How Squid Rewrite RNA to Reshape Thousands of Proteins

A squid hatches in cold coastal water, and the same gene that builds a transport motor in its nerve cells can end up producing a slightly different motor than it would in warmer water. The DNA is the same. The animal is the same. What changes is a message that sits for a few minutes between the genome and the protein, and that message gets rewritten by the cell itself. In the longfin inshore squid, Doryteuthis pealeii, that rewriting happens thousands of times over, and until this autumn nobody had a protein-level map of what it does.

The map arrived on 15 October 2026, when a team led from the University of California, San Francisco, published it in the journal Cell. The paper, "Functional diversification of the cephalopod proteome by RNA editing," follows a preprint posted in December 2025. Its first author is Jack Moen, its corresponding author is Danielle Swaney, and the author list also includes Joshua Rosenthal, who has studied squid RNA editing for decades, and Nevan Krogan, a UCSF systems biologist. The question the team asked is simple to state and hard to answer: when a squid edits its RNA, does it actually change what its proteins do?

A recipe that gets corrected on the way to the kitchen

To see why the question matters, start with how a cell uses a gene. DNA stays in the nucleus. When a cell needs a protein, it makes a temporary copy of the relevant gene in the form of messenger RNA, and a ribosome reads that copy three letters at a time, each triplet (a codon) calling for one amino acid. Strings of amino acids fold into proteins. The usual picture is a one-way street: the gene decides the protein, and any change to the protein has to come from a change to the gene.

RNA editing breaks that rule. The enzymes involved, called ADARs (adenosine deaminases acting on RNA), find the letter adenosine inside certain RNA molecules and chemically convert it to inosine. A ribosome reads inosine as if it were guanosine, a different letter. If the converted letter sits inside a codon, the codon can call for a different amino acid, and the finished protein differs from what the gene alone would predict. Researchers call such changes "recoding."

One of the people involved in the new study describes it with a kitchen analogy: the genome is a printed recipe, and editing swaps raisins for chocolate chips just before the instructions reach the baker. The analogy is imperfect, because the swap happens in a copy, not in the original, and it can happen in some copies and not in others. That is the whole point. The same gene can yield a mixture of protein versions, and the cell can vary the mixture.

All animals have ADAR enzymes, from jellyfish relatives to humans. In most of them recoding is a sideshow. A large survey of human genes found that only about 3% of messages carry a recoding site at all, and only around 25 human transcripts carry one that is conserved across mammals. Coleoid cephalopods, the group that contains squid, octopus and cuttlefish, are the exception, and by a wide margin.

What was already known about squid editing

The squid story began with a surprise. In 2015, researchers from Tel Aviv University, the University of Connecticut and the Marine Biological Laboratory reported in eLife that the squid nervous system carries 57,108 recoding sites and that the majority of the transcripts they studied were edited. Editing was concentrated in genes with neuronal and cytoskeletal roles. Before that paper, the general assumption in the field was that recoding is extremely rare.

In 2017, a follow-up in Cell by Noa Liscovitch-Brauer, Eli Eisenberg, Rosenthal and colleagues compared squid, two octopus species and a cuttlefish with the primitive nautilus and with the sea hare Aplysia. The coleoids had on the order of 80,000 to 130,000 editing sites in protein-coding regions; the nautilus and the sea hare had orders of magnitude fewer. The sites were concentrated in nerve tissue and affected proteins involved in how neurons fire and how they are shaped. About 1,100 sites were shared by all four coleoids, a sign that evolution had kept them for a reason.

That 2017 study also exposed a cost. ADARs need a stretch of double-stranded RNA around the target letter, often spanning hundreds of nucleotides, and a single change in the surrounding sequence can switch editing off. The authors found that the DNA around editing sites mutates less than the rest of the genome. "Mutation is usually thought of as the currency of natural selection," Rosenthal told University of Chicago News at the time, in explaining that these animals appear to suppress mutation to preserve flexibility at the RNA level. In plain terms, cephalopods seem to have traded some freedom to evolve their DNA for the ability to adjust their proteins on the fly.

