Patients were randomly selected based on positive SARS-CoV-2-RNA test results by RT-PCR from throat swab samples as well as absence of co-morbidities or pregnancy as assessed by a physician (lung diseases, cardiovascular diseases, cancer, autoimmune diseases, therapeutic immunosuppression). Patient recruitment followed a public announcement for voluntary study participation via the Facebook website of the Klinikum r. d. Isar, the University hospital of the Technical University of Munich. Blood sampling occurred 2 weeks after positive SARS-CoV-2-RNA test results in convalescent patients and on two additional timepoints in 2-week intervals. Patient characteristics are outlined in Figure 1. Clinical COVID-19 scores were assigned to each individual patient according to a questionnaire covering a wide range of 15 symptoms (Figure 1D), each scored 0-3 according to its degree of severity (range of clinical scores 0-45). All patients provided informed consent for this study. Ethical approval was obtained from the Ethics Review Board of the Technical University of Munich (164/20S and 147/20S) and the Ethics Review Board of the University Hospital Jena (Reg.-Nr.: 2020-1997-Material). All work involving humans was carried out in accordance with the Declaration of Helsinki for experiments involving humans.

Peripheral blood mononuclear cells (PBMC) were isolated immediately after blood collection into heparinized collection tubes (within 30 min) by density gradient centrifugation using Biocoll Separating Solution (Merck). They were analyzed immediately by flow cytometry (BDFortessa) using the following antibodies (CD45RA-BUV737, clone HI100; CD4-BV785, clone RPA-T4; CD25-BV650, clone BC96; CD8-BV510, clone RPA-T8; CD14-PacificBlue, clone HCD14; CD19-PerCP-Cy5.5, clone HIB19; CD56-FITC, clone MEM-188; CD197-PE-Cy7, clone G043H7; CD127-PE/Dazzle594, clone A019D5; CD3-PE, clone SK7; γδTCR-APC, clone B1), all from Biolegend. Dead cells were excluded by staining with Hoechst (Sigma-Aldrich). For assessment of cellular markers of acute infection such as plasmablasts, activated CD8 T cells and cTfh cells, frozen PBMC from the same patient cohort were analyzed. Cells were stained with fluorescently labeled antibodies and samples were acquired using the spectral flow cytometer Cytec Aurora (Cytek^® Biosciences). The antibodies are indicated in Figure S8B and were all purchased from Biolegend. Unmixing was performed using the software provided by the Cytec Aurora. FlowJo v10 was used for further analysis of the data.

PBMC were cultured in RPMI-1640 medium supplemented with 2 mM glutamine, 1% (vol/vol) non-essential amino acids, 1% (vol/vol) sodium pyruvate, penicillin (50 U/ml), streptomycin (50 μg/ml; all from Invitrogen) and 10% (vol/vol) fetal calf serum (Biochrom). PBMC were stimulated with plate-bound anti-CD3 (2 μg/ml, clone TR66; Enzo Life Science) and anti-CD28 (1 μg/ml CD28.2; BD Biosciences) for 3 days before intracellular cytokine staining and harvesting of supernatant.

For intracellular cytokine staining, human cells were restimulated for 5 hours with PMA and ionomycin with brefeldin A added for the final 2.5 hours of culture (all Sigma-Aldrich). The cells were fixed and permeabilized with Cytofix/Cytoperm (BD Biosciences) according to the manufacturer’s instructions. The cells were stained with anti-IL-10-PerCP-Cy5.5 (clone JES3-9D7), anti-TNFa-PE-Cy7 (clone Mab11), anti-Granzyme B-FITC (clone GB11), anti-IL-17A-Pacific Blue (clone BL168), anti-IFNγ-PE-Cy7 (clone 4S.B3), anti-GMCSF-APC (clone BVD2-21C11) (all from BioLegend); and anti-IL-4-PE (clone MP4-25D2) and anti-IL-2-APC (clone MQ1-17H12) (both from BD Biosciences). Cells were analyzed with a BD LSRFortessa (BD Biosciences). Flow cytometry data were analyzed with FlowJo software (Tree Star).

