In January 2020, a research team in Cambridge, Massachusetts, downloaded a viral sequence that had been posted online from China, picked the slice that mattered, and within days had finalized the design of a vaccine. There was no virus to grow, no chicken eggs, no decade of trial and error against a moving target. The hard intellectual work was a file. By the time that vaccine — and the one designed in parallel by Pfizer and BioNTech — reached the world a year later, the bottleneck had moved somewhere no one outside the field expected: a freezer.

The shots had to travel at roughly seventy degrees below zero — the cold chain the Comirnaty arc walked in full (CS-COMIRNATY-05). Not the cold of a kitchen freezer, which sits a little below freezing, but ultra-cold, packed in dry ice, in specialized thermal shippers that most pharmacies and many hospitals did not own. A medicine that could be invented in a weekend could not, at first, be reliably delivered to a rural clinic two counties away.

That inversion is the whole story of how mRNA is made and moved, and it is where this column ends. The earlier stages followed an mRNA program from an idea through its biology, its trials, and its approval. The final stage is the unglamorous one — chemistry, manufacturing, and controls, plus everything that happens after the drug is on the market (SP-12). For mRNA, that stage carries a real paradox: the modality flipped the usual economics of making a drug, and in doing so it created a brand-new problem at the loading dock.

For Once, the Active Ingredient Is the Easy Part

Read more

The number that made Comirnaty an approved vaccine was not a survival curve or a lab assay. It was 170. One hundred seventy people, out of 43,448 who had rolled up a sleeve, got sick enough with COVID-19 to be counted as confirmed symptomatic cases. Of those 170, eight had received the vaccine. The other 162 had received salt water. From that lopsided split — eight against one hundred sixty-two — came the figure that circled the world: 95.0% effective.

That is a strange way to prove a drug works. A drug for a disease usually proves itself by making sick people better, and you measure the improvement directly in the people you treated. A vaccine cannot work that way, because the people it is meant to help are not sick yet, and most of them never will be in the window you are watching. So the whole logic of the trial inverts. You take a large, mostly healthy population, split it in two, give one half the real thing and the other half a placebo, send everyone back to ordinary life, and wait. You are not measuring whether anyone gets better. You are counting who gets the disease at all — and comparing the two tallies.

That inversion has three parts worth separating: how an mRNA vaccine proves it works at the scale of a whole population, why the proof is built out of counted illnesses rather than calendar time, and what the famous percentage actually means once you look at it closely.

A Prevention Trial, Not a Treatment Trial

Read more

A Phase 1 trial usually asks the body a blunt question: how much of this can you take before it hurts you? For an mRNA vaccine, that question barely makes sense. The two COVID vaccines that worked best settled on doses more than threefold apart — Comirnaty at 30 micrograms, Spikevax at 100 — for what was, at the protein level, essentially the same target. Neither dose was the largest the body could tolerate. Each was the dose at which the body responded best without paying too high a price in fever and sore arms.

That gap is not sloppiness, and it is not a rounding difference. It is a clue that mRNA dose-finding obeys a different logic from almost everything that came before it. With a pill, the molecule that does the work is the molecule you swallow, and you can measure it in the blood. With an mRNA product, the thing you inject is an instruction. The drug that actually acts — the protein — is built inside the patient’s own cells, in amounts that run through each person’s biology. So the early trials cannot simply find the maximum survivable dose and stop. They have to read the immune system’s answer.

Three doses — 30, 100, and 12 micrograms — show why that answer is never obvious in advance.

The Small-Molecule Baseline

Read more

In August 2022, a COVID vaccine that already had a marketing license in hand was re-pointed at a brand-new pair of viral variants and shipped — and it did not require a brand-new approval to do it. The original product had been licensed. The updated one, encoding the BA.4 and BA.5 spike proteins alongside the ancestral strain, went out as a supplement to that existing license: a paperwork amendment, not a fresh application built from the ground up. The same thing happened again with the 2023–24 update to the XBB.1.5 lineage, and again later for the JN.1 and KP.2 lineages. Each time, the regulator picked the target — much as it does every year for the seasonal flu vaccine — and the manufacturer changed what the product encoded, then filed an update against a license it already held.

