One Letter, No Cut: In Vivo Base Editing

The Cut Is the Problem

In the summer of 2024, a baby boy in Pennsylvania started to die from the food he was fed. His parents called him KJ. Within days of birth, ordinary protein, the protein in milk, was poisoning him. His body couldn’t get rid of ammonia, the toxic waste left over when the body breaks protein down, and the ammonia kept climbing toward the levels that swell the brain and kill. His disease is so rare that maybe one baby in a million is born with its severe form, and about half of those infants don’t survive infancy. There was no drug for it.

So a team of doctors and scientists built one. A medicine for exactly one patient on Earth, made from scratch in roughly six months, and dripped into KJ’s vein in early 2025. The most surprising thing about that medicine is what it didn’t do. It never cut his DNA.

That cuts against almost everything the public has been told about gene editing. The technology that made editing famous is CRISPR, and the popular image of CRISPR is a pair of molecular scissors. The image is earned. In 2012, Jennifer Doudna and Emmanuelle Charpentier showed that a system bacteria use to chop up invading viruses could be reprogrammed to cut DNA at any address a scientist chose. The targeting is the genius of it. A short guide molecule acts like a postal code, steering a cutting protein to one exact spot among the three billion letters of the human genome. Doudna and Charpentier shared the 2020 Nobel Prize in Chemistry for it, and the Nobel committee called their invention the “genetic scissors.”

But scissors cut, and a cut in DNA is more violent than it sounds. DNA is a double strand, the famous twisted ladder, and classic CRISPR slices clean through both rails at once. To the cell, a clean break through both strands is an emergency: the signal that a chromosome has shattered. So it drops everything to glue the ends back, fast and without a blueprint. In the rush it often loses a few letters or jams in extra ones at the splice. Sometimes it deletes long stretches to either side, or stitches the wrong ends together and scrambles a whole region. The break gets sealed, but the neighborhood around it can come back damaged. If all you want is to switch a gene off, that’s tolerable. Broken is broken. But if you’re trying to repair one, where the whole point is landing on an exact sequence, it’s a serious liability.

So a different idea took hold: keep CRISPR’s genius, throw away its blade. Keep the postal-code targeting that finds one spot in three billion letters. But on arrival, instead of cutting, quietly rewrite a single letter of the code in place. In 2016, a Harvard chemist named David Liu and his lab built the first tool that could. They took CRISPR’s targeting machinery, dulled the scissors so they could grip the DNA without slicing through it, and bolted on a second part: a chemical tool that converts one DNA letter directly into another. DNA is written in just four letters (A, C, G, and T), and a huge number of genetic diseases come down to one of those letters being wrong. Liu’s tool turns the wrong letter into the right one without ever breaking the strand. He puts the contrast plainly. The original CRISPR was a pair of scissors. This is a pencil.

Where that pencil works turns out to matter enormously. The first CRISPR medicine ever approved treats sickle-cell disease the cautious way. Doctors remove a patient’s blood-making cells, edit them in a dish, check the result, and infuse the corrected cells back. Editing outside the body lets you inspect the work before you commit to it. The harder and more ambitious route is to edit inside the living patient (in vivo, Latin for “in the living body”) by injecting the machinery and letting it hunt down its target among the body’s own organs. There’s no dish, no inspection, no taking it back. Once the injection is in, the editor changes whatever it reaches, in a body you can’t pause. That’s the danger. It’s also the only way to reach the organs you can’t remove and return: the liver, the brain, the eye, the muscle. In vivo is where the stakes are highest, and where the real frontier is.

KJ’s disease lives in the liver. His condition, CPS1 deficiency, knocks out the first step of the urea cycle, the chemical assembly line the liver uses to pack ammonia into urea and flush it out in urine. The instructions for the enzyme that starts that line are written in a gene, and both of KJ’s copies were misspelled, one error inherited from each parent. The enzyme never worked. Every gram of protein he digested made ammonia he couldn’t clear. Babies with the severe, newborn form are kept alive on ammonia-scavenging drugs and a near-proteinless diet while they wait for a liver transplant. Many never reach one.

The team that took his case (at Children’s Hospital of Philadelphia and the University of Pennsylvania, led by the physician Rebecca Ahrens-Nicklas and the gene-editing researcher Kiran Musunuru) read out KJ’s exact misspelling and designed a base editor to correct one of his two faulty copies. Rewrite the single wrong letter, switch the dead enzyme back on. To get it to his liver, they packaged the editor not as a finished protein but as instructions the cell could read, sealed inside tiny bubbles of fat. Injected into a vein, those fat bubbles drain naturally to the liver and dump their cargo there. The editor reached the one organ that needed it.

Here’s why the case is a landmark. The whole therapy (designed, tested in animals, manufactured to clinical standards, and cleared by regulators) came together in about six months. Take a moment and let that sink in. That’s a fraction of the time even a fast gene-editing program usually needs, and the FDA cleared it about a week after the application was filed. The drug was a literal one-off, named for the boy it was built for. KJ got his first infusion in late February 2025, and follow-up doses that spring.

The early results were enough to publish. Within weeks KJ could handle more protein. His scavenger drugs were cut by roughly half. He caught the ordinary viral infections that would normally spike his ammonia into a crisis, and weathered them without one. He started hitting developmental milestones. He went home.

His doctors are careful about what they claim. The follow-up is short, and they can’t yet safely biopsy his liver to count how many cells were actually fixed. The work was presented in May 2025 and published in the New England Journal of Medicine.

But the lesson is the one the whole field has been circling. The dramatic move, molecular scissors slicing through the double helix, is what won the Nobel Prize and the headlines. The edit that actually reached inside a human body, the one that rebuilt a gene in a dying infant, is the quiet one that refuses to cut.

From Scissors to Pencil

CRISPR didn’t start in a lab. It started as a survival trick in bacteria. A microbe that lives through a viral attack keeps a souvenir of the encounter, a short snippet of the virus’s genetic code, filed away inside its own DNA like a mugshot in a police archive. When the virus comes back, the microbe pulls the mugshot, recognizes the intruder, and sends a protein to slice the viral code apart. It’s a crude immune system. And the crucial part is that it remembers.

Around 2012, biologists realized the mugshot could be forged. Hand the system any genetic address you want, and its slicing protein will travel to that one spot among three billion letters and cut. A search function for the genome, steerable to any target. That single trick turned an obscure bacterial defense into the most powerful tool biology had seen in a generation, the discovery that would win a Nobel Prize. But the cut was also the catch. A clean slice through both rails of the double helix is exactly the move that scrambles a gene. And a tool meant to mend a gene one letter at a time can’t afford to make it.

The way around it came out of a single lab in Cambridge, Massachusetts: the group run by David Liu, a scientist working in the borderland between chemistry and biology. The insight was to stop thinking like a surgeon and start thinking like a chemist. Each letter of DNA isn’t an abstract symbol. It’s a small molecule, a particular cluster of atoms. Rearrange a few of those atoms and one letter quietly becomes another, no cutting required.

In 2016 the lab built a tool that did exactly this, for a single swap. It clamped CRISPR’s targeting onto the DNA (close enough to grip, but with the cutting blades switched off) and then deployed a borrowed enzyme that works like a chemical eraser. The enzyme scrubs one specific chemical tag off the letter C, and once that tag is gone, C becomes a molecule the cell reads as T. The first such eraser was lifted, improbably, from a rat. There was one more sleight of hand to make the change stick. A fresh chemical edit on one strand is the kind of irregularity a cell’s proofreaders might notice and reverse. So the tool also makes a tiny scratch in the opposite strand, a nick that fools the cell into treating the rewritten strand as the correct master copy and rebuilding the other to match. The edit sets, like ink drying. No break, no scramble, no foreign DNA. Just a C rewritten as a T, in place.

But one tool isn’t the whole alphabet. DNA’s letters pair off in a fixed way (A always sits opposite T, C always opposite G), so there are only a handful of possible single-letter corrections, and the 2016 eraser could manage just one of them. The correction geneticists wanted most was out of reach. Roughly half of the single-letter typos known to cause human disease follow one pattern: a spot that should read G has decayed into an A. Undoing that means turning the stray A back into a G. And here the field ran straight into a wall. Nowhere in any living thing, no bacterium, no plant, no animal, does an enzyme exist that performs that conversion on DNA. Evolution never needed one, so it never built one. The eraser for C had been waiting inside a rat, ready to borrow off the shelf. The tool for A didn’t exist anywhere on Earth. So Liu’s lab made the one nature had skipped.