Cytokines and chemokines in serum or cell supernatants were measured using the multiplex bead array system Bio-Plex human Cytokine Group I Panel 7-Plex, with added single kits for human RANTES (CCL5), human IP-10 (CXCL10), human IFN-b, IFN-a2, and IL27p28 (Bio-Rad Laboratories GmbH), following the manufacturer’s recommendations. Data were acquired using the Luminex100 machine with BioPlex Manager 6.1 software. Standard curves were fitted using the logistic-5PL regression type.

Three patients with mild (m29, m42 and m43 with severity scores 7, 7 and 5) and two patients with severe (m10 and m11 with severity scores 18 and 27) disease course from the first sampled time point were chosen for RNA-seq analysis. Both groups contain patients of different gender and ages to equalize across cohorts. The two patients with a severe disease course had been hospitalized.

Samples were stained with anti-human CD14-PerCP (clone HCD14), anti-human CD45RA-FITC (clone HI100), and anti-human CD3-PE (clone SK7), all from BioLegend. Memory T Cells were defined as CD3⁺ CD45RA^– population and sorted with the BD FACSMelody including dead cell exclusion with Hoechst 33258. Cells were sorted directly into 96 well plates prefilled with 1X TCL lysis buffer consisting 50% 2X buffer TCL (QIAGEN), 1% 2-Mercaptoethanol (Gibco), and Nuclease-free water (QIAGEN) for RNA extraction.

Library preparation for bulk-sequencing of poly(A)-RNA was done as described previously (45). Briefly, barcoded cDNA of each sample was generated with a Maxima RT polymerase (Thermo Fisher) using oligo-dT primer containing barcodes, unique molecular identifiers (UMIs) and an adaptor. Ends of the cDNAs were extended by a template switch oligo (TSO) and full-length cDNA was amplified with primers binding to the TSO-site and the adaptor. NEB UltraII FS kit was used to fragment cDNA. After end repair and A-tailing a TruSeq adapter was ligated and 3’-end-fragments were finally amplified using primers with Illumina P5 and P7 overhangs. In comparison to Parekh et al. (45), the P5 and P7 sites were exchanged to allow sequencing of the cDNA in read1 and barcodes and UMIs in read2 to achieve a better cluster recognition. The library was sequenced on a NextSeq 500 (Illumina) with 63 cycles for the cDNA in read1 and 16 cycles for the barcodes and UMIs in read2.

RNA-seq data was processed using the published Drop-seq pipeline (v1.0) to generate sample- and gene-wise UMI tables (46). The human reference genome (GRCh38) was used for alignment. Transcript and gene defined according to GENCODE version 34 [doi: 10.1101/gr.135350.111].

For qRT-PCR frozen PBMC were thawed and sorted for CD3⁺ (CD3-PE, clone SK7, Biolegend) and Hoechst^– T cells (Sigma). RNA isolation was performed with the RNeasy Mini Kit (QIAGEN). Reverse transcription was performed with the high capacity cDNA reverse transcription kit (Applied Biosystems). qRT-PCR was performed with the Taqman Gene Expression Assay probe OASL FAM (HS00984387_ml) and IFI27 FAM (HS01086370_ml), 18S FAM (HS03928990_g1) (all Applied Biosystems) on the Roche Lightcycler 480 platform.

Protein-coding genes were selected from the count table using the Biomart annotation for the human genome from April 2021. Differential expression analysis was performed using the DESeq2 [version 1.28.1, (47)] R (version 4.0.2, R Core Team) package. Technical replicates were combined using the collapseReplicates() function of DESeq2. Wald tests were performed using the DESeq() function. Gene expression levels were contrasted between patients with a mild COVID-19 course and patients with a severe COVID-19 course. Genes with an adjusted p-value < 0.05 and a fold-change ≥ 2 or ≤ 0.5 were considered differentially expressed.

Library size-normalized regularized log-transformed expression values were computed using the rlog() function in the DESeq2 package. We further refer to these values as “expression values”. Principal component analysis (PCA) was performed on the expression values of the DEGs using the robpca() function of the rospca [version 1.0.4, (48)] R package. Heatmaps were generated using the pheatmap R package [version 1.0.12, (49)].

