Dysregulation of the Leukocyte Signaling Landscape during Acute COVID-19

The global COVID-19 pandemic has claimed the lives of more than 450,000 US citizens. Dysregulation of the immune system underlies the pathogenesis of COVID-19, with inflammation mediated local tissue injury to the lung in the setting of suppressed systemic immune function. To define the molecular mechanisms of immune dysfunction in COVID-19 we utilized a systems immunology approach centered on the circulating leukocyte phosphoproteome measured by mass cytometry. COVID-19 is associated with wholesale activation of a broad set of signaling pathways across myeloid and lymphoid cell populations. STAT3 phosphorylation predominated in both monocytes and T cells and was tightly correlated with circulating IL-6 levels. High levels of STAT3 phosphorylation was associated with decreased markers of myeloid cell maturation/activation and decreased ex-vivo T cell IFN-gamma production, demonstrating that during COVID-19 dysregulated cellular activation is associated with suppression of immune effector cell function. Collectively, these data reconcile the systemic inflammatory response and functional immunosuppression induced by COVID-19 and suggest STAT3 signaling may be the central pathophysiologic mechanism driving immune dysfunction in COVID-19.

samples were first barcoded by incubation with Cell-ID 20-Plex Pd Barcoding Kit (Fluidigm) following manufacturers protocol. Up to 10 samples were pooled, washed x2 with CyFACS then resuspended in 0.5 mL CyFACS. Fc Receptors were blocked by adding 45 uL TruStain FcX (Biolegend) at RT for 10 min.
Surface antibody cocktail was then added (see Extended Data Table 1) and sample incubated for 1 hour on ice. Cells were then washed in 10 mL of CyFACS buffer x1.
For intracellular staining cell pellet was resuspended in -20° C methanol to a final concentration of 5x10^6 cells/ml and incubated at -20° C for 30 min, then washed 2x in 10 mL of ice-cold CyFACS buffer.
Cells were resuspended in 1mL CyFACS buffer then incubated for 60 minutes on ice with intracellular antibody cocktail (see Extended Data Table 1). Cells were washed x2 with 10 mL CyFACS buffer, then resuspended in FACS buffer containing 2% PFA and Cell ID Intercalator Solution (Fluidigm, San Francisco, CA) following manufacturers protocol for DNA staining.
Cells were analyzed on a Fluidigm CyTOF 2 Mass Cytometer. Samples were first resuspended in water+10% EQ Four Element Calibration Beads (Fluidigm) then acquired with an event rate of ~500 events/sec. CyTOF data were analyzed using the FlowJo software platform (BD, Franklin Lakes, NJ).
All antibodies were commercially conjugated to heavy metal isotopes except for clones B1.1, WM53, ICRF44, VI-PL2 and 7C9. For these clones unconjugated antibodies were purchased and conjugated using MAXPAR Antibody Labeling Kit (Fluidigm, San Francisco, CA) following manufacturers protocols.
Soluble Molecule Determination. For plasma studies, cellular elements of whole blood were pelleted by centrifugation at 1000 xg for 7 minutes at room temperature; soluble phase was aspirated, aliquoted and stored at -80° C. Cytokine and chemokine levels were determined by multiplex bead array (Cytokine Human Magnetic 35 Plex Panel, ThermoFisher Scientific) following manufacturers protocol. Samples were acquired on a Luminex FLEXMAP 3D instrument system.
Enzyme-linked Immunospot (ELISpot) assay. Peripheral blood mononuclear cells (PBMCs) were isolated from 5 mL fresh whole blood as previously described 22 . Total number of PBMCs was determined using a Vi-Cell™ viability analyzer (Beckman Coulter, Brea, CA, USA). Flow cytometry was performed for cell typing, staining for CD3, CD4, CD8 and CD14. Detection of ex vivo production of IFN-ɣ was assessed by ELISpot using precoated plates (ImmunoSpot by Cellular technology Limited (CTL), Cleveland, OH, USA) as per manufacturers protocols. Cells were incubated in serum-free media (CTL) with 500 ng/mL of anti-CD3 (Bio-legend clone HIT3a) with 2.5 µg/mL of anti-CD28 (Bio-legend clone CD28.2) antibodies in a total well volume of 200 µg for 2.5 x 10 4 PBMCs. Following overnight incubation at 37° C in 5% CO2, biotinylated detection antibody, streptavidin bound alkaline phosphatase and developer solution were applied as per manufacturer instructions. ELISpot analysis was performed using a CTL series 6 ImmunoSpot Universal Analyzer with Immunospot 7.0 professional software (CTL Analyzers, Shaker Heights, OH).
