Patients diagnosed with COVID-19 and admitted to the COVID-19 ward of the IRCCS Sacro Cuore Don Calabria Hospital between March and April 2020 were consecutively included. Demography, clinical characteristics and laboratory findings upon admission were retrieved from electronic medical records ( Tables 1 , 2 ). Patients were classified into three categories of severity based on a modified WHO score (20) to take into account both the type of ventilation administered and the lung function expressed through the Horowitz index (PaO₂/FiO₂ ratio - arterial oxygen partial pressure to inspired oxygen fraction). Patients were thus classified as follows: mild (score 4), if no oxygen therapy was administered or PaO₂/FiO₂ ≥300; moderate (score 5), if oxygen supplied by mask or nasal prongs, or PaO₂/FiO₂ between 150–299; severe (score≥6), if oxygen administered by NIV, high flow or intubation, or PaO₂/FiO₂ <150.

All patients signed written informed consent. The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Ethical Committee of Verona and Rovigo provinces under protocol no. 63471/2020.

Whole blood and serum samples used for experimental analyses were collected upon admission and stored at -80°C until further use.

COVID-19 diagnosis was performed by reverse-transcriptase real time PCR (RT-qPCR) on nasopharyngeal swab according to WHO guidelines (available at https://www.who.int/publications/i/item/diagnostic-testing-for-sars-cov-2) and applying the CDC 2019-nCoV rRT-PCR Diagnostic Panel assay and protocol (available at https://www.fda.gov/media/134922/download) (21). IgG anti-SARS-CoV2 nucleocapsid protein (IgG-N) and IgM anti-SARS-CoV2 spike protein (IgM-S), were measured in patients’ serum using a chemiluminescent microparticle immunoassays (CMIA) (Abbott, Ireland) following manufacturer’s instructions. Results were expressed as assay index, i.e. sample relative light unit (RLU)/calibrator RLU. The assay was considered positive for indices >1.4 or ≥1 for IgG-N and IgM-S, respectively.

Flow cytometry analyses were performed on whole blood samples collected in EDTA, aliquoted and stored at -80°C in 10% DMSO (v/v). The list and concentration of mouse anti-human monoclonal antibodies employed (BD Biosciences, San Jose, CA, USA) are reported in Table S1 . Samples were prepared as recommended elsewhere (22). Briefly, three test tubes containing different combinations of fluorescence-labelled antibodies were used for each patient. One hundred and twenty (120) μL of whole blood were added to each panel-tube, vortexed and incubated at room temperature for 20 minutes in the dark. Erythrocytes were lysed and sample fixed with 1X FACS Lysing solution (BD Bioscience) for 10 minutes in the dark. Samples were washed twice with PBS + 0.5% BSA (w/v) and cell pellet was finally suspended in 200 μL of PBS + 0.5% BSA. Data acquisition was performed using a CytoFlex flow cytometer (Beckman Coulter, Brea, CA, USA) with the CytExpert software v2.3 (Beckman Coulter). The stopping rule was set at 100000 events in the CD45⁺ gate or, when this criterion could not be met due to severe leukopenia, the entire sample was acquired. The number of cells/µL was determined by the instrument based on the volume of sample consumed. The selection of surface markers and the gating strategy employed are reported in Figure S1 . Data were analysed with the Kaluza software v2.1 (all from Beckman Coulter, Brea, CA, USA).

The serum concentration of IFN-α, IFN-γ, IL-2, IL-4, IL-5, IL-6, IL-9, IL-10, IL-12p70, IL-17A, TNF-α, GM-CSF was measured using the MACSPlex Cytokine 12 kit (Miltenyi Biotec). Samples were prepared as recommended by the manufacturer. Briefly, samples were centrifuged at 10000g for 5 minutes at 4°C to remove large debris and the supernatant was then diluted 1:4 with sample buffer provided within the kit. Data were acquired on a CytoFlex flow cytometer (Beckman Coulter) at a 20 µL/min flow rate. Acquisition stopping rule was set at 4000 events in the bead gate or 180µl of acquired sample. Exported data were analysed with Flowlogic software (Inivai Technologies) and the median intensity in APC was used to extrapolate cytokine concentrations. To all samples having concentration values out of range (OOR), an arbitrary value corresponding to half of the lowest measured concentration was assigned.

