key: cord-0825541-ih81b6l3
authors: Khan, Mona; Yoo, Seung-Jun; Clijsters, Marnick; Backaert, Wout; Vanstapel, Arno; Speleman, Kato; Lietaer, Charlotte; Choi, Sumin; Hether, Tyler D.; Marcelis, Lukas; Nam, Andrew; Pan, Liuliu; Reeves, Jason W.; Van Bulck, Pauline; Zhou, Hai; Bourgeois, Marc; Debaveye, Yves; De Munter, Paul; Gunst, Jan; Jorissen, Mark; Lagrou, Katrien; Lorent, Natalie; Neyrinck, Arne; Peetermans, Marijke; Thal, Dietmar Rudolf; Vandenbriele, Christophe; Wauters, Joost; Mombaerts, Peter; Van Gerven, Laura
title: Visualizing in deceased COVID-19 patients how SARS-CoV-2 attacks the respiratory and olfactory mucosae but spares the olfactory bulb
date: 2021-11-03
journal: Cell
DOI: 10.1016/j.cell.2021.10.027
sha: 8757b1317aa056f41353393b9455acf47288bdaa
doc_id: 825541
cord_uid: ih81b6l3

Anosmia, the loss of smell, is a common and often the sole symptom of COVID-19. The onset of the sequence of pathobiological events leading to olfactory dysfunction remains obscure. Here, we have developed a postmortem bedside surgical procedure to harvest endoscopically samples of respiratory and olfactory mucosae and whole olfactory bulbs. Our cohort of 85 cases included COVID-19 patients who died a few days after infection with SARS-CoV-2, enabling us to catch the virus while it was still replicating. We found that sustentacular cells are the major target cell type in the olfactory mucosa. We failed to find evidence for infection of olfactory sensory neurons, and the parenchyma of the olfactory bulb is spared as well. Thus, SARS-CoV-2 does not appear to be a neurotropic virus. We postulate that transient insufficient support from sustentacular cells triggers transient olfactory dysfunction in COVID-19. Olfactory sensory neurons would become affected without getting infected.

Olfactory dysfunction was recognized early in the COVID-19 pandemic (Eliezer et al., 2020; Lüers et al., 2020; Vaira et al., 2020b) , and is a strong and consistent symptom associated with a positive COVID-19 test (Sudre et al., 2021) . Well into the second year of the pandemic (Wang et al., 2020) , there is no explanation in sight as to how SARS-CoV-2 mutes or alters the sense of smell (Lechien et al., 2021; Vaira et al., 2021; Whitcroft and Hummel, 2020; Xydakis et al., 2020 Xydakis et al., , 2021 ). An unresolved question is whether the olfactory nerve can provide SARS-CoV-2 with a route of entry to the brain (Butowt et al., 2021) .

Soon after SARS-CoV-2 made its entry on the scene, the expression patterns of the virus cell entry genes ACE2 and TMPRSS2 were characterized in the human and mouse olfactory system (Brann et al., 2020; Fodoulian et al., 2020) . The inference was drawn that sustentacular cells but not olfactory sensory neurons (OSNs) might be the susceptible cell type in the olfactory epithelium (OE) (Cooper et al., 2020) . But, puzzlingly, two of the other six human coronaviruses, SARS-CoV (Fung and Liu, 2019) and the endemic HCoV-NL63 (van der Hoek et al., 2004; Hofmann et al., 2005) , also use ACE2 for cell entry but do not commonly cause olfactory dysfunction (Zugaj et al., 2021) . SARS-CoV-2 replication in sustentacular cells of COVID-19 patients remains to be demonstrated.

Historically, histological and molecular studies of normal and diseased human olfactory mucosa (OM) and olfactory bulb (OB) have been few and far between. Harvesting samples of suitable quality and unambiguous identity has proved problematic, both from living and deceased patients. Macroscopically the OM cannot be distinguished from the respiratory mucosa (RM). Anatomically the OM is made up of an archipelago of islands of various sizes scattered amidst RM high up in the nasal cavity within the olfactory cleft (de Rezende Pinna et al., 2013; Engström and Bloom, 1953; Escada, 2013; Kachramanoglou et al., 2013; Kern, 2000; Naessen, 1970; Salazar et al., 2019) . In contrast to the OM in laboratory rodents, the human OM is not a uniform sensory sheet (Morrison and Costanzo, 1990, 1992) . Patches of aneuronal epithelium are intercalated with patches containing abundant OSNs in human OE (Holbrook et al., 2005 (Holbrook et al., , 2011 Tanos et al., 2017) . The OM consists of OE and lamina propria J o u r n a l P r e -p r o o f 5 (LP), and these two layers are bound tightly together by a basal lamina. Simply put, it is not possible to collect samples of pure human OM, let alone of pure human OE. Furthermore, OB biopsies cannot be taken from living patients due to the intracranial position and debilitating consequences of the intervention. Harvesting OM and OB in the conventional setting of an autopsy is often feasible only after a long postmortem interval (PMI), particularly in COVID-19 patients who may still be contagious (Gagliardi et al. 2021; Matschke et al., 2020) . Analysis of samples procured after long PMIs is clouded by limitations resulting from autolysis of cells and tissues (Meinhardt et al., 2021) .

We reasoned that, to achieve a drastic reduction of the PMI, tissue harvesting best be performed bedside, soon after death. We further reasoned that, to investigate how the sequence of pathobiological events leading to olfactory dysfunction is initiated, the study design must accommodate the inclusion of patients in an acute phase of the infection, enabling us to catch the virus as it strikes.

