Plasma Proteome Fingerprints Reveal Distinctiveness and Clinical Outcome of SARS-CoV-2 Infection

Background: We evaluated how plasma proteomic signatures in patients with suspected COVID-19 can unravel the pathophysiology, and determine kinetics and clinical outcome of the infection. Methods: Plasma samples from patients presenting to the emergency department (ED) with symptoms of COVID-19 were stratified into: (1) patients with suspected COVID-19 that was not confirmed (n = 44); (2) non-hospitalized patients with confirmed COVID-19 (n = 44); (3) hospitalized patients with confirmed COVID-19 (n = 53) with variable outcome; and (4) patients presenting to the ED with minor diseases unrelated to SARS-CoV-2 infection (n = 20). Besides standard of care diagnostics, 177 circulating proteins related to inflammation and cardiovascular disease were analyzed using proximity extension assay (PEA, Olink) technology. Results: Comparative proteome analysis revealed 14 distinct proteins as highly associated with SARS-CoV-2 infection and 12 proteins with subsequent hospitalization (p < 0.001). ADM, IL-6, MCP-3, TRAIL-R2, and PD-L1 were each predictive for death (AUROC curve 0.80–0.87). The consistent increase of these markers, from hospital admission to intensive care and fatality, supported the concept that these proteins are of major clinical relevance. Conclusions: We identified distinct plasma proteins linked to the presence and course of COVID-19. These plasma proteomic findings may translate to a protein fingerprint, helping to assist clinical management decisions.


Introduction
SARS-CoV-2 has become a pandemic, affecting more than 219 million people (October 2021), afflicting a significant percentage of infected individuals with symptoms severe enough to require hospitalization. The treatment of COVID-19 patients is challenging because of the clinical variability of the disease, ranging from asymptomatic to severe courses, with potentially fatal outcomes [1][2][3][4]. Several biomarkers and protein signatures have been suggested that can predict the course of COVID-19, particularly a severe clinical outcome, e.g., intensive care unit (ICU) treatment, mechanical ventilation (MV), and multi organ failure [5][6][7][8][9]. the time of admission to the ED or consecutively during the hospital stay. Hematology was measured on a Sysmex hematology analyzer within an XN-2000 configuration in EDTA tubes, clinical chemistry parameters were determined on Roche cobas c701, e602, and e801 analyzers within a cobas 6000/8000 configuration in heparin plasma or serum. Coagulation testing was performed using citrate plasma and a STAR MAX analyzer (Stago Germany GmbH, Düsseldorf, Germany).

Fingerprinting the Circulating Proteome
To measure the circulating proteome, aliquots of standard of care samples were used that were taken from patients at presentation to the ED. In cases of hospitalization (n = 53), additional samples were collected from the latest time a blood sample was taken (either prior to discharge or death) and at the mean of hospital stay (total of 3 samples). For 10 cases, the hospital stay was covered by a total of 7 plasma samples each taken at 7 consecutive times.
Plasma levels of altogether 177 different proteins were analyzed using Olink's panels Cardiovascular II (Lot #B02623) and Inflammation (Lot #B02608) that employ their proximity extension assay (PEA) technology [20]. Briefly, one microliter of plasma was incubated with a set of probes, each consisting of an antibody conjugated to a specific DNA oligonucleotide. Once a protein is recognized by a pair of probes, the DNA oligonucleotides of the antibody pairs, now in close proximity, are allowed to hybridize to each other and are extended by enzymatic polymerization. The newly formed DNA molecule is then amplified and quantified by real-time PCR. The PCR results were analyzed as normalized protein expression (NPX) values on a log2 scale. NPX values were obtained by normalizing Cq-values against extension controls, inter-plate control, and a correction factor. A high NPX value corresponds to a high protein concentration and expresses relative quantification between samples but represents no absolute quantification that would allow comparing the levels of different proteins to another. Results from the proteome analysis did not impact the treatment of patients.