Functional enrichment analysis was performed with the enrichGO() function of the clusterProfiler R package [version 3.16.1, (50)]. The annotation of the human genome in R´s org.Hs.eg.db database (version 3.11.4) was used as background. Enriched gene ontology terms were summarized using REVIGO (51) and visualized with the treemap() function of the treemap R package (version 2.4.2).

Error bars indicate the standard error of the mean (SEM); 2way ANOVA tests with Holm-Sidak’s multiple comparisons tests were performed to compare mildly versus more severely affected patients longitudinally over time. p values of 0.05 or less were considered significant. Analyses were performed using GraphPad Prism 9.

Seventy-four patients were randomly selected based on the criteria of a positive SARS-CoV-2-RNA viral test result, lack of comorbidities, and balanced sex distribution (37 female, 37 male patients) (Table S1 and Figure 1A). The mean age of our patient cohort was 47 years, ranging from 19 to 80 (Figure 1B).

We established a clinical COVID-19 severity scoring system including 15 symptoms that have been reported to occur during this disease (52). Quantitative scoring of each symptom (range 0-3) allowed to achieve a final COVID-19 severity score with a mean of 14.5 in our cohort (range: 0-45) (Figure 1C). More than 50% of all patients presented with unspecific symptoms such as headache, cough, myalgia, fever and a sore throat. More than 50% of all patients presented with chemosensory alterations such as ageusia and anosmia, which have been proposed to be more characteristic of SARS-CoV-2 infection and to even occur in otherwise asymptomatic infections (Figure 1D) (53). In our patient cohort, the chemosensory loss of taste and smell was consistently associated with other symptoms and always occurred very early in COVID-19 (not shown). Increased age has previously been reported to represent an independent risk factor for severe COVID-19 disease course and death (5). Accordingly, we observed slightly higher disease scores in correlation with higher age in men. Interestingly, this age-severity correlation was not significant for women (Figure 1E). Although men have occasionally been reported to develop more severe COVID-19 (36), severity between men and women was equally distributed in our cohort (Figure 1F).

We also categorized our patient cohort according to scoring systems established for clinical monitoring of COVID-19, such as the Ordinal Scale for Clinical Improvement by the WHO R&D Blueprint. Our patient cohort covered WHO scores 1-4 in the Ordinal Scale for Clinical Improvement, which ranges from 0-8. In line with many previous reports that concluded an association of very severe disease courses with comorbidities (5), our patient cohort did not require intubation and thus lacked very severe disease outcomes (Table S1). Our clinical symptom scale correlated significantly with the WHO ordinal scale for clinical improvement (r= 0,530; p<0,0001) (Figure S1).

Our patient cohort covered a range of COVID-19 severities (WHO 1-4) in the absence of comorbidities and immunosuppression, thus excluding critical predictive markers and immunological abrogations that have been previously correlated with a severe disease course. Based on these highly controlled patient characteristics, phlebotomies were scheduled approximately two weeks after the start of symptoms and two more times at 14-day intervals. At the time of blood donation, no remaining disease symptoms except for occasional fatigue were recorded in any of the patients (Figure 1G). In sum, we present for the first time a large COVID-19 patient cohort with a spectrum of symptoms and disease severities, which is characterized by absence of comorbidities and thus by absence of a very strong bias that might have influenced immune parameters in previous COVID-19 immune monitoring studies.

There is substantial evidence that a dysregulated host immune response plays a critical role in disease severity in several viral infections including SARS-CoV-2 (25, 54). Comorbidities and associated therapies might predispose an individual’s immune system to a more severe disease course and even death as suggested by multiple previous studies (55, 56). However, how SARS-CoV-2 infection translates into a mild versus severe disease course in healthy individuals remains elusive. The host’s response to infection is mirrored by its immune signatures in the blood. We therefore set out to characterize the cellular immune cell composition in our convalescent COVID-19 patient cohort and, thereby, to correlate immunophenotypes with disease severity.

