SARS-CoV-2 infection causes dopaminergic neuron senescence

Summary COVID-19 patients commonly present with signs of central nervous system and/or peripheral nervous system dysfunction. Here, we show that midbrain dopamine (DA) neurons derived from human pluripotent stem cells (hPSCs) are selectively susceptible and permissive to severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection. SARS-CoV-2 infection of DA neurons triggers an inflammatory and cellular senescence response. High-throughput screening in hPSC-derived DA neurons identified several FDA-approved drugs that can rescue the cellular senescence phenotype by preventing SARS-CoV-2 infection. We also identified the inflammatory and cellular senescence signature and low levels of SARS-CoV-2 transcripts in human substantia nigra tissue of COVID-19 patients. Furthermore, we observed reduced numbers of neuromelanin+ and tyrosine-hydroxylase (TH)+ DA neurons and fibers in a cohort of severe COVID-19 patients. Our findings demonstrate that hPSC-derived DA neurons are susceptible to SARS-CoV-2, identify candidate neuroprotective drugs for COVID-19 patients, and suggest the need for careful, long-term monitoring of neurological problems in COVID-19 patients.


In brief
Yang and colleagues demonstrate that SARS-CoV-2 can infect hPSC-derived midbrain dopamine (DA) neurons and induce cellular senescence.Several drug candidates were identified to rescue SARS-CoV-2 infection and DA neuron senescence.Inflammation and cellular senescence were also identified in substantia nigra tissue of COVID-19 patients.

INTRODUCTION
Abnormal neurological manifestations are increasingly recognized in patients with COVID-19, which most commonly include anosmia, dysgeusia, and headache, followed by seizures, stroke, and acute inflammatory polyradiculoneuropathy, also known as Guillain-Barre syndrome. 1Furthermore, an increased risk for addi-tional neurological and psychiatric disorders has been reported in a large retrospective cohort at 6 months post diagnosis. 2Recent studies using hPSC-derived organoid models have established that choroid plexus cells within the CNS are highly susceptible to severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection. 3,44][5][6] We previously developed a (legend continued on next page) human pluripotent stem cell (hPSC)-derived organoid/cell-based platform to evaluate the tropism of SARS-CoV-2.Using this platform, we found that hPSC-derived midbrain dopamine (DA) neurons-representing one of the main neurodegeneration targets in Parkinson's disease (PD)-are permissive to SARS-CoV-2 infection. 7Conversely, under identical experimental conditions, we found that hPSC-derived cortical neurons are not permissive to SARS-CoV-2 infection, 7 supporting the notion that not all neuronal populations are equally permissive to viral infection.
Here, we set out to define how DA neurons respond to SARS-CoV-2 infection and to determine the molecular changes induced by SARS-CoV-2 infection.

