Chemokines, soluble PD-L1, and immune cell hyporesponsiveness are distinct features of SARS-CoV-2 critical illness

Abstract

Critically ill patients manifest many of the same immune features seen in coronavirus disease 2019 (COVID-19), including both “cytokine storm” and “immune suppression.” However, direct comparisons of molecular and cellular profiles between contemporaneously enrolled critically ill patients with and without severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) are limited. We sought to identify immune signatures specifically enriched in critically ill patients with COVID-19 compared with patients without COVID-19. We enrolled a multisite prospective cohort of patients admitted under suspicion for COVID-19, who were then determined to be SARS-CoV-2-positive (n = 204) or -negative (n = 122). SARS-CoV-2-positive patients had higher plasma levels of CXCL10, sPD-L1, IFN-γ, CCL26, C-reactive protein (CRP), and TNF-α relative to SARS-CoV-2-negative patients adjusting for demographics and severity of illness (Bonferroni P value < 0.05). In contrast, the levels of IL-6, IL-8, IL-10, and IL-17A were not significantly different between the two groups. In SARS-CoV-2-positive patients, higher plasma levels of sPD-L1 and TNF-α were associated with fewer ventilator-free days (VFDs) and higher mortality rates (Bonferroni P value < 0.05). Lymphocyte chemoattractants such as CCL17 were associated with more severe respiratory failure in SARS-CoV-2-positive patients, but less severe respiratory failure in SARS-CoV-2-negative patients (P value for interaction < 0.01). Circulating T cells and monocytes from SARS-CoV-2-positive subjects were hyporesponsive to in vitro stimulation compared with SARS-CoV-2-negative subjects. Critically ill SARS-CoV-2-positive patients exhibit an immune signature of high interferon-induced lymphocyte chemoattractants (e.g., CXCL10 and CCL17) and immune cell hyporesponsiveness when directly compared with SARS-CoV-2-negative patients. This suggests a specific role for T-cell migration coupled with an immune-checkpoint regulatory response in COVID-19-related critical illness.

INTRODUCTION

Severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) is the novel coronavirus that causes coronavirus disease 2019 (COVID-19; 1). There have been over 250,000,000 people infected with SARS-CoV-2 and 5,100,000 deaths attributed to COVID-19 worldwide (2). Recent trials have identified immune-modulating therapies such as corticosteroids, IL-6 receptor antagonists, and JAK inhibitors that improve organ function and survival in patients with severe COVID-19 (3–5). Despite these effective therapies, mortality rates for patients admitted to the intensive care unit (ICU) remain high and range from 10% for subjects supported on the high-flow nasal cannula to 40% in patients who require invasive mechanical ventilation (3–5). A better understanding of the host immune response to SARS-CoV-2 in critically ill patients is necessary to better select therapeutic targets for critically ill patients with and without COVID-19.

There is an ongoing debate about the role of “cytokine storm” and “immune suppression” in COVID-19 pathophysiology, particularly in critically ill patients with the most severe disease. The systemic levels of proinflammatory cytokines such as IL-6 and TNF-α are clearly elevated in critically ill patients with COVID-19 compared with healthy individuals. However, studies comparing plasma cytokine levels from critically ill patients with COVID-19 with those from historical cohorts of subjects with bacterial sepsis and acute respiratory distress syndrome (ARDS) suggest a less robust early inflammatory response in patients with COVID-19 (6–9). Monocyte endotoxin tolerance and T-cell exhaustion have also been described in patients with COVID-19 (10–16), however, these are well-established responses that occur in critically ill patients without COVID-19 (17–22). No direct comparison of immune profiles between contemporaneously enrolled patients with SARS-CoV-2- and non-SARS-CoV-2-associated critical illness has been reported.

To address these limitations in our understanding of immune responses specific to COVID-19-related critical illness, we enrolled a prospective multisite cohort of critically ill patients suspected of having COVID-19 and measured molecular and cellular profiles. Our primary goal was to identify distinct as well as convergent immune responses between critically ill patients with or without SARS-CoV-2. We sought to test whether systemic interferon-induced responses, early innate and adaptive immune cell hyporesponsiveness, and a dysregulated relationship between immune cell phenotypes (e.g., chemokine receptors) and circulating mediators (e.g., chemokines) are key features in patients with SARS-CoV-2-associated critical illness compared with other forms of critical illness such as bacterial sepsis and ARDS.

MATERIALS AND METHODS

Study Population and Design

We conducted a prospective cohort study of patients admitted to three hospitals in Seattle, WA, between March 16, 2020 and May 16, 2021 (23). Clinical follow-up data were collected up to June 16, 2021. This study was granted a waiver of informed consent by the University of Washington Human Subjects Division given minimal risk, the urgency of COVID-19 research, and because research staff was unable to approach patients and legal authorized representatives given the limited supplies of personal protective equipment (STUDY 09763). Subjects were eligible if they had symptoms suggestive of SARS-CoV-2 infection (fever, respiratory symptoms including cough/shortness of breath, or sore throat), were admitted to an ICU with clinical suspicion for COVID-19, and had one of the following: 1) any respiratory support with supplemental oxygen or an oxygen saturation of < 94% on ambient air; or 2) any chest radiographic abnormality. Exclusion criteria included age ≤ 18 yr, pregnancy, or current incarceration. All subjects were enrolled and had plasma collected within 24 h of ICU admission. Peripheral blood mononuclear cells (PBMCs) were collected and analyzed from a subset of subjects with ARDS between 48 and 96 h after study enrollment. SARS-CoV-2-positive subjects were classified based on a positive SARS-CoV-2 RT-PCR nasal swab clinical test.