Statistics
Identifying relevant parameters. The aim was to identify proteins that indicated differences with regard to patient classification before hospitalization and to clinical endpoints. For every parameter, measured values were binned into "low", "middle", "high" with the same population size for each quantile and two half-populated quantiles were added for the upper and lower outliers. The two rows (comparing two classes of patients) of the corresponding cross table should overlap as little as possible, determined by the chi 2 -independence test of those cross tables. A classification was defined as "likely not independent from the measured parameter", if the p-value was <0.001, approximately one fifth of 1/184 (total number of parameters including analytes spotted on both PEA-plates).
Restricting to clinically relevant parameters. The quantized chi 2 -test led to a set of identified parameters. To reduce the risk that correlations between parameter quantile and clinical events were due to overfitting, the data set was further restricted to parameters which correlated to at least one cohort assignment (COVID-19 yes/no, hospitalization yes/no) as well as to at least one of the clinical endpoints (ICU admission, requirement for MV, TE, death).
Principal component analysis (PCA). Based on the matrix M of the selected clinically relevant parameters versus the first measurements of hospitalized COVID-19 patients (n = 53), a singular value decomposition (PCA) was performed with M = U T ∑V.
The prognostic and diagnostic strength of the circulating proteins was compared using area under the receiver operating characteristics (AUROC), calculated with the R package pROC. Standard of care parameters and baseline characteristics were compared using Mann-Whitney U and Fisher's exact tests.
All statistical analyses were performed on R version 3.6.1 and MATLAB/Octave.

Patient Stratification, Clinical and Standard of Care Laboratory Assessment
Altogether, n = 141 patients at the ED were enrolled and allocated to the following major subgroups ( Figure 1): (1) patients with negative SARS-CoV-2 PCR results but with clinical symptoms compatible with COVID-19 (COVID-19 negative group, n = 44); (2) patients with positive SARS-CoV-2 PCR results but without hospitalization (non-hospitalized COVID-19 group, n = 44); (3) patients with positive SARS-CoV-2 PCR result and hospitalization (hospitalized COVID-19 group, n = 53). Within the hospitalized COVID-19 group, 24 patients were admitted to the ICU, 16 with indication for MV ventilation, 6 experienced TE, and 7 patients died. Baseline demographics for all patient groups are shown in Supplemental Table S1. The groups did not show any clinically significant or unexpected differences with regard to sex and major comorbidities. The prognostic and diagnostic strength of the circulating proteins was compared us ing area under the receiver operating characteristics (AUROC), calculated with the R package pROC. Standard of care parameters and baseline characteristics were compared using Mann-Whitney U and Fisher's exact tests.
All statistical analyses were performed on R version 3.6.1 and MATLAB/Octave.

Patient Stratification, Clinical and Standard of Care Laboratory Assessment
Altogether, n = 141 patients at the ED were enrolled and allocated to the following major subgroups (Figure 1): (1) patients with negative SARS-CoV-2 PCR results but with clinical symptoms compatible with COVID-19 (COVID-19 negative group, n = 44); (2) pa tients with positive SARS-CoV-2 PCR results but without hospitalization (non-hospital ized COVID-19 group, n = 44); (3) patients with positive SARS-CoV-2 PCR result and hos pitalization (hospitalized COVID-19 group, n = 53). Within the hospitalized COVID-19 group, 24 patients were admitted to the ICU, 16 with indication for MV ventilation, 6 ex perienced TE, and 7 patients died. Baseline demographics for all patient groups are shown in Supplemental Table 1. The groups did not show any clinically significant or unexpected differences with regard to sex and major comorbidities.  At admission to the emergency department, the initial clinical laboratory data revealed no major differences in clinical chemistry parameters when symptomatic non-COVID-19 patients were compared with combined hospitalized and non-hospitalized COVID-19 patients (Supplemental Table S2). Only white blood cell (WBC) counts were lower in COVID-19 patients compared to non-COVID-19 patients. Notably, the standard of care parameters CRP, procalcitonin (PCT), and lactate dehydrogenase (LDH) did not differ, although they were frequently reported as COVID-19-associated analytes [12]. Thus, these analytes did not present a pathognomonic COVID-19 pattern for patients in the ED ( Figure 2). As expected, our comparison of hospitalized and non-hospitalized COVID-19 patients found significantly higher LDH, CRP, PCT, and troponin T (TNT) concentrations and lower estimated glomerular filtration rates (eGFRs) among the COVID-19 inpatients (Supplemental Table S2) [12]. of care parameters CRP, procalcitonin (PCT), and lactate dehydrogenase (LDH) did not differ, although they were frequently reported as COVID-19-associated analytes [12]. Thus, these analytes did not present a pathognomonic COVID-19 pattern for patients in the ED ( Figure 2). As expected, our comparison of hospitalized and non-hospitalized COVID-19 patients found significantly higher LDH, CRP, PCT, and troponin T (TNT) concentrations and lower estimated glomerular filtration rates (eGFRs) among the COVID-19 inpatients (Supplemental Table S2) [12].

