not move; they disappeared, because the host already had an equivalent system, or because that function was no longer needed inside a cell. A minority actually moved. A smaller minority stayed.
The movement can be measured directly. In 1990, researchers placed a gene in yeast mitochondria that could function only after reaching the nucleus. They measured its escape into the nucleus at roughly twice per hundred thousand cell divisions. Movement in the opposite direction, from nuclear DNA into mitochondria, was at least about a hundred thousand times rarer.
That hundred-thousand-fold asymmetry matters. This is not random diffusion in both directions. It is a one-way street: material moves toward the nucleus and almost never comes back.
A tobacco experiment in 2003 used an equally ingenious design. Researchers put a gene in the chloroplast genome that could be expressed only after entering the nucleus. Among roughly a quarter of a million offspring, they obtained sixteen independent integration events. That works out to about one pollen grain in sixteen thousand carrying transferred chloroplast DNA.
This is still happening in humans too. Everyone’s nuclear DNA contains hundreds of old pieces of mitochondrial DNA, known as nuclear mitochondrial DNA segments.
One population-genomics study found 141 previously unrecorded insertion sites. Different populations, and even different individuals, carried different recent insertions. Larger-scale analyses have also detected newly occurring insertions, at a rate of roughly one in every few thousand births.
In other words, about one newborn in a few thousand carries a piece of DNA that has just moved from a mitochondrion into the nucleus. The process has not stopped. It is happening right now. These fragments sometimes cause trouble for forensic scientists and researchers working on ancient DNA: an old fragment in the nucleus can be mistaken for mitochondrial DNA.
V The Part That Will Not Transfer
If transfer is still happening, and has been happening for roughly two billion years, the question becomes: why are thirteen left? Before answering, though, we need to establish something else. How are those thirteen proteins made today?
The vast majority of mitochondrial proteins are made in the cytoplasm and then brought into the mitochondrion. Its membranes have specialized channels. A protein carries an “address label” that is recognized; the protein is pulled through the membrane and folds into shape inside. In principle, then, any mitochondrial gene could move to the nucleus, as long as its product could still be sent back.
Why, then, can thirteen of them not be sent back? There are several explanations, none of which explains everything.
The first is local control. Proposed by the British researcher Allen in 1993, this idea holds that if core components of the respiratory chain are expressed under the control of genes inside the same mitochondrion, they can respond directly to local conditions. The nucleus, far away, may not know how busy that particular mitochondrion is, or whether something has gone wrong there.
One clear point in its favor is that the most consistently retained genes are largely those encoding the respiratory chain’s core membrane proteins.
What it does not explain is why mitochondrial genomes also retain genes for the RNA and ribosomal proteins needed for translation, which do not participate directly in electron transport. On the other hand, mitochondrial variants that have lost the respiratory chain have mostly lost their genomes too, just as the idea predicts.
The second explanation: too oily to get through. We need a picture here. A protein is a chain of amino acids, some of which like water and some of which avoid it.
The core components of the respiratory chain are embedded in a membrane. Their surfaces therefore need to avoid water; otherwise they could not stay in the membrane. So what happens to an extremely water-averse protein in the watery environment of the cytoplasm?
It clumps. Just as oil gathers into a droplet in water, the molecules stick to one another and will not come apart. Or the protein inserts itself into the wrong membrane, or gets stuck in an import channel and cannot pass through.
A 2002 study provided direct support. In legumes such as soybean, a mitochondrial gene had naturally relocated to the nucleus, and its protein had become markedly less water-averse. Experimentally reducing this water aversion could also improve import. In 2016, researchers tried moving two such genes into the nucleus in human cells, then directing the proteins they produced back to the mitochondria. Remarkably, they had some success.
What this explanation cannot account for is why genes for water-friendly proteins, as well as RNA genes, remain in mitochondria. Nor does it explain why proteins with similar degrees of water aversion have moved in one lineage but stayed in another.
