ACE2-containing defensosomes serve as decoys to inhibit SARS-CoV-2 infection

Few studies, if any, have analyzed exosomes in mucosal tissues of patients during an ongoing infection. This is especially true in the setting of SARS-CoV-2 infection where access to lower airway samples were limited due to concerns about possible risks for health care professionals [13–20]. In the airways, ACE2 expression exists as a gradient, with the highest levels in the nasal passageway, down to the lower airways where it is exclusively expressed on type II pneumocytes [21]. To determine whether ACE2+ exosomes are generated and are present in the respiratory tract, we analyzed bronchoalveolar lavage fluid (BALF) samples from 80 critically ill COVID-19 patients [22] using biochemical fractionation to enrich for exosomes followed by flow cytometry [9]. All patients in our cohort were admitted to the intensive care unit (ICU) due to respiratory failure requiring invasive mechanical ventilation and tested positive for SARS-CoV-2 infection [22]. Applying stringent gating criteria for exosomes, we observed remarkable interindividual variation in the proportion of exosomes with surface ACE2 and the levels of ACE2 present on each exosome measured by mean fluorescence intensity (MFI) (Figs 1A and S1A–S1C). We found that 10 patients had at least 2-fold higher exosome levels in their BALF compared to the mean exosome levels from the other patients. These data suggest that there is variation between individuals in how much ACE2 is loaded into each exosome as well as how many ACE2+ exosomes are produced by each person.

We then asked whether level of ACE2 on exosomes or proportion of ACE2+ exosomes correlate with clinical parameters such as patient age, sex, disease severity, infection status, and other comorbidities linked to SARS-CoV-2 infection such as diabetes or hypertension. Approximately 76.2% of the patients were male, mostly nonwhite (53.8%), with a mean age of 62 years (SD, 14.2). Hypertension and diabetes were reported in 52.5% and 38.3% of patients, respectively (S1 Table). Two patients were excluded from analysis because they were not within 3 standard deviations from the mean (47.96 days) for length of stay in the ICU. As a sensitivity analysis, we also ran the regressions using ACE2 MFI as a predictor of ICU stay outcome, resulting in the same trends in the covariates (S3 Table). Compiling the information regarding the proportion of exosomes that contain ACE2 in patient BALF, we ran different models using clinical data to test whether there was any association with ACE2+ exosomes and length of stay in the ICU, as there was no correlation with mortality. We performed a linear model and a negative binomial model to test this relationship, and we controlled for age, gender, culture results of BAL and blood, diabetes, and hypertension. We also included in a sensitivity analysis model an interaction term between gender, sex, and hypertension, as they are usually correlated. We expressed the coefficients without interactions for simplicity because the estimates did not significantly change. The linear model resulted in negative correlation between length of stay in the ICU and the proportion of ACE2+ exosomes found in the BALF (Table 1, Fig 1B). The negative binomial model predicted similar results (S2 Table). To account for increased length of stay in the ICU due to complications or subsequent infections, we also used ventilation days as an outcome and found that proportion of ACE2+ exosomes remained a significant predictor (S4 and S5 Tables). Also, we did not detect a correlation between ACE2+ exosomes and viral RNA (S1G Fig).

These analyses uncovered a threshold where patients with a proportion of ACE2+ exosomes lower than the mean (38.78%) generally have increased lengths of stay in both the ICU and hospital and increased days on a ventilator (Figs 1 and S1D). Individuals with diabetes (or both diabetes and hypertension) clustered below the average proportion of ACE2+ exosomes (Fig 1C), suggesting common comorbidities may influence the proportion of ACE2+ exosomes. Individuals with hypertension have been shown to have reduced levels of circulating ACE2 with the adverse consequence of increased concentration of the vasoconstricting molecule Angiotensin II due to activation of ACE [23]. However, neither diabetes nor hypertension were associated with statistically significant differences in ACE2 levels on exosomes (S1E and S1F Fig).

