key: cord-0268224-q5h5w8oo
authors: Wu, Timothy Ting-Hsuan; Travaglini, Kyle J.; Rustagi, Arjun; Xu, Duo; Zhang, Yue; Jang, SoRi K.; Gillich, Astrid; Dehghannasiri, Roozbeh; Martinez-Colon, Giovanny; Beck, Aimee; Wilk, Aaron J.; Morri, Maurizio; Trope, Winston L.; Shrager, Joseph B.; Quake, Stephen R.; Kuo, Christin S.; Salzman, Julia; Kim, Peter S.; Blish, Catherine A.; Krasnow, Mark A.
title: Activated interstitial macrophages are a predominant target of viral takeover and focus of inflammation in COVID-19 initiation in human lung
date: 2022-05-10
journal: bioRxiv
DOI: 10.1101/2022.05.10.491266
sha: 57ff15cac58cda637856bd82a1284a796ecbdbc4
doc_id: 268224
cord_uid: q5h5w8oo

Early stages of deadly respiratory diseases such as COVID-19 have been challenging to elucidate due to lack of an experimental system that recapitulates the cellular and structural complexity of the human lung, while allowing precise control over disease initiation and systematic interrogation of molecular events at cellular resolution. Here we show healthy human lung slices cultured ex vivo can be productively infected with SARS-CoV-2, and the cellular tropism of the virus and its distinct and dynamic effects on host cell gene expression can be determined by single cell RNA sequencing and reconstruction of “infection pseudotime” for individual lung cell types. This revealed the prominent SARS-CoV-2 target is a population of activated interstitial macrophages, which as infection proceeds accumulate thousands of viral RNA molecules per cell, comprising up to 60% of the cellular transcriptome and including canonical and novel subgenomic RNAs. During viral takeover, there is cell-autonomous induction of a specific host interferon program and seven chemokines (CCL2, 7, 8, 13, CXCL10) and cytokines (IL6, IL10), distinct from the response of alveolar macrophages in which neither viral takeover nor induction of a substantial inflammatory response occurs. Using a recombinant SARS-CoV-2 Spike-pseudotyped lentivirus, we show that entry into purified human lung macrophages depends on Spike but is not blocked by cytochalasin D or by an ACE2-competing monoclonal antibody, indicating a phagocytosis- and ACE2-independent route of entry. These results provide a molecular characterization of the initiation of COVID-19 in human lung tissue, identify activated interstitial macrophages as a prominent site of viral takeover and focus of inflammation, and suggest targeting of these macrophages and their signals as a new therapeutic modality for COVID-19 pneumonia and progression to ARDS. Our approach can be generalized to define the initiation program and evaluate therapeutics for any human lung infection at cellular resolution.

Lower respiratory infections are one of the leading causes of death worldwide 1,2 , accelerated by the current Coronavirus Disease 2019 (COVID-19) pandemic 3 . Most such infections, including COVID-19, start innocuously in the upper respiratory tract and become dangerous when they reach alveoli [4] [5] [6] [7] [8] [9] , the site of gas exchange, but the critical transition to life threatening pneumonia and acute respiratory distress syndrome (ARDS) has been difficult to elucidate. For practical and ethical reasons, such early and key steps in human pathogenesis have been inferred, with rare exception 10,11 , from examination of late or end stage patient lung lavage, biopsy, or autopsy specimens, using classical histopathological methods 12-14 and recently single cell multi-omic profiling [15] [16] [17] [18] [19] [20] .

These approaches provide a picture of COVID-19 pneumonia at unprecedented cellular and molecular resolution, and have suggested models of pathogenesis involving not only infection of the alveolar epithelium but also implicating alveolar capillaries, macrophages and other myeloid cells 18, 19, [21] [22] [23] and production of various inflammatory cytokines and chemokines 15, 19 . It remains unclear which cells are the direct virus targets in the human lung, and the nature of their virus-induced host response -in particular, the origin and sequence of molecular signals that initiate, sustain, and propagate the inflammatory cascade that leads to COVID-19 ARDS 4 .

These early pathogenic events hold the key to understanding and preventing the transitions to the deadly and systemic forms of COVID-19, but we know little about them. This is due to difficulty accessing human lung tissue at this critical transition, and the sheer number of lung (>58) and alveolar (14) cell types potentially involved. This cellular complexity has made pathogenic mechanisms challenging to empirically address even in the most sophisticated human lung organoid systems [24] [25] [26] [27] and animal models [5] [6] [7] 20, [28] [29] [30] [31] .

Here we describe an experimental model of SARS-CoV-2 infection that allows systematic interrogation of the early molecular events and pathogenic mechanism of COVID-19 at cellular resolution in native human lung tissue. We determine the cellular tropism of SARS-CoV-2 and its distinct and dynamic effects on host cell gene expression for individual lung cell types. The most prominent target is a population of interstitial macrophages in which the virus takes over the transcriptome and induces a specific host interferon program along with seven chemokines and cytokines that can signal to a diverse array of lung immune and structural cell types that express the cognate receptors. We propose that this early focus of lung inflammation is a key step in the transition to the deadly and systemic forms of COVID-19 and a potential new therapeutic target.