We first quantified the relative distribution of multiple immune cell subsets by multi-color flow cytometry of fresh PBMC (gating strategy, Figure S2). Convalescent COVID-19 patients were stratified into mild and a severe disease course based on their clinical severity scores (mild: ≤ 15, severe: ≥16) and assessed at three different time points after disease onset (2, 4, 6 weeks). The cut-off disease score value of 15 was chosen based on the mean disease severity score of 14.5 across all patients. We furthermore stratified our patient cohort into patients who required hospitalization or not as a surrogate for a mild versus more severe disease course (Table S1). For hospitalized patients inflammation markers such as C-reactive protein were assessed during ongoing COVID-19 two weeks prior to study enrollment (Table S1).

Surprisingly, the relative proportions of most immune cell types did not differ despite the divergent history of disease symptoms between the two patient groups (Figure 2A and Figure S3). However, CD14⁺ monocytes were significantly reduced two weeks after infection in the severely affected COVID-19 patients (Figure 2A). This alteration in the myeloid compartment was only detectable in early convalescent patients since monocyte frequencies did no longer differ from those of mildly affected COVID-19 patients upon later assessments at 4 and 6 weeks after infection. Relative frequencies of total lymphocytes and their subsets such as CD3⁺ T cells, γδ T cells, B cells, NK cells, NKT cells and Treg cells did not differ between COVID-19 patients with a mild versus more severe disease course at any time point analyzed (Figure 2A). Our longitudinal sampling of matched patient material also enabled us to conclude stability of most immune parameters across time in both severely or mildly affected patients.

To corroborate these unexpected findings, we also correlated the relative frequencies of the immune cell subsets to the respective COVID-19 symptom scores for each individual patient (Figure S3). We restricted our analysis to the first time point (2 weeks after infection). We did not observe any change in the relative frequency of any of the adaptive immune cell populations over the range of symptom scores in accordance with our previous analysis (Figure S3). The proportion of myeloid cells within the PBMCs did, however, significantly decline with increased COVID-19 severity (r=0.29, p=0.03). This was also the case for the fraction of CD14⁺ monocytes within the myeloid cell population (r=0.28, p=0.03) (Figure S3). We then excluded patients aged 60 and older to test if absence of this independent risk factor, as we showed in the case of men, would explain the negative correlation with myeloid cells or even unravel any new correlation with immune cell frequencies. In fact, the same results were recapitulated with a patient cohort spanning the age of 19 to 59 (mean 39 years) (data not shown). Interestingly, this demonstrates absence of any major perturbations in the relative frequencies of the main immune cell types, in particular adaptive immune cells, despite the range of COVID-19 severities in patients without relevant comorbidities. The physiological range of cell counts and frequencies is indicated by the dotted lines. Only the frequency of myeloid cells, and in particular CD14⁺ monocytes, correlated negatively with COVID-19 disease severity.

Contrary to relative immune cell proportions, the analysis of absolute numbers of immune cell subsets in PBMC revealed profound reductions in almost all immune cells in the severely affected COVID-19 patient cohort (Figure 2B). This showed that SARS-CoV2-induced pancytopenia across various immune cell types is masked if analysis of immune perturbations is restricted to relative frequencies of immune subsets. Only γδ T cells and B cell numbers were not affected by the disease course. This absolute loss of a wide spectrum of immune cell types in more severely affected patients was detectable in early convalescent patients (2 weeks after infection) but gradually recovered to levels indistinguishable from mildly affected patients after another 2 and 4 weeks. This suggests that even the adaptive immune compartment does not maintain major quantitative cellular alterations upon SARS-CoV2 infection. Immune signatures and biomarkers identified by multi-parameter flow cytometry are detectable in early convalescence only (Figure 2B).