RESULTS
hPSC-derived DA neurons are susceptible and permissive to SARS-CoV-2 infection To examine the impact of SARS-CoV-2 infection on DA neurons, postmitotic DA neuron marker (NURR1: GFP), reporter hPSCs were differentiated toward a DA neuron fate using a previously established strategy. 8,9Our previous work found that hPSC-derived midbrain DA neurons are susceptible to a vesicular stomatitis DGluciferase virus pseudotyped with the SARS-CoV-2 Spike protein incorporated at the surface of the viral particle (SARS-CoV-2-entry virus). 7To focus our study on purified DA neurons in culture, we sorted NURR1-GFP + cells at day 25 of differentiation.In this study, DA neuron identity was validated by immunofluorescent staining with NURR1-GFP, tyrosine-hydroxylase (TH), MAP2, and FOXA2 showing co-expression of TH and FOXA2 at day 40 (Figures S1A and S1B).We used those highly purified populations of DA neurons, resulting in up to 90% of the NURR1 + cells exhibiting DA neuron identity, consistent with the purity reported in our previous single-cell real-time quantitative PCR study (Figure S1C), 8 to monitor their response to SARS-CoV-2 infection.First, we validated the expression of ACE2, the SARS-CoV-2 receptor, in NURR1: GFP + DA neurons by immunostaining (Figures S1D and S1E).Then, DA neuron susceptibility to SARS-CoV-2 entry was further confirmed using the SARS-CoV-2-entry virus 10,11 resulting in robust luciferase activity in infected DA neurons (Figure S1F).Immunostaining demonstrated that around 5% of NURR1:GFP + cells were positively stained by luciferase antibody (Figures S1G and S1H).
Next, hPSC-derived purified DA neurons were infected in vitro with SARS-CoV-2 (USA-WA1/2020, multiplicity of infection [MOI] = 0.2).At 48 h post infection (hpi), real-time quantitative PCR analysis using primers targeting subgenomic N transcripts detected significant amounts of viral replication at the RNA level in infected hPSC-derived DA neurons (Figure 1A).Immunostaining for SARS-CoV-2 nucleocapsid protein (SARS-N) antibody confirmed robust SARS-CoV-2 infection in approximately 8% of NURR1:GFP + DA neurons at 72 hpi (Figures 1B and 1C).Early time points showed a trend toward increased SARS-CoV-2 infection to TH + DA neurons from 24 to 48 hpi (Figures 1D and  1E).Moreover, endpoint titer determining assays confirmed productive infection in hPSC-derived DA neurons with the robust generation of infectious virus at 24, 48, and 72 hpi (Figure 1F).Furthermore, transmission electron microscopy (TEM) detected the presence of viral particles in SARS-CoV-2 infected hPSCderived purified DA neurons (Figure 1G).In addition, we tested both low MOI (MOI = 0.2) and high MOI (MOI = 1) infection of DA neurons and detected higher percentage of SARS-N + cells at 72 hpi of high MOI condition than that at low MOI condition (Figures S1I and S1J).ACE2 blocking antibody prevented virus infection at both low and high MOI conditions (Figures S1I and  S1J), suggesting that SARS-CoV-2 infection of hPSC-derived DA neurons is dependent on ACE2 receptor interactions.Since the percentage of SARS-CoV-2 infection in DA neurons is not very high, we cannot fully exclude the possibility that the initially infected DA neurons can produce interferon, which protects the non-infected DA neurons.
RNA sequencing (RNA-seq) analysis was then applied to compare mock-infected or SARS-CoV-2 infected hPSC-derived purified DA neurons.Robust viral infection was detected in SARS-CoV-2 infected DA neurons (Figure 1H).Moreover, plotting these datasets by principal-component analysis (PCA, Figure 1I) and clustering analysis (Figure 1J) demonstrated that the infected DA neurons occupied a distinct transcriptional space compared with mock-infected DA neurons.Interestingly, the expression of midbrain DA neuron markers such as NR4A2 (encoding protein NURR1), FOXA2, and LMX1A, and in particular, A9 (substantia nigra) DA neuron subtype-enriched markers DKK3 and LMO3 12 , were decreased in SARS-CoV-2 infected samples (Figure 1K).
To study DA subtype-specific vulnerability to SARS-CoV-2 infection, we performed scRNA-seq from hPSC-derived purified DA neurons at day 40 in mock or SARS-CoV-2 infected (MOI = 0.2) neurons.Although we ran the scRNA-seq from GFP + purified cells corresponding to NR4A2 expression, due to the detection sensitivity 13,14 and decreased expression of NR4A2 upon SARS-CoV-2 infection, we were not able to detect NR4A2 in all cells.To further confirm the infection of DA neurons, we analyzed both TH and NR4A2 positive cells as well as all cells (Figures 2 and  S2).Clustering analysis identified four cell populations expressing DA neuron markers, including MAP2, NR4A2, and LMX1A (Figures 2B and S2B), and we mark those clusters as LMO3 high cluster-1, LMO3 high cluster-2, LMO3 high cluster-3, and CALB1 high cluster (Figures 2A-2C and S2A-S2C).One cluster shows high expression of CALB1, an A10 DA neuron marker, 15,16 whereas the other three clusters expressed higher level of LMO3, which is considered an A9-DA neuron marker 12 (Figures 2B and  S2B).There is no other cell population detected based on scRNA-seq analysis.Together, it suggests hPSC-derived NURR1-GFP-sorted neurons are a relatively pure DA neuron population, mainly consisting of A9-like DA neurons, the subtype of substantia nigra DA neurons most affected in PD. 12 (J) Clustering analysis of mock or SARS-CoV-2 infected purified NURR1:GFP-DA neurons at 48 hpi (MOI = 0.2).(K) Heatmap of expression levels of DA neurons and A9 DA neuron marker genes in the mock or SARS-CoV-2 infected purified NURR1:GFP H9-DA neurons at 48 hpi (MOI = 0.2).Data were presented as mean ± SD. p values were calculated by unpaired two-tailed Student's t test.***p < 0.001.See also Figures S1 and S3.(legend continued on next page) SARS-CoV-2 viral transcripts, including SARS-CoV-2-M, SARS-CoV-2-N, SARS-CoV-2-S, and SARS-CoV-2-E, are highly detected in SARS-CoV-2 infected cells (Figures 2D and S2D) in particular LMO3 high cluster-1, LMO3 high cluster-2, and LMO3 high cluster-3.By contrast, those viral transcripts are absent in the CALB1 high cluster (Figures 2E and 2F).Dot plots showed the decrease of A9 DA neuron marker expression (Figures 2G, 2H, S2E, and S2F), which is consistent with the bulk RNA-seq results (Figure 1K).Finally, quantitative RNA in situ hybridization further confirmed the decrease of A9 marker LMO3, but not A10 marker CALB1 expression following SARS-CoV-2 infection, indicating the loss of cell identity of A9-type DA neurons upon SARS-CoV-2 infection (Figure 2I).
To further validate the susceptibility and permissiveness of hPSC-derived DA neurons to SARS-CoV-2 infection and replication, we derived DA neurons from one additional human embryonic stem cell (hESC) line (MEL-1) and one PD-control induced PSC (iPSC) line (2 copy of SNCA DNA, 17 equivalent to healthy control).Real-time quantitative PCR and immunostaining confirmed the robust SARS-CoV-2 infection of DA neurons derived from MEL-1 hESCs (Figures S3A-S3C) and PD-control iPSCs (Figures S3D and S3E).Together, these experiments confirm that hPSC-derived DA neurons are susceptible to SARS-CoV-2 and support productive infection.
Finally, we also examined the impact of SARS-CoV-2 infection on hPSC-derived cortical neurons (Figure S3F).Consistent with our previous reports using SARS-CoV-2 entry virus, 7 there are few SARS-N positive cells detected in hPSC-derived cortical neurons at 72 hpi at MOI = 0.2 (Figure S3G).Besides, no obvious transcriptional changes were observed following SARS-CoV-2 exposure of hPSC-derived cortical neurons (Figure S3H).