For almost any other kind of medicine this would be strange to the point of being illegal. A new drug is a new drug; you do not get to swap out the active ingredient and keep the old approval. The reason mRNA gets to behave this way is not a loophole and not regulatory generosity. It is a consequence of what the modality is — visible in what an IND normally contains, in what changes when the medicine is made of information, and in where that leverage finally runs out.

What an IND Actually Contains

Read more

A small-molecule drug spends its preclinical life being interrogated about the trouble it might cause on the way out. Does it block the hERG channel — a particular potassium gate in heart-muscle cells whose accidental inhibition can lengthen the heartbeat’s electrical cycle and, rarely, trigger a dangerous arrhythmia? Does it inhibit or induce the CYP enzymes — the liver’s cytochrome P450 family, the workhorses that chemically dismantle most oral drugs — and thereby collide with whatever else the patient is taking, raising or crashing those other drugs’ levels? Does the body, in chewing the molecule apart, produce a reactive metabolite — a chemically aggressive fragment that latches onto proteins and damages tissue? These are the standard hazards a Stage-6 program screens for (SP-06). Each one follows from the same starting condition: the drug is a foreign chemical the body must recognize, metabolize, and clear.

An mRNA drug mostly fails that premise. Its active ingredient is a strand of messenger RNA — the same kind of molecule a cell makes by the millions every second — and once it has been read and translated, it degrades into ordinary nucleotides, the body’s own building blocks. Its delivery vehicle, the lipid nanoparticle (LNP) that carries the fragile RNA into cells, breaks down into lipids. There is no persistent foreign small molecule to clear, so the hERG question, the CYP question, the reactive-metabolite question largely fall away. The gauntlet does not get harder. It inverts. The old hazards recede, and two concerns that barely registered for a small molecule move to the center: immunogenicity — whether and how the immune system reacts to the drug — and biodistribution — where in the body the drug actually goes.

One Property, Opposite Signs

Read more

A cell takes in a particle of messenger RNA, wraps it in a bubble of membrane, and pulls it inside. So far the system is working as designed. Then comes the part nobody advertises: of all the mRNA the cell has just swallowed, only a few percent will ever reach the place it needs to go. The rest is sorted, packaged, and quietly shipped back out, or routed for destruction, never having done anything at all. Roughly ninety-seven to ninety-nine of every hundred molecules of cargo are wasted.

This is not a manufacturing defect. It is a central, unsolved inefficiency of the entire mRNA modality, and the single problem that more drug-development money and chemistry have gone into than any other part of these medicines. The inversion is stark. The thing an mRNA drug is for — the protein it tells your cells to make — turns out to be the cheap part, fixed early and easily copied. The hard part, the part worth a fortune, the part that is genuinely difficult to make and difficult to imitate, is the lipid particle that carries the message and, most of the time, fails to deliver it.

The earlier cells in this column built the foundation. The first established that an mRNA drug is, at bottom, information — a sequence that tells a cell which protein to build — and that the modality had two great bottlenecks: getting the cell to read the message without sounding its innate-immune alarm, and getting the message inside in the first place (M-MRNA-01). A later cell established that the encoded sequence, the protein you want made, is fixed all the way back at hit discovery, and is the near-free, near-commodity input to everything downstream (M-MRNA-04). This cell picks up the obvious next question. If the sequence is settled and nearly free, then where does the engineering value actually go? The answer is everything around the sequence — and that is what lead optimization means for an mRNA drug.

Two Arenas, Neither of Them the Protein

Read more

In January 2020, a Chinese laboratory posted the genome of a new coronavirus to a public database. Within days — not months, days — researchers at the U.S. National Institutes of Health had finished designing the molecule that would become the active ingredient in the first authorized COVID-19 vaccines. They did not screen anything. They did not grow cells, fish through a library, or run an assay. They opened a file, made a small set of changes to a protein sequence already worked out for related viruses, and that was the design. The hard part, by the time the genome arrived, had been done years earlier and for a different disease entirely.

That is a reversal of the usual order. In most of drug discovery, the moment a target is chosen is the moment the real searching begins. The first molecule that does something useful — the hit — has to be found, and finding it is typically slow and expensive. Here the molecule existed, in a sense, before anyone had ever seen the virus. It is possible because the expensive part of hit discovery had already happened years earlier — which does not make the problem easy, only relocated. What hit discovery normally costs shows where it went.

