Serum samples of 49 healthy individuals who were not previously exposed to SARS-CoV-2 (judged from no detectable symptoms and no presence of SARS-CoV2 RNA or anti-SARS-CoV-2 antibodies) were analyzed and compared to sera from 68 convalescent individuals after a previous mild COVID-19 disease (mild symptoms that did not require hospitalization during the acute disease) and 25 acutely ill COVID-19 patients with severe symptoms that were treated at the ICU at the time of blood sampling. The convalescent and the healthy individuals were examined from April to November 2020 (acquired with less than 10 infected persons per 100,000 inhabitants). The acutely ill COVID-19 patients were examined in December 2020 and January 2021. All further details on the study design and the participating patients can be found in previous publications (43, 44). All samples were collected before vaccination against COVID-19 was available. Characteristics of the three patients groups are also given in Table 1 , and the cytokine profile of the three groups is given in Table 2 .

For the detection of sIL-6R, sgp130, IL-11, and sCD25 in human serum, DuoSet ELISA Kits (R&D System) were used according to manufacturers’ instructions. Where necessary, samples were diluted to stay within the detection range of the ELISA kit. The detection limits were as follows: 31.2 pg/ml IL-11, 6.2 ng/ml sIL-6R, 62.4 ng/ml sgp130, and 7.8 pg/ml sCD25. IL-6 serum levels have been published previously (43).

Statistical analyses were performed using IBM SPSS Statistics (Version 29). Normal distribution was evaluated by the Shapiro-Wilk test and the Kolmogorow-Smirnow test. Spearman rank correlation tests were used to evaluate the correlations between protein serum levels, BMI, age, sex and the number of known pre-existing conditions. One-way analyses of variances (ANOVA) for all five serum proteins were applied. Consecutively, Tukey’s multiple comparison tests (i.e., test of contrast-coefficients) as post hoc analyses were used to test for differences between healthy control participants and different groups of patients. Multiple contrast-tests are superior to ANOVA and mean comparisons with t-Tests in two samples regarding power and information (45). All p-values are two-tailed, and a p-value below 0.05 was considered as statistically significant.

The mass action law was applied to estimate the concentrations of the dimer of IL-6 and sIL-6R and the trimer of IL-6, sIL-6R and sgp130 in serum. First, based on the measured amounts of IL-6 and sIL-6R, the concentration of the IL-6:sIL-6R dimer was calculated. In the next step, based on the result of this calculation and the measured concentration of sgp130, the concentration of the IL-6:sIL-6R:sgp130 trimer was calculated.

The equilibrium concentration of IL-6, sIL-6R and IL-6:sIL-6R dimer is described by Equation 1,

To calculate the concentration IL-6:sIL-6R dimers, Equation 1 was rearranged to [Dimer], which results in Equation 2,

Assuming that one sgp130 protein binds to one IL-6:sIL-6R dimer, the concentration of the IL-6:sIL-6R:sgp130 trimer was calculated accordingly using Equation 3.

Molar concentrations of IL-6 (23.7 kDa), sIL-6R (50 kDa), and sgp130 (100 kDa) in serum were calculated for each patient based on the molecular weights of the three proteins. The dissociation constants of the IL-6:sIL-6R dimer and the IL-6: sIL-6R:sgp130 trimer are K_D1 = 0.5 nM (46) and K_D2 = 0.05 nM (47, 48), respectively.

The publicly available single-cell RNA sequencing dataset of PBMCs from COVID-19 patients and controls (49) was downloaded from the fastgenomics repository (Schulte-Schrepping_2020_COVID19_10x_PBMC dataset, as.h5ad file). Quality control was performed using scanpy version 1.7.2 (50), which included filtering out cells with fewer than 200 genes detected and genes expressed in fewer than 3 cells. Clustering analysis and cell type assignment information were used from the dataset. The expression levels for genes of interest were interrogated and visualized using scanpy’s built-in plotting functions. Differential gene expression analysis was done by normalizing the total raw counts per cell, log transforming the data and performing the comparison between “control” and “severe” groups, for specific cell types, using scanpy’s built-in functions.

