Non-neural expression of SARS-CoV-2 entry genes in the olfactory epithelium suggests mechanisms underlying anosmia in COVID-19 patients

David Brann; Tatsuya Tsukahara; Caleb Weinreb; Darren W. Logan; Sandeep Robert Datta

An analysis of single-cell gene expression data from 10 studies was performed to investigate the expression of genes coding for coronavirus entry proteins in neurons from a range of brain regions and sensory systems. Processed gene expression data tables were obtained from scSeq studies that evaluated gene expression in retina (GSE81905) (79) inner ear sensory epithelium (GSE115934) (80, 81) and spiral ganglion (GSE114997) (82) , ventral midbrain (GSE76381) (83), hippocampus (GSE100449) (84) , cortex (GSE107632) (85) , hypothalamus (GSE74672) (86), visceral motor neurons (GSE78845) (87), dorsal root ganglia (GSE59739) (88) and spinal cord dorsal horn (GSE103840) (89) . Smart-Seq2 sequencing data from Vsx2-GFP positive cells was used from the retina dataset. A subset of the expression matrix that corresponds to day 0 (i.e. control, undisturbed neurons) was used from the layer VI somatosensory cortex dataset. A subset of the data containing neurons from untreated (control) mice was used from the hypothalamic neuron dataset. From the ventral midbrain dopaminergic neuron dataset, a subset comprising DAT-Cre/tdTomato positive neurons from P28 mice was used. A subset comprising Type I neurons from wild type mice was used from the spiral ganglion dataset. The "unclassified" neurons were excluded from the visceral motor neuron dataset. A subset containing neurons that were collected at room temperature was used from the dorsal root ganglia dataset. Expression data from dorsal horn neurons obtained from C57/bl6 wild type mice, vGatcre-tdTomato and vGlut2-eGFP mouse lines was used from the spinal cord dataset. Inspection of all datasets for batch effects was performed using the scater package (version 1.10.1) (77) . Publicly available raw count expression matrices were used for the retina, hippocampus, hypothalamus, midbrain, visceral motor neurons and spinal cord datasets, whereas the normalized expression data was used from the inner ear hair cell datasets. For datasets containing raw counts, normalization was performed for each dataset separately by computing pool-based size factors that are subsequently deconvolved to obtain cell-based size factors using the scran package (version 1.10.2) (90) . Violin plots were generated in scater. Fig. 4 . Expression of genes coding for CoV-2 entry-related proteins in neurons forming part of deeply sequenced scSeq datasets from various brain regions and sensory systems.