Here, we have developed a postmortem bedside surgical procedure, which we adapted from an endoscopic technique of skull base surgery, to harvest RM and OM tissue samples and whole OBs. We visualized how SARS-CoV-2 attacks the olfactory system by combining the RNAscope platform of ultrasensitive single-molecule fluorescence in situ RNA hybridization with fluorescence immunohistochemistry (IHC). We identified ciliated cells in the RM and sustentacular cells in the OM as the major target cell types for SARS-CoV-2 replication in the nasal mucosa. A subset of cases showed viral RNA in the leptomeningeal layers surrounding the OB, but invariably the OB parenchyma was spared from infection. The absence of evidence for infection of OSNs and of OB neurons suggests that SARS-CoV-2 is not a neurotropic virus. We postulate that infected sustentacular cells transiently provide insufficient support to OSNs, structural and/or physiological.

J o u r n a l P r e -p r o o f 6

We designed a 24/7 workflow initiated by a health care worker of an intensive care unit (ICU) or a ward placing a phone call to a team of Ear, Nose and Throat (ENT) physicians shortly after death of a COVID-19 patient ( Figure 1A ). The ENT team wore personal protective equipment (Van Gerven et al., 2020) and performed an endoscopic surgical procedure at the bed of the deceased patient with a preassembled mobile unit consisting of a monitor, light source, camera, and endoscopic equipment. This concept was the foundation of a clinical study called ANOSMIC-19, ANalyzing Olfactory dySfunction Mechanisms In COVID-19. We included a cohort of 68 patients who died from or with COVID-19 in the University Hospitals

Leuven (Leuven, Belgium) or in the General Hospital Sint-Jan Brugge-Oostende AV (Bruges, Belgium) between May 2020 and April 2021 ( Figure 1B ). In parallel we included 15 control patients and two convalescent COVID-19 patients who died in a hospital several months after recovering ( Figure S1 ). Our cohort of COVID-19 cases is representative of the rather uniform phenotype of deceased COVID-19 patients (Patel et al., 2021; Van Aerde et al., 2020) : predominantly men suffering from multiple comorbidities, most commonly obesity or overweight, diabetes mellitus type 2, and hypertension.

We adapted the postmortem bedside surgical procedure from the endoscopic endonasal transcribriform approach in skull base surgery (Kassam et al., 2005) (Video S1). Briefly, to harvest samples of the RM, we resected separately the inferior, middle, and often also the superior turbinates of the nasal cavity with Heymann nasal scissors ( Figure 1C ). Next, to harvest samples of the OM, we dissected the lining of the olfactory cleft including the superior part of the septum and the cribriform plate with a sickle knife, while transecting the fila olfactoria ( Figure 1D ). Subsequently, we removed the bony part of the anterior skull base with a hammer and chisel instead of a drill, avoiding aerosol formation in these patients, some of whom might still have been contagious. After making a longitudinal incision of the dura mater, we detached the OB from the overlying part of the brain using a ball probe, J o u r n a l P r e -p r o o f 7 ensuring atraumatic removal of the tissue, and transected the OB from the olfactory tract as posteriorly as possible ( Figure 1E ). We performed the procedure on the left and right nasal cavity, and the identical procedure on control patients.

In summary, we drastically reduced the PMI: the median was 67 minutes for COVID-19

ICU patients, 85 for COVID-19 ward patients, and 89 for control patients.

We reasoned that visualizing the target cell types of an RNA virus ought to be conducted first and foremost by RNA in situ hybridization. We opted for the RNAscope technology, which visualizes a single RNA molecule as a dot or "punctum", plural "puncta" (Wang et al., 2012) .

Fluorescence RNAscope can be combined with fluorescence IHC, which visualizes an antigen as immunoreactive (IR) signal. Often IR signal diffusely fills a cell and consequently outlines its contours, facilitating cell type identification. which encodes a transcription factor involved in ciliogenesis, is a marker for ciliated cells, whose cilia continuously sweep the overlying mucus to the nasopharynx. EPCAM, a cell adhesion molecule, labels ciliated cells in the RE and cells of mucus-producing glands and their ducts in the LP. The mucin MUC5AC, a gel-forming glycoprotein protecting the RM, is a marker for goblet cells, and MUC5AC-IR signal also identifies secreted blobs of mucus.

The OM is a minor constituent of the nasal mucosa ( Figure 2B (Barnes et al., 2020) . An OSN is shown harboring puncta for OR5A1, the major receptor for β-ionone, a key aroma in food and beverages (Jaeger et al., 2013) . Puncta for an OR gene assume a characteristic pattern resembling the shape of a cherry.

The OB resides within the cranial cavity ( Figure 2C ). It receives ipsilateral input from fila olfactoria, bundles of OSN axons that course through a few dozen holes in the sieve-like cribriform plate (Favre et al., 1995; López-Elizalde et al., 2018; Vasvári et al., 2005) . TUBB3, a component of microtubules, is a classical marker of neurons and axons (Lee et al., 1990; Zapiec et al., 2017) . TUBB3-IR OSN axons coalesce into glomeruli in the OB. The surface of the OB is covered snugly with pia mater, a thin leptomeningeal layer that is IR for SSTR2A, somatostatin receptor 2 (Boulagnon- Rombi et al., 2017; Menke et al., 2015) . The other leptomeningeal layer is the arachnoid, a spider web-like structure that connects to the dura mater, the tough outer meningeal layer close to the skull. Cerebrospinal fluid circulates continuously between the pia mater and the arachnoid.

