Looking for the “immunological dark matter” of Dr. Karl Friston

Will Bates
14 min readJun 10, 2020


Is the epidemic is turning down on its own accord?

The first chart we need to explain is this one:

In every country, the epidemic grew, turned around, and started to die away, almost like clockwork.

If it’s turning down, how will we know when it’s over? The correct metric for this is a dark-sounding statistic known as excess death, or excess mortality. The CDC computes excess death weekly, and has for decades. Death is seasonal; people die in winter from influenza. To compute “excess” death for a given week in the current year, the CDC subtracts the average number of deaths in that same week in recent years.

the epidemic will be over when “excess death” falls below threshold

The CDC’s current chart for excess death in the U.S., despite a 4-week reporting lag, also makes it clear the the epidemic has turned around. The chart show deaths attributable to Covid-19 in blue, others in green. Red crosses mark the weeks that had “excess” death; note the weeks with excess death during the January, 2018 flu season:

Source: CDC. Death data runs 3–4 weeks behind.

The pandemic (or first wave of it, for the pessimistic) is over when “excess death” falls below threshold (the orange line on the chart).

It’s important to pause and say what this does and does not mean. It does not mean there will be no new cases. It does not mean no one will die of Covid; some people will. It does not mean Covid will “no longer be a problem.” Covid-19 will, for the foreseeable future, remain a problem just as heart disease, cancer, car accidents, and suicide remain problems.

This week, the proponents of social distancing, lockdown, and other “non-pharmaceutical interventions” have taken a sort of victory lap. In a paper published in Nature on 8 May, the modelers from Imperial College, London estimated that shutdowns saved about 3.1 million lives in 11 European countries, including 500,000 in the United Kingdom. With actual deaths in the UK currently totaling 40,883, they are apparently sticking with their model’s March prediction of 550,000 UK deaths, and giving social distancing credit for the difference.

A study from the Global Policy Laboratory at the University of California at Berkeley, also published Monday in Nature, estimated that the United States, had it not imposed shutdowns and other measures, would have seen an additional 4.8 million diagnosed infections and 60 million actual infections.

Yet before we take full credit for taming Covid-19, we need to remind ourselves that homo sapiens comes hard-wired to believe in its agency, and also habitually believes that events that follow each other in time are causally connected: x is followed by y, therefore x caused y — post hoc, ergo propter hoc (“after this, therefore because of this”).

The case for formal lockdowns, for example, is in retrospect troubling. A lockdown can be viewed as an experiment: the hypothesis was that they would reduce case counts, hospitalizations, ICU usage, and death. And, indeed, all those those things happened.

The problem is with the timing. Those things happened too fast. In the UK, for example, deaths from Covid-19 peaked on April 8th:

Yet the UK only went into its nation-wide lockdown on 23 March. “If you work backwards 21 days before that,” observed Dr. Carl Heneghan, Director of the Centre for Evidence-Based Medicine (CEBM) at Oxford, on a BMJ talk program, “that suggests the infections actually peaked pre-lockdown.” (The time it takes Covid-19 to progress from infection through incubation to symptoms and death is at least 21 days, and probably longer. March 23 to April 8 is 16 days.)

Contrary to the Imperial College finding, which uses a complicated counterfactual “what if” model, the University of Oxford’s Blavatnik School of Government in May constructed a “stringency index,” a composite measure of government containment policies, from banning large gatherings to closing schools. Looking at the different countries of Europe, the Oxford researchers could find little correlation between the stringency of restrictions and the severity of the pandemic as measured by excess death.

Deaths in selected European countries. Blue line is multi-year average death rate. Source: Bloomberg.

So if the epidemic is turning down, and we don’t want to give human intervention all the credit, what else could it be? Do we blame the weather? Or is the virus just getting tired?

Actually, both those things deserve serious consideration. They are even related.

A recent observation by Italian doctors that the virus is getting weaker has become so politicized we need to briefly restate the facts.

Dr Massimo Clementi, director of the Microbiology and Virology Laboratory at the San Raffaele hospital in Milan in Lombardy, has been measuring viral load in incoming Covid patients. He plans to publish his data, but has not done so yet. (The week of June 15 for publication has been mentioned recently).