Plasma Proteins Associated to COVID-19 Status
The levels of 177 individual circulating proteins were determined in the four study groups at the time of admission (Supplemental Table S3, including full names and abbreviations of analyzed proteins). As highlighted in Figure 3, within this proteome analysis, there were only 14 mainly inflammatory proteins (CXCL10, CXCL11, CXCL5, Gal-9, INF-gamma, IL-18, IL-18R1, LIF-R, MCP-2, MCP-3, MERTK, MMP-1, PD-L1, and TNF) that differed significantly (p < 0.001) between COVID-19 patients (hospitalized and non-hospitalized) compared to those with a symptomatic non-COVID-19 diagnosis. This suggests pathognomonic proteome changes triggered by SARS-CoV-2, but also highlights that only 8% were changed in the cross-sectional analysis when compared to other symptomatic patients ( Table 1). The assumption that these proteins define a pathophysiological signature of COVID-19 was underlined by the finding that no significant differences of protein levels were detected for all different 177 proteins analyzed between the reference subgroup and the symptomatic non-COVID-19 groups (significance level of p < 0.001, Supplemental Table S3). 8% were changed in the cross-sectional analysis when compared to other symptomatic patients ( Table 1). The assumption that these proteins define a pathophysiological signature of COVID-19 was underlined by the finding that no significant differences of protein levels were detected for all different 177 proteins analyzed between the reference subgroup and the symptomatic non-COVID-19 groups (significance level of p < 0.001, Supplemental Table S3).

Comparison of Hospitalized and Non-Hospitalized COVID-19 Patients
Next, we focused on protein levels with regard to differences between COVID-19 patients that were discharged from the ED based on clinical evaluation (non-hospitalized COVID-19 group) or were subsequently hospitalized (hospitalized COVID-19 group). We identified a subset of protein analytes that were significantly (p < 0.001) elevated in hospitalized COVID-19 patients (ADM, CTSL1, HGF, IL-27, IL-6, KIM1, MERTK, MMP-1, MMP-12, OPG, TNFRSF10A, and TRAIL-R2) ( Figure 4) (medians, interquartile ranges, and p-values are shown in Supplemental Table S3). The shortlist proteins include known markers of cellular degradation, hormones, as well as interaction with microbes and viruses. Figure 3. Proteome analyses-PEA values of discriminatory analytes with significant differences (p < 0.001) between non-COVID-19 and COVID-19 patients. Data are visualized by violin plots, which are a combination of kernel density plots, and box plots depicting median, 25th, and 75th percentiles. A high-normalized protein expression (NPX) value resembles a high protein concentration and expresses the relative quantification between samples. NPX values are given on a log-scale. Green = non-COVID-19; yellow = COVID-19 without hospitalization; red = COVID-19 with hospitalization; black dots: outliers. Table 1. Selected protein levels analyzed by proximity extension assay (PEA), number of valid measurements, interquartile ranges, and significances (p-values) of the non-COVID-19 and the COVID-19 group at the time of admission to the ED. PEA proteome values are selected on the basis of significant differences either between non-COVID-19 patients and COVID-19 patients, or within the COVID-19 patient group regarding significant predictive impact (refer to Table 2). Only p-values < 0.001 are depicted in bold.