What Finding a Hit Normally Means

Read more

On January 11, 2020, Chinese researchers posted the genetic sequence of a new coronavirus to an open database. Eleven months later, on December 11, 2020, the US authorized a vaccine built from that sequence for emergency use. Between those two dates a finished medicine was designed, tested in tens of thousands of people, and shipped. The fastest any vaccine had ever moved before — for mumps, in the 1960s — was about four years.

The natural reaction to “eleven months” is suspicion. It sounds like corners were cut. They were not. Most of the work that makes a vaccine slow had already been done, quietly, over the preceding twenty-five years — and almost none of it was about coronaviruses. It was about a single, stubborn problem: how to get a cell to read a message you hand it, without the body destroying the message first and inflaming itself in the process.

That problem, and the way it was finally solved, is what defines an entire class of medicine, and the rest follows from it: what kind of drug an mRNA drug actually is, why it took a quarter of a century to become possible, and why, once it was possible, it could move so fast. The Comirnaty vaccine most readers know is one worked example, not the story (CS-COMIRNATY-01). The story is the platform underneath it.

What the Drug Actually Is

Read more

The boxes were the size of a carry-on suitcase, and they were packed with dry ice. Inside each one sat trays of small glass vials, and clipped to each shipment was a sensor with a GPS chip and a thermometer, reporting back to a control tower that watched, around the clock, for the one thing that could ruin everything: the temperature ticking up. Pfizer designed these thermal shippers to do a single stubborn job — hold their contents at roughly seventy degrees below zero Celsius, all the way from a factory in Michigan or Belgium to a freezer in a hospital basement, and to keep doing it for up to ten days if they had to. A box held up to five trays. Each tray held 195 vials. Each vial held five doses. And every one of those doses had to be kept far colder than any household freezer can reach — down in the range you would otherwise find only in the dead of an Antarctic winter.

This is the part of the story that almost no one was prepared for.

Read more

For most of 2021, the same shot in the same vial was being described, simultaneously and often in the same breath, as two opposite things. To one person it was not really approved — authorized, sure, but not the way a normal medicine is. To another it was fully vetted, the most scrutinized vaccine in history. What is strange, and worth sitting with, is that both descriptions were accurate. They were not opinions arguing past each other. They were two true statements about the same object, separated by a piece of regulatory vocabulary that almost no one outside the agency had ever needed to know.

The vaccine had been going into American arms since December of 2020. By the time the FDA granted it full approval on August 23, 2021, the country had already administered nearly two hundred million doses of it. That sequence sounds backwards if you assume approval comes first and use comes after. It usually does. But in a declared emergency, there is a second door — a faster one, with its own conditions for opening — and for the eight months between December and August, the emergency door was the only one that had been opened.

Earlier installments in this arc followed the molecule itself: the engineered strip of messenger RNA wrapped in fat (CS-COMIRNATY-02), and the roughly eleven months in which it went from a viral genome posted online to an emergency-authorized vaccine going into millions of arms (CS-COMIRNATY-01). This installment is about something less visible than the molecule and, as it turned out, harder to manage than the chemistry. It is about the two legal documents that authorized the same shot, what each one demanded, and why the gap between them became one of the most consequential communication failures in modern public health.

There is a thread running through this whole curriculum about a single idea: that a drug, legally speaking, is its authorizing document (SP-11). The molecule in the vial is not the medicine. The medicine is the molecule plus the official paper that says what it is, who may receive it, on what evidence, and with what claims printed on the label. Most drugs have exactly one such document. This vaccine had two, in sequence — and the difference between them is the entire subject of what follows.

Two Doors Into the Same Building

Read more

By the second week of November 2020, the candidate from the last installment (CS-COMIRNATY-02) was no longer a question of design. It was a question of proof. The full-length spike, locked in its prefusion shape by the two proline substitutions; the messenger RNA rewritten with a modified nucleoside so the body’s alarm systems would let it through; the whole thing wrapped in a lipid nanoparticle and dosed at thirty micrograms — all of that was settled. It worked in cells. It worked in mice and monkeys. It raised the right antibodies in the first few hundred people who received it. None of that was the same as knowing whether it stopped anyone from getting sick.