SARS-CoV-2 infection induces senescence of DA neurons
Ingenuity pathway analysis of genes differentially expressed between mock and SARS-CoV-2 infected hPSC-derived DA neurons highlighted the cell cycle, DNA replication, and senescence pathways as the top upregulated pathways in SARS-CoV-2 infected DA neurons (Figure 3A).Gene set enrichment analysis (GSEA) further confirmed the enrichment of senescence pathway in SARS-CoV-2 versus mock-infected purified hPSCderived DA neurons (Figure 3B).Further analysis of RNA-seq data of purified hPSC-derived DA neurons using NURR1-GFP in mock-infected or SARS-CoV-2-infected conditions identified robust induction of chemokine/cytokine transcripts and inflammation-related genes (Figures S3I and S3J).Senescence-associated genes are also significantly upregulated in SARS-CoV-2 infected DA neurons (Figure S3K).In stark contrast, the senescence pathway was not significantly enriched in SARS-CoV-2 infected lung organoids, pancreatic cells, liver organoids, and cardiomyocytes (Figure S3L).
A key feature of senescent cells is the activation of the senescence-associated secretory phenotype (SASP).There was an upregulation of SASP associated genes, including CCL2, CCL20, CSF1, CXCL11, GDF15, IGF2R, IL1B, IL6ST, IQGAP1, and TNFRSF11B upon SARS-CoV-2 infection in DA neurons (Figure 3C).Acidic lysosomal senescence-associated b-galactosidase (SA-b-gal) activity, a biomarker of cellular senescence, 8 was also upregulated in SARS-CoV-2 infected hPSC-derived DA neurons (Figure 3D).Transmission EM detected lipofuscin in SARS-CoV-2 infected DA neurons, another senescence-associated marker of DA neurons 18 (Figure 3E).In addition, SARS-CoV-2 infected DA neurons also showed increased nuclear size (Figure 3F).Real-time quantitative PCR analysis showed the upregulation of IGFBP7 and downregulation of LMNB1, genes reported to be associated with senescence in DA neurons 8,19 (Figure 3G).scRNA-seq analysis confirmed the upregulation of IGFBP7 and CDKN1A (encoding protein P21) and downregulation of LMNB1 (Figures 3H and 3I).Western blotting further validated the upregulation of P21 and downregulation of LMNB1 at the protein level (Figure 3J).Moreover, SARS-CoV-2 infected DA neurons showed additional senescence-associated phenotypes, including increased accumulation of lysosomes as indicated by LysoTracker staining (Figure 3K), mitochondrial dysfunction as indicated by EM and MitoTracker staining (Figures 3L and 3M), and increased protein oxidation as indicated by ROS staining (Figure 3N).Consistent with the data of H9 hESC-derived DA neurons, MEL-1 hESC-derived DA neurons also showed increased b-gal staining and lysosomal accumulation upon SARS-CoV-2 infection (Figures S3M and S3N).By contrast, in hPSC-derived cortical neurons, SARS-CoV-2 infection neither increased b-gal staining nor upregulated SASP associated genes, which is consistent with our previous data showing a lack of susceptibility of cortical neurons to SARS-CoV-2 7 (Figures S3O and S3P).
Since DA neurons are one of the major cell types degenerated in PD, and DA neuron senescence has been shown to be associated with PD, 20 we next monitored how PD-iPSC-derived DA neurons responded to SARS-CoV-2 infection.SNCA, the first gene  (legend continued on next page) associated with familial PD, encodes the protein a-synuclein (a-Syn). 21Copy-number variations (CNVs) in the SNCA gene have been identified in patients with familial PD, 22,23 and SNPs in SNCA are also strongly associated with sporadic PD. 24 Here, we used isogenic SNCA 17 4 copy, 2 copy, and 0 copy iPSCderived DA neurons to determine their response to SARS-CoV-2 infection.First, SNCA copy numbers do not affect the differentiation toward MAP2 + FOXA2 + LMX1 + TH + DA neurons (Figure S4A).Immunostaining showed that SNCA 4 copy iPSC-derived DA neurons displayed higher total a-Syn expression compared with SNCA 2 copy iPSC-derived DA neurons, whereas SNCA 0 copy iPSC-derived DA neurons showed no expression when using an antibody against total a-Syn (Figure S4B).Interestingly, real-time quantitative PCR analysis detected increased levels of SARS-CoV-2 subgenomic N RNA in SNCA 4 copy iPSC-derived DA neurons compared with SNCA 2 copy and SNCA 0 copy iPSCderived DA neurons at 48 hpi (Figure S4C).Although the mechanism of increased SARS-CoV2 susceptibility in SNCA 4 copy DA neurons remains to be determined, previous studies have shown that the SARS-CoV-2 S protein can bind to TLR4, which is upregulated in PD, 25 expressed on the neuron surface, and which facilitates the viral entry. 26,27ext, we assessed senescence and disease associated phenotypes in mock and SARS-CoV-2-infected DA neurons derived from isogenic SNCA 4 copy, 2 copy, and 0 copy hiPSC lines.SARS-CoV-2 infection increased both the percentage of b-gal + cells (Figure S4D) and lysosomal accumulation (Figure S4E) in all iPSC-derived DA neurons.However, the percentage of b-gal + cells (Figure S4D) and lysosomal accumulation (Fig- ure S4E) increased progressively along with the copy number of SNCA both in mock condition and upon SARS-CoV-2 infection.SNCA 4 copy iPSC-derived DA neurons with SARS-CoV-2 infection showed the highest b-gal activity and lysosomal content (Figures S4D and S4E).There is previous evidence that a-Syn has a higher binding affinity to SARS-CoV-2 N protein and that SARS-CoV-2 proteins accelerate the aggregation of a-Syn. 28,29Western blotting confirmed that SARS-CoV-2 infection increases total a-Syn expression and SARS-CoV-2 infected SNCA 4 copy iPSC-derived DA neurons showed the highest a-Syn expression (Figure S4F).
High-throughput screening to identify drug candidates blocking SARS-CoV-2 induced senescence of DA neurons To identify drug candidates that may protect DA neurons from SARS-CoV-2-induced senescence, we screened DA neurons against a library of Food and Drug Administration (FDA)approved drugs supplied at 10 mM. 6 h post-treatment, H9 hPSC-derived DA neurons were infected with SARS-CoV-2 at MOI = 0.2 and analyzed at 72 hpi for b-gal activity.The hPSCderived DA neurons used for drug screening showed a purity of about 80% NURR1-GFP + cells (Figures S5A and S5B).Compounds with a Z score < À2 were defined as primary hit drugs (Figure 4A).The hits were further evaluated for potency and cytotoxicity at different concentrations.Three drugs, riluzole (Figures 4B and 4E), metformin (Figures 4C and 4F), and imatinib (EC 50 = 3.25 mM, CC 50 = 17.14 mM, Figures 4D and 4G), reduced b-gal activity in a dose-dependent manner without inducing cytotoxicity.Next, we confirmed the drugs' activities on DA neurons derived from the purified NURR1-GFP + cells.DA neurons treated with either 10 mM riluzole, 50 mM metformin, or 10 mM imatinib showed a significant decrease in the percentage of b-gal + cells as compared with DMSO treatment (Figures 4H and 4I).Real-time quantitative PCR analysis showed a decrease in the senescence-pathway associated gene IGFBP7 and an upregulation of LMNB1 for each of the three drugs (Figure 4J).Moreover, these three drugs also decreased lysosomal accumulation compared with DMSO-treated cells (Figures 4K and 4L).
RNA-seq analysis was applied to determine the transcriptional changes induced by the drug candidates versus DMSO in DA  (legend continued on next page) neurons upon SARS-CoV-2 infection.Plotting these datasets by PCA (Figure 4M) and by performing clustering analysis (Figures S5C-S5E) demonstrated that DA neurons treated with drug candidates + SARS-CoV-2 occupied a distinct transcriptional space compared with DMSO + SARS-CoV-2-treated DA neurons.Importantly, the genes involved in SASP phenotype (Figure 4N) and senescence pathway (Figures S5F-S5H) were downregulated in riluzole, metformin, or imatinib treated DA neurons in the presence of SARS-CoV-2 infection.

