ABSTRACT
Neuropsychiatric manifestations are common in both acute and post-acute phase of SARS-CoV-2 infection, but the mechanism of these effects is unknown. Here, we derive human brain organoids with innately developing microglia to investigate the cellular responses to SARS-CoV-2 infection on a single cell level. We find evidence of limited tropism to SARS-CoV-2 for all major cell types and observe extensive neuronal cell death that also include non-infected cells. Single cell transcriptome profiling reveals distinct responses in microglia and astrocytes that share features with cellular states observed in neurodegenerative diseases, includes upregulation of genes with relevance for synaptic stripping, and suggests altered blood brain barrier integrity. Across all cell types, we observe a global translational shut-down as well as altered carbohydrate metabolism and cellular respiration. Together, our findings provide insights into cellular responses of the resident brain immune cells to SARS-CoV-2 and pinpoint mechanisms that may be of relevance for the neuropathological changes observed in COVID-19 patients.
INTRODUCTION
A significant proportion of patients infected with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the causative agent of coronavirus disease 2019 (COVID-19), display acute neurological and psychiatric symptoms1, with autopsy studies detecting the presence of SARS-CoV-2 RNA and protein in brain tissue of around half of the examined COVID-19 patients, despite low expression in neural cells of the major receptor (angiotensin-converting enzyme 2; ACE2,) for SARS-CoV-2 entry.2 Survivors commonly report persistent central nervous system (CNS)-related symptoms, such as memory loss, sleep disturbances3, and depressive symptoms,4 and a subset demonstrate structural brain abnormalities.5 Further, the risk of receiving a neurological or psychiatric diagnosis related to cognitive impairment or behavioral abnormalities, including dementia, psychosis, and major depressive disorder, is increased up to at least 6 months after infection.3, 6, 7 Long-lasting neurobehavioral and cognitive impairments are also common features of other RNA viruses with a neuro-invasive potential and the CNS-damaging processes can also continue after virus elimination.8 So far, the mechanisms causing neurocognitive sequelae after a viral encephalitis are poorly understood, but recent studies in rodents suggest that microglia, the resident immune cells of the brain parenchyma, play a pivotal role by inducing neuronal apoptosis and stripping neurons of synaptic terminals through complement-dependent signaling.9, 10
Reprogrammed induced pluripotent stem cells (iPSCs) from easily-accessible human somatic cells can be used to derive 3D cellular aggregates in vitro referred to as organoids, which aim to provide mechanistic models that more precisely capture the broad repertoire of cell-cell interactions and complex spatial patterns that define the functions of complex tissues and organs such as the human brain.11–13 Brain organoids have also been successfully employed to study neurotropism and neurotoxic effects of viruses such as the Zika virus,14, 15 and now more recently SARS-CoV-2.16–,20 However, as microglia are of non-ectodermal origin,21, 22 and as the predominate germ layer fate in embryoid bodies is ectodermal, brain organoids typically lack innately developing microglia. This has limited the mechanistic understanding of microglial cellular responses in viral CNS infection, such as in COVID-19, as well as the understanding of the cellular responses of astrocytes and neurons in the context of surveying microglia. To address this limitation, we applied a recently-developed protocol that generates brain organoids containing innately-developing microglia,23 adapted it to miniaturized bioreactors, and used it to characterize the temporal dynamics of single cell responses in a human experimental model upon SARS-CoV-2 infection.
RESULTS
SARS-CoV-2 neurotropism and cell death in brain organoids containing resident immune cells
With modifications (see Methods and Supplementary Fig. 1a), the protocol reported by Ormel et al.23 was used to generate brain organoids. At day 66, these undirected organoids were found to contain PAX6+ cells (neural progenitor cells; NPCs), β-tubulin+ cells (neurons), SOX10+ cells (neural crest), GFAP+ cells (astrocytes), and IBA1+ cells (microglia); Supplementary Fig. 1b-e. We exposed the organoids to active SARS-CoV-2 virus for 2h at an estimated multiplicity of infection (MOI) of 0.3, after which they were washed and transferred into fresh media to monitor the course of infection. Supernatants were harvested at 6, 24, 48, 72 hours post-infection (hpi) for viral replication assessment. Viral RNA-dependent RNA polymerase (Rdrp) and nucleocapsid (N) transcription were measured in supernatants via RT-qPCR and we observed a time dependent increase for both the transcripts (Fig. 1a and Supplementary Fig. 2a), indicating that the virus is capable of initiating RNA replication of its genomic material within the cells of the organoid, but at a slower pace as compared to Vero-E6 cells (Supplementary Fig. 2a). To validate that SARS-CoV-2 could assemble and release complete viral particles in the organoids, we performed plaque formation assays using supernatants from infected organoids and observed release of infectious viral particles as early as 6hpi, although the number of plaque forming units (PFUs) then promptly decreased (Fig. 1b).
