A total of 42 COVID-19 patients in this retrospective cohort study were enrolled from Beijing Ditan Hospital from March 13, 2020 to April 25, 2020. All enrolled patients were confirmed to be infected with SARS-CoV-2 by RT-PCR assays. This study was approved by the Committee of Ethics at Beijing Ditan Hospital, Capital Medical University. M/M patients were mild and moderate patients, and S/C patients were severe and critical patients, according to the guidelines on the diagnosis and treatment of new coronavirus pneumonia (version 7) by the National Health Commission of China issued on March 3, 2020. These 42 patients included 32 mild or moderate (M/M) patients and 10 severe or critical (S/C) patients. All baseline medical record information including demographic data and clinical characteristics were obtained within the first day after hospital admission ( Table S1 ). Blood samples were first collected within 3 days of the hospital admission and once every 3–7 days during hospitalization. The median age of the patients was 37 years (range 20–75) with 50% men and 50% women. Among these 42 patients, the most common were hypertension (five cases), diabetes (three cases), chronic pulmonary disease (three cases), chronic kidney disease (one case), cardiovascular disease (one case). Other clinical details are shown in Table S1 .

This study was approved by Committee of Ethics at Beijing Ditan Hospital, Capital Medical University [NO. JDLKZ (2020) D (036)-01] with informed consents acquired from all enrolled patients. This study complied with all relevant ethical regulations for work with human participants, and informed consent was obtained. Samples were collected from patients who provided informed consent to participate in the study.

The PBMCs were collected in EDTA at the indicated time points. PBMCs were separated by density gradient centrifugation with lymphocyte separation solution. Serum samples were collected in serum separation tube. The blood was centrifuged at 2,000 rpm for 10 min at 20°C, and the serum was stored at −80°C and thawed at the time of assays. All samples were processed and analyzed within 24 h of collection.

PBMCs were incubated with directly conjugated antibodies for 30 min at 4°C. The cells were then washed before flow cytometric analysis. Antibodies used were anti-human CD3-BUV737, CD4-BUV395, PD-1-BV711, CD38-FITC, GITR-BV605 (BD Biosciences, San Diego, CA, USA), CD8-BV510, CTLA-4-BV786, OX40-APC-Fire750, 4-1BB-BV421, HLA-DR-AF700 (BioLegend, San Diego, CA, USA), TIGIT- PE-Cy7, LAG-3-APC, ICOS-PE, (Ebioscience, San Diego, CA, USA), and the corresponding isotype controls. Data acquisition was performed on an LSR Fortessa flow cytometer (BD Biosciences), and data analysis was performed using FlowJo Software (Tree Star, Ashland, OR, USA).

PBMCs, isolated as described above, were resuspended to 1×10⁶ cells/ml in PBS. The cells were surface-stained with CD3-BV786, CD38-BUV737, HLA-DR-PE, CCR7-BV421, CD45RA-AF700, CD71-APC-H7 (BD), CD4-APC-Fire750, CD8-BV510 (BioLegend) for 30 min in the dark at 4°C, followed by fixation and permeabilization. After permeabilization, cells were stained with ki67-FITC, Granzyme B-AF700, T-bet-BV421, BAX-FITC, Bcl2-PE (BD Biosciences), Eomes-PE-Cy7 (Ebioscience), perforin-APC (BioLegend) antibodies for 30 min in the dark at room temperature. Following staining, cells were washed and acquired on an LSRFortessa.

The serum of 27 COVID-19 patients were assayed for the two multiplexed bead immunoassays. First, we tested 45 ProcartaPlex Human Cytokine/Chemokine/Growth Factor Panel (Invitrogen, Calsbad, CA, USA), including BDNF, Eotaxin/CCL11, EGF, FGF-2, GM-CSF, GROα/CXCL1, HGF, NGFβ, LIF, IFNα, IFNγ, IL-1β, IL-1α, IL-1Rα, IL-2, IL-4, IL-5, IL-6, IL-7, IL-8/CXCL8, IL-9, IL-10, IL-12p70, IL-13, IL-15, IL-17α, IL-18, IL-21, IL-22, IL-23, IL-27, IL-31, IP-10/CXCL10, MCP-1/CCL2, MIP-1α/CCL3, MIP1β/CCL4, RANTES/CCL5, SDF-1α/CXCL12, TNFα, TNFβ/LTA, PDGF-BB, PLGF, SCF, VEGF-A, and VEGF-D. Second, we tested the 14 ProcartaPlex Human ImmunoOncology Checkpoint Panel (Invitrogen), including BTLA, GITR, HVEM, IDO, LAG-3, PD-1, PD-L1, PD-L2, TIM-3, CD28, CD80, 4-1BB, CD27, and CD152. All data were acquired according to the manufacturer’s protocol using Luminex MAGPIX^® instrument (Luminex Co., Austin, TX, USA) and analyzed using ProcartaPlex Analyst 1.0 software (Invitrogen).