There is only one way to find that out, and it is not clever. You give the real thing to a very large number of people, give a saltwater placebo to an equally large number, and then you wait — not for a result you engineer, but for the disease itself to come and find them. You count who gets sick in each group. The vaccine cannot summon that data. Only the pandemic can.

So consider the scale of what had to be assembled. Across the United States, Argentina, Brazil, Germany, South Africa, and Turkey, 43,448 people had rolled up a sleeve and received their injections — a two-dose course, the shots twenty-one days apart — with no way of knowing which arm they were in.

Half were assigned the vaccine. Half got salt water. Every one of them then went back to daily life in a world where conventional clinical trials had stalled — and agreed to report it, swab by swab and symptom by symptom, every time they felt unwell. This was the pivotal trial. In the language of the pipeline, this is Stage 10: the point at which a drug stops being promising and becomes proven, or fails to. It is the stage where the curriculum elsewhere calls a Phase 3 trial “the bet” (SP-10) — the single most expensive, most consequential experiment a developer runs, the one that converts a beautiful hypothesis into a number a regulator can act on.

Here the bet was unusual. In this curriculum, you will meet Gleevec’s tiny Phase 1 — whose landmark result came from just fifty-four adequately dosed patients, so stark it announced itself in a few weeks — and Keytruda’s sprawling, shape-shifting early trial that kept adding new, pre-planned cohorts as it went. Those were studies that surprised people with biology. This one’s surprise was logistical. The biology, by November 2020, was almost anticlimactic. The hard part was building a clean, enormous, randomized experiment fast enough to matter while the thing it was measuring was killing thousands of people a day.

And then, in mid-November, the envelope was opened.

Read more

In the autumn of 2013, Barney Graham and Jason McLellan published a paper about a virus that had nothing to do with a pandemic, because there was not yet a pandemic to have anything to do with. The virus was respiratory syncytial virus — RSV — a common cause of severe infection in infants, and a vaccine target that had defeated everyone who tried it for half a century. Their paper, in Science, was about a protein on the surface of that virus, and about a shape.

RSV, like many viruses, gets into a cell using a surface protein that fuses the virus’s membrane with the cell’s. That fusion protein is a kind of loaded spring. Before it strikes, it sits in a tense, high-energy arrangement called the prefusion conformation. After it strikes — after it has driven the two membranes together — it snaps into a slack, low-energy arrangement called the postfusion conformation, and it cannot go back. The two shapes are the same chain of amino acids folded two completely different ways, and the immune system reacts to them differently. The most powerful protective antibodies, the ones that actually stop the virus, recognize the prefusion shape. The postfusion shape shows the immune system mostly the wrong things to learn.

Read more

The story everyone thinks they know runs like this: a strange new virus appeared, the world locked down, and within a year — impossibly, almost magically — a new kind of vaccine arrived to meet it. A vaccine made of genetic material. A vaccine designed, some accounts marvel, in a single weekend. The speed was the headline. The speed was also, for many people, the reason to be afraid: anything built that fast, the worry went, must have skipped something.

That story is true in its particulars and backward in its shape.

The shape it gets wrong is the most important part. The vaccine did not appear in a year. The biology underneath it — the unglamorous, decades-long project of teaching a fragile molecule to do useful work inside a human body — had been underway since the early 1990s. By the time the SARS-CoV-2 genome was posted online in January 2020, the hard part was already done. What looked like a sprint was the last leg of a marathon that almost nobody had been watching, run for years by scientists who were, for most of that time, doubted, defunded, and quietly told they were wasting their careers.

To understand why the timeline collapsed the way it did — and to answer the question that this particular story gets weaponized around more than almost any other in modern medicine — we have to start with what the vaccine actually is. Not what it feels like. What it is.

A Recipe, Not a Rewrite

Read more

Almost every medicine is made in a place the patient will never visit, from materials that have nothing to do with the person who will swallow or be injected with it. A tablet is pressed by the million. A vial of antibody is filled from a vat that fed a thousand other vials. The factory does not know who you are, and it does not need to. The product exists before the patient does.

A CAR-T therapy like Carvykti is built the other way around. CAR stands for chimeric antigen receptor — a receptor stitched together from parts, then placed onto a patient’s own immune cells so those cells will recognize and attack the patient’s cancer. The crucial word is own. The raw material is the patient. White blood cells are collected from one specific person, engineered, grown, and returned to that same person, and to no one else. The manufacturing run begins and ends in the same body. For this one class of drug, the factory floor is, in effect, the individual.

