Intranasal SARS-CoV-2 infection of golden hamsters deposits viral RNA without infectious virus in dorsal root ganglia and spinal cord, producing a milder but longer-lasting mechanical hypersensitivity than influenza A virus and a neuropathic dorsal root ganglion transcriptome at 31 days, from which ILF3 emerges as an analgesic target validated in mouse pain models.
Serafini RA, Frere JJ, Zimering J, Giosan IM, Pryce KD, Golynker I, Panis M, Ruiz A, tenOever BR, Zachariou V. SARS-CoV-2 airway infection results in the development of somatosensory abnormalities in a hamster model. Science Signaling. 2023. Volume 16, issue 784, article eade4984.
DOI 10.1126/scisignal.ade4984. PMID 37159520. PMCID PMC10422867.
Intranasal SARS-CoV-2 infection of golden hamsters deposits viral RNA without infectious virus in dorsal root ganglia and spinal cord, producing a milder but longer-lasting mechanical hypersensitivity than influenza A virus and a neuropathic dorsal root ganglion transcriptome at 31 days, from which ILF3 emerges as an analgesic target validated in mouse pain models.
Sensory symptoms including pain, tingling and numbness occur both during acute COVID-19 and persistently in post-acute sequelae, yet the mechanism was unclear and it was not established whether SARS-CoV-2 reaches the peripheral sensory nervous system at all. Using the golden hamster model of respiratory infection, which reproduces COVID-19 without host or virus adaptation, the authors tracked viral material, host transcription and sensory behaviour in parallel, with influenza A virus as a comparator respiratory RNA virus. Viral nucleocapsid transcripts and Isg15 rose sharply in cervical and thoracic dorsal root ganglia and spinal cord within one day of infection and largely resolved by four to seven days, while plaque assay recovered infectious virus from lung and from no neural tissue. In situ hybridisation placed Spike RNA near nuclei in both neuronal and satellite glial compartments, and nucleocapsid protein was not detectable in these tissues. Behaviourally, influenza produced pronounced mechanical hypersensitivity at one day that resolved by four, whereas SARS-CoV-2 produced a slower, milder decline in withdrawal threshold that reached significance at four days and, unlike influenza, returned at 28 days in both sexes. Transcriptionally, acute SARS-CoV-2 dorsal root ganglia were dominated by neuronal signalling pathways and influenza by interferon pathways, and at 31 days SARS-CoV-2 produced 1065 differentially expressed genes with a neuropathic character. Upstream regulator analysis nominated ILF3, and its inhibitor YM155 reduced hypersensitivity in three mouse pain models.
Abnormal somatosensation accompanies both neuroinvasive and non-neuroinvasive viral infections, with mechanisms that differ by virus. Herpesviruses persist in dorsal root ganglia and drive abnormal sensory neuron activity on reactivation, while HIV induces sensory neuropathy through viral protein interaction with axons together with secondary inflammation. For coronaviruses the mechanism was much less understood. Whether SARS-CoV-2 crosses the blood-brain barrier and infects the central nervous system was unresolved, with ultrastructural studies of post-mortem tissue describing particle-like structures while other studies failed to recover replication-competent virus from brain. Hamster work, including from these groups, had found viral RNA in olfactory bulb, cortex, brainstem and cerebellum during active infection without evidence of infectious material, suggesting that airway replication disseminates viral RNA and inflammatory debris that provoke antiviral transcription in distal tissues.
The paper identifies the specific gap it addresses. Despite extensive study of central nervous system involvement, little clinical or preclinical work had examined penetration of SARS-CoV-2 into the peripheral nervous system and particularly the dorsal root ganglia, and reports on viral transcripts in cerebrospinal fluid conflicted. Earlier coronavirus literature in rodents, including hemagglutinating encephalomyelitis virus and mouse hepatitis virus, had described replication in dorsal root ganglia and spinal cord invasion, attributing neurological dysfunction either to direct viral damage or to collateral damage from antiviral immune responses.
Do somatosensory abnormalities after respiratory SARS-CoV-2 infection arise from exposure of dorsal root ganglion and spinal cord neurons to mature virus, or to circulating inflammatory material and viral RNA, and does the resulting transcriptional state differ from that produced by another respiratory RNA virus in a way that explains a different sensory time course.
The design rests on three comparisons layered onto one animal model.
The first is virus against virus. Influenza A virus is a respiratory RNA virus that also provokes a systemic inflammatory response and clinical myalgia, so running it in parallel separates whatever is specific to SARS-CoV-2 from what any acute respiratory infection produces. That comparison is what makes the neuronal versus interferon distinction in the transcriptomes interpretable, and it is what establishes that the persistent phenotype is not simply a consequence of having been acutely ill.