Comparison of Hospitalized and Non-Hospitalized COVID-19 Patients
Next, we focused on protein levels with regard to differences between COVID-19 patients that were discharged from the ED based on clinical evaluation (non-hospitalized COVID-19 group) or were subsequently hospitalized (hospitalized COVID-19 group). We identified a subset of protein analytes that were significantly (p < 0.001) elevated in hospitalized COVID-19 patients (ADM, CTSL1, HGF, IL-27, IL-6, KIM1, MERTK, MMP-1, MMP-12, OPG, TNFRSF10A, and TRAIL-R2) (Figure 4) (medians, interquartile ranges, and p-values are shown in Supplemental Table S3). The shortlist proteins include known markers of cellular degradation, hormones, as well as interaction with microbes and viruses.

Proteome Analyses in Hospitalized COVID-19 Patients with Regard to Clinical Outcome
Using samples collected at the initial admission to the ED, we investigated the associations of the circulating proteins with outcomes in hospitalized COVID-19 patients represented by four clinical events: ICU treatment, need for mechanical ventilation (MV), TE, and death (Table 2). Only the plasma levels of CCL23, IL6, MCP-1, MCP-3, PD-L1, and TRAIL-R2 were significantly higher (p < 0.001) in patients who subsequently received ICU treatment. Furthermore, a subset of five proteins (DCN, IL6, MCP-1, MCP-3, and TRAIL-R29) was significantly higher (p < 0.001) in patients with subsequent MV. Interestingly, the COVID-19 specific elevation of MCP-3, predicted, in addition to ICU care also the need for ventilation. Finally, the protein levels of ADM (p < 0.0001) and LPL (<0.0002) were strongly associated with fatal outcome.
Interquartile ranges for protein levels as well as standard of care parameters CRP, PCT, WBC, and LDH with regard to clinical outcome are shown in Table 2 and AUROC values in Supplemental Table S4. A number of analytes had high AUROCs with respect to fatal outcome, in particular ADM (0.87), CXCL10 (0.82), CXCL11 (0.82), DCN (0.80), IL-27 (0.82), and TNF (0.80). Noteworthy, no standard of care parameter (CRP, PCT, WBC, and LDH) was significantly accurate for the analyzed outcomes.

A Proteomic Fingerprint Predicts Outcome
To further increase the predictive power of the plasma proteomics, we combined the different individual proteins into a single model. Based on our mathematical approach to identify relevant parameters, we selected five patient-discriminatory proteins (ADM, IL-

Proteome Analyses in Hospitalized COVID-19 Patients with Regard to Clinical Outcome
Using samples collected at the initial admission to the ED, we investigated the associations of the circulating proteins with outcomes in hospitalized COVID-19 patients represented by four clinical events: ICU treatment, need for mechanical ventilation (MV), TE, and death ( Table 2). Only the plasma levels of CCL23, IL6, MCP-1, MCP-3, PD-L1, and TRAIL-R2 were significantly higher (p < 0.001) in patients who subsequently received ICU treatment. Furthermore, a subset of five proteins (DCN, IL6, MCP-1, MCP-3, and TRAIL-R29) was significantly higher (p < 0.001) in patients with subsequent MV. Interestingly, the COVID-19 specific elevation of MCP-3, predicted, in addition to ICU care also the need for ventilation. Finally, the protein levels of ADM (p < 0.0001) and LPL (<0.0002) were strongly associated with fatal outcome.
Interquartile ranges for protein levels as well as standard of care parameters CRP, PCT, WBC, and LDH with regard to clinical outcome are shown in Table 2 and AUROC values in Supplemental Table S4. A number of analytes had high AUROCs with respect to fatal outcome, in particular ADM (0.87), CXCL10 (0.82), CXCL11 (0.82), DCN (0.80), IL-27 (0.82), and TNF (0.80). Noteworthy, no standard of care parameter (CRP, PCT, WBC, and LDH) was significantly accurate for the analyzed outcomes.