The work fell largely to a researcher named Nicole Gaudelli, and her method wasn’t blueprint-and-build. It was closer to breeding. She started with the nearest thing nature offered: an enzyme that performs a similar atom-swap, but only on RNA (DNA’s short-lived cousin), never on DNA itself. Then she forced it to change careers. She made enormous numbers of slightly mutated copies of the enzyme, loosed them inside bacteria rigged so that only a copy able to edit DNA could survive, killed off the failures, and bred the survivors. Round after round, seven generations of this molecular husbandry in all, the enzyme drifted toward a skill it had never possessed. This was evolution run at lab speed and bent to a purpose: the same blind process that took billions of years to invent the first enzymes, compressed into a matter of weeks and aimed at a single target.

What Gaudelli ended up holding was genuinely new under the sun: a lab-grown protein that reaches into DNA and converts an A into a molecule the cell reads as G. Fixed onto the same disarmed CRISPR targeting, it became the adenine base editor, reported in 2017. With both editors in hand, the lab could now, in principle, correct the majority of the single-letter misspellings behind inherited disease. The trick of forcing a protein to evolve a function nature never gave it would become one of the field’s most powerful and most demanding instruments.

A discovery like that doesn’t sit quietly in a journal. Within months of the second editor’s debut, the same Cambridge scientists were assembling a company to turn these molecular pencils into medicines. Beam Therapeutics was founded in 2017 by Liu and two other gene-editing pioneers, Feng Zhang and Keith Joung. The pace from bench to business was startling. The first editor appeared in print in 2016. The company was incorporated the next year. By February 2020 Beam was trading on the public markets, having raised $180 million in its stock debut on top of more than $200 million from private backers. A laboratory curiosity had become a company worth billions in under four years, money committed on the promise of the science years before any of these editors had touched a single patient. Beam cast a wide net, aiming to edit cells both outside the body (drawn out, fixed in a dish, checked, and returned) and, harder, inside it.

The bolder bet belonged to a second company. Editing cells in a dish and reinfusing them is the cautious route, and it works for blood, which can be drawn and given back. But the heart and the liver can’t be taken out, corrected, and put back. They have to be edited inside the living patient or not at all. Harder problem, far larger prize. In 2018 a Boston cardiologist named Sekar Kathiresan left his hospital post to chase it, founding Verve Therapeutics with the gene-editing researcher Kiran Musunuru. Verve aimed squarely at the world’s leading cause of death: heart disease. The plan was audacious in its plainness: a single injection that permanently switches off a gene in the liver called PCSK9, whose normal job is to keep cholesterol circulating in the blood. Switch the gene off, and cholesterol falls and stays down. One shot meant to replace a lifetime of daily pills.

In 2022 Verve dosed the first human being ever to receive an in vivo base editor, injecting its therapy into a patient with an inherited form of dangerously high cholesterol. The patient’s cholesterol fell sharply, the first proof that a base editor could be sent into a living person, find its target organ, and work. In June 2025 the pharmaceutical giant Eli Lilly agreed to buy the company in a deal worth up to $1.3 billion. A wager that one permanent edit could do what daily medication has done for decades.

None of this is mere backstory. The choices locked in during those first few years still shape every in vivo base-editing medicine being built today.

The choice not to cut, to rewrite chemically instead of slicing, is why these therapies sidestep the broken-chromosome chaos that ordinary CRISPR risks. The choice to deliver the editor as fleeting instructions rather than a permanent fixture (a burst of molecular directions the cell follows once and then breaks down) is why the machinery does its job and disappears, instead of lingering for months and making stray edits no one wanted. And the choice to aim first at the liver was no accident of nerve: the fatty bubbles that ferry the editor through the bloodstream happen to drain there, which made the liver the one organ the technology could reliably reach before any other.

KJ’s corrected liver and Verve’s silenced cholesterol gene are two answers to the same early question. The pencil, the disappearing ink, and the liver as the first page were all written into the field’s design inside a single Boston lab and the handful of companies it spun out. All before a single patient had ever been dosed.

Rewriting a Single Letter

Everything else (the delivery, the companies, the patients) exists to serve one tiny act: changing a single letter of DNA from a wrong one to a right one. That’s the whole game. So let’s get concrete about what that act actually is, and why it’s so hard to pull off. Start with the letter, and why a single one can matter so much.

A gene is a stretch of DNA that spells out the recipe for one protein, and proteins are what build and run the body. The cell reads a gene the way you read a sentence, except every word is exactly three letters long. Each three-letter word, a codon, names one building block to clip onto a growing chain. Snap the blocks together in the order the gene dictates and the finished chain folds into a working part: an enzyme, a hormone, a fiber of muscle. Change one letter and you change one word. Usually that means one wrong building block in the chain, or a word that now reads stop here, in which case the protein gets cut off half-finished.

Sickle-cell disease is the textbook case. In the gene for hemoglobin (the protein that carries oxygen inside red blood cells), a single letter in the sixth codon is wrong. That lone substitution swaps one building block for another, and the swap is enough to make the hemoglobin molecules stack into stiff rods that warp soft round cells into rigid crescents. Cells that snag in blood vessels and cause a lifetime of pain and organ damage. One letter out of billions. One wrong building block out of hundreds. A devastating disease.

Sickle cell isn’t unusual in its cause, only in how cleanly it shows the rule. Of the more than 75,000 genetic variants catalogued as causes of human disease, the single largest class is exactly this: a one-letter typo, a single rung of the DNA ladder carrying the wrong character. Thousands of distinct conditions trace back to such typos. A tool that could reliably correct one chosen letter in one chosen gene, inside a living person, would be a key that fits an enormous number of locks.

And that’s precisely the thing scissors can’t do. The genius of CRISPR is finding the spot; its limit is what it does on arrival. A clean cut through both rails of the ladder, followed by the cell’s frantic patch-up, leaves the local sequence garbled. And because the cell reads in strict three-letter steps, losing or gaining even a letter or two knocks every downstream word out of register, so the rest of the recipe reads as nonsense. For one purpose that’s exactly what you want. To switch a gene off, to silence one whose product is harmful, scrambling its opening lines into gibberish is the whole point. The recipe becomes unreadable, no working protein gets built, the gene goes dark. This is why classic CRISPR excels at disabling genes.

By contrast, correcting a typo is the opposite job. You’re not trying to wreck a word. You’re trying to fix one character and leave everything around it pristine. Blowing up the misspelled word and letting the cell improvise a repair is no way to restore an exact spelling. The job demands a tool that can settle onto the single wrong letter and change just that one, with no break in the strand at all.

A three-part machine

Enter the base editor. To pull this off it has to manage three separate feats, so it’s built from three parts fused into a single molecule you can inject. It has to find one address among billions of letters. It has to stop there and hold absolutely still. And it has to perform the chemical rewrite. Each part owns one of those jobs.

The first part is the address-finder. It works like a set of GPS coordinates written in the same chemical language as the target: a short single strand, a molecular cousin of DNA called RNA, carrying a run of about twenty letters that spell out, in complement, the exact sequence the editor is meant to land on. This guide RNA threads along the genome until its twenty letters meet their matching partners in the DNA and lock on, the way a key slides home into the one lock whose pins it fits. Load the machine with a different guide and you send it to a different address. The aim is reprogrammable by rewriting twenty letters, nothing more.

The guide carries the coordinate. But something has to grab the DNA once you arrive and hold the position. That’s the second part: the very protein that, in ordinary CRISPR, does the cutting: a molecular machine called Cas9. In a base editor, though, its blades are deliberately broken. A small change to its structure leaves Cas9 able to clamp onto the double helix and grip it firmly, but no longer able to slice it. Parked on the target like a car with its engine running and its wheels locked, going nowhere, cutting nothing.

Holding the right spot accomplishes nothing on its own. The rewrite needs a third part, and this is the piece CRISPR never carried. Bolted onto the parked Cas9 is a small enzyme, a chemical tool whose entire purpose is to perform one specific reaction on one specific kind of letter. It’s the business end. Everything else exists to ferry it to the right place and hold it there long enough to work.

The bubble

Here’s where it gets clever, because the three parts share a problem. The chemical tool can only act on a letter that’s exposed: naked, unpaired, hanging free. But in intact DNA no letter is naked. Every rung of the ladder is a pair, each letter clasped to its partner across the rung, A to T and G to C, the two strands wound shut around each other. A letter buried inside a closed rung is out of reach. So how does the tool get a naked letter? It makes one, in the very act of binding.

When the guide RNA pairs with its matching DNA strand, it shoulders in between the two rails: to grab one strand, it has to pry that strand away from its partner. Locally, over a short stretch, the double helix unzips. One strand is now held by the guide. The other strand, its partners suddenly stripped away, is left dangling loose and single. This open pocket of unwound DNA is the bubble. And the loose strand inside it is exactly what the chemical tool has been waiting for: a short run of exposed, unpaired letters, finally reachable. The machine doesn’t merely find its target. In clamping down, it pops open its own workspace.