The third explanation: different codes. DNA encodes amino acids in groups of three letters, using a codebook shared by almost all organisms.
Several mitochondrial lineages, however, have reassigned the meanings of particular triplets. In human mitochondria, a triplet that means “stop” elsewhere is read as an amino acid. So if a mitochondrial gene moves unchanged into a nucleus that uses the standard code, it may be translated into different amino acids, or its translation may stop too soon.
This is no universal answer either. In many lineages, the differences from the nuclear code are limited. After transfer, mutations can also alter the relevant triplets. And the reassignment of the code may itself have occurred after large numbers of genes had already moved.
There are other constraints too: import signals, the base composition of the genes themselves, and the need for local assembly. There is no consensus on a single explanation for everything. One direct way to weigh these explanations against one another is to ask whether the genes retained in different lineages converge on the same small set.
All thirteen protein-coding genes in human mitochondria encode core membrane components of the respiratory chain and the rotary device: seven belong to the first proton pump in the respiratory chain, three to the last, two to the rotary device, and one sits in between.
All are embedded in the inner mitochondrial membrane and participate directly either in pumping protons outward across it or in letting protons flow back to turn the rotor. They are the innermost working parts of the dam system described in Part III. In other words, the genes that are hardest to move encode precisely the proteins most deeply embedded in the membrane.
This tiny set strongly fits the tendency for core membrane components to stay behind. Other organisms, however, do not all retain these same thirteen.
The mitochondrial genome of a protist sequenced in 1997 has about 69,000 base pairs and retains around ninety-seven genes. It is one of the known mitochondria closest to the gene-rich ancestral state. Another retains about a hundred. Some lineages keep several times as many genes as animals do.
The strongest partial convergence across mitochondria occurs among a few classes: core subunits of cytochrome oxidase and the cytochrome reductase complex, and core membrane subunits of the energy-synthesizing apparatus. Exactly which subunits remain encoded in the mitochondrion varies by lineage.
What stays behind is neither a random handful nor a list identical in every organism. The same few kinds of components remain, but each lineage keeps a different selection. This suggests that the constraints work in combination rather than singly.
Different lineages retain overlapping but nonidentical sets of genes. Both shared constraints and lineage-specific differences therefore matter. “Water aversion and membrane assembly” and “local control” may both contribute to the explanation. The lists alone cannot establish that either mechanism is sufficient; calling the move unfinished cannot substitute for investigating how it works.
At the other end lies complete loss. In 2016, researchers reported an anaerobic, single-celled eukaryote that had no mitochondria at all, and no structures derived from them.
It lives in the gut of a chinchilla, the rodent often kept as a pet. This was not merely a failure to find something under the microscope. Its genome and transcriptome also lacked the entire set of signature proteins needed for mitochondrial protein import and iron–sulfur cluster formation.
Nor is it a surviving branch from before mitochondria were acquired. Evolutionary evidence shows that its ancestors once had mitochondria and later lost them completely. One key replacement was the acquisition of a system in the cytoplasm that allowed iron–sulfur clusters, previously dependent on mitochondria for their synthesis, to be made there instead.
All known eukaryotic lineages, then, trace back to ancestors that had mitochondria, while a few lineages can subsequently lose them altogether. The genome can disappear before the structure does.
Giardia, a common culprit when dirty drinking water gives you diarrhea, has a reduced mitochondrion without a genome, mainly retaining the function of iron–sulfur protein assembly. The organelle in Trichomonas also usually has no genome of its own, but it retains a mitochondrial-derived protein import system and some energy metabolism.
Full mitochondria, reduced forms that retain energy metabolism, reduced forms mainly devoted to iron–sulfur assembly, and complete absence: these are a series of different endpoints. A simple division into “has mitochondria” and “has no mitochondria” cannot describe them.