The activity of IL-6 is controlled by sIL-6R and sgp130 proteins, which are present in rather high amounts in human blood. The amounts of sIL-6R and sgp130 remain mostly constant, even in patients with inflammatory diseases (51). In contrast, only few pg/ml IL-6 can be detected in healthy individuals, but these amounts can rise by several orders of magnitude during inflammation and infection (24). We have previously reported that the balance between sIL-6R and sgp130, which form a buffer system to neutralize small amounts of systemic IL-6, is disturbed in type 2 diabetes patients (27). We have additionally shown that such alterations occur also in patients with severe COVID-19 infections (42). In order to investigate whether such a phenomenon is also present in mild convalescent COVID-19 patients, we quantified IL-6, sIL-6R and sgp130 via ELISA in serum samples from 49 healthy unexposed individuals (HD), 68 mild COVID-19 convalescents (MC) and 25 acute severe COVID-19 patients (ICU, Table 1 ). As shown in Figure 1A , IL-6 levels were expectedly low in the HD group (14.7 ± 1.3 pg/ml), only marginally increased in the MC group (21.9 ± 2.1 pg/ml), but highly and significantly elevated in the ICU group (186.1 ± 67.9 pg/ml, p<0.0001, Tables 2 , 3 ). Similarly, patients in the ICU group had significantly increased sIL-6R serum levels (26.2 ± 3.9 ng/ml, p<0.0001, Figure 1B , Tables 2 , 3 ) compared to the HD group (11.0 ± 1.2 ng/ml, Figure 1B ). Importantly, sIL-6R levels were also significantly increased in the MC group (17.2 ± 1.5 ng/ml, Figure 1B ), despite the fact that their COVID-19 infection had caused only mild symptoms and was several months ago. When we analyzed sgp130 serum levels in the same patient samples, no significant differences between the HD group (252.1 ± 5.6 ng/ml), the MC group (248.7 ± 6.2 ng/ml) and the ICU group (267.6 ± 9.8 ng/ml) were detected ( Figure 1C ). The significant main effects on group serum levels remain significant even after controlling for covariates (sex, age, BMI, quantity of previous illnesses). Only for sIL-6R age was a significant covariate (F(1, 139) = 6,738*), but had no impact on the main effect.

Previous studies reported both increased and decreased sgp130 levels in severe COVID-19 patients (42, 52). In summary, we find that IL-6 and sIL-6R serum levels are significantly increased in acute severe COVID-19 patients, and that increased sIL-6R levels can also be detected in mild COVID-19 convalescents several months after their infection, suggesting a long-term effect on sIL-6R generation, and thus IL-6 function due to the altered buffer, in these patients.