At two time points (24 and 72hpi), we then fixed infected and uninfected organoids (day 66) for immunohistochemical (IHC) analyses. At 24hpi, a strong signal for the viral nucleocapsid protein (NP) was observed mainly in the periphery of the infected organoids, whereas at 72hpi, we observed NP staining in cellular cytoplasm or processes of infected cells (Fig. 1c). On average, around 3% of the cells in the superficial layers of the organoids were found to be NP-positive at 72hpi (Supplementary Fig. 2c). Caspase3 (Casp3) staining revealed extensive cell death in the infected organoids as compared to the uninfected organoids, occurring already at 24hpi (Fig. 1d). Further, Casp3+ cells were not just limited to superficial layers like NP+ cells (Supplementary Fig. 2b), and the number of Casp3+ cells greatly exceeded the number of NP+ cells at both timepoints (Supplementary Fig. 2c-d), then indicating viral-related cell death beyond toxicity occurring as a result of direct infection.
Next, we investigated infectivity and cell death in the context of cellular identity. IHC staining on infected organoids at 72hpi showed viral NP overlapping with PAX6+, NeuN+ (neuronal lineage), and GFAP+ cells (astrocytic), in addition to SOX10+ cells (Neural crest/schwann cells), and OLIG2+ cells (glial progenitors); Fig. 2a-d). Casp3+ staining was foremost observed in neurons (Supplementary Fig. 2f), but also included other cell types (Supplementary Fig. 2g). In addition, we detected IBA1+ cells (i.e., microglia) that stained positive for NP (Fig. 2e) although the scarcity of these cells (slightly below 1% on average) obstructed meaningful quantifications. Therefore, we enriched for microglia from two infected organoids using magnetic-activated cell sorting (MACS) and CD11b beads. qRT-PCR on cell lysates then confirmed the presence of immature microglia (AIF1 expression, no TMEM119 expression; Supplementary Fig. 2e) including N gene copies (Fig. 2f). Further, we derived induced human microglia-like cells (iMGs) in 2D-culture,24 and exposed these cells to live SARS-CoV-2 virus (MOI=0.01). At 24hpi, we detected dsRNA+ cells (Fig. 2g), indicating the presence of virus inside iMGs.
Single cell characterization of brain organoids containing resident immune cells
Droplet-based encapsulation of single cells using the 10X Genomics chromium system allowed us to profile and compare transcriptomes of individual cells isolated from SARS-CoV-2 infected brain organoids as compared to uninfected control organoids. Given the extent of cell death in the organoids exposed to an estimated MOI of 0.3, we decreased the MOI to 0.1. Further, to achieve more mature microglia, we cultured the organoids up to 160 days. Given the low microglia counts in day 66 organoids we also enriched for CD11b-positive cells (MACS). In this way we were able to obtain good-quality transcriptomic data for fresh cells isolated from three experimental conditions: uninfected control organoids (7257 cells), 24hpi organoids (15254 cells), and 72hpi organoids (2825 cells), respectively (Supplementary Fig. 3a). Then, we integrated pre-processed data from each condition and performed unsupervised graph-based clustering to obtain cellular clusters with similar transcriptomic identities across conditions (Fig. 3a, Supplementary Fig. 3b). Supervised inspection of top differentially expressed genes (DEGs) per cluster and comparison to transcriptomes of cells from the developing human brain confirmed clusters comprised of radial glia (RG) (cyclingRG: EGFR, TOP2A, MKI67 (Supplementary Fig. 3c); RG1: EGFR, OLIG2, OLIG1; RG2: NES, and oRG: HOPX, FABP7), intermediate progenitors (IPCs: PAX6, EOMES), immature (GAD2, PAX6) and mature neurons (DCX, SLC17A7) as well as more mature microglia (AIF1, LAPTM5, P2RY12), astrocytes, (GFAP, AQP4), choroid-plexus cells (TTR), perivascular cells (DCN, LUM), and endothelial precursors (CLDN5) (Fig. 3b-c, Supplementary table 1a). We also confirmed the transcriptomic identity of microglia derived from organoids by correlating to single-cell data of primary microglia (Fig. 3d-e).25–27
Confirming the expression of SARS-CoV-2 entry factors
We then assessed the expression of previously identified entry factors for SARS-CoV-2 in the generated single cell data. Cells belonging to choroid plexus and a subset of neurons (expressing midbrain dopaminergic markers) showed relatively higher ACE2 expression, although overall basal RNA expression and protein levels were relatively low (Fig. 4a-b and Supplementary Fig. 3d). Similarly, proteases involved in viral S protein priming 28 (TMPRSS2 and TMPRSS4) showed low expression, whereas FURIN 29 and infection potentiating factor24 (NRP1, Neuropilin 1) had a wider cell type distribution which included astrocytes, perivascular cells, endothelial cells and microglia (Fig. 4b). Microglia also expressed other predicted entry factors (CTSL and CTSB, Cathepsin L and B), purported to act as substitutes for TMPRSS225 (Fig. 4b). Next, we identified infected cells by aligning cellular viral transcripts to the whole SARS-CoV-2 genome in the infected conditions. As expected, given the decreased MOI, the percentage of infected cells were low (0.1-0.2% of sequenced cells), although we were able to qualitatively identify infected neurons, radial glia, perivascular and choroid plexus cell types. For further downstream analyses, we did not make any distinction between infected and non-infected cells, instead focusing on overall cellular responses in the infected conditions.