A Proteomic Fingerprint Predicts Outcome
To further increase the predictive power of the plasma proteomics, we combined the different individual proteins into a single model. Based on our mathematical approach to identify relevant parameters, we selected five patient-discriminatory proteins (ADM, IL-6, MCP-3, TRAIL-R2, PD-L1) that were predictive for COVID-19 rule-in and/or hospitalization and at least for one of the following events: ICU treatment, TE, MV, and death. The principal component analysis (PCA) of these proteins (ADM, IL-6, MCP-3, TRAIL-R2, PD-L1) is shown in Figure 5. To illustrate the proteome-based clustering of patient groups in terms of clinically relevant events, we selected those three proteome parameters from the identified proteins, which predicted hospitalization (ADM, IL-6, TRAIL-R2). We plotted this projected plasma proteomics data into a three-dimensional space ( Figure 6A). For the analysis using ED samples only, we included all COVID-19 patients (non-hospitalized and hospitalized, n = 97). For the middle and last sampling time point, we only projected the data from the hospitalized patients. The scatter plots reveal a robust clustering of the different clinical groups and even more distinct separation during the course of COVID-19 infection.
Similarly, the high-resolution kinetic ( Figure 6B) of the selected markers ADM, IL-6, MCP-3, TRAIL-R2, and PD-L1 demonstrated prognostic value from ED blood samples. The relative levels of the individual markers fluctuated slightly during hospitalization, a disadvantage for diagnostics, with the exception of TRAIL-R2, which increased in nearly all of patients with fatal outcome during hospital stay. However, by combining these markers into a panel using PCA, the compound fingerprint remained stable over the hospitalization time ( Figure 6C) and was clearly not dependent of medical intervention (ICU, MV, and medication). The data that support the findings of this study are available on request from the corresponding author. A correlation matrix between the selected markers and the standard of care infection markers WBC, CRP, and PCT is shown in Supplemental Figure S1. To illustrate the proteome-based clustering of patient groups in terms of clinically relevant events, we selected those three proteome parameters from the identified proteins, which predicted hospitalization (ADM, IL-6, TRAIL-R2). We plotted this projected plasma proteomics data into a three-dimensional space ( Figure 6A). For the analysis using ED samples only, we included all COVID-19 patients (non-hospitalized and hospitalized, n = 97). For the middle and last sampling time point, we only projected the data from the hospitalized patients. The scatter plots reveal a robust clustering of the different clinical groups and even more distinct separation during the course of COVID-19 infection.
Similarly, the high-resolution kinetic ( Figure 6B) of the selected markers ADM, IL-6, MCP-3, TRAIL-R2, and PD-L1 demonstrated prognostic value from ED blood samples. The relative levels of the individual markers fluctuated slightly during hospitalization, a disadvantage for diagnostics, with the exception of TRAIL-R2, which increased in nearly all of patients with fatal outcome during hospital stay. However, by combining these markers into a panel using PCA, the compound fingerprint remained stable over the hospitalization time ( Figure 6C) and was clearly not dependent of medical intervention (ICU, MV, and medication). The data that support the findings of this study are available on request from the corresponding author. A correlation matrix between the selected markers and the standard of care infection markers WBC, CRP, and PCT is shown in Supplemental Figure S1.

Discussion
Our in-depth and targeted analysis of the immune and cardio-metabolism related plasma proteins described the molecular phenotypes related to a specific pathophysiology of COVID-19 and outlined that circulating proteins can predict clinical outcome. The high predictive accuracy, in particular with regard to fatal outcome, together with stable protein kinetics provide a strong rationale to further evaluate the identified proteins for clinical diagnostics. Moreover, we present a composite risk fingerprint that robustly predicts COVID-19 mortality in patients at initial presentation in the ED. Our study confirms previously suggested immune-markers, such as TNF and IL-6 [21,22], as well as hitherto unrecognized proteins as novel indicators for further disease progression.

How Can Proteomic Fingerprinting Help Clinicians Better Understand COVID-19?
Organ failure in severe COVID-19 cases seems to be due to this excessive reaction, often described as a cytokine storm, rather than due to viral virulence [23][24][25]. Routinely measured markers, such as hematological, hemostaseological and clinical chemistry parameters, were extensively evaluated [24,26]. However, while some provide prognostic information, their lack of specificity for COVID-19 pathophysiology and their poor performance for long-term clinical outcome limits their clinical value.