Plasma Mediator Measurements

We measured immune mediator concentrations from blood collected from an EDTA tube using electrochemiluminescent immunoassays per the manufacturer’s instructions (Meso Scale Discovery) [V-Plex Proinflammatory Panel 1 (K15049D); V-Plex Chemokine Panel 1 (K15047D); V-Plex Cytokine Panel 1 (K15050D); V-Plex Th17 Panel 1 (K15085D); and R-Plex sPD-L1 (F214C)]. All plasma samples underwent two freeze-thaw cycles before analysis. The percentage coefficients of variation (% CV) and limits of detection for all mediators we tested are shown in Supplemental Table S1 (all supplemental material is available at https://doi.org/10.6084/m9.figshare.19699990.v1). Analytes that did not meet any of the following quality control parameters were excluded from subsequent analysis: 1) intraplate % CV > 20%; 2) interplate % CV > 20%; or 3) > 10% of samples with a measurement below the lower limit of detection.

Cytometric Assays

We isolated PBMCs from peripheral blood by using cell preparation tubes per the manufacturer’s instructions (Becton, Dickinson and Company). PBMCs were cryopreserved in FBS/10% DMSO freezing solution. On the day of staining, samples were thawed and incubated in Thaw Media (10% FBS in RPMI) with DNase (50 units/mL) for 5 min, pelleted, and resuspended in Thaw Media. We stimulated cells with either phorbol 12-myristate 13-acetate with Ionomycin (PMA/Iono) or anti-CD3/CD28 beads because T-cell cytokine secretion can be influenced by different in vitro stimulation conditions (24). PMA/Iono activates protein kinase C, which bypasses the T-cell receptor (TCR) complex to activate several intracellular signaling pathways. Anti-CD3/CD28 beads activate T cells via the TCR complex. Additional stimulation conditions included no stimulation (Thaw Media only) and lipopolysaccharide (LPS at 1 µg/mL). The dose and duration of the LPS stimulation were based on prior studies demonstrating strong human monocyte intracellular IL-8 staining after treatment with 1 µg/mL LPS for 4 h (25, 26).

For the PMA/Iono stimulations, cells from each subject were transferred into 96-well round-bottom plates, treated with 100 µL of the stimulation condition (PMA/Iono at 50/500 ng/mL, LPS at 1 µg/mL, or thaw media only), and incubated for 1 h at 37°C. After the first incubation, we added 1x Brefeldin A and Monensin (BioLegend) to each stimulation condition, and the cells were cultured for an additional 3 h at 37°C. We washed the cells with cell stain buffer (CSB; Fluidigm) and then barcoded the different stimulation conditions from each subject with a β-2 microglobulin (B2M, BioLegend) antibody conjugated to a unique cadmium metal (Maxpar MCP9 Antibody Labeling Kit, Fluidigm).

For anti-CD3/CD28 stimulations, LEAF purified anti-human CD3 (Biolegend, clone OKT3) antibodies were placed in a 48-well plates (200 µL at 1 µg/mL). Plates were stored overnight before staining at 4°C. On the day of staining, samples were thawed and incubated as aforementioned. Half of the cells from each subject were transferred into the wells with anti-CD3, and we added 100 µL of LEAF purified anti-human CD28 (BioLegend, clone 28.2) at 4 µg/mL, 200 µL of LPS at 1 µg/mL, and 20 µL at 1X of each Brefeldin and Monensin in 10% of FBS media. The other half of the cells were placed in a different well and resuspended with 10% of FBS media. Samples were incubated for 3 h at 37°C. We washed the cells with CSB and then barcoded the stimulated condition with CD45-111Cd and unstimulated condition with CD45-89Y (Fluidigm). Cells from each subject were combined for further processing.

We stained the cells for viability exclusion with a 100 µM cisplatin solution (Enzo Life Science) for 1 min at room temperature and quenched with CSB. We analyzed the following cell surface markers (clones) in unstimulated samples: CD45 (HI30), CD3 (UCHT1), CD4 (RPA-T4), CD8 (RPA-T8), CD19 (HIB19), human leukocyte antigen (HLA)-DR (L243), CD56 (NCAM16.2), PD-1 (EH12.2H7), CXCR3 (G025H7), CD33 (WM53), and PD-L1 (29E.2A3). Samples were washed with CSB, followed by fixation and permeabilization with 1X nuclear antigen staining buffer (Fluidigm) for 20 min at room temperature. The cells were washed twice with nuclear antigen staining perm (Fluidigm). We analyzed the following intracellular markers (clones) that were shared between the two stimulation conditions: IFN-γ (B27), TNF-α (Mab11), IL-8 (E8N1), and IL-6 (MQ2-13A5). Cells were washed, fixed in Maxpar Fix and Perm Buffer containing Intercalator-Ir at 125 nM (Fluidigm), and stored at 4°C. On the day of acquisition, cells were washed with CSB and ultrapure water, resuspended in ultrapure water containing EQ four element calibration beads (Fluidgim) and acquired by Helios cytometry time-of-flight (CyTOF) system running CyTOF Software (Fluidigm) at a rate of 300–500 events per second. FCS files were subsequently generated using CyTOF Software.

Covariable and Outcome Definitions

We abstracted clinical data from the electronic medical record into standardized case report forms. Acute physiology, age, chronic health evaluation (APACHE III) and sequential organ failure assessment (SOFA) severity of illness scores were calculated based on the original instruments (27, 28). ARDS was defined by the 2012 Berlin definition (29). Chest radiographs were evaluated for the presence of bilateral opacities consistent with ARDS by a board-certified radiologist blinded to clinical status. Ventilator-free days (VFDs) were defined as the total number of days alive and free of invasive mechanical ventilation in the 28 days following ICU admission (30–34). Patients who died before day 28 were considered to have VFDs = 0. We calculated the change in 8-point ordinal score from baseline (enrollment) to Day 7 as described in prior COVID-19 interventional trials (35). If a subject died before Day 7, an ordinal scale score of eight was carried forward to Day 7. If a subject was discharged before Day 7, their ordinal scale score at discharge was carried forward to Day 7. Primary admission diagnosis in SARS-CoV-2-negative subjects was adjudicated by review of the medical record by a critical care physician blinded to all other data at the time of hospital discharge.

