Circulating Interleukin-8 Dynamics Parallels Disease Course and Is Linked to Clinical Outcomes in Severe COVID-19

Abstract

Severe COVID-19 frequently features a systemic deluge of cytokines. Circulating cytokines that can stratify risks are useful for more effective triage and management. Here, we ran a machine-learning algorithm on a dataset of 36 plasma cytokines in a cohort of severe COVID-19 to identify cytokine/s useful for describing the dynamic clinical state in multiple regression analysis. We performed RNA-sequencing of circulating blood cells collected at different time-points. From a Bayesian Information Criterion analysis, a combination of interleukin-8 (IL-8), Eotaxin, and Interferon-γ (IFNγ) was found to be significantly linked to blood oxygenation over seven days. Individually testing the cytokines in receiver operator characteristics analyses identified IL-8 as a strong stratifier for clinical outcomes. Circulating IL-8 dynamics paralleled disease course. We also revealed key transitions in immune transcriptome in patients stratified for circulating IL-8 at three time-points. The study identifies plasma IL-8 as a key pathogenic cytokine linking systemic hyper-inflammation to the clinical outcomes in COVID-19.

Keywords: COVID-19; IL-8; biomarker; cytokines; machine learning; survival.

Figure 1

Circulating cytokines and blood oxygenation in severe COVID-19 patients. (A) Experimental schema for the study. (B) Correlation (Pearson) clustering of plasma abundance of cytokines shown in order of increasing blood oxygenation (represented by area under curve of the SpO2/FiO2 ratio kinetics over 7 days following recruitment.

Figure 2

Machine learning to derive a minimal predictive model to describe blood oxygenation using plasma cytokine abundance in severe COVID-19 patients. (A) Bayesian information criterion applied to multiple regression model to derive SFR_7dAUC based on plasma concentration of cytokines at time-point 1 (day of recruitment). The minimization of BIC value and evolution of R2 values are shown in the right panel with elimination of cytokines to derive a minimally effective model. Red * indicates the cytokines derived from the analysis. (B) Correlation of the three cytokine BIC-Y index with actual SFR_7dAUC values in severe COVID-19 patients. Pearson’s R-value is shown, *** p < 0.0005. (C) Receiver operator characteristic curve (black line) of BIC-Y values to derive remission versus non-remission (death); the marked point denotes the specificity and sensitivity with the derived cut-off. (D) Survival of patients until day 30 post-enrolment is compared in a Kaplan–Meier curve between patients with BIC-Y values above (black line) or below (red line) the cut-off derived from the ROC curve shown in (C). Surviving patients were censored on day 30 post-enrolment. (E) Hospital stay duration (time to remission) of the patients from both groups (BIC-Y values above or below cut-off) since the day of enrolment are plotted in an ascending Kaplan–Meier curve. Deaths and non-remission at day 30 post-enrolment were censored. For the outcome comparisons shown in (D,E), the Mantel–Cox log-rank test was performed. The number of patients at risk on different days and the Mantel–Haenszel hazard ratio is shown.

Figure 3

Plasma level of IL-8 and clinical outcomes in severe COVID-19 patients. (A) Receiver operator characteristic curve (black line) of plasma levels of IL-8 to derive remission versus non-remission (death), the marked point denotes the specificity and sensitivity with the derived cut-off (red circle). The red dotted line denotes the diagonal. (B) Survival of patients till day 30 post-enrolment is compared in a Kaplan–Meier curve between patients with plasma level of IL-8 below (IL8_lo, black line) or above (IL8_hi, red line) the cut-off derived from the ROC curve shown in (A). Surviving patients were censored on day 30 post-enrolment. (C) Comparison of SpO2/FiO2 ratio kinetics over 7 days following plasma sampling between IL8_lo (black line) and IL8_hi (red line) subgroup of patients. * p < 0.05, ** p < 0.005, *** p < 0.0005 from unpaired t-tests. (D) Hospital stay duration (time to remission) of the patients from both groups, IL8_lo (black line) and IL8_hi (red line), since the day of enrolment is plotted in an ascending Kaplan–Meier curve. Deaths and non-remission at day 30 post-enrolment were censored. For the outcome comparisons shown in (B,D) Mantel–Cox log-rank test was performed. Number of patients at risk on different days and the Mantel–Haenszel hazard ratio is shown. (E) Mantel–Haenszel hazard ratio for Kaplan–Meier analysis for survival in subgroups of patients, for IL8_lo patients, compared with IL8_hi patients. Analyses showing Mantel–Cox log-rank test p-values <0.05 are indicated in green.