To define the early events of SARS-CoV-2 infection in human lung, we cut thick sections (~300-500 µm "slices") of fresh lung tissue procured from therapeutic surgical resections or organ donors, and placed the slices in culture medium containing DMEM/F12 and 10% FBS (Fig. 1a) . We then infected them with SARS-CoV-2 (USA-WA1/2020) at a multiplicity of infection (MOI) of 1 for two hours, and the cultures continued for 24 or 72 hours to allow infection to proceed. Plaque assay of culture supernatants demonstrated production of infectious virions that increased between 24 and 72 hours of culturing (Fig. 1b, Extended Data Fig.1 ).

Productive infection was abrogated by pre-inactivation of the viral stocks with heat or ultraviolet light, or by addition to the slice culture medium of 10 µM remdesivir, an RNA-dependent RNA polymerase inhibitor used as a COVID-19 therapeutic (Fig. 1b) .

To characterize viral and host gene expression during SARS-CoV-2 infection, slices were dissociated and analyzed by single-cell RNA sequencing (scRNA-seq, 10x Genomics), adapting the methods we and others previously used to construct a comprehensive transcriptomic atlas of the healthy human lung 32-38 to capture, quantify, and map SARS-CoV-2 viral gene expression along with host gene expression in each profiled lung cell 39, 40 . The number of viral RNA molecules detected per infected cell spanned a wide range (Fig.1c, d) , with the vast majority (~99%) of profiled cells from infected lung slices containing few or no detected viral RNA molecules (Fig. 1d) . But the rest of the cells (~1%) expressed tens to hundreds of viral RNA molecules per cell at 24 hours, and by 72 hours the distribution had shifted to even higher values with rare cells (~0.01%) accumulating thousands of viral RNAs per cell (Fig. 1d) , paralleling the increase in virus production during this period (Extended Data Fig.1 ). As with infectious virions, viral RNA levels determined by scRNA-seq were diminished by inactivation of the virus stocks with heat or ultraviolet irradiation, or by addition to the culture medium of remdesivir (Fig. 1c) .

We also investigated the junctional structure and processing of the viral RNA molecules by analyzing our scRNA-seq dataset using the SICILIAN framework 41 , which identifies RNA sequencing reads that map discontinuously in a genome, such as reads that span splice junctions of eucaryotic mRNAs or the subgenomic junctions of the nested SARS-CoV-2 mRNAs. We detected canonical subgenomic junctions among the rare sequence reads outside their 3' ends, confirming generation of canonical SARS-CoV-2 mRNAs in the lung slice cultures (Fig. 1c, right panel). In addition, we identified dozens of novel subgenomic junctions, indicating widespread generation of diverse non-canonical subgenomic viral RNAs along with canonical subgenomic forms during lung infection (Fig. 1c, Extended Data Fig. 2, Table S1 ). These noncanonical junctions included three that spliced the standard viral 5' leader sequence to a novel downstream site, as well 494 junctions between two novel internal sites in the genome and 479 junctions between an internal and 3' site (the most abundant non-canonical species detected, consistent with the strong 3' end bias of 10x technology). Some of these non-canonical RNAs are predicted to encode novel viral proteins, or alter potential regulatory sequences in the 3' noncoding region of the viral mRNA. Heat inactivation or ultraviolet irradiation of the viral stocks, or remdesivir addition, abrogated formation of both canonical and non-canonical junctions (Fig.   1c , right panel). Together, these data demonstrate that lung cultures support ongoing, productive viral replication.

The cellular tropism of a virus -the set of host cells that allow viral entry and replication -is among the most characteristic and significant determinants of virulence. Historically tropism has been inferred from autopsy specimens, often weeks, months, or even years after disease onset. More recently, tropism has been predicted from expression patterns of entry receptors identified by biochemical or functional screening in heterologous cell types 42 . For SARS-CoV-2, a small subset of lung epithelial types (AT2, ciliated, AT1, club, and goblet cells) were predicted to be the major direct targets for SARS-CoV-2 based on their expression of the canonical SARS-CoV-2 receptor ACE2 and protease TMPRSS2 18, 24, 32, 43 . However, studies of COVID-19 autopsy lungs have detected viral gene products in various epithelial and endothelial cells, fibroblasts, and myeloid cells, indicating widespread viral presence at least in end-stage disease 18, 19 .

To determine SARS-CoV-2 lung cell tropism empirically and directly compare infection of lung cell types in their natural context, we first used the most sensitive and specific markers from our molecular atlas of the healthy human lung 32 to identify the cell types present in the cultured lung slices from their transcriptomic profiles, and then assessed their viral RNA levels in the infected cultures. Of the 176,382 cells with high quality transcriptomes obtained from infected lung slices of four donor lungs, along with those of the 112,359 cells from mockinfected slices (cultured without viral addition) and 95,389 uncultured control cells (from freshly-cut lung slices), we identified 55 distinct molecular lung cell types distributed across the major tissue compartments (Fig. 2a, Extended Data Fig. 3 , Table S2 ). These included most (46 out of 58, 80%) of the cell types described in the healthy human lung 32 plus 5 additional types of lymphocytes (e.g., CD4+ cytotoxic T lymphocytes, γδ T cells, regulatory T cells, tissue-resident memory CD8+ T cells, GZMK+CD8+ T cell; Fig. 2a , blue) along with culture-induced proliferative states of signaling alveolar type 2 (AT2-s) cells, NK cells, and dendritic cells (DCs) and several culture-induced proliferative and activation states of fibroblasts, which could not be ascribed to any previously defined fibroblast types (Fig. 2a, grey) . The only cell types not recovered after culturing were rare myeloid types (e.g., IGSF21+ DCs, TREM2+ DCs, classical monocytes), which may egress from the slices or not survive during culture (Extended Data Table S2 ).