The second is material against material. Quantitative PCR for viral transcripts, plaque assay for infectious virus, in situ hybridisation for Spike RNA and immunohistochemistry for nucleocapsid protein are applied to the same tissues, so the question of whether virus replicates in sensory tissue or only deposits RNA there can be answered rather than assumed. Lung serves as the internal positive control for both the plaque assay and the protein stain.
The third is acute against chronic. Tissues and behaviour were sampled at 1 and 4 days, when viral RNA and interferon signalling are present, and again at 28 to 31 days, after clearance, which is the interval relevant to persistent symptoms. Both sexes were included at the late time point.
Target identification then proceeds by computational prediction followed by pharmacological test in a different species and different injury contexts. Upstream regulator analysis was filtered deliberately, retaining regulators predicted to be less active at time points where hypersensitivity was lower, on the reasoning that inhibiting such a regulator might be analgesic, and the selection of ILF3 among the candidates was driven by the availability of a systemically deliverable and clinically tested inhibitor. Testing that inhibitor in complete Freund's adjuvant inflammation, in interferon beta injection and in paw incision separates a general analgesic effect from one specific to interferon-driven sensitisation, and paired locomotor and weight measurements guard against mistaking sedation or illness for analgesia.
Viral RNA and an interferon-stimulated transcript rise transiently in sensory tissue. Nucleocapsid transcripts and Isg15 increased at 1 day after infection in cervical dorsal root ganglia and spinal cord and in thoracic dorsal root ganglia and spinal cord, and were cleared or returned toward baseline in most samples by 4 to 7 days (Figure 1A to 1H).
No infectious virus reaches sensory tissue. Plaque assay at 3 days recovered virus from lung homogenates of infected animals only, and from no dorsal root ganglion or spinal cord sample (Figure 1I).
Spike RNA is present in dorsal root ganglia and spinal cord without detectable viral protein. In situ hybridisation at 1 day showed Spike puncta around DAPI-labelled nuclei, consistent in dorsal root ganglia with satellite glial cells, and in Rbfox3-labelled neuronal spaces, absent in mock animals (Figure 2A and 2B). Nucleocapsid protein was not notably detected in dorsal root ganglia while it was confirmed in lung (Figure 2C). These are representative images from two animals per group.
The two viruses produce different acute sensory time courses. Influenza produced robust mechanical hypersensitivity at 1 day that had completely subsided by 4 days, while SARS-CoV-2 produced a gradual decline in withdrawal threshold reaching significance only at 4 days, and influenza hypersensitivity at 1 day was significantly greater than that of SARS-CoV-2 at the same time (Figure 3A).
The two viruses produce different acute dorsal root ganglion transcriptomes. SARS-CoV-2 gave 344 differentially expressed genes at 1 day and 63 at 4 days, against 82 and 18 for influenza (Figure 3B). Pathway analysis placed axonal guidance signalling and synaptogenesis signalling as the top enriched pathways for SARS-CoV-2 at 1 day and neuroinflammation signalling among the top five at 4 days, while influenza pathways were consistently generic antiviral (Figure 3C). Shared upregulated genes between the two viruses were primarily antiviral (Figure 3D). Quantitative PCR confirmed bidirectional regulation of neuropathy-associated and pronociceptive genes including Sema3b, Vegfa and Rgs4 at 1 day and Mx1, Irf7, Slc6a4 and Rgs18 at 4 days (Figure 3E).
Hypersensitivity returns after clearance and only for SARS-CoV-2. At 28 days, well after viral clearance, SARS-CoV-2 infected hamsters of both sexes showed substantial mechanical hypersensitivity while influenza and mock animals were normal (Figure 5A).
A neuropathic transcriptome emerges late. At 31 days, thoracic dorsal root ganglia from SARS-CoV-2 infected animals showed 1065 differentially expressed genes, 170 up and 895 down (Figure 5B), with decreased synaptogenesis signalling and involvement of EIF2, mTOR, opioid and SNARE signalling pathways (Figure 5C). Disease association analysis linked these genes primarily to neuro-oncological and neurodegenerative conditions (Figure 5D), and affected transcripts included tubulin isoforms, myelin proteins, activity-related channels, extracellular matrix proteins and cytokine and interferon-related proteins (Figure 5E).
Cell composition inference suggests a shift. Deconvolution of the 31-day data indicated reduced transcriptomic contribution from myelinating Schwann cells and increased contribution from macrophages (Figure 5F). The authors read this as suggesting a proinflammatory state with potentially impaired myelination, which is an inference from bulk deconvolution rather than a direct cell count.
