Prolonged and extended impacts of SARS-CoV-2 on the olfactory neurocircuit

The impact of SARS-CoV-2 on the olfactory pathway was studied over several time points using Syrian golden hamsters. We found an incomplete recovery of the olfactory sensory neurons, prolonged activation of glial cells in the olfactory bulb, and a decrease in the density of dendritic spines within the hippocampus. These data may be useful for elucidating the mechanism underlying long-lasting olfactory dysfunction and cognitive impairment as a post-acute COVID-19 syndrome.

www.nature.com/scientificreports/ revealed that SARS-CoV-2 RNA and viral antigen can be detected in the brain 12 . Moreover, we found that the OE thickness transiently decreased at 5 dpi but recovered fully by 21 dpi, as was the case for the SC numbers (Figs. 1c, 3a,b). Nevertheless, the density of mature OSNs did not completely recover up to 42 dpi, suggesting that the maturation of OSNs may be delayed and/or incomplete (Fig. 3a,c).
Our data indicate that the DM is the most vulnerable to SARS-CoV-2 in the OE. This is reasonable, as ACE2 receptors are predominantly expressed in this region 13 . The DM is considered to be included in the zone 1 of the www.nature.com/scientificreports/ zonal categorization 8 . Interestingly, the OSNs in zone 1 only express the endocellular enzyme NAD(P)H quinone oxido-reductase 1 (NQO1) 14 . NQO1 acts as an antioxidant but also mediates ROS generation, being involved in increasing certain forms of neuronal damage after injury 9 . This led us to analyze the relationship between NQO1 expression patterns and neural damage caused by SARS-CoV-2 infection. The numbers of OMP-positive cells decreased after infection and were not completely restored in both NQO1-positive and -negative regions ( Fig. 4a-c). However, the damage was more severe in NQO1-positive regions than in NQO1-negative regions (Fig. 4d). Furthermore, the presence of macrophages, which are labeled using the ionized calcium binding adaptor molecule 1 (Iba1), suggest prolonged activation in the lamina propria within NQO1-positive regions ( Fig. 4e-g). In contrast, macrophages quickly returned to a normal level within NQO1-negative regions. These data suggest that NQO1-positive areas were more susceptible to SARS-CoV-2 than the negative areas. Such uneven damages to the OE may induce the unusual pattern of odor-induced activity in the OB and contribute to development of parosmia during recovery process. These trends were also observed in the next olfactory relay center, the OB, in which the OSNs have synaptic contact with multiple neurons within the glomerulus (Fig. 5a,b). In the OB, the density of the OSN axon terminal of each glomerulus was significantly decreased within NQO1-positive regions, but not in the NQO1-negative regions (Fig. 5c,d). Interestingly, the size of the glomeruli themselves decreased not only in NQO1-positive but also in NQO1-negative regions (Fig. 5e,f), suggesting that the multiple mechanisms may be in play that impacts OB damage. These data indicate that SARS-CoV-2 infection impacts odor information processing within the whole OB, but especially prominent in NQO1-positive regions. Next, we carefully examined the effects of SARS-CoV-2 on the profile of glial activities (Figs. 6a, 7a, 8a) throughout the multiple layers of the OB: the olfactory nerve layer (ONL) where the OSN axon shaft are densely packed, the glomerular layer (GL) where the OSN axon synapse to the relay neurons, and the external plexiform layer (EPL) where the projection neurons interact with the local interneurons. Remarkably, the Iba1 signal peaked at 17 dpi (Fig. 6b), in contrast to the OE, in which it peaked at 5 dpi, indicating that microglia in the OB became activated more slowly than macrophages in the OE. Interestingly, the ONL, occupied by OSN axons, showed significantly strong microglia activation, even at 5 dpi. This indicates that the microglia in the ONL respond to the damage of the OSN axons from the OE quickly and directly, and this damage would provide an indirect secondary impact on the postsynaptic circuits in the GL and EPL. Consistent with the impact of SARS-CoV-2 on the OSN axon terminals within the OB (Fig. 5), the impact on the NQO1-positive regions was significantly stronger within the ONL at 5 dpi and within the GL at 17 dpi, as compared to that of the NQO-1-negative www.nature.com/scientificreports/ regions (Fig. 6b,d,f). However, the area occupied by glial fibrillary acidic protein (GFAP)-expressing reactivated astrocytes was enlarged at 17 dpi within the GL (Fig. 7b,c,e), although the dynamics of GFAP-positive cells were not different based on NQO1 expression (Fig. 7d,f). This indicates that astrocyte activity is strongly associated with the activity of microglia in the GL compared to that in other layers. Next, we examined the impact of SARS-CoV-2 on higher brain areas, including the piriform cortex (PC) and the hippocampus (Figs. 8,9). In the PC, which is one of the olfactory cortices receiving direct inputs from the OB, GFAP-expression appeared transient in the pia mater but persisted in the layer 1 of the PC up to 42 dpi (Fig. 8b). Activated macrophages labeled by Iba1 within the pia mater increased in numbers at 5 and 17 dpi, although the Iba1-positive area was unchanged in layer 1 (Fig. 8c,d). The peak of this activity was not correlated to the presence of SARS-CoV-2 within the nasal cavity (Fig. 1a), indicating that glial activation in the PC may not be induced directly by the virus through the olfactory pathway but by local and/or systemic inflammatory responses. In fact, inflammatory cytokines are systemically elevated in both severe and non-lethal COVID-19 models 15 . Given that ACE2 is expressed in the Bowman's glands of the lamina propria 16 and that macrophages there were activated after SARS-CoV-2 infection (Fig. 4f,g), the immune response in the lamina propria may induce the elevation of intravascular inflammatory cytokines, triggering the activation of glial cells and macrophages in the CNS. In the hippocampus, GFAP-positive astrocytic endfeet were easily detectable around the blood vessels of the apical dendritic region at 5, 8, 17, and 42 dpi (Fig. 9b), as compared to those of the basal region. The peak Iba1-signal in the basal region was achieved by 8 dpi, with the area being restored by 42 dpi (Fig. 9c,d). In contrast, the Iba1-signal in the apical region remained strong until 42 dpi (Fig. 9c,d,e). These data indicate that SARS-CoV-2 infection in the nostril triggered the activation of microglia and astrocytes, even in the hippocampus, and that the www.nature.com/scientificreports/ impacts are significantly different in individual layer. Thus, an interesting question could be posed as to whether the activated microglia and astrocytes could induce any changes in the neuronal circuits? There are many reports that reveal glial cells induce synaptic modulation 17 , synaptic loss 10 , synaptic plasticity 18 , and change of synaptic density 19 . These changes may be associated with dementia 20 . Therefore, we analyzed dynamics of spine density within the hippocampus using Golgi stained brain sections (Fig. 9f,g). Basal dendritic spines in the hippocampal CA1 were extremely stable. In contrast, the density of apical dendritic spines was significantly decreased at 42 dpi (Fig. 9h). This finding may be associated with the prolonged activation of microglial cells, most notably significant at 42 dpi (Fig. 9e). These results suggest that intranasal inoculation of SARS-CoV-2 induces glial cell activation and changes spine density within the higher brain regions, including the hippocampus. These may underlie long-lasting sequela of SARS-CoV-2 infection including depression, memory impairments, and brain fog 26 , although more evidences to show synaptic dysfunction should be acquired. Future www.nature.com/scientificreports/ studies using behavioral battery test for cognitive function, should determine whether these changes impact animal behavior and if so, how long it is required for recovery. In summary, we report that single priming with SARS-CoV-2 resulted in long-lasting effects in hamsters, not only on the OE but also on the brain regions critical for cognitive function. We found drastic CNS changes including glial activation and synaptic dynamics, without any SARS-CoV-2 antigen by immunostaining. Our study may be of importance for better understanding the mechanism(s) driving the long-lasting olfactory dysfunction and cognitive impairment as a post-acute COVID-19 syndrome.  Animal experiments. Six-week-old female Syrian golden hamsters were purchased from Charles River.