Having shown that IL-6 and sIL-6R serum levels are increased in acute severe COVID-19 patients, we sought to determine how much of the IL-6 is trapped in inactive complexes and how much is free and able to do harmful activities, thereby potentially contributing to COVID-19 pathology. IL-6 binds to the sIL-6R with an affinity of 500 pM (46), and the resulting IL-6/sIL-6R complex binds to sgp130 with a higher affinity of 50 pM (47, 48). Whereas the IL-6:sIL-6R complex acts as an agonist and constitutes the pro-inflammatory part of the IL-6 biology (termed IL-6 trans-signaling) (17), the tripartite complex IL-6:sIL-6R:sgp130 is inactive, as it can no longer bind to cells in the body. Thus, not the pure IL-6 and sIL-6R levels are important to determine the possible impact of IL-6 and sIL-6R in a given disease, but rather how much active IL-6:sIL-6R complex and how much free IL-6 is present in the serum of a patient. In order to calculate this, we converted the amounts of IL-6, sIL-6R and sgp130 from ng/ml into pM ( Table 1 ) and calculated first how much IL-6:sIL-6R complex can be found in the patients. As shown in Figure 2A , unexposed healthy controls had 0.17 ± 0.03 pM IL-6:sIL-6R complexes in their serum. Serum levels of patients IL-6:sIL-6R complexes in the MC group were higher (0.33 ± 0.04 pM), but this difference was not statistically significant. In contrast, IL-6:sIL-6R complex levels were significantly higher in ICU patients (3.09 ± 1.1 pM, p<0.001, Figure 2A ). This is caused by the increase in both IL-6 and sIL-6R in these patients. Additionally, sIL-6R levels are always in molar excess over IL-6 levels, and thus an increase in IL-6, which we have seen in the ICU patients ( Figure 1A ), is the major driver of complex formation. Afterwards, we determined the amounts of the trimeric IL-6:sIL-6R:sgp130 complexes. As sgp130 levels are equal in all three groups ( Figure 1C ), the tripartite complex follows the same pattern with significantly elevated levels in ICU patients (3.04 ± 1.08 pM, p<0.001) compared to healthy controls (0.17 ± 0.02 pM) and mild convalescent patients (0.32 ± 0.04 pM, Figure 2B ). Importantly, after calculation of the formed complexes, the free IL-6 levels were still significantly elevated in the ICU patients (4.77 ± 1.78 pM, p<0.0001) compared to healthy controls (0.45 ± 0.04 pM) and the mild convalescents (0.59 ± 0.06 pM, Figure 2C ). Similarly, trans-signaling competent IL-6:sIL-6R complexes were still present after formation of the inactive tripartite complexes, which were significantly increased in ICU patients (0.06 ± 0.02 pM, p<0.0001) compared to healthy controls (0.003 ± 0.001 pM) and mild convalescents (0.006 ± 0.001 pM, Figure 2D ). In summary, our results show that ICU patients on the one hand have more IL-6 that is inactivated in IL-6:sIL-6R:sgp130 complexes compared to the other two groups, but on the other hand have still significantly more free IL-6 and biologically active IL-6:sIL-6R complexes than healthy controls and mild convalescents.

Cytokines can only act on cells that express the required receptors on their cell surface. Despite this fact, the question which cell types express which cytokine receptors is still largely unexplored, and whether the expression pattern of cytokine receptors is altered during disease states is also unclear (24). In order to obtain insights into this question, we used public single cell RNA sequencing data of immune cells derived from peripheral blood from severe COVID 19 patients and healthy controls published previously (49). We used the same strategy to identify different immune cell subsets as the original authors and analyzed gene expression in 23 distinguishable immune cell subsets ( Figure 3A ). Intriguingly, with the exception of one B cell subset in the healthy controls, immune cells did not significantly express IL6, which fits to the assumption that immune cells are not the only source of IL-6 production [ Figure 3B and (53)]. IL6R expression, in contrast, was detected in monocytes, neutrophils and dendritic cells, whereas expression in T and B cells was less pronounced. Moreover, differential gene expression analysis revealed that IL6R expression was significantly increased in total monocytes and reduced in neutrophils of COVID-19 patients ( Figure 3C ; Supplementary Figure S1 ). IL6ST, which encodes the signal-transducing receptor gp130, was more uniformly expressed throughout the investigated cell types and its expression was significantly downregulated in neutrophils and upregulated in monocytes, CD4+ and CD8+ T cells and B cells from severe COVID-19 patients ( Figure 3D ; Supplementary Figure S1 ).