In summary, our rapid approach of tissue sample procurement allowed us to generate, from 100% of cases, high-quality confocal images combining RNAscope with IHC. 

and SARS-CoV-2-orf1ab-sense (orf1ab-sense). The sense probes detect negative-sense RNAs, with puncta occurring perinuclearly (Chandrashekar et al., 2020; Liu et al., 2020) . We also did IHC with an antibody against nucleocapsid. Figure S2 shows negative controls for the probes and the antibody.

To provide suitable context for the examination of the samples of olfactory cleft mucosa, we first examined the RM samples ( Figure 3 ).

We detected viral presence in the RM of 30 of the 68 (44%) COVID-19 cases. Henceforth we refer to this subset as the "informative" cases. They died within 16 days after diagnosis of COVID-19 by reverse-transcription, quantitative polymerase chain reaction (henceforth abbreviated as PCR), except for COVID #29, the immunosuppressed recipient of a solid organ transplant who died 29 days after diagnosis ( Figure S3 ). We did not detect SARS-CoV-2 puncta in the RM, OM, or OB of the other 38 ("non-informative") COVID-19 cases, of the two convalescent COVID-19 cases, and of the 15 control cases. For COVID #9 through #70, we carried out rapid antigen tests on nasopharyngeal (NP) swabs that we took endoscopically prior to the procedure (Video S1 at 0'31"), and found a high concordance with the RNAscope data ( Figure S4A ). For 11 COVID-19 cases, we obtained PCR data on a second NP swab that we took preprocedurally ( Figure S4B ).

We identified ciliated cells as the major target cell type for SARS-CoV-2 in the RM of 27 of the 30 (90%) informative cases, and cells lining gland ducts in the LP in 4 (13% Figure 3A ).

J o u r n a l P r e -p r o o f 10 EPCAM-IR labels epithelial cells, and MUC5AC-IR signal labels mucin-producing cells and identifies blobs of mucus ( Figure 3B ). ACE2-IR signal forms a discontinuous thin band at the luminal surface of the RE, and puncta for TMPRSS2 abound throughout the RE ( Figure 3C ).

In the RE of COVID #7, nucleocapsid-IR signal diffusely fills an uninterrupted apical row of cells ( Figures 3D and 3E ). The timeline of infection is exceptionally well defined for this patient, who died 78 hr after diagnosis, which was preceded by two negative PCR results from NP swabs taken 3 and 6 days earlier. Consistent with the acute phase of the infection, nucleocapsid-IR cells harbor perinuclear orf1ab-sense puncta. Perinuclear N-sense puncta cluster with orf1ab-sense puncta in nucleocapsid-IR cells in COVID #7 ( Figure 3F ) and with densely packed orf1ab puncta in COVID #27 ( Figure 3G ). In COVID #51, perinuclear Nsense puncta cluster with M and FOXJ1 puncta ( Figure 3H ). In COVID #39, perinuclear orf1ab-sense puncta cluster with FOXJ1 puncta within an individual ciliated cell ( Figure 3I ).

Perinuclear S-sense puncta cluster with FOXJ1 puncta in KRT7-IR cells in COVID #7 ( Figure   3J ).

In 4 of the 30 informative cases, cells lining gland ducts in the LP were infected. In COVID #29, #63, and #67, only the ducts were infected, and in COVID #60 both the RE and the ducts were infected. KRT8-IR cells lining gland ducts in the LP harbor densely packed N puncta in COVID #29 ( Figure 3K ) and orf1ab-sense puncta in COVID #63 ( Figure 3L ).

Initially COVID #63 was included as a control case, with a negative PCR result from a NP swab taken 82 hr prior to the time of death, but tested PCR-positive on a swab that we took postmortem. COVID #63 has the shortest period between diagnosis and death in our cohort.

In summary, the RM is a major site of infection for SARS-CoV-2 and represents a vast area of cells susceptible to virus entry and replication (Wölfel et al., 2020; Zou et al., 2020) . We obtained Variant of Concern-specific PCR or sequence data for 35 COVID-19 cases, among whom are COVID #60 (infected with a non-Alpha lineage) and COVID #68 (infected with Alpha). In the RE of COVID #60 (a patient with an active oncological condition who died 40 hr after diagnosis), a fraction of cells harboring FOXJ1 puncta are diffusely filled with nucleocapsid-IR signal, and most of these cells harbor perinuclear orf1ab-sense puncta ( Figure 4A , top). In the RE of COVID #68 (a patient with an active oncological condition who died 5 days after diagnosis), perinuclear S-sense puncta cluster with FOXJ1 puncta in nucleocapsid-IR cells ( Figure 4A , bottom). A mix of two BaseScope probes for the wild-type or deletion form of S yielded either a teal or red precipitate in the RE of COVID #60 or of COVID #68, respectively ( Figure 4B ). A mix of two BaseScope probes for the wild-type or deletion form of orf1ab supported this binary genotyping ( Figure 4C ).

In summary, we have developed a post-hoc assay for differential diagnosis of infection with Alpha vs. non-Alpha lineages in fixed tissue samples.

Next, we analyzed samples from olfactory cleft mucosa ( Figures 5 and S5 Figure 5A ).

ACE2-IR crest-like stripes cap an array of intertwined KRT8-IR sustentacular cells and TUBB3-IR OSNs ( Figure 5B ). An image of COVID #22, who died 26 days after diagnosis and had no detectable SARS-CoV-2 puncta in any tissue sample, showcases the three major cell types of the OE ( Figure 5C ). Puncta for GPX3, which encodes a glutathione peroxidase, label sustentacular cells from apical to basal and Bowman's gland cells in the LP. Puncta for ANO2, which encodes the chloride channel in the olfactory signal transduction pathway, label the middle layer of OSNs. KRT5/6-IR signal labels the basal layer of cells.