Microglial responses to SARS-CoV-2 infection
To define microglial responses to SARS-CoV-2 exposure, we first performed pseudo-bulk differential gene expression testing across the three experimental conditions, i.e., the uninfected control condition, 24hpi and 72hpi. An unbiased hierarchical clustering approach generated two unique groups of DEGs (labeled module A and B). Module A comprised of genes whose expression on average decreased at 24hpi and then at 72hpi displayed a more pronounced increase, while module B genes on average displayed reduced expression at 24hpi and a more dramatic decrease at 72hpi (Fig. 4c). While module B was largely comprised of genes encoding for cytosolic ribosomal proteins (Supplementary Fig. 4a), we detected several interferon-stimulated genes (ISGs) in module A such as ISG15, MX1, MX2, OAS2, OAS3, RSAD2 and BST2 (Fig. 4c, Supplementary Fig. 4b). Nonetheless, we detected no clear upregulation of type I or type II interferon genes (IFNA, IFNB) in microglia, at either 24hpi or at 72hpi, although several pathogen sensors and upstream effectors of interferon signaling were upregulated including nuclear factor kappa B (NFKB1) (Fig. 4d). At 24hpi, we detected upregulation of pathways (such as oxidative phosphorylation, release of calcium into cytosol, monovalent inorganic ion transport, lysosomal disruption and hypoxia; Supplementary Table 2) that can serve as activating stimuli for primed inflammasome components (NLRP3, PYCARD, CASP1), and are required for the production and maturation of pro-inflammatory cytokines (Fig. 4d).30 Taken together, this result suggests that microglia, without the support of infiltrating myeloid cells, respond with an incomplete antiviral type-I and type-II interferon response to SARS-CoV-2 exposure.
In the infected conditions, we also observed a significant upregulation of pathways related to neurodegenerative diseases, including Alzheimer’s disease (AD) and Parkinson disease (PD) (Supplementary Fig. 4c). This led us to more closely examine the microglial transcriptional state after exposure to SARS-CoV-2 in relation to previously described disease-associated microglial activation states. First, we compared SARS-CoV-2-exposed microglia to such activation states observed in experimental murine models. Injury responsive microglia (IRM) have been described in mice close to experimentally induced focal demyelination (7 days post lesion) and are to a large extent defined by upregulation of ISGs,31, 32 while the disease-associated microglia (DAM) described in mouse models of AD and ALS are thought to restrict neurodegeneration and disease development.33 SARS-CoV-2 exposed microglia then displayed a significant enrichment for genes associated with the IRM activation state (p=6.9×10-24, Fig. 4e) and less so for the unique DAM signature genes (p=0.11), while still displaying upregulation of genes implicated in AD such as APOE, CTSZ and GRN with a non-significant decrease of core microglial homeostasis genes (P2RY12, CX3CR1, P2RY13, CD33, TXNIP; Supplementary Fig. 4d). In fact, APOE and CTSZ were observed to be commonly dysregulated across all three signatures (IRM, DAM and SARS-CoV-2), then suggesting unique as well as shared activation mechanisms across these microglial states.
Further, we compared the SARS-CoV-2 microglial transcriptomic signature to previously identified human microglial populations identified in autopsy and surgical brain tissues obtained from subjects in the Memory and Aging Project (MAP),34 and found an enrichment with genes in so-called interferon responsive microglia (cluster 4 in this dataset) (Fig. 4f), defined by ISG15 and with increased expression of multiple sclerosis (MS) and AD susceptibility genes. To exclude effects of tissue processing, we also compared microglia from the organoids with a cluster defined by general cellular distress due to tissue processing (cluster 3) but observed no enrichment (Fig. 4f).
Actin-cytoskeletal remodeling pathways, essential for promoting migration and phagocytosis,35 were also upregulated in microglia exposed to SARS-CoV-2 (Fig. 4g, Supplementary Fig. 4c). Consistent with this upregulation in microglia, and the extensive neuronal cell death in infected organoids, even non-infected neurons at 72hpi downregulated the expression of ‘don’t-eat-me’ signals such as CD46 and CD200 (Supplementary Fig. 4d).36 Consequently, we also observed increased microglial expression of genes associated with microglia-mediated phagocytosis,37–40 such as CD68, TREM2, ITGB5, CD47, MSR1, CALR, as well as genes previously identified to be directly involved in pathological synaptic stripping of viral encephalitis such as C3, C3AR1 and FCGR3A8 (Supplementary Fig. 4e).
In summary, this suggest that microglia in response to SARS-CoV-2 exposure adopt transcriptional states that are unique but also overlap with signals observed in neurodegenerative diseases, and that these states are likely to include both protective — e.g., clearing of debris and defense as well as detrimental — e.g., inducing excessive neuronal apoptosis and synaptic stripping responses.