We previously showed that exosomes are induced by TLR9 activation in response to bacterial DNA or synthetic agonist CpG-A. We found that SARS-CoV-2 infection of A549^ACE2+ cells also induced robust production of ACE2+ exosomes (Fig 2A). Similar to our observations with bacteria, SARS-CoV-2-induced exosome production was impaired by ATG16L1 knockdown (Fig 2A). Therefore, the enhanced production of ACE2+ exosomes display a key hallmark of defensosomes—induced in response to an infectious agent in a manner dependent on ATG16L1.

Fig 2. TLR ligands and SARS-CoV-2 induce exosome production.

(A) ACE2⁺ A549 (NT) and ACE2⁺ A549 cells knocked down for ATG16L1 using shRNA (ATG16L1 KD) were infected with TCID50 SARS-CoV-2 USA-WA1/2020 for 72 h. Exosomes were isolated from supernatants and quantified in exact counts (left) and fold change (right). (B) Quantification of exosomes from A549 cell-culture supernatant by flow cytometry 16 h after stimulation with 32 μg of oxidized mtDNA (ox mtDNA), mtDNA products without oxidized bases (non-ox mtDNA), or genomic DNA (gDNA) (n = 4–10). (C) Quantification of exosomes from A549 cells stimulated for 16 h with 1 μg/mL 3′3′-cGAMP (n = 4), 2′3′-cGAMP (n = 9), 0.5 μM CpG-A (n = 8), and 0.5 μM CpG-B (n = 7). CpG-A will be referred to as CpG from this point on. (D) Quantification of exosomes from A549 cells stimulated with various dilutions of IFN-β for 16 h (n = 3–6). (E) Quantification of exosomes from A549 cells stimulated with various dilutions of IFN-λ for 16 h (n = 3–6). (F) Quantification of exosomes following stimulation of A549 cells with 100 U/mL IFNα for 16 h (n = 5–9). (G) Exosome quantification of IFNAR/γR/λR TKO A549 cells stimulated with 500 U/mL IFN-β for 16 h compared to non-transduced (NT) A549 cells (n = 7–11). (H) Quantification of exosomes from NT A549 or TKO A549 cells stimulated with CpG for 16 h (n = 3–6). (I) Quantification of exosomes from A549 cells stimulated with HMW or LMW Poly I:C with LyoVec at 50, 100, 500 ng/mL (n = 2–7). (J) Exosomes were quantified from A549 cells pretreated with 100 U/mL IFNα for 24 h, followed by stimulation with 10 μg/mL LMW or HMW PolyI:C. (K) Quantification of exosomes from A549 cells and TLR7-expressing A549 cells stimulated with 1 μg/mL R848 for 16 h (n = 3–5). (A–K): Exosomes: CD9+ CD63+ CD81+ events. Measurements were taken from distinct samples. Error bars represent mean ± SEM and at least 2 independent experiments were performed. Underlying data can be found in S1 Data. C–F, I, J One-way ANOVA with multiple comparisons with Dunnett’s post-test. A, G, H, K Two-way ANOVA using Dunnett’s post-test. * P  ≤  0.05; ** P ≤ 0.01. HMW, high molecular weight; LMW, low molecular weight; ns, not significant; NT, non-transduced; SARS‑CoV‑2, Severe Acute Respiratory Syndrome Coronavirus 2; TKO, triple knockout.

https://doi.org/10.1371/journal.pbio.3001754.g002

Although SARS-CoV-2 is not predicted to directly activate DNA sensors, damage during respiratory tract infection can lead to production of oxidized mitochondrial DNA (ox-mtDNA) [24,25], which is known to be a potent ligand for TLR9 [26]. Stimulation with ox-mtDNA increased exosome production in A549 cells, but this was not observed with sheared genomic DNA (gDNA) or DNA fragments without any oxidized bases (Fig 2B). We also tested CpG-B, which can activate TLR9, but unlike CpG-A, lacks a phosphodiester backbone and poorly induces IFNs [27]. CpG-B stimulation failed to elicit exosomes from A549 cells (Fig 2C). SARS-CoV-2 has been shown to indirectly activate the cytosolic DNA-sensing pathway dependent on cGAS/STING [28]. However, ligands for cGAS/STING (2′3′-cGAMP and 3′3′-cGAMP) failed to induce exosomes (Fig 2C).