Riluzole, metformin, and imatinib blocking SARS-CoV-2 infection induced DA neuron senescence by blocking SARS-CoV-2 infection
The lead compounds might decrease senescence by blocking SARS-CoV-2 infection or by rescuing senescence pathway directly.To distinguish between these two possibilities, DA neurons were again treated with 10 mM riluzole, 50 mM metformin, or 10 mM imatinib and then infected with SARS-CoV-2 (pre-treatment).At 48 hpi, real-time quantitative PCR analysis demonstrated that riluzole, metformin, and imatinib all decreased viral RNA (Figure 5A), a finding further validated by immunostaining using an antibody against the SARS-CoV-2 nucleocapsid protein (Figures 5B and 5C).To determine the therapeutic potential of the drug candidates, we infected DA neurons with SARS-CoV-2 first, and then treated cells with 10 mM riluzole, 50 mM metformin, or 10 mM imatinib at 4 hpi (post-treatment).Consistent with pre-treatment, the candidate drugs significantly decreased both replicating viral RNA (Figure 5D) and the percentage of SARS-N + cells (Figures 5E and 5F) when applied at the post-infection time point.Moreover, these three drug candidates also blocked virus infection in MEL-1-derived DA neurons (Figures S6A-S6C).To examine if these drugs can rescue senescence phenotype independent of virus infection, we used SATB1 knockout (KO) hPSCs, which have been reported to show senescence phenotype when differentiated toward DA neurons. 8Neither of these three drugs decreased b-gal staining or significantly changed the expression of IGFBP7 and LMNB1, suggesting that these three drugs do not directly rescue the SATB1 KO-induced senescence phenotype (Figures S6D and S6E).Together, this suggests that riluzole, metformin, and imatinib prevent SARS-CoV-2 infection induced DA neuron senescence by blocking SARS-CoV-2 infection.
We further examined the molecular mechanisms of the antiviral activities of these three drugs.Our previous studies have reported imatinib can block SARS-CoV-2 viral entry in lung organoids through binding ACE2 receptor. 30Luciferase activity assay following infection of SARS-CoV-2-entry virus to DA neurons showed decreased luciferase activity in imatinib treated condition compared to mock, indicating that imatinib can block SARS-CoV-2 viral entry in DA neurons (Figure 6A).GSEA identified the downregulation of fatty acid biosynthesis pathway in riluzole-treated condition (Figures 6B and 6C).Several studies have shown the important role of fatty acid biosynthesis in SARS-CoV-2 infection, 29,31,32 suggesting that riluzole might block SARS-CoV-2 infection by inhibiting fatty acid biosynthesis.GSEA also identified the upregulation of the adenosine monophosphate-activated protein kinase (AMPK) pathway upon metformin treatment (Figures 6D and 6E).Consistent with our findings, a recent study identified metformin as a potential COVID-19 therapeutic agent and implicated increased AMPK signaling as protective to SARS-CoV2. 33We also tested the three drug candidates on hPSC-derived cortical neurons and did not detect obvious cell cytotoxicity at concentrations effective to protect DA neurons (Figure S6F).

