Inflammatory profiles across the spectrum of disease reveal a distinct role for GM-CSF in severe COVID-19

Abstract

While it is now widely accepted that host inflammatory responses contribute to lung injury, the pathways that drive severity and distinguish coronavirus disease 2019 (COVID-19) from other viral lung diseases remain poorly characterized. We analyzed plasma samples from 471 hospitalized patients recruited through the prospective multicenter ISARIC4C study and 39 outpatients with mild disease, enabling extensive characterization of responses across a full spectrum of COVID-19 severity. Progressive elevation of levels of numerous inflammatory cytokines and chemokines (including IL-6, CXCL10, and GM-CSF) were associated with severity and accompanied by elevated markers of endothelial injury and thrombosis. Principal component and network analyses demonstrated central roles for IL-6 and GM-CSF in COVID-19 pathogenesis. Comparing these profiles to archived samples from patients with fatal influenza, IL-6 was equally elevated in both conditions whereas GM-CSF was prominent only in COVID-19. These findings further identify the key inflammatory, thrombotic, and vascular factors that characterize and distinguish severe and fatal COVID-19.

Fig. 1. Plasma mediators at the time of study enrollment demonstrated a broad exaggerated immune response in patients hospitalized with COVID-19.

Clustered heatmap of 33 immune mediators in plasma samples collected from patients hospitalized with COVID-19 at the time of study enrolment. Missing mediator data were imputed and values were scaled within each mediator. Rows and columns were split by K-means clustering. Each patients’ column is additionally annotated with data on disease outcome (“Severity”) as one of the following outcome groups: not requiring oxygen support (‘3′, n=132), requiring oxygen via face mask or nasal prongs (‘4’, n=106), requiring non-invasive ventilation or high-flow nasal canulae oxygen (‘5′, n=79), requiring invasive mechanical ventilation (‘6/7’, n=85) or fatal disease (‘8’, n=69). Columns are additionally annotated with patient age, sex and duration of illness at the time of sample collection (“Onset”).

Fig. 2. Antiviral, coagulation, and inflammation associated mediators distinguished severity groups early in disease.

Plasma samples from the time of study enrolment were analyzed for levels of the antiviral cytokines (A) IFN-α, (B) IFN-γ, and (C) the interferon-induced chemokine CXCL10 in healthy control (HC, n=15), patients with COVID-19 not requiring hospitalization (‘1/2’, n=39), and hospitalized patients with COVID-19 that would: not require oxygen support (‘3′, IFN-α n=32, other mediators n=132), require an oxygen face mask (‘4’, IFN-α n=23, other mediators n=106), require non-invasive ventilation or high-flow nasal cannulae (‘5′, IFN-α n=19, other mediators n=79), require invasive mechanical ventilation (‘6/7’, IFN-α n=19, other mediators n=85) or progress to fatal disease (‘8’, IFN-α n=14, other mediators n=69). Mediators associated with coagulation and endothelial injury were also quantified in these plasma samples; (D) D-dimer, (E) Angiopoietin-2, and (F) von-Willebrand factor A2 (vWF-A2). Similarly, mediators associated with inflammation were quantified: (G) IL-6; (H) GM-CSF; and (I) EN-RAGE/S100A12. Violin plots display medians (solid lines) and interquartile ranges (dashed lines). Data were analyzed for statistical significance using Kruskal-Wallis tests with Dunn’s tests for multiple comparisons between all groups. *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001.

Fig. 3

Plasma mediators in COVID-19 were coordinated around IL-6 and GM-CSF and influenced by age. (A) Correlogram of the association between plasma mediator levels at the time of enrolment in all patients hospitalized with COVID-19 (n=465). (B) Network analysis showing the mediator-to-mediator correlation profile. Nodes represent mediators and the coloring of edges between nodes represents the Spearman correlation coefficient (R_S) connecting them. (C) Inflammatory mediator levels within an outcome group, stratified as those ≥ or < than 70 years of age. Data in panel a were analyzed using Spearman’s rank correlations with correction for multiple testing; significant correlations are denoted by a circle, the color of which denotes the Spearman’s R value. Data in panel C were analyzed using Mann-Whitney U tests with P-value adjustment for false discovery rate. Violin plots display medians (solid lines) and interquartile ranges (dashed lines).

Fig. 4. Longitudinal analysis of plasma mediator levels demonstrated a progressive immune response and an exaggerated inflammatory signature in fatal COVID-19.

Plasma levels of (A) IFN-γ, (B) CXCL10, (C) Angiopoietin-2, (D) D-dimer, (E) GM-CSF, and (F) EN-RAGE/S100A12 over the course of disease in patients with fatal COVID-19. Plasma mediator levels of (G) IL-2, (H) IL-6, (I) GM-CSF, (J) D-dimer, and (K) von-Willebrand factor A2 (vWF-A2) within the first 4 days of symptom onset in patients in severity groups 6/7 or 8 (“Severe”, n=22) and groups 3, 4, or 5 (“Moderate”, n=54). Linear regressions with 95% confidence intervals are shown in panels A-F. Data in panels G-K were analyzed for statistical significance using Mann-Whitney U tests, where solid lines denote the median values and dashed lines denote the interquartile ranges.

Fig. 5

GM-CSF and IL-1α were elevated in fatal COVID-19 relative to influenza. (A) Median Z-scores for each mediator between healthy controls (HC, n=15) and patients with fatal influenza (n=20) or fatal COVID-19 (severity 8, n=69). (B) Levels of IL-6, GM-CSF, IL-1α, IL-1β, Thrombomodulin, and vWF-A2 in plasma samples from patients with fatal influenza or COVID-19. Data were analyzed for statistical significance using Mann-Whitney tests with between groups. ***P<0.001, ****P<0.0001.