Cellular SARS-CoV-2 viral RNA levels across the 55 human lung cell types in the infected cultures are shown in Fig. 2a . Although 10-20 viral RNA molecules were detected in about one-third of the molecular cell types in the infected cultures, cells with high viral levels (hundreds to thousands of SARS-CoV-2 UMI per cell) were rare and restricted to six cell types.

One was AT2 cells, a predicted target of SARS-CoV-2. The others had not been previously shown to directly support viral replication or production, and included myofibroblasts, lipofibroblasts, two molecular types of T cells and NK cells, and macrophages. Macrophages were the most prominent lung targets, accounting for 75% of cells with 50 or more viral UMI per cell. However, even for macrophages, such cells represented only a small proportion of the recovered cell type (0.5% of all macrophages), indicating inefficient access or entry or a sensitive subpopulation (see below). One caveat to this tropism analysis is that identities could not be assigned to 16% of cells with 50 or more viral UMI per cell because they did not robustly express cell type markers ("unidentified" cell types, Fig. 2a) , presumably due to downregulation of the host transcriptome during viral takeover. Most cells with high viral load were detected in cultures at 72 but not 24 hours after infection, indicating that the intervening 48 hours is the critical period of viral RNA amplification in most lung cell types.

To validate these lung cell tropism results, visualize the infected cells, and localize foci of viral replication, we performed multiplexed single molecule fluorescence in situ hybridization (smFISH) of the infected lung slices to simultaneously detect positive strand viral RNA (S gene probe), negative strand viral RNA (replication intermediate, Orf1ab gene probe), the canonical viral receptor ACE2, and markers of the infected cell types detected in scRNA-seq (Fig. 2b,c) .

We found both positive and negative strand viral RNA in AT2 cells (SFTPC + EPCAM + ), myofibroblasts (ASPN + COL1A2 + ), macrophages (PTPRC + MARCO + ) and exceedingly rarely, CD4 T cells (PTPRC + CD3 + CD4 + ). We also detected cells filled with viral mRNA molecules but no negative strand RNA, the early replication intermediate, or any of the cell-type markers in our panel; these are likely cells at an advanced stage of viral takeover, nearing cell lysis. Infected cells were generally scattered throughout the infected lung tissue, but rare clusters were detected such as an infected macrophage associated with two CD4 T cells (Fig. 2d) .

For AT2 cells, myofibroblasts, and T cells, the cells with high viral load were rare in the tissue sections, as in the scRNA-seq tropism analysis. In contrast, infected macrophages were more abundant and showed a broad and seemingly continuous range of viral RNA molecules. Some macrophages (PTPRC + MARCO + ) showed a few (1-3 puncta) positive-strand viral RNA molecules but no negative strand viral RNA, whereas others expressed a few (1-3 puncta) negative strand viral RNA molecules alongside a wide range (1 to dozens of puncta) of positive strand viral RNA molecules (Fig. 2c) .

We reasoned that the macrophages in the lung slice cultures with SARS-CoV-2 RNA levels spanning several orders of magnitude -from tens to thousands of UMIs in the scRNA-seq analysis ( Fig. 2a) and from one to dozens of puncta detected by smFISH (Fig. 2c ) -were infected cells harboring active intermediates that had progressed to different stages of infection, those with highest RNA loads having progressed furthest in the infection cycle. This is consistent with our finding that cells harvested 72 hours after infection generally had higher viral RNA levels than those harvested at 24 hours (Fig. 1d) .

To resolve the apparent heterogeneity in the infected macrophages, we further clustered the gene expression profiles of macrophages in the lung slices and found that they separated into three distinct clusters (Fig. 3a) . One had higher expression of genes involved in functions ascribed to mature alveolar macrophages, including antigen-presentation major histocompatibility complex class II (MHCII) genes (HLA-DPA1, HLA-DPB1, HLA-DQA1, HLA-DQB1, and CD74) and genes involved in lipid homeostasis (LPL, APOC1, FABP3, FABP4, and HPGD; Fig. 3c ,d) 44 . smFISH showed cells expressing these markers were larger and rounder in morphology than the others, and localized to the alveolar airspace (Fig. 3b ,e,f), hence we refer to them as "alveolar macrophages" (AMs). Another cluster we call "interstitial macrophages" (IMs) expressed lower levels of the classical AM markers including LPL, APOC1, FABP3, FABP4, and HPGD but were enriched for a different set of genes including the monocyte marker CD14 (Fig. 3c,d) , and localized interstitially (Fig. 3b ,e,f, Extended Data Fig. 4 ). The third cluster was transcriptionally similar to interstitial macrophages but also expressed genes known to be activated by NF-KB signaling (NFKBIA, NFKBIZ, and IL1B), inflammation (IER3, EREG, TIMP1, STAB1), and hypoxia-induced factor HIF1A; we call them "activated interstitial macrophages" (Fig. 3c,d, a-IMs) . Although a-IMs were detected in the uncultured control lung slices, they were a minor population. However, upon culturing, almost all IMs became activated;

indeed, this was among the most robust transcriptional changes we observed in the control (Fig. 3g ). Viral takeover reached up to 60% of an a-IM transcriptome (ratio of viral to total UMIs in a cell), whereas it never exceeded 2% of an AM transcriptome (Fig. 3g) .

Thus, in a-IMs, SARS-CoV-2 can infect and amplify its RNA until it dominates the host transcriptome, whereas viral RNA takeover does not occur in AMs.