Astrocytic responses to SARS-CoV-2 infection
In response to SARS-CoV-2, astrocytes displayed DEGs enriched for mechanisms involved in cell cycle, carbon metabolism and disorders such as Huntington’s disease (HD) and PD (Fig. 5a). Already at 24hpi, three subclusters of astrocytes could be identified: AS-0, AS-3 and AS-4 (Fig. 5b). AS-0 cells expressed relatively higher levels of anti-viral ISGs such as IFIT1, ISG15, and SCRG1, while AS-3 and AS-4 cells had a proliferative profile (TOP2A, MKI67) and exhibited lower expression levels of GFAP (Fig. 5b-d). Subcluster AS-2, comprising of cells sampled at 72hpi, showed elevated levels of GFAP and STAT3, indicative of reactive astrogliosis,41 along with phagocytosis related genes such as MEGF8 and ABCA1 (Fig. 5c). We also observed upregulation of metal ion (zinc, copper, iron) homeostasis pathways involving metallotheionins (MT2A, MT1X, MT1E) in AS-2 astrocytes at 72hpi (Fig. 5d).
In neurodegenerative diseases, astrocytes have been shown to express markers of a putative reactive neurotoxic state commonly referred to as A1.42 Given astrocytic DEGs from infected organoids were enriched for such diseases, we integrated single-cell transcriptomic data from A1 astrocytes (stimulated with IL-1A, TNF-A and C1q)43 and SARS-CoV-2 exposed astrocytes via canonical correlation analysis (CCA). Although the majority of SARS-CoV-2 exposed astrocytes (AS-0, AS-1, AS-2) clustered separately from A1 astrocytes, the proliferative clusters (AS-3 & AS-4), as well as a subset of AS-0 astrocytes showed clustering similarities with a portion of A1 astrocytes (Fig. 5d). Furthermore, higher expression levels of metallotheionins, as in AS-2 cluster (72hpi), have been observed in astrocytes from HD patients.44 In summary, similar to the observed effects in microglia, we observe responses in astrocytes to SARS-CoV-2 exposure that share features with cellular states observed in neurodegenerative diseases, but also a strong temporal heterogeneity that includes unique cellular states.
Signaling impacting barrier integrity and responses in choroid plexus cell types
Potential entry routes to the CNS for SARS-CoV-2 includes the blood-brain barrier (BBB) or the blood-cerebrospinal-fluid-barrier (BCSFB), of which both involve cell types such as endothelial cells, pericytes, astrocytes and choroid plexus-epithelium. Upon encountering pathogens, these barrier cells act in concert to activate and regulate signal transduction pathways that aid invasion of peripheral immune cells and restoring of CNS homeostasis.45 Correspondingly, at 24hpi we found increased ligand-receptor communication between astrocytes, neurons, microglia, perivascular, and endothelial cells (Fig. 6a), whereas at 72hpi, both microglia and neurons significantly reduced their communication with most other cell types, while choroid plexus cell types gained interactions (Fig. 6b). Sub-clustering of choroid plexus related cells (Fig. 6c) showed two control enriched clusters (CP-0 and CP-2), consisting of genes involved in maintaining solute homeostasis (AQP1), barrier or extracellular matrix (ECM) integrity (TIMP1, LGALS1, COL1A2, COL2A1, COL3A1) and amyloid clearance (TTR, MT3), amongst others (Fig. 6d). At 24hpi, we observed an enrichment of metallotheionins (MT1M, MT1E, MT1X, MT1F, MT2A: CP-6), similar to what we observed in astrocytes at 72hpi, whereas genes involved in viral defense response (SAMHD1), regulation of reactive oxygen species (ROS) (SOD3) and ECM organization (FBLN1, SYNE1) were enriched at 72hpi (CP-5) (Fig. 6d). In addition to choroid plexus, differential expression analysis showed significant increase in expression of matrix metalloproteases and ECM-regulatory enzymes (MMP14, MMP9, LOX) at 72hpi in microglia, astrocytes and perivascular cells, whereas endothelial cells reduced expression levels of solute carrier SLC2A1 (Fig. 6e), all in agreement with signaling that promotes compromised CNS barriers. Astrocytes, pericytes and choroid plexus related cells were also observed to downregulate VEGF signaling, a potent inducer of BBB permeability, at 24hpi but upregulated it at 72hpi, coinciding with increased signaling for leukocyte chemotaxis and activation in astrocytes, as well as antigen processing and presentation in choroid plexus (Fig. 6e). In aggregate, this result suggests that cell types implicated in barrier functions change their communication upon SARS-CoV-2 infection, indicative of compromised integrity, as well as altered secretome and signaling with peripheral immune cells.
Metabolic dysregulation across cell types
Protein accumulation in the endoplasmic reticulum (ER) following viral infection or excessive production of secretory proteins can induce ER stress and initiate counter measures in the form of reduced translation and export to ER along with unfolded protein responses (UPRs).46 In exposed organoids, we detected an upregulation of pathways related to UPRs (IRE1, PERK), ER stress, proteasomal degradation and autophagy, among cell types such as microglia, astrocytes, choroid plexus and perivascular cells (Supplementary Table 2). Further, a partial translational inhibition could be observed in microglia (module B, as discussed before), as well as most of the other cell types (Supplementary Table 2, Fig. 4g). We also observed a downregulation of the ROS metabolism across cell types, which can lead to dysregulation of the antioxidant cellular systems and result in oxidative stress.47 SARS-CoV-2 then relies on excessive glucose for its replication,48 and shunts glucose from glycolytic pathways to fuel oxidative phosphorylation and ATP production in a process known as the Warburg effect.49 In our dataset, pathways pertaining to cellular respiration (oxidative phosphorylation, electron transport chain coupled to ATP synthesis) were enriched at 24hpi in neurons, while those of glucose and pyruvate metabolism were downregulated (Supplementary Table 2). Furthermore, transcription of glucose transporter, GLUT1 (SLC2A1), lactate dehydrogenase (LDHA) along with monocarboxylate transporter, MCT4 (SLC16A3), was downregulated in astrocytes at 24 and 72hpi, indicating deficits in glucose export and redox cycling necessary for neuronal energization50 (Supplementary Table 2). This suggests that following exposure to SARS-CoV-2, CNS cell types undergo major stress that results in metabolic changes which are likely to impact cellular communication such as the important interplay between astrocytes and neurons.