A highly informative case is COVID #8, who died four days after diagnosis. TUBB3-IR cells (OSNs) do not contain nucleocapsid-IR signal, and sustentacular cells harbor UGT2A1 puncta ( Figure 5D ). N puncta diffusely fill a great many sustentacular cells spanning the width of the OE from apical to basal; interestingly, KRT8-IR signal identifies a patch of uninfected sustentacular cells, whereas infected sustentacular cells are low on or negative for KRT8-IR signal ( Figure S5A ). N puncta are densely packed in cells with the typical shape of sustentacular cells (resembling a wine glass with a twisted stalk touching the basal lamina), and the wider apical parts of infected sustentacular cells are intermingled with those of uninfected sustentacular cells harboring UGT2A1 puncta and capped with IR signal for ERMN, a sustentacular cell marker ( Figures 5E and S5B ). That infected sustentacular cells are low on or negative for puncta or IR signal for a given marker is consistent with SARS-CoV-2-elicited decay of host mRNAs and inhibition of host protein translation (Banerjee et al., 2020; Burke et al., 2021; Finkel et al., 2021; Schubert et al., 2020; Zhang et al., 2021) .

This multipronged viral takeover is illustrated by a single infected sustentacular cell standing out among uninfected sustentacular cells ( Figure 5F ): this cell is devoid of GPX3 puncta, is J o u r n a l P r e -p r o o f 13 filled diffusely with nucleocapsid-IR signal from apical to basal, and harbors perinuclear orf1ab-sense puncta.

We exhaustively searched for the presence of sense puncta and nucleocapsid-IR signal in

OSNs but failed to find it. By way of example of the negative evidence, S-sense puncta occur in the apical layer along with KRT8-IR signal, whereas puncta for the pool of four OR gene probes occur in the middle layer ( Figure 5G) . A high-magnification image shows that OSNs harboring puncta for the probe pool do not harbor perinuclear S-sense puncta ( Figure 5H ).

The apical layer harboring S-sense puncta and containing KRT8-IR signal is mutually exclusive with the middle layer of OSNs harboring puncta for the probe pool ( Figure 5I ). An individual OSN harboring puncta for the OR gene OR7C1 as well as several TUBB3-IR cells surrounding it do not harbor S-sense puncta ( Figure 5J ).

We confirmed these observations in another case, COVID #7. We identified OSNs with CNGA2 puncta and TUBB3-IR signal (Figures S5C and S5D) or GNAL puncta ( Figure S5E ).

Among several uninfected sustentacular cells harboring puncta for SOX2 (Durante et al., 2020), two cells harbor perinuclear orf1ab-sense puncta ( Figure S5F ). Sustentacular cells harboring densely packed N puncta stand out by the depletion of KRT18-IR signal (Figures S5G and S5H) . In COVID #57 ( Figure S5I ) and COVID #25 ( Figure S5J ), the infected OE is damaged, with swaths of tissue sloughing off; it may well be at the verge of desquamation.

In summary, sustentacular cells are the major target cell type in the olfactory mucosa. We failed to find evidence for infection of OSNs. The pattern of infection of the OM is patchy. Figure 6B ). There is no significant difference in nucleus counts between the two types of AOIs ( Figure 6C ).

Each AOI was UV-illuminated individually to photocleave the WTA probes for collection and sequencing. The normalized expression counts for orf1ab in ORF1ab High vs. ORF1ab Low

AOIs fit well with our visual judgment of the confocal scans ( Figure 6D ). The S and orf1ab

counts have a Pearson's correlation coefficient of 0.998 ( Figure 6E ). Differential expression modeling of 9,262 genes detected in at least 20% of the AOIs reveals that in ORF1ab High AOIs, the normalized expression counts for sustentacular cell markers GPX3, KRT8, and In summary, the intra-slide approach of spatial whole-transcriptome profiling revealed no changes in OR gene expression levels in OE patches of high vs. low viral load in COVID #8.

Consistent with the absence of evidence for infection of OSNs, we failed to find evidence for viral invasion of the OB parenchyma. Surprisingly, we discovered viral RNA within the leptomeningeal layers surrounding the OB in 11 of the 30 (37%) informative cases ( Figure 7 ).

In COVID #16 (a patient with an active oncological condition who died 8.5 days after diagnosis), a tiled confocal image of a sagittal section of a whole OB shows SSTR2A-IR signal labeling the pia mater and the arachnoid, and TUBB3-IR signal labeling incoming OSN axons and OB neurons ( Figure 7A ). A high-magnification image shows N puncta at the side of the pia mater abutting the OB ( Figure 7B ). In an adjacent section of the same OB, densely packed N puncta occur within a segment of the pia mater together with abundant nucleocapsid-IR signal, a combination that may reflect free virions, but not in the OB parenchyma ( Figure 7C ). In another section, S puncta occur within an obliquely cut blood vessel defined by PECAM1 puncta in endothelial cells ( Figure 7D ). In COVID #7, N puncta occur in a swath of SSTR2A-IR pia mater that is partially detached, but not in the OB parenchyma ( Figure 7E ). In COVID #27 (who died 93 hr after diagnosis), densely packed M puncta occur in the pia mater covering the OB and outside the confines of a blood vessel harboring PECAM1 puncta, but not in the OB parenchyma ( Figure 7F ). In COVID #60 (a patient with an active oncological condition who died 40 hr after diagnosis), a leptomeningeal sample near the OB that includes the transition zone to the dura mater contains abundant N and S puncta scattered among SSTR2A-IR signal ( Figure 7G ).