It is possible that IFNs generated downstream of TLR9 activation mediate exosome production. Stimulation of A549 cells with IFN-β induced exosomes in a concentration-dependent manner, where at higher concentrations, the number of exosomes began to decrease (Fig 2D), which may be due to ISGylation and degradation of the compartment where exosomes are generated [29]. IFN-α and IFN-λ2 also induced exosomes (Fig 2E and 2F). As expected, IFN-β failed to elicit exosomes from triple knockout (TKO) A549 cells that lack receptors necessary for recognition of all 3 major classes of IFNs (IFNAR, IFNGR, and IFNLR) (Fig 2G). In contrast, TKO A549 cells were still able to produce exosomes in response to CpG-A (Fig 2H). Thus, IFN stimulation is sufficient to induce exosome production, but are not required downstream of nucleic acid sensing.

SARS-CoV-2 infection generates viral ssRNA and dsRNA that may serve as signals to induce exosomes. When transfected into cells, low and high molecular weight versions of the dsRNA mimetic poly I:C preferentially activate the cytosolic RNA sensors RIG-I and MDA-5, respectively, but we found that neither induced exosomes (Fig 2I). In A549 cells, the endosomal dsRNA sensor TLR3 is expressed at low levels; IFN-α treatment induces TLR3 to functionally active levels, allowing A549 cells to become sensitive to untransfected poly I:C [30]. Following pretreatment with IFNα for 24 h, A549 cells produced 10- to 20-fold more exosomes in response to poly I:C without a transfection agent (Fig 2J). A549 cells do not express the endosomal ssRNA sensors TLR7 or TLR8. We found that ectopic expression of TLR7 led to exosome induction upon stimulation with a synthetic agonist, R848 (Fig 2K). Therefore, multiple innate immune ligands associated with SARS-CoV-2 infection and IFNs induce exosome production.

An IFN-inducible splice variant of ACE2 (dACE2) missing the N-terminal region necessary for SARS-CoV-2 binding has been identified [31–33]. Using an antibody targeting the C-terminus of ACE2 (Abcam 15348), we readily detected full-length ACE2, but not dACE2 protein in exosomes induced by bafilomycin or IFNα in various cell lines using western blotting (S6 Fig), consistent with reports indicating that dACE2 is post-transcriptionally unstable [32]. We found that the antibody used for the flow cytometry experiments (R&D AF933) exclusively detected full-length ACE2. Hence, we can conclude that full-length ACE2 is incorporated into exosomes from BALF. With available RNA-seq data on the cellular fraction of BALF from our patient cohort [22], we were able to detect low expression of dACE2 in 50 out of 80 patients (S4A Fig). dACE2 expression did not correlate with disease severity (length of ICU stay), but positively correlated with the proportion of ACE2+ exosomes and ACE2 MFI on exosomes found in the BAL (S4B–S4D Fig). dACE2 expression only negatively correlated with 2 genes previously identified to have a positive correlation with dACE2 from data on bronchial tumors in the TCGA-LUSC database (S5A–S5T Fig) [33]. As dACE2 lacks the catalytic component of full-length ACE2 and the binding site for Spike and is expressed at low levels, we believe dACE2 has no direct impact on exosome neutralization of SARS-CoV-2, but may be a reflection of IFN responses.

Given that IFNs induce exosome production (Fig 2D–2F), we next asked whether ACE2+ exosome production was linked to the IFN response in our cohort of patients. We performed differential gene analysis on “high” ACE2+ exosome producers or individuals with “high” ACE2 levels on their exosomes compared to their “low” counterparts. We found that “high” ACE2+ exosome producers had an enrichment of genes found in the Coronavirus pathogenesis pathway and activation of cytosolic PRRs, and a down-regulation of pathways related to EIF2 signaling and oxidative phosphorylation. ACE2 MFI “high” individuals had an enrichment in IFN and PI3K/AKT signaling pathways (S7–S9 Figs). These findings reinforce the association between antiviral signaling and defensosome production.