Statistical Analysis

All molecular and cellular measurements were made by research staff blinded to the clinical data set. Patient characteristics were summarized using standard descriptive statistics. Plasma mediator concentrations were log₂ transformed to facilitate parametric statistical analyses. CyTOF data were preprocessed in FlowJo v10 (FlowJo LLC) by manually gating on intact/singlet/live/CD45⁺ cells. All CyTOF analyses after preprocessing were performed on inverse hyperbolic sine transformed data. We excluded samples that had less than 200 monocyte or T-cell events in each stimulation experiment. Univariable analyses on CyTOF data assessed the differences in the proportions of specific cell populations as percentages of total cell populations using Mann–Whitney U tests. We tested for associations between nasal swab SARS-CoV-2 PCR threshold cycle (Ct) values and mediator concentrations using multiple linear regression adjusting for age and sex.

Our primary analysis tested for associations between plasma mediator concentrations and SARS-CoV-2-status. We used multivariable regression with robust standard errors adjusting for age, sex, ethnicity, corticosteroid administration, tocilizumab administration, and APACHE III score to test for associations between mediator concentrations and SARS-CoV-2 status. The point estimates are expressed as adjusted fold-change in mediator concentration between the two groups. To control type I error from multiple hypothesis testing, we considered a Bonferroni corrected P value < 0.05 adjusted for 27 statistical tests (biomarkers) as statistically significant. We performed a sensitivity analysis comparing only SARS-CoV-2-negative subjects with a primary diagnosis of bacterial pneumonia, bacterial sepsis, or ARDS with all SARS-CoV-2-positive subjects. In secondary analyses, we stratified by SARS-CoV-2 status and tested for associations between immune mediators and clinical outcomes by adjusting for age, sex, corticosteroid administration, tocilizumab administration, and interval of days between SARS-CoV-2-positive test date and blood sample collection date. We considered a Bonferroni corrected P value < 0.05 as statistically significant in these analyses. We incorporated an interaction term for the product of SARS-CoV-2 infection status, and each plasma mediator we measured in the outcome analyses to identify which mediator-to-outcome relationships were modified by SARS-CoV-2 status. We did not correct for multiple hypothesis testing in the exploratory interaction term analysis. We performed multiple linear regression with robust standard error estimates to test for associations between mediators and bins of VFDs or change in ordinal scale score from baseline to Day 7. We estimated the relative risk (RR) of hospital mortality using Poisson regression with robust standard error estimates. All analyses were performed in R version 3.6.2.

RESULTS

Study Population

We obtained plasma samples and clinical data from patients (n = 326) within 24 h after admission to the ICU. Patient characteristics at the time of enrollment are shown in Table 1. A significantly higher proportion of the SARS-CoV-2-positive subjects were Hispanic/Latinx. Other baseline demographics were similar between the two groups. The median interval of time between initial SARS-CoV-2 testing and enrollment was 1 day in SARS-CoV-2-negative patients and 3 days in SARS-CoV-2-positive patients. SARS-CoV-2-negative patients were primarily admitted with non-COVID-19 pneumonia (27%), non-COVID-19 acute lung injury (11%), exacerbation of a chronic lung disease (15%), or cardiac dysfunction (18%). The median 8-point ordinal scale score was the same in both groups. Approximately, about 45% of subjects in both groups were supported with invasive mechanical ventilation at the time of enrollment, though the proportion of patients with ARDS was higher among SARS-CoV-2-positive patients. The mean APACHE III and SOFA scores were similar between the two groups. Clinical events and outcomes assessed 28 days after enrollment are shown in Supplemental Table S2. Hospital mortality was higher for SARS-CoV-2-positive (36%) compared with -negative (24%) subjects.

Table 1.

