COVID-19 induces neuroinflammation and loss of hippocampal neurogenesis

Infection with the Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) is associated with onset of neurological and psychiatric symptoms during and after the acute phase of illness1–4. Acute SARS-CoV-2 disease (COVID-19) presents with deficits of memory, attention, movement coordination, and mood. The mechanisms of these central nervous system symptoms remain largely unknown. In an established hamster model of intranasal infection with SARS-CoV-25, and patients deceased from COVID-19, we report a lack of viral neuroinvasion despite aberrant BBB permeability, microglial activation, and brain expression of interleukin (IL)-1β and IL-6, especially within the hippocampus and the inferior olivary nucleus of the medulla, when compared with non-COVID control hamsters and humans who died from other infections, cardiovascular disease, uremia or trauma. In the hippocampus dentate gyrus of both COVID-19 hamsters and humans, fewer cells expressed doublecortin, a marker of neuroblasts and immature neurons. Despite absence of viral neurotropism, we find SARS-CoV-2-induced inflammation, and hypoxia in humans, affect brain regions essential for fine motor function, learning, memory, and emotional responses, and result in loss of adult hippocampal neurogenesis. Neuroinflammation could affect cognition and behaviour via disruption of brain vasculature integrity, neurotransmission, and neurogenesis, acute effects that may persist in COVID-19 survivors with long-COVID symptoms.


Hamsters intranasally infected with SARS-CoV-2 and patients deceased from acute COVID-19 do not exhibit viral neuroinvasion
The Golden Syrian hamster is naturally susceptible to SARS-CoV-2 infection. Intranasal inoculation results in mild-to-moderate disease with labored breathing, ru ed fur, weight loss, and hunched posture 5. To assess acute neuroin ammation following SARS-CoV-2 infection, 5-6-week-old male hamsters were infected intranasally with 2 x 10 5 plaque forming units (PFU) of a fully infectious SARS-CoV-2 isolate (strain 2019-nCov/USA-WA1/20202). Whole heads of uninfected and infected hamsters were collected throughout the acute infectious period and at one week after viral clearance, which occurs in the lungs at 5-7 days post-infection (dpi) 18 . High levels of SARS-CoV-2 RNA was detected within the hamster ethmothurbinates at 2-4 days post infection (dpi), and completely cleared by 8 dpi (Fig. 1a,b). As previously reported 19 , viral RNA was detected only within K18 + sustentacular cells of the olfactory neuroepithelium (ONE) (Fig. 1c), which might impact olfactory sensory neuron (OSN) function via loss of calcium signaling 20 . SARS-CoV-2-infected sustentacular cells exhibited decreased expression of K18 + compared with uninfected ONE (Supplementary Video 1) and were found sloughed off into the nasal cavities of infected hamsters (Supplementary Video 2). Consistent with lack of OSN infection and neuroinvasion via the olfactory route, no viral RNA was detectable in the acute and recovered hamster olfactory bulb (OB), cortex, hippocampus, and medulla oblongata at any timepoint post-infection ( Supplementary Fig. 1a).
Consistently with our hamster ndings, we previously assessed 41 COVID-19 subjects who underwent autopsy and neuropathological exam, which we reported to show no viral RNA in brain tissue, as detected by RNAscope or RT-PCR 14 (Supplementary Fig. 1b).

SARS-CoV-2 infection is associated with widespread BBB disruption
Disruption of the BBB may occur during infection with respiratory viruses 21 . To evaluate effects of SARS-CoV-2 on BBB integrity in the hamster model, we assessed brain levels of extravasated serum IgG via immunohistochemistry. At 3-4 dpi there was a signi cant increase in IgG in the CNS parenchyma compared to naïve animals, which gradually decreased by 14 dpi (Fig. 1d). Further assessment of IgG extravasation revealed regional differences in BBB permeability following infection. While all regions examined, which include OB, cortex, hippocampus, and medulla oblongata, showed some degree of BBB disruption, the hippocampus suffered the most signi cant changes compare to naïve hamsters (Fig. 1e). The medulla oblongata exhibited little BBB disruption at early timepoints, but signi cant detection of IgG at 5 dpi (Fig. 1e). Similar persistent alterations in IgG immunoreactivity were observed in the OB and cortex, albeit to a lesser degree when compared to the hippocampus ( Supplementary Fig. 2a).