In summary, SARS-CoV-2 does not appear to be a neurotropic virus, in the sense that it does not infect OSNs and OB neurons.

J o u r n a l P r e -p r o o f 16

We have here taken a virocentric view of COVID-19, from the viewpoint of SARS-CoV-2 acutely attacking the human olfactory system. We identified sustentacular cells as the main target cell type in the OM, failed to find evidence for infection of OSNs and of the OB parenchyma, and discovered viral RNA in the leptomeningeal layers surrounding the OB.

Our cohort consisted of patients who died from or with COVID-19 in two major hospitals over a period of 12 months spanning the first three waves of the pandemic in Belgium. We consistently kept the PMI at approximately one hour. None of the 85 cases had to be excluded because of poor staining quality. As the onset of symptoms is not always clear or even known and is subject to patient recall, we chose to report the period until death starting from the time the NP swab was taken that led to the diagnosis.

The 30 informative cases died at a median of 8.8 days (Q1-Q3: 4-12) after diagnosis, Of the 30 informative cases, 9 (30%) displayed ongoing viral replication at the time of death, as judged by the presence of sense puncta: COVID #7, #8, #27, #39, #51, #60, #63, #67, and #68. These patients died within 8.5 days after diagnosis. Our panel of seven RNAscope probes and the nucleocapsid antibody represents a stringent criterion for assessing virus replication. As sustentacular cells have phagocytic activity (Suzuki et al., 1996) , the mere demonstration of nucleocapsid-IR (or spike-IR signal) in sustentacular cells J o u r n a l P r e -p r o o f 17 is insufficient to call these cells infected: the signal may reflect phagocytosis of debris from infected cells.

The major target cell type in the RM are ciliated cells (Ahn et al., 2021; Hou et al., 2020; Lee et al., 2020; Sungnak et al., 2020; Ziegler et al., 2021) . The expression pattern of the receptor can predict which cells can be infected but does not mean that all cells that express this receptor or even the cells with the highest expression level are the major targets (Weiss, 2020). A secretory form of ACE2 may explain some of these discrepancies (Yeung et al., 2021) . Neuropilin-1 expression in olfactory epithelial cells has been invoked as a cofactor facilitating SARS-CoV-2 cell entry and infectivity (Cantuti-Castelvetri et al., 2020).

In the same vein, the popular interpretation of the absence of expression of ACE2 and We applied spatial whole-transcriptome profiling to the OE of COVID #8 to address quantitatively the hypothesis of an indirect effect on OR gene expression. This method is complementary and orthogonal to the RNAscope analysis. We took an analytical approach of intra-patient, intra-slide profiling, and interrogated multiple AOIs within the OE. Our interpretation of the GeoMx WTA data is that the relative contribution of RNA from the J o u r n a l P r e -p r o o f 19 infected subpopulation of sustentacular cells is reduced in AOIs with high vs. low viral loads.

The nonstructural protein nsp1 of SARS-CoV-2 elicits a rapid decay of host mRNAs (Burke et al., 2021; Finkel et al., 2021) , consistent with our observations that infected sustentacular cells are low in or devoid of puncta for marker genes such as UGT2A1, GPX3, and SOX2.

An AOI can be regarded as a tiny, directed biopsy of a few hundred cells, and expression counts of RNA from an AOI are normalized. Therefore, the anti-correlated increase in normalized expression counts for OSN markers genes does not reflect upregulation of gene expression in OSNs but mRNA decay in infected sustentacular cells. OR genes do not undergo changes in gene expression -neither down nor up. To confirm and extend these findings, it will be necessary to investigate cases who died later after diagnosis and still had OE that was infected.

Admittedly, the absence of evidence for infection of OSNs does not constitute evidence of absence. We leave the possibility open that OSNs may become infected and support viral replication in a subset of patients, or in certain disease courses or phases.

OSNs do not appear to offer SARS-CoV-2 a route straight to the brain from the nasal cavity via the OB. An intriguing observation was our finding of SARS-CoV-2 puncta in the leptomeningeal layers surrounding the OB in 11 of the 30 informative cases. We speculate that these puncta reflect RNA within free extracellular virions instead of intracellular viral (Balcom et al., 2021) , such as by prompting the generation of autoantibodies against neural antigens (Song et al., 2021a) . It is tempting to speculate that this viral RNA presence may contribute to olfactory dysfunction by perturbing signal propagation via the olfactory tract from the OB to the cerebral cortex.

Here too, the absence of evidence for invasion of the OB parenchyma does not equal evidence of absence of invasion. In any case, our data do not support the neurotropic properties and neuroinvasive capacity that have been attributed by some to SARS-CoV-2 (Song et al., 2021b) .

The pathogenesis of olfactory dysfunction in COVID-19 may turn out to be multifactorial and heterogeneous among patients. There need not be a single mechanism explaining all cases of olfactory dysfunction. We favor a pathobiological mechanism whereby the sequence of events that ultimately mutes or alters the sense of smell is initiated when infected sustentacular cells no longer provide sufficient support, structural and/or physiological, to

OSNs. They may even harm OSNs, such as through paracrine effects of chemokines The 23andMe COVID-19 initiative reported on a genome-wide association study comparing loss of smell or taste with no loss of smell or taste among nearly 70,000 probands with a positive SARS-CoV-2 test (Shelton et al., 2021) . A single associated locus was J o u r n a l P r e -p r o o f 21 identified, comprising the UGT2A1 and UGT2A2 genes encoding UDP glucuronosyltransferase enzymes. In rat, UGT2A1 is involved in odorant metabolization, which aids in olfactory signal termination (Lazard et al., 1991) . Our findings of UGT2A1 puncta support a role of sustentacular cells in COVID-19 associated olfactory dysfunction.