Subject characteristics

Characteristic	SARS-CoV-2-Negative (n = 122)	SARS-CoV-2-Positive (n = 204)	P Value*
Demographics and comorbidities
Age, yr (mean ± SD)	56 ± 17	55 ± 16	0.66
Female, n (%)	45 (37%)	68 (33%)	0.55
Race, n (%)
American Indian	5 (4%)	9 (4%)	0.99
Asian	7 (6%)	25 (12%)	0.08
Black/African American	22 (18%)	28 (14%)	0.34
Pacific Islander	1 (1%)	4 (2%)	0.65
White	79 (65%)	123 (60%)	0.48
Unknown	8 (7%)	15 (7%)	0.99
Ethnicity, n (%)
Hispanic/Latinx	11 (9%)	67 (33%)	<0.01
Baseline comorbidity, n (%)
Asthma	26 (21%)	33 (16%)	0.30
COPD	33 (27%)	17 (8%)	<0.01
CKD	31 (25%)	41 (20%)	0.27
Cirrhosis	12 (10%)	13 (6%)	0.29
CAD	23 (19%)	21 (10%)	0.04
DM	34 (28%)	65 (32%)	0.90
HTN	66 (54%)	106 (52%)	0.72
Characteristics upon ICU admission
Admission information
Hospital transfer, n (%)	35 (29%)	102 (50%)	<0.01
COVID-test to enrollment, median days (IQR)	1 (0–1)	3 (1–9)	<0.01
Primary diagnosis†
Non-COVID pneumonia (viral or bacterial)	33 (27%)	NA	NA
Non-COVID ALI (ARDS, aspiration, and contusion)	13 (11%)	NA	NA
Exacerbation of chronic lung disease (asthma, COPD, bronchiectasis, and ILD)	18 (15%)	NA	NA
Cardiac dysfunction (CHF, MI, arrhythmia, and cardiac arrest)	22 (18%)	NA	NA
Nonpulmonary sepsis (NSTI, bacteremia, cholangitis, and peritonitis)	16 (13%)	NA	NA
Other	20 (16%)	NA	NA
8-Point ordinal score, n (%)
4 (hospitalization)	19 (16%)	28 (14%)	0.74
5 (any supplemental oxygen)	29 (24%)	35 (17%)	0.15
6 (HFNC or NIPPV)	18 (15%)	51 (25%)	0.04
7 (Invasive MV or ECMO)	56 (46%)	90 (44%)	0.82
Median 8-point ordinal score (IQR)	6 (5–7)	6 (5–7)	0.64
ARDS, n (%)	26 (23%)	80 (39%)	<0.01
P/F ratio – median (IQR)	134 (93–184)	90 (69–133)	<0.01
RALE score, means ± SD	23 ± 13	26 ± 11	0.36
SOFA, means ± SD	7.8 ± 4.5	7.3 ± 4.6	0.34
Respiratory SOFA, means ± SD	2.0 ± 1.6	2.3 ± 1.7	0.09
Cardiovascular SOFA, means ± SD	2.1 ± 1.6	1.8 ± 1.7	0.20
APACHE III, means ± SD	78 ± 29	73 ± 30	0.09
Treatments at enrollment
Corticosteroids, n (%)	41 (34%)	136 (67%)	<0.01
Tocilizumab, n (%)	0 (0%)	7 (3%)	0.05
Remdesivir, n (%)	0 (0%)	77 (38%)	<0.01
IV antibiotics, n (%)	95 (78%)	104 (51%)	<0.01
Vasopressors, n (%)	49 (40%)	84 (41%)	0.91

ALI, acute lung injury; APACHE III, acute physiology, age, chronic health evaluation; ARDS, acute respiratory distress syndrome; CAD, coronary artery disease; CHF, congestive heart failure; CKD, chronic kidney disease; COPD, chronic obstructive pulmonary disease; COVID, coronavirus disease; DM, diabetes mellitus; ECMO, extracorporeal membrane oxygenation; HFNC, high-flow nasal cannula; HTN, hypertension; ILD, interstitial lung disease; IQR, interquartile range; MI, myocardial infarction; MV, mechanical ventilation; NIPPV, noninvasive positive pressure ventilation; NSTI, necrotizing soft tissue infection; P/F, $\text{P}{\text{a}}_{\text{O}}{}_{{}_2}$/$\mathrm{\text{F}}{\mathrm{\text{I}}}_{{\mathrm{\text{O}}}_2}$ ratio; RALE, radiographic assessment of lung edema score (75); SARS-CoV-2, severe acute respiratory syndrome coronavirus-2; SD, standard deviation; SOFA, sequential organ failure assessment. *Statistical comparisons were done by a t test, Wilcoxon rank-sum, or Fisher’s test as appropriate. †Adjudicated by review of the medical record by a critical care physician blinded to all other data.

IFN-Induced Plasma Mediators Distinguish Subjects with SARS-CoV-2

We compared 27 plasma immune mediators between subjects with or without SARS-CoV-2 adjusting for age, sex, ethnicity, corticosteroid administration, tocilizumab administration, and APACHE III score to identify important similarities and differences in immune signatures between well-matched ICU patients with or without COVID-19. Plasma levels of C-X-C motif chemokine ligand 10 (CXCL10), soluble programmed death-ligand 1 (sPD-L1), interferon-γ (IFN-γ), C-C motif chemokine ligand 26 (CCL26), C-reactive protein (CRP), and tumor necrosis factor-α (TNF-α) were markedly higher in SARS-CoV-2-positive compared with -negative patients (Bonferroni adjusted P values < 0.05) (Fig. 1A and Supplemental Table S3). The plasma levels of CXCL10 and sPD-L1 remained significantly higher in SARS-CoV-2-positive subjects versus SARS-CoV-2-negative subjects when we restricted our comparison to subjects with a diagnosis of either pneumonia, sepsis, or ARDS (Fig. 1B and Supplemental Table S4) or only subjects with ARDS (Supplemental Table S5). Higher plasma concentrations of IFN-γ and CXCL10 were associated with higher SARS-CoV-2 viral burden as measured by lower nasal Ct values (Supplemental Fig. S1). There was a trend between higher nasal swab SARS-CoV-2 viral burden and higher plasma sPD-L1 and TNF-α levels (Supplemental Fig. S1).

Figure 1.

Interferon-induced mediators distinguish subjects with SARS-CoV-2. Volcano plots display differentially expressed proteins between SARS-CoV-2-positive and -negative subjects adjusted for age, sex, ethnicity, APACHE III score, tocilizumab, and steroid administration. Dashed line indicates Bonferroni-corrected P value = 0.05. Dotted line indicates a nominal P value = 0.05. A: proteins measured from plasma in all subjects (SARS-CoV-2-positive: n = 204; SARS-CoV-2-negative: n = 122). B: proteins measured from plasma in SARS-CoV-2-positive subjects (n = 204) vs. SARS-CoV-2-negative subjects with a primary diagnosis of pneumonia, sepsis, or ARDS (n = 62). APACHE III, acute physiology, age, chronic health evaluation; ARDS, acute respiratory distress syndrome; SARS-CoV-2, severe acute respiratory syndrome coronavirus-2.