We next examined BBB permeability in human COVID-19 brain tissue samples via detection of brinogen, a blood coagulation protein whose CNS deposition is implicated in a wide range of neurological disease and injuries associated with BBB disruption 22 . In a subset of COVID-19 decedents (n=7) compared with age and sex-matched controls (Table 1) deceased from other infections (n=3) or cardiovascular disease (n=2), we observed increased brinogen in the hippocampus (Fig. 1f), and a smaller non-signi cant increase in medulla (Fig. 1f) and OB ( Supplementary Fig. 2b). Together, these data suggest that SARS-CoV-2 infection may lead to region-speci c alterations in human BBB integrity.
SARS-CoV-2 infection is associated with aberrant expression of brain cytokines Loss of BBB integrity may permit CNS entrance of cytokines or immune cells, which, in turn, may activate glial cells 23 . Given the observed increased BBB permeability in SARS-CoV-2 infected hamsters and humans, we examined microglia activation and cytokine expression by glia and neurons.
Medulla oblongata from SARS-CoV-2-infected hamsters showed activated microglia within the inferior olivary nuclei (ION), displaying larger cell bodies and thicker processes than those detected in uninfected tissues, and increased levels of ionized calcium-binding adapter molecule 1 (IBA1) at 4 dpi, which remained elevated at 14 dpi (Fig. 2a). In vertebrates, ION are found in the medulla underneath the superior olivary nucleus, and coordinate signals from the spinal cord to the cerebellum, regulating motor coordination and learning via integration of glutamatergic synaptic inputs 24 . In the medulla ION, IBA1 + activated microglia of SARS-CoV-2-infected hamsters exhibited increased expression of IL-1β at 2-5 dpi, which returned to baseline by 8 dpi, compared to uninfected hamsters ( Fig.  2b). Unfortunately, there are no commercially available antibodies to detect IL-6 in hamster tissues.
Analysis of COVID-19 patients revealed increased microglial activation, and IL-1β expression within the ION compared with control patient tissues ( Fig. 2c-f). Elevated expression of IL-6 was detected in ION neurons, suggesting neuronal cytokine production (Fig. 2g), as previously reported 27 .
Hippocampal tissue from COVID-19 patients exhibited subtle changes in IBA1 + expression and microglial production of IL-1β compared with control patients (Fig. 3c-g). Similar to the medulla ION, we also observed neurons to be the main source of IL-6 in COVID-19 patient hippocampi compared to controls (Fig. 3i).
OB of both SARS-CoV-2-infected hamsters (2-5 dpi) and humans exhibited elevated expression of IL-1β within IBA1 + microglia compared with uninfected control tissues ( Supplementary Fig. 4a,c). In hamster somatosensory cortex, co-localization of IL-1β within IBA1 + microglia was not signi cantly elevated ( Supplementary Fig. 4b). In hamsters, OB and somatosensory cortex showed similar microglial activation, but this was not observed in OB of COVID-19 patients versus controls ( Supplementary Fig. 3a-d). However, OB from COVID-19 patients exhibited elevated neuronal expression of IL-6 compared with controls ( Supplementary Fig. 4d). IL-6 has been shown to have both pro-and anti-in ammatory roles 28 ; further investigation of its role in COVID-19 is needed. Altogether, these data suggest neurons are important players in controlling neuroin ammation in humans affected by SARS-CoV-2.
We determined whether astrocytes contribute to neuroin ammation in the CNS of infected hamster and humans. Using a SOX9 antibody, we observed no persistent changes in astrocyte cell numbers in OB, ION, and hippocampus of infected hamsters compared to naïve animals ( Supplementary Fig. 5). In COVID-19 human tissue samples, GFAP + expression was increased in the medulla compared to controls ( Supplementary Fig. 6a), and signi cant increases in IL-1β expression by GFAP + astrocytes were observed in the OB of COVID-19 patients when compared to controls. No differences in astrocytes were observed between COVID-19 and control patients in the hippocampus and medulla oblongata (Fig. 3h, Supplementary Fig. 6a).
