How the research program developed

Sixty-three publications, from doctoral training in 2003 to 2025, look at first like several separate careers, one on the alarm system cells use against viruses, one on microRNAs, one on how influenza schedules its genes, one on rebuilding viruses into instruments, and one on COVID-19. Two habits run through all of them. The first is that an immune response is interesting for how much, which parts and how soon, rather than for whether it happens. The second is that a virus can be rewritten, and that a change written into a virus reaches whatever the virus reaches and faces the same selection. Neither is announced as a plan in any single paper, and both are visible across many.

Interferon, and why proportion is the question

When a cell detects a virus it switches on protective genes and releases interferon, a signal warning its neighbours. The textbook version is a switch, and this work treats that as too coarse.

The oldest thread begins in training-period work. tenOever 2007, with the Maniatis laboratory at Harvard, found that mice missing one enzyme made normal amounts of interferon and still failed to turn on about a third of the genes interferon is supposed to activate. The response is not one thing. Schmid 2010 then showed that much of the antiviral program runs even in mice that cannot receive interferon at all, and Ng 2011 explained how a cell picks which version to run.

Thirteen years later the same question was asked of a new virus. Blanco-Melo 2020 compared SARS-CoV-2 against five other respiratory viruses and found not an absent response but a badly proportioned one, low interferon alongside vigorous inflammation that did not depend on interferon. Nilsson-Payant 2021, in the paper on the NF-kappa B footprint, then inverted that reading by showing the inflammatory arm is something the virus requires in order to replicate rather than something it failed to suppress.

Restraint matters as much as activation. Han 2018 and Manivasagam 2025, led by the Manicassamy laboratory, found a protein holding antiviral genes quiet that respiratory viruses destroy within forty minutes of entry.

An evolutionary question answered in two halves

Plants, worms and insects defend themselves by shredding viral genetic material with small RNA guides. Vertebrates keep most of that machinery but appear to use it for ordinary gene regulation instead, relying on interferon. Whether that appearance was correct was disputed.

The negative half came first. Backes 2014 built a virus carrying a poxvirus enzyme that destroys the cell's small RNAs, reasoning that if those RNAs were a defence, removing them should help the virus. It did not help, in cells or in mouse lungs, and the result held in mice unable to respond to interferon, which rules out interferon masking the effect. The authors called this a strong argument rather than proof, since absence of an activity is hard to demonstrate.

The positive half is more interesting. Benitez 2015, in the paper on reconstructed silencing, built the missing defence artificially into influenza and found it protects mice completely, including mice that cannot use interferon at all. The system vertebrates do not use would have worked, which removes the simplest explanation for why they gave it up. tenOever 2016 proposes that a body-wide small RNA defence would need an enzyme whose mere presence sets off the interferon alarm, and calls this a hypothesis.

Rebuilding a virus in order to ask something

A host microRNA silences any message carrying a matching sequence. Put such a sequence into a virus and it grows only in cells lacking that microRNA, so whether it replicates becomes a property of the cell it lands in.

Perez 2009 used that to build a vaccine candidate that grows normally in eggs and is crippled in mammals. Its more durable contribution was showing that influenza leaves the silencing machinery intact, which is what makes it usable as a tool at all. What follows is linked by citations in the papers themselves rather than by resemblance. Langlois 2012 turned the element into an instrument, silencing influenza only in immune cells so one animal carried both a permissive and a non-permissive compartment, and Langlois 2013 turned it into a safety layer for risky influenza experiments, using a microRNA present in humans and absent from ferrets, so a virus can be studied in ferrets and cannot grow in people.

A parallel line ran the other way. Varble 2010 tested the claim that an RNA virus cannot produce a microRNA. It can, once the genome is rearranged so the hairpin is cut out of a discarded piece, and that rearranged segment became the space into which almost everything later was inserted. A suggestion in the discussion of Langlois 2012 in Molecular Therapy, that a virus could carry a library of such elements so selection inside an animal picks the winners, was implemented in Varble 2013 and rebuilt in influenza in Benitez 2015, the screening paper. Its winner was a sensor called MDA5, which two earlier studies elsewhere had concluded plays no part in influenza detection. Both are right about different things, and the lesson is that an interferon readout can miss what a sensor contributes.

Where the program changed its mind

Shapiro 2010 in RNA reported that a virus replicating outside the nucleus can make a microRNA without the nuclear enzyme normally required. Shapiro 2012 in RNA, using a cleaner genetic method, found that enzyme absolutely required after all, and said so. The correction mattered more than the original claim, because it drew attention to something odd, that the enzyme moves out into the cell body during infection. Shapiro 2014 and Aguado 2017 followed that thread and found it restricts viruses by gripping structured viral RNA rather than cutting it into guides, and that relatives of the protein from bacteria, archaea, yeast and a sea squirt do the same. That position belonged to neither side of the original dispute.

Hoagland 2021 found antiviral activity in hamster brain, gut and olfactory tissue where SARS-CoV-2 barely reached, and suggested, as speculation, that stray viral material was drifting there and setting off alarms. Carrau 2023 from the same laboratory tested that and found something else. Interferon made in the airway enters the bloodstream and primes every other organ in advance. Three manipulations agreed. A steroid delaying the airway response let the virus reach liver, spleen and gut, intravenous injection bypassing the airway infected organs it normally spares, and infecting the airway first protected those organs. The response far from the lung is therefore not evidence of the virus being there. It is the lung protecting the rest of the body.

What stopped

An honest account includes work that did not continue. The line on small influenza RNAs that switch the viral copying machine between modes ran from Perez 2010 to 2012 and stopped, and how those RNAs are made is still unknown. The delivery vector of Schmid 2014 in the Journal of Virology reached mouse airways and no paper here shows it silencing anything in an animal. Heaton 2014, which tagged infected cells permanently and found airway cells that survive infection and stay inflamed, produced one paper, and the design was developed elsewhere. Varble 2014, which showed airborne influenza transmission can be founded by as few as two virus particles, left the filtering barrier unidentified. And Guzmán-Solís 2021, on viral genomes from teeth in Colonial Mexico City, has no predecessor or successor.

Where it landed

Two practical positions rest on animal experiments and are short of human evidence, which the papers state. Interferon delivered into the airway early lowers virus, illness and onward transmission in hamsters, and it works because the airway warns the rest of the body. And the choice of drug target decides whether an antiviral also helps the immune system, since Nilsson-Payant 2021, in the paper on nucleoprotein availability, found two compounds that block influenza equally well differing in whether they leave the cell alarmed.

The methodological result travels furthest. tenOever 2019 sets out the design vocabulary with one criterion applied throughout, whether a rebuilt virus is as fit as the original, because a crippled recombinant tells you about a different virus than the one you meant to study.

Publications referenced

Read the detailed intellectual history →