Peripheral and central pain-relevant tissues largely counter-regulate. Comparing 31-day dorsal root ganglion upstream regulators with striatum and thalamus data from the same model in prior work, most of the top 15 regulators showed similar predicted states in the two brain regions and generally opposite states in dorsal root ganglia, with PTPRR and MIR17HG commonly upregulated and FIRRE commonly downregulated between dorsal root ganglia and thalamus (Figure 5G).
ILF3 was nominated and validated pharmacologically. Nine upstream regulators met the filtering criterion, and ILF3 was selected from among those not previously studied in pain because an inhibitor, YM155, exists. Ilf3 transcript itself did not change by whole-tissue quantitative PCR, which the authors read as indicating the change is at the protein or localisation level. YM155 at 5 milligrams per kilogram raised thermal and mechanical withdrawal thresholds in complete Freund's adjuvant treated mice both acutely and with sustained effect over days (Figure 4C to 4F), prevented interferon beta induced mechanical hypersensitivity when given from the time of injection and reversed it when started a day later at a lower dose (Figure 4I and 4J), and reduced hypersensitivity after paw incision when given prophylactically (Figure 4K), without changes in locomotion beyond 30 minutes or in weight (Figure 4B, 4G, 4L). Toxicity was lethal at 20 milligrams per kilogram and variable at 10, and male mice were more sensitive than female mice at 5.
Comparison against published pain model datasets showed genes commonly upregulated between SARS-CoV-2 and complete Freund's adjuvant at 1 and 4 days and between SARS-CoV-2 and spared nerve injury at 1 day, a shared extracellular matrix remodelling set upregulated across all three acute conditions, and a strong counter-regulation between spared nerve injury and 31-day SARS-CoV-2 in genes related to nervous system development, myelination and synaptic transmission (Figure 6A to 6D).
The study does not establish a mechanism linking viral material in sensory tissue to persistent hypersensitivity, and the authors are explicit that the association is correlative. Their summary states that the presence of viral material and the resulting interferon response and gene expression changes correlated with the prolonged hypersensitivity signature.
What the data constrain is the following. SARS-CoV-2 RNA reaches cervical and thoracic dorsal root ganglia and spinal cord within a day of intranasal infection, in both neuronal and satellite glial compartments, in the absence of recoverable infectious virus and without detectable nucleocapsid protein, so local replication is not supported and the material is best described as disseminated viral RNA. The acute transcriptional consequence in dorsal root ganglia is predominantly neuronal in character and differs from the predominantly interferon-driven response to influenza, even though both viruses provoke systemic inflammation, so the difference is not simply a matter of how much inflammation occurs. The late phenotype, both behavioural and transcriptional, appears after viral RNA has cleared and is specific to SARS-CoV-2 among the two viruses tested.
The interpretive hypothesis the authors advance, that certain acute SARS-CoV-2 transcriptional changes counteract interferon-induced sensitisation through a stronger neuronal adaptation signature and thereby produce the milder acute phenotype, is stated as a hypothesis that motivated the pathway analysis rather than as a result. Likewise the suggestion that the late phenotype reflects altered excitability, cytoskeletal architecture, extracellular remodelling and myelination is presented as what the transcriptome and deconvolution may indicate.
For ILF3 the causal claim is narrower and better supported but also displaced from the virus. YM155 reduces hypersensitivity in three mouse models including one driven purely by interferon beta, which supports ILF3 activity as a contributor to interferon-associated sensitisation. It was not tested in infected hamsters, so ILF3 inhibition is not shown to relieve SARS-CoV-2 associated hypersensitivity. The authors also note that YM155 is thought to act on ILF3 subcellular localisation rather than on its expression, and Ilf3 transcript was unchanged in the tissue.
The work establishes that a purely airway infection can leave a durable transcriptional and behavioural mark on the peripheral sensory nervous system without any infectious virus reaching that tissue, which reframes post-viral sensory symptoms as a consequence of disseminated viral material and host response rather than requiring neuroinvasion. Comparing two respiratory RNA viruses in the same model and at the same time points is what allows a virus-specific claim to be made at all. The authors also propose the hamster respiratory SARS-CoV-2 model as a preclinical chronic pain model suitable for evaluating pharmacological treatments, and the ILF3 result illustrates the pipeline they intend, moving from an upstream regulator prediction in infected tissue to a validated analgesic effect in conventional injury models.