Data Analysis. Cell number, cell frequency, subpopulation frequency and signal intensities were compared across healthy, moderate COVID-19 and severe COVID-19 cohorts by non-parametric Kruskall-Wallis ANOVA, followed by correction for a false discovery rate (FDR) using the method of Benjamini, Krieger and Yekutieli. Features with <1% FDR were considered significant. Post-hoc bivariate comparison was done by Dunn's multiple comparison test. To correct for batch effects across CyTOF runs, signal intensities were normalized to a common reference sample by dividing the experimental sample geometric mean by the reference sample geometric mean for each feature. Normalized signal intensities were used for downstream analysis. Cytokine data was analyzed by Kruskall-Wallis ANOVA followed by correction for a false discovery rate (FDR) using the method of Benjamini, Krieger and Yekutieli. Features with <1% FDR were considered significant. Post-hoc bivariate comparison was done by Dunn's multiple comparison test. ELISpot data was analyzed by Kruskall-Wallis ANOVA. Post-hoc bivariate comparison was done by Dunn's multiple comparison test.

Correlation between CyTOF features and soluble markers and ELISpot results was done with Spearman
Rank Order Correlation.

Results:
Mass-Cytometry Defines the Immunologic Pathways Induced by COVID-19. To define the effects of SARS-COV2 infection on the intracellular signaling environment of circulating leukocytes, we recruited a cohort of 63 subjects with COVID-19. COVID-19 severity was defined using modified WHO disease severity criteria. Critically ill patients requiring intensive care were characterized as "Severe"; patients requiring inpatient care but not intensive care were defined as "Moderate". Blood was isolated from 20 subjects with moderate illness and 43 subjects with severe illness. All samples were obtained within 3 days of hospital presentation. A cohort of healthy donors served as controls. Peripheral blood was analyzed by mass cytometry. For subsets of these patients, plasma cytokine levels and T cell interferon gamma production by ELISpot assay were assayed ( Figure 1A). Demographic characteristics of all cohorts are shown in Extended Data Table 2. Subjects with severe COVID-19 had a longer length of stay   as compared to subjects with moderate COVID-19. 58% of patients with severe COVID-19 required   mechanical ventilation, and within the severe COVID-19 cohort, mortality was 37%. Consistent with prior studies, clinical laboratory values demonstrated that COVID-19 was associated with leukocytosis ( Figure 1B). We also identified anemia in both moderate and severe COVID-19 patients. Severe COVID-19 was associated with elevated serum creatinine, although there was no significant difference between subjects with moderate vs. severe disease. Both moderate and severe COVID-19 was associated with a coagulopathy manifest as an elevated INR; further, subjects with severe COVID-19 had increased INR as compared to moderately ill subjects ( Figure 1B). Complete clinical laboratory profiles are shown in Extended Data Table 3.
We used mass-cytometry to define the immunologic features associated with COVID-19 disease. We analyzed blood using a panel of heavy-metal conjugated antibodies recognizing both cell surface antigens and intracellular signaling molecules coupled with mass cytometry (See Extended Data Table 1).
Cell type was defined based on cell surface phenotype by on manual gating using canonical markers (Extended Data Figure 1). Within each population, we measured the signal intensity of 15 intracellular signaling molecules which are key mediators of T cell and monocyte response to invading pathogens.
Altogether, we defined 95 cellular features that identify the dominant response of immune effector cells to COVID-19 infection (Extended Data Table 4). To understand the overall structure of the CyTOF data we performed factor reduction by Principal Component Analysis. Figure 1C shows an X-Y plot of the first two principal components, which together account for 45% of the overall variance in the dataset.
Healthy subjects were tightly arrayed along a common line. In contrast, subjects with COVID-19 formed a distinct cluster with severely ill subjects having a higher variance than that of moderately ill subjects.
