A century of people have quoted the Gallops Island result. Almost nobody has tried to explain it. A search of the modern medical literature turns up one historian who describes the experiments accurately and offers no mechanism, and a handful of papers that cite the bulletin in passing. What follows is therefore an assembly rather than a summary: the four explanations that the evidence supports, each with what is known for and against it.
Rosenau named the first two himself, in print, in 1919.
One: the volunteers were immune
This is the strongest explanation and the experimenters’ own first thought. Here is their discussion, written in 1919:
It is possible that all our volunteers resisted infection because of a natural or an acquired immunity. If this be true, then we have an indication of a much higher degree of immunity to this disease than is generally assumed.
And Rosenau, from the podium in June 1919: “Our volunteers may not have been susceptible. They may have been immune. They had been exposed as all the rest of the people had been exposed to the disease, although they gave no clinical history of an attack.”
The facts in the report support it as far as the report can. Every one of the 62 came from the Deer Island camp, which had recorded 186 cases of influenza in an average population of 1,058 men between 7 September and 3 November 1918 — an incidence of 176 per 1,000 among the men who were ill enough to be recorded. The report says that “in varying degree all of these men had been exposed to the infection at this station, and in some instances also at preceding stations and places.”
What nobody in 1918 could know is how many more had been infected without being recorded, because an unrecorded infection was simply invisible. A man can meet influenza, fight it off with a day of feeling slightly unwell, and be protected afterwards, and in 1918 there was no test on earth that could find that out.
Modern work gives the size of the gap. A model of twelve English populations through the three waves of 1918 and 1919, fitted to repeat attacks in 24,706 people, found that before the summer wave only about half the population was susceptible, with prior immunity in adult populations running between 38 and 66 per cent, and that only a quarter to two thirds of exposed susceptible people reported symptoms at all. In a barracks that had just been through an epidemic, the proportion already protected would have been higher than that, not lower.
The Boston experimenters’ own counter-argument was San Francisco: their colleagues at Angel Island had used men quarantined for a month who had never met the disease, and had also failed. That is fair, and it is weaker than it looks — only twenty-six of those fifty men were actually inoculated with patients’ material, and all fifty had been given three doses of a vaccine containing five billion Pfeiffer’s bacilli each a month earlier. The San Francisco series is told in its own companion piece.
There is also a modern finding that makes this explanation harder to dismiss on any screening. In ferrets, pre-existing immunity to one influenza subtype has been shown to abolish airborne transmission of another — and that protection was “independent of neutralizing antibodies,” which means it would not show up in the standard blood test that modern challenge studies use to select volunteers. Sixty-two men who had just lived through an epidemic could have been substantially immune in a way that neither a 1918 questionnaire nor a 2013 antibody titre would reveal.
Two: the material was collected too late
Rosenau’s second explanation, also in the 1919 discussion:
Epidemiological evidence points to the likelihood that influenza is most communicable during its early stages. Most of our material was obtained during the first, second, or third days of the disease, sometimes as early as the eighth or tenth hour after the beginning of symptoms. In no case, however, did we obtain material during the period of incubation. If our volunteers were susceptible, then it could be argued that the material used did not contain the virus.
This turns out to be close to right, and for a reason nobody could have stated then. Modern challenge studies show that virus shedding begins twelve to twenty-four hours before symptoms appear and tapers off before the symptoms do. The most infectious person in any room is usually somebody who does not yet know they are ill. An experiment can only use donors it can identify, and it can only identify them once they are ill, which is to say after the best moment has passed.
The experimenters worked this out for themselves within three months. The second Boston series went after the freshest material it could get — eight hours after onset in one experiment, four hours in another — and its formal conclusion singles out transmission “from patients in the early stages of this disease, probably within less than 12 hours from onset.” That is a remarkable thing to have inferred in 1919 from nothing but the pattern of one’s own failures.
Three: the route and the dose were wrong
This one is modern and the experimenters could not have reached it.
Everything done at Gallops Island was aimed at the nose and the throat: sprays, drops, swabs, atomisers, eye instillations, and men breathing and coughing into faces from two inches. That was not an oversight. The nose and throat were where Pfeiffer’s bacillus lived, and the experiment was built on the assumption that Pfeiffer’s bacillus was the thing being moved.