In contrast, the plasma levels of key immune mediators previously associated with ARDS and sepsis such as interleukin-6 (IL-6), interleukin-8 (IL-8), C-C motif chemokine ligand 2 (CCL2), interleukin-10 (IL-10) were not significantly different between patients with and without SARS-CoV-2 (Fig. 1 and Supplemental Tables S3, S4, and S5). The plasma levels of IL-6 in both SARS-CoV-2-negative and -positive subjects were much higher than those recently reported in healthy individuals (36). There was no relationship between CCL26, CRP, or IL-6 and SARS-CoV-2 Ct viral burden (Supplemental Fig. S1). Taken together, our analysis strongly suggests the plasma immune profile in critically ill patients with SARS-CoV-2 is specifically characterized by interferon-induced immune mediators such as CXCL10 and sPD-L1.

Plasma sPD-L1 is a Novel Biomarker for COVID-19 Severity

We further explored differences in immune responses of critically ill subjects with or without SARS-CoV-2 by testing for associations between plasma immune mediators and clinical outcomes in multivariable analyses. Higher plasma levels of sPD-L1 and TNF-α were associated with fewer VFDs in SARS-CoV-2-positive subjects (Bonferroni adjusted P values < 0.05; Fig. 2 and Supplemental Table S6). Higher plasma levels of sPD-L1 and TNF-α were also associated with a worsening of ordinal scale score from enrollment to day 7 (Supplemental Table S7) and hospital mortality (Supplemental Table S8). In contrast, neither of these mediators were associated with VFDs in SARS-CoV-2-negative subjects (Fig. 2 and Supplemental Table S6), even at a nominal P value threshold of 0.05. Key immune mediators previously associated with COVID-19 severity such as IL-6, IL-10, and CCL2 were nominally associated with fewer VFDs in SARS-CoV-2-positive patients (P < 0.05). The relationship between severity of the respiratory failure and lymphocyte chemoattractants, such as CXCL10 and C-C motif chemokine ligand 17 (CCL17) (TARC), was significantly different (nominal P value for interaction < 0.05) in subjects with and without SARS-CoV-2 (Fig. 2 and Supplemental Table S6). For example, higher plasma levels of CCL17 were associated with fewer (worse) VFDs and worsening in ordinal score from enrollment to Day 7 in SARS-CoV-2-positive subjects. Conversely, higher plasma levels of CCL17 were associated with more (better) VFDs and improvement in ordinal score in SARS-CoV-2-negative subjects (Supplemental Tables S6 and S7). Other lymphocyte chemoattractants, such as macrophage inflammatory protein-1-α (MIP-1α) and macrophage inflammatory protein-3-α (MIP-3α), were not significantly associated with mortality in SARS-CoV-2-negative subjects, however, these chemokines were associated with mortality in SARS-CoV-2-positive subjects (Supplemental Table S8). These results suggest a distinct role for T-cell chemokines in COVID-19 immunopathology versus other critical illness syndromes such as bacterial sepsis and ARDS.

Figure 2.

Distinct associations between plasma immune mediators and ventilator-free days (VFDs) in SARS-CoV-2-positive and -negative subjects. Forrest plots display the β (95% CI) for tertiles of VFDs according to each plasma mediator adjusted for age, sex, corticosteroid administration, tocilizumab administration, and interval of days between SARS-CoV-2 test and enrollment. Mediators are ranked by P values for associations with VFDs in the SARS-CoV-2-positive subjects (point estimates for each mediator are provided in Supplemental Table S6). Dashed line (and red color) indicates Bonferroni-corrected P values < 0.05. Dotted line (and gold color) indicates nominal P values < 0.05. P values for interaction were calculated by incorporating an interaction term for the product of SARS-CoV-2 infection and log₂ plasma mediator level in the statistical model (regress “mean”: vfds ∼ log₂(mediator) X sars-cov-2-status + age + sex + covid-test-interval + corticosteroids + tocilizumab). CI, confidence interval; SARS-CoV-2, severe acute respiratory syndrome coronavirus-2.

SARS-CoV-2 is Characterized by Peripheral Blood Mononuclear Cell Hyporesponsiveness

We analyzed PBMCs collected from a subset of patients with the most severe form of acute hypoxemic respiratory failure, ARDS (n = 48), to determine whether immune cell activation or suppression was distinct between well-matched ARDS patients with or without COVID-19. The characteristics of this subset of patients are shown in Supplemental Table S9. The median interval of time between SARS-CoV-2 testing and PBMC sampling was 3.5 days longer in SARS-CoV-2-positive compared with -negative subjects. All other clinical characteristics, including illness severity and degree of hypoxemia [arterial partial pressure of oxygen ($\text{P}{\text{a}}_{\text{O}}{}_{{}_2}$)/$\mathrm{F}{\mathrm{I}}_{{\mathrm{O}}_2}$ ratio] were similar between SARS-CoV-2-positive and -negative patients. Biaxial gating of cell populations is shown in Supplemental Fig. S2.

We assessed immune cell responsiveness in CD33⁺ monocytes, CD4⁺ T cells, and CD8⁺ T cells in vitro by measuring intracellular cytokine staining after 4 h of stimulation with either LPS, PMA/ionomycin, or anti-CD3/CD28 beads. The proportion of CD33⁺IL-6⁺ and CD33⁺TNF-α⁺ peripheral blood monocytes in response to LPS stimulation was significantly lower in SARS-CoV-2-positive compared with SARS-CoV-2-negative subjects (Fig. 3A). The percentage of CD4⁺TNF-α⁺ T cells collected from SARS-CoV-2-positive subjects was significantly lower in response to both PMA/ionomycin as well as anti-CD3/CD28 beads (Fig. 3B). There was a trend toward lower percentages of CD4⁺IFN-γ⁺ T-cell staining in response to either PMA/ionomycin or anti-CD3/CD28. The percentage of intracellular CD4⁺IFN-γ and CD4⁺ TNF-α T cells in response to any T-cell stimulation (either PMA/Ionomycin or anti-CD3/CD28) was lower in SARS-CoV-2-positive versus -negative patients (Fig. 3B). The percentage of CD8⁺TNF-α⁺ cytotoxic T cells from SARS-CoV-2-positive subjects was lower in response to stimulation compared with SARS-CoV-2-negative patients, but there was no difference in intracellular IFN-γ staining (Fig. 3B).