That is a strange fact, and it bends everything downstream of it. It is the reason this medicine costs what it costs, takes as long as it takes, and is built one batch at a time with no spare. Manufacturing is usually the most anonymous stage in the long road from idea to therapy (SP-12). Here it is the only stage that touches the patient directly — because the patient is what it is made of.

The Loop That Returns To Where It Began

Read more

When Carvykti was approved in 2022, there was nothing to stock on a shelf. The medicine for any given patient would not exist until that patient’s own cells were drawn — no batch to inspect, no vial to release lot by lot, no interchangeable unit at all. A pharmacy shelf is built on sameness: every vial of a conventional medicine is interchangeable with every other, and that interchangeability is what makes approval simple to enforce. The regulator clears the formula, the factory makes it by the million, and the bottle that reaches you could have reached anyone.

An autologous CAR-T therapy breaks that arrangement at the root. “Autologous” means the starting material is the patient’s own cells; “CAR-T” means those cells — a kind of white blood cell called a T-cell — have been re-engineered to carry a chimeric antigen receptor (CAR), a synthetic protein built to recognize a marker on the cancer. The cells are collected from one specific person, shipped to a manufacturing site, modified, grown, and returned to that same person. There is one batch per patient, and the patient is the product. Until someone’s cells are drawn, the medicine for them does not exist.

That single fact reshapes what “approval” can even mean. There is no vial to stock, no shelf inventory to inspect, no interchangeable unit to release lot by lot. So the approval cannot govern a thing on a shelf. It has to govern places, procedures, and identity — where the therapy may be delivered, by whom, and how each person’s cells are kept traceably their own. The stamp that elsewhere lands on a product instead lands on a system (SP-11).

A Medicine That Is a Procedure

Read more

In CARTITUDE-4, the patients who received cilta-cel — a chimeric antigen receptor T-cell therapy, or CAR-T, made from a patient’s own immune cells — did so well that the trial could not name its central number. Median progression-free survival (PFS), the point at which half the group has had their cancer grow or has died, was simply not reached in the cilta-cel arm: by the time the analysis closed, fewer than half had progressed or died. In the group given standard regimens, that same milestone arrived at 11.8 months. The hazard ratio (HR) — a single number comparing the moment-to-moment risk of progression or death between the two groups — was 0.26, meaning roughly a 74% reduction in that risk for the patients who got the cell therapy. A later analysis, with nearly three years of follow-up, held that advantage — a hazard ratio of 0.29, about a 71% reduction — and added a significant survival benefit as well: a hazard ratio for death of 0.55, roughly a 45% lower risk of dying.

That result was not a surprise — it was the evidence the single arm had owed since accelerated approval. Cilta-cel had already reached the market on a narrower kind of evidence — a single-arm study, with no comparison group, judged on how many patients responded. Accelerated approval on that basis came with a condition attached: run the randomized trial that proves the benefit is real, against an alternative, in living patients. CARTITUDE-4 was that trial. It did two things at once. It supplied the proof the single arm could only promise, and it carried the therapy out of the last-resort setting toward earlier treatment. Doing both also exposed something no ordinary drug’s confirmatory trial has to reckon with.

What a Single Arm Could Only Promise

Read more

Ninety-seven people enrolled in the trial called CARTITUDE-1 (CS-CARVYKTI-03), and every one of them got the drug. There was no second group. No one drew a placebo, no one was sorted into a comparison arm receiving the standard of care, no coin was flipped to decide who would and would not be treated. The trial had a single arm, which is the plain technical name for exactly that: one path through, the same intervention for all comers. And on the strength of that single-arm study — 97 patients, no control group, run as a combined Phase 1b/2 — the therapy won approval.

The entire architecture of a rigorous drug trial is usually the comparison: you give the new thing to one group, give something else to another, and measure the gap. Remove the second group and you seem to have removed the experiment. Yet for a particular kind of therapy given to a particular kind of patient, the single arm is not a shortcut or a lapse. It is the defensible design — and, at the same time, an incomplete one. Both of those are true, and holding them together is the whole point.