Our finding that ACE2+ exosomes in patient BALF display an inverse correlation with length of hospital stay raised the possibility that they play a role during viral infection akin to our previous description of defensosomes during bacterial toxinosis. Therefore, we tested whether addition of exosomes can protect Vero E6 cells, which are highly susceptible to SARS-CoV-2 infection (Fig 3A). Exosomes were isolated from A549^ACE2+ cells (Fig 3B) and mixed with human SARS-CoV-2 USA WA1/2020 isolate at an MOI of 0.01. We included a polyclonal neutralizing antibody targeting spike (nAb) as a positive control. Addition of ACE2+ exosomes led to a dose-dependent reduction in viral nucleoprotein (NP) staining by immunofluorescence microscopy at 24 h post-infection (Figs 3C and 3D and S2A), whereas ACE2− exosomes from untransduced control A549s failed to inhibit infection. Also, addition of the microvesicle fraction that does not contain exosomes (5,000xg) did not decrease infection levels (Fig 3C). We also tested neutralization in human airway epithelial cultures (HAECs), an air-liquid interfact (ALI) model that contains all major cell types found in the bronchial epithelium (basal, ciliated, secretory), organized in pseudostratified architecture [34]. Addition of ACE2+ exosomes mixed with SARS-CoV-2 to the apical side (“lumen”) blocked infection of HAECs, while ACE2− exosomes did not show any significant effect (Figs 3E and S2B).

Fig 3. ACE2+ exosomes protect against SARS-CoV-2 infection.

(A) Workflow for exosome neutralization assay used to assess protective effects of mixing SARS-CoV-2 and exosomes. Created with BioRender.com. (B) Representative histogram of cell-surface ACE2 on A549 and A549^ACE2 cells. (C) Representative immunofluorescence images of Vero E6 cells infected with SARS-CoV-2, pellet from 5,000xg spin, approximately 20 K exosomes from ACE2− cells, approximately 20 K exosomes from ACE2+ cells, neutralizing antibody, or with media alone stained for viral N protein. Scale bar: 0.625 mm, (D) Percentage of infected Vero E6 cells based on positive staining for N protein after infection with SARS-CoV-2 mixed with exosomes isolated from A549^ACE2 (teal) or untransduced A549 cells (black). Compiled means ± SEM from 5 experiments, values normalized to infection with SARS-CoV-2 alone. (E) Representative immunofluorescence images of HAECs infected with mock (media only control), SARS-CoV-2 alone, approximately 11 K (1) or approximately 22 K (2) ACE2+ exosomes with virus, approximately 150 K (3) ACE2− exosomes with virus, and neutralizing antibody with virus (nAb). Images were taken 72 hpi. Scale bar: 0.156 mm. (F) Representative flow cytometry plots of exosomes isolated from Vero E6 cells stimulated with CpG. FSC: forward scatter, SSC: side scatter. (G) Quantification of exosomes from Vero E6 cells and Vero E6 cells transduced with a non-targeting shRNA for ATG16L1 stimulated with PBS (n = 4) or 1 uM CpG-A (n = 4) for 16 h. (H) Quantification of infected cells after infection of Vero E6 cells with SARS-CoV-2 only, SARS-CoV-2 mixed with supernatants of cells pretreated overnight with CpG-A (induced), or SARS-CoV-2 added to CpG-A pretreated cells following removal of supernatant (wash). Exosomes: CD9, CD63, CD81+, PKH67+ events. Means ± SEM of at least 2 independent experiments. G Two-way ANOVA with Dunnett’s post-test compared to NT PBS. H One-way ANOVA with Dunnett’s post-test compared to PBS. Error bars represent the mean ± SEM of at least 2 independent experiments, with measurements taken from distinct samples. Underlying data can be found in S1 Data. ** P ≤ 0.01; **** P  ≤  0.0001. HAEC, human airway epithelial culture; hpi, hours post infection; ns, not significant; SARS‑CoV‑2, Severe Acute Respiratory Syndrome Coronavirus 2.