In view of the superficial location of sustentacular cells, which present ACE2 receptors to virions within the mucus, the mucosal immune system (Iwasaki, 2016) may not be able to prevent infection of these cells. It may have to condone a brief phase of viral replication in sustentacular cells of convalescent COVID-19 patients during re-infection or fully vaccinated individuals during breakthrough infection (Yewdell, 2021) . Therefore prior natural infection or vaccination may not be fully protective against olfactory dysfunction upon subsequent exposure to SARS-CoV-2.

Understanding the mechanisms whereby human sustentacular cells normally support OSNs in countless ways may yield clues for therapeutic interventions aimed at preventing, alleviating, or curing olfactory dysfunction in COVID-19. The spotlight ought to be shone on the unsung heroes of the sense of smellthe humble sustentacular cells.

The scope of the study was limited to visualizing how SARS-CoV-2 attacks the nasal mucosa and whether it invades the OB parenchyma. We took the viewpoint of the virus and not of the 

Further information and requests should be directed to the Lead Contact, Peter Mombaerts (peter.mombaerts@gen.mpg.de).

This study did not generate new unique reagents.

 Clinical data about the patients are confidential, subject to compliance with applicable personal data protection laws, and not publicly available. For a patient who was initially called control #11, diagnosis of COVID-19 was made postmortem by PCR on a nasopharyngeal swab we took during the postmortem bedside surgical procedure and in parallel through our RNAscope and IHC analyses. We then renamed control #11 as COVID #63 but did not reassign number 11 to the next control case.

The collection, processing, and disclosure of personal data, such as patient demographic, health, and medical information, are subject to compliance with Regulation (EU) 2016/679, also referred as the General Data Protection Regulation, and the Belgian Law on the protection of natural persons regarding the processing of personal data.

Therefore, combinations of data deemed to be identificatory to specific persons cannot be disclosed.

J o u r n a l P r e -p r o o f 56 Comorbidities were categorized in accordance with international recommendations.

Overweight is as a body mass index (BMI) >25 kg/m 2 , and obesity as a BMI ≥30 kg/m 2 .

Presence of diabetes mellitus type 2 includes previously known and newly diagnosed patients, based on Hb1Ac ≥6.5% or active treatment on admission. Former smokers, defined as having ceased smoking >6 months prior to inclusion, are not considered smokers in Figure 1B . Hypertension is defined as grade 1 hypertension, or treatment with antihypertensive drugs. Chronic kidney disease is defined as the presence of kidney damage or a glomerular filtration rate of <60 ml/min/1.73m 2 for >3 months. Chronic lung disease includes obstructive lung disease (chronic obstructive pulmonary disease, asthma), interstitial lung disease, pulmonary fibrosis, and pulmonary hypertension. Cardiovascular disease comprises heart conditions (such as valvular disease, heart failure, arrhythmias, cardiomyopathies, coronary artery disease), cerebrovascular antecedents, and history of pulmonary embolism.

Patients were considered immunocompromised if one of the following criteria was met: (1) an active oncological condition, defined as presence of a solid tumor or hematologic malignancy <6 months prior to inclusion;

(2) immunosuppressive drugs as maintenance therapy, including corticosteroids and chemotherapy; (3) recipient of a solid organ transplant.

For ICU patients, the Sequential Organ Failure Assessment (SOFA) score, the Acute Physiology And Chronic Health Evaluation II (APACHE II) score, and the arterial-to-inspired oxygen (P aO2 /F IO2 ) ratio were calculated daily. The highest SOFA and APACHE II scores and lowest P aO2 /F IO2 ratio were extracted from the patient file as indicators of disease severity while on ICU.

The cause of death of COVID-19 patients was classified into one out of three categories.

(1) Death from COVID-19: hypoxic respiratory failure secondary to COVID-19 pneumonia, 

Samples of respiratory and olfactory cleft mucosa and whole olfactory bulbs were harvested bedside by ENT surgeons via an endoscopic endonasal approach soon after the death of the patient. A 4 mm 0° endoscope (Karl Storz), connected with a camera and monitor and light source, was used throughout the procedure allowing optimal visualization and assistance.

To harvest respiratory mucosa samples, the inferior turbinate, middle turbinate, and superior turbinate were resected bilaterally with Heymann nasal scissors.

 The inferior turbinate is attached to the lateral nasal wall over its entire length (5-6 cm). Prior to cutting its attachment, the inferior turbinate was in-fractured by a Cottle elevator allowing optimal positioning of the Heymann scissors.

 The middle and superior turbinates each have a vertical, anterior attachment to the skull base and a horizontal, more posterior attachment to the lateral nasal wall. For both turbinates, the anterior attachment was cut first with Heymann or endoscopic scissors, followed by the posterior attachment.

 To harvest the mucosa in toto from the resected turbinate bone, a dissection in the subperiosteal plane was performed with a Cottle elevator.

 Samples of each turbinate were transferred into separate pots containing 10%

formalin.

To harvest olfactory cleft mucosa samples, the lining covering the olfactory cleft including the superior part of the septum and the cribriform plate was resected.

 An elliptical incision was made with a sickle knife running over the superior part of the septum, the cribriform plate and the area of the vertical attachment of the medial and superior turbinates, thus covering the full olfactory cleft region.