The patients in CARTITUDE-1 had multiple myeloma, a cancer of the plasma cells in the bone marrow. They were not newly diagnosed. Every one had already been through at least three prior lines of therapy — three full treatment regimens, tried and exhausted. Ninety-nine percent were refractory to their most recent therapy, meaning the cancer had kept growing through the last thing given to it. Eighty-eight percent were triple-class refractory: resistant, simultaneously, to the three major classes of drugs that myeloma treatment relies on. These were people for whom the standard menu had run out.

What the Trial Was Measuring

Read more

Cilta-cel, the CAR-T marketed as Carvykti, is dosed by body weight — a fixed number of engineered cells per kilogram. What that dose becomes inside a patient is not fixed at all. Two people given the identical weight-based count can diverge completely. In one, the cells multiply modestly and the week passes with a fever and little else. In another, the same starting count multiplies into a population orders of magnitude larger, and within days that patient is in an intensive-care bed with a plunging blood pressure. The number infused was the same. The dose, in any sense that matters, was not.

This is the strange arithmetic of a chimeric antigen receptor T-cell — a CAR-T, the patient’s own T-cells drawn out, reprogrammed to carry a chimeric antigen receptor (a CAR, the engineered protein that lets the cell recognize the tumor), and returned. A CAR-T is a living drug. It expands inside the body and persists there. And because it expands, the quantity that ends up acting on the patient is not the quantity anyone administered. It is a quantity the patient’s disease produces.

That single fact bends the entire logic of a first-in-human trial, the stage whose job is to learn what a new drug does in a person for the first time (SP-08). For most drugs, the first-in-human question is how high a dose a body can tolerate. For a CAR-T, the dose within a patient is not the variable under the developer’s control. The disease is.

Why The Dose Sets Itself

Read more

A small-molecule drug can be written down. Its application to begin human testing — the Investigational New Drug application, or IND, the package a sponsor files for permission to give a therapy to people for the first time — opens with the thing itself: a chemical structure drawn bond by bond, a molecular weight, a list of impurities measured to fractions of a percent, and a description of how the compound is synthesized so that the batch made next year is the same as the batch made today. Identity, for that kind of drug, is a picture. You can hold two vials from two factories and say, with a spectrometer to back you, that they contain the same substance.

An autologous chimeric antigen receptor T-cell therapy — a CAR-T like Carvykti, built from the patient’s own T-cells reprogrammed with a chimeric antigen receptor (a CAR) so that they recognize and kill cells bearing a chosen target — cannot be written down that way. There is no fixed molecule to draw. The product is a particular person’s living cells, collected from their blood, engineered, and grown; the next patient’s product starts from different cells and ends up different. No two batches are identical, and none of them is a substance in the sense a chemist would accept. So the IND for a CAR-T has to answer the identity question a completely different way — not by saying what the product is as a structure, but by specifying what it must measure up to before anyone is allowed to receive it.

What Stands In For A Molecule

Read more

In the standard efficacy experiment behind a CAR-T therapy — the kind that supported Carvykti, and every approved CAR-T before it — a mouse a few weeks old carries a human tumor under its skin, put there on purpose. Into the mouse goes a dose of engineered human T-cells — immune cells drawn from a patient, rewired in the lab to recognize that tumor. Over days, the mass shrinks. Measured with calipers, it gets smaller and then it is gone. By the logic of preclinical testing, that is a clean result: the engineered cells find the human cancer and clear it, which means the receptor on their surface does what it was designed to do.

Now ask the mouse the other question — the one that actually keeps the developers awake. When this therapy reaches people, its most feared complication is a body-wide inflammatory surge. Does the mouse show any sign of that surge? It does not. The blood of these animals stays quiet. One of the chemical signals that drives the surge in patients sits below the level a lab instrument can even detect. The model that proves the cells kill the tumor is, by construction, unable to show how they might harm.

That gap — efficacy modelable, defining toxicity invisible — is the preclinical problem for this therapy. For a conventional drug, the preclinical stage is where most dangers get caught before anyone is dosed. For an autologous CAR-T product — a chimeric antigen receptor therapy built from a patient’s own cells — the stage is reorganized from the ground up. The things easiest to model are not the things you most need to know.

The Gauntlet That No Longer Applies

Read more