https://doi.org/10.1371/journal.pbio.3001754.g003

We next tested whether inducing exosome production protects against viral infection without the need of adding exogenous exosomes from donor cells. We found that pretreatment with CpG-A, which we confirmed induces exosome production by Vero E6 cells in a manner dependent on ATG16L1 (Fig 3F and 3G), led to a reduction in cells infected with SARS-CoV-2 (Fig 3H). Removing the exosome-containing supernatant prior to infection led to similar levels of SARS-CoV-2 infection as unstimulated cells (Figs 3H and S2C). Further, knockdown of ATG16L1 in either A549 cells or Vero E6 cells did not impact viral replication levels in either cell type (S2D and S2E Fig). Vero E6 cells generally cannot produce IFNs [35,36], although hantavirus infection was shown to elicit IFNλ from these cells [37]. To formally demonstrate that the exosome fraction of the supernatant from CpG-A-treated Vero E6 cells contains the antiviral activity, we biochemically isolated exosomes and tested whether they inhibit SARS-CoV-2 infection of target cells. The enriched exosomes were sufficient to reduce viral infection, and the degree of inhibition was even greater than the unmanipulated supernatant. Together, these results show that defensosomes can inhibit SARS-CoV-2 infection.

We next used electron microscopy techniques to visualize the interactions between SARS-CoV-2 virions and defensosomes. Viral particles with predicted spherical shape and uniform size of 80 to 100 nm in diameter were detected by negative stain imaging in control suspensions containing SARS-CoV-2 virions alone (Fig 4A) [38]. With the addition of exosomes, we detected structures with the characteristic collapsed morphology of exosomes clustered around particles with virion morphology. The difficulty in distinguishing exosomes and virions due to their similar shape and size precluded quantification of the data. Therefore, we used immunogold labeling in which exosomes and SARS-CoV-2 virions were identified by an anti-CD63 antibody coupled to 18-nm gold particle and an anti-NP antibody coupled to a nanogold particle (1.4 nm), respectively (Fig 4B). We observed an enrichment of CD63 labeling around SARS-CoV-2 virions when incubated with ACE2+ exosomes, but not ACE2− exosomes (Fig 4C). The pronounced clustering of CD63 in the ACE2+ exosome-virus mixtures may be indicative of multiple exosomes bound to the virions (or vice versa); however, we cannot definitively conclude the number of exosomes due to the lack of contrast of the exosome membranes.

Fig 4. SARS-CoV-2 virions associate with ACE2+ exosomes.

(A) TEM of negative stained exosomes and virus alone. Arrowheads: putative exosomes. Scale bars: 200 nm. Red arrows mark the characteristic collapsing of exosomes during TEM imaging. (B) Double immunogold labeling TEM images of SARs-CoV-2 virions mixed with ACE2+ exosomes (right panel, scale bar: 50 nm) or ACE2− exosomes (left panel). Virions are labeled using an antibody for SARS-CoV-2 N, followed by nanogold-conjugated Fab (black triangle outline), which fills the entirety of the virion giving it the appearance of the larger dark mass. Exosomes are labeled using an antibody against CD63, followed by 18-nm colloidal gold-conjugated IgG (filled white triangles). Bottom panels are the same images as the top panels, with exosome membranes marked with dotted yellow outlines denotes to enhance their visibility. (C) Quantification of CD63 structures per virion present in a cluster containing both markers. ACE2+ (n = 68 clusters), ACE2− (n = 57). Error bars represent mean ± SEM from at least 2 independent experiments. Underlying data can be found in S1 Data. Unpaired t test with Welch’s correction. * P  ≤  0.05; ** P ≤ 0.01; *** P  ≤ 0.001; **** P  ≤ 0.0001. ns, not significant; SARS‑CoV‑2, Severe Acute Respiratory Syndrome Coronavirus 2; TEM, transmission electron microscopy.

