Abstract
Severe cases of COVID-19 are characterized by excessive inflammation. Here we report on an inpatient cohort where plasma cytokines were measured and tested for association with future need for mechanical ventilation. Hierarchical clustering, Kaplan-Meier curves, and odds ratios demonstrated that the cluster of IL-13 (OR: 1.57), IL-7 (OR: 1.04) and bFGF (OR: 1.04) was predictive for intubation, independent of age, gender and comorbidities.
Introduction
SARS-CoV-2 is a human coronavirus currently causing the global pandemic COVID-19. As of June 14th, there were more than 7.6 million cases worldwide and 400,000 fatalities [1]. Complications of COVID-19, the disease caused by SARS-CoV-2, include pneumonia and acute respiratory distress syndrome. Severe disease is accompanied by excessive production of cytokines, commonly referred to as cytokine storm, and may ultimately result in significant lung damage. Cytokine storm, or cytokine release syndrome (CRS) is often characterized by high levels of IL-6, TNFα, IL-1β, and IL-10, which have all been observed to be elevated in patients with COVID-19 [2,3]. Complications due to cytokine storm can include shock and multiple organ failure and can be life-threatening. Additionally, patients with comorbidities are more likely to have worse disease outcomes [4]. For example, chronically elevated proinflammatory cytokines in diabetics may be a cause of more severe disease due to COVID-19 [5]. The presence of CRS in severe COVID-19 highlights the nature of immune dysregulation in SARS-CoV-2 infection, and suggests an immunopathological response that may be targetable by therapeutics However, studies from around the world do not all observe the same immune profiles within their cohorts, and so it may be important to understand the variability of patient responses, and consider whether personalized approaches may need to be taken.
Methods
57 patients who tested positive for COVID-19 were found using the University of Virginia Medical Center’s electronic database. Any remnant EDTA-plasma samples identified by the clinical laboratory within 48 hours of the time of diagnosis or hospitalization were collected. Blood was centrifuged at 1300 × g for 10 minutes, then plasma was aliquoted and stored at −80°C until testing. Demographics (age, gender, race), comorbidities, hospitalization status, lab results, and other clinical information was obtained from the electronic medical record (EMR). The collection of biological specimens and de-identified patient information was approved by the University of Virginia Institutional Review Board (IRB-HSR #22231 and 200110).
Cytokine concentrations in plasma were measured using Bio-Plex Pro Human Cytokine 27-plex Assay (R & D Systems, Minneapolis, MN) on a Bio-Plex 200 suspension array system (Bio-Rad, Hercules, CA, USA). Cytokines detected were IL-1β, IL-1ra, IL-4, IL-5, IL-6, IL-7, IL-8, IL-9, IL-10, IL-12(p70), IL-13, IL-17, Eotaxin, bFGF, G-CSF, IFN-γ, IP-10, MCP-1(MCAF), MIP-1α, PDGF-bb, MIP-1β, RANTES and TNF-α. IL-2, IL-15, VEGF, and GM-CSF were below the limits of detection in most patients.
COVID-19 patients who ultimately required mechanical ventilation were identified through the EMR. The cytokine levels at the time of COVID-19 diagnosis or admission were compared between patients who did and did not require ventilatory support using a Mann-Whitney U test. A Mann-Whitney U test was also performed to compare biomarkers between COVID-19 positive patients with elevated blood glucose (≥126 mg/dl; n = 19) and patients with normal glucose (<125 mg/dl; n = 19). Patient cytokine levels were separated into quartiles and log-rank tests were used to generate Kaplan-Meier curves. Odds ratios were calculated using general logistic regression, and confidence intervals that did not overlap one were considered statistically significant. For hierarchical clustering, the pheatmap function in the pheatmap library was used (R-project.org). Cytokines were scaled by row (patients), and the clustering was calculated using the complete linkage method. Statistical analyses were performed using GraphPad Prism and R.
Results
To understand the relationship between the different measured cytokines in our cohort, we used hierarchical clustering to generate a heatmap for patients and cytokines (Figure 1A). Cytokines that were clustered in close proximity with one another had similar expression dynamics within patients, compared to cytokines more distant. IL-6, a cytokine of high interest for therapeutic intervention in COVID-19 patients, clustered most closely with IL-5, which is a type 2 cytokine that among other properties, induces eosinophils [6]. We additionally observed clustering between IL-8 and IL-10, which may be due to their induction during CRS. Clustering away from these cytokines, were others including IL-13, IL-7, and IL-1β, suggesting these may be differently regulated responses. Of note there was little visible separation between patients who were ventilated compared to those who were not through clustering analysis, and no significant observable pattern in the heatmap. This highlighted the heterogeneity in responses between patients.