Figure 3.

SARS-CoV-2 is characterized by circulating immune cell hyporesponsiveness. The panels show individual percentages (dots); medians (thick line); and interquartile ranges (error bars). Statistical analyses were performed with Mann–Whitney tests. Biaxial gating of cell populations is shown in Supplemental Fig. S2. A: percentage of monocytes positively staining for IL-8, IL-6, or TNF-α as a proportion of total CD33⁺ monocyte events in response to LPS stimulation. B: percentage of cells positively staining for either IFN-γ or TNF-α as a proportion of all CD3⁺CD4⁺ or CD3⁺CD8⁺ events in response to PMA/ionomycin or anti-CD3/CD28 stimulation. The combined data set includes values from either stimulation. C: percentage of CXCR3⁺ (ligand: CXCL10) cells as a proportion of total CD4⁺ or CD8⁺ T cells, respectively. Percentage of PD-L1⁺ or PD-1⁺ as a proportion of total CD33⁺ monocytes or CD8⁺ T cells, respectively. LPS, Lipopolysaccharide; SARS-CoV-2, severe acute respiratory syndrome coronavirus-2.

We next identified cell surface phenotypes related to the chemokine receptor CXCR3 (receptor for CXCL10) and PD-L1/PD-1 given our findings that plasma CXCL10 and sPD-L1 were strongly associated with SARS-CoV-2-status. We found that the proportion of CD4⁺CXCR3⁺ T cells was significantly lower in SARS-CoV-2-positive versus -negative subjects (Fig. 3C). The proportion of CD8⁺CXCR3⁺ was numerically lower in SARS-CoV-2-positive versus -negative subjects, however, this difference was not statistically significant. There were no associations between the percentages of CD33⁺PD-L1⁺ monocytes or CD8⁺PD-1⁺ T cells and SARS-CoV-2-status. A higher percentage of CD4⁺PD-1⁺ T cells was associated with worse VFDs among patients with ARDS (Supplemental Fig. S3). Previous landmark studies of critically ill patients with bacterial sepsis and severe trauma have identified a key role for immune cell hyporesponsiveness (17–19). Our findings suggest that subjects with severe COVID-19 ARDS have impaired immune effector cell function even when compared with similarly ill SARS-CoV-2-negative ARDS subjects.

Correlations between Immune Cell Phenotypes and Circulating Proteins Are Distinct in SARS-CoV-2

We performed exploratory analyses comparing the correlations between plasma mediators and immune cell phenotypes in SARS-CoV-2-negative and -positive patients to better understand whether typical relationships observed in sepsis and ARDS between cellular phenotypes and systemic inflammation are altered in severe COVID-19. Prior in vitro studies have demonstrated that CXCR3 is strongly downregulated in response to high levels of CXCL10 (37). We observed that CD4⁺CXCR3⁺ cells were negatively correlated with plasma CXCL10 levels in SARS-CoV-2-negative subjects (Fig. 4, top). In contrast, CD4⁺CXCR3⁺ cells were positively correlated with plasma CXCL10 levels in SARS-CoV-2-positive patients, suggesting normal downregulation of CXCR3 in response to high levels of CXCL10 may be dysregulated in COVID-19.

Figure 4.

Correlations between immune cell subsets and soluble mediators in SARS-CoV-2-negative and -positive subjects. Matrices displaying the correlation coefficients (r) between immune cell subsets (Y-axis) and plasma mediator levels (X-axes). Colors represent the correlation with scale indicating value of Pearson’s r correlation.

The relationship between immune cell responsiveness and plasma mediator levels was also highly distinct between SARS-CoV-2-negative and -positive subjects. Expression of PD-L1/PD-1 on monocytes and T cells was positively correlated with higher plasma inflammatory mediator levels in SARS-CoV-2-negative subjects (Fig. 4, middle), which is consistent with the well-described upregulation of checkpoint proteins to dampen systemic inflammation (38). In contrast, the positive correlation between checkpoint expressing immune cells and higher levels of plasma inflammatory proteins was lost in SARS-CoV-2-positive subjects. The inverse relationship between CD8⁺ T-cell intracellular cytokine staining and plasma inflammatory mediator concentrations seen in SARS-CoV-2-negative patients was also lost in SARS-CoV-2-positive subjects (Fig. 4, bottom). Taken together, we have identified divergent correlations between CXCR3 and its ligand CXCL10 as well as checkpoint proteins and systemic inflammation. These findings point toward a potential disconnect between high systemic levels of inflammation and appropriate mechanisms to downregulate lymphocyte migration/homing and inflammation in subjects with severe COVID-19.

DISCUSSION

We comprehensively analyzed the molecular and cellular signatures of critically ill patients with SARS-CoV-2 and compared these signatures with those from contemporaneously enrolled patients with non-SARS-CoV-2 critical illness. This unique study design allowed us to directly address three key questions about severe COVID-19 immunopathology: 1) are the systemic inflammatory profiles in severe COVID-19 and their relationship with clinical outcomes distinct from other forms of critical illness; 2) are immune cells activated or hyporesponsive in severe COVID-19 compared with other forms of critical illness; and 3) are the relationships between immune cell phenotypes (e.g., chemokine receptors) and plasma mediators (e.g., chemokines) distinct in severe COVID-19 compared with other forms of critical illness? Our study builds upon prior studies in COVID-19 that used historical controls and less well-characterized patient cohorts (39) and supports a model whereby COVID-19 and associated severity are marked by high circulating levels of sPD-L1, high circulating levels of lymphocyte chemoattractants such as CXCL10, and mononuclear cell hyporesponsiveness.