To characterize the host cell response during viral takeover, we computationally ordered the infected macrophages according to the principal components that best correlated with viral RNA levels and takeover to reconstruct what we refer to as "infection pseudotime" ( (Table S3) .

Alveolar macrophages showed a distinct and more limited response to the virus (Table   S3 , Fig. 4f ). During infection pseudotime, only a handful of genes were specifically induced, including APOC1, FDX1, IFI27, HLA-DRB1, serine proteases SERPINA1 and SERPING1, and To predict the cellular targets of the inflammatory signals induced by SARS-CoV-2 infection of a-IMs, we used the single cell gene expression profiles of the infected lung slices to produce a map of cells expressing the cognate receptors (Fig. 4g ). Viral induction of CXCL10 in a-IMs predicts communication to and recruitment of broad classes of CD4 and CD8 T cells via the cognate receptor CXCR3, consistent with our observation by smFISH that T cells interacted directly with infected a-IMs. (Fig. 2d ). Viral induction of CCL2 predicts recruitment of specific DC subtypes (maDCs, mDC2) through CCR2, and induction of CCL8 could recruit neutrophils and create a self-amplifying circuit with macrophages via CCR1 (Fig. 4h) . The specific viral induction of IL6 and IL10 along infection pseudotime that can signal through its cognate receptors to most other structural (IL6) and immune (IL6, IL10) cell types of the lung indicates that infected a-IMs could broadcast inflammatory signals widely in the lung (Fig. 4h ).

We conclude that infection and takeover of a-IMs induces a robust cell intrinsic response to the virus and creates an immune signaling center and focus of inflammation in early SARS-CoV-2 infection, whereas infected AMs restrict viral RNA amplification and generally suppress their communication to other immune cell types.

Activated interstitial macrophages did not express detectable levels of ACE2, the canonical SARS-CoV-2 receptor, by scRNA-seq (Extended Data Fig. 6 ) or by smFISH (Fig. 2c) .

To explore the mechanism of SARS-CoV-2 entry into human lung macrophages, we modified a recombinant Spike-pseudotyped lentivirus system 58 to carry a nanoluciferase (NLuc) bioluminescence and tdTomato fluorescence dual readout (lenti-S-NLuc-tdT, Fig. 5a ). Lenti-S-NLuc-tdT enables sensitive detection of Spike-mediated viral entry over eight orders of magnitude by NLuc assay, and its fluorescent reporter allows viral detection in individual live cells by flow cytometry with minimal interference from cellular autofluorescence.

To test if SARS-CoV-2 Spike can mediate entry into human lung macrophages, we purified EPCAM -CD31 -CD206 + macrophages from freshly resected human lung tissue. The primary lung macrophages survived in culture for up to a week with minimal decrease in viability. When the purified lung macrophages were exposed to lenti-S-NLuc-tdT at 1 TCID25 (titer at which 25% of HeLa-ACE2/TMPRSS2 were infected) and then cultured for 48 hours, infection was demonstrated by robust expression of luminescence (1-1.5×10 5 RLUs), whereas the control lentivirus lacking SARS-CoV-2 spike protein did not elicit measurable luminescence ( Fig. 5b) .

Treatment of the purified lung macrophages with hydroxychloroquine, a lysosomal protease inhibitor, or cytochalasin D, a potent inhibitor of phagocytosis previously shown to block macrophage entry of bacteria and virus 59,60 , did not block infection by lenti-S-NLuc-tdT across a wide range of concentrations (10 -2 -10 2 uM) (Fig. 5c ). This indicates that Spikemediated entry into lung macrophages does not require phagocytosis. We next performed neutralization assays using three potent anti-Spike monoclonal antibodies (mAbs) at various stages of preclinical and clinical testing, including an ACE2-competing anti-RBD antibody (COVA2-15). Although each of these mAbs robustly inhibited lenti-S-NLuc-tdT infection of HeLa-ACE2/TMPRSS2 at nanomolar concentrations ( Fig. 5d) , none reduced lenti-S-NLuc-tdT infection of purified lung macrophages (Fig. 5e) . Thus, Spike-mediated entry into lung macrophages occurs by a potentially novel mechanism that does not require phagocytosis or the ACE2-interacting receptor binding motif (RBM) of the SARS-CoV-2 Spike protein. Our scRNA-seq and smFISH analyses indicate that SARS-CoV-2 can also infect alveolar macrophages, as previously surmised 21, 23 . However, in contrast to activated interstitial macrophages, there is neither viral takeover of the host cell transcriptome nor cell-autonomous induction of a substantial inflammatory response. This could be due to effective host control or destruction of the virus in alveolar macrophages, or to viral evasion of detection 61 

Fresh, normal lung tissue was procured from organ donors that have exhausted therapeutic Table S4 .

In each case, proximal (airway) and distal (alveolar) regions were resected and cut into 300-500 µm slices with platinum coated double edge blade (Electron Microscopy Sciences 7200301) manually. Both airway and alveolar slices (3 or 4 total) were cultured in the same well in a 12well plate with or without precoating of 500 µL of growth factor reduced Matrigel (Corning 

VeroE6 cells were obtained from ATCC as mycoplasma-free stocks and maintained in L-glutamate, 20mM HEPES, 2X antibiotic/antimycotic (Life Technologies), and 0.24% sodium bicarbonate (Sigma S8761)) and 2.4% Avicel (FMC Biopolymer). Plates were then returned to the incubator for 72h (VeroE6) or 48h (VeroE6/TMPRSS2) prior to overlay removal, washing with PBS, fixation with 70% ethanol, and staining with 0.3% crystal violet (Sigma). For the timecourse (Extended Data Fig.1 ), lung slices were infected and washed. At 24h the supernatant was harvested, stored frozen, and replaced completely with fresh media. At 72h, the supernatant was harvested and stored frozen. The supernatants were then thawed, and plaque assays performed on the same plate as above. 