DISCUSSION
Using brain organoids with innately developing microglia, we observe limited tropism for SARS-CoV-2 in NPCs, mature neurons, glial progenitors, as well as in glia of astrocytic, microglial, and schwann/oligodendrocyte lineage. Despite a low percentage of infected cells at a modest virus load, we observe extensive cell death that also occurs in uninfected neurons. Single cell transcriptome profiling reveals distinct responses in microglia and astrocytes but also cellular states that overlap with disease-associated phenotypes observed in neurodegenerative diseases, including upregulation of genes implicated in microglia-mediated synaptic stripping in post-encephalitis models, and suggests compromised blood-brain barrier integrity. Across all cell types, we further observe a global translational inhibition as well as altered carbohydrate metabolism and cellular respiration.
Previous studies using undirected brain organoids to study SARS-CoV-2 have yielded variable results in terms of infectivity.16–20 To some degree, this discordance may be a result of differences regarding virus load, between-subject variability, as well as the cellular composition of the organoids. In this study, we observe that most cell types were sparsely infected at a modest virus load. In agreement with previous reports,19, 51 we over time observe an increase in viral gene copies in the supernatants of infected organoids. Moreover, when we perform plaque formation assays using the same supernatants, the number of plaques per unit volume rapidly decreased at later time points. This suggests that the cells either have an inefficient assembly and shedding of viral particles, or that the massive cell death observed as early as 24hpi decreases the availability of viable cells to sustain the production of fully assembled viral particles.
Existing protocols for generating brain organoids have several constraints that also limit our interpretations. Importantly, brain organoids most closely resemble the developing fetal brain rather than the mature adult brain. In this context, we cannot exclude that important difference exists in the cellular responses and tropism between immature and mature brain cells, and primarily our model should be considered to recapitulate the responses to SARS-CoV-2 of the developing fetal brain. Nonetheless, the observe cellular responses that to a large degree are consistent with molecular mechanisms that pertain to the observed clinical phenotypes in the aftermath of COVID-19 infection in adults. Further, it is unclear to what extent the used virus loads recapitulate in vivo exposure. To avoid supra-physiological exposure, we therefore use modest virus titers with estimated MOIs in between 0.1 and 0.3. Finally, our model does not include peripheral immune cells and thus cannot capture effects related to infiltrating immune cells. However, by excluding infiltrating monocyte-derived cells the model is optimized for assessing microglial responses then avoiding difficulties with distinguishing these resident cells from infiltrating cells that can obtain a largely overlapping signal in the CNS environment.
In summary, we here provide an experimental approach for modeling viral encephalitis that includes tightly orchestrated responses of microglia and astrocyte in the context of neuronal circuits. Challenging this model with live SARS-CoV-2 virus, we observe initial glial responses to virus exposure suggesting that the resident immune cells of the brain can contribute to CNS damage through pathological processes such as synaptic stripping and neuronal apoptosis. A key next step will be to determine the clinical importance of such molecular processes, and specifically whether they may contribute to the neurocognitive and neuropsychiatric symptoms observed in a subset of COVID-19 patients. If so, these microglia-containing organoid models may provide an opportunity to evaluate microglia-targeted therapeutics aimed at minimizing or preventing these COVID-19 sequelae.
MATERIALS AND METHODS
Ethics
All individuals signed a written informed consent before participating in the study, as approved by the Institutional Review Board of Partners HealthCare (Boston, MA, USA) and the Regional Ethical Review Boards in Stockholm, Sweden. All relevant ethical regulations were followed when performing the study.
Vero E6 cells
Vero E6 (ATCC-CRL-1586) cells were maintained in Dulbecco’s modified eagle medium (DMEM, Cytiva) supplemented with 5 % heat-inactivated fetal bovine serum (FBS, Cytiva), 100 U/mL penicillin and 100 μg/mL streptomycin (Cytiva).
iPSC reprogramming
Two healthy human iPSC lines (males) were obtained from the MGH Neurobank. Briefly, iPSC colonies were obtained using mRNA reprogramming in a feeder-free culture system as described previously.52 Stable iPSCs were expanded in NutriStem XF medium (Biological Industries) and on biolaminin 521 LN-coated (BioLamina) plates. iPSCs were then purified using MACS with anti-TRA-1–60 MicroBeads (Miltenyi Biotec) on LS columns. All fibroblasts and iPSCs were screened and found negative for Mycoplasma and stained positive for pluripotency markers like octamer-binding transcription factor 4 (POU domain, class 5, transcription factor 1) and TRA-1– 60.
