473 Genes and 149 Mysteries

In-depth articles · Aug 18, 2026 · 10 min read · By No Design Editorial

Venter's team stripped a cell to the bone and it still needed 473 genes, and could not say what a third of them do. What that licenses, and what it does not.

The short version

• Stripped to the bone, the smallest cell anyone has built still needed 473 genes, and a third of them could not be given a specific function.
• The first design failed outright. Quasi-essential genes and synthetic lethal pairs meant the requirements could not be read off the sequence, even with the whole sequence in hand.
• 473 is the floor for a pampered parasite in rich medium. A free-living bacterium in the open ocean needs 1,354 genes.
• The honest claim is narrow: nobody can currently name a simpler working autonomous system. Not that none could exist.

In 2016 a team led by Clyde Hutchison and Craig Venter published the result of years of work in Science. They had taken a synthetic version of the Mycoplasma mycoides genome, a strain called JCVI-syn1.0 carrying 901 genes, and set about cutting away everything the cell did not need. What they ended up with was JCVI-syn3.0: a genome of 531,560 base pairs carrying 473 genes, 438 of them coding for proteins and 35 for RNAs. They had removed 428 genes, roughly half the genome by length, and the cell was still alive.

It is a landmark experiment, and it gets quoted a great deal in arguments about design. Some of that quoting is careless. So before the number is used for anything, it is worth going through what the experiment was, what it was not, and what a decade of follow-up work has done to it.

First, what it was not

The headlines in March 2016 said scientists had created synthetic life. They had not, and the design argument does not need them to have.

What was chemically synthesised was the genome. The DNA was assembled in yeast and then transplanted into a living recipient cell, Mycoplasma capricolum, which supplied the membrane, the ribosomes, the transfer RNAs, the polymerases and the entire metabolic inventory required to read the new instructions and act on them. Nothing was built out of non-living chemistry. The paper's own wording is careful about this: it describes using whole-genome design and chemical synthesis to minimise an existing genome. The object of the verb is the genome, not the cell.

Two more things belong in the same paragraph of honesty. Mycoplasma is a host-dependent parasite that had already lost most of its biosynthetic machinery through ordinary evolution long before anyone in a laboratory touched it, so the project began from an unusually stripped baseline rather than from a generic bacterium. And syn3.0 was grown in a rich medium. The authors say so themselves, in as many words: "The work described here has been conducted in medium that supplies virtually all the small molecules required for life."

That sentence should be read slowly, because it means 473 is not the minimum number of genes for life. It is the minimum for a degenerate parasite living in a warm incubator with no competitors, no predators, no temperature swings, and a supply of nearly every molecule it would otherwise have to make for itself.

Conceding that costs nothing, and it is worth noticing which way it moves the number. Take away the incubator and the requirement goes up, not down. The smallest genome of any cell known to replicate independently in nature belongs to Pelagibacter ubique, the bacterium that dominates the open ocean, and it runs to 1,308,759 base pairs and 1,354 genes, with no pseudogenes and no recent duplications: a genome genuinely honed by selection rather than merely decayed. That is the realistic floor for something that has to earn its own living.

The part that gets skipped: the design failed

The most interesting thing in the paper is not the final number. It is what happened on the way to it.

The team began by using transposon mutagenesis to work out which genes the cell could survive without, then designed a genome that deleted all of them at once. That design failed to produce a viable cell. Not a sickly cell, not a slow one. No cell.

Two reasons emerged. The first was that essentiality is not a yes or no property. Alongside genes that are strictly essential and genes that are dispensable there is a large class the authors call quasi-essential: genes whose loss does not kill but impairs, across a continuum from barely noticeable to severe. Remove enough of them together and the cell drops below the threshold at which it can be kept alive at all.

The second reason is the more striking. Synthetic lethal pairs: two genes providing alternative routes to the same necessary end. Delete either one and the cell is fine, because the other covers for it. Delete both and the cell dies. Every experiment that tests genes one at a time will report both as expendable, and the report will be wrong.