To corroborate these conclusions, we reanalyzed all patient data using alternative disease severity definitions. This was important considering that our convalescent patient cohort, characterized by absence of relevant comorbidities, contained only patients scoring 1-4 in the ordinal scale for clinical improvement by the WHO (range: 0-8), which excludes severe disease in hospitalized patients (https://www.who.int/publications/i/item/covid-19-therapeutic-trial-synopsis) (Table S1). To account for the closer similarity of COVID-19 severity in our patient cohort and the overall milder disease course in the absence of cases with WHO scores greater than 5, we also stratified our patients into those with and without prior hospitalization during acute COVID-19. This showed significantly reduced frequencies of almost all analyzed immune cell subsets in patients, which had been hospitalized, at the early timepoint 1 (Figure S4A), which resolved at later stages. Only γδ-T cells and NK T cell frequencies did not display significant early differences in hospitalized versus non-hospitalized patients. Fewer significant differences in hospitalized versus non-hospitalized COVID-19 patients were observed upon analysis of absolute cell numbers (Figure S4B). Longitudinal analysis of these parameters across different timepoints within the mildly versus severely affected patient cohort revealed overall stability of immune parameters. However, the relative decline in lymphocytes was significantly restored over time (Figure S4A) in severely affected patients with stability across time in mildly affected patients. We also observed a significant increase in the percentage of CD3⁺ T cells between timepoint 1 and 3 in severly affected patients but not mildly affected patients. Also relative and absolute frequencies of NKT cells significantly increased from timepoint 1 to timepoint 3 within the severely affected patient group (Figure S4A).

Memory T cells represent the repository of an individual’s antigen experience (26). A gradual loss of antigen inexperienced naïve T cells over a lifetime in accordance with a history of antigenic challenges is associated with increased frequencies of memory and effector T cells (57). The memory T cell pool contains distinct populations of central and effector memory T cells that differ in their homing capacities, functions and history of antigen encounters (25, 58). Evidence of T cell activation in both CD4⁺ and CD8⁺ T cells has recently been reported in COVID-19 patients (14, 59). We therefore decided to investigate the distribution of distinct T cell differentiation states in the CD4⁺ and CD8⁺ T cell compartment in patients following SARS-CoV-2 infection. Naïve (T_Naive), central memory (T_CM), effector memory (T_EM) and terminally differentiated effector memory cells expressing CD45RA (T_EMRA) were identified based on the differential expression of CD45RA and CCR7 within CD4⁺ and CD8⁺ CD3⁺ T cells (CD19^–, CD56^–, γδ TCR^–) as described before (Figure S2) (58). Interestingly, we did not observe any difference between the mildly and severely affected patient cohort in the relative frequency of T cell differentiation states across the spectrum from naïve to T_EMRA cells at any time-point 2, 4, and 6 weeks after infection (Figure 3A). CD4⁺ and CD8⁺ T_Naive cells did, however, reveal a negative correlation across the full range of disease severity scores (CD4⁺ cells: r= -0.2974, p=0.02; CD8+ T cells: r= -0.3078, p=0.02) (Figures S5A, B). In CD8⁺ but not CD4⁺ T cells, this was associated with a concomitant increase in T_EM cells (Figures S5A, B). T_CM and T_EMRA cell frequencies did not correlate with disease scores. Taken together, the distribution of distinct T cell differentiation states in early convalescent COVID-19 patients was similar between the mild and severe COVID-19 patient cohort and across the range of disease severity scores, with significant changes mainly observable within T_Naive cells (Figures S5A, B).

Analysis of absolute T cell numbers, on the other hand, revealed a consistent significant reduction of naïve, T_CM and T_EM cells at the early time-point (Figure 3B). T_EMRA cells did not display any difference between the mildly and severely affected COVID-19 cohort. In the CD4⁺ T cell compartment a significant upregulation of T_EMRA cells was observed between the first and last timepoint within the severely affected patient cohort. Already 2 weeks later, no shifts in the absolute frequencies of T cell differentiation states were observable except for CD8⁺ T_Naive cells, indicating that SARS-CoV2 induced T cell alterations occur early and are not maintained on the T cell subset level (Figure 3B).