Figure 4

Re-analyses of single-cell RNA sequencing data to show expression of IL-8 and its receptor among all cells. (A–D) Abundance of IL-8 transcript (CXCL8) among different subsets of cells isolated from peripheral blood (A,B) and bronchoalveolar lavage fluid (C,D), compared between mild (A,C) and severe (B,D) COVID-19 patients. (E–H) Abundance of transcript for IL-8 receptor (CXCR1) among different subsets of cells isolated from peripheral blood (E,F) and bronchoalveolar lavage fluid (G,H), compared between mild (E,G) and severe (F,H) COVID-19 patients. Data analyzed from public datasets GSE163668 (blood cells) and GSE145926 (cells from bronchoalveolar lavage).

Figure 5

Re-analyses of single-cell RNA sequencing data to show individual immune cells expressing IL-8 and its receptor in severe COVID-19 patients. (A) Abundance of IL-8 transcript (CXCL8) among different cell subsets from peripheral blood of severe COVID-19 patients, defined by expression of characteristic transcripts. (B) Abundance of the transcript for IL-8 receptor (CXCR1) among different cell subsets from peripheral blood of severe COVID-19 patients, defined by expression of characteristic transcripts. Data analyzed from public dataset GSE163668. (C) Abundance of IL-8 transcript (CXCL8) among different cell subsets from bronchoalveolar lavage of severe COVID-19 patients, defined by expression of characteristic transcripts. (D) Abundance of the transcript for IL-8 receptor (CXCR1) among different cell subsets from bronchoalveolar lavage of severe COVID-19 patients, defined by expression of characteristic transcripts. Data analyzed from public dataset GSE145926.

Figure 6

Analyses of immunocellular transcriptome from peripheral blood mononuclear cells from severe COVID-19 patients. (A–C) Differentially expressed genes (more than 2-fold change) shown in a volcano plot, analyzed from RNA sequencing data generated from peripheral blood mononuclear cells, compared between patients falling into the IL8_hi group at time-point T1 (N = 9), designated T1_hi, and patients falling into the IL8_lo group at time-point T1 (N = 8), designated T1_lo (A), between patients falling into the IL8_hi group at time-point T2 (N = 7), designated T2_hi, and patients falling into the IL8_lo group at time-point T2 (N = 6), designated T2_lo, (B) and, finally, between patients who were classified as IL8_hi at time-point T1 (N = 5), designated T1_hi, and ended up falling into the IL8_lo group at time-point T3, designated T3_lo (C). (D–F) Network depictions of major pathway groups enriched by genes upregulated in T1_hi compared to T1_lo (D), upregulated in T2_hi compared to T2_lo (E), and upregulated in T3_lo compared to T1_hi (F). (G–I) Network depictions of major pathway families enriched by genes downregulated in T1_hi compared to T1_lo (G), downregulated in T2_hi compared to T2_lo (H), and downregulated in T3_lo compared to T1_hi (I).

Figure 7

Plasma IL-8 dynamics parallel disease course and stratify final outcomes in severe COVID-19. (A) Sankey chart showing subgroups of severe COVID-19 patients variably falling into either IL8_hi or IL8_lo groups at three different time-points, stratified in terms of final outcomes. (B) Plot shows data from receiver operator characteristics analyses for plasma IL-8 levels at three time-points in terms of final fatal outcomes, showing the respective cut-off values (black line, right Y axis represented as percent) as well as sensitivity (blue line) and specificity (green line) for those cut-off values (left Y axis represented as plasma concentration of IL-8).