The "on the fly" part is not a figure of speech. In 2012, Sandra Garrett and Rosenthal showed in Science that potassium channels from an Antarctic octopus and a tropical octopus differ by only a handful of amino acids once edited, and that editing helps explain how the channels work at very different temperatures. Eleven years later two papers in Cell in 2023 pushed the idea further. A team at UC San Diego led by Kavita Rangan and Samara Reck-Peterson found that opalescent squid raise the editing of the motor protein kinesin as the water gets colder, and the recoded kinesins performed better in the cold. A study of octopus by Rosenthal, Eisenberg and colleagues found that cooling the animals changed editing at more than 13,000 codons in the nervous system. Editing, in other words, is a dial the animal turns in response to the environment.

The gap: edits are easy to count and hard to interpret

All of this earlier work shared a limit. Editing is detected by sequencing RNA and noticing that a letter reads G where the genome says A. That tells you an edit exists in the RNA. It does not tell you whether the edit survives into a protein, whether the cell tolerates the protein, or whether the new version does anything different. RNA is not the product; protein is. Between the two sit steps where cells can discard, silence or ignore a message.

The quantity that eventually matters is called the proteome, the full set of proteins in a cell or tissue. Reading the proteome means mass spectrometry: proteins are cut into short fragments called peptides, the peptides are weighed with very high precision, and software works out which amino acid sequence matches each weight. If the sequence contains the amino acid that the edited codon would produce, rather than the one the genome predicts, that is direct evidence the edit made it into a real protein.

This is what the UCSF team did at scale. According to the paper's abstract, they confirmed the translation of 7,997 recoded sites at the protein level. That number is far smaller than the tens of thousands of sites seen in RNA, and that is expected. Mass spectrometry only catches peptides it can detect, so the figure is a floor, not a ceiling. What it gives is a large set of edits whose protein products are real and can be studied one at a time.

Our earlier news report on the paper, based on the authors' own comments to the press, adds that the work looked at nearly 8,000 sites across more than 2,000 proteins in two tissues, the optic lobe, which processes vision, and the gills, a simpler respiratory organ. The optic lobe showed more frequent and more complex editing, which fits the long-standing link between editing and the cephalopod nervous system, the most complex among invertebrates.

From a list of edits to a list of effects

A catalogue of edited sites would be a useful resource but not a discovery. The abstract claims more: that recoding reshapes the proteome in four specific ways. The authors combined the proteomics with biochemical and cellular assays, which means that for chosen proteins they made the edited and unedited versions and compared them in cells and in test tubes. The four properties they report are stability, subcellular localisation, post-translational modifications and enzymatic activity.

Stability is how long a protein survives before the cell breaks it down. A single amino acid swap can make a protein more fragile or more durable, so the same gene can produce a short-lived and a long-lived version. Localisation is where a protein ends up. Cells run on address labels, short stretches of amino acids that tell a protein to go to the nucleus, the membrane or a particular organelle. Edit the label and the protein can arrive somewhere else. In the news coverage, edited proteins were described as moving from the outskirts of the cell toward the nucleus.

Enzymatic activity is the plainest of the four. Enzymes are proteins that do chemistry, and their active sites are tuned to the shape and charge of a few amino acids. Change one of them and the enzyme can speed up, slow down or switch to a different target. The fourth property, post-translational modification, deserves its own explanation.

Switches that editing can create or erase

After a protein is built, cells decorate it with small chemical tags. Adding a phosphate group can turn a protein on or off; adding a chain of a small protein called ubiquitin can mark it for destruction or change where it goes. These tags attach at particular amino acids in particular sequence contexts. The enzymes that add them recognise a pattern.