What Are the Advantages of Our Analysis Design?
To gain a greater overview we included not only patients with suspected COVID-19 and proven SARS-CoV-2 infection in pharyngeal swabs, but also patients with COVID-19 typical symptoms, but negative SARS-CoV-2 PCR (symptomatic COVID-19 negative group). COVID-19 positive patients were further subdivided into groups requiring hospitalization or not. In the COVID-19 positive and hospitalized group, we isolated three and up to seven longitudinal samples per patient in the hospitalization follow-up. By comparing non-COVID-19 and COVID-19 patients presenting with similar symptoms to the ED, the study design has the advantage of describing proteomic aspects of the pathophysiology of COVID-19, i.e., to identify proteins and proteome changes specific to COVID-19 patients and not for a respiratory symptom pattern. In the COVID-19 patient group, protein markers can further be evaluated in predictive models, either as early time point markers or in protein kinetics during disease course.
SARS-CoV-2 infection rapidly triggers the release of acute phase reactants, such as INF gamma and interleukins, which is considered a general characteristic of COVID-19 [27]. Indeed, some of these pathognomonic parameters have previously been shown to be associated with SARS-CoV-2 at some point between initial infection and clinical deterioration [23,24,28].
Interestingly, only few of the COVID-19 plasma markers are also related to disease outcome. This indicates that immune activation is a feature of symptomatic COVID-19 rather than a sign for deterioration. Thus, elevated cytokine and interleukin levels, including IL-2, IFN gamma, and TNF described in several studies [24,28] are pathognomonic for SARS-CoV-2 infection in general, but seem not primarily related to the feared and potentially lethal hyperactivation of the immune system.
Nevertheless, to draw conclusions of the immunological essence of COVID-19 is difficult and might not only depend on the phase of infection, but also be influenced by the individual host response, as well as host-dependent co-factors, such as age, comorbidities, and coinfections.

Which Plasma Proteins Indicate a Severe Outcome in COVID-19 Patients?
Our model identified 16 markers that were associated with high significance (<0.001) with clinical events (hospitalization, mechanic ventilation, thromboembolic event, or death). Remarkably, only nine markers (ADM, CCL23, DCN, IL6, LPL, MCP1, MCP3, PD-L1, and TRAIL-R2) were associated with clinical events after hospitalization, and of them, only the cytokine MCP-3 and the immune inhibitory receptor ligand PD-L1 were COVID-19 specific. This fits to the concept that COVID-19 triggers a cascade of immune response followed by a secondary cytokine wave in some patients, which is responsible for the fatal progression [24,25].
In our study, the circulating hormone ADM was the strongest single predictive marker that was associated with COVID-19 mortality and outperformed all other immune markers as well as laboratory parameters in its predictive power (AUROC = 0.87). These findings support further focused evaluation of ADM's pathophysiological role in organ failure and COVID-19 and its application as powerful diagnostic marker and potentially therapeutic target in COVID-19 patients [29].
Although ADM and other markers had a strong association with clinical outcomes on their own, we developed at a composite marker combining the circulating hormone ADM, the cytokines IL-6 and MCP-3, the apoptosis related receptor TRAIL-2R, and the immune inhibitory receptor ligand PD-L1 in order to further increase the predictive performance.
The combination of several markers to a COVID-19 risk score balances inter-and intra-patient variability due to the individual immune repertoire, comorbidities, and actual therapy, reflected by genetic aspects influencing the circulating proteomes [30]. Potential technical imprecisions can also be adjusted. The robustness of our composite COVID-19 risk PCA analysis is impressively illustrated in the kinetics during hospitalization. With the exception of the apoptosis-related protein TRAIL-R2, which was dynamically increased in those patients with fatal outcome, the other markers displayed a rather steady state. Stable high levels of the composite score were seen in the high-risk patients, despite differences in clinical treatment, e.g., admission to the ICU with or without indication for MV. We believe that refinement, by inclusion and exclusion of markers, and further standardization will increase the clinical utility for a potential diagnostic assay.