His data was seen in advance, however, by Dr Alberto Zangrillo, the head of intensive care at the same hospital. Dr Zangrillo is a controversial media doctor in Italy, and has been since he was personal physician to Silvio Berlusconi.

On a Sunday, 31 May talk show on RAI, the Italian state television channel, Zangrillo talked about Clement’s findings, saying “swabs that were performed over the last 10 days showed a viral load … absolutely infinitesimal compared to the ones … a month or two ago.” He added something to the effect that it was as if the old virus was gone.

That was enough to bring the World Health Organization down on his head. The next day WHO issued a statement “we need to be exceptionally careful not to create a sense that all of a sudden the virus by its own volition has now decided to be less pathogenic. That is not the case at all.” Various experts chimed in, leading to such headlines as Global experts go head-to-head over claims the coronavirus ‘no longer exists clinically’ (CNBC, 2 June).

Yet the skeptical experts are being disingenuous, since all of them believe that viruses evolve over time to become less pathogenic. What they were really saying was they don’t believe it could be happening with SARS-CoV-2 so soon.

When a virus first appears in a new host, after jumping from one species to another, it is often extraordinarily virulent. There are a number of examples: Ebola, SARS, myxomatosis in European rabbits, Lassa hemorrhagic fever and bird flu.

For the virus, however, there is an evolutionary trade-off between virulence and transmission. When the virus can easily rip through a population (such as a nursing home), viral evolution favors strains that reproduce quickly and shed more. This is the “chicken barn” scenario, seen with tragic results in avian flu.

For SARS-CoV-2, there is a known correlation between the amount of virus that can be measured in a patient’s nose and throat (the “viral load”) and the clinical severity of that patient’s case of Covid-19. Viral load is what Dr Clementi was measuring to explain his clinical observation.

Yet as the virus takes more time to find new susceptible hosts — either because they are physically distant, or intervening ones are immune — viral evolution starts to favor milder strains, which remain infectious longer.

Patrick Shaw Stewart, who normally studies influenza (his blog is here), came up with this illustration of this process:

If the WHO were less interested in enforcing its party line on health messaging, and more interested in science, it might have asked Dr Zangrillo what the weather was like in Milan.

Covid likes it cold. Like other respiratory viruses, it’s more active at lower temperatures (33–35°C) than body temperature (37°C). This is why it is at home in the nose and throat, which are among of the coldest parts of the body (as are the toes, which explains the symptom of “Covid toes”).

Despite “everybody knowing it” there is, surprisingly, no settled explanation for the seasonality of influenza. Stewart has taken the trouble of giving numbers to four different hypotheses:

  • M1 humans crowd together more indoors in cold weather
  • M2 the virus can survive outside the body for longer in winter
  • M3 immune defenses in the respiratory tract are weaker in winter than in summer
  • M4 involves evolution of the virus

M2, about the survival rate of the virus outside the body, suggests infection would be rare in the Tropics, and it is not. The equator passes right through Brazil.

Stewart is very taken by a 2019 Columbia University study that found that many, indeed most, people carried many respiratory viruses all year-round: as many tested positive in summer as in winter. This would imply that seasonal differences are not about transmission, at least for influenza.

We’ll have more to say about M3, immune defenses, below. For the seasonality of influenza, Stewart favors M4, which is the temperature-dependent viral tropism (TDVT) hypothesis.

SARS-CoV-2 most likely has an “RNA thermometer” — an RNA segment that responds to temperature changes with three-dimensional conformational changes that alters gene expression. Virtually all respiratory viruses do.

The TDVT hypothesis is that respiratory viruses evolve to prefer replicating in the cooler nose and throat. The evolutionary advantage to the virus is that the host keeps spreading it by coughs and sneezes. As temperatures rise in the air, and thus in the nose and throat, the virus replicates less.

Patrick Shaw Stewart suggested in May “milder CoV-2 strains that are more temperature-sensitive may arise spontaneously in the next few weeks or months.”

In March, by the way, the average temperature in Milan was 9.4° C (49° F); in May, 19.4° C (67° F), a ten degree Celsius difference.

Along with evolution and temperature, there is a third possibility: something we don’t fully understand going on with immunity.