The chemistry

What happens inside that bubble is the heart of the whole technology, and it’s gentler than the word rewrite suggests. No letter gets torn out and swapped for another. The tool performs a tiny piece of chemical surgery on the letter already sitting there. Each of the four DNA letters is, underneath the abstraction, a particular small molecule: a specific cluster of atoms with a specific shape. What separates one letter from a close chemical relative can be a single tag: a small nub of atoms jutting off the molecule. Pluck off that nub and the letter’s shape shifts just enough that the cell’s reading machinery mistakes it for a different letter entirely. That plucking is the whole reaction.

It even has a name: deamination. It means snipping away one particular tag, a little cluster built around a nitrogen atom. Take that tag off the letter C and C is converted into a molecule called uracil; uracil is neither C nor quite T, but it’s shaped enough like T that when the cell reads the strand, it reads a T. Do the same to A and A becomes a molecule called inosine, which the cell dutifully reads as G. In neither case is the ladder broken, and no foreign letter is spliced in. A C is quietly shaved down until the cell calls it a T. An A is shaved down until the cell calls it a G. The strand stays whole the entire time, and once the cell copies the new reading and locks it in, the change is permanent. Indistinguishable from a letter that had been correct all along.

The editing window

But here’s the catch, and it comes straight from the bubble. The pocket of exposed single-stranded DNA isn’t large. It lays bare only a short run of letters (in the standard editors, a stretch about five letters wide), and only the letters that happen to fall inside that run can be reached and rewritten. This narrow zone is the editing window, and it sits at a fixed spot: a few letters in from one end of the roughly twenty-letter stretch the machine grips. Everything outside it stays clasped shut, unreachable, untouched.

That same window is the source of base editing’s precision and the fence around its power. The precision is obvious: the tool isn’t loosed on the whole gene, or even the whole twenty-letter target, but on a defined handful of letters in a known position. The constraints are subtler, and there are two.

First, the letter you want to fix has to land inside the window. Which means you have to park the machine at just the right distance, and the machine can only park where a particular short signpost of letters sits in the DNA right beside the target. Plenty of letters in the genome simply have no legal parking spot within reach.

Second, the window can’t tell two identical letters apart. If your target C has another C sitting a rung or two away, inside the same window, the tool may well rewrite both: an unwanted “bystander” edit riding along with the intended one. The chemistry is very choosy about which kind of letter it changes, and completely blind to which copy.

So the same five-letter window that makes base editing surgical also boxes it in. It can change a single chosen letter with a precision no scissors can match, but only when the right letter sits, alone of its kind, in exactly the right small pocket of exposed DNA. Most of the engineering that came afterward, and most of the difficulty in aiming these tools at real diseases, amounts to a long campaign to widen that pocket, slide it somewhere new, or sharpen its aim.

The Enzyme That Didn’t Exist

Nature had four billion years to invent an enzyme that rewrites the DNA letter A directly into another letter. It never bothered.

Why would it? That conversion solves a human problem, not one any living thing ever faced. So to build the adenine editor, the lab couldn’t just pull a ready-made tool off nature’s shelf, the way it had for the cytosine editor. It had to produce a protein that had never existed anywhere on Earth. And that’s far harder than it sounds, because you can’t simply design one to order.

A protein is a long chain folded into one precise three-dimensional shape, and the shape is everything. It decides whether the molecule can grip a particular letter of DNA and perform a particular chemical nip on it.

Here’s the problem. Nobody can reliably read a chain’s sequence and predict the shape it’ll fold into, let alone run the logic backwards: name the shape and the chemical trick you want, and calculate the exact chain that delivers them. The reason is the sheer size of the search. A protein just a couple hundred building blocks long can be strung together in more ways than there are atoms in the observable universe. Searching that haystack by hand, placing atoms one at a time, is hopeless.

Breeding an enzyme

So the researchers didn’t search. They let evolution search for them.

The method is called directed evolution, and it’s a celebrated invention in its own right. The chemist Frances Arnold pioneered it in 1993 and won a share of the 2018 Nobel Prize in Chemistry “for the directed evolution of enzymes.” The idea is to run the same loop that drives natural evolution, except inside a test tube and aimed at a goal. Start with a protein that already does something close to what you want. Make millions of slightly mutated copies of its gene. Subject them to a test that only the better copies survive. Keep the winners, mutate those, run the test again. Round after round, the surviving population creeps toward the target. Not because anyone designed the improvements, but because the failures get weeded out and the lucky accidents pile up.

The elegance is in the outsourcing. Evolution is, at bottom, a search, a way of turning up rare useful arrangements in that unimaginably vast space of possible proteins, and it searches without understanding anything at all. It never calculates which mutation will help. It just tries enormous numbers of them and lets the test sort the few that work from the multitude that don’t.

The researchers never had to know, atom by atom, how their finished enzyme did its job. They only had to design a test that the right enzyme would pass, then let generation after generation of bacteria hurl mutations at the problem until one stuck. What came out the far end was a genuinely man-made protein, carrying a scatter of mutations no human designer would ever have thought to choose, doing a chemical job that nothing in the living world had ever done.

The permanence trick

Inventing the enzyme was the first clever move. Making its work last was the second, and it needed a different kind of cunning entirely.

The moment the chemistry converts one letter into another, a freshly altered letter is sitting in a strand of DNA. And a cell doesn’t treat its DNA as a notebook to scribble in. It treats it as the master archive (the original from which every other molecule in the cell is copied), and it guards that archive constantly. Repair crews patrol the strands, looking for anything irregular and putting it back the way it was. To those crews, a base editor’s handiwork looks exactly like damage. Left alone, the cell would scrub the edit out within hours. So permanence had to be engineered, against two separate defenses.

The first defense falls on the cytosine editor. The letter it creates on the way to a T is uracil, a letter that belongs in RNA and never in DNA. To a cell, uracil in DNA is an unmistakable alarm bell, and it keeps a dedicated enzyme on patrol to hunt down stray uracil and snip it straight out. That very efficiency, the thing that normally protects the genome, would rip the edit back out before it could set. The countermeasure is a tiny molecular plug, borrowed (of all places) from a virus that preys on bacteria, bolted directly onto the editor. The plug jams the uracil-removing enzyme and holds the patrol at bay just long enough for the change to become permanent.

The second defense is subtler, and outwitting it is the real sleight of hand. After an edit, one rung of the DNA ladder is mismatched: a new letter now sits across from the old partner it no longer fits. The cell runs a proofreading crew built precisely to catch mismatches like this. But the crew faces a dilemma every time. A mismatch means one of the two strands is wrong. Which one? Fix the wrong strand and you save the information. Fix the right strand and you destroy it. The cell’s rule of thumb is to trust the older strand and suspect the newer one, and it tells them apart by looking for a fresh cut. The strand carrying a recent nick is the one still being assembled, so that’s the strand to overwrite. The editor turns this rule into a trap. It deliberately nicks the unedited strand, the one still holding the original, “correct” letter. The proofreading crew reads that nick as a confession, decides the unedited strand is the unreliable copy, and dutifully rewrites it to match the edited strand. The cell’s own machinery copies the new letter across into both rungs and seals it.

The payoff of getting the letter into both rungs is permanence in the fullest sense. The edit is no longer a change balanced on a single strand, waiting to be noticed and reversed. It’s now simply the sequence, indistinguishable from a letter that had been there since birth. And every time the cell divides and copies its DNA, the correction gets copied along with everything else, inherited by every descendant cell. One quiet chemical nip, protected by a viral plug and locked in by a faked repair signal, becomes a permanent line in the archive.

What it can’t reach

For all that ingenuity, base editing can only do so much. The ceiling is built into the chemistry. The four DNA letters fall into two shape-families: two letters of one build, two of another. The nip the editor performs, shaving a small tag off a letter, can only ever convert a letter into its same-family partner: C into T, or A into G, and the reverse. Geneticists call these gentle, within-family changes transitions, and between the two editors all four of them are covered.

The other kind of change is the problem. Swapping a letter for one from the opposite family (an A into a C, or a G into a C) would mean tearing out the whole letter and dropping in a structurally different one, which the shaving trick simply can’t do. Nor can a base editor add a missing letter or delete an extra one. Those harder rearrangements are off the table. And that boundary cuts straight through some of the diseases the field most wants to cure. Take sickle-cell disease, the cleanest single-letter disorder of all. It turns out to need exactly the kind of swap base editing can’t make at the site of the original error: a cross-family change, not a within-family one.

The gap was glaring enough that Liu’s lab built a second, more versatile tool to close it. Reported in 2019 and led by a researcher named Andrew Anzalone, prime editing works less like a chemical eraser and more like the find-and-replace in a word processor. It can write a short new stretch of sequence into a chosen spot, performing all twelve possible letter swaps and inserting or deleting letters besides. In principle it reaches nearly nine in ten of the disease variants catalogued so far. The catch is that it’s a more elaborate machine, so it carries its own difficulties into the body.