https://doi.org/10.1371/journal.pbio.3001754.g004

Lastly, we applied cryo-electron microscopy and tomography to confirm these associations between SARS-CoV-2 virions and ACE2+ exosomes. Virions were easily distinguished by the presence of protrusions that resembled the club-like head of the spike protein and exosomes were larger single-membraned vesicles, spherical or ellipsoid in shape. Focusing on the area between the viral and exosome membranes, we were able to visualize an elongated structure resembling the post-fusion form of spike and ACE2 in some tomograms (S3A and S3B Fig, S2 Data and S3 Data). Tomographic reconstruction also captured several virions around larger exosomes as well as potential virions inside an exosome (S4 Data). Collectively, these data demonstrate that the presence of ACE2 on exosomes increases binding and clustering of virions.

The characteristics of this cohort have recently been described [22]. Briefly, we enrolled adults (>18 years old) admitted to the ICU due to respiratory failure requiring invasive mechanical ventilation with confirmed diagnosis of SARS-CoV-2 infection by reverse transcriptase polymerase chain reaction (RT-PCR). Lower airway samples (bronchoalveolar lavage (BAL)) were obtained during clinically indicated bronchoscopy performed for airway clearance or for percutaneous tracheostomy placement. Surviving subjects signed informed consent to participate in this study while samples and metadata from subjects who died or were incapacitated were de-identified and included in this study. All patients or their legal representative agreed to participate via our NYU IRB approved protocol (IRB no. s16-00122/01598). The study protocol was approved by the Institutional Review Board of New York University.

A549 cells were purchased from ATCC. A549^ACE2 cells were generated using lentiviral transduction of a human ACE2 cDNA-expressing plasmid (backbone: pLV-EF1a-IRES-Puro (Addgene 85132)) as previously described [34]. TLR7+ A549 were generated by lentiviral transduction of pcDNA3-TLR7-YFP (a gift from Doug Golenbock (Addgene plasmid # 13022; http://n2t.net/addgene:13022; RRID:Addgene_13022)). A549^ACE2 cells were selected using 1 μg/mL puromycin. Cells were cultured with 10% Dulbecco’s modified Eagle medium (DMEM (Gibco), 10% fetal calf serum (FCS), 1% non-essential amino acids (NEAA), 1% Penicillin/Streptomycin) and filtered using a 0.22 μm filter unit (Millipore). Vero E6 cells were purchased from ATCC (cat. No. CLR-1586) and maintained in same media as A549 cells, but including 10 mg/mL Amphotericin B. 293FTs were maintained in same media as A549 cells. Lentivirus was produced by transfecting 10-cm plate of 293FTs with 10 μg of plasmid using Lipofectamine 3000. Media was exchanged the next day and lentivirus was collected the following day; 1 mL lentivirus was used to transduce A549 cells using Polybrene (Millipore). TKO (IFNAR1/IFNLR/IFNGR1) A549 cells were generously provided by Marco Binder at the German Cancer Research Center (DKFZ), Heidelberg, Germany [63].

Maintenance of HAECs were done as described in [34]. At 48 h prior to infection, 6- to 8-week-old HAECs were washed apically twice for 30 min each with prewarmed PBS (Ca/Mg+). At 1 h prior to infection, cultures were washed apically twice for 30 min each with prewarmed PBS. Each culture was infected with a viral dilution containing with 1.35E+05 PFU with or without exosomes for 2 h at 37°C. The inoculum was removed, and the cultures were washed 3 times with prewarmed PBS. For each washing step, PBS was added to the apical surface, and cultures were incubated at 37°C for 30 min before the PBS was removed. The third wash was collected and stored at −80°C for titration by plaque assay on Vero E6 cells. Infectious particles were collected every 24 h (up to 72 h) by adding 60 μl of prewarmed PBS, incubating at 37°C for 30 min, and washes were stored at −80°C. Wells were fixed by submerging in 10% formalin for 24 h and washed 3 times with PBS before analysis by immunofluorescence. Fixed HAECs were quenched 50 mM NH₄Cl (in PBS), permeabilized using a solution of, 0.1% Saponin, and blocked with 2% BSA. Cultures were stained with anti-SARS nucleocapsid protein antibody that cross reacts with SARS-CoV-2 N (1:1,000, cat. no. 200-401-A50; Rockland) and goat-anti-rabbit Alexa Fluor 488 and DAPI.

