The research program

The subject

This laboratory studies the earliest decisions a host makes about a virus. Not whether infection is detected, which is usually settled within minutes, but what follows from detection. Which genes are induced, in which cells, how fast, for how long, and at what cost to the organism. Alongside that runs a methodological commitment, that the most direct way to interrogate a virus is to rewrite it, because a modification encoded in a viral genome travels wherever the virus travels and is subject to the same selection.

Sixty-three publications between 2003 and 2025 span influenza A virus, SARS-CoV-2, vesicular stomatitis virus, Sindbis virus, dengue virus, human cytomegalovirus and others. They organize into six research areas and thirty-four themes, but the areas are a convenience. Two questions recur across all of them.

The interferon response is a system with settings

The textbook account of innate antiviral immunity describes a pathway. A sensor detects viral nucleic acid, kinases activate transcription factors, interferon is made, neighboring cells induce hundreds of interferon-stimulated genes, and an antiviral state follows. Each element of that account is correct and the summary is misleading, because it treats a graded and combinatorial system as a switch.

The oldest thread here is training-period work in the Hiscott and Maniatis laboratories identifying the IKK-related kinases IKKepsilon and TBK1 as the activity that phosphorylates IRF3 and IRF7. The consequential finding came next. Mice lacking IKKepsilon produce normal interferon beta and nonetheless fail to induce roughly a third of the interferon-stimulated gene repertoire, because interferon itself activates IKKepsilon, which phosphorylates STAT1 at serine 708 and determines which response elements ISGF3 will occupy. The response is divisible, and something decides how it is divided.

Work in the independent laboratory pursued that. IRF7 and ISGF3 were shown to read overlapping response elements, so a substantial antiviral program is still induced when type I and type III interferon signaling are both removed, which means interferon secretion is one route to the antiviral state rather than the only one. Phosphorylation of STAT1 serine 708 was shown to block assembly of the STAT1 homodimer while leaving the STAT1 and STAT2 heterodimer intact, biasing a shared pool of subunits toward the type I response and away from the type II response, in work co-led with the Maniatis laboratory. And sustained IRF7 activity was shown to induce a kinase that redirects IRF3 out of homodimers and into IRF3 and IRF7 heterodimers, broadening the transcriptional output as a threat persists. The picture is of a small set of transcription factor subunits combinatorially allocated, where the allocation rather than the trigger determines the response.

If the program is expensive, it should be restrained, and it is, in three ways the corpus documents. It is repressed during homeostasis by the capicua repressor complex occupying interferon and interferon-stimulated gene loci, which respiratory virus entry degrades within minutes through EGFR signaling, a discovery led by the Manicassamy laboratory. It is buffered once triggered, since cells lacking the stress granule nucleators G3BP1 and G3BP2 respond to double-stranded RNA with excessive sensor activation and MAVS-dependent apoptosis, work led by the Hur laboratory. And it is incompatible with some cell states, since driving an interferon-stimulated gene program in human induced pluripotent stem cells produces lasting transcriptional change and impaired differentiation.

A virus can be rewritten, and then it becomes an instrument

A host microRNA silences any transcript carrying a complementary site. Writing such a site into a viral genome therefore makes replication conditional on the cell the virus lands in, because the silencing is supplied by the host rather than the virus.

The first application inserted response elements for a mammalian-ubiquitous microRNA into the influenza nucleoprotein coding sequence, attenuating the virus in mice while leaving growth in embryonated eggs intact, which is a workable live attenuated vaccine design. Its more durable contribution was establishing that influenza infection and NS1 leave the silencing machinery functional, without which none of what followed would work.

What followed is connected by citation rather than by resemblance. Restriction to a cell type turned the design from an attenuation strategy into an instrument, silencing influenza specifically in hematopoietic cells so that one animal carried both a permissive and a non-permissive compartment, which showed that replication inside antigen-presenting cells is dispensable for CD8 T cell priming yet responsible for much of the RIG-I-dependent interferon response in vivo. The same design applied to dengue virus identified hematopoietic cells as the dominant amplification compartment. Applied across species rather than cell types, using a microRNA abundant in human airway and absent from ferret airway, it produced a biocontainment layer permitting influenza transmission studies in ferrets with viruses that cannot replicate in people.

A parallel line asked whether an RNA virus could produce a microRNA at all. It can, once segment 8 is rearranged so the hairpin is excised from a dispensable transcript, and that rearranged segment became the cassette into which most later modifications were inserted. A suggestion in the discussion of one of these papers, that a virus could carry a library of such elements so that selection inside an infected animal identifies the ones that matter, was implemented as an in vivo RNA interference screen in Sindbis virus and rebuilt in influenza. That influenza screen identified MDA5 as a contributor to the antiviral response despite two prior studies elsewhere concluding it plays no part in influenza sensing, and the reconciliation is instructive. Both conclusions are right about different readouts, because a sensor can contribute to the response without contributing to interferon induction.