But make no mistake. Even hemmed in, base editing’s reach is enormous. Among the tens of thousands of single-letter typos catalogued as causes of human disease, the within-family transitions are the clear majority. Roughly three in five. The adenine editor alone, in principle, could reverse close to half of every disease-causing point mutation on record, and the cytosine editor adds a large slice more. An enzyme that didn’t exist a decade ago, fused to a trick that turns a cell’s own repair crews into unwitting accomplices, had put the majority of humanity’s single-letter genetic typos within conceptual reach. And it did it without ever cutting the strand.

The Delivery Problem

The molecular pencil works.

Drop a base editor into a dish of cells and it will hunt down its target among three billion letters, settle onto the single wrong one, and rewrite it. Reliably. The chemistry is no longer what’s keeping these therapies out of patients.

What’s keeping them out is humbler, and a lot less glamorous. Plumbing.

A human body is not a dish. It’s a sealed, churning system of some thirty trillion cells, sorted into organs, most of them walled off behind membranes and barriers and nowhere near a needle. To fix a gene in a living person, a large and fragile molecular machine has to travel out of a syringe, through the bloodstream, past every wrong cell, into the right ones, and then deep inside them, all the way to the nucleus where the DNA is kept.

The edit is the easy part. Delivery is the unsolved engineering problem. It’s the reason the whole field has advanced as slowly, and as narrowly, as it has.

The cargo problem

The obvious way to smuggle genetic cargo into a cell is to borrow the experts.

Viruses have spent billions of years perfecting exactly one trick: slipping their own genes into the cells of a host. So tame one (strip out the genes that make it dangerous, load your own in their place) and you’ve got a delivery truck that drives itself into cells. Gene therapy has leaned on one such tamed virus for years, a small, mild one called adeno-associated virus, or AAV, that causes no known disease in people. It has already ferried approved therapies into the eye, restoring sight in a form of inherited blindness, and into the body, to halt a fatal muscle-wasting disease of infancy.

But AAV has a flaw that’s fatal for this particular job. Its cargo hold is tiny.

The virus is a small protein shell, and it can pack in only about 4,700 letters of genetic code. That’s the entire budget. And a base editor blows the budget. The targeting protein alone (the disarmed scissors that grips the DNA) runs to roughly 4,100 letters of code; bolt on the chemical enzyme that performs the rewrite, plus the guide that aims it, plus the genetic “on switch” the virus needs to make any of it run, and the full package is comfortably too big to fit inside the shell. It’s like shipping a refrigerator in an envelope.

Researchers have put real effort into shrinking the machine, hunting for more compact targeting proteins in other bacteria, or splitting the editor across two separate viruses to be reassembled inside the cell. But those are workarounds for a truck that was never built to haul a load this size.

The fat bubble

The vehicle that actually carried base editing into living people came from a completely different direction. And most of the world already has it coursing through their arms. It’s the technology behind the mRNA COVID-19 vaccines: the lipid nanoparticle, a microscopic bubble of fat.

Here’s the trick those vaccines proved at planetary scale. You take a set of temporary genetic instructions (messenger RNA, the short-lived working copy a cell normally makes from a gene and then reads to build a protein) and wrap them in a droplet of specially designed fat, small enough to slip into a cell. The droplet is built from four kinds of fatty molecule layered together, and the crucial one carries a faint, switchable electrical charge that helps the bubble fuse with a cell’s oily outer skin and spill its contents inside.

The COVID vaccines wrapped the instructions for one harmless viral protein, teaching the immune system a face to recognize. A base-editing therapy wraps a different set of instructions: the blueprint for the editor itself, plus its targeting guide. The cell takes in the bubble, reads the blueprint, and builds the editing machine on its own factory floor. A machine too bulky for any virus to deliver, assembled on-site from a recipe instead of shipped in whole.

And sending instructions instead of a permanent gene buys you a quiet safety bonus. This is the second reason to favor the fat bubble over the virus. When a virus delivers an editor, it parks a working copy of the editor’s gene inside the cell, and that gene keeps issuing orders, churning out fresh editor molecules for months, sometimes years. Every one of those molecules is another chance to drift to the wrong address and nick a healthy gene. The longer the machine stays in production, the more of these stray edits pile up.

Messenger RNA does the opposite. It’s built to be disposable. The cell reads each instruction a handful of times and then shreds it, so the editor flares into existence, does its work over a few days, and is gone. It’s a hit-and-run. The tool shows up, makes the change, and clears out before it can cause much trouble. And because the instructions are RNA, they never get filed into the cell’s DNA archive the way a virus’s genes can, so there’s no risk of the delivery vehicle itself jamming a foreign chunk into the genome. For a permanent edit, a brief, self-erasing burst of the machine that makes it is exactly what you want.

The liver’s lucky address

Which raises the question the entire field is organized around: once the bubble is in the bloodstream, where does it actually go? Left to itself, it goes to the liver. And that single fact has shaped this whole first generation of medicines.

The reason is a happy accident of the body’s own housekeeping. Part of the liver’s day job is filtering fats out of the blood, so it’s built to grab fatty particles. The moment a lipid nanoparticle hits circulation, it gets coated with a blood protein called ApoE, whose normal job is to flag fat droplets for pickup. Liver cells are studded with the receptor that recognizes ApoE (the molecular catcher’s mitt they use to haul fat in from the blood), so they reach out, grab the ApoE-coated bubbles, and pull them inside, mistaking the therapy for ordinary cargo. No steering required. Injected into a vein, the bubbles drift downstream and the liver inhales them. The organ isn’t so much targeted as it is the path of least resistance.

Which is why the opening wave of in vivo base-editing therapies reads like a catalog of liver diseases. The liver is the body’s central chemical plant, and a surprising number of the proteins that cause trouble elsewhere are manufactured right there. The protein that keeps cholesterol high in the blood is made in the liver. So is the enzyme whose absence poisoned the infant KJ. So is the rogue protein behind transthyretin amyloidosis, a nerve-and-heart disease that the leading injectable editing programs aim to silence at its liver source.

The cleanest showcase so far is a base-editing therapy for alpha-1 antitrypsin deficiency, a disorder in which a single misspelled liver protein both fails at its job in the lungs and clogs the liver that makes it. In a trial reported in 2025, one infusion corrected the misspelling across a large share of a patient’s liver cells. Within a month, more than 90 percent of the relevant protein circulating in the blood was the corrected version, and the disease-causing form had dropped by roughly four-fifths. And because the liver pumps this protein straight into the bloodstream, you could read the edit’s success right off a blood draw. Rare, hard proof that an injected base editor had reached its target organ and rewritten the gene.

Zip codes for everywhere else

But the liver’s lucky address is also a cage. Everything the technology most wants to reach next sits outside the liver: the muscles that waste away in muscular dystrophy, the neurons behind inherited brain disease, the blood-forming stem cells in the marrow that could cure sickle cell with a single shot, the light-sensing cells of the retina. A bubble that drains helplessly to the liver will never get to any of them.

Reaching those tissues means overriding the default: writing a new molecular address onto the particle so it bypasses the liver and docks somewhere else. Scientists call this the problem of tropism (where a particle naturally goes), and re-addressing the bubble is the frontier on which the next decade of this technology will be won or lost.

Two strategies are furthest along. The first is to change the bubble’s own recipe. A team led by Daniel Siegwart at UT Southwestern found that adding a fifth fatty ingredient to the standard four-part droplet (a molecule carrying a deliberate electrical charge) re-sorts where the particles end up, steering them to the lungs or the spleen instead of the liver, with the destination tunable by how much of the extra ingredient you add.

The second is to bolt a targeting tag onto the bubble’s outer surface, a molecule cut like a key for a single lock, designed to dock only on the cell type you want. The most striking early results are in the blood-forming stem cells. Bubbles studded with a tag that grips a marker found almost exclusively on those stem cells have delivered their cargo to roughly nine in ten of them in animals, editing those stem cells inside a living mouse without ever drawing a single cell out of its body.

And for the few organs sealed inside their own compartments (the eye, the brain, the fluid around the spinal cord) there’s a cruder option that sidesteps the bloodstream entirely. Inject the bubbles straight into the enclosed space, where they can’t drain to the liver because there’s nowhere else for them to go.

None of this is solved. Steering a nanoparticle to the muscle or the brain, at a dose that’s both effective and safe, is genuinely beyond reach today. And most of the diseases this technology dreams of curing live in exactly those hard-to-reach places. The chemistry of the edit (the part that drew the Nobel attention) is by comparison the finished portion of the work. What stands between a base editor and most of the patients who need one is not whether the pencil can write. It’s whether the letter ever reaches the right address.