Brain organoid cultures
Undirected brain organoids were prepared from single cell suspension of human iPSCs as previously described.23, 53 Briefly, embryoid bodies (EBs) were generated by seeding 9000 single cells in each well of a low attachment 96-well U-bottom plate with Y-27632 ROCK inhibitor (10 μM) for one day. EB media was replaced on day 5 with Neural induction media in the same 96-well plate. On day 10-11, EBs were embedded in 30μl Matrigel (Corning) using sheets of dimpled parafilm and incubated for 20min at 37°C as previously detailed. Matrigel embedded single EBs were then either transferred to a 24-well plate in 3ml of expansion media per well, on an orbital shaker (90 rpm), or to a 12-well miniaturized bioreactor (largely overlapping with the design proposed by Qian et al.),11 until further use.
Induced microglia-like cells
iMGs were derived from monocytes (donated from one healthy male) using established methods previously described in detail.24, 52 Briefly, whole blood was collected into vacutainer cell preparation tubes containing sodium citrate as an anticoagulant (Becton, Dickinson and Company) and processed as per the manufacturer’s instructions. Peripheral blood mononuclear cells (PBMC) were isolated, washed twice with PBS by centrifugation and suspended in heat-inactivated fetal bovine serum (FBS; Sigma) containing 10% DMSO (Sigma). The cell suspension was then divided into aliquots, transferred into cryovials and cryopreserved by using a liquid nitrogen cooled freezer. Generation of induced microglia-like cells (iMG) from PBMCs were carried out using methods previously described,24 with minor modifications. Briefly, cryopreserved PBMC samples were transferred from liquid nitrogen freezer to a 37 °C water bath. Once the cell suspension had been thawed, it was gently pipetted into 10 ml of pre warmed complete RPMI medium (CM) consisting of basal RPMI-1640 supplemented with 10% heat-inactivated FBS and 1% penicillin-streptomycin (P/S; Thermo Fisher Scientific). Following centrifugation (300 g for 5 min at room temperature), the supernatant was discarded, and the cell pellet was resuspended in the appropriate volume of CM.
Isolated PBMCs were cultured at a density of 5×105 cells/1 ml CM on 24-well plates coated with Geltrex (Thermo Fisher Scientific). After 24 h of incubation, the media was replaced with RPMI-1640 supplemented with 1x glutamax (Life Technologies), 1% P/S and 0.1 μg ml−1 of interleukin (IL)-34 (R&D Systems) and 0.01 μg ml−1of granulocyte macrophage colony-stimulating factor (GM-CSF; R&D Systems). After 7 days, fresh media was added. At day 11, cells were used for infection.
Virus isolate
SARS-CoV2 (GenBank: MT093571.1), originally from the Public Health Agency of Sweden, was propagated on 90% confluent Vero E6 cells for 3 days at 37 °C. Cell debris were removed by centrifugation at 300 RCF for 5 minutes and the viral supernatant was aliquoted into cryovials and stored at −80 °C. Viral titers were quantified by PFU assay.
Viral infection
All experimental studies involving infectious SARS-CoV2 were performed within the biosafety level 3 (BSL3) facility at Karolinska Institute. From single cell dissociation experiments, approximately 1×106 cells were found to be present in a D65 organoid. Based on this, 3×105 PFU/ml of SARS-CoV2 was used to infect organoids for 2h (MOI 0.3) in a spinning bioreactor. Following viral exposure, the organoids were washed twice with PBS and transferred into fresh medium (2ml) to follow the course of infection. To monitor viral replication, 100μl of medium was collected at specified time points, centrifuged at 300g for 5 min and the supernatant was collected in trizol for qPCR analysis. For single cell RNA sequencing (scRNA-seq), the viral load was reduced to 1 ×105 PFU/ml (MOI 0.1) in order to reduce the percentage of cell death observed with the previous MOI.
Plaque forming unit assay
PFU assays was performed on 24-well cell culture plates seeded with 2×105 Vero E6 cells per well. A serial dilution of the inoculum medium (consisting of either the viral stock or experimental samples in DMEM) was prepared and 200μL used to infect each well. The plates were then incubated for 1 hour at 37 °C, 5% CO2. After the incubation period, the wells were washed twice with PBS and 1 mL of overlay medium (2:3 mix of 3% high density carboxymethyl cellulose and complete DMEM medium) was added to each well and the plates incubated for 3 days at 37 °C, 5% CO2. The plates were subsequently inactivated with 1 mL of 10% formaldehyde solution overnight at RT, washed twice with PBS, stained with 200μL of crystal violet solution for 30 min at room temperature and the plaques counted.