Statistical analyses were performed using STATA software v14.0 (StataCorp LP, TX, USA) or SAS EG v7.1 (SAS Institute Inc., NC, USA), and plots generated with GraphPad Prism v8.3.0 (GraphPad Software, CA, USA). Non-parametric tests were applied according to data distribution. Differences in cell or cytokine concentration were assessed using the Mann-Whitney U test or the Kruskal-Wallis test followed by Dunn’s post-test and Bonferroni correction for multiple comparisons. The Spearman coefficient was used to evaluate correlations. Significance level was set at p-value <0.05 and all tests were two-tailed. A linear regression model was used to investigate the effect of disease severity on T cell subsets and cytokines, after log transformation of variables. Significant univariable models were then adjusted for the effects of covariates of interest in multivariable linear regression models. The ability of circulating cells and cytokines to predict the probability of clinical aggravation was explored using Firth logistic regression. Significant regressors (p-value from univariable analyses <0.2) were dichotomized using Receiver Operating Characteristic (ROC) curve cut-off analysis and included in multivariable Firth logistic regression models. Cut-offs were defined maximizing the sum of sensitivity and specificity (Youden’s J statistic) ( Table S6 ). Areas Under the Curve (AUC) were compared by the DeLong test. Goodness-of-fit measures were estimated for these models.

The demographic and clinical characteristics of COVID-19 patients classified as mild, moderate or severe are reported in Table 1 . No differences in gender frequencies were observed across the three groups, while a small, although significant, difference in patients’ age was observed, with severe and moderate patients being older than mild patients. Considering clinical information, the frequency of subjects presenting with fever was comparable between the three groups. Comorbidities were significantly more frequent among patients suffering from severe or moderate COVID-19 (89% for both groups) compared to the mild form (48%), with cardiovascular diseases being the most common co-morbidity. The number of patients receiving treatment was the same across the three groups and the majority of them were receiving either hydroxychloroquine or a combination of hydroxychloroquine and antivirals, as a reflection of standard of care at the time of the study. The frequency of subjects worsening during hospitalisation or having a fatal outcome was significantly higher among severe patients (89% and 56%, respectively) compared to the other groups.

The principal laboratory findings obtained on blood samples collected at the same time as those tested here are reported in Table 2 . In agreement with already reported data, disease severity was associated with a general leucocytosis and lymphopenia (23). Severe patients presented a raised concentration of inflammatory markers, i.e., C-reactive protein (CRP) and IL-6, while other markers proposed to be related with SARS-CoV-2 infection severity, including ferritin, D-dimer and angiotensin-converting enzyme were not. Blood creatinine was also higher in severe disease compared to milder forms. No differences in the viral load, expressed as RT-qPCR Ct value, were observed between the three groups. The frequency of patients with positive IgM-S and IgG-N antibodies did not differ across the three groups.

The gating strategy adopted to determine the frequency of leucocytes and lymphocyte subsets is reported in Figure S1 . A marked reduction in the absolute number of total lymphocytes and T cell subsets (CD4⁺, CD8⁺ T, Th1, Th17 and Treg) was recorded in severe COVID-19 patients compared to mild or moderate subjects, while the number of granulocytes was significantly increased in severe patients ( Figure 1 and Table S2 ).

A multivariate linear regression analysis confirmed the significant association between disease severity and the amount of the different lymphocytes, independently of the effect of age, gender, presence of fever or comorbidities, and ongoing treatment, considered as potential effect modifiers ( Table 3 ). In addition, patients’ gender showed a significant association with CD4⁺ T cells, while the presence of fever and comorbidities were associated with Th1 variability.