The recovery was to abandon whole-genome design and split the genome into eight segments that could be synthesised and tested independently, mixing reduced segments with unreduced ones in combination, through three rounds of design, build and test.

Sit with what that means. This team had the complete sequence of the organism, decades of Mycoplasma literature, the best synthesis technology on the planet, and a specific, well-defined question: which of these genes does the cell actually need? They could not answer it by inspection. They had to find out by breaking things and seeing what died. A genome is not a parts list from which the requirements can be read off. It is a network whose behaviour was not derivable, even from complete knowledge of its parts.

The 149, stated precisely

Now the figure this article is named after, because the popular version of it is looser than the paper.

The team sorted all 473 genes into five classes by how confidently a function could be assigned: 232 equivalog, where the function is essentially certain; 58 probable; 34 putative; 84 generic; and 65 unknown. The 149 is the last two classes added together. "Generic" means the protein is clearly identifiable as, say, a kinase, but with no consistent clue as to what it acts on. "Unknown" means it could not be reliably categorised at all.

So the accurate sentence is not "a third of the genes were complete blanks." It is that a third could not be given a specific biological function, and that about 65 of them, roughly one gene in seven, were genuinely opaque. That is a smaller claim than the one usually made, and it is still a remarkable one, because these are genes the cell dies without.

Venter put it plainly in the announcement: "Our attempt to design and create a new species, while ultimately successful, revealed that 32% of the genes essential for life in this cell are of unknown function." Elsewhere he was blunter. "We're truly like aliens from another planet trying to work out what these parts do." And: "It's a very humbling set of experiments."

Hutchison's line from the same announcement is the one that ought to be quoted alongside those, and rarely is: "Our goal is to have a cell for which the precise biological function of every gene is known." That is a scientist naming an unmet target, not conceding a permanent barrier. Quoting Venter's humility while omitting Hutchison's ambition would be a cheap way to win a point.

And what has happened since

The 149 has not stood still, and any argument built on it should say so.

By 2019 the count of unannotated proteins had fallen to around 112, largely through database updates and homology transfer rather than through anyone doing new biochemistry. By 2022 a survey aimed squarely at the problem reported roughly 90 of the strain's 452 protein-coding genes still unannotated. By 2025, protein language models searching for remote relatives could propose a functionally annotated homologue for about 93 per cent of the minimal genome, leaving something like thirty proteins without even a computational guess.

That is real progress and it should be stated as such. But it is worth being precise about the kind of progress it is. A predicted gene ontology term and a detected remote homologue are not the same thing as a demonstrated biochemical activity.

The clearest experimental gain of the decade is the 2021 follow-up on cell division. It turned out syn3.0 could not reliably do the single most basic thing a cell does: it grew into filaments, vesicles and irregular shapes rather than dividing cleanly. Adding nineteen genes back produced a usable strain, syn3A, and of those, seven were needed to restore normal shape and division: two recognised division genes, a hydrolase whose substrate is unknown, and four membrane proteins of unknown function. That assigns a phenotype to five proteins. It does not tell us what they do.

The honest summary is this. The number of genes with no functional assignment has fallen from 149 to roughly 90 over six years, computation now proposes candidates for most of the rest, and only a handful have been experimentally pinned down. A gap that is measurably closing is a poor foundation for a claim that it cannot be closed, and this article does not make that claim.

The objection that actually matters

Here is the serious reply, and it should be granted in full.

A minimal modern cell is not the first cell. JCVI-syn3.0 sits at the end of something like four billion years of a lineage refining itself, discarding what it stopped needing and elaborating what it kept. Its requirements are the requirements of a late, specialised, thoroughly evolved system. They tell you nothing directly about what the earliest replicator had to have, and the earliest replicator may have been far simpler.