Immunohistochemistry
Organoids were fixed with 4% paraformaldehyde for 20 min at RT followed by 24h in 30% sucrose in PBS. Organoids were then embedded in OCT (VWR) and frozen at –80°C. Cryosections (16 um) of brain organoids were incubated with blocking solution for 1h (10% of Normal Donkey serum in 0.3% Triton X) in 1xPBS. Blocked cryosections were incubated with the respective primary antibodies, diluted in blocking solution and incubated overnight at 4°C: anti-ACE2 (rabbit, Nordicbiosite ASJ1B6222, 1:100), anti-β-tubulin (mouse, Promega G712A, 1:500), Cleaved caspase-3 (rabbit, Cell Signaling, 1:300), anti-PAX6 (mouse, Developmental Studies Hybridoma Bank, 1:100), anti-SOX10 antibody (goat, R&D AF2864, 1:50), anti-OLIG2 (goat, R&D AF2418, 1:100), anti-GFAP (mouse, Sigma G3893, 1:100), anti-IBA1 (goat, Abcam AB5076, 1:100), anti-IBA1 (rabbit, Wako 019-19741, 1:100) and SARS-CoV-2 (2019-nCoV) Nucleoprotein / NP Antibody (rabbit, Nordicbiosite 158-40143, 1:200). Cryosections were washed three times with 1xPBS and incubated with the secondary antibody for 1h at RT. Secondary antibodies (all conjugated to Alexa Fluor 488, 555 and 647) were purchased from Life Technologies and used at a 1:500 dilution. Then, samples were washed three times with (1xPBS), incubated for 5 minutes with fluorescent nuclear DAPI stain (VWR; 1:5000), and mounted with DAKO (Life Technologies). Samples were imaged using a LSM800 confocal microscope.
Image quantification
Immunofluorescence images were analyzed using the CellProfiler software to measure and classify cells according to the expression of the selected markers. First, the software was trained to automatically segment the images into: (1) cells (DAPI positive objects); (2) NP or casp3 objects, followed by relating objects (1) and (2). For cell type specific quantifications, (2) was related to (3) objects segmented by cell type specific markers stained by antibodies indicated above. Finally, the related objects were expressed as a percentage of DAPI positive cells.
Quantitative PCR (qPCR)
RNA extraction was performed using DirectZol RNA-Miniprep Kit (Zymo Research Inc.), following the manufacturer’s protocol. Samples collected in TRI reagent were applied to Zymo-spin columns. DNase I treatment was performed for samples with genomic DNA. Bound RNA was washed, diluted in 30ul nuclease-free water and stored at −80°C. Quality and concentration of extracted RNA was determined using a NanoDrop (Thermofischer Scientific). Total RNA was reverse-transcribed to cDNA using the High-Capacity RNA-to-cDNA Synthesis kit (Thermofischer Scientific) following manufacturer’s protocol. Reverse transcription reactions were carried out in 20ul reaction volume containing 1ug of input RNA with the following thermal cycler conditions: 37°C (60 minutes) + 95°C (5 minutes) + 4°C (∞). cDNA was further diluted to 1:3 and used as templates for PCR reactions.
PCR reactions were performed using the StepOnePlus™ Real-time PCR system (Applied Biosystems, Thermofischer Scientific) with PowerTrack™ SYBR Green Master Mix. Expression of viral mRNA was assessed by evaluating threshold cycle (Ct) values. Relative expression levels of N gene in cell lysates were normalized against a housekeeping gene, GAPDH (glyceraldehyde-3-phosphate dehydrogenase), according to the Delta delta Ct method. Absolute quantification of viral gene expression levels in the supernatant were performed using the standard curve method. Primer sequences used in this study are:
GAPDH Forward – GGTGGTCTCCTCTGACTTCAACA
GAPDH Reverse – GTGGTCGTTGAGGGCAATG
N gene Forward – CATTGGCATGGAAGTCACAC
N gene Reverse – TCTGCGGTAAGGCTTGAGTT
RdRp Forward – CGCATACAGTCTTRCAGGCT
RdRp Reverse – GTGTGATGTTGAWATGACATGGTC
Dissociations for single cell RNA sequencing
Briefly, three organoids per condition (control, 24hpi, 72hpi) were washed twice with DPBS without ions and cut into small pieces using a sterile scalpel. Single cell suspension was generated using neural dissociation kit (Miltenyi Biotec) according to manufacturer’s protocol. Microglia was enriched using CD11b magnetic microbeads (Miltentyi Biotec) using MACS. Viability and cell numbers were assessed for both MACS-enriched and flow-through fractions before proceeding with the 10x protocol.
Single-cell RNA-sequencing
Single cell suspensions were loaded onto a Chromium controller chip v3.1 (10X Genomics), with a target output of 6000-7000 cells per channel. GEM generation, barcoding, cDNA amplification and library preparation was performed using the Single-cell 3’ Gel Bead and Library v3.1 kit (10X Genomics) at the Eukaryotic Single Cell Genomics facility (SciLifeLab, Sweden). Amplified cDNA and final libraries were evaluated on a Bioanalyzer for quality control and sequenced on Illumina NovaSeq 6000 platform.
Data processing
Sequenced data was processed through the CellRanger Software (5.0, 10X Genomics) and transcripts were aligned to a combined reference of Human GRCh38-3.0.0 and SARS-Cov-2 genome (GenBank: MT093571.1). Filtered feature-barcode matrices from CellRanger were further subjected to quality control measures where low-quality cells with < 200 uniquely expressed genes and high percentage of mitochondrial reads (>20%) were filtered out. Doublets were identified using scDblFinder package and removed with caution. Genes expressed in fewer than 5 cells were filtered out. Filtered count matrices were merged and analyzed downstream using the Seurat package. Expression data per experimental condition was normalized using the regularized negative binomial regression method implemented in SCTransform, while regressing out the difference between the G2M and S phase scores, which served as a confounding factor in our dataset. Anchors across conditions were identified using FindIntegrationAnchors function and samples were integrated using canonical correlation analysis (CCA) implemented in the IntegrateData function of Seurat. Linear data compression using principal component analysis (PCA) was performed on 3000 highly variable genes. Top 30 PCs were used as the input to perform non-linear dimensionality reduction using UMAP.