Having found that IL-6 and sIL-6R serum levels are increased in acute severe COVID-19 patients, we sought to investigate whether this is a specific effect or whether similar proteins are also increased in the serum of these patients. IL-11 is the closest related protein to IL-6, as it belongs to the same cytokine family (12). IL-11 binds to a unique non-signaling IL-11 receptor (IL-11R) before it, like IL-6, recruits a homodimer of gp130 for signaling. Furthermore, previous work had shown that several respiratory viruses, e.g. respiratory syncytial virus (RSV), parainfluenza virus type 3 (PIV3) and rhinovirus (RV) 14 were potent inducers of IL-11 (54). In order to investigate whether COVID-19 infection would result in increased IL-11 levels, we first analyzed the above mentioned single cell RNA sequencing dataset. Intriguingly, IL11 expression was very low or completely absent in healthy and diseased individuals ( Figure 4A ), which is also reflected by the very small expression dots of the individual cell types ( Figure 4B ). When we used a specific ELISA to detect IL-11 in the same serum samples we had analyzed previously, we found no significant differences between IL-11 serum levels in patients from the HD group (389.3 ± 151.5 pg/ml), the MC group (210.7 ± 58.8 pg/ml) and the ICU group (161.7 ± 91.6 pg/ml) ( Figure 4C ; Tables 2 , 3 ). These findings rule out a major role of IL-11 in COVID-19 infection and underline that induction of IL-11 is not a uniform cellular response to viral infection.

We have recently shown that the soluble form of CD25 (sCD25) is generated by proteolytic cleavage of the membrane-bound CD25 by the metalloproteases ADAM10 and ADAM17 (29). Because the same proteases are responsible for the majority of the sIL-6R found in human serum (18), we investigated CD25 expression and sCD25 levels in our cohorts. While CD25 expression in healthy individuals was rather low and only moderately detectable in a distinct cell population, the levels were strongly increased in severe COVID-19 patients, being significantly upregulated in CD4+ T cells [ Figure 5A ; Supplementary Figure S1 , (55)]. We further mapped the increased expression to one B cell cluster ( Figure 5B ). When we analyzed sCD25 levels via ELISA, there was no significant difference between the HD group (854.1 ± 82.7 pg/ml) and the MC group (765.4 ± 43.0 pg/ml, Figure 5C , Tables 2 , 3 ). However, sCD25 levels were significantly increased in serum samples from the ICU group compared to both the HD and the MC group, which is in good agreement with previous results [p < 0.0001, Figure 3C , Tables 2 , 3 , (56)]. These results show that increased generation of soluble cytokine receptors in severe COVID-19 patients is not restricted to sIL-6R generation, but occurs for other cytokine receptors as well and might even be a general phenomenon. However, in contrast to sIL-6R ( Figure 1B ), we detected no increase in sCD25 in the MC group compared to healthy controls, which might be caused by different transcriptional and post-transcriptional mechanisms controlling expression and/or proteolysis of the two cytokine receptors.

Our data so far revealed differences for several serum proteins between healthy controls, mild convalescent and severely ill COVID-19 patients ( Table 3 ). In order to determine how these data are correlated to possible confounders like age, sex, body mass index (BMI) or the number of known pre-existing conditions ( Table 1 ), we coded the disease state (meaning whether the sample was from the HC, MC or ICU group) as one variable and performed a correlation analysis. We observed a positive correlation of the disease state with the age of the patients (r=0.33, p<0.01) and the number of known pre-existing conditions (r=0.42, p<0.01, Table 4 ). The numbers of known pre-existing conditions also highly correlated with the age (r=0.38, p<0.01) and the BMI of the patient (r=0.3, p<0.01, Table 4 ). These findings fit to previous studies showing that older age, higher body weight and known diseases increase the likelihood to experience a severe COVID-19 infection (6). We also noted significant correlations between the disease state and IL-6 serum levels (r=0.5, p<0.01), sIL-6R serum levels (r=0.33, p<0.01) and sCD25 levels (r=0.35, p<0.01), which is expected by our analysis shown in Figures 1 , 2 . These correlations remain significant after controlling for the possible confounders (sex, age, BMI, quantity of previous illnesses). Intriguingly, the number of known pre-existing conditions was also highly correlated with IL-6 (r=0.25, p<0.01), sIL-6R (r=0.21, p<0.05) and sgp130 (r=0.29, p<0.01, Table 4 ) serum levels, underlining that pre-existing diseases are an important predictor of COVID-19 disease course.