SARS-CoV-2 is detected in autopsy substantia nigra samples of COVID-19 patients
A key question is whether the selective vulnerability of hPSCderived DA neurons and the resulting senescence and inflammatory responses are reflected in any cognate changes in the brain of human COVID-19 patients.To answer the question, we first collected human substantia nigra autopsy samples from 6 COVID-19 patients and 3 age-matched controls (cohort 1).Immunostaining for TH was used to confirm substantia nigra identity and the presence of DA neurons (cohort 1, Figure S7A).Next, we performed RNA-seq analysis on RNA isolated from formalinfixed paraffin-embedded (FFPE) autopsy samples of 6 COVID-19 patients and 3 age-matched controls.Remarkably, the same transcriptional signatures identified in SARS-CoV-2 infected DA neurons in vitro (Figures S3I, S3J, S3K, and 3C) were observed in COVID-19 autopsy samples, including the induction of chemokine/cytokine (cohort 1, Figure 7A), inflammation (cohort 1, Figure 7B), senescence-associated (cohort 1, Figure 7C) genes, and SASP associated genes (cohort 1, Figure 7D).The RNA-seq data also showed expression for several SARS-CoV-2 transcripts across the 6 substantia nigra samples from severe COVID-19 patients, compatible with the presence of virus (cohort 1, Figure 7E).Immunostaining showed the co-localization of SARS-N and DA neuron marker TH (cohort 2, Figure S7B).However, we also detected very low levels of viral RNA by realtime quantitative PCR in frozen tissue samples from other brain regions from these same autopsies (data not shown), which could potentially represent virus in leptomeningeal or intracerebral vessels. 34Immunostaining showed accumulation of p-a-Syn (cohort 1, Figure 7F) in COVID-19 samples compared with the expression in non-COVID-19 samples, which indicates a potential link of PD phenotypes within the substantia nigra of COVID-19 patients despite the lack of acute clinical symptoms specific to midbrain dysfunction. 34e also collected another set of human substantia nigra samples, including 8 controls, 13 COVID-19, and 8 PD(D) donors, to assess whether the DA neurons in the substantia nigra pars compacta are damaged or reduced in number due to SARS-CoV-2 infection (cohort 2).We quantified neuromelanin (NM)-positive neurons, TH-immunoreactive neurons, and TH-immunoreactive threads using an in-house QuPATH script.Demographics, main clinical and pathological features of the 8 controls, 13 COVID-19, and 8 PD(D) donors in cohort 2 are presented in Tables S1, S2, and S3.None of cohort 2 COVID-19 patients had obvious PD symptoms before testing positive with SARS-CoV-2 infection.The percentage of males was different between groups (50%, 84%, and 25% in controls, COVID-19, and PD(D), respectively).Age at death was not significantly different between groups (p > 0.05).Disease duration for COVID-19 was between 5 and 30 days in the COVID-19 groups, whereas the disease duration for PD in the PD(D) was between 10 and 17 years after the start of subjective complaints.SARS-CoV-2 infection was confirmed in the COVID-19 donors using a quantitative PCR test at the time of hospital submission.As by definition, the PD(D) group showed higher Braak aSyn stages, Braak neurofibrillary stages, and CERAD scores and higher levels of AD-related pathologies (all p < 0.001) than COVID-19 and control group.The density of NMcontaining and TH-immunoreactive neurons was decreased in COVID-19 and PD(D) brain donors compared with age-matched controls (NM+: À58%, p = 0.008% and À71%, p < 0.001, respectively; TH+: À44%, p = 0.035; À64%, p < 0.002).The percentage area of TH-immunoreactivity threads was also decreased in COVID-19 and PD(D) donors (TH+ À55%, p = 0.006; 60%, p = 0.006, respectively) (cohort 2, Figures 7G and 7H).Although follow-up studies in additional, independent cohorts of severe COVID-19 patients will be needed to further corroborate our results, the overall findings of our study on the selective vulnerability of hPSC-derived DA neurons in vitro, the associated inflammatory and cell senescence responses observed in DA neurons in vitro, and the inflammatory, cell senescence, and potentially degenerative responses in COVD-19 patient samples in vivo argue that these results may be of clinical relevance.