All fresh (non-cultured and non-infected) tissue was processed at BSL2, and all cultured or infected tissue was processed in BSL3.

BSL2: Normal lung tissue was obtained as described for the slice cultures. All tissues were received and immediately placed in cold PBS and transported on ice directly to the research lab. benzonase (EMD Millipore 706643)) and incubated with manual or automatic rocking at 37℃ for 1 hour, followed by serum neutralization of Liberase and elastase activity with 10% FBS in cold DMEM/F12 media. For infection 1 only, the tissue was then dissociated by running m_lung_02 on a gentleMACS dissociator inside the BSC. The tissue was then mashed through a 100 µM filter with a syringe insert (Falcon), and the filter was washed with additional cold DMEM/F12 with 10% FBS to recover any remaining cells. The cellular suspension was spun at 4℃ at 300 x g for 5 minutes, washed, and exposed to 1mL cold ACK lysis buffer (Sigma) for 1 minute on ice. The lysis buffer was neutralized by dilution with 5 mL cold DMEM/F12 with 10% FBS, after which the cells were pelleted and resuspended in DMEM/F12 with 10% FBS, and the cells were stained with Trypan blue (Sigma T8154), sealed out of the BSC, and counted manually. For all steps, cells were kept at 0-4℃ using a cold block (Eppendorf Isotherm system).

10x mRNA capture, library construction, and sequencing BSL2: Cells isolated from normal lung tissue were captured in droplet emulsions using a 10x

Chromium Controller (10x Genomics). cDNA was recovered and libraries were prepared using 10x Genomics 3' or 10x Genomics 5' Single Cell V3.1 protocol (infections 1,2,4 and 5 were sequenced using 3' chemistry, while infection 3 used both 3' and 5' technology), as described 32 .

Sequencing libraries were analyzed (Agilent TapeStation D4150, using regular and high sensitivity D5000 ScreenTape) and balanced, and sequenced on a NovaSeq 6000 (Illumina).

BSL3: The 10x Genomics Single Cell protocols were performed as before, with the following modifications for BSL3. The 10x Genomics, 3' or 5' Single Cell v3.1 master mix was prepared outside the BSC. Within the BSC, cells prepared as above were added to the master mix in PCR tubes (USA Scientific 1402-4708) in a 96-well cold block (Ergo 4ti-0396) and the 10x chip was loaded per manufacturer's instructions, sealed, and processed in a 10x Chromium Controller in the BSC. The resultant cell/bead emulsions were loaded into PCR tubes and transferred immediately to a pre-warmed (53℃) PCR machine for cDNA synthesis carried out at 53℃ for 45 minutes, then 85℃ for 5 minutes, then 60℃ for 15 minutes (plaque assays showed that exposure of SARS-CoV-2-infected samples at 60℃ for 20 minutes in this manner rendered the sample non-infectious). After cDNA synthesis, samples were transferred out of the BSL3 for cDNA recovery, amplification, and sequencing library preparation as above.

Sequencing reads from single cells isolated using 10x Chromium were demultiplexed and then aligned to a custom built Human GRCh38 (GENCODE v30) and SARS-CoV-2 WA1 (GenBank:

MN985325.1) reference using Cell Ranger (version 5.0, 10x Genomics).

Expression profiles of cells from different subjects were clustered separately using Python software package Scanpy (v1.7.2). For host genes, normalization was performed as described 32 ;Unique molecular identifiers (UMIs) were normalized across cells, scaled per 10,000 using the "sc.pp.normalize_total" function, converted to log-scale using the "sc.pp.log1p" function, and highly variable genes were selected with the "sc.pp.highly_variable_genes" function with a dispersion cutoff of 0.5, and scaled to unit variance and zero mean (z-scores) with the "scanpy.pp.scale" function, clipping values exceeding standard deviation 10. Principal components were calculated for these selected genes with the "sc.tl.pca" function. Clusters of similar cells were detected using the Leiden method ("tl.leiden" function) for community detection including only biologically meaningful principle components, as described 32 , to construct the shared nearest neighbor map ("sc.pp.neighbors") and an empirically set resolution, visualized by uniform manifold approximation and projection (UMAP; "tl.umap" function).

Cells were iteratively clustered as described 32 , with the following modifications. After separating clusters by expression of tissue compartment markers, cultured cell types generally segregated from their non-cultured counterparts. When possible, we assigned cell types to the canonical cell types using the most sensitive and specific markers identified in the human lung cell atlas 1 . For culture-induced subtypes that showed substantial transcriptional change, a representative marker gene was prepended to their canonical identity (e.g., IRF1+ aCap). If the transcriptional change caused the cell type to lose markers that define their canonical identity, we named them based on the general type that could be assigned, and prepended a representative marker gene (e.g., KLF6+ Endo). If most of the cluster-specific markers were ribosomal or mitochondrial genes, they were labeled as low quality (e.g., Stromal-LQ). If most of the expressed genes were viral and we could not distinguish which cell type the cluster belonged to due to downregulation of marker genes, they were designated "infected" (e.g., Infected-LQ). Cells from different subjects with the same annotation were merged into a single group for all downstream analyses. Cell types that were exclusively found to be culture induced were grouped as "culture induced" (e.g., Induced Fibroblast) for viral tropism analysis. Some native subtypes characterized by subtle transcriptional differences could not be resolved by droplet-based 10x sequencing (e.g., proximal subtypes for basal or ciliary cells, molecular subtypes of bronchial vessel cells, mast/basophils), and several rare (neuroendocrine cells, ionocytes) or anatomically-restricted cell types (e.g. serous cells in submucosal glands) were absent from the profiled lung tissue.