These data suggest that, while astrocytes may be involved in post-infection neuroin ammation in the OB, a region proximal to the location of viral replication, microglia and neurons appear to be the main players in more remote brain regions, like the medulla and hippocampus. Taken altogether, our ndings show that, despite lack of viral neuroinvasion, SARS-CoV-2-infected individuals develop microglial activation and cytokine expression in brain regions associated with olfactory function, motor coordination, memory and learning, possibly inducing neuropsychiatric signs and symptoms.

Loss of hippocampal neurogenesis in hamsters and humans infected with SARS-CoV-2
The formation and consolidation of new memories occurs primarily within the hippocampus and relies on the integrity of a trisynaptic circuit between the entorhinal cortex, dentate gyrus (DG), and Cornu Ammonis (CA) 29 . Spatial learning, in particular, relies on the link between synapses within the CA3 region and rates of adult neurogenesis, which occurs via generation of new neurons from neural progenitor cells  37 , as well as DCX + /Ki67 + cells, that were almost completely absent in the SGZ at 5 dpi (Fig. 4a,b). Cell numbers normalized to pre-infection levels after day 5. There was no change in DCX + /Ki67 + cell numbers in the rostral migratory stream in infected hamster compared to uninfected controls ( Supplementary Fig. 7). These data suggest that effects of neuroin ammation and IL-1β on neurogenesis are speci c to the DG.
To address the relevance of these ndings in humans, we compared numbers of DCX + neuroblast and DCX + /NeuN + immature neurons in the SGZ and granule cell layer (GCL) of a subset of COVID-19 decedents (n=8), and age-and sex-matched non-COVID-19 controls (n=8, Table 1). Adult hippocampal neurogenesis has been shown to persist into adulthood in healthy humans 38,39 , with reports suggesting DCX + immature neurons are present into the ninth decade of life 40 . Because DCX + cells migrate from the SGZ into the GCL as they mature and start expressing NeuN 41 , DCX + /NeuNcells in the SGZ are more likely neuroblasts and DCX + /NeuN + cells located in the GCL are more likely immature neurons 42 . We quanti ed these populations separately in the SGZ and GCL (Supplementary Fig. 8), and found fewer DCX + /NeuNneuroblasts in the SGZ of COVID-19 patients compared to non-COVID-19 controls (Fig. 4c,d).
Conversely, DCX + /NeuNcells in the GCL and DCX + /NeuN + immature neurons in SGZ and GCL were not fewer in COVID-19 versus control subjects (Fig. 4c,d). In mice neurogenesis takes approximately four weeks, and in monkeys it takes six months 43 , which is the time-frame these mammals carry their pregnancies. Thus, human neurogenesis might take nine months, and the more mature DCX + /NeuN + cells, and DCX + /NeuNthat have already migrated into the GCL, might have been generated weeks or months before the infection. On the other hand, cytokine surge and hypoxia appear to affect the more immature DCX + /NeuNcells located in the SGZ, possibly halting NPC differentiation into neuroblasts or reducing neuroblast survival. These hypotheses could be further tested in animal models where cell lineage can be traced, or in human brain via RNA velocity 44 .
We detected no signi cant effect of age on the number of DCX + /NeuN + and DCX + /NeuNcells in SGZ and GCL in COVID-19 patients or non-COVID-19 controls ( Supplementary Fig. 9), in line with previous ndings on the persistence of adult hippocampal neurogenesis in older subjects with no chronic neuropsychiatric illness [38][39][40]45 .

Limitations
Limitations of this study include the small sample sizes, and lack of reagents for detecting neural cell markers and additional in ammatory factors in hamsters. COVID-19 and non-COVID patients suffered hypoxia, which has been associated with activation of microglia in the absence of viral infection, and can also affect DCX + cells 26 . Although most of our controls did not have a history of intubation, given their medical history of cardiovascular disease and infections, they likely experienced elevated cytokines and hypoxic damage. Hamsters develop milder disease without hypoxia, which is different from humans who died from COVID-19. Since hamsters exhibited similar CNS damage as humans, brain alterations are most likely the result of in ammation associated with COVID-19.