These PCA plots demonstrate that our CyTOF features can segregate healthy donors from moderately and severely ill COVID-19 patients. To identify specific features associated with COVID-19 disease, we used the non-parametric rank-order Kruskall-Wallis test to compare the distribution of each feature across groups (healthy controls, moderate COVID-19, severe COVID-19) and corrected for a 1% false discovery rate (FDR) using the Benjamini, Krieger and Yekutieli approach. After false-discovery correction, we identified 43 features with differential variance across condition (healthy, moderate COVID-19 and severe COVID-19, see Extended Data Table 4). To visualize the interaction between pvalue, false discovery and effect size we first calculated an effect size for each feature shown in Extended Data Table 4. Effect size was estimated using a modified z-score, calculated as the difference in means between subjects with severe COVID-19 and healthy controls, divided by the standard deviation of the healthy control population. Figure 1D shows volcano plot of effect size vs. p-value, with features reaching an FDR of 1% shown in green.
To define the immunologic effects of COVID-19, on the absolute numbers of different immune effectors comprising innate and adaptive immunity, we first evaluated circulating leukocyte frequency and number. Using canonical surface markers to define leukocyte types, we measured the frequency and absolute number of circulating leukocyte cell populations. Consistent with prior studies, we found that COVID-19 was associated with an increase in neutrophil frequency, although there was a trend toward increased neutrophil numbers, this trend did not reach a statistically significant threshold. We did measure a significant decrease in frequency of both innate and adaptive leukocytes including monocytes, CD4 T cells, CD8 T cells, plasmacytoid dendritic cells (pDC) and NK cells. These differences in frequency translated to absolute leukopenia during COVID-19 for most of the cell populations measured, including monocytes, CD4 and CD8 T cells and pDC ( Figure 2). We detected no differences in B-cell frequency or number. To further characterize the effect of COVID-19 on the leukocyte phosphoprotein landscape we measured the magnitude and direction of the signaling effect induced by COVID-19. Effect size was defined by a modified z-score calculated as the difference in median signaling intensity between severe COVID-19 and healthy controls, divided by the standard deviation in the control population. COVID-19 was associated with wholesale increases in signaling phosphoprotein levels. Consistent with hierarchical clustering data, we measured the largest effect sizes measured for STAT3 and STAT1 ( Figure 4A). As compared to healthy donors, we found only increased levels of signaling molecule phosphorylation, and in no case was COVID-19 associated with decreased phosphoprotein signal intensity as compared to healthy donors ( Figure 4A).
Within the innate immune compartment, we evaluated changes in signaling phosphoprotein levels in monocytes, NK cells and neutrophils. In monocytes, we found that as compared to healthy donors both moderate and severe COVID-19 was associated with increased phosphorylation of STAT3, CREB, MAPKAP2 and ERK; there was no difference between moderate and severe COVID-19. We saw broader but less robust activation in NK cells, including elevated levels of phosphorylated STAT1, STAT3, ERK, MAPKAP2, CREB and IκB. In neutrophils, phospho-STAT1 and phospho-ERK distinguished healthy donors for patients with COVID-19 ( Figure 4B).
Among T cells, we detected broad increases in signaling protein phosphorylation in both CD4 and CD8 T cells. Notably in both populations, STAT3 phosphorylation predominated, with a >7-fold increase in phospho-STAT3 in CD4 T cells and a >3-fold increase in CD8 T cells in severe COVID-19 subjects vs. healthy donors ( Figure 4C). Phospho-STAT3 levels in CD4 and CD8 T cells distinguished between healthy donors, moderate COVID-19, and severe COVID-19. STAT1, MAPKAP2, CREB and IκBα were also increased in severe COVID-19 in both T cell populations. In both CD4 and CD8 T cells, MAPKAP2 and ITK/BTK phosphoprotein levels were increased in moderate vs. healthy donors. In contrast with the innate immune cell populations, we did not detect significant levels of ERK phosphorylation in CD4 or CD8 T cells ( Figure 4C).
During COVID-19 Inflammatory Cytokines are associated with STAT3 and ERK phosphorylation in Circulating Leukocytes. To define the interactions between the leukocyte phosphoproteome and the systemic inflammatory response during COVID-19, we measured plasma levels of 36 cytokines and chemokines in a subset of the subjects using a multiplexed assay (see Extended Data Table 5).
To understand the pathways driving cellular activation in neutrophils, monocytes and NK cells, we evaluated the relationship between phosphoprotein levels and circulating cytokine levels by measuring the correlation between phosphoprotein levels and cytokine levels across all samples (healthy, moderate COVID-19 and Severe COVID-19). Figure 5B shows a heatmap of Spearman correlation coefficients and 5C shows scatter plots of a subset of highly correlated features. In monocytes and NK cells, phospho-STAT3 levels were strongly correlated with circulating levels of IL-6 and CXCL10. In neutrophils phospho-STAT1 was highly correlated with IL-6 and CXCL10 as were levels of phospho-STAT1 in neutrophils. In contrast, monocyte phospho-ERK levels were most correlated with plasma levels of IL-8 ( Figure 5A/B).