The abstract reports that recoding "creates or ablates" such sites. That is a mechanistic consequence. If an edit turns a codon for an amino acid that cannot hold a tag into one that can, a new switch appears. If it turns the amino acid that holds a tag into one that cannot, a switch disappears. The authors say the effects reach direct changes in protein function and in protein-protein interactions, since a phosphate or ubiquitin tag is often the thing another protein latches onto.

This matters because tags are how a cell responds to its situation within seconds. A squid that can edit away a regulatory switch, or build a new one, is changing its wiring without touching the genome. It is also a reminder of why this is a genuinely different strategy from ordinary evolution: a mutation in DNA changes every copy of a protein in every cell of every descendant, while an edit changes some copies in some tissues at some times.

One ubiquitin ligase and the power plants of the cell

Among thousands of edited proteins, the paper picks one for a detailed case study: the enzyme MARCHF5. It belongs to a family called E3 ubiquitin ligases. An E3 ligase is the part of the ubiquitin system that chooses which protein gets tagged. The cell has hundreds of different E3s, and each one has a list of targets, called substrates. Changing an E3 therefore changes the tagging of many other proteins at once.

The authors report that recoding of MARCHF5 drives widespread changes in the ubiquitin tagging of its substrates and perturbs mitochondrial homeostasis. Mitochondria are the organelles that generate most of a cell's usable energy, and their upkeep, including the disposal of worn-out parts, depends on constant tagging and untagging of proteins on their surface. A shift in one ligase can therefore ripple through energy production and mitochondrial health.

"We found effects on protein activity, stability, localization, and even how they interact with other proteins," Moen told ScienceAlert. "The downstream effects even go so far as to regulate organelle health." The MARCHF5 result is the clearest example of that last sentence: an edit in one enzyme, a change in how many of its targets are tagged, a change in how a whole compartment behaves. Not many studies in this field have followed an edit that far.

Press coverage of the paper adds a comparison that is worth treating with care. It suggests recoding may make squid mitochondria more like human ones, more stable under changing conditions and better at recycling their own parts, and may touch the creatine system that supplies short bursts of energy for fast escape movements. Those are readings of the data, not measured performance of living squid, and the paper's own summary stops at saying that RNA editing can influence organelle function.

Short proteins nobody had listed, and a number bigger than the universe

Two other results in the coverage show how much of the squid proteome had been invisible. First, the team identified about 5,000 short proteins that were absent from the reference squid proteome and from Swiss-Prot, a curated database of proteins covering many organisms. Such short proteins are hard to find because standard gene-prediction software tends to ignore them, and mass spectrometry that does not look for them will miss them. Whether they all have a function is not established by the number alone.

Second, some of the edits fall in the ADAR genes themselves. That suggests a feedback loop, in which the editors modify their own activity, a plausible control system for something this powerful. The details of that loop are one of the open questions the work leaves.

The third number is the one that gets quoted. Moen said that, counting every site found in the sequencing data, the theoretical count of possible protein variants comes to about 10²⁴⁸, a figure far larger than the number of atoms in the observable universe. It is an upper bound built by multiplying independent choices at every edited site. Real cells edit some sites together and others never together, so most of those variants will never exist. "They describe an enormous potential diversity," Moen said of the figure, and how much of it the animals use is, in the coverage's words, unknown.

Why squid, and why not us

If editing is so useful, why do mammals not do it? The 2017 comparison suggests an answer that has two parts. First, editing needs RNA structure. Because ADARs work on double-stranded RNA, every recoded site depends on a stretch of sequence around it folding the right way. Holding that structure in place restricts how freely the surrounding DNA can mutate. In coleoids, windows of about 200 nucleotides around recoding sites cover roughly a quarter to two-fifths of protein-coding sequence, depending on species. For a genome, that is a heavy constraint, and the authors estimated that it purges several percent of the mutations that would otherwise show up.