When the Spell-Check Slips

A pill is a negotiation you can always walk away from. Get headaches, and you stop swallowing it. Within days or weeks it washes out of your blood and the effect fades. A base edit offers no such exit. It rewrites the cell’s master copy of a gene. And then every time that cell divides, the change is copied faithfully into each new cell, and into their descendants after that, for as long as the person lives. No antidote. No tapering dose. No washout. When the editor changes the letter it was aimed at, that permanence is the entire point: one infusion, a lifetime of correction. When it changes a letter it wasn’t aimed at, the same permanence becomes the nightmare: a mistake that can’t be recalled. A stray edit that nudged a healthy cell a single step toward becoming a tumor would sit in the genome for decades, quietly copied into every daughter cell, with no way to pull it back.

That asymmetry, the gift and the hazard riding on the very same property, is why the field has become almost obsessive about one question: when the spell-check reaches in to fix a single letter, what else does it touch?

The first way it goes wrong is the aim. The machine finds its target by matching a short string of about twenty letters against the destination address. But the genome runs to three billion letters, and in a stretch that long there are bound to be near-misses, spots that read almost the same as the target, off by a letter or two. Every now and then the editor settles onto one of these look-alikes and makes its change in the wrong gene entirely. That’s an off-target edit. The GPS docking at a house that looks like yours, on a street with the same name, in the wrong town.

There turned out to be a second, sneakier route to the same kind of harm, and it blindsided everyone. The chemical tool that does the rewriting doesn’t actually wait for the guide’s permission. It will shave a letter off any exposed, unpaired stretch of DNA that drifts past it, so long as the letter is the right kind. Most of the time the only exposed stretch is the little bubble the machine pries open at its target. But a living cell is forever unzipping its own DNA, to copy it and to read it, briefly baring short patches all across the genome. A loose editor molecule pounces on those, scattering edits at random, with zero regard for the address.

Two teams pinned this down in 2019 with an elegant trick. They edited just one cell of a two-celled embryo and let it grow, so the edited half of the animal had a perfect genetic twin in the unedited half. Any new typo present on one side but absent on its twin had to be the editor’s doing. The cytosine version of the tool, they found, sprinkled random changes across the genome at more than twenty times the natural mutation rate. By contrast, the adenine version stayed down near background. The flaw wasn’t in the targeting. It was an enzyme too eager to do its one job. And the fact that the adenine editor came through clean was a quiet reassurance for the in-body therapies built around it.

Even with perfect aim, the editor can overshoot. The rewrite isn’t surgical down to a single letter. It happens across a small window a few letters wide, and the chemistry, picky as it is about which kind of letter it will change, can’t tell one copy from another. So when a second letter of the same kind sits next to the target inside that window, both get changed: the intended fix plus an unwanted passenger. These bystander edits aren’t off in some distant gene. They land right next to the bullseye, inside the very gene you’re trying to repair, and every now and then they wreck it as thoroughly as the disease did.

A third kind of collateral damage skips DNA altogether. The cell rarely works straight from its master recipes. Instead it makes cheap, disposable copies of whichever genes it needs at the moment, short-lived working drafts written in RNA (DNA’s close chemical cousin), and reads those. And because the letters of RNA are nearly identical to the letters of DNA, the same shaving tool that trims one will happily trim the other. It does so lavishly. When researchers looked in 2019, a cytosine editor had quietly rewritten tens of thousands of letters across the RNA pool, touching somewhere between a third and a half of all the genes that were switched on. The adenine editor, for all its tidiness on DNA, scrambled RNA too. While the tool is active, a cell can churn out subtly wrong proteins from these corrupted drafts.

But here’s the redeeming detail: RNA is trash by design. The cell shreds each working copy within minutes to hours and prints fresh ones. So once the editor itself is gone, the RNA damage flushes out and the cell goes back to making clean copies. It’s collateral damage with an expiration date, the exact opposite of a DNA edit, which never expires.

The field has pushed back on three fronts.

The first is to build better enzymes. By breeding and tuning the chemical tool, researchers have produced versions that act over a far narrower window (in some, as little as one or two letters instead of five), which starves the bystander problem. They’ve also built versions that barely grip RNA at all, cutting the stray RNA edits by hundreds to thousands of times. And the same redesigns tend to calm the over-eager enzyme that scattered random edits through the genome in the first place.

The second front is timing, and the therapies already lean on it. The editor shows up not as a permanent fixture but as a brief, self-erasing burst of instructions. It flares up, does its work over a few days, and is gone before it can do much wandering.

The third front is simply looking, harder and deeper than anyone used to. You can’t limit damage you can’t see. So the field has built ever more sensitive ways to comb a treated genome, reading out billions of letters to flag any change that shouldn’t be there, or zeroing in on exactly the spots where the machine was caught prying the DNA open and checking each one for a stray edit. A bad edit might hide in one cell out of ten thousand. Finding it takes deliberate, exhaustive searching. And even the best search can only push the limit of detection lower. It can never quite prove the number of hidden edits is zero.

All of which sharpens to one uncomfortable question: who’s on the table?

The arithmetic of risk swings entirely on the answer. For a dying infant, or for a patient whose inherited cholesterol will trigger a heart attack in their forties, a small chance of a stray edit is an easy trade. The disease in front of them is by far the bigger threat.

But the boldest dream of in vivo editing is something else entirely. It’s to switch off PCSK9 in millions of basically healthy middle-aged people, sparing them a heart attack they might never have suffered in the first place. That’s far closer to a vaccine than to a cure. Nature even hints it should be safe: a rare few are born with a broken PCSK9 gene, carry very low cholesterol their whole lives, and stay perfectly healthy.

But that reassurance is about the intended edit. It says nothing about the accidental ones.

Vaccinating the healthy is held to a brutal standard, precisely because the person started out fine. Any permanent harm is harm the treatment manufactured out of nothing. A one-in-a-thousand chance of seeding a cancer might be a reasonable gamble for someone facing death this year, and an unthinkable one for someone facing no near-term danger at all. Same edit, same odds, opposite verdict.

And the calculus only hardens with scale. Spread an intervention across millions of well people and even a tiny per-person risk turns, with grim certainty, into a real count of people harmed.

So far, importantly, the trouble in human trials has come not from off-target edits but from the rest of the procedure. In Verve’s first cholesterol-editing study, the serious events traced back to the delivery and to the patients’ own advanced disease. One volunteer suffered a temporary crash in blood platelets and a spike in liver enzymes within days of the infusion, both of which resolved on their own. Two others had cardiac events. One was a fatal cardiac arrest five weeks later, which independent monitors judged to stem from the severe heart disease the patient already carried, not from the edit. The other was a heart attack the day after dosing, which was deemed potentially related to the treatment, in a patient who likewise carried critical underlying coronary disease. Verve rebuilt the delivery, and its second-generation version has so far stayed clean: no treatment-related serious events in its first fourteen patients, and cholesterol cut by as much as sixty-two percent.

The measurable edits, in other words, look reassuring. It’s the ones that might be slipping past the measurements (sitting silent and permanent in an otherwise healthy genome, waiting thirty years to matter) that keep the safety bar for vaccinating the well so far above the bar for rescuing the sick.

A Vaccine for Heart Attacks

Sometime in 2012, a healthy 42-year-old named Senthil came home from a run and collapsed. A little over a week later he was dead. He had high cholesterol, he’d had a heart attack, and the minutes his brain spent starved of oxygen finished what the blocked artery started.

Senthil’s younger brother happened to study the genetics of heart attacks at Harvard. His name is Sekar Kathiresan, and he had already watched this disease cut through his own family. Senthil’s death turned a question he’d been asking into an obsession: why do hearts fail like this, without warning, in people who feel perfectly fine?

This is not some rare misfortune like the one that nearly killed the infant KJ. Heart disease is the single largest cause of death on the planet: some 18 million people a year, nearly 1 in 3.

In 2018 Kathiresan staked his career on a conviction most cardiologists would have called naïve: that heart attacks can be stopped before they ever start. His company, Verve, set out to do it in the most audacious way imaginable. Not a daily pill taken for life. One injection, given once.

“I tried to turn that negative energy into Verve,” he later said, “to make sure what happened to Senthil doesn’t happen to others.”

The injection needed a target. That target was handed to Verve by a quirk of human biology that two researchers in Texas spent years chasing down.

Around the turn of the millennium, Helen Hobbs and Jonathan Cohen at UT Southwestern started combing the genes of thousands of ordinary Dallas residents, hunting for the people whose cholesterol sat at the extreme ends of the range, and asking what in their DNA put them there. One gene kept surfacing: PCSK9. About 1 in 50 of the Black participants carried a version that was simply broken. Switched off by a typo, producing no working protein at all. These people walked around with strikingly low LDL (the so-called bad cholesterol) their entire lives.

Then the researchers tracked their hearts, and the payoff was staggering. Reported in 2006: carriers of a dead PCSK9 gene suffered 88% fewer cases of coronary heart disease than everyone else. And these weren’t health fanatics. They were an ordinary cross-section who smoked and carried diabetes and high blood pressure at the usual rates.