The point holds even for the deepest ancestor anyone can reconstruct. The last universal common ancestor, LUCA, is not the first life. It is the most recent organism from which everything now alive descends, which is a different thing entirely. A 2016 reconstruction attributed 355 protein families to it and described an anaerobic, heat-loving, hydrogen-dependent organism, and even that stripped-down portrait already includes a working genetic code, ribosomes and an ATP synthase. A 2024 study pushed LUCA back to around four and a quarter billion years ago and gave it a gene repertoire in the ordinary prokaryotic range. Whatever LUCA was, it was a survivor's snapshot taken well after the interesting part had happened.

And simpler self-replicating things do exist. In 2009 Gerald Joyce's laboratory reported a pair of RNA enzymes that catalyse each other's synthesis and can amplify exponentially and indefinitely, with no proteins and no cell. Jack Szostak's laboratory has shown fatty-acid vesicles that grow and divide under simple physical conditions with no genes at all, and non-enzymatic copying of short RNA templates. None of that is nothing, and anyone who tells you origin-of-life research has produced no results is not reading it.

So the claim this article makes is narrow, and the narrowness is the point. It is not that a simpler working system could not exist. It is that nobody can currently name one. The Joyce replicator does not copy a sequence; it joins two oligonucleotides that a chemist supplied, in a buffer a chemist prepared, and take the chemist away and nothing runs. The Szostak vesicles divide, but they carry nothing and inherit nothing. Between those and the simplest thing that keeps itself going unaided there is a distance nobody has crossed, and the researchers working on it say so in their own papers.

That is a claim about the state of knowledge, and an experiment could overturn it tomorrow. It has not been overturned yet, and it has not been for a long time.

The number this article is not going to use

Every one of those 473 genes specifies a protein, and every protein has to fold into one particular three-dimensional shape or it does nothing at all. This is the point at which design writing usually reaches for a very large number, and it is worth explaining why this one will not.

The figure in circulation comes from Douglas Axe's 2004 paper in the Journal of Molecular Biology, which estimated the prevalence of sequences adopting a functional enzyme fold at around one in 10⁷⁷, and from Stephen Meyer's composite of one in 10¹⁶⁴, which multiplies that by the odds of correct peptide bonding and of uniformly left-handed amino acids. Axe's paper is real and peer-reviewed. But it estimated fold rarity for one specific enzyme domain, and extrapolating from it to proteins in general is contested. The sharpest counter is Keefe and Szostak's 2001 paper in Nature, which fished through random sequence libraries and found functional ATP-binding proteins at a rate of roughly one in 10¹¹. Eleven orders of magnitude and seventy-seven are not the same claim. Meyer's multiplication also treats as independent several factors that are not obviously independent.

There is a second problem with the whole move, and the honest version of the argument has to face it. Evolution does not draw sequences at random from the whole of sequence space. It works within existing protein families, along gradients, from starting points that already fold. A calculation that models it as a single blind draw is answering a question nobody is asking.

What survives all of that is smaller, and it needs no probability estimate at all. The 473 genes have to be there together. The experiment showed that you cannot even establish which ones matter by testing them one at a time, because the pairs cover for each other and the quasi-essentials fail gradually rather than cleanly. Whatever produced this, it had to arrive at a working combination, and the combination is the unit.

What the experiment licenses

Not "life is impossible without a designer." That is more than one experiment can carry, and a claim that outruns its evidence is worth less than the evidence it outran.

What it licenses is this. The floor of autonomous life, measured rather than imagined, is high, and it came out higher than the people who set out to measure it expected. The parts cannot be identified one at a time, because their necessity is relational rather than individual. A third of the essential genes could not be explained when the cell was built, and a decade later most of those still have predicted functions rather than demonstrated ones. And the closer anyone looks, the more the requirement rises: a free-living cell in the open ocean needs roughly three times what the pampered one in the incubator needs.

A team of the best people in the field, working with the full sequence in front of them, built the simplest living thing anyone has ever built, and then could not say what a third of it was for. That does not end a debate. It is a fairly precise measurement of how far we are from the bottom.

From the Introduction to No Design.

Read the rest of the argument at No Design, and find the book and the wider project at How to Muslim.

Tags: Minimal Cell, Genome, Origin of Life, Quran