Second, the payoff has to be large enough to justify the loss. In a mammal, the nervous system evolves through changes in DNA, and those are inherited and refined over millions of years. A cephalopod can add a layer of change that is reversible, different in each tissue and tunable by temperature. The authors of the 2017 paper argued that for coleoids the need to make specific changes in the neural transcriptome outweighs the standard route of neuronal evolution. The new paper does not settle why only coleoids went this way; the news report on it says the answer lies partly in the structure of their ADAR enzymes and partly in reasons that need further research. A squid ADAR2 splice variant with an extra double-stranded RNA binding region, noted in 2017, is one candidate.

The strategy also has a limit built into it. An edit lives in a copy of the message, not in the genome, so it is not inherited: each generation of squid has to edit its messages again from scratch.

What the paper does not show

A few cautions follow from how the study was built. The proteomics was done in two tissues from one species, the optic lobe and the gills, so the findings say nothing yet about the stellate ganglion that drives the squid's mantle, the arms, or the animal at different temperatures. The abstract shows effects on stability, localisation, modifications and activity for selected proteins, but it also states that the functional consequences of most recoding events remain unknown, and, as far as the abstract shows, it follows up in detail on only a few of them. For the thousands of other sites, the paper is a map and not a verdict.

The key mitochondrial result, MARCHF5, comes from cellular assays in which the edited and unedited versions were compared, not from live squid whose mitochondria were measured. Whether that same shift shows up in a squid, and whether it helps the animal, remains to be tested. And the headline diversity figure is a theoretical ceiling.

It is also worth being clear about what this article rests on. Cell's full text was not available to us at the time of writing, and the preprint server declined automated access, so the account above relies on the published abstract, our news report, press coverage that quotes Moen, and the earlier open literature. Numbers that are not in the abstract come from that coverage and are attributed to it. Anyone who needs methods, sample sizes or the full list of tested proteins should consult the paper itself.

What comes next

The immediate value is practical: a set of nearly 8,000 sites where the protein-level consequence of an edit can be tested, and an approach that others can apply to octopus and cuttlefish. The 2023 temperature studies showed that editing responds to the environment in specific proteins, kinesin in squid and thousands of codons in octopus; this paper supplies the machinery to ask what the same dial does to stability, location and tagging across a whole proteome.

There is also a research direction that has nothing to do with squid. In the same 2023 work, the UC San Diego team tested cephalopod-style substitutions in human kinesin and found that many changed how far and how fast the motors move without breaking them. Rangan said in the university's news release that cephalopods "may be able to show us where to look and what changes to make." Recoded sites in MARCHF5 and its relatives could offer the same kind of hint for the ubiquitin system, which is implicated in many human diseases.

The larger point is conceptual. A genome is often described as the book of an organism, fixed at birth. The squid shows an alternative: a book with a margin where the reader can write corrections, differently in every chapter and every season. This paper is the first proteome-scale look at what those marginal notes do. It shows that they are not noise. Whether they are the secret of the cephalopod brain is a bigger claim, and it is not one this work makes. Its claim is narrower and better supported: the edits change the proteins, the proteins change what the cell does, and in at least one case that reaches all the way to the mitochondria.

More science coverage is in our Science section.

Sources

• Moen et al., "Functional diversification of the cephalopod proteome by RNA editing," Cell, October 2026
• Preprint of the same work, bioRxiv, 16 December 2025
• Alon et al., "The majority of transcripts in the squid nervous system are extensively recoded by A-to-I RNA editing," eLife, 2015
• Liscovitch-Brauer et al., "Trade-off between transcriptome plasticity and genome evolution in cephalopods," Cell, 2017
• University of Chicago News, April 2017
• Rangan et al., RNA recoding in cephalopods tailors microtubule motor protein function, Cell, 2023
• UC San Diego Today, 8 June 2023
• Wang, Li and Wu, commentary on the 2023 Cell papers, Signal Transduction and Targeted Therapy, 2023