DISCUSSION
Advancements in hPSC technology allow for the study of hostvirus interactions in human, disease-relevant cells. 354][5][6] Here, we differentiated DA neurons from hPSCs using a published protocol. 8,9Using NURR1-GFP + sorted cells at day 25 of the differentiation, we confirmed that SARS-CoV-2 can infect DA neurons.Since the majority of the cells are DA neurons as indicated by immunostaining, single-cell real-time quantitative PCR, and scRNAseq, we do not expect other cell types to meaningfully contribute to the reported senescence phenotypes.scRNA-seq analysis indicates that SARS-CoV-2 infects DA neurons characterized by high expression of LMO3 and a lack of CALB1 expression.Therefore, DA neurons with an A9-like subtype identity may be particularly vulnerable to SARS-CoV-2, similar to the vulnerability of A9 neurons in the substantia nigra most affected in PD. 12 Here, we report that SARS-CoV-2 infection triggers cellular senescence in DA neurons.Previous work indicates that senescence of DA neurons can function as a contributing factor in PD pathogenesis. 8As DA neuron dysfunction is also linked to lethargy and anhedonism, 36 its role in the post-COVID lethargy/syndrome or long COVID may deserve further study.
Interestingly, SNCA 4 copy iPSC-derived DA neurons show increased SARS-CoV-2 infection compared with the isogenic 2 copy or 0 copy iPSC-derived DA neurons.This might be due to the increased TLR4 expression, which has been reported to facilitate viral entry. 26,27,37Consistently, SNCA 4 copy iPSCderived DA neurons infected by SARS-CoV-2 showed the highest b-gal activity, lysosome accumulation, and total a-Syn, suggesting the potential synergistic or add-on activities of SARS-CoV-2 infection and PD causal mutations.
The FDA-approved drugs, riluzole, metformin, and imatinib, were identified to block SARS-CoV-2-mediated DA neuron senescence, which might function through inhibiting viral infection.Although imatinib was also identified to block SARS-CoV-2 entry in our hPSC-derived lung organoid-based screen, 30 riluzole has not been previously linked to SARS-CoV-2 infection.The use of metformin has been associated with a decrease in the mortality of COVID-19 patients with obesity and/or type 2 diabetes. 38,39Recently, metformin has been identified as a potential COVID-19 therapeutic agent, and we found it can activate AMPK signaling pathway in DA neurons, which is consistent with recently published papers. 33However, we cannot exclude other, yet unknown, mechanisms for the anti-viral activities of these three drugs.Future studies are needed to dissect the mode of action and translational potential of these compounds and whether they are capable of reversing infection-induced neuropathology.
Overall, our data highlight DA neurons as a possible target for SARS-CoV-2 infection, which in turn may trigger an inflammatory and cellular senescence response.Although we observed a comparable inflammatory and senescence signature in SARS-CoV-2 infected hPSC-derived DA neuron cultures in vitro and in autopsy samples in vivo (cohort 1), we cannot exclude the possibility that other cell types such as astrocytes or microglia or other pathological changes such as hypoxic state could contribute to the inflammatory and senescence signatures in the substantia nigra samples.Given our findings, we posit that over the coming years, there is a need to closely monitor COVID-19 patients for an increased risk of developing PDrelated symptoms.

Limitations of the study
The detection of SARS-CoV-2 viral antigen in brain autopsy samples is highly controversial.Although we are currently still trying to generate EM to assess evidence for direct viral infection in substantia nigra neurons, those studies are technically extremely challenging given the low number of DA neurons and difficulties in sample availability and quality.In addition, due to the biosafety requirements of COVID-19 autopsy samples, we are not capable of performing extensive analysis, such as scRNA-seq, on autopsy samples of COVID-19 patients.The co-staining of the TH neuron marker and SARS-N viral antigen was initially conducted on an autopsy sample, but further validation with a larger sample size is necessary to confirm these findings and to rule out technical challenges such as senescence-related lipofuscin expression that could interfere with reliable SARS-N detection.Furthermore, our study represents the first detailed, quantitative pathological analysis of midbrain DA neurons in severe COVID-19 patients and is limited to the cohort 2 of patients presented here, representing the first wave of COVID-19 cases.Additional analyses using independent and more recent cohorts of patients are needed to further validate the inflammatory and degenerative phenotypes observed.Finally, by focusing on cellular senescence in 72-h infection model, there is the possibility that the observed phenotypes may be selflimiting, arising in response to acute stress, and potentially resolving with cessation of infection.
In this study, we have shown that DA neurons can be infected by SARS-CoV-2 virus.However, only a small percentage of cells ($5%) are infected at the experimental endpoint (48 hpi).This may be due to bystander effects, as we cannot exclude the possibility that the production of interferons by the initially infected cells protects surrounding cells against subsequent rounds of infections.Nevertheless, infected DA neurons exhibit a senescence phenotype that results in the secretion of toxic cytokines, which may trigger a SASP.SASP may further cause inflammation and disrupt tissue structure and function locally and contribute to chronic inflammation throughout the body.However, the clinical relevance the level of individual patients is as of yet unknown given that the infection appears to affect only a relatively small percentage of DA neurons.
For infectivity measurement, Vero-ACE2-TMPRSS2 cells were seeded at 20,000 cells/well and incubated at 37 C/5% CO 2 overnight.Supernatants collected at indicated time points were thawed from -80 C. Serial dilutions of the supernatant were performed from 1:2 dilution to 1:2048 dilution.100ul of dilutions were overlaid on the overnight monolayer of Vero-ACE2-TMPRSS2 cells.Cells were incubated with virus at 37 C/5%CO 2 for 70h.Endpoint titers were calculated for each sample by visualizing virus cytopathic effects using light microscopy (ECHO Revolve: 10x objective).Vero-ACE2-TMPRSS2 infected with MOI 0.1 of WA1 virus isolates were used as positive controls for the assay.

b-Gal Staining
The identification of senescent cells is based on an increased level of b-galactosidase activity.The assay followed Senescence b-Galactosidase Staining Kit (#9860, CST).

Western blotting
Cells were collected in Pierce RIPA buffer (Thermo Fisher Scientific) plus HALT protease inhibitor cocktail (1:100) (Thermo Fisher Scientific) and lysates loaded on 12% NuPage Bis-Tris pre-cast gels (Thermo Fisher Scientific).After separation by electrophoresis, proteins were transferred to 0.2 mm nitrocellulose membranes (Thermo Fisher Scientific).Membranes were blocked with 5% milk in TBS +0.1% Tween and incubated with primary antibody overnight.Information for primary antibodies is provided in Table S5.Membranes were washed and incubated with secondary antibody for 1 h at room temperature in 5% milk-TBS-0.1% Tween and developed using Super-Signal West Pico PLUS chemiluminescent substrate (Thermo Fisher Scientific).Human b-Actin was employed as an internal reference.