J o u r n a l P r e -p r o o f 58  A subperiosteal dissection was initiated with a sickle knife on the medial side (superior part of the septum) and lateral side (vertical attachment of the turbinates) simultaneously, progressively extending to the center (cribriform plate), where the mucosa is attached only by the remaining fila olfactoria.

 After transection and tearing of the fila olfactoria, the mucosa was harvested in one or a few pieces. All pieces were transferred into a single container with 10% formalin.

To harvest whole olfactory bulbs, an adapted transcribriform approach was performed at the end of the procedure.  After full exposure of the bony skull base, the adapted transcribriform approach was performed. Compared to the conventional approach, the opening made in the bony skull base is smaller: extending from lateral to the anterior attachment of the middle and superior turbinates until the septum (width) and from the anterior ethmoidal artery until the anterior wall of the sphenoid (length). Resection of the bony skull was performed with hammer and chisel. Cold instruments were used instead of powered instruments, such as a high-speed drill with rinsing system, to avoid aerosol formation in these patients, who might still have been contagious at the time of death.  The exposed dura mater was incised longitudinally and paramedially to avoid damage to the overlying olfactory bulb. After the olfactory bulb was exposed, blunt resection with a ball probe allowed harvesting of the full length of the olfactory bulb, often including an attached part of the olfactory tract. Therefore, the transection was made as posteriorly as possible.

J o u r n a l P r e -p r o o f 59 The ENT surgeons wore powered air-purifying respirator masks and personal protective equipment during the surgical procedure on COVID-19 patients.

In October 2020, the amended ANOSMIC-19 study protocol implemented systematically the use of rapid antigen tests ( 

Tissue sections were stained using a fully automated H&E platform (Dako CoverStainer, Agilent).

The 

Individual DCC files were aggregated and checked for probe-level quality prior to data analysis. The data consist of two pools totaling 18,953 probes and 18,704 genes. For the WTA, each gene is mapped to a single probe. For the COVID-19 spike-in, which includes probes for S and ORF1ab, there were five probes per target. Target counts were generated for the multiple-probe genes by taking the geometric mean of their counts after removing probes that did not pass the Grubb's outlier test (alpha = 0.01). Sample-level quality control was also performed. Each sample was screened to ensure greater than 50% sequencing saturation. The negative probe geometric means for each sample were visually checked to ensure there were no pool dropouts.

For each sample, two values of Limit of Quantification (LOQ) were derived, one for each pool. LOQ for a given gene is defined as the geometric mean of pool-specific negative probes times the geometric standard deviation of negative probes raised to a power of 2 (i.e., LOQ2). These LOQ values were used as a basis of filtering genes that are expressed near background. We required that a given gene needed to be above LOQ2 in at least 20% (i.e, in J o u r n a l P r e -p r o o f 64 ≥ 4 of the 17 AOIs). The filtered expression data were then normalized. Specifically, the 75th percentile of target counts for each AOI was computed and each of these values was divided by the geometric mean of the 75th percentile values of all 17 AOIs to generate normalization factors. Targets count values for a given AOI were then divided by their sample-specific normalization factor. GeoMx profiling data are available on GEO at GSE176080.

In Figures 6A and 6B , the geometric mean of the log 2 normalized ORF1ab expression (9.92) was used as the basis of bifurcating samples into ORF1ab Low AOIs (n=7) and ORF1ab

High AOIs (n=10). In Figure 6C , a Welch Two Sample t-test was used to determine whether there was a significant difference in nucleus counts between ORF1ab Low AOIs (n= 7) and

ORF1ab High AOIs (n= 10). In Figure 6D , a Welch Two Sample t-test was used to determine whether there was a significant difference in normalized log 2 ORF1ab expression between ORF1ab Low AOIs (n= 7) and ORF1ab High AOIs (n= 10). In Figure 6E , a linear model using the base stats package in R was used to regress log 2 normalized ORF1ab expression against log 2 normalized S expression (n = 17 AOIs). In Figure 6F , differential expression analysis was performed for each gene by regressing the log 2 normalized gene expression by viral load (two levels) using R. Raw p values were adjusted for multiple hypothesis testing using a Benjamini-Hochberg False Discovery Rate of 5%

(https://www.jstor.org/stable/2346101?seq=1). In Figure 6G , a bar plot was presented in GraphPad Prism v9.2. In Figure 6H, J o u r n a l P r e -p r o o f

In the Discussion, an independent-samples Mann-Whitney U test was run to determine if there were significant differences in the time from diagnosis to death between informative cases (n=30, median 8.8 days, IQR 7.4) and non-informative cases (n=38, median 21.1 days, IQR 26.4), with n the number of cases in each group, and IQR the interquartile range. 