Nature had already run the experiment. Switch off this one gene and you appear to be protected from the world’s leading killer for as long as you live, at no apparent cost.

To see why switching off PCSK9 works such magic, you have to picture how the body clears cholesterol from the blood.

The liver is the body’s cholesterol filter. It does the filtering with tiny grabbers studded across the surface of its cells. Each one reaches into the passing bloodstream, latches onto a particle of LDL, and hauls it inside to be broken down. Then, thriftily, it rides back up to the surface to grab another. A single grabber clears particle after particle, recycled over and over for hours.

PCSK9 is the saboteur of this system. It’s a protein, also made by the liver and dumped into the blood, and its whole purpose is to clamp onto those grabbers and wreck the recycling. Instead of letting a grabber return to the surface, PCSK9 drags it down into the cell’s shredder to be destroyed. Every PCSK9 molecule means one fewer grabber pulling cholesterol out of the blood.

So the logic is simple. Make a lot of PCSK9 and the filter runs short of grabbers, cholesterol backs up, and levels climb. Make little or none (as Hobbs and Cohen’s lucky carriers do) and the grabbers survive, multiply, and scrub the blood clean. The gene is a dial that controls how aggressively the body clears its own bad cholesterol. And evolution set that dial, in most of us, higher than is good for us.

Verve’s bet is to reach into an ordinary adult’s liver and turn that dial down to where the protected carriers were born. Switch off PCSK9 with a single edit, and in effect enroll a normal person into Hobbs and Cohen’s 88% club.

The contrast with what medicine already offers is the entire pitch.

Doctors can already lower cholesterol several ways, and all of them work — the daily statin pill, an antibody injected under the skin every few weeks, an RNA-based shot given twice a year. Drugmakers even sell medicines that block PCSK9 directly, which is itself proof the target is sound. But every one of these is a subscription. Stop paying and the protection lapses; within weeks the cholesterol creeps back.

And people stop constantly. Roughly half of patients quit their cholesterol drugs within a year. Not out of recklessness, but because swallowing a pill every single day, for decades, to fight a disease that gives you no symptoms until it gives you a catastrophe, is something humans are simply bad at.

A one-time edit answers that failure at its root. Nothing to remember, nothing to refill, nothing to quit.

Verve dosed its first patients in 2022 — people born with an inherited condition that drives cholesterol so high it triggers heart attacks in early adulthood — and watched their bad cholesterol fall.

The first attempt stumbled. And how Verve handled the stumble is part of what makes the story instructive.

The original therapy ran into a safety signal in early patients — a worrying, if temporary, reaction serious enough that the company shelved the program. But the trouble didn’t trace to the edit itself. It traced to the delivery: the burst of fatty particles needed to flood the liver was hard on the body. So rather than walk away, Verve re-engineered the vehicle. They fixed onto the particles a small sugar tag that docks specifically onto liver cells, so the therapy homes in on its target organ more precisely and a gentler dose can do the job. The reworked version, VERVE-102, proved far better tolerated. It’s the candidate the whole enterprise now rides on.

This is what an honest failure looks like in a field where failures are usually buried. A real safety signal, disclosed, diagnosed, and engineered around in the open — with the lead horse quietly swapped for a sounder one.

By 2025 the bet had attracted the kind of money that signals a technology has crossed from dream to plausible product.

In June, Eli Lilly — a pharma giant that has built its recent fortunes on drugs for the heart and metabolism — agreed to buy Verve outright. And the terms reveal how cautiously even believers place these wagers. Lilly would pay roughly $1B in cash up front, at $10.50 a share, plus a contingent promise of up to $3 more per share — another $300M or so — payable only if the reworked therapy reaches a late-stage human trial within 10 years. All in, up to about $1.3B. It closed that July.

The contingency is the telling part. Even Lilly, committing real money, hedged a slice of the price against the single hardest milestone still ahead — pushing a permanent edit through the long, costly, late-stage trials that prevention demands.

Hidden inside what looks like an ordinary cholesterol drug is a genuinely radical idea. It’s the one the company is named for.

Medicine, as practiced, waits for disease. The chest pain, the narrowed artery, the first heart attack — and only then does it intervene. Verve’s logic runs the other way. The Dallas carriers weren’t spared because they got a good drug late in life. They were spared because their cholesterol was low from birth, across every one of their decades, so the slow gumming-up of the arteries never got started in the first place. A single edit given to a healthy 40-year-old aims to reproduce exactly that — not to unclog a damaged vessel, but to keep it from clogging at all, years or decades before any symptom could appear.

That’s not really treating heart disease. It’s vaccinating against it. One shot, delivered to someone who feels perfectly well, to head off a threat that hasn’t yet arrived.

But vaccinating the healthy changes who carries the risk. And three hard questions stand between Verve’s idea and your family doctor’s office.

The first is durability. The edit is permanent in any cell that receives it, faithfully copied into all of that cell’s descendants. But a person is not one cell, and the liver slowly renews itself over the years. Whether a single childhood-style shot truly holds for a 40-year lifetime, or fades as edited cells give way to unedited ones, is something nobody can know yet. The longest human follow-up so far runs to roughly 2 years, and the cholesterol reductions have held. Encouraging — and nowhere close to a lifetime.

The second is long-term safety in people who started out healthy. This pushes the bar very high. Any harm from the edit is harm done to someone who wasn’t sick to begin with.

The third is whether regulators will ever bless a permanent, un-take-back-able change to the body as a tool of mere prevention.

The cautious path — the one Verve has walked — is to start with the sickest, those whose inherited cholesterol almost guarantees an early heart attack, where the math of risk and benefit tilts sharply in the edit’s favor. Proving the same trade makes sense for a basically healthy person in their 40s is a different and far longer road. Heart attacks prevented are events that take years to not happen. So the trials have to be vast and slow, tracking thousands of people for the better part of a decade just to show the shot earned its keep.

None of this is settled. What Verve has established is narrower, and still remarkable: that an editor can be sent into a living human being, find the liver, switch off a chosen gene, and drive down the number that predicts heart attacks — all from a single injection.

Whether that becomes a true vaccine against the world’s deadliest disease, or stays a powerful treatment for the unlucky few born most at risk, is the question the next decade will answer.

Beam and a Baby Named KJ

A scientific breakthrough is not a medicine.

The molecular pencil that rewrites a single letter of DNA was invented in a university lab. But a university lab doesn’t manufacture drugs to clinical standard, doesn’t push them through years of trials, doesn’t carry them to a regulator’s desk. That’s what companies do.

And one company has staked its entire existence on base editing: Beam Therapeutics, built by the very people who invented the pencil, for the sole purpose of turning their tool into medicine.

Most drug companies hedge. They spread their bets across many kinds of treatment so that no single failure can sink them. Beam did the opposite. It’s what the industry calls a pure-play — a company whose fortune rides on one idea, the molecular pencil, working inside human beings.

If base editing fails, Beam fails. There is no second platform to fall back on.

A bet that big only survives if other people back it with their wallets. And the loudest vote of confidence came from one of the largest drugmakers on Earth.

In January 2022, Pfizer agreed to pay Beam $300 million in cash up front. Not for a finished product. Merely for the right to work together on editing three still-secret genes behind rare diseases of the liver, muscle, and brain. If those programs panned out, the partnership could be worth up to $1.35 billion in all, plus a cut of any future sales.

Pfizer was buying a promise: that a tool which had barely touched a patient could be aimed, organ by organ, at one disease after another. And when the four-year collaboration ran its course at the end of 2025, Pfizer made good on the bet — it took a worldwide license to one of the liver programs and committed to carrying it the rest of the way.

Money handed over on that scale, long before the science has proven itself in people, is what turns a laboratory trick into an industry.

Beam has used that backing to build something no single lab ever could: a war chest. By early 2026 it was sitting on roughly $1.2 billion in cash — enough to fund itself to the end of the decade and to push several therapies toward patients at once.

Because it’s not one bet. It’s several.

One program corrects the blood cells of people with sickle-cell disease — drawn out, edited, given back — and is closing in on the point where Beam can formally ask regulators to approve it. Another goes after an inherited disorder in which the liver can’t process a common building block of protein. And the flagship — the one meant to prove the pencil can work inside a living person, not just in a dish — targets the lung-and-liver disease alpha-1 antitrypsin deficiency.

That flagship has matured from a single dramatic result into something far more convincing: a track record.

When Beam first reported, in 2025, that one infusion had rewritten the faulty gene across a patient’s liver and flooded the blood with corrected protein, it was a proof of principle — among the first times anyone had corrected a disease-causing mutation in a living human being. Not switched the gene off. Actually repaired it.

The danger with any “first” is that it turns out to be a fluke, or fades within months. By early 2026 it had done neither. Beam had treated close to thirty patients and followed some of them for a year and a half. The correction held. The toxic form of the protein fell by more than 80%. And every patient given the dose the company chose to advance climbed above the level doctors consider protective.