Immunohistochemistry
In brief, paraffin-embedded tissue blocks of the SN were cut into 6 mm-thick consecutive sections with a microtome and incubated with mouse-anti tyrosine hydroxylase (TH, dilution 1:800 in TBS-Trition 0.1%, Immunostar, Hudson, USA) at 4 C overnight.TH immunoreactivity was visualized with Vector SG grey (Vector, California, United State) followed by counter-staining with fast red (Vector, California, United State), and dehydrated in a series of ethanol, xylene, then mounted with Entellan.

Tissue delineation and image analysis
Images were taken using a whole-slide scanner (Vectra Polaris, 20x objective) and quantified using QuPath 0.2.3 (https://qupath.readthedocs.io/en/stable/index.html).The SN was delineated according to the anatomical landmarks on TH-stained sections which were counterstained with fast-red.Briefly, the SN was outlined based on the coordinates from the Atlas of Human Brainstem and previous literature (Paxinos and Huang, 1995 42 ; Parkkinen et al., 2011 43 ; Dijkstra et al., 2014 44 ).In brief, the dorsal border of SN was delineated ventrally to the red nucleus and along the white matter tracts with extension to the dorsal-lateral part of the cerebral peduncle and medially reaching the ventral tegmental area (VTA).The ventral border of SN was depicted by the cerebral peduncle, connecting both the lateral and medial end of the dorsal border.
To analyze the TH immunoreactivity, we used an in-house QuPath script.Briefly, an object classifier was used to identify and count TH positive cells and neuromelanin only cells (without TH staining).Once counted, they were subtracted from the annotation, and then a pixel classifier was used to quantify the % area occupied by TH threads.The following outcome measures were defined for the SN: (1) TH positive cell count, (2) neuromelanin only cell count and (3) TH threads % area.

Real-time quantitative PCR
Total RNA samples were prepared from cells and DNase I treated using TRIzol according to the manufacturer's instructions.To quantify viral replication, measured by the expression of sgRNA transcription of the viral N gene, one-step real-time quantitative PCR was performed using SuperScript III Platinum SYBR Green One-Step qRT-PCR Kit (Invitrogen) with primers specific for the TRS-L and TRS-B sites for the N gene as well as ACTB as an internal reference.Real-time quantitative PCR reactions were performed on an Applied Biosystems QuantStudio 6 Flex Real-Time PCR Instrument (ABI).Delta-delta-cycle threshold (DDCT) was determined relative to ACTB levels and normalized to mock infected samples.Error bars indicate the standard deviation of the mean from three biological replicates.The sequences of primers/probes are provided in Table S6.

RNA-Seq
Cell infections were performed at the described MOI in DMEM supplemented with 0.3% BSA, 4.5 g/L D-glucose, 4 mM L-glutamine and 1 mg/ml TPCKtrypsin and harvested 24 hpi.Total RNA was extracted in TRIzol (Invitrogen) according to the manufacturer's instructions.RNAseq libraries of polyadenylated RNA were prepared using the TruSeq Stranded mRNA Library Prep Kit (Illumina) according to the manufacturer's instructions and sequenced on an Illumina NextSeq 500 platform.The resulting single end reads were checked for quality (FastQC v0.11.5) and processed using the Digital Expression Explorer 2 (DEE2) 45 workflow.Adapter trimming was performed with Skewer (v0.2.2). 46 Further quality control done with Minion, part of the Kraken package. 47The resultant filtered reads were mapped to human reference genome GRCh38 using STAR aligner 48 and gene-wise expression counts generated using the ''-quantMode GeneCounts'' parameter.BigWig files were generated using the bamCoverage function in deepTools2 (v.3.3.0). 49or RNA preparation with human exome enrichment, total RNA samples were prepared from formalin-fixed and paraffinembedded autopsy substantia nigra tissues followed by DNaseI treatment using manufacturer's instructions (Qiagen RNeasy FFPE kit Cat# 73604).100 ng total RNA was prepared using NEB Next Ultra II RNA Library Prep Kit without polyA selection or RNA depletion, then the libraries were enriched with twist human exome probes and reagents.
For RNA preparation with COVID-19 panel enrichment, 100ng total RNA was prepared using NEB Next Ultra II RNA Library Prep Kit without polyA selection or RNA depletion, then the libraries were enriched with IDT COVID-19 Capture Panel probes and reagents.
For analysis, the human genome GRCh38 was used to generate the hisat2 index and then alignment from the raw fastq file.The quantification of reads for each individual gene were counted using feartureCounts.Then the clustering and differential gene expression analysis were using the edgeR package after import the expression and phenotype data in R.