For the UMIs that aligned to the SARS-CoV-2 viral genome, raw UMIs were either directly converted to log scale ("log10(Viral UMIs + 1)") or explicitly divided by total cellular UMIs but not log-converted ("Viral UMIs"). Viral takeover trends were visualized by non-parametric Local Regression (LOESS, R stats version 3.6.2).

For viral pseudotime analysis, computations were performed in R using the Seurat package (v3).

Infected alveolar macrophages (AMs) and activated interstitial macrophages (a-IMs) from infection 1 were grouped, and counts were normalized using the 'SCTransform' command.

Principal component analysis was performed using the 'RunPCA' command with default parameters and visualized with 'DimHeatmap'. To identify the major axes of variation within the infected macrophage subtypes that best correlated with SARS-CoV-2 RNA levels, the principal components with significant contribution from SARS-CoV-2 counts (among the top 15 genes with highest loadings) were selected for further inspection. PC.1 was found to be associated with AMs and aIMs were assigned respective pseudotime values that were normalized between 0 and 1.

To detect viral subgenomic RNA junctions, we ran SICILIAN 41 , a statistical wrapper that takes alignment files from a spliced aligner and calls true positive RNA splice junctions by employing a statistical model. SICILIAN assigns an empirical p-value to each detected junction in a 10x dataset, quantifying the statistical confidence of each detected junction being a truly expressed RNA junction. We used STAR v.2.7.5a as the aligner and aligned fastq files from all infections to our custom built Human GRCh38 (GENCODE v29) and SARS-CoV-2 WA1 (GenBank: MN985325.1) reference. STAR was run in two-pass mode, in which the first pass identifies novel splice junctions and the second pass aligns reads after rebuilding the genome index with the novel junctions and its parameters were tuned to avoid bias against non-GTAG junctions as previously shown 67 .

BSL2: Samples were fixed in either 10% neutral buffered formalin, dehydrated with ethanol and embedded in paraffin wax, as described 32 .

BSL3: Slices not taken for digestion were washed with PBS and transferred to 15 mL conical tubes containing 10% neutral buffered formalin (Sigma) and held at 4°C for 72 hours prior to transfer out of the BSL3. Slices were then transferred to 15 mL conical tubes containing PBS prior to dehydration. Sections (6 μm) from paraffin blocks were processed using standard pre-treatment conditions for each per the RNAscope multiplex fluorescent reagent kit version 2 (V2) Assay (Advanced Cell Diagnostics, (ACD)), or immunostaining RNAscope co-detection assay in which antibody labeling was carried out after RNAscope V2 assay, or RNAscope HiPlex Assay protocols. 

Lung tissue was obtained as described above for the slice cultures. All tissues were received and 

To create lenti-S-NLuc-tdT, Spike pseudotyped lentiviruses encoding a nanoluciferase-tdTomato reporter were produced in HEK-293T cells (5 × 10 6 cells per 10-cm culture dish) by cotransfection of a 5-plasmid system as described previously 58 . Based on the original lentiviral backbone plasmid (pHAGE-Luc2-IRES-ZsGreen, Addgene 164432), we replaced the Luc2-IRES-ZsGreen reporter with a cassette encoding H2B fused to Nanoluciferase (Promega) to minimize background luminescence, followed by a T2A self-cleaving peptide, and tdTomato fluorescent protein using gBlock synthesis (Integrated DNA Technologies). The 5-plasmid system includes a packaging vector (pHAGE-H2B-NanoLuc-T2A-tdTomato), a plasmid encoding full-length Spike with a 21-residue deletion on the C-terminus (pHDM SARS-CoV-2-Spike∆21), and three helper plasmids (pHDM-Hgpm2, pHDM-Tat1b, and pRC-CMV_Rev1b). Macrophages were seeded in white-walled, clear-bottom 96-well plates (10,000-20,000 cells per well) 1-day before the assay (day 0). On day 1, mAbs were serially diluted in DMEM/F12 medium and then mixed with lentivirus (diluted in DMEM/F12 medium, supplemented with polybrene, 1:1000, v/v) for 1 hr before being transferred to macrophages. Culture medium was replenished 4 hr post-infection. On day 3, cells were rinsed with DPBS, and then luminescent signals were read out with the nanoluciferase substrate (Nano-Glo) as above.

All heat maps and plots with single cell expression data include every cell from indicated types, unless otherwise stated in the figure legend (numbers available in Supplementary Table 4 ). Dot plots were generated using a modified version of Scanpy's 'pl.dotplot' (Fig. 2a, Fig. 6 ). Scatter plots for infection pseudotime were generated with ggplot2's 'geom_point' function, and trend lines were plotted with parameter 'method = "loess"' (Figs. 3g, 4da-f, Extended Data Fig. 5e ) Violin plots were generated with Scanpy's 'pl.violin' function (Fig. 1c, left panel) , or Seaborn's 'violinplot' and 'stripplot' functions (Fig. 2a) and show proportion of single cells at indicated expression levels.