To decide whether a structure is still there, we need to look at its remaining metabolism, protein import, and genetic integration, not just its ancestor’s most conspicuous function. Losing photosynthesis does not make a chloroplast vanish at once; the malaria parasite is an example. Losing a respiratory chain does not make a mitochondrion-derived structure vanish at once either.
VI What Else Is New?
The “eukaryotic” in eukaryotic cell means that it has a true nucleus: two membranes enclose the genetic material, separating it from the rest of the cell. The DNA in a single human nucleus would stretch to about two meters, yet it must fit inside a nucleus only a few micrometers across. Bacteria and archaea lack this wrapping and are collectively called prokaryotes.
A eukaryotic cell has more than an extra mitochondrion and a nucleus. It has an entire system of internal membranes that can pack cargo into vesicles and send it from place to place, and a cytoskeleton that can change shape. Many can also swallow other things whole. Where all this came from is the hardest part of eukaryotic origins.
There are three broad accounts of where the nucleus came from. One holds that the host’s own membrane folded inward around its genetic material. Today, the outer nuclear membrane is continuous with another membrane system, the endoplasmic reticulum: a large expanse of folded membrane specialized for making proteins and lipids. Nuclear pores and the protein coats around cargo vesicles also share the same kind of fold. Both observations are compatible with this account. But continuity today cannot reconstruct the geometry of the ancient process. Inward folding and the inside-out growth described in Section II can both produce continuity between the nuclear envelope and the endoplasmic reticulum, while telling opposite histories.
A second account traces the nucleus to a large DNA virus. Such viruses enter a cell and enclose a membrane-bound replication zone of their own, separating replication from translation. It looks a little like a nucleus.
But these viruses and eukaryotic cells have exchanged genes for a long time. Shared proteins do not tell us who gave what to whom, and no donor tree as stable as the mitochondrial one has emerged.
Viruses certainly can build this sort of thing, though. In 2017, researchers found that a giant virus infecting bacteria constructs a nucleus-like compartment out of protein once inside its host, enclosing its DNA for replication. It even uses a microtubule-like protein filament to push the compartment into the middle of the bacterium.
A third account proposes that the nucleus itself arose through endosymbiosis. Yet no small genome recognizable as that of an independent bacterium or archaeon can be found inside the modern nucleus.
None of these three accounts has won out. What we can establish is something else: the last eukaryotic common ancestor already had complex nuclear pores. A nuclear pore is assembled from hundreds to a thousand or more proteins, making it one of the largest protein complexes in a cell, and can let more than a thousand molecules pass each second. In 2010, researchers found homologues of its major components across the genomes of diverse eukaryotes.
The cytoskeleton gives a more clear-cut result. Textbooks used to say that prokaryotes had no cytoskeleton. That statement has been overturned. In 2001, researchers solved the structure of a bacterial protein and found it almost identical to actin, the protein our muscles rely on to contract. It could form filaments too.
Another bacterial protein, involved in cell division, had its structure solved in 1998. It shared the same fold as tubulin.
These proteins and their eukaryotic counterparts belong to the same ancient families, but have evolved separately for a very long time; neither is the direct precursor of the other. The parts for building a cytoskeleton therefore existed before eukaryotic cells. The Asgard archaea from Part II also carry genes for cytoskeletal and membrane-trafficking functions once thought unique to eukaryotes.
Organizing those parts into a network that could push membranes around, transport cargo, and pull chromosomes apart came later. The spindle that pulls chromosomes toward opposite sides during cell division is built from microtubules.
More of the eukaryotic novelty lies in expanding these families, finding new ways to assemble them, and combining molecular motors with regulation. In nerve cells, molecular motors shoulder cargo-filled vesicles and walk along microtubules, eight nanometers per step. To get from the spinal cord to a toe, the journey can be a meter long.
The same applies to the internal membrane system. Comparisons of genomes across eukaryotes show that their last common ancestor already possessed a highly developed membrane-trafficking apparatus, not merely a simple endoplasmic reticulum. Some parts have deeper homologues in archaea, but a great many families expanded within the eukaryotic branch itself.
