Every autumn a vaccine arrives at pharmacies and clinics that looks exactly like the one from the year before. It isn't the same vaccine. This year it isn't even close: according to the CDC, all three virus components of the 2026-2027 flu shot were updated from last season's version, including a change aimed at a variant called subclade K, which spread widely last winter.
Readers of this column know my habit by now. The calendar supplies the nouns; biology supplies the verbs. The flu shot is the best example I know of a medicine organized around a single verb, and the verb is mutate. So before the season starts in earnest, here is the mechanism behind why the shot from last fall is, biologically speaking, out of date.
A coat that keeps changing
Start with what your immune system actually sees when influenza arrives. It doesn't inspect the whole virus. It sees the outside, mainly two proteins studding the virus's surface: hemagglutinin and neuraminidase, the H and the N in names like H3N2. Antibodies, the immune system's Y-shaped tags, work by gripping particular shapes on those proteins. A vaccine is essentially a preview of those shapes, shown to the body ahead of time so the right antibodies are ready and waiting.
The trouble is that the shapes don't hold still. As influenza copies itself, small mutations creep into its genes, and some of them alter those surface proteins. The CDC calls this antigenic drift: small changes that pile up continually, until antibodies made against an older version stop gripping as well as they used to. Drift is the main reason the CDC gives for reviewing the vaccine's composition every single year. The virus doesn't need to overpower your immune system. It only needs to change its coat until your immune system stops recognizing it.
There's a second, rarer process called antigenic shift, an abrupt swap that produces surface proteins almost nobody has immunity to. That's how pandemics like 2009's begin. Drift is the everyday version: quieter, slower, and relentless.
The variant that arrived late
Last year gave us a textbook case. According to the World Health Organization, a notably different version of H3N2 emerged in August 2025 and spread rapidly. Scientists classified it as J.2.4.1; everyone else calls it subclade K. The timing was what hurt. The 2025-2026 Northern Hemisphere vaccine had already been designed around an earlier relative, and subclade K showed up, in the words of the Canadian epidemiologist Danuta Skowronski, "at the tail end of Southern Hemisphere's season." She was precise about what kind of change it was: "This is a major drift, not a shift."
I had to read the next part twice when I first learned it: a mismatched vaccine doesn't stop working. It works less. WHO says subclade K contributed to an earlier-than-usual season start in many countries. In the United States, CDC estimates through the end of February counted somewhere between 24 and 47 million flu illnesses and 310,000 to 710,000 hospitalizations, with the adult hospitalization rate the third highest since the 2010-2011 season and children's the second highest. Preliminary vaccine effectiveness came in at 14 to 48 percent for children and 22 to 34 percent for adults under 65, lower than in previous seasons. A partial match is still a match. But those numbers are what a moving target costs.
A forecast made in February
So why not simply aim at whatever is circulating? Because the aiming happens long before the season. Twice a year, WHO convenes experts to review data from its Global Influenza Surveillance and Response System, a network of 166 institutions across 136 countries that has been watching influenza since 1952. The meeting for the Northern Hemisphere happens in February. Then comes the slow part. Most flu vaccine is still made by a method more than 70 years old: the chosen viruses are injected into fertilized chicken eggs, incubated for several days to multiply, then harvested, inactivated, purified, and tested, a process that consumes enormous numbers of eggs.
That makes a flu vaccine less a cure than a forecast. Every February, scientists look at which versions of the virus are gaining ground and bet on which one will dominate a winter that's still most of a year away. Some years, a variant shows up after the bet is placed.
This year's bet was placed on February 27, when WHO announced its recommendations after a four-day consultation: new H1N1, H3N2, and B components, with the H3N2 pick chosen to match subclade K. On March 12, the FDA's vaccine advisory committee voted 7 to 0 to recommend the same strains for American vaccines. I noticed that detail because of where this column began. In March, I wrote that the United States had left the WHO, and that the flu shot in an ordinary American pharmacy is redesigned every year from samples gathered through that global network. This spring, the American choice still matched the global one, strain for strain. The machinery of the forecast kept running, and much of our medicine cabinet rests on it.
Evolution with a calendar
Here is what I find strangely beautiful about all this: you can watch evolution happen on a schedule. Textbooks tend to present natural selection as something slow, finches and fossils and millions of years. Influenza runs it in months. A mutation appears in one infected person; if it helps the virus slip past existing antibodies, that version spreads faster than its relatives, until it accounts, as subclade K did, for most of the influenza viruses reported across several regions. The selection pressure is us: our collective immunity pushes the virus toward whatever shape we haven't seen yet. The vaccine pushes back, and the virus answers the following year. It's an arms race with a standing meeting.
It also means nobody, including me, can tell you what this winter holds. The new shot is aimed at last season's dominant virus, which is the best information anyone has and still a guess. WHO's advisers meet again later this month to pick strains for the Southern Hemisphere's 2027 season, and the surveillance labs never stop sequencing in between. Somewhere, the next variant may already be copying itself, one small mistake at a time.
This isn't medical advice; that belongs to your own doctor. It's a description of an engine. Antibodies remember shapes, and the virus's entire strategy is to change shape. The flu shot is rebuilt every year because the flu rebuilds itself every year. Thresholds don't negotiate, and neither does drift. The only thing we get to choose is whether we keep updating our side of the match. I'll be watching the verb.