What Are Clinical Implications and Do Possible Therapeutic Targets Exist?
A better understanding of the underlying mechanisms resulting in distinct proteome changes and the proteins involved may enable targeted therapeutic approaches. Three of the five key proteins identified in this study in PCA sorting analyses have already been identified as therapeutic targets in COVID-19, and clinical trials are currently ongoing.
Inhibitors of IL-6 and its receptors, such as Tocilizumab, have been approved by the FDA for treating chimeric antigen receptor (CAR)-T-cell mediated cytokine release syndrome (CRS), and have been proposed as a new therapeutic strategy for other types of cytokine wave including sepsis [31]. Indeed, recent studies provided data in favor of improved outcome in critically ill patients with COVID-19 receiving IL-6 antagonists [32][33][34], while other studies produced mixed results [35,36].
ADM is a peptide hormone modulating endothelial function, potent vasodilator, and immune modulator [37], correlating with mortality in septic patients [38]. Like in our study, elevated levels of ADM were also observed in severe COVID-19 cases [39]. The anti-ADM antibody Adrecizumab stabilizes and maintains the endothelial barrier function and is thus supposed to reduce capillary leakage in septic shock, which is also described as a pathomechanism in COVID-19 [40,41]. In a small case series with eight critically ill COVID-19 patients, the administration of Adrecizumab seemed to favor the clinical outcome [42]. [43]. The expression of its cognate ligands PD-L1 (CD 274) and PD-L2 (CD 273) can be increased by various viruses [44]. Elevated blood levels of the soluble PD-L1 isoform seem to inhibit T-cell functions and thus be a marker of T-cell-exhaustion [45].

PD-1 (CD 279) is a regulator of T cell responses and immunological tolerance
In our study, PD-L1 was not only one of our composite markers in PCA analyzes predicting clinical outcome but was also significantly higher in COVID-19 patients compared to the symptomatic non-COVID-19 group. This extends current findings of significant difference of soluble PD-L1 levels at hospital admission between mild/moderate and severe/critical COVID-19 cases [45,46]. The PEA technology has several advantages compared to conventional antibody based assays [20,47] and possibly "omic"-technologies [16]: (1) the technique is ultrasensitive allowing parallel detection of multiple proteins comparable to commercially available singleplex immunoassays; (2) the use of paired antibodies and matching of complementary DNA tags to detect and report each protein at a high specificity; (3) the required plasma volume is minimal, in the range of microliters; and (4) the system is flexible to exclude and include new targets. Moreover, the technical reproducibility is highly indicated by the strong correlation (correlation coefficient 0.972-0.997) of the levels of same targets from different plates.
Although we believe our findings are solid and supported by previous research, the results should be interpreted with caution because of its observational nature. Most associations of markers and outcomes were highly significant, but ultimate confirmation can only be obtained in independent COVID-19 patient cohorts. The same is true for the construction of a risk score based on the combination of prognostic markers, which may imply overfitting. However, recent studies applying also PEA technology identified an overlap of proteins with our results in non-COVID-19 and COVID-19 patients in the ED, when comparing COVID-19 patients with healthy controls, e.g., CXCL10, CXCL11, IFN gamma, MCP-2, MCP-3, and TNF [48], and regarding HGF in hospitalized vs. nonhospitalized patients [46]. Finally, it needs to be stressed that our clinical data were obtained early in the pandemic when clinical evaluation and management of COVID-19 patients, in particular when presented to the ED, was driven by empirical considerations. Since then, optimized management has decreased the morbidity and mortality of COVID-19.

Concluding Remarks
In summary, we demonstrated diverse immune traits that were associated with the pathophysiology and the clinical outcome of COVID-19. As such, increases of proinflammatory markers underline disease severity dependent changes in the plasma proteome. The dynamic signatures of markers related to immune response and vascular homeostasis broaden our understanding of the systemic reaction to COVID-19 and will help to optimize biological and clinical assessment of proteins in the context of COVID-19. It is important to note that we employed a high significance level (p < 0.001) in our evaluations of proteomics data because we analyzed a high number of analytes in the samples from the different cohorts. This approach was used to avoid detecting random differences in the study population. While many aspects are still enigmatic in the pathogenesis of COVID-19, our study stresses the potential to monitor the suggested mediators of inflammation for informed risk stratification and alternative intervention.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/ 10.3390/v13122456/s1, Supplemental Figure S1: Correlation matrix between the selected markers and the standard of care infection markers WBC, CRP, and PCT. Blank cells were statistically insignificant. Supplemental Table S1: Baseline demographics of patients included in this study. Supplemental Table S2: Clinical chemistry, hematology, and hemostaseology of patient groups at