What we can be confident of is that the last eukaryotic common ancestor already had mitochondria, a nucleus, nuclear pores, complex membrane trafficking, and a fully developed cytoskeleton together. The order in which they arose remains disputed. The 2025 study discussed in Section II places mitochondria last.
Linear chromosomes brought another problem. Most bacterial chromosomes are circular and have no ends. The ends of a linear chromosome get a little shorter each time it is copied.
Eukaryotic cells cushion that loss with stretches of repeated sequence at the ends, then use a specialized enzyme to replenish them. The enzyme is called telomerase. Discoveries concerning it earned the 2009 Nobel Prize, and one of the three laureates was Szostak, whom we met studying the origin of life in Part III.
In most of our somatic cells, telomeres gradually shorten over repeated divisions. Most cancer cells turn the enzyme back on. Chromosome ends are not the same in all eukaryotes, however, and the human arrangement cannot simply be projected onto the ancestor.
VII Evidence Overturned
For the fossils, we can borrow the division from Part I: Level A is more widely accepted; Level B is plausible but not yet secure.
By about 1.65 billion years ago, we have eukaryotic microfossils that are widely accepted and belong in Level A. They are tiny sacs tens of micrometers across, with elaborate surface ornament, projections, and branches. Under a microscope they look like little spiny sea urchins or balls trimmed with lace. Researchers call microfossils of this uncertain affinity acritarchs.
It is not their size that identifies them as eukaryotic. It is their wall structures and their capacity for varied forms, which known prokaryotic cells would struggle to produce. One set comes from rock formations in the Yanshan region of northern China.
Another Level A fossil is Bangiomorpha: filaments of cells that grow attached to a surface and closely resemble certain growth forms of modern red algae. It is among the oldest known multicellular organisms assignable to a living group. In 2018, researchers redated its rock formation from about 1.2 billion to about 1.047 billion years ago, making it some 150 million years younger. It comes from an island in the Canadian Arctic. We will meet it again in Part VII.
Level B contains a particularly beautiful piece of evidence. In 2017, Swedish researchers used synchrotron tomography to reveal the three-dimensional cellular structure of two kinds of fossils in roughly 1.6-billion-year-old rocks from central India. They interpreted them as red algae. Some cells even contained small, diamond-shaped structures, which the researchers took to be a component of chloroplasts involved in carbon fixation.
If that interpretation holds, the red-algal record jumps back more than five hundred million years. But we have morphology alone, without independent molecular information, so for now it stays in Level B.
Next comes the story of a line of evidence overturned in full. In 1999, Brocks, a young German then working on his PhD in Australia, and his colleagues reported in Science that they had found molecules called steranes in roughly 2.7-billion-year-old rocks from Western Australia.
Steranes are the stable carbon skeletons left by cholesterol-like molecules after long burial. These molecules, including the cholesterol on your blood-test results, are widespread in eukaryotes but rare in bacteria.
Finding steranes in rock is therefore usually taken as a trace of eukaryotic life. If the molecules were as old as the rock, eukaryotes were already around 2.7 billion years ago, almost a billion years before the earliest eukaryotic fossils.
The problem that emerged was not that the instruments had misread the molecules. It was whether the molecules really were the same age as the rock. In 2008, researchers measured carbon isotopes in individual particles of organic matter in the rocks and argued that the molecules had entered only after the rocks underwent thermal metamorphism about 2.2 billion years ago. Later fluids had very likely carried them in.
In 2012, several laboratories joined forces to drill fresh cores from the same Western Australian formations under unprecedentedly clean conditions, then analyzed them independently. The results, published in 2015, showed that sterane concentrations inside the fresh cores were much the same as in the procedural blanks. In other words, the amounts were comparable to the background reading when no sample was put through the procedure. On the outer surfaces of cores collected by the old methods, concentrations were more than an order of magnitude above the blanks.