GraphPad5 (GraphPad Software, La Jolla, CA, USA) or SPSS (IBM Corporation, New York, NY, USA) was used for statistical calculations. Data are expressed as the mean ± standard deviation (SD) and percentage (frequency), and the normality of each variable was evaluated using the Kolmogorov–Smirnov test. In cases of two normally distributed data, the comparison of variables was respectively performed using unpaired or paired two-tailed Student’s t tests. One-way ANOVA test followed by Tukey’s multiple comparisons test or Holm-Sidak’s multiple comparisons test was performed for comparing two more independent or matched samples. When the data were not normally distributed, the comparison of variables was performed with a Mann–Whitney U test or a Wilcoxon matched-pairs signed-rank test for unpaired and paired data, respectively. For comparing two more samples, a Kruskal–Wallis test or Friedman test followed by Dunn’s multiple comparisons test was applied for independent and matched samples. Comparisons of patient characteristics were analyzed using Fisher’s exact test (categorical variables) or Kruskal–Wallis test (continuous variables). P and correlation coefficient values were obtained using the Spearman’s correlation test. For all analyses, P values <0.05 were considered statistically significant.

We recruited 42 confirmed COVID-19 patients who received care at Beijing Ditan Hospital. The clinical courses of these cases included 32 mild or moderate (M/M) and 10 severe or critical (S/C) cases have been described in Table S1 . Twenty age- and gender-matched healthy donors (HDs) were enrolled as controls. Blood samples were first collected within 3 days of the hospital admission and once every 3–7 days during hospitalization. The study was approved by the Committee of Ethics at Beijing Ditan Hospital, Capital Medical University, Beijing, China.

To determine the activated status of CD8⁺ T cells, we first analyzed co-expression of HLA-DR and CD38, which are key markers of CD8⁺ T cell activation during viral infection. Based on expression of CD38, three subpopulations were defined among activated HLA-DR⁺CD8⁺ T cells: HLA-DR⁺CD38⁻ (fraction I), HLA-DR⁺CD38^dim (fraction II), HLA-DR⁺CD38^hi (fraction III), as shown in Figure 1A . Patients of S/C group were found to have an obvious peak in HLA-DR⁺CD38^hi CD8⁺ T cell (fraction III) within 2–3 weeks post onset ( Figures S1A, B ). We then observed a significantly higher percentage of HLA-DR⁺CD38^hi CD8⁺ T cells in all patients including M/M and S/C groups at the peak point, compared to healthy controls. Moreover, the percentage of HLA-DR⁺CD38^hi CD8⁺ T cells was dramatically higher in S/C patients than M/M patients (23.48 vs. 3.203%, Figure 1B ). HLA-DR⁺CD38^dim CD8⁺ T cells showed a similar trend but less increase. Consequently, the ratio of HLA-DR⁺CD38^dim to HLA-DR⁺CD38^hi CD8⁺ T cells was significantly higher in M/M than S/C COVID-19 patients in CD8 T cells ( Figure S2 ). In contrast, no significant difference was found in HLA-DR⁺CD38⁻ subset among three groups.

We further examined the longitudinal flow cytometric data in eight S/C and eight M/M cases. S/C patients developed elevated HLA-DR⁺CD38^hi CD8⁺ T cells early in the infection and displayed persistently high percentage of this population (peak 43%) during the whole course of hospitalization. In contrast, percentage of HLA-DR⁺CD38^hi CD8⁺ T cell in M/M cases increased slightly and transiently in the course of illness. As expected, there were no differences in kinetics of HLA-DR⁺CD38⁻ and HLA-DR⁺CD38^dim CD8⁺ T cells between S/C and M/M patients ( Figure 1C ). Furthermore, we combined all flow data of each patient (32 M/M and 10 S/C cases) and plotted their fluctuation patterns against the time point post onset. Consistently, these aggregating data showed that percentage of HLA-DR⁺CD38^hi CD8⁺ T cells rather than the other two subsets was persistently higher in S/C patients than in M/M cases during hospitalization ( Figure 1D ). Meanwhile, we observed no significant changes of HLA-DR⁺CD38^hi CD4⁺ T cells in S/C and M/M patients with COVID-19 infection ( Figure S3 ). Overall, our data showed that persistent accumulation of HLA-DR⁺CD38^hi CD8⁺ T cells was associated with severity of COVID-19.