Bar plots were generated in Excel (Fig. 1c, right panel, 3f) . Histogram plots were generated using Seaborn's 'histplot' function with log scale transformation on both x-axis and y-axis (Fig.   1d , lower panel). Cumulative distribution plot was generated using Seaborn's 'ecdfplot' function and plotted on a Matplotlib''logit' scale which implements the logistic distribution (in Fig. 1d, upper panel). Arcplots depicting number of subgenomic junctions was plotted using a custom Python function (available on Github). Differentially expressed genes along infection pseudotime were computed by taking the top 250 genes that contributed to each pseudotime trajectory (see Methods), and further tested using pseudotimeDE's 'runPseudotimeDE' function without subsampling testing against the asymptotic null distribution, with exact p-values indicated in Table S3 . Differentially expressed genes between "Late" vs. "Early" macrophages along infection pseudotime were computed using Seurat's 'FindMarkers' function implemented using the default Wilcoxon Rank Sum test, with exact p-values indicated in Table S3 .

Immunostaining and smFISH experiments were performed on at least 2 human or mouse subjects distinct from the donors used for sequencing, and quantifications were based on at least 10 fields of view in each. For smFISH, fields of view were scored manually, calling a cell positive for each gene probed if its nucleus had at least three associated expression puncta. No statistical methods were used to predetermine sample size. The experiments were not randomized and investigators were not blinded to allocation during experiments and outcome assessment.

Raw sequencing data, UMI tables, cellular annotation metadata, Seurat objects, and scanpy objects are being deposited and will be released as soon as possible (at latest, upon acceptance of this manuscript).

Code to reproduce the analyses and figures are being deposited and will be released as soon as possible (at latest, upon acceptance of this manuscript Nluc luciferase values are presented as mean ± s.d from two independent experiments; values were normalized to control (non-neutralized) viral infections in each plate. Tables   Table S1 . Summary of viral subgenomic junction discovery from single lung cells using SICILIAN Table S2 . Human lung cell cluster identities and their abundances in each dataset Table S3 . Differentially expressed genes along infection pseudotime trajectory for alveolar and interstitial macrophages Table S4 . Clinical summaries of donors or patients of surgical resection Identities and abundances of each cell type called in each human lung slice infection dataset (10x scRNA-seq) are shown, tabulated by condition (Truncated). Genes that are differentially expressed continuously along pseudotimeAM trajectory, and differentially expressed in late vs. early activated interstitial macrophages (AMs). 

Leading causes of death -Global summary estimates

A novel coronavirus outbreak of global health concern

SARS-CoV-2 pathogenesis

The pathogenicity of SARS-CoV-2 in hACE2 transgenic mice

Pathogenesis of SARS-CoV-2 in Transgenic Mice Expressing Human Angiotensin-Converting Enzyme 2

SARS-CoV-2 infection of human ACE2-transgenic mice causes severe lung inflammation and impaired function

Comparative pathogenesis of COVID-19, MERS, and SARS in a nonhuman primate model

Respiratory disease in rhesus macaques inoculated with SARS-CoV-2

Endothelial cell infection and endotheliitis in COVID-19

Histopathology and ultrastructural findings of fatal COVID-19 infections in Washington State: a case series

Pathological findings of COVID-19 associated with acute respiratory distress syndrome

Single-cell landscape of bronchoalveolar immune cells in patients with COVID-19

A single-cell transcriptomic landscape of the lungs of patients with COVID-19

The spatial landscape of lung pathology during COVID-19 progression

COVID-19 tissue atlases reveal SARS-CoV-2 pathology and cellular targets

A molecular single-cell lung atlas of lethal COVID-19

Immune memory from SARS-CoV-2 infection in hamsters provides variant-independent protection but still allows virus transmission

Circuits between infected macrophages and T cells in SARS-CoV-2 pneumonia

FcgammaR-mediated SARS-CoV-2 infection of monocytes activates inflammation

Inflammasome activation in infected macrophages drives COVID-19 pathology

Progenitor identification and SARS-CoV-2 infection in human distal lung organoids

A Single-Cell Atlas of the Human Healthy Airways

A cellular census of human lungs identifies novel cell states in health and in asthma

An integrated cell atlas of the human lung in health and disease. bioRxiv

A single-cell atlas of the peripheral immune response in patients with severe COVID-19

Multi-omic profiling reveals widespread dysregulation of innate immunity and hematopoiesis in COVID-19

Specific splice junction detection in single cells with SICILIAN

SARS-CoV-2 Cell Entry Depends on ACE2 and TMPRSS2 and Is Blocked by a Clinically Proven Protease Inhibitor

Single-cell meta-analysis of SARS-CoV-2 entry genes across tissues and demographics

Lipid-Associated Macrophages Control Metabolic Homeostasis in a Trem2-Dependent Manner

Innate Immune Landscape in Early Lung Adenocarcinoma by Paired Single-Cell Analyses

Therapy-Induced Evolution of Human Lung Cancer Revealed by Single-Cell RNA Sequencing

Single-cell RNA-seq reveals ectopic and aberrant lung-resident cell populations in idiopathic pulmonary fibrosis

Treatment of alveolar macrophages with cytochalasin D inhibits uptake and subsequent growth of Legionella pneumophila

A novel role for phagocytosis-like uptake in herpes simplex virus entry

SARS-CoV-2 evades immune detection in alveolar macrophages

CoV-2 cellular tropism

IL-6 increases endothelial permeability in vitro

COVID-19: the vasculature unleashed

How does SARS-CoV-2 cause COVID-19?