The study also found that organic matter genuinely belonging to those ancient rocks had long since been cooked beyond recognition by geothermal heat, while the previously reported steranes were relatively “young.” Those steranes no longer served as evidence for eukaryotes 2.7 billion years ago. Brocks himself, an author of the 1999 paper, was among the authors who overturned it. Later, using the same class of molecules, he read evidence of a great expansion of algae in the oceans from rocks about 650 million years old.
Contamination can enter at three points. Drilling, sample surfaces, and laboratory handling can all introduce modern or younger organic matter. Over long geological periods, subsurface fluids can carry younger organic matter into older rocks. And even with no modern contamination at all, a molecule that arrived after deposition cannot establish what was living at the time of deposition.
New molecular evidence appeared in 2023. Researchers reported primitive steroid molecules dating back as far as about 1.64 billion years. These correspond to early intermediates in the synthesis of cholesterol-like molecules today, made by a shorter pathway than the modern one; simulation experiments established the correspondence. The first half of the paper’s title was “Lost world of complex life.” The organisms that made the molecules may have included bacteria as well as a long-extinct assemblage near the root of the eukaryotic lineage that flourished in the seas for hundreds of millions of years.
This is about a billion years younger than the 1999 claim, and concerns a different kind of molecule. It does not restore the old claim. When a line of evidence has been overturned, new evidence that later appears along the same line does not automatically inherit the earlier conclusion.
VIII One Origin, and the Thirteen That Still Have Not Left
What did mitochondria give eukaryotic cells? In 2010, Lane, a biochemist at University College London, and Martin, who had proposed the hydrogen hypothesis, offered a famous answer: energy.
Their argument ran like this. To produce more energy, a bacterium must expand the membrane it uses for respiration. That respiratory membrane is the cell membrane wrapped around the whole bacterium.
Expanding it means enlarging the entire cell. But as the cell grows, it still has the same single genome, now serving a larger volume. A eukaryotic cell, by contrast, can have hundreds or thousands of mitochondria, each with its own respiratory membrane and each retaining only a tiny genome. Most of the proteins are encoded by a much larger nucleus.
Their calculation put the energy available per gene in a eukaryotic cell at about two hundred thousand times that in a prokaryotic cell. This was a model’s result, not a measurement made gene by gene. Lane later developed the idea in a bestselling popular-science book.
The objections were quantitative too. Lynch, an American researcher of genome evolution, has long argued that genomes often grow larger and more complex through random drift in small populations, rather than being driven there by selection.
In 2015, Lynch and Marinov calculated the full cost of a gene: copying it, transcribing it, making its protein, and replacing that protein. They argued that genes themselves consumed too little of a cell’s energy budget to create a barrier of “no mitochondria, no affordable large genome.” A 2017 comparison across species, scaling measured metabolic power by cell volume, surface area, and biomass, likewise found no order-of-magnitude leap in energy output per unit volume attributable to mitochondria.
There is also a straightforward mismatch. A typical bacterium has three or four thousand protein-coding genes. Most living eukaryotes have between six thousand and forty thousand. The actual increase is only about one and a half to ten times, not two hundred thousand times.
If the theoretical surplus is really that large, why did gene counts grow only a fewfold? Perhaps energy alone does not determine the number of genes. Perhaps the model used the wrong denominator. Neither possibility has been ruled out.
A third challenge comes from a bacterium you can see with the naked eye. A French researcher first collected it in a mangrove swamp on the Caribbean island of Guadeloupe in 2009. More than a decade later, in 2022, researchers formally described this giant sulfur bacterium. An average cell is over nine millimeters long, and the longest approach two centimeters: the size and shape of an eyelash. Its name means “magnificent sulfur pearl.” Ordinary bacteria measure only a few micrometers. This one is thousands of times larger.