Our first question was whether systemic inflammatory profiles in severe COVID-19 are distinct from other forms of critical illness such as sepsis or ARDS. To answer this question, we enrolled a contemporaneous control group of critically ill SARS-CoV-2-negative subjects with the same inclusion/exclusion criteria as SARS-CoV-2-positive subjects and compared their respective plasma inflammatory biomarker levels. We found that plasma CXCL10 and sPD-L1 levels are significantly higher in critically ill SARS-CoV-2-positive versus -negative subjects, confirming a key role for interferon-related mediators in severe COVID-19. In contrast, plasma IL-6, IL-8, IL-10, and IL-17A levels were not significantly different between critically ill subjects with or without SARS-CoV-2. The lack of a significant difference in these mediators between SARS-CoV-2-positive and -negative subjects persisted even when we limited our analysis to only subjects with sepsis, pneumonia, or ARDS (Supplemental Tables S4 and S5). Our finding that levels of IL-6 and IL-8 are not significantly different between ICU patients with and without COVID-19 diverge from some [but not all (40, 41)] studies that found the levels of these proinflammatory mediators are markedly lower in patients with COVID-19 compared with historical levels from non-COVID-19 ARDS or sepsis patients (6–9). Comparing the plasma biomarker levels between cohorts of subjects with COVID-19 versus historical cohorts of patients with sepsis and ARDS enrolled before the pandemic is very challenging because there have been secular changes in clinical practice over the past two decades (42, 43), there have been significant differences in care provided in the setting of a stressed healthcare system during the pandemic (44–46), and the performance characteristics of molecular measurements across different platforms is highly variable (47). Indeed, plasma inflammatory biomarker levels in patients with ARDS had already been decreasing in the 20 years just before the COVID-19 pandemic (48). These factors all highlight the importance of our report, which clarifies which mediators are similar and different between contemporaneously enrolled critically ill subjects with and without COVID-19.

Our plasma biomarker findings also significantly expand upon prior COVID-19 studies by identifying distinct associations between immune mediators and respiratory-related clinical outcomes in SARS-CoV-2-positive compared with SARS-CoV-2-negative subjects (49–60). We found that elevated interferon-induced chemoattractants are differentially associated with more severe respiratory failure based on SARS-CoV-2 status (Fig. 2 and Supplemental Table S6). For instance, higher plasma levels of CCL17 were nominally associated with fewer VFDs in SARS-CoV-2-positive subjects, yet were nominally associated with more VFDs in SARS-CoV-2-negative subjects. These findings are consistent with multiple genome-wide association studies that have identified a COVID-19 respiratory failure susceptibility locus in the chromosome 3p21.31 region, which includes numerous chemokine receptor genes (61–63).

One of our most novel findings was the strong association between higher plasma sPD-L1 levels and more respiratory failure as measured by fewer VFDs, progression in ordinal scale score, and mortality. These findings build upon smaller cross-sectional analyses, suggesting plasma sPD-L1 levels are higher in subjects with more severe disease at the time of hospital arrival and suggest a key role for sPD-L1 in COVID-19 disease progression (64, 65). Notably, these findings from the plasma are somewhat discordant with our previous work that found higher PD-L1 gene expression in alveolar macrophages isolated from subjects with ARDS are associated with more VFDs (66). We have previously identified significant differences between lung and blood mononuclear cell gene expression patterns in subjects with ARDS (67), and suspect compartment-specific differences in immune responses may account for these discordant findings. Further work analyzing biospecimens collected from the lower respiratory tract in patients with SARS-CoV-2 is necessary to clarify the role of PD-L1 in COVID-19 immunopathology (39).

Our second question was whether monocyte endotoxin tolerance and T-cell exhaustion are distinct features of severe COVID-19. We found that SARS-CoV-2-positive subjects with ARDS exhibited features of both monocyte endotoxin tolerance and T-cell exhaustion compared with SARS-CoV-2-negative critically ill subjects with ARDS. Monocyte endotoxin tolerance and T-cell exhaustion are well-established responses that occur in critically ill patients (17–22). Thus, the overall severity of illness can highly confound associations between immune cell hypofunction and COVID-19 (68). Our study design allowed for a direct comparison between similarly ill COVID-19 and non-COVID-19 patients. We found that monocytes and T cells collected from subjects with severe COVID-19 are hypofunctional when compared with patients with severe non-COVID ARDS, suggesting that immune cell tolerance and exhaustion are distinctive features of severe COVID-19 beyond generalized critical illness.

We integrated our plasma biomarker and immune cell phenotype data sets to determine whether the relationships between immune cells and circulating mediators are distinct in severe COVID-19 compared with other forms of critical illness. In SARS-CoV-2-negative critically ill subjects, we found the chemokine receptor CXCR3 was negatively correlated with CXCL10 (CXCR3 ligand), PD-L1/PD-1 was positively correlated with high systemic inflammatory mediator levels, and cellular responsiveness was inversely correlated with high inflammatory mediator concentrations (Fig. 4). These findings are consistent with prior in vitro and translational studies performed in patients with non-COVID sepsis and ARDS (17, 18). In contrast, these correlations were lost in critically ill subjects with SARS-CoV-2 (Fig. 4), suggesting that derangements in negative regulatory processes may play a unique role in COVID-19 pathogenesis. Our finding that CD8⁺ T-cell intracellular cytokine responses are not correlated with plasma CXCL10 or IL-6 levels in patients with SARS-CoV-2 are consistent with prior publications implicating CD8⁺ T-cell phenotype and function as playing a crucial role in controlling the initial acute SARS-CoV-2 infection (69). Previous studies have also identified associations between CXCR3⁺ and PD-L1⁺/PD-1⁺ cell subsets in patients with COVID-19 compared with healthy controls (70–73). Our study has clarified the role these immune cell subsets play in COVID-19 compared with other conditions such as ARDS and suggest that altered T-cell trafficking/migration and dysregulated tolerance programs play a key role in severe COVID-19 pathogenesis.