We applied a similar approach to define the signaling pathways activated in circulating CD4 and CD8 T cells. We again sought to correlate protein phosphorylation with circulating cytokine levels. In both CD4 and CD8 T cells, STAT1 and STAT3 phosphorylation were strongly correlated with IL-6, CXCL10, IL-8 and IL-1RA levels ( Figure 6A/B). Consistent with the canonical signaling downstream of the IL-6 receptor 28, 29, 30 , CD4 T cell pSTAT3 was most highly correlated with circulating IL-6 levels, with a Spearman R of 0.85. This suggest that circulating IL-6 drives CD4 T cell STAT3 activation during acute COVID-19.
Signaling Phosphoprotein Levels Correlate with Immune Cell Functional Metrics. Severe COVID-19 is associated with a dysregulation of the myeloid cell compartment characterized by increased numbers of immature neutrophils in circulation, a shift to HLA-DR-lo monocytes and defects in the function of mature neutrophils in the circulation. Prior work has characterized significant heterogeneity in the circulating neutrophil compartment induced by COVID-19, including accumulation of neutrophils expressing decreased levels of HLA-DR, CD15, CD11b. Consistent with this, we detected decreased mean surface expression of HLA-DR and CD15 on the neutrophils during both moderate and severe COVID-19 as compared to healthy donors. CD11b significantly decreased in severe COVID-19 as compared to both healthy donors and subjects with moderate COVID-19 disease. We also evaluated the relationship between changes in phosphoprotein levels and surface marker expression. In neutrophils, we found that surface CD11b levels were negatively correlated with intracellular phospho-ERK (pERK) levels, with lower mean levels of CD11b in samples with higher pERK levels. Surface expression of CD15 on neutrophils was more tightly (and negatively) correlated with phospho-STAT1 levels ( Figure 6B).
Similarly, we found lower levels of CD11b, HLA-DR, CD14 and CD33 on the monocyte surface ( Figure 6C).
In monocytes, we again measured a negative correlation between increased signaling phosphoprotein levels and surface marker expression. In monocytes, CD11b levels were negatively correlated with phospho-STAT3 levels and HLA-DR levels correlated with phospho-CREB levels ( Figure 6D).
We have previously shown that critical illness including COVID-19 is associated with adaptive immune dysfunction as measured by defects in T cell IFN-γ production in response to ex vivo T cell receptor stimulation 22 . Consistent with these data, we found that COVID-19 was associated with attenuated production of IFN-γ after stimulation of peripheral blood mononuclear cells with agonistic antibodies against CD3/CD28, suggesting a defect in T cell function during COVID-19 as compared to healthy controls ( Figure 6E). We then assessed the relationship between phosphoprotein levels and T cell function. We found that increased pSTAT3 levels in both CD4 and CD8 T cells was associated with decreased production of IFN-γ in response to CD3/CD28 stimulation, suggesting increased STAT3 activation as a mechanism for defects in T cell function during COVID-19.

Discussion:
We deployed a systems biology approach centered on a mass-cytometry phosphoproteome assay combined with multiplexed measurement of circulating cytokine levels and ex-vivo stimulated T cell function by ELISpot to define the effect of COVID-19 on the intracellular signaling landscape of circulating leukocytes during acute COVID-19. We find that COVID-19 is associated with a wholesale increase in signaling protein phosphorylation in both myeloid and lymphoid cell populations across multiple signaling pathways. In all cell populations there was a strong, positive correlation between an activated phosphoproteome signature and circulating inflammatory cytokine levels. In particular, STAT3 activation was tightly correlated with plasma IL-6 levels. In myeloid cells signaling protein phosphorylation levels were negatively correlated with surface markers of activation and maturation.
Similarly, in T cells, increased STAT3 phosphorylation was specifically associated with defects in IFN-γ production. These data reconcile the coincident inflammatory cytokinemia and functional immunosuppression induced by COVID-19 and suggest modulation of STAT3 as a potential therapeutic avenue to restore T cell function during severe COVID-19.
Consistent with past studies 16, 17, 18 , we found that COVID-19 was associated with an increase in neutrophil frequency and a concomitant decrease in other circulating leukocyte populations including monocytes, basophils and both CD4 and CD8 T cells. The increased neutrophil frequency was associated with a decreased expression of surface markers of maturation including CD11b, HLA-DR and CD15.