What is now known is that influenza delivered as a fine aerosol — small enough to stay airborne and be carried into the lower airway — infects at a far smaller dose than the same virus squirted into the nose, and is more likely to produce a typical illness with fever and cough. Work on naturally infected people found infectious virus in the fine aerosol of ordinary breath, without coughing or sneezing, and concluded that infection of the upper and lower airways behaves almost independently. A swab in the nose is sampling, and treating, the wrong compartment.
Against this: the ward experiment of 26 November 1918 put ten men two inches from the faces of coughing patients for the best part of an hour, which is as good an aerosol exposure as any experiment has ever contrived deliberately. If fine aerosols are the route, that arm should have worked.
Four: nobody could tell what they had
This is the explanation the experimenters stated most clearly, and the one least often quoted. From the conclusions of the second series, on why no definite conclusion was possible: “the uncertainty of our diagnosis in recipients and donors on account of the lack of decisive criteria as to what is influenza.”
There was no test. Not for the donors, so the material may not have been influenza at all in some arms; not for the recipients, so an infection without much illness would have been recorded as a failure; and not for the great fog of other respiratory illness that was moving through the same men at the same time. The second series had a quarter of its volunteers reporting sick with tonsillitis in the observation week, before anything was done to them.
What the programme could do, repeatedly and convincingly, was transmit streptococcal tonsillitis, and it could do that precisely because the streptococcus could be cultured from the donor and cultured again from the recipient. The arm with a laboratory test behind it worked. The arms without one did not. That is not a coincidence about biology; it is a fact about measurement.
The modern repeat
In 2013 an Anglo-American consortium ran the Gallops Island experiment again, with everything the twentieth century had learned. The study is known as EMIT, and it reported in 2020.
They screened 496 volunteers for low antibody levels against the strain they meant to use. They inoculated 52 of them intranasally with influenza A/Wisconsin/67/2005, and 42 became infected — 81 per cent, with 25 of them showing influenza-like illness. Giving a person influenza on purpose, that is to say, is straightforward, and has been for decades.
Then they housed 75 screened, seronegative recipients with those donors for four days, in a quarantine unit with the temperature, humidity, carbon dioxide and ventilation held at the values of an ordinary indoor winter. Forty recipients wore face shields and sanitised their hands every fifteen minutes, so that only aerosols could reach them; thirty-five did nothing of the kind and could touch their faces freely.
They expected a secondary attack rate around 16 per cent. They got one confirmed transmission in 75 people: 1.3 per cent overall, 2.9 per cent in the unprotected group and zero in the shielded one, a difference that means nothing at those numbers. No recipient developed a fever.
Their own reading, which is as close as the modern literature comes to a verdict on 1918:
The overall low SAR suggests that Donors in this model were minimally contagious and prevents definitive assessment of the modes of transmission.
And, on the specific route Gallops Island was built to test:
Minimal transmission despite days of prolonged exposure to Donors shedding virus into nasal secretions provides strong evidence that contact and large droplet transmission are not important in this model.
Two of their findings bear directly on Rosenau. Only eleven of their 42 infected donors had any detectable virus in their exhaled breath at all, and the amount those donors put into the air was two to four orders of magnitude below what naturally infected people in the community produce. An inoculated donor is not an infectious donor. The men coughing into volunteers’ faces at Chelsea in 1918 were naturally infected, which is in their favour; but they were also, by then, several days into their illness.
The authors’ conclusion was that this kind of experiment may simply be the wrong instrument, and that watching real transmission in real places — comparing dormitories with good and bad ventilation, say — is now the more promising approach. Which is to say: a hundred years on, the recommendation is to stop building the experiment and go back and look at the ship.
What that leaves
Four explanations, none sufficient alone, all pulling the same way: a group of men who were probably not susceptible, given material that had probably passed its most infectious hours, by a route that is probably not the efficient one, and judged by an endpoint that could not reliably tell infection from a sore throat.
And underneath all four, the uncomfortable finding that the modern repeat produced: that person-to-person transmission of influenza, watched closely under controlled conditions, is a rarer and more conditional event than a century of posters about coughs and sneezes would suggest. The posters are not wrong — influenza plainly does spread, and the Yacona lost 80 men out of 95 in a fortnight to prove it. What nobody has yet managed, in 1918 or in 2013, is to make it happen on demand in front of witnesses.