Possible ATP trafficking by ATP-shuttles in the olfactory cilia and glucose transfer across the olfactory mucosa

Nasal ciliated cells are primary targets for SARS-CoV-2 replication in the early stage of COVID-19

Acute and chronic neurological disorders in COVID-19: potential mechanisms of disease

SARS-CoV-2 disrupts splicing, translation, and protein trafficking to suppress host defenses

Expert curation of the human and mouse olfactory receptor gene repertoires identifies conserved coding regions split across two exons

High-plex spatially resolved RNA and protein detection using Digital Spatial Profiling: a technology designed for immuno-oncology biomarker discovery and translational research

The identification and topographical localisation of the olfactory epithelium in man and other mammals

Natural history, trajectory, and management of mechanically ventilated COVID-19 patients in the United Kingdom

SARS-CoV-2 one year on: evidence for ongoing viral adaptation

Clinical outcomes for patients with anosmia 1 year after COVID-19 diagnosis

Anatomy of the olfactory mucosa

SARS-CoV-2 Nsp1 binds the ribosomal mRNA channel to inhibit translation

The UGT2A1/UGT2A2 locus is associated with COVID-19-related anosmia

Continuous and discontinuous RNA synthesis in coronaviruses

Divergent and self-reactive immune responses in the CNS of COVID-19 patients with neurological symptoms

Neuroinvasion of SARS-CoV-2 in human and mouse brain

Anosmia, ageusia, and other COVID-19-like symptoms in association with a positive SARS-CoV-2 test, across six national digital surveillance platforms: an observational study. Lancet Digit Health

SARS-CoV-2 entry factors are highly expressed in nasal epithelial cells together with innate immune genes

Supporting cells as phagocytes in the olfactory epithelium after bulbectomy

Isolation of putative stem cells present in human adult olfactory mucosa

Olfactory epithelium histopathological findings in long-term coronavirus disease 2019 related anosmia

Anosmia and ageusia: common findings in COVID-19 patients

Psychophysical evaluation of the olfactory function: European multicenter study on 774 COVID-19 patients

COVID-19 in the olfactory system and their implications

Dysregulation of brain and choroid plexus cell types in severe COVID-19

Individuals cannot rely on COVID-19 herd immunity: Durable immunity to viral disease is limited to viruses with obligate viremic spread

Soluble ACE2-mediated cell entry of SARS-CoV-2 via interaction with proteins related to the renin-angiotensin system

A ventral glomerular deficit in Parkinson's disease revealed by whole olfactory bulb reconstruction

Nsp1 protein of SARS-CoV-2 disrupts the mRNA export machinery to inhibit host gene expression

Impaired local intrinsic immunity to SARS-CoV-2 infection in severe COVID-19

For codetection of RNA and protein, IHC was performed after the final step of HRP blocker application in the RNAscope Multiplex Fluorescent Detection protocol. Slides were blocked in 10% donkey serum (Sigma-Aldrich, Cat#S30-100ML) in 0.1% Triton/PBS at room temperature for 1 hr. The following primary antibodies were diluted in 2% donkey serum in

Cat#AF933) at 1:100, EpCAM (Abcam, Cat#ab32392) at 1:100, ERMN (Thermo Fisher Scientific, Cat#PA5-58327) at 1:100, Cytokeratin 5/6 (Novus Biologicals, Cat#NBP2-77439) at 1:200, Cytokeratin 7 (Novus Biologicals, Cat#NBP2-44813) at 1:200, Cytokeratin 8 (R&D Systems, Cat#MAB3165) at 1:200

Cat#40143-R001) at 1:100, Somatostatin receptor subtype 2A/SSTR2A (Biotrend, Cat#NB-49-016-50ul) at 1:4000, and TuJ1/TUBB3 (BioLegend, Cat#801202) at 1:100 for OM sections and 1:400 for OB sections. Slides were then washed in 0.1% Triton/PBS 3 x 5 min each followed by incubation with appropriate secondary antibodies at 1:500 in 2% normal donkey serum in 0

Cat#A-11056), Alexa Fluor Plus 647 donkey anti-rabbit (Thermo Fisher Scientific, Cat#A32795), Alexa Fluor Plus 647 donkey anti-mouse (Thermo Fisher Scientific, Cat#A32787), Alexa Fluor Plus 555 donkey anti-rabbit (Thermo Fisher Scientific, Cat#A32794), and Alexa Fluor Plus 555 donkey anti-mouse (Thermo Fisher Scientific, Cat#A32773). Slides were washed in 0.1% Triton/PBS 3 x 5 min each followed by DAPI (Thermo Fisher Scientific, Cat#D1306) application for nuclei staining. Slides were mounted in Mount Solid antifade (abberior, Cat#MM-2011-2X15ML). For IHC only, slides were pretreated in target retrieval reagent (Advanced Cell Diagnostics, Cat#322000) for 3 min in a steamer

BaseScope probes were designed for a 9-nucleotide deletion encoding amino acids SGF 3675-3677 of the ORF1ab gene (Advanced Cell Diagnostics, Cat#1055881-C1 for wildtype and Cat#1055871-C2 for deletion) and a 6-nucleotide deletion encoding amino acids HV 69-70 of the S gene (Advanced Cell Diagnostics, Cat#1055861-C1 for wildtype and Cat#1055851-C2 for deletion). The BaseScope assay was performed according to manufacturer's protocols using the BaseScope Duplex Reagent Kit (Advanced Cell Diagnostics, Cat#323800). Tissue pretreatment was performed in the same way as in the fluorescence RNAscope experiments. Slides were mounted in VectaMount permanent mounting medium (Vector Labs

The fixed frozen slide was baked at 37°C for 1 hr, fixed for 30 min in 10% neutral buffered formalin (Electron Microscopy Sciences

Cat#848568-C3) and the GeoMx slide prep protocol on a Leica Bond Rxm. The slide was then incubated with GeoMx WTA (NanoString Technologies, Cat#121401102) and COVID-19 spike-in reagents at 37°C overnight to allow the probes to hybridize to their RNA targets. Following incubation, morphology marker antibodies anti-Pan-cytokeratin (Novus Biologicals, Cat#NBP2-33200AF488) referred to as pan-KRT, anti-cytokeratin 8/18 (Novus Biologicals, Cat#NBP2-34655AF488) referred to as KRT8/18, and DNA dye Syto 83