Seven cytokines were found to be elevated in COVID-19 patients that ultimately required ventilation (Figure S1). These included FGF basic (bFGF) (p=0.0006), IL-7 (p=0.0006), IFN-γ (p=0.0102), MIP-1α (p=0.0195), IL-13 (p=0.0159), G-CSF (p=0.0368), and IL-17 (p=0.0448). IL-1ra (p=0.0554) and IL-1β (p=0.0707) were also trending towards being significantly higher in the ventilated patients. For the seven cytokines with elevated expression in ventilated patients, we performed a logistic regression to calculate the odds ratio for ventilation status in the patients, while controlling for gender, age, and presence of comorbidities. For cytokines with significant odds ratios, we then divided the expression levels into halves or quartiles and generated Kaplan-Meier curves for time-to-ventilation after admission to the hospital. From this, three cytokines were found to have significant odds ratios and Kaplan-Meier curves: IL-13, IL-7, and bFGF (Figure 1B-D). Patients in the top half percentile of IL-13 were significantly more likely to require ventilation (p = .005). Additionally, IL-13 levels were predictive of ventilation, with the highest odds ratio of 1.58. IL-7, which is an important for induction and maintenance of lymphocytes [7] was also predictive of ventilation (OR: 1.04) and patients in the top three percentiles for IL-7 expression were significantly more likely to require ventilation (p = .015). Finally, bFGF was also moderately, but significantly, predictive for ventilation status (OR: 1.04), and the upper three quartiles were also associated with ventilation (p = .018). When considering these three cytokines together, we also observed that they clustered closely with one another in Figure 1A.
Finally, to test whether diabetes mellitus was associated with higher levels of proinflammatory cytokines, COVID-19 positive patients were separated by whether they had normal (<126 mg/dl) or elevated glucose (≥126 mg/dl). Of the cytokines tested, IFN-γ (P=0.0011) and IL-1ra (p=0.0042) were significantly higher in patients with elevated glucose levels (Supplemental Figure 2A-B).
Discussion
The most important finding of this report is the discovery of an immunophenotype associated with future risk for ventilatory support in patients with COVID-19. Infection with SARS-CoV-2 has been associated with a dysregulated immune response, primarily characterized by elevated levels of cytokines such as IL-6, TNFα, IL-1β, and IL-10 in patients with severe disease [2,3]. The association of these and other proinflammatory cytokines with severity has suggested that cytokine storm may be a pathologic driver of disease, and that immune responses may be targetable options for therapeutics. To observe the relationships between these proinflammatory cytokines in our UVA COVID-19 positive patient cohort, we performed hierarchical clustering analyses of their cytokine levels.
Here we report that the type 2 cytokine IL-13 is predictive of the future need for ventilation in patients with COVID-19, as well as IL-7 and bFGF. Notably, these three cytokines clustered together in the dendrogram, as well as significantly correlated with each other by linear regression (data not shown), suggesting some common regulatory pathway or response. The proximity of IL-13 to IL-7 may suggest that there is IL-7-induced type 2 Innate Lymphocytes (ILC2) or CD4+ Th2 population that produces IL-13. IL-13 is often considered an anti-inflammatory cytokine, but within the lung is known to promote eosinophilia and tissue remodeling [8]. It can also increase the production of the IL-1 receptor antagonist (IL-1ra) [9,10], which functions to inhibit IL-1β mediated inflammation. Additionally, IL-1ra absolves inhibition of bFGF by IL-1β [11], and may explain the close clustering between IL-13 and bFGF. In our cohort, however, IL-13 was not positively correlated with IL-1ra (data not shown), so further studies to understand this pathway will be required.
Given the interest in elevated IL-6 in patients in other studies, we also tested which cytokines clustered with IL-6, and saw that IL-5 was most closely linked, suggesting an association between IL-6 and type 2 immunity, even though IL-6 itself was not associated with future need for ventilation in this cohort.
Additionally, we observed that patients with elevated blood glucose had increased IFN-γ and IL-1ra indicating a proinflammatory response. Patients with diabetes mellitus are known to have excessive inflammation and have been found to have worse outcomes due to COVID-19 [6]. IFN-γ is part of the Th1 immune response and an activator of macrophages, therefore an increase in IFN-γ may indicate hyperactivation of macrophages resulting in hyperinflammation of tissues [8,12]. IFN-γ also recruits other inflammatory cells that may induce a cytokine storm [12]. Studies with an increased sample size will be important in understanding the importance of IL-13 and type 2 immunity in patients with diabetes.
Overall, we show that a previously underappreciated cytokine, IL-13, independent of gender, age, or comorbidity, is associated with worsened outcomes from COVID-19, highlighting that type 2 immunity may be important to disease. Additionally, higher levels of proinflammatory IFN-γ were observed in patients with elevated glucose, suggesting comorbidities may contribute to excessive inflammation. While functional studies to test the specific contribution of these cytokines to disease will be crucial, the identification of their significant predictive capabilities for severe disease increases our understanding of this infection. Such immunophenotyping may improve diagnostic capabilities, allowing us to prioritize treatment, and utilize patient-specific approaches.
Data Availability
The data will be available online when the peer-reviewed manuscript is published.
Potential Conflicts of Interest
W.P. is a consultant for TechLab, Inc. and Syneos Health.
Acknowledgements
We thank Judi Allen for discussion, Pan Tongvichit for excellent technical support and Genevieve Lyons and Jennie Ma for biostatistics consultation. This work was supported by NIH grants R01 AI124214, UL1TR003015, the Global Infectious Diseases Institute and by the Henske family.