Reanalysis of patients who were categorized into hospitalized versus non-hospitalized, revealed a similar picture (Figures S6, S7). The bias for patients with a more divergent disease severity course stressed, in particular, the pronounced reduction in T_Naive cells in more severely affected patients across several timepoints. This was especially noteworthy in the CD8⁺ T cell compartment. Longitudinal analysis over all three timepoints showed an increase of CD8⁺ T cell frequencies between timepoint 1 and 3 within the severely affected patient group and stability among mildly diseased patients (Figure S6B).

We also assessed key immune populations relevant to acute infection, including activated CD4⁺ and CD8⁺ T cells, circulating T follicular helper cells (cTfh) and antibody secreting plasmablasts and plasma cells (gating strategy, Figure S8). Although T cell activation markers such as HLA-DR and CD38 showed a higher expression tendency in the more severe group, especially early after infection (TP1), the difference in the frequency of activated CD4⁺ and CD8⁺ T cells between mild and more severely affected patients was not significant (Figure 3C). There was also no difference in the frequency of cTfh cells in both patient groups (Figure 3D, left panel). Activation markers showed a tendency for higher expression in cTfh cells in the more severely affected COVID-19 patients at the early time point, but this was not a statistically significant difference (Figure 3D, right panel). This was also the case for plasmablasts (CD27⁺ CD20^lo CD38⁺CD138^–, left panel) but not for plasma cells (CD27⁺ CD20^lo CD38⁺ CD138⁺, right panel) (Figure 3E).

T cell cytokine production is tailored to the specific elimination of pathogens and represents a surrogate of previous antigen exposure and the history of signal integration (25). We reasoned that SARS-CoV2 infection and consecutive perturbations of blood and tissue homeostasis in conjunction with disease manifestations of different quality and intensity would translate into changes in T cell functions. We therefore set out to longitudinally investigate the cytokine expression profiles of patients who were mildly or severely affected by COVID-19 (Figures 4A, B). In order to assess the functional capacity for cytokine production by T cells, we stimulated PBMC from our cohort for 72 hours with CD3 and CD28 mAbs and then performed intracellular cytokine staining of CD3 gated T cells. Interestingly, we did not observe any significant difference in the expression of the Th1, Th2, Th17 and Th_GM-CSF, signature cytokines, which identify T helper cell subsets, i.e. IFN-γ, IL-4, IL-17A and GM-CSF respectively (26, 61, 62). This was the case for all timepoints 2, 4, and 6 weeks after infection (Figure 4A). No significant changes across the different timepoints were observed for mildly or severely affected patients. We also assessed the cytokine profile of CD8⁺ gated T cells. TNF-α, IL-2, granzyme B and IL-10 expression levels, which characterize cytotoxic T cell functionalities, did not differ between mildly and severely affected patients at any time-point. Expression levels were within the physiological range of interindividual variation and therefore suggested absence of any functional abrogation in T cell cytotoxicity in our patient cohort (Figure 4A).

The degree of cytokine co-expression has previously been reported to correlate with the efficacy of effector functions and to have an important bearing on protection (63, 64). We next assessed the polyfunctionality of CD4⁺ T helper and CD8⁺ cytotoxic T cells based on cytokine coexpression patterns. Interestingly, no significant difference in the fraction of single, double, triple or quadruple cytokines was observed at any time-point despite the divergent disease course between the two patient groups. This also applied to the type of overlapping cytokines (Figure 4B).

Taken together, this indicates that bulk T helper and cytotoxic T cell populations display comparable effector functions with respect to cytokine production after a mild versus severe COVID-19 disease course in the absence of comorbidities and do not leave an immunologic footprint indicative of the preceding disease course.

To corroborate our findings, we also assessed patients categorized according to their hospitalization status (Figure S9). Hospitalized patients displayed significantly upregulated IL-10 expression compared to non-hospitalized patients at early (TP1) and late (TP3) timepoints by cytotoxic T cells with a similar trend at TP2. T helper cells from hospitalized patients showed higher IL-4 expression at the late timepoint (TP3) (Figure S9). Longitudinal analysis across different timepoints within the hospitalized and non-hospitalized patient groups revealed stable cytokine expression levels (Figure S9). This indicates that patient stratification with an emphasis for highlighting the most severely affected patient group unmasks cytokine alterations. They demonstrate anti-inflammatory features in more severely affected COVID-19 patients.