A correlation matrix was computed to assess whether lymphocyte immuno-phenotype shows a linear relation with clinical and biochemical parameters, independently of disease severity ( Figure 2A ). The Horowitz index showed a significant correlation (p<0.01) with total lymphocytes and CD3⁺ T cells (moderate positive correlation), as well as with CD4⁺ and CD8⁺ T cells, Th1, Th17 and Treg (weak positive correlation), indicating a relation between a reduced lung function and reduced circulating cells of the T compartment. Significant negative correlations were observed between inflammatory markers (CRP, ferritin and IL-6) and the majority of the cell subsets analysed ( Figure 2A ). In particular, the most relevant correlations having a Spearman r coefficient < -0.6 were observed for CRP and CD4⁺ T cells, Th1, Th17 or Treg ( Figure 2B ). Finally, total lymphocytes, T cells, CD4⁺, Th1, Th17 and Tregs showed a significant relation with RT-qPCR Ct values, pointing out that higher viral loads are accompanied by a lower absolute count of T cell subsets ( Figure 2A ). Considering anti-SARS-CoV-2 antibodies, the absolute number of the different cell types analysed did not differ between patients with negative or positive serology, neither for IgM-S nor for IgG-N antibodies ( Table S3A ). Moreover, in patients with positive serology, the cell count did not correlate with IgM-S or IgG-N indices ( Table S3B ).

The systemic levels of selected cytokines were measured in serum samples collected on the same day as blood samples used for the immuno-phenotype analysis. IL-2 and IL-9 were excluded from further analyses as more than 90% of the measured samples had OOR values. The systemic concentration of all measured cytokines did not vary according to COVID-19 severity ( Table S2 ) however, when evaluated at the individual level, it was possible to identify within each severity group, clusters of patients displaying raised levels of all or most of the assessed cytokines ( Figure S2 ). Among the measured cytokines, only a few correlated significantly with the number of circulating lymphocyte populations, although the strength of the correlation was overall weak ( Table S4 ). Only IL-6 and CD4⁺ T cells showed a moderate negative correlation (Spearman rho = -0.54, p-value <0.0001).

To evaluate the potential association between our experimental data and patients’ clinical course, our cohort was re-classified according to the progression of patients’ status as improved or worsened compared to the clinical conditions established on the day of sample collection. Clinical aggravation was determined as an increased requirement of oxygen compared to baseline or death during hospitalisation. On this base, we included 37 patients with an improved clinical course and 23 whose conditions worsened. Amongst the latter, 43.5% ultimately died from COVID-19.

Total lymphocytes, B and T cells and T subsets (except for CD8⁺ T cells), were significantly reduced in worsening patients; CD4⁺ T cells and Th1 cells showed the strongest differences between the two groups (p ≤ 0.0006) ( Figure 3A and Table S5 ). Additionally, worsening patients displayed significantly higher systemic concentrations of IFN-α (p value=0.0175) ( Figure 3B and Table S5 ). Among the clinical and biochemical parameters already proposed as associated with disease severity, only the viral load, blood creatinine and PaO₂/FiO₂ ratio were able to differentiate between the two groups ( Figure 3C and Table S5 ).

The multivariable logistic regression analysis revealed that subjects harbouring decreased CD4⁺, Th1, Th2 and Tregs have a significantly increased risk of worsening during the hospitalisation, after adjusting for gender, presence of comorbidity and the administration of treatment during hospitalisation ( Table 4 ). In particular, patients with baseline CD4⁺ T cells ≤ 136.7 cells/µL have 6.5 times the odds of progression compared to subjects with higher count. Similarly, Th1 count ≤ 18.34 cells/µL, Th2 count ≤ 5 cells/µL and Treg ≤ 30 cells/µL are associated with increased risk of 7.9, 4.4 and 6.8 times, respectively. Similarly, patients with baseline PaO2/FiO2 ratio ≤186 have significantly higher odds of worsening. The multivariable analysis also confirmed a significantly higher risk for men to undergo clinical aggravation compared to women. The combination of gender and Tregs through a ROC curve selection analysis significantly improved the discriminatory ability of gender, with AUC increased from 0.75 to 0.82 (p=0.044) ( Figure 4 ).