An evolutionary question, answered in two halves

Plants, nematodes and arthropods defend themselves against viruses by cleaving viral RNA with small RNA guides. Vertebrates retain Dicer, Argonaute and the rest of the machinery and appear to use it only for gene regulation, relying on interferon instead. Whether that appearance is correct was genuinely disputed, and this laboratory argued both sides of the evidence.

The negative half came from arming vesicular stomatitis virus with a poxvirus enzyme that destroys the cell's Argonaute-loaded small RNAs. If those RNAs constituted a defense, removing them should have helped the virus. It did not help, in cells, in primary macrophages, in mice, or in mice unable to respond to interferon, which excludes interferon masking the effect. The authors described this as a strong argument rather than a proof, since absence of an activity is hard to demonstrate.

The positive half is the more interesting result. Rebuilding the missing defense artificially, either by repurposing host microRNAs as virus-specific guides or by having the virus encode a small interfering RNA against itself, attenuated influenza A virus in mice by more than five logs and prevented disease, with no requirement for type I interferon signaling whatsoever. The system vertebrates do not use would have worked. That removes the simplest explanation for its absence and leaves a real question, for which the corpus offers a hypothesis, that a body-wide RNA silencing defense requires proteins whose presence would itself trigger the interferon system, and two mechanistic constraints. Escape from RNA-guided targeting tracks with a virus's capacity for polymerase template switching rather than with genome polarity, and where the virus cannot supply escape the host editing enzyme ADAR1 can, by destroying target sites through adenosine to inosine editing.

A related finding came out of a correction. An early report that cytoplasmic microRNA processing proceeds without the nuclear enzyme Drosha was revised by the same laboratory two years later, using cleaner genetics, to show Drosha absolutely required. The correction mattered more than the original claim, because it drew attention to Drosha leaving the nucleus during infection. Following that produced the finding that Drosha restricts positive-strand RNA viruses by binding structured viral RNA and impairing the polymerase, separably from microRNA biogenesis and from its own catalysis, and that RNase III homologues from bacteria, archaea, yeast and a urochordate do the same. That is a position neither side of the original dispute had held.

Influenza as a self-timing circuit

A third line concerns how influenza A virus controls its own gene expression without dedicated regulatory proteins, using timing and stoichiometry instead. Small viral RNAs corresponding to segment termini were found to load into the polymerase PA subunit and act as segment-specific allosteric enhancers of genome synthesis. The inefficient 5 prime splice site of segment 8 was shown to function as a timer, causing the nuclear export protein to accumulate slowly as a minor product, with both raising and lowering that rate attenuating the virus. Nucleoprotein availability was shown to determine not only whether full-length genomes are made but whether aberrant, immunostimulatory replication products accumulate, which means replication competence and immune invisibility are coupled and two drugs that inhibit influenza equally can differ in whether they leave the cell alarmed. And the same splicing step proved to be a family-wide vulnerability, since archaeal L7Ae proteins block production of spliced influenza A, influenza B and isavirus transcripts without disturbing host splicing.

Pandemic work, and what it inherited

The SARS-CoV-2 work begins abruptly in early 2020 and draws on capabilities built for other purposes, comparative transcriptional profiling, hamster models and a habit of asking about proportion. The founding observation was that SARS-CoV-2 provokes not an absent host response but an imbalanced one, low type I and type III interferon alongside strong chemokine and IL-6 induction. The mechanism followed, that the virus engages NF-kappa B at chromatin without engaging the interferon transcription factors, and that this inflammatory arm is a viral requirement rather than a failure of suppression, since loss of p65 or p50 abolishes replication. The therapeutic implication followed from that, that intranasal type I interferon lowers viral load and disease in hamsters given before or after challenge. And the explanation for why the infection stays respiratory followed later still, that interferon generated by airway replication circulates and primes every organ in advance, so blunting or bypassing the airway response permits viremia and productive infection of liver, kidney, spleen and brain. That last result corrected an earlier proposal from the same laboratory that distal inflammation reflected disseminated viral RNA.

Much of the remaining pandemic work is collaborative and led elsewhere, including the phosphoproteomic surveys, the kinase substrate profiling, the human pluripotent stem cell tropism panel, the airway chip and the humanized mouse. The sequelae work is co-led and shows that a persistent inflammatory program remains in olfactory tissue a month after clearance, that olfactory receptor gene expression collapses in neurons the virus never enters, and that sensory abnormalities outlast detectable virus.

What this program has not settled

The corpus is explicit about its own boundaries. No viral antagonist explaining the low interferon phenotype of SARS-CoV-2 was identified. The enzyme responsible for cytoplasmic pri-microRNA processing was never found. How small viral RNAs are generated remains unknown, and that line stopped in 2012. The delivery vector reached mouse airway and was never shown to silence anything in an animal. The recipient-side barrier that reduces an airborne influenza inoculum to as few as two founder genomes was never characterized. And the two clearest translational positions, intranasal interferon and the choice of antiviral target, rest on animal experiments and are short of human evidence, which the papers themselves state.