COVID-19 is associated with accumulation of peripherally circulating immature low-density neutrophils (LDN) 14 , previously characterized to have low-to-intermediate surface expression of CD11b 13 and HLA-DR 13,14,15 . Similarly, CD15 is a marker of neutrophil antimicrobial function, with decreased CD15 expression being associated with high TB microbial burden and treatment failure 31 . We also found decreased expression of monocyte maturation markers including CD33, CD14 and HLA-DR. This observation is consistent with studies showing that IL-6 is a suppressive cytokine to monocyte CD33 and CD14 levels 32 . This shift in surface marker phenotype likely reflects emergency hematopoiesis 33 with mobilization of immature neutrophils from the bone-marrow driven both by margination of mature neutrophils to the lungs and mobilization of immature cells by inflammatory cytokines including IL-6 and IL-8 34 .
To our knowledge this is the first description of the signaling phosphoproteome in peripheral blood leukocytes during acute infection with SARS-COV2. COVID-19 was associated with an overall increase in signaling protein phosphorylation, and in no case was there a decreased protein phosphorylation in COVID-19 vs. controls. We found that circulating leukocyte phosphoprotein signature alone can distinguish subjects with COVID-19 from healthy donor controls. In all cell populations, STAT3 was the predominant signaling pathway activated during COVID-19. STAT3 is the predominant downstream target of the IL-6 receptor 28, 30 . COVID-19 induces high levels of circulating IL-6, and we found that phospho-STAT3 levels were tightly correlated with plasma IL-6 levels. Although we detected upregulation of multiple signaling pathways, the specific increase in STAT3 phosphorylation during COVID-19 was striking. STAT3 has also been also been shown to inhibit the Type-I IFN antiviral response. COVID-19 is associated with increased circulating levels of IFN-alpha 26 , and our analysis did measure a 2-fold increase in IFNalpha in severe COVID vs. healthy donors (although these data did not meet the threshold for a 1% FDR, see Extended Data Table 5). Transcriptomic studies detected expression of Interferon-stimulated genes (ISG) in mild and moderate COVID-19, but an impaired Type-I interferon response in severe COVID-19 44,45 . Others reproduced these results, finding that severe influenza was associated with upregulation of ISG which was not seen in severely ill COVID-19 patient 26 . STAT3-mediated inhibition of the IFN response is independent of STAT3 transcriptional activity and is mediated through negative regulation of the cytosolic dsRNA receptor MDA5 44 . MDA5 serves as a pattern recognition receptor for SARS-CoV2 underlying the IFN response during COVID-19 46 , and the decreased ISG expression during COVID-19 may result from increased STAT3 mediated inhibition of MDA5 activity.
We acknowledge several limitations in our study. Foremost, our study design provides limited clinical information for our cohort of COVID-19 subjects, and we have not associated our findings with clinical outcomes from COVID-19. We have segregated our patients into "moderate" and "severe" COVID-19 disease, and we do report clinical outcomes including ICU admission, ventilator days and hospital length of stay are segregate based on disease severity, and we report differences in signaling phosphoproteome between our cohorts, but our limited sample size and limited clinical data constrains our ability to do a controlled analysis of the relationship between specific phosphoproteome features and clinical outcome. We predict that phospho-STAT3 levels on hospital admission may have prognostic significance, but this hypothesis will require further investigation. Our data analysis approach also has limitations. We have analyzed our CyTOF data by manual gating of canonical markers. We have only compared signaling protein phosphorylation across a limited set of major circulating populations. We recognize that by aggregating together leukocyte subpopulations, we may obscure differences in phosphoprotome activation in leukocyte subpopulations. The data utilized for this study will be made available in publicly accessible database, and we encourage collaborating investigators to undertake further analysis to elucidate these differences.
Given these limitations, these data demonstrate that COVID-19 causes an acute dysregulation of the signaling landscape of circulating leukocytes, defined primarily by exaggerated phosphorylation of STAT3. In both myeloid and lymphoid cell populations, pSTAT3 levels are correlated with evidence of functional immunosuppression, suggesting STAT3 blockade as a potential therapeutic pathway to restore immunocompetence during severe COVID-19. Prior reports have advocated STAT3 inhibition to restrain inflammation-mediated tissue damage 10

Figure 1
Severe