Discussion
This study identi es neurobiological mechanisms of CNS damage in SARS-CoV-2 infection. In hamster and humans, BBB disruption, cytokine expression, activated microglia and loss of hippocampal neurogenesis, may contribute to neuronal dysfunction/loss, and ultimately to neurocognitive or psychiatric symptoms. The absence of these brain changes in non-COVID patients deceased from other infections or cardiovascular disease suggests effects of neuroin ammation and hypoxia in COVID-19 patients may be speci c to SAR-CoV-2 infection.
Brain alterations were transient in hamsters, peaking after viral clearance in the nasal cavity. In humans, we do not know how long elevated in ammatory markers persist in subjects who survive the disease. The persistence of neuropsychiatric symptoms in long-COVID cases suggests that neuronal damage may be prolonged. Brain imaging studies investigating in ammation markers in post-COVID patient are warranted. Studies using positron emission tomography (PET) radiotracers for the brain translocator protein (TSPO) 46 located on microglia and astroglia, and transcranial near-infrared spectroscopy (NIRS) to assess mitochondrial function 47,48 , may reveal useful for gathering data on indices of brain in ammation levels in post-COVID patients.
In hamsters and humans, BBB disruption and cytokine expression by microglia, astrocytes, and neurons occurred in a region-speci c fashion and were not associated with viral neuroinvasion 14 . Therefore, cytokines appear to be the main mediator of BBB disruption and cellular damage. NPC IL-1β receptors have been implicated in reducing neurogenesis in murine models of avivirus encephalitis 34,50 . These viruses, however, lead to signi cant CNS in ltration of mononuclear cells, which has not been observed in studies of COVID-19.
Fewer DCX + neuroblasts in the SGZ of COVID-19 patients compared with non-COVID-19 controls suggests either decreased NPC maturation or increased neuroblast death. DCX expression has been hypothesized to occur in mature granule neurons that might be de-differentiating in pathological conditions 51 . However, this kind of cells would be more likely DCX + /NeuN + cell located in the GCL, which were unaffected. The time-course of human pathological ndings remains unknown, and if similar to hamsters, one would expect that after the initial cytokine surge, neurogenesis might recover, as cognitive symptoms and anosmia subside in many patients. Nevertheless, long-COVID symptoms have been widely reported 12 . There is the possibility that, in some individuals, the neurogenic niche might not have enough multipotent progenitors for neurogenesis to resume after this insult, as the multipotent progenitor pool of SOX2 + cells is smaller in aging humans 38 . If this was the case, some patients might not be able to recover, and COVID-19 may result in chronic neuropsychiatric symptoms, as observed in several clinical studies 2,8,13 .
Given the likely predominant role of neuroin ammation in the mechanism of brain damage in COVID-19, anti-IL-6 and anti-IL-1β therapies, currently under investigation 52  image of blood-brain permeability in the hamster brain 2 dpi, showing staining for IgG (green) and DAPI (blue), followed by quanti cation of IgG intensity in the CNS parenchyma (white outline). e.
Representative images of IgG detection (green) within hamster MO and hippocampi at naïve, 2, 3, 4, 5, 8, and 14 dpi, and nuclear stain, DAPI, (blue), followed by quanti cation of IgG intensity in their respective regions. f. Representative image of blood-brain permeability in the MO and hippocampus of control and COVID-19 patient tissue, depicting detection of brinogen (green) and DAPI (blue), followed by quanti cation of brinogen intensity. Data were pooled from at least two independent experiments. Scale bars, 50 μm (10x), 20 μm (20x) or 10 μm (63x). Data represent the mean ± s.e.m. and were analysed by two-way ANOVA or Student's t-test.  presented as microscopy with IBA1 (red), IL-1β (green) and DAPI (blue) and percent IL-1β+ area and IL-