Our study’s main strength is its use of a well-characterized and large clinical cohort of contemporaneously enrolled critically ill patients with and without SARS-CoV-2. This study design allowed us to identify molecular and cellular immune signatures that are specific to COVID-19. Despite these strengths, there are some important limitations. The specimens in our study were collected within a narrow time window indexed to when a patient presented to the ICU. Although this well-defined sampling period captured subjects when their symptoms were becoming most severe and organ failures most pronounced, it is possible the similarities and differences we observed in immune mediator levels between SARS-CoV-2-positive and -negative patients may have been influenced by a longer latency between SARS-CoV-2 infection and onset of critical illness compared with non-COVID-19 causes of critical illness (74). Our contemporaneous control group did not include significant numbers of patients with non-SARS-CoV-2 respiratory viruses such as influenza. It is possible the immune signatures we have identified are not specific to SARS-CoV-2, but rather to respiratory viruses more generally. Nonetheless, our findings have refined our understanding of the host response to COVID-19 among a broad population of critically ill patients. Our study’s sample size was determined by the number of subjects we enrolled into our cohort during a 14-mo period and was not based on a prospective power calculation. Therefore, our analysis of subjects with ARDS may have been underpowered to detect statistically significant differences in mediators such as IL-6, IL-10, and IL-17A between subjects with and without SARS-CoV-2 (Supplemental Table S5). Nevertheless, our sample size should have been sufficient to detect previously reported differences in these mediators between ARDS due to COVID-19 versus historical controls with ARDS before the COVID-19 pandemic (6–9). Finally, our findings should be validated in separate prospective clinical cohorts.

Overall, our findings support an immunologic model whereby COVID-19 and associated severity is marked by high circulating levels of interferon-stimulated mediators (sPD-L1 and CXCL10), immune cell hyporesponsiveness, and dysfunctional antiviral adaptive immune responses such as lymphocyte homing and tolerance. We highlight multiple mediators whose associations with clinical outcomes are equivalent across critically ill subjects with and without SARS-CoV-2, suggesting strategies for targeting pathways such as IL-6 that have shown benefit in severe COVID-19 may also be efficacious in subpopulations of non-COVID critically ill patients whose IL-6 levels are elevated. Our identification of an association between sPD-L1 and clinical outcomes is novel. Therapies to augment immune cell function in certain subtypes of critically ill patients with COVID-19 may be beneficial. These findings highlight pathophysiologic factors that might be shared or different between critically ill patients with or without COVID-19 and may have implications for the design of therapeutic trials in critical illness due to SARS-CoV-2 as well as other forms of critical illness such as bacterial sepsis or ARDS.

DATA AVAILABILITY

The data that support the findings of this study will be made available upon reasonable request from the corresponding author.

SUPPLEMENTAL DATA

Supplemental Tables S1–S9 and Supplemental Figs. S1–S3: https://doi.org/10.6084/m9.figshare.19699990.v1.

GRANTS

This work was funded by the following grants: National Institutes of Health-National Heart, Lung, and Blood Institute (NHLBI) K23HL144916 (to E.D.M.), National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) K23DK116967 (to P.K.B.), NHLBI F32HL 158088 (to N.A.S.), National Institute of Allergy and Infectious Diseases (NIAID) K08AI119142 (to S.L.F.), NHLBI R03HL141523 (to C.M.), and NHLBI R01HL149676 (to C.M.); Bill and Melinda Gates Foundation (to P.K.B. and M.M.W.); The CDC Foundation (to P.K.B., L.E., and M.M.W.); and the Firland Foundation (to T.E.W.).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

E.D.M., P.K.B., M.M.W. and C.M. conceived and designed research; E.D.M., P.K.B., N.A.S., J.L., L.M., S.E.H., S.R.P., A.W., C.A.-V., M.A.M., T.L., X.-Y.C., S.S., C.B., M.O., S.S.S., A.S., D.M.L., N.K., A.G., E.B., G.C., M.K.B., S.P., S.L.F., L.E., S.A.L., and T.E.W. performed experiments; E.D.M., P.K.B., N.A.S., L.M., S.E.H., S.R.P., A.W., C.A.-V., M.A.M., T.L., X.-Y.C., S.S., C.B., M.O., S.S.S., A.S., D.M.L., N.K., A.G., E.B., G.C., M.K.B., S.P., S.L.F., L.E., S.A.L., T.E.W., M.M.W., and C.M. analyzed data; E.D.M., P.K.B., N.A.S., J.L., L.M., S.E.H., S.R.P., A.W., C.A.-V., M.A.M., T.L., X.-Y.C., S.S., C.B., M.O., S.S.S., A.S., D.M.L., N.K., A.G., E.B., G.C., M.K.B., S.P., S.L.F., L.E., S.A.L., T.E.W., M.M.W., and C.M interpreted results of experiments; E.D.M. and P.K.B. prepared figures; E.D.M., P.K.B., and C.M. drafted manuscript; E.D.M., P.K.B., M.M.W., and C.M. edited and revised manuscript; E.D.M., P.K.B., N.A.S., J.L., L.M., S.E.H., S.R.P., A.W., C.A.-V., M.A.M., T.L., X.-Y.C., S.S., C.B., M.O., S.S.S., A.S., D.M.L., N.K., A.G., E.B., G.C., M.K.B., S.P., S.L.F., L.E., S.A.L., T.E.W., M.M.W., and C.M. approved final version of manuscript.