Serum cytokines have previously been profiled extensively in various diseases to identify biomarkers and predictive markers of disease progression. Increases of systemic cytokine concentrations usually indicate very strong aberrations of immunity since cytokines act as local mediators. Recent investigations in COVID-19 patients have reported elevated levels of inflammatory mediators such as IL-6, which suggests a contribution of this and other cytokines as well as their innate cell sources to the systemic cytokine response that is characteristic of the COVID-19 acute respiratory distress syndrome (ARDS) (65, 66).

So far, most studies have been conducted in hospitalized patients, leaving systemic cytokine signatures in convalescent patients largely unexplored. Importantly, no information on serum cytokines is available to date in a COVID-19 cohort without comorbidities that could bias serum analysis. We therefore decided to profile serum cytokines in convalescent patients longitudinally to identify if and how long immune cytokine responses of a particular quality would be detectable systemically and whether they would correlate with previous disease severity (Figure 5).

We profiled 22 cytokines in the serum of mildly and severely affected convalescent patients at three distinct time-points (2, 4, 6 weeks after infection) and assessed serum cytokines with a multiplex bead array system (Bio-Plex). Interestingly, we did not observe any significant difference in serum cytokine concentrations between the mildly and more severely affected COVID-19 cohort at any timepoint for any of the cytokines (Figure 5). No significant variation was observed across different timepoints within mildly or severely affected patients. This was corroborated across disease severity scores for the early time-point (Figure S10). These data were unexpected considering the divergent range of disease symptoms covering asymptomatic and hospitalized patients prior to serum analysis. However, transient early changes in innate cytokine release could have potentially masked differences between the patient groups upon analysis at the convalescent stage.

Convalescence was overall associated with rather uniform immune signatures across a wide range of disease severities. This was corroborated using several different disease severity scoring strategies such as disease symptom severity scores, hospitalization status as well as the WHO Ordinal scale for Clinical Improvement (Figures S11, S12). Although significant alterations in the distribution of immune cell subsets and T cell differentiation states were observed early after disease resolution, they did not persist for longer than two weeks in most cases.

Finally, we used RNA-seq to assess global gene expression differences between the memory T-cells of three random patients with mild COVID-19 disease course and two with severe COVID-19 disease course (patient characteristics in Tables S1, S2). We generated an average of ~4 million read pairs for each patient (see Table S3) and successfully mapped ~90% of them to the human genome, quantifying 16,282 (protein-coding) genes. PCA on the expression values of these genes revealed that the patients clustered according to the severity of the disease (Figure 6A; Materials & Methods). In particular, the first principal component (PC1) separated the patients with a mild COVID-19 course from those with a severe course and accounted for ~40% of the variance in the data. Surprisingly, we only identified 31 differentially expressed genes (DEGs) between the two patient groups (Table S3; Materials and Methods), with all of them (30) being downregulated in the more mildly affected patients (Figures 6B, C). DEGs were mainly associated with cell cycle processes such as regulation of mitotic cell cycle phase transition and regulation of cell cycle phase transition (Figure 6E), indicating a less vigorous expansion of memory T-cell in more mildly affected COVID-19 patients. Of note, OASL was significantly upregulated in severely affected COVID-19 patients (Figure 6C). This bioinformatic transcriptome analysis with few patient samples was further validated by targeted qRT-PCR in more patients. IFN-inducible oligoadenylate synthetases-like (OASL) protein has previously been reported to have antiviral activity by mediating RIG-I activation, thus contributing to host antiviral responses (67). Likewise, antiviral activity has previously been attributed to interferon Alpha inducible Protein 27 (IFI27) (68), which was significantly upregulated upon transcriptomic analysis (Figure 6C). Comparison of three further mildly versus severely affected patients by qRT-PCR analysis showed a tendency for IFI27 upregulation, which did, however, not reach statistical significance (Figure 6D). A role of these molecules as potential biomarkers for previously mildly versus severely affected patients at the convalescent stage needs to be investigated in more detail in the future.