Micrographs were collected using Leginon [64,65] at a dose rate of 11.21 e⁻/Ų/s with a total exposure of 3.50 s, for an accumulated dose of 39.23 e⁻/Ų. Intermediate frames were recorded every 0.07 s for a total of 50 frames per micrograph. Frames were aligned using Motioncor2 [66].

Tilt series were collected using SerialEM [67] at a dose rate of 11.21 e⁻/Ų/s with a total exposure of 0.25 s for an accumulated dose of 2.79 e⁻/Ų per tilted image. Intermediate frames were recorded every 0.05 s for a total of 5 frames per micrograph. Movies were collected in super-resolution mode. Tilt series were acquired in a bidirectional pattern starting from −18°, with a range from −60° to 60° in 3° step with a total dose of 114e⁻/Ų.

Frames were aligned using Warp [68], binned to the physical pixel size of 1.408 Å, and constructed into a tilt series stack for further processing. Micrographs within each tilt series were aligned with EMAN2 [69] and constructed into a tomogram at binning 4 (5.63 Å/pixel) using Fourier inversion. Tomograms were denoised using Topaz [70]. Denoised tomograms were inspected using IMOD [71] and clipped to appropriate size. Movies were made from selected tomograms using ImageJ. Automated segmentation was performed using EMAN2 to segment membranes, and further 3D rendering and refinement of Spike proteins was done manually on ORS Dragonfly (Object Research Systems, Quebec, Canada).

Raw reads can be found using Bioproject accession: PRJNA688510. Sequencing reads were mapped to the reference genome (hg38) using the STAR aligner (v2.5.0c) [72]. Alignments were guided by a Gene Transfer Format (GTF) file. The mean read insert sizes and their standard deviations were calculated using Picard tools (v.1.126) (http://broadinstitute.github.io/picard). The read count tables were generated using HTSeq (v0.6.0) [73], normalized based on their library size factors using DEseq2 [74], and differential expression analysis was performed. transcript per million (TPM) values for transcript abundances were calculated using kallisto (v. 0.46.2) [75].The read per million (RPM) normalized BigWig files were generated using BEDTools (v2.17.0) [76] and bedGraphToBigWig tool (v4). Gene set enrichment analysis was performed using GSEA tool (PMID: 16199517). To compare the level of similarity among the samples and their replicates, we used 2 methods: principal-component analysis and Euclidean distance-based sample clustering. All the downstream statistical analyses and generating plots were performed in R environment (v3.1.1) (https://www.r-project.org/). The pathway analyses and signaling networks for S8 and S9 Figs were generated through the use of QIAGEN IPA (QIAGEN, https://digitalinsights.qiagen.com/IPA) [77].

Statistical analyses were performed using GraphPad Prism 9.0 (GraphPad Software, San Diego, California, USA; https://www.graphpad.com/). All graphs show the mean and standard error of the mean (SEM). An unpaired 2-tailed t test was used to evaluate differences between 2 groups with Welch’s correction. One-way ANOVA with Dunnett’s post-test was used to evaluate experiments involving groups of 3 or more. Two-way ANOVA with Dunnett’s or Šidák’s post-test was used to evaluate groups with more than one variable. Correlation was analyzed using Pearson r. For each test, a p-value <0.05 is considered statistically significant. Clinical data was analyzed using R-studio. We performed descriptive analysis extracting data from EHR for these patients and summarizing this data with R. Linear model was used with the outcome as length of stay in the ICU. As a sensitivity analysis, we also performed a negative binomial model. with ACE2 MFI as a predictor instead of % ACE2 positivity, which yielded the same results.