On that strength Beam locked in a dose and started assembling a pivotal trial — the large, final study on which formal approval hangs.

A result was hardening into a drug.

All of this is base editing as the pharmaceutical industry has always understood a drug. One product. Built to a fixed recipe. Sold to every patient who carries the same disease. Make it once, make it identically, make it by the thousand.

Baby KJ’s medicine broke that mold completely.

A drug with a market of one

KJ’s medicine had a market of exactly one human being. No one else on Earth carries his precise pair of misspellings, so no one else could ever take his drug. It even bore a name built from his own: kayjayguran abengcemeran, or “k-abe” for short, the second half lifted straight from the base editor at its heart.

Here was a treatment designed, manufactured, and infused for a single patient. No production line behind it. No market in front of it.

That’s not a smaller version of an ordinary drug. It’s a different kind of object. And it forces a strange question, one the field is only beginning to confront: if a treatment is built for one person and will never be given to a second, is it still a “drug” in any sense the word has ever carried?

Conjuring such a thing in months took a relay of helpers no single company could supply — most of them working for nothing.

The fatty bubble to ferry the editor into the liver came from a firm in Canada that had spent years perfecting exactly that kind of bubble. The genetic instructions for the editor itself were manufactured at a facility in North Dakota. The custom guide — the short address label that steers the editor to KJ’s exact misspelling and nowhere else in three billion letters — was made by a company in Iowa. The vials converged on Philadelphia from three places, each component built for one baby, each donated free of charge.

A therapy like this normally takes a year and a half to produce. The team did it in roughly six months. And the cost, stripped of the usual overhead and run on emergency speed, came in far below what a conventional drug runs.

Swap one component

The deeper lesson of KJ is hidden in how that machine is assembled. It’s built like a tool with interchangeable parts.

Notice that two of its three pieces barely change from one patient to the next. The fatty delivery bubble is the same for everyone — every liver-bound editor has to reach the same organ. The bulk of the editing machine, the instructions for the pencil itself, is reusable too. Only one small piece is truly custom: the guide, the roughly 20-letter address that dictates where in the genome to go.

Swap that single component, and in principle the very machine that fixed KJ’s gene could be retargeted at an entirely different patient’s mutation.

The medicine stops being a fixed product. It becomes a platform — a chassis you build once, into which each new patient slots one bespoke part.

This inverts the oldest logic of medicine. Ordinarily a patient is matched to a drug that already exists, and if none fits, there’s nothing to offer. Here the drug is generated to fit the patient. The disease names the target, and the target is simply typed into the one swappable piece.

KJ’s own doctors, Rebecca Ahrens-Nicklas and Kiran Musunuru, are now trying to turn that idea into a working system. Instead of seeking approval one patient at a time — impossibly slow, one trial per child — they’ve proposed a single sprawling study that would accept patients with any of seven related disorders, the whole family of conditions in which the liver’s ammonia-clearing assembly line is broken, as long as a given patient’s typo is one a base editor can fix.

Treat a handful of them, show the method works, and the aim is for regulators to bless not each individual medicine but the platform itself. Approve the machine, and let the swapped guide ride along inside it.

A regulator reshaping itself

Regulators were never built for this. The entire apparatus of drug approval assumes a medicine given to thousands, proven in large trials, run twice over for confidence. You can’t run a thousand-person trial for a drug with one patient.

In February 2026 the FDA moved to close that gap. It proposed a draft framework it called the “plausible mechanism” pathway. For a tailor-made therapy aimed at an ultra-rare disease, where a conventional trial is simply impossible, approval could rest instead on a different kind of evidence: that the treatment precisely targets the known genetic cause of the disease, that the underlying biology is well understood, and that the therapy demonstrably corrects the broken pathway.

This is the rare sight of a regulator reshaping itself around a medicine that is really a method. The agency’s own leaders pointed to cases like KJ’s as the reason the old two-trial habit “no longer makes sense.”

And yet KJ’s case ran into a wall that no breakthrough has climbed. His therapy was affordable only because a chain of companies donated their labor and a dying infant’s clock collapsed the timeline. Neither of those is a business model.

Mass-produced medicine got cheap precisely because it’s mass-produced. One recipe, millions of doses, the cost spread across all of them. A medicine hand-built for a single patient is the exact opposite — and hand-built things are slow and ruinously expensive.

The hope is that platformization closes the gap. Once the chassis and the manufacturing steps are standardized, and only the cheap, swappable guide changes from patient to patient, the cost of each new therapy might fall within reach.

But that’s a hope, not yet a fact.

“We need to be able to scale this,” Ahrens-Nicklas has said. “The technology is there, we just need the infrastructure to evaluate it and deliver it.” Musunuru is blunter about the stakes: reach that point, he warns, or “this whole thing falls apart.”

KJ proved that a drug can be conjured for one child in half a year and rewrite his fate. Whether the same machine can be retargeted, patient after patient, at a price the world can actually pay — that’s the question on which the entire dream now turns.

The Next Five Years

Every base edit that has ever reached a living human being has landed in the same organ. The cholesterol programs, the rare-metabolic rescues, the gene rebuilt inside a dying infant — all of them ride on the same piece of luck. A tiny bubble of fat, injected into a vein, drifts downstream and gets swallowed by the liver.

The next five years come down to two questions. Can the edit get out of the liver and reach the rest of the body? And among the companies racing to get there, who puts a finished medicine in a patient’s hands first?

The early answer to the second one comes with a twist: the first in-body gene-editing therapy likely to win approval isn’t a base editor at all. It’s a cutting one.

The liver is the easy part

Inside the liver, the technology is on solid ground. The next few years are mostly a matter of finishing trials that have already shown their hand.

Take Lilly’s cholesterol editor, the single shot that switches off the liver gene keeping bad cholesterol high. It has come through its early human studies cleanly, and a mid-stage trial is set to begin by the end of 2026. But the final proof — that it can spare an otherwise-healthy person a heart attack — lies years beyond that, in a vast, slow study that has to sit around and wait for heart attacks not to happen.

Closer to the finish line is the base editor for alpha-1 antitrypsin deficiency, a disorder of the lung and liver. It’s now being built into the large pivotal trial on which approval depends, and it may well become the first base editor ever approved for use inside the body.

Then there are the rarest metabolic conditions, the family of urea-cycle disorders that nearly killed the infant KJ. These are being knit into a single platform, one meant to be retargeted from one child’s misspelling to the next.

Notice the pattern. The liver wave is the derisked wave: one organ, one delivery vehicle, and each new disease mostly a matter of swapping the address.

Beyond the liver

The harder prize is everywhere else, in the tissues a vein-injected bubble simply can’t reach.

Sickle-cell disease is the sharpest example. We can already cure it: draw a patient’s blood-making stem cells out of the body, edit them in a dish, put them back. But that route demands chemotherapy to wipe out the marrow first and a hospital stay measured in weeks. The dream is to skip all of it — one injection that finds those stem cells deep in the bone and edits them right where they sit. In mice, a fatty bubble wearing a molecular tag that grips a marker sitting almost exclusively on blood-making stem cells has delivered its cargo to roughly nine in ten of them. In humans, it has never been done.

Muscle and the nervous system are harder still. The first attempt to edit muscle inside the body — a one-time deletion aimed at Duchenne muscular dystrophy, the muscle-wasting disease — cleared regulators to begin human testing only in early 2026. And again, it works by cutting, not by base editing.

The brain is further off yet. Base editors injected directly into the skulls of mice have extended the lives of animals dying of a fatal brain disease by half. But nothing of the kind has been tried in a person.

The rivals

Two companies are racing alongside the base-editing pioneers, and each is wielding a different tool.

The first uses ordinary CRISPR — the cutting kind — inside the body. Not to rewrite a gene, but simply to break it and switch it off for good. Intellia Therapeutics has aimed this at two diseases, both driven by a harmful protein that the liver pours into the blood.

The first is hereditary angioedema, a disorder of sudden, dangerous swelling attacks. Intellia’s one-time injection disables the gene behind the attacks, and in the spring of 2026 it became the first in-body gene-editing therapy of any kind to succeed in a final-stage trial. That puts it on course to reach the market — quite possibly as the first in-body editing medicine ever approved. A cutting one, not a pencil.

The second program went after transthyretin amyloidosis, a disease of the nerves and heart. It ran headlong into the hazard that shadows the entire field. In the autumn of 2025, a patient given the therapy suffered severe liver injury and, weeks later, died. Regulators paused the trial. The case was tangled — the immediate cause was an infection from a perforated ulcer — but the liver damage tied to the treatment was real, and it froze the program for months.

The second rival carries the more powerful tool. Prime editing can do what base editing can’t. Not just the gentle within-family letter swaps, but the harder cross-family swaps, and the outright insertion or deletion of letters. In principle that brings the large majority of known disease mutations within reach. It’s the natural successor for everything a base editor has to leave alone — including, awkwardly, the very change that would fix sickle-cell disease at its root.