Its genome has more than eleven million base pairs, larger than those of most bacteria, and a single cell carries hundreds of thousands of copies. Its DNA and protein-making machinery are concentrated in numerous small, membrane-bound sacs. A huge vacuole occupies most of the cell’s interior, squeezing the active material into a thin layer around the outside and keeping diffusion distances short.
What it overturned was the old size record for bacteria. The largest previously known bacterium had measured less than a millimeter; this one was about fifty times larger. What it did not overturn was the need for special structures to solve the problems of diffusion and gene expression in a giant bacterium. The enormous vacuole, the hundreds of thousands of genome copies, and the little sacs show that those problems are very real.
Nor can those sacs yet be called homologues of a nucleus. What we can say is that mitochondria profoundly changed the spatial organization of energy production in eukaryotic cells. That is not disputed. Whether they were an indispensable threshold, one that could not be crossed without them, is still debated.
As for which group is more abundant, Part IV has already done the accounting. By carbon, eukaryotes make up the overwhelming majority, largely thanks to land plants. By cell number, prokaryotes dominate the Earth overall by an enormous margin.
How many times did the eukaryotic form of organization arise? Living eukaryotes share a great many complex, homologous systems: nuclei, nuclear pores, cytoskeletons, membrane-trafficking systems, and a mitochondrial ancestor. All can be traced to the same last common ancestor.
Count lineages with the basic organization of modern eukaryotes, then, and only one origin on Earth can be confirmed. That fact cannot tell us the probability of this organization arising on other planets. Our sample size is one.
But endosymbiosis itself has happened more than once: once for chloroplasts, independently in Paulinella, several times in the secondary incorporations that produced complex chloroplasts, and again, quite recently, for the nitrogen-fixing organelle. “Basic eukaryotic organization traces back to one origin” and “endosymbiosis happened only once” are not the same statement.
Now we can return to this essay’s construct. In this series, a construct is not merely a human-made criterion. Anything that draws a boundary and determines what counts as being inside it qualifies. Here, it is a unit at a higher level of organization.
A free-living cell is incorporated into another cell and loses the ability to reproduce independently. Its genome is dismantled; most of its functions are relocated or deleted. Once integration is complete, the higher-level unit begins to be reproduced and selected as a whole.
For the first time in this series, the unit on which selection acts has changed levels. In the earlier essays, when a construct could no longer accommodate something, the response was to patch it up, expand it outward, or replace it with a new construct at the same level. This essay takes a different route: a change of level. Two units that could once reproduce separately become a new unit that reproduces together.
What a construct can neither accommodate nor eliminate is called the remainder in this series. Here, the remainder is the little handful of genes that will not transfer. It is not a case of “the move is not finished yet.”
The history of mitochondrial incorporation stretches back roughly two billion years. Human mitochondria still retain thirteen protein-coding genes. Mitochondrial DNA fragments continue to enter the nucleus, at a measurable frequency of about once in every few thousand births. But a fragment arriving there does not mean the nucleus can now make a working protein and send it back to the mitochondrion. Moving DNA into the nucleus and relocating a function are not the same step.
Two billion years. Once in every few thousand births. Thirteen left. This is no isolated case. Paulinella’s young chromatophore still retains about a quarter to thirty percent of its ancestral gene repertoire. Chloroplasts still retain more than a hundred genes. These organelles, with genomes of their own, show that incorporation need not empty out the original genetic system.
In almost every cell in your body, those thirteen genes are still there right now, on that little circle of mitochondrial DNA. Bacterial-style machinery reads them using a codebook altered in a few places. This is the remainder in one of its purest forms: sequenced in full, acknowledged in full, and still there despite the continuing attempts to move it. As for why it will not go, no single explanation accounts for everything.
Extreme water aversion, the need for local control, limits on transport and assembly, base composition, lineage-specific codes: each explains part of it. The same few kinds of components remain, but each lineage keeps a different selection.
This construct achieved something those in the earlier essays did not: it changed levels. What it could not do was dissolve the incorporated partner completely.