Enhanced isolation of SARS-CoV-2 by TMPRSS2-expressing cells

The Architecture of SARS-CoV-2 Transcriptome

Potent neutralizing antibodies from COVID-19 patients define multiple targets of vulnerability

Isolation of potent SARS-CoV-2 neutralizing antibodies and protection from disease in a small animal model

Engineering SARS-CoV-2 neutralizing antibodies for increased potency and reduced viral escape. bioRxiv

Cellular host factors for SARS-CoV-2 infection

Strategy for purification, culture, and infection of human lung macrophages with a SARS-CoV-2 pseudotyped lentivirus. (Left panel) Diagram of lenti-S-NLuc-tdT, a lentivirus pseudotyped to express SARS-CoV-2 Spike (S) protein on its surface and also engineered to express the reporter gene (boxed) encoding nuclear-targeted nanoluciferase (H2B-Nluc) and tdTomato fluorescent protein, separated by a self-cleaving T2A peptide

Human lung tissue obtained from surgical resections or organ donors were dissociated fresh, then labeled for magnetic activated cell sorting (MACS) using antibodies for the indicated surface markers

Lung macrophages (CD31EPCAM -CD206 + ) were cultured (37 o C, DMEM/F12 medium with 10% FBS and pen/strep) and infected by lenti-S-NLuc-tdT. After 4 hours, free virions were washed off and infection continued for 48 hours before quantification of infection by expression level of a lenti-S-NLuc-tdT reporter gene

Each pseudovirus was tested at two concentrations (diluted 1:1 or 1:2 with growth media). (c) Effect of phagocytosis inhibitor cytochalasin D or lysosome inhibitor hydroxychloroquine (HCQ) on lenti-S-NLuc-tdT infection of purified lung macrophages. Cells were pre-exposed for 2 hr to the indicated concentration of inhibitor before addition of lenti-S-NLuc-tdT for 4 hr before washing. NLuc luminescence values measured 48-hr after infection were normalized to control (non-neutralized) viral infections in each plate

ACE2/TMPRSS2 control cells (c) or purified human lung macrophages (d). An anti-HIV CD4 Na, CD4 naïve T cells

CD8 tissue resident memory T cells

MP, macrophage; pDC, plasmacytoid dendritic cell; mDC, myeloid dendritic cell; maDC, mature dendritic cell; Mono C, classical monocyte; Mono NC, nonclassical monocyte; Mono Int

Extended Data Figure 4. Localization and morphology of interstitial and alveolar macrophages in the lung. Additional examples as in Fig. 3e of RNAscope single molecule fluorescence in situ hybridization (smFISH) and immunostaining of alveolar (AM) and interstitial (IM) macrophages in non-cultured human lung of case 2, detecting general macrophage antigen CD68 (green, protein), AT1 antigen RAGE (white, protein), AM marker FABP4 (cyan, RNA), and IM marker RNASE1 (red, RNA)

Extended Data Figure 5. Distinct viral pseudotime trajectories in interstitial and alveolar macrophages

Uniform Manifold Approximation and Projection (UMAP) projection of alveolar macrophages (AM) and activated interstitial macrophages (a-IM) in infected human lung slices from 10x scRNA-seq from infections 1 and 4, as in Fig. 3. (b) Normalized expression of SARS-CoV-2 RNA in each cell as shown by the heat map scale. (c) Normalized value of a-IM viral pseudotime value as shown in Fig. 4a. (d) Normalized value of AM viral pseudotime value as shown in Fig. 4a. (e) Total viral RNA expression (log10(Viral UMIs + 1)) graphed against viral

Extended Data Figure 6. Expression of receptors and other SARS-CoV-2 host factors in different lung macrophage subtypes

Dot plot of scRNA-seq results of freshly profiled human lung slice cultures from cases 1 and 4, as in Fig. 3 showing for each indicated macrophage subtype (AM, alveolar macrophage; IM, interstitial macrophage; a-IM, activated interstitial macrophage) the fraction of expressing cells (% Expression) and mean expression value among expressing cells (ln(UP10K+1) of key proviral host factors in the SARS-CoV-2 replication cycle previously identified in CRISPRbased functional genetic screens 71 . Genes are grouped based on different steps of the viral life cycle (black font) and their normal cellular functions (colored font). Dots representing genes differentially up-regulated in a-IMs are outlined in red, and dots representing genes differentially up

Summary of viral subgenomic junction discovery from single lung cells using SICILIAN GeneA JuncPosR1A GeneB JuncPosR1B Junction_type numReads SARS-CoV-2_S 21761 SARS-CoV-2_S 21789 Internal 168

SARS-CoV-2_L 66 SARS-CoV-2_ORF6 27385 Canonical 27 unknown 1383 SARS-CoV-2_N 29494

32 (out of 982) novel junctions that pass SICILIAN from all infection experiments are shown, ordered by number of reads supporting the junction. The 5' and 3' positions of the junctions are indicated, along with their classification as Canonical, L-novel, Internal, or 3' UTR (see main text and methods)

We are grateful to the lung donors and Donor Network West for whole lung tissue, and to an anonymous financial donor for construction of a new BSL3 facility. We thank Amy Kistler