Human DNA Has a Hidden Layer That Controls Protein Levels

Think about the word "colour." And then think about "color." Same meaning. Different spelling. Now imagine that, depending on which spelling your body uses inside a gene, it ends up making four times less of a crucial protein. That's not a metaphor — that's what a team of researchers at the University of Colorado School of Medicine found, and they published it in Cell Reports in December 2023. The discovery settles a long-running argument about why certain genes consistently underproduce — and the answer isn't what most biologists expected.

A Puzzle That Kept Getting Swept Under the Rug

Here's something that might surprise you. Your DNA doesn't just have one way to write "build leucine here." It has six. Most amino acids — the building blocks of every protein in your body — have multiple three-letter codes that all mean the same thing. Scientists call them synonymous codons. For most of the 20th century, the working assumption was: synonymous means interchangeable. A rose by any other name, and all that.

Then cracks started appearing. Researchers noticed that genes using certain spellings reliably produced less protein than genes using others. The first instinct was to blame the messenger RNA — the molecule that ferries gene instructions from DNA to the protein-building machinery. Slow spellings, it turned out, speed up mRNA destruction. That seemed like a tidy enough explanation. But it kept not quite adding up. Studies on a well-known cancer gene called KRas showed protein shortfalls far too large to blame on mRNA loss alone. Something else was cutting output. Nobody could pin it down.

Why Scientists Were Only Looking at Half the Problem

Lead researcher Chloe Barrington, working in Olivia Rissland's lab, ran what sounds like an almost simple experiment. Her team built two versions of several genes — one using fast-read spellings, one using slow-read spellings — then measured both the messenger RNA produced and the actual protein that came out the other end. Same gene. Different spelling. Everything else identical.

The mRNA gap was real, but modest. The protein gap was something else entirely. For one set of genes, the mRNA difference was 2.3-fold. The protein difference came out at 9.4-fold. That's not a rounding error — that's a factor of four sitting unexplained on the table. The same pattern held across multiple genes, in human cells and in fly cells, lab line after lab line. The slow spellings were doing something to protein output that had nothing to do with mRNA getting destroyed faster. A whole extra layer of suppression, invisible to anyone just counting mRNA.

How Does a Different Spelling Actually Reduce What Gets Built?

This took the team down a long list of "not that." Not the protein falling apart faster — they measured protein lifetimes and found no meaningful difference. Not the building machinery falling off midway through a gene — ribosome profiling showed the machines ride all the way to the end. Not a cellular alarm triggered by traffic jams — knocking out that alarm system made no difference to the numbers.

What they eventually traced it to is something much earlier in the process. There's a small protein called eIF4E that acts essentially like a key in the ignition of protein production. It has to grab onto the front end of a messenger RNA before any building machinery can even assemble. With slow-spelled genes, eIF4E gripped significantly less tightly — meaning fewer build jobs ever got started at all. The team confirmed this with a smart side experiment: they rewired their gene so that protein production could start via a completely different route, one that skips eIF4E entirely. That bypass route? Totally unaffected by the slow spelling. The penalty vanished. The slow spelling suppresses output specifically by making it harder to start — not harder to finish.

Live-cell imaging of individual mRNA molecules sealed it. Each slow-spelled transcript had a median of 5.4 building machines running on it at once, versus 7.4 on fast-spelled ones. The slow-spelled mRNAs didn't go dark. They just ran with a smaller crew.

Why This Changes How We Make Medicines and Treat Disease

mRNA vaccines — the same platform behind the Pfizer and Moderna COVID shots — are already built on spelling optimization. The logic has always been: pick fast-read spellings so the body produces lots of protein quickly, and stabilize the mRNA so it sticks around long enough to do its job. Drug developers have been doing this for years. What this study adds is a second dial that nobody was turning: how easily does the cell decide to start reading the mRNA in the first place.

Nail that, and you might squeeze meaningfully more protein out of the same mRNA dose — which matters when you're designing vaccines, cancer immunotherapies, or any treatment that relies on the body making a specific protein on demand. There's a cancer angle too. KRas, the gene mutated in roughly a third of all human cancers, happens to use a lot of slow spellings. Part of why it underproduces protein compared to its close relatives might be this exact bottleneck. Whether that's ever targetable therapeutically is a long way off. But it's a new lens on a very old problem.

What Nobody Has Figured Out Yet

The big open question is deceptively simple: how does the spelling of a word in the middle of a gene reach back to the very beginning and weaken the grip of eIF4E? The mRNA-destruction pathway has a known messenger — a protein called CNOT3 that literally sits inside a slow-moving ribosome and sets off a chain reaction. No equivalent has been found for the initiation side. Something is sensing the slow reading and signaling upstream. Nobody knows what it is.

The study also only tested specially built reporter genes, not the full catalog of natural human genes. That's a harder experiment to run cleanly, since real genes carry all sorts of other features that muddy the analysis — and the researchers say so outright. Still, the pattern showing up in other published data strongly suggests this isn't a quirk of their constructs. It looks like something cells have been doing all along, quietly, while scientists were mostly just counting mRNA and calling it a day.

• Spelling matters, not just the message — The same gene, written with slow-read spellings, can produce four times less protein than a fast-read version, even when mRNA levels are nearly identical.
• The door closes before it opens — Slow spellings weaken the grip of a key protein that has to latch onto the mRNA before any building begins, throttling output at the very first step.
• Two brakes don't just double up — mRNA destruction and this new initiation block are separate systems; disable one, and the other compensates harder, which means future therapies will need to account for both.

"Our results reveal a potent mechanism of regulation by codon usage where nonoptimal codons repress further rounds of translation." — Barrington et al., Cell Reports, 2023.