Clustering and cell-type identification
We constructed a k-nearest neighbor (KNN) graph based on Euclidean distance in 30 PCs and performed unsupervised graph-based clustering using the Louvain algorithm (modularity resolution = 0.8). Preliminary clustering of 39,808 cells identified 21 clusters yielding a clear separation of neuro-ectodermal clusters. Non-neuroectodermal clusters that did not express known marker genes related to cell types present in the brain, were removed. The remaining 26,148 cells were then subjected to second-level clustering in an integrated space (resolution= 0.6), yielding 16 final clusters. The clustering was visualized using UMAP embedding in two dimensions. Top differentially expressed genes conserved across conditions for the final clusters were identified using a non-parametric Wilcoxon rank-sum test in FindConservedMarkers function. P-values were adjusted based on Bonferroni correction and genes with at least 25% cluster-specific expression, >0.25 average log-fold change, and FDR<0.01 were chosen to identify the clusters. All cells were assigned scores based on cell type-specific gene modules to further identify broad cell types such as neurons, astrocytes, microglia and oligodendrocytes. A priori set of markers curated based on previous studies of cerebral organoids and developing fetal brain was explored to manually annotate the clusters. Additionally, we leveraged existing published fetal brain and organoid scRNA-seq datasets and undertook a correlation-based method (Spearman) to compare our annotated clusters to transcriptional profiles of previously annotated cell types.
Differential expression testing and functional interpretation
We performed differential gene expression analyses on each cluster across conditions on log-normalized values using the MAST package in R. DEGs were considered significant if their Benjamini-Hochberg-adjusted P-value was < 0.05 and a log2 fold change was > 0.25. Significant DEGs were used to perform GO term overrepresentation analysis using the enrichR package. Gene set enrichment analysis (GSEA) was performed on ranked DEGs using fgsea package for pathway analysis (KEGG, GO:BPdatabase). Gene sets were limited by minSize=3 and nPerm=10000. Normalized enrichment scores were calculated and plotted for pathways with adjusted p-values < 0.05. Statistical significance of association of DEGs with gene signatures from published datasets was performed using Fischer’s exact test to obtain a p-value (<0.05, adjusted with BH method) and Odds ratio (OR). Number of genes expressed in the respective cell type was used as the genomic background.
Intercellular interaction
Cellular crosstalk was inferred via ligand-receptor pair expression using CellphoneDB package (version 2) in python. Statistical analysis was performed within the package with default parameters (1000 iterations) and no subsampling was done.
Statistics
The assumptions of each used test were checked. All reported p-values are two sided and type of statistical test is reported in the figure legends or in the main text.
DATA AND CODE AVAILABILITY
Processing of data and downstream analysis was performed in R (version 4.0.3). Figures used in this manuscript were generated in Python (version 3.6.12). Raw single-cell RNA sequencing data is deposited into GEO database (XXXX). Source code to reproduce the findings are available in our repository on Github (https://github.com/SellgrenLab/organoid-Covid19). All other data is available from the corresponding authors upon request.
AUTHOR CONTRIBUTIONS
C.M.S, S, A.O.O, S.K.G, and S.M conceived the project. R.H.P and S.D.S provided iPSCs. A.O.O derived the organoids with help from S. N.R.S and S performed the infections and related assays supervised by A.R, with help from L.S. A.O.O optimized and performed IHCs, imaging and quantifications with S. S.K.G optimized the qPCR assays and performed them with S.M. F.O derived microglia for 2D assays. S.M designed the scRNA-seq experiments with inputs from S and C.M.S. Dissociations for scRNA-seq was performed by S and N.R.S. S.M analyzed scRNA-seq data. S and S.M interpreted the data. S, S.M and C.M.S wrote the manuscript with inputs from the other co-authors.
COMPETING INTEREST
The authors declare no competing interests.
ACKNOWLEDGEMENT
We thank the study participants. We are also grateful to the participant core facilities at Karolinska Institutet; The Biosafety Level (BSL)-3 laboratory at Biomedicum, BIC, and SciLife Lab. We are thankful to Judit Ozsvar, an intern at the Sellgren laboratory at time of the experiments, for her help with cell culture work, as well as to Gretchen Majkowitz for valuable feedback on the manuscript. This work was supported by grants from Hjärnfonden postdoktorala stipendier (S.: PS2019-0063), the Swedish Research Council (C.M.S.: 2017-02559), Karolinska Institutet (C.M.S.: KID), regional agreement on medical training and clinical research between Stockholm County Council (A.L.F, C.M.S.), One Mind Foundation/Kaiser Permanente (C.M.S.), and Marianne and Marcus Wallenberg Foundation (C.M.S.).