Single cell RNA-Seq
The mock and SARS-CoV-2 infected hPSC-derived DA neuron cells were dissociated into single cells using Accutase at 37 C for about 30 min, and then neutralized with culture medium.The dissociated cells were pelleted and resuspended in PBS with 0.04% BSA.The resuspended cells were then placed through a 40 mm filter to obtain a single cell suspension.The single cell suspension was processed with the Chromium Single Cell 5' Reagent Kit v3 (10x Genomics, # 1000263) using a 10X Genomics Chromium Controller following the instruction of the user guide.The final libraries were assessed by Agilent Technology 2100 Bioanalyzer and sequenced on Illumina NovaSeq sequencer with pair-end 2x100 cycle kit (28+10+10+90).
Sequencing and gene expression UMI count matrix generation The FASTQ files were imported to a 10X -data analysis pipeline (v3.0.2) to align reads, generate feature-barcode matrices and perform clustering and gene expression analysis.In a first step, cellranger mkfastq demultiplexed samples and generated fastq files; in the second step, cellranger count aligned fastq files to the reference genome and extracted gene expression UMI counts matrix.The expression matrix was then imported into Seurat for further clustering analysis.To exclude poor quality cells that might result from cell death or other membrane damages, we filtered cells based on the number of expressed genes detected, the sum of UMI counts and the proportion of mitochondrial genes.We visualized QC metrics and used these to filter cells.We filtered cells that have unique feature counts over 6000 or less than 200 and the cells that have >5% mitochondrial counts.Finally, 4702 cells which passed the quality control were used for the following analysis.We normalized the sum of UMI counts for each cell to the median of all cells.After normalization, the UMI count data were log-transformed.Then, the data were centered for each gene by subtracting the average expression of that gene across all cells.The top enriched genes in each cluster were used to identity the cell types.And the expression levels of selected genes were visualized and demonstrated with the function Vlnplot, FeaturePlot and Dotplot.

In situ hybridization
Adherent cells plated in a glass-bottom plate are fixed and permeabilized and stained for a protein of interest (TH; tyrosine hydroxylase) in order to locate RNA puncta signals within a mature DA neuron.Following protein detection, a fluorescent in situ hybridization (FISH) and branched DNA amplification technology is used to amplify the signal detection of an RNA transcript.In the first step, a gene-specific oligonucleotide target probe binds to the target RNA sequence.Signal amplification is then achieved through a series of sequential hybridization steps.After two sequential amplifying steps, a fluorescent dye is introduced to hybridize to their corresponding amplifier molecules.RNA signals in dots are visualized using confocal microscopy with 63X oil lenses.All the images in z-stacks were projected and obtained using Imaris software.Projected images were analyzed for quantification.
High Throughput Chemical Screening hPSC-derived DA neurons were cultured in 384-well plates at 10,000 cells/50 ml medium/well until Day 40.Compounds from an inhouse FDA-approved drug library (Prestwick) were added at 10 mM.DMSO treatment was used as a negative control.hPSC-derived DA neurons were further infected with SARS-CoV-2 (MOI=0.1).After 72 hpi, hPSC-derived DA neurons were harvested for b-galactosidase assay using Senescence b-Galactosidase Staining Kit (#9860, CST) protocol.
To calculate EC50 and CC50, cells were stained with b-Galactosidase Staining Kit and normalized to DMSO-treated condition.To calculate CC50, the cell survival was monitored by DAPI and normalized to DMSO-treated condition.The efficacy and cytotoxicity curves were calculated using Prism GraphPad Prism 7.0.

QUANTIFICATION AND STATISTICAL ANALYSIS
N=3 independent biological replicates were used for all experiments unless otherwise indicated.Data was presented as mean ± STDEV.For the comparison between two samples, P values were calculated by unpaired two-tailed Student's t test unless otherwise indicated.For the comparison between DMSO or drug candidates-treated samples, P values were calculated by one-way ANOVA using Dunnett's test with a set up control.To examine between-group differences regarding clinical and pathological features, we used independent sample t-tests for continuous variables, chi-square tests for categorical variables and Kruskall-Wallis tests for ordinal variables.Group differences in the neuromelanin neuron density, TH density and %TH threads were tested using a KrusKall-Wallis test.Data analysis was performed using GraphPrism version 9.5.1.n.s.indicates a non-significant difference.*p<0.05,**p<0.01 and ***p<0.001.

Figure 4 .
Figure 4. Riluzole, metformin, and imatinib rescue SARS-CoV-2 induced senescence of DA neurons (A) Primary screening results.x axis is the compound number.y axis is the Z score.Red line is Z score < À2, which means the luminescent signal is lower than average-2 3 SD.

Figure 5 .
Figure 5. Riluzole, metformin, and imatinib block SARS-CoV-2 infection (A) Real-time quantitative PCR analysis of total RNA from DMSO or drug candidates-pre-treated purified NURR1:GFP H9-DA neurons following SARS-CoV-2 infection (MOI = 0.2) for viral N sgRNA at 48 hpi.The graph depicts the mean sgRNA level normalized to ACTB.n = 3 independent biological replicates.(B and C) Representative confocal images (B) and quantification of the percentage of SARS-N + cells (C) of DMSO or drug candidates-pre-treated purified NURR1:GFP H9-DA neurons following SARS-CoV-2 infection (MOI = 0.2) at 72 hpi.Scale bars, 50 mm.n = 3 independent biological replicates.(D) Real-time quantitative PCR analysis of total RNA from DMSO or drug candidates-post-treated purified NURR1:GFP H9-derived DA neurons after SARS-CoV-2 infection (MOI = 0.2) for viral N sgRNA at 48 hpi.The graph depicts the mean sgRNA level normalized to ACTB.n = 3 independent biological replicates.(E and F) Representative confocal images (E) and quantification of the percentage of SARS-N + cells (F) of DMSO or drug candidates-post-treated purified NURR1:GFP H9-DA neurons after SARS-CoV-2 infection (MOI = 0.2) at 72 hpi.Scale bars, 50 mm.n = 3 independent biological replicates.Data were presented as mean ± SD. p values were calculated by one-way ANOVA using Dunnett's test with a setup control.***p < 0.001.See also Figure S6.