Interestingly, serum levels of IL-6 (r=0.21, p<0.05), sIL-6R (r=0.31, p<0.01) and sgp130 (p=0.35, r<0.01, Table 4 ) were positively correlated with age, a correlation that we had previously not observed for these proteins in type 2 diabetes patients (27). Additionally, IL-11 was recently shown to be an important regulator for systemic adipogenesis (57), and the only significant factor that was correlated to IL-11 serum levels in our samples was the BMI (r=-0.19, p<0.05, Table 4 ). Importantly, sIL-6R levels were significantly correlated with sgp130 levels (r=0.34, p<0.01, Table 4 ), which we had also previously seen in healthy individuals as well as patients with type 2 diabetes and which led us to conclude that these two proteins constitute a natural occurring buffer system that is able to neutralize low amounts of IL-6 in the circulation (27).

Because it is challenging to determine correlations between three groups, we now analyzed COVID-19 patients (combining the MC and the ICU) and healthy controls separately regarding correlations with confounding factors. The number of known pre-existing conditions was still highly correlated with age (r=0.44, p<0.01) and BMI (r=0.33, r<0.01, Table 5 ) among people infected with COVID-19, but this was not the case for the healthy unexposed controls ( Table 5 ). IL-6 serum levels in the patients correlated with age (r=0.29, p<0.01) and the number of known pre-existing conditions (r=0.26, p<0.05, Table 5 ), which was both not the case for the healthy unexposed controls, underlining that IL-6 levels are only relevant under pathological conditions ( Table 5 ). We further observed significant correlations of IL-11 serum levels with sex (r=0.31, p<0.01) and inverse correlations with age (r=-0.34, p<0.01) and BMI (r=-0.24, p<0.05, Table 5 ). Again, such correlations were not detected in the healthy control samples ( Table 5 ). In contrast, the correlation of sIL-6R serum levels with age was seen in patients (r=0.21, p<0.05) and even stronger in healthy controls (r=0.44, p<0.01, Table 5 ). Furthermore, sIL-6R serum levels significantly correlated with sgp130 levels, both in patients (r=0.3, p<0.01) and in healthy individuals (r=0.48, p<0.001, Table 5 ). Serum levels of sCD25 correlated significantly in COVID-19 patients with age (r=0.42, p<0.01), the number of known pre-existing conditions (r=0.37, p<0.01) and the IL-6 serum levels (r=0,29, p<0.01, Table 5 ). None of these correlations were detected in the healthy unexposed controls. In summary, these data clearly show that confounding factors that correlate with serum levels of different analyzed proteins can be differentiated between COVID-19 patients and healthy controls that have not been infected with SARS-CoV-2.

Having shown these differences between patients and healthy controls, we next sought to determine whether correlations differed between mild COVID-19 convalescent (MC group) and acute severe COVID-19 patients (ICU group). In the MC group, the number of known pre-existing conditions was still significantly correlated with both the age (r=0.31, p<0.05) and the BMI (r=0.31, p<0.05, Table 6 ) of the patients, whereas in the ICU group only the BMI correlated with the pre-existing conditions (r=0.45, p<0.05, Table 6 ). Interestingly, the IL-11 serum levels in the MC group correlated significantly with sex (r=0.26, p<0.05) and were inversely correlated with age (r=-0.41, p<0.01) and BMI (r=-0.38, p<0.01, Table 6 ). A significant correlation between sIL-6R and sgp130 serum levels were detected in ICU patients (r=0.41, p<0.05, Table 5 ), but not in the convalescent patients (r=0.24, p>0.05, Table 6 ). Furthermore, IL-6 and sIL-6R serum levels were significantly inversely correlated in acute severe COVID-19 patients (r=-0.44, p<0.05, Table 6 ).

In conclusion, our analysis shows that the serum profiles of mild convalescent and acute severe COVID-19 patients differ significantly.