The hamster model, the infection protocol and the longitudinal transcriptomic framing follow directly from the laboratory's prior work on systemic and brain consequences of SARS-CoV-2 in the same model, and the striatum and thalamus datasets used in the cross-tissue comparison come from that earlier study. The paper extends the same question, what a respiratory infection does to distal tissue in the absence of infectious virus there, from the central nervous system to the peripheral sensory nervous system. The tenOever contribution covers study design, the infection work, virology and supervision, with Frere as co-first author, while the sensory neuroscience, behavioural pharmacology and mouse pain models come from the Zachariou laboratory with Serafini as co-first author, and the two are joint corresponding authors and joint supervisors.
Category 3 synthesis. Read together with the laboratory's earlier hamster brain work and with its influenza work, this paper contributes to a recurring question in the corpus about how a localised respiratory infection produces effects in tissues the virus does not productively infect. That is a cross-paper theme and this study alone establishes it only for dorsal root ganglia and spinal cord.
The infection model is the golden hamster, and all of the infection biology, transcriptomics and infection-associated behaviour come from that species, while every pharmacological test of ILF3 inhibition was performed in mice using non-infectious injury models, so the therapeutic claim is not demonstrated against SARS-CoV-2 associated hypersensitivity in any animal. Acute transcriptomic and imaging work used male hamsters only, with females included at the 31-day behavioural time point, and the 31-day sequencing was male only. Group sizes are small throughout, with n of 3 to 8 for quantitative PCR, n of 4 for acute sequencing, n of 3 for 31-day sequencing, n of 4 per group for the acute behaviour and n of 2 per group for the in situ hybridisation and immunohistochemistry images. Absence of infectious virus rests on plaque assay at a single time point, 3 days, and absence of viral protein on immunohistochemistry at 1 day, so neither excludes low-level or transient events below assay sensitivity. Sensory assessment is limited to mechanical withdrawal threshold in hamsters, with thermal testing performed only in the mouse models, and biosafety constraints required a linear perceived intensity scale rather than standard logarithmic force measurements for hamsters, which the authors state was done to accommodate a low number of animals at high variable forces. Pathway, upstream regulator and disease association results are predictions from bulk tissue transcriptomes, and the cell composition shift is a deconvolution inference rather than a measured change in cell numbers. Ilf3 transcript did not change, so the upstream regulator prediction was not confirmed at the level of the molecule itself in the infected tissue. YM155 is a single compound with documented toxicity at higher doses and sex-dependent sensitivity, and its selectivity for ILF3 in these assays was not independently established. The cross-species meta-analysis compares hamster infection data against mouse injury data generated elsewhere. Only one SARS-CoV-2 isolate and one influenza strain were used, and the latest time point examined is 31 days.
A nose-only SARS-CoV-2 infection in hamsters left viral RNA but no live virus in sensory nerve tissue, and produced touch sensitivity lasting a month after clearance.
Many people report lingering pain and altered sensation after COVID-19. In hamsters infected through the nose, viral RNA but no infectious virus appeared in sensory nerve clusters within a day. Influenza caused sharp, brief touch sensitivity, while SARS-CoV-2 caused milder sensitivity that grew and then returned a month later, long after the virus was gone. Gene activity in those nerve clusters shifted toward a nerve-injury pattern, and a drug targeting one predicted regulator relieved pain in mice.
Intranasal SARS-CoV-2 in golden hamsters raised viral nucleocapsid transcripts and Isg15 in cervical and thoracic dorsal root ganglia and spinal cord by one day, resolving by four to seven days, while plaque assay recovered infectious virus only from lung and nucleocapsid protein was undetectable in neural tissue. In situ hybridisation placed Spike RNA in both neuronal and satellite glial compartments. Influenza A virus produced strong mechanical hypersensitivity at one day that resolved by four, whereas SARS-CoV-2 produced milder hypersensitivity that reached significance at four days and reappeared at 28 days in both sexes. Acute dorsal root ganglion transcriptomes were neuronal in character for SARS-CoV-2 and interferon-dominated for influenza, and at 31 days SARS-CoV-2 produced 1065 differentially expressed genes with reduced myelin and synaptogenesis signatures and increased macrophage contribution by deconvolution. Upstream regulator analysis nominated ILF3, whose inhibitor reduced hypersensitivity in three mouse pain models, though not in infected animals.
co-led. Frere 2022 and Serafini 2023 are co-corresponding with the Zachariou laboratory, Zazhytska 2022 was led with the Lomvardas and Overdevest groups, and Heaton 2014 is co-led with Peter Palese and tenOever as joint senior and corresponding authors