GFAP NQO1 Hoechst
Syrian golden hamsters were anesthetized with isoflurane and inoculated bilateral-intranasally with 10 5 50% tissue culture infectious dose (TCID50) of SARS-CoV-2 diluted in 100 µl of phosphate-buffered saline (PBS) or PBS as a mock control. The anesthetized hamsters were held vertically, and virus-containing PBS was intranasally inoculated according to breathing rhythm by micropipette 4,5,26 . Mock controls were sacrificed 42 days after inoculation. All hamsters were housed in the animal biosafety level-2 (ABSL-2) and ABSL-3 facilities within the Galveston National Laboratory at the University of Texas Medical Branch (UTMB). All animal studies are reviewed and approved by the Institutional Animal Care and Use Committee at UTMB and are conducted according to the National Institutes of Health guidelines. This study is reported in accordance to ARRIVE guidelines (https:// arriv eguid elines. org/ arrive-guide lines).
Immunohistochemistry. Hamsters were euthanized using a high-flow rate of CO 2 followed by thoracotomy at each time point to collect samples. The heads of hamsters were fixed in 10% buffered formalin for 7 days before removal from the BSL-3 facility and followed by incubation in phosphate buffered saline (PBS). After OBs were extracted from the skull, right OBs were sectioned into 100 µm thick sections using a vibratome and left OBs were proceeded to the tissue Golgi-Cox procedures. The nasal tissues including the olfactory epithelium were decalcified with 10% EDTA (pH 7.0) for 7 days at 37 °C with gentle shaking. Decalcification samples were dehydrated in 100% ethanol three times and embedded in paraffin. Serial coronal sections (5 µm thick) were cut Golgi-Cox staining. To quantify the number of dendritic spines, brain slices were treated with the Rapid GolgiStain Kit according to the manufacturer's instructions (ND Neurotechnologies, Inc., PK401) with minor modifications. Since sample conditions, especially after formalin fixation, were significant and critical variables for Golgi staining [28][29][30][31] , modified Golgi-Cox staining was performed as described in previous reports 31,32 . Briefly, the brain slices were incubated in the impregnation solution for 14 days in 37 °C. The impregnation solution was mixed using a 1:1 ratio of Solutions A and B, which contain mercuric chloride, potassium dichromate, and potassium chromate. The solution was prepared 24 h before use, to allow precipitate formation, and covered with aluminum foil and kept in 37 °C. The sample was replaced in a sucrose-based solution (solution C) for 10 min at room temperature. The sections were transferred and agglutinated onto gelatin-coated slides. The sections were rinsed twice for 5 min each in distilled water to remove the impregnation solution. The sections were then incubated in the staining solution (Solution D: Solution E: distilled water, 1:1:2) for 10 min at room temperature.

