COVID-19 disease is characterized by strong inflammation. Individual responses to infection vary enormously, ranging from asymptomatic infection to fatal disease. Here, we embarked on extensive longitudinal measurement of soluble analytes in serum, collected in two waves, to obtain insights in the pathophysiology of the disease but also aiming to identify markers with predictive value at hospital admission, that could guide clinical decision making.
Circulating analyte levels were measured by multiplex bead arrays on all available samples and results analysed in several ways using advanced, longitudinal computational models. As continuous scaling was critical, all sample collection time points were related to the onset of symptoms, rather than to hospital admission as most others assessing inflammatory profiles have done4. Firstly, we analysed levels of circulating analytes at the first available time point since study inclusion and compared these to outcome in categories (moderate, severe and fatal). Secondly, we correlated analyte levels at inclusion to the severity at that time point, as calculated with the SCODA severity-score21. Thirdly, predictive models were identified at time of inclusion aiming to predict outcome (survival vs fatal), disease severity (ICU vs no-ICU admission; high vs low max SCODA disease score), and duration of disease (long or no vs short recovery time). Fourthly, longitudinal models were developed to assess differences in analyte levels over time in the different disease outcome groups. Fifthly, disease severity was also modelled as continuous parameter over time to correlate with soluble analyte levels. Sixthly, analytes correlating with disease progression were compared to those associated with convalescence. Finally, soluble analytes at hospital discharge and outpatient follow-up were correlated to levels of healthy controls.
Surprisingly and intriguingly, the majority of analytes identified in all analysis strategies were very similar and typically overlapping. Analyte levels were characteristically higher with increasing disease severity, but the type of markers associated with different COVID-19 stages was very similar. Key markers that were upregulated in serum from hospital admitted individuals with COVID-19 disease were HGF, CXCL13, sCD30, CXCL16, IL-6, sCD163, IL-11, CXCL10, IL-16, CX3CL1, sIL-6Rβ, Osteopontin and Chitinase3like1, all associated with inflammatory responses, including cytokine storms, TLR activation and macrophage activation. Pathway analysis of all markers associated with disease severity over time also identified ‘pathogen induced cytokine storm signaling pathway’ as top hit, IL-10 and IL-17 signaling as prominent hits as well as ‘macrophage classical activation signaling pathway’. Even during convalescence, both at hospital discharge and outpatient follow-up, those markers were significantly increased compared to the healthy control population, suggesting persisting inflammation. Advanced longitudinal models, both based on outcome defined groups and continuous models on severity-scores also identified major associations with macrophage activation syndrome. Analytes identified at the earliest possible time point (study inclusion) were remarkably similar to those at the most severe disease state, indicating that ongoing inflammation is the major determinator of circulating analyte profiles. When comparing samples during disease progression and convalescence in association to severity-scores, more analytes were detected during disease progression compared to convalescence. However, the markers significantly correlating with disease severity during convalescence, are consistent with ongoing inflammation rather than resolution of inflammation or recovery. HGF and CXCL13 were also previously identified as markers of severe COVID-19 and good predictors of ICU admission5. In particular HGF has been described to dampen inflammation as counter-regulator of many pro-inflammatory cytokines and may even promote tissue repair5. In our data, HGF was also amongst the strongest correlates with severe disease at inclusion into the study, at maximum severity, longitudinally and during recovery but not at hospital discharge or during convalescence. Others have described increased HGF and KRT19 levels during convalescence and suggested they have potential value for prediction of lung impairment post-COVID-193. The observed differences in HGF associations with disease may result from cohort differences e.g. in disease severity but also in the definitions used to group the patients as well as potentially in timing of sample collection.
In our cohort, the major difference between wave-1 and 2 was the high proportion of dexamethasone usage in the 2nd wave, whereas none of the patients in the 1st wave received steroids early in disease. We observed strong differences in circulating pro-inflammatory markers between the waves, with more pro-inflammatory cytokines upregulated in wave-1. In severely-ill patients 33 analytes were significantly different between wave-1 and 2, likely reflecting differential dexamethasone treatment. Similarly, also in COVID-19 patients who received dexamethasone circulating levels of pro-inflammatory cytokines decreased 3–4 days after treatment start23. In our cohort IFN-γ and CXCL10 were significantly decreased in wave-2 compared to wave-1, possibly also as a result of the steroid treatment. Most other analytes were also less increased in wave-2 compared to wave-1, fully supporting these observations.
Our study had several limitations, most importantly the fact that only a single cohort of COVID-19 patients was analysed, without replication cohort. Although in some aspects the two waves could be considered two subcohorts from the larger cohort, with differences in duration of disease and routine treatment, which were also analysed independently. In addition, we have analysed our data with several different approaches (longitudinal vs cross-sectional, group classification vs continuous severity assessment) and identified the same marker signatures, which in a way could serve as within cohort validation. To the best of our knowledge, no datasets are publicly available with longitudinal data that would allow replication of our associations between circulating analytes and disease severity. Unfortunately, we were not able to include patients with similar clinical presentation but uninfected by SARS-CoV2. Albeit our primary goal was to assess analyte levels in relation to disease severity in COVID-19 patients for optimal treatment stratification, it would have been highly relevant to assess the same in patients with other causes of respiratory distress or other inflammatory conditions. Our cohort was heterogenous, in time since symptom onset, disease severity, treatment policies etc, but nevertheless we identified strong associations between circulating analytes and disease severity. Finally, our analysis was limited to patients that were hospital admitted and did not include less severe, or earlier stage patients with SARS-CoV2 infection, which would have been a valuable addition in particular for evaluation of the predictive signatures.
In summary, longitudinal measurement of circulating analytes combined with daily assessment of disease severity using a specifically developed score demonstrated that COVID-19 disease reveals strong pro-inflammatory profiles, most probably as a result of strong macrophage activation, with dominant roles of IL-6, IL-10 and IL-17 signaling cascades, which are associated with disease severity and outcome.