This speculation, like the one about weaker virus strains, is also politically incorrect. Dr Rupert Beale, head of the cell biology of infection laboratory at the Francis Crick Institute in London, reminds us in an opinion piece in The Guardian on 5 June: Less than 10% of people in Britain are immune to coronavirus. There’s no room for mistakes.

On May 29, the New York Times said much the same thing: In Battling Outbreak, Herd Immunity Remains Distant Objective. The authors take 60% as the percentage of the population needed for herd immunity. They contrast that to recent seroprevalence survey results, which show that at most 20% of the population has antibodies against SARS-CoV-2 in their blood. The dire conclusion is that 80–90% of the population remains susceptible.

One of the early models predicted a turn-down of the sort we are seeing. The Oxford Evolutionary Ecology of Infectious Disease group produced results from a classic “S-I-R model” in the third week of March. It’s worth quoting from the original text: “the epidemic wave … should have an approximate duration of 2–3 months.”

This “Oxford model” suffered what would have to be called a public relations disaster: it was released one week after Professor Neil Ferguson’s “Imperial College” (London) model, where Ferguson and his group famously projected that, without intervention, 2.2 million Americans and a half million Britons would die of Covid-19. Those dire projections got everyone’s attention, including that of UK Prime Minster Boris Johnson and U.S. President Donald Trump. The model is generally credited with prompting policy about-faces by those two governments.

The Oxford model, whose lead author was Dr Sunetra Gupta, languished largely ignored. It was a classic S-I-R model, in which individuals in the population are placed into one of three categories: susceptible (S), infected (I), or recovered ( R ). Individuals become infected, recover or die at rates estimated from empirical data:

The equations governing the March Oxford model

The reason the epidemic curve turns down in a S-I-R model is the “R,” the recovered or immune. An “S-I” model without “R” captures the early stage of an epidemic reasonably well, but never turns down:

An S-I (no R) model.

It is the build-up of immunity that gives epidemic outbreaks their characteristic bell shape. And a build-up of immunity would be a very parsimonious explanation for what we are seeing across multiple countries.

There’s a different twist on this that is possible. Rather than seeing a build-up of immunity, the S-I-R- model also fits well if the size of the susceptible population (the “S”) is actually much smaller than we think.

This is the idea of Dr Karl Friston, a neuroscientist at University College London. Dr Friston is an provocative character. In his research on the brain, he pioneered a type of modeling that worked backwards from rather detached observable data, such as images from MRI scans, toward hidden brain states.

Such modeling is not intuitive, but in today’s technology, perfectly mainstream. Siri and Alexa make sense of what you are saying using hidden Markov models (HHMs), in which the sequence of frequencies produced by your voice, a very raw series of emissions, are the observed data. The model maps these into a sequence of more meaningful states, such as consonant followed by a vowel.

Dr Friston is a founding member of “Independent SAGE,” a group of experts critical of the UK government’s official advisory body, the Scientific Advisory Group for Emergencies (SAGE). Independent SAGE, led by David King, from 2000 to 2007 the UK’s chief scientific officer under Labour, is somewhat in the British tradition of a “shadow cabinet” made up of members from the opposition party.

On 31 May, in an interview with The Guardian, Friston raised eyebrows by stating that the path of the epidemic in Germany could only be explained by some kind of “immunological dark matter.”

“This is like dark matter in the universe,” he added. “We can’t see it, but we know it must be there to account for what we can see.”

In several recent interviews, Friston elaborated only slightly. He doesn’t claim to know what the dark matter is. But he remains convinced that the models work only if there is some portion of the population who is, in his words, “out of the game.”

The concept of immunity has gotten a hard time during the pandemic. Those concerned with health messaging believe that even mentioning immunity is bad thing, in that mistaken beliefs about it may encourage what to them is irresponsible behavior. All news items about antibody tests now seem to require several ritual disclaimer paragraphs, which warn that antibodies in the blood do not imply immunity and, even if they do, there is no telling how good that immunity is, or how long it will last. All of which makes discussing immunity somewhat like discussing genetics when the party line in the Soviet Union was Lysenkoism. But we will try.