But versatility has a price, and the price is bulk. The prime-editing machine is a more elaborate contraption than a base editor: a larger protein lashed to a longer, fussier guide that has to carry both the destination address and the replacement text. And the bigger the machine, the harder it is to fold into a delivery vehicle and smuggle into a cell. That extra heft is exactly why prime editing trails base editing into the body.

The company built on it, Prime Medicine, has only just begun human trials. Its first patients have shown that the tool can work in a person at all. And true to form, it has pointed its first in-body programs at the one easy address — the liver — going after a copper-overload disorder called Wilson disease, and the same lung-and-liver disease several of its competitors are chasing.

What could still go wrong

None of this guarantees broad impact. Three hazards could stall it.

Safety. A base edit is permanent. It gets copied into every descendant of every cell it touches, so a mistake can never be recalled. And the only way to know a one-time edit is safe across a lifetime is to watch patients for a lifetime — follow-up that doesn’t yet exist. The death in the cutting-therapy trial was a blunt reminder that an in-body editor can do grave harm. A comparable event in a base-editing program — above all one given to people who weren’t even sick yet — could freeze the field for years.

Durability. The liver renews itself continually. Whether a single edit holds for the forty years a true cure implies, or slowly washes out as unedited cells crowd back in, is something no one can answer yet. The longest human follow-up we have runs only a few years.

Money. The first CRISPR cure to reach the market, an edit-and-return therapy for sickle-cell disease called Casgevy, lists at about $2.2 million. More than two years after approval, fewer than a hundred patients worldwide had received it — partly because the process is so punishing that clinics struggle even to collect enough of a patient’s own cells. An injection given inside the body sidesteps that particular bottleneck, since there are no cells to harvest. But it does nothing for the deeper problem. The most personalized edits of all — a medicine built for one patient’s unique misspelling — can’t be mass-produced even in principle. That’s precisely what makes them so ruinously expensive.

So the near-term picture splits cleanly in two.

The liver-bound therapies are nearly ordinary medicine now. Several will likely be approved within the five-year window, and the first in-body editing medicine of any kind may arrive sooner still — though it will break a gene rather than rewrite it. Everything past the liver — blood, muscle, brain — still waits on a delivery breakthrough that has worked in mice and in no one else.

The pencil can already write a single correct letter into a human gene and change a life. Whether it can be carried to every address that needs it — durably, safely, and at a price the world can actually pay — is the work of the next five years. And very likely the decade after.

Medicine Becomes Programmable

Forget the cure for a second. The thing that strikes me isn’t any single patient — it’s a pattern that matters far more than any one cure.

The machine that fixed the infant KJ’s liver and the machine being tested against high cholesterol are, underneath, the exact same machine. Same delivery bubble. Same editing apparatus — the disarmed targeting protein, and the chemical tool bolted onto it. The only piece that actually changes from one disease to the next is the guide, the short string of letters that tells the editor where to land.

Everything else is fixed hardware.

The guide is software.

That’s the real revolution inside base editing, and it has almost nothing to do with any particular cure.

For all of medical history a drug has been one specific molecule built to do one specific thing — a shape that fits one lock, a chemical that blocks one reaction. Discovering each one took years and a mountain of failures. A reprogrammable editor breaks that mold. The same physical machine, loaded with a different twenty-letter guide, can in principle be aimed at a completely different gene. You don’t invent a new drug for each disease. You retype the address.

Medicine starts to behave less like manufacturing and more like writing software. One platform, reprogrammed at will.

The scope here is staggering. Roughly 7,000 diseases are classified as rare. The great majority are genetic, and around 95% have no approved treatment whatsoever — most are simply too uncommon for any company to justify the cost of developing a conventional drug. A huge share of them trace back to exactly the kind of single-letter misspelling a base editor is built to fix.

Under the old model, each of these was a separate, near-hopeless project. Under the new one, they collapse into a long list of addresses waiting on the same machine, each needing only its own guide.

Managing vs. Mending

It also changes what medicine is even trying to do.

Most of what doctors prescribe manages symptoms. It doesn’t fix causes. A diabetic injects insulin every day because the body no longer makes enough, but the injection does nothing to repair whatever broke in the first place. Blood-pressure pills, inhalers, the daily cholesterol tablet — most chronic-disease care is a lifelong holding action against a problem that never goes away.

A genetic correction is a different category of thing altogether. It reaches the root cause — the misspelled instruction itself — and rewrites it. Done once, in principle done for good. You stop bailing the boat and patch the hole. A lifetime of management collapses into a single afternoon’s repair.

Take that logic to its extreme and a drug’s market shrinks to a single human being. KJ’s medicine was built for one patient and will never be given to a second. This is the purest form of what the field calls an n-of-1 therapy — a treatment with a trial population of one.

For the millions of people scattered across thousands of ultra-rare conditions, each individually too rare to interest any drugmaker, n-of-1 is the only plausible road to a cure. And the reprogrammable platform is the only reason it’s even conceivable. If the machine is fixed and only the guide changes, building a therapy for one patient becomes a matter of reading their misspelling and typing it in — not inventing a medicine from scratch.

The Price of Bespoke

Conceivable is not the same as affordable.

Hand-built things are expensive, and a medicine made for one person has no economies of scale to soften the blow. There’s no production run to spread the cost across. And even mass-produced gene therapies already sit at the far edge of what medicine charges today. Lenmeldy, a one-time treatment for a fatal childhood nerve disease, lists at $4.25 million — the most expensive drug in the world. A gene therapy for hemophilia runs about $3.5 million. And those are made by the hundred.

A therapy built for a single patient has, by definition, a market of one to absorb its entire cost.

Here the technology slams into a wall that has nothing to do with biology. A cure that exists but costs millions isn’t a cure most of the world can reach.

The injustice is sharpest for diseases that aren’t rare at all. Sickle-cell disease afflicts millions, the overwhelming majority in sub-Saharan Africa and other places where a multimillion-dollar therapy is pure fantasy. A treatment priced like a private jet, aimed at a disease of the poor, doesn’t close a health gap. It widens one.

For programmable medicine to matter beyond a fortunate few, the cost curve has to bend, and bend hard. The platform logic is the best reason to think it can: standardize the shared hardware, automate the manufacturing, and let the only custom part be the cheap, swappable guide. Whether that actually happens is the entire difference between a technology that reshapes global health and one that stays a boutique luxury for the well-insured.

The Lines Not Yet Drawn

Then there’s the unease, which grows in step with the power.

A base edit can’t be undone. It’s written into the genome and copied into every descendant of every cell it touches, for the rest of a life. That permanence is a gift when the edit is right and a liability when it’s wrong. And the more ambitious the use — editing people who aren’t sick yet, to head off a disease that might never have come — the higher the bar climbs. Every harm is now a harm done to someone who was perfectly fine.

There’s a deeper line running beneath even that one, and so far the world has held it.

Every therapy I’ve described edits the body’s ordinary cells — somatic cells, the kind that die with the patient. The correction helps that one person and goes no further. It’s never passed to their children. The line society has actually drawn sits at the germline: editing an embryo, egg, or sperm, so the change flows into every future generation.

In 2018 a Chinese scientist named He Jiankui crossed it. He used CRISPR to alter the embryos of twin girls, disabling a gene called CCR5 to try to make them resistant to HIV. The response was near-universal condemnation. He was stripped of his post and thrown in prison.

Somatic correction of disease has since won broad acceptance, and now underpins hundreds of clinical trials. Heritable editing remains banned across most of the world.

And past even the germline sits the line nobody has had to face yet: enhancement. The same pencil that repairs a broken gene could, in principle, “improve” a working one. There the technology stops being medicine and becomes something society has barely begun to argue about. The machinery itself draws no distinction between fixing a typo and rewriting a perfectly good sentence to taste.

The restraint, for now, lives entirely in the hands holding the pencil.

Scissors vs. Pencil

Remember how this field began. The molecular scissors that slice the double helix won the Nobel Prize and captured the public imagination. The cut is still what most people picture when they hear gene editing.

But the cut is the blunt instrument. It breaks, it disables, it scrambles. The edit that refuses to cut is the subtle one. It settles onto one wrong letter and quietly changes it. And it may reshape medicine far more profoundly than the famous cut ever will.

Because what it offers isn’t a single cure. It’s a method. A machine built once and reprogrammed forever, a way to treat the genome the way a programmer treats code: find the bug, fix the line, ship the patch.

A baby in Pennsylvania has already had his own broken line rewritten — in half a year, from scratch. And the infrastructure to do the same for the next patient, and the next thousand, is being assembled right now.

The pencil works. The remaining work — bending the cost curve, drawing the lines, reaching the organs past the liver — is the work of building, not of proving the thing is possible.

That part is already done.