Analysis of immunohistochemistry.
To minimize the anatomical variations of the nose, three coronal sections covered with olfactory neuroepithelium were analyzed every 500 µm intervals. The OE consists of three cell types: olfactory sensory neurons, supporting cells, and basal stem cells. For each OE, coronal sections of the OE were divided into 4 regions: dorsomedial (DM), dorsolateral (DL), ventromedial (VM) and ventrolateral (VL). The thickness of olfactory epithelium was measured between the surface of neuroepithelium and the line of basal cells. We counted the numbers of OMP-positive cells for mature olfactory sensory neurons. To evaluate the numbers of supporting cells (SCs), we defined and counted the SCs as the columnar as the cells located proximal to the nasal cavity, as indicated in a previous report 33 . To investigate the presence of macrophages in the lamina propria, Iba1-positive cells were counted. All analyses were performed for each group per 100 µm OE length. Three regions were randomly selected, with the regions being at least 100 µm apart, and the averaged values were used. To quantify the densities of the Iba1-and GFAP-positive area, we defined the positive area as one that exceeds two SDs of the mean background intensity. Subsequently, binary images were overlapped onto the original images and the densities were automatically measured using ImageJ software. OMP-positive cells (Fig. 4d), Iba1-postive areas (Fig. 6d,f,h), OMP-positive areas (Fig. 5d), and GFAP-positive areas (Fig. 8d,f) were adjusted to normalize the average values in the NQO1-negative. Iba1-positive areas in the hippocampus (Fig. 9e) were adjusted to normalize the values in the basal region.
Analysis of dendritic spine density. The location of CA1 was identified using a previously published brain map 34 . Pyramidal neurons in the CA1 were selected and three apical and three basal dendritic segments were analyzed 35 . For analysis of Golgi-impregnated neurons, the criteria are the following, according to a previous report 36 : (1) dendrites with a length of at least 50 µm, (2) consistent and dark impregnation along the entire www.nature.com/scientificreports/ extent of dendrites, and (3) relative isolation from neighboring impregnated dendrites. Densities of dendritic spines were calculated as the mean numbers of spines per 10 µm per dendrite per neuron in individual animals per group.

Statistical analysis.
To prevent arbitrary analysis, all data were randomly and blindly analyzed by two trained human operators. R (https:// www.r-proje ct. org/) and custom-written Python scripts were used for statistical analysis. Mean ± SD was used to indicate biological variations. In Figs. 4d, 6d, f, h, and 9e, and Figs. 5d,e, 7d,f, Welch's t-tests were used. Compared to mock groups, comparisons among multiple groups (Figs. 1c, 4c, g, 6c, e, g, and 9d, and Figs. 3b,c, 5c,e, 7c,e, 8c, d) were evaluated with one-way ANOVA and followed by Dunnett's post hoc test. In Fig. 9h, one-way ANOVA and followed by Tukey's post hoc test were performed. P-values for figures are depicted as follows: *p < 0.05, **p < 0.01, ***p < 0.001 (Supplementary Information).