Recent studies show we need to broaden our concept of “immunity.” We might use the word “resistance” instead, or in addition. The antibody seroprevalence numbers — 10%, 20%, whatever — are a lower bound in this conception, not the true measure.

The first clue that we need a broader concept of immunity came from some unexpected results in the search for neutralizing antibodies. The blood of recovered Covid-19 patients is potentially valuable. Blood plasma containing the antibodies developed naturally can be given by transfusion to the critically ill. “Convalescent serum” is a very old form of therapy.

Those naturally-created antibodies are also of intense interest to anyone trying to develop a vaccine. The purpose of almost any vaccine is to provoke the recipient’s body into creating artificial antibodies that are “close enough” that they neutralize the real, wild virus.

To determine how neutralizing a candidate antibody is, the antibody is mixed with a “pseudo virus” resembling SARS-CoV-2 in its essentials and the neutralizing effect measured.

In the first such study of neutralizing antibodies [link], Fan Wu and others at Fudan University in Shanghai recruited subjects who had definitely had Covid-19, and had recovered from the disease.

To their surprise, 30% of those who had recovered from Covid-19 didn’t have antibodies in their blood at a level that mattered.

A 33% number for this came up at Rockefeller University Hospital in New York [link], where researchers were trying to find “elite neutralizers.” These are a minority of people whose antibodies are sufficiently potent they might be used for therapeutic drugs. Eli Lilly has such a drug in now in clinical trial.(Interestingly, a Seattle study [link] found a child who was an elite neutralizer whose antibodies were off the chart: “a potency that exceeded the limit of our assay.”)

Yet a third study of patients who had confirmed Covid-19 infections, this one in Lübeck, Germany [link], found: “In about 30% of the patients with mild to moderate symptoms, no significant antibodies could be detected in two consecutive analyses.”

The Rockefeller University researchers speculated that the patients’ immune systems may have removed the infection before antibodies could be produced.

The innate immune system is the first line of defense against infection in vertebrates. For viruses, it generates host type I interferon (IFN), which potentially can clear them out. Unfortunately many viruses, including coronaviruses, have evolved strategies to counteract IFNs.

The adaptive immune system, involving antibodies, typically kicks in later. For SARS-CoV-2, with its long incubation time, an interplay between the two immune systems is potentially dangerous.The adaptive immune system can come in too soon and try too hard. This is the overreaction known as a cytokine storm, which kills healthy cells and causes tissue damage.

There is third possibility, which was initially overlooked because the “novel coronavirus” was assumed to be just that, novel. T cells are simple and almost “blind.” They can see only so-a called “billboard protein,” a peptide they recognize as belonging to some previous virus. When they find a cell surface with the billboard peptide sticking out of it, they assume the cell is infected and kill it off.

In a paper published in Cell on 14 May, researchers at the La Jolla Institute for Immunology in California found that T cells in blood drawn from people between 2015 and 2018 recognized and reacted to fragments of the Sars-CoV-2 virus. “The most reasonable hypothesis,” explained Alessandro Sette, one of the paper’s authors, “is that this reactivity is really cross-reactivity with the cousins of Sars-CoV-2 — the common cold coronaviruses which circulate very broadly.”

An earlier study from a group at the Charité hospital in Berlin found 34% of healthy volunteers had T cell reactivity to proteins in Sars-CoV-2.

Is cross-protection from other related viruses Dr Friston’s “immunological dark matter?” Perhaps.

Even after a vaccine, SARS-CoV-2 will be endemic in the human population. “Eradicating” the virus is wishful thinking: Only two viral diseases have ever been eliminated by human intervention, smallpox (variola major) and rinderpest, a viral disease related to measles that infects livestock.

If we are going to live with SARS-CoV-2, we need to learn, respect, and — if possible — take advantage of its ways. A little humility on the part of homo sapiens won’t hurt. We need, to paraphrase the Serenity Prayer (written by theologian Reinhold Niebuhr but popularized by Alcoholics Anonymous), the courage to change the things we can, the serenity to accept what we cannot, but — most importantly — the wisdom to know the difference.

Follow this blog on Twitter @Will_Bates_sci



Will Bates

Will Bates writes about science, technology, and business. His journalism has appeared in the New York Times, the Wall Street Journal, and numerous magazines.