The error threshold on genome length

Why can a genome never grow longer than its own copying accuracy permits?

Genomes vary enormously in length, and the variation is not random. RNA viruses top out around thirty thousand bases. Bacteria run to millions. We run to billions. If a longer genome buys more capability, why did the RNA viruses not simply get longer?

The tempting answer is that they had no need, or no time. I would like to test a harder possibility instead: that there is a ceiling, that the ceiling is set by nothing more than how accurately the thing copies itself, and that no amount of selection can raise it.

Let us treat a genome as a message being sent through a noisy channel, where the transmission is replication and the noise is mutation. Ask a simple question: how many errors does one copy introduce?

If the per-base error rate is u and the genome is L bases long, the expected number of mistakes per copy is simply u times L. So the interesting quantity is never the error rate alone and never the length alone — it is their product, the mutations per genome per replication.

Now what happens as that product approaches one? Roughly, every offspring differs from its parent somewhere. The exact sequence that natural selection has painstakingly assembled — the best one, the master sequence — is copied faithfully only rarely, and it is being produced anew from its own mutants even more rarely.

Here is where you have to be careful, because there is an obvious objection: surely selection just keeps killing the bad copies? True, and selection is doing exactly that. But selection can only act on what it can distinguish, and it can only preserve the master sequence if that sequence is regenerated faster than it is degraded. Manfred Eigen worked this out formally in 1971. The population settles not on one sequence but on a cloud of related mutants — a quasispecies — and the master sequence persists in that cloud only while u times L stays below roughly the natural logarithm of its selective advantage over the average mutant.

That logarithm is the reason the answer is so blunt. A tenfold fitness advantage gives you a budget of about 2.3; a hundredfold gives about 4.6. Because the logarithm crushes even large fitness differences into small numbers, the practical constraint is that u times L must be of order one, so L can be no more than roughly one over u. Copy at one error in ten thousand bases and you may carry about ten thousand bases. No more, whatever the selection pressure.

And past that point, what happens? Not gradual decay. The transition is sharp: heritable information dissolves into the mutant cloud, and the population drifts through sequence space rather than sitting on a peak. Eigen called this the error catastrophe.

Now test the prediction against the world, because this is where it gets satisfying. RNA viruses copy at roughly one error in ten thousand to a hundred thousand bases and carry genomes of about ten thousand bases — sitting right on the line. DNA organisms with proofreading polymerases and mismatch repair achieve error rates orders of magnitude lower, and their genomes are correspondingly longer. And the pointed case: coronaviruses have the largest known RNA genomes, near thirty thousand bases, and they are the RNA viruses that carry a proofreading exonuclease, nsp14-ExoN. Disable it experimentally and mutation rates rise while the genome becomes unstable. Does the correlation between fidelity machinery and permitted length look like a coincidence to you?

But notice the trap this sets. To copy more accurately you need proofreading enzymes. Proofreading enzymes must be encoded, which requires a longer genome. A longer genome requires better fidelity. Eigen named this circularity his paradox, and how the first replicators escaped it — through compartmentalisation, cooperating short sequences, or something not yet proposed — is a genuinely open question in origin-of-life research, not a solved one.

Three qualifications, since this is a model that is often overstated.

First, the threshold is not a hard wall in nature. The clean version assumes an unchanging fitness landscape, a well-mixed asexual population and one distinct master sequence. Recombination, sex, population structure and neutral mutations all shift where the edge falls. The scaling L of order one over u is robust; the precise coefficient is not.

Second, "error catastrophe" is a contested label for what happens to a real virus pushed over the edge. Mutagenic drugs such as ribavirin and favipiravir do drive viral populations to extinction, but most workers now describe the effect as lethal mutagenesis — simply too many lethal mutations for the population to sustain itself — rather than as Eigen's information-loss transition. The two are distinct mechanisms with similar outcomes, and the distinction is still argued.

Third, sitting near the threshold is not obviously a failure. A high mutation rate is also how an RNA virus explores antigenic space fast enough to outrun immunity, so the observed rates may reflect a fidelity-adaptability balance rather than a limitation the virus would remove if it could.

• How compartmentalisation and cooperating replicators are proposed to break Eigen's paradox.
• Why Drake's rule — a roughly constant mutation rate per genome per replication across microbes — falls out of the same argument.
• How the fidelity-versus-adaptability trade-off shapes drug resistance.

Every term the collection defines is gathered in the glossary.