Next, we applied the longitudinal data of all patients and analyzed the correlation between dynamic changes of circulating HLA-DR⁺CD38^hi CD8⁺ T cells and laboratory parameters ( Table 1 ). We found percentage of HLA-DR⁺CD38^hi CD8⁺ T cells was negatively correlated with absolute counts of lymphocytes, total T cells, CD4, CD8 T cells, B cells, and NK cells, but not neutrophil and monocyte counts. We also observed significant negative correlations between percentage of HLA-DR⁺CD38^hi CD8⁺ T cells and hemoglobin (R=−0.546, P<0.0001). Additionally, coagulation-related parameters including platelet count, D-dimer, prothrombin time (PT), and activated partial thromboplastin time (APTT) were detected. We showed that percentage of HLA-DR⁺CD38^hi CD8⁺ T cells was significantly associated with D-dimer, which was indicated to correlate with COVID-19 severity (R=0.452, P=0.0003).

We further found positive correlations between HLA-DR⁺CD38^hi CD8⁺ T cells and levels of C-response protein (CRP) and serum amyloid A (SAA), suggesting systemic inflammation (R= 0.475, P<0.0001; R=0.565, P<0.0001). Moreover, the percentage of HLA-DR⁺CD38^hi CD8⁺ T cells was found to have positive correlations with aspartate transaminase (AST) and total bilirubin (TB) in COVID-19 patients (R=0.397, P<0.0001; R=0.398, P<0.0001), and a strong negative correlation with albumin (R=−0.481, P<0.0001). These data suggested that high percentage of HLA-DR⁺CD38^hi CD8⁺ T cells was involved in liver injury induced by COVID-19. Consistently, levels of lactate dehydrogenase (LDH) and creatinine (CRE) were correlated with the percentage of HLA-DR⁺CD38^hi CD8⁺ T cells, respectively (R=0.643, P<0.0001; R=0.354, P=0.0003), indicating myocardial and renal injury ( Table 1 ). HLA-DR⁺CD38^dim and HLA-DR⁺CD38⁻ CD8 T cells showed no correlation with the clinical characteristics above. Collectively, these results suggested the involvement of HLA-DR⁺CD38^hi CD8⁺ T cells in immune disorders and tissue injury in COVID-19 patients.

To assess the phenotypic status of HLA-DR⁺CD38^hi CD8⁺ T cells from COVID-19 patients, we performed additional stains on selected 20 samples from 18 patients. We determined the developmental stage of HLA-DR⁺CD38^hi CD8⁺ T cells through dissecting T cells into naïve (T_N: CD45RA+, CD27+, CCR7+), central memory (T_CM: CD45RA−, CD27+, CCR7+), transitional memory (T_TM: CD45RA−, CD27+, CCR7−), effector memory (T_EM: CD45RA−, CD27−, CCR7−), and effector T cells (T_E: CD45RA+, CD27−, CCR7−). HLA-DR⁺CD38^hi CD8⁺ T cells consisted of enhanced percentage of T_TM, constant proportion of T_CM and T_EM, and decreased T_N and T_E ( Figure 2 ).

Consistent with activation, HLA-DR⁺CD38^hi CD8⁺ T cells (fraction III) displayed significantly higher levels of CD69, an early activation marker, compared with fractions I and II. We also evaluated expression of costimulatory molecules, including inducible T-cell costimulator (ICOS), OX40, TNF receptor superfamily member 9 (4-1BB), and glucocorticoid-induced tumor necrosis factor receptor (GITR). HLA-DR⁺CD38^hi CD8⁺ T cells showed elevated expression of ICOS, OX40, 4-1BB, and GITR compared to fraction I. The level of OX40 in fraction III was higher than in fraction II, while fraction II expressed the highest levels of 4-1BB and GITR ( Figure 3 ). Interestingly, we also found HLA-DR⁺CD38^hi CD8⁺ T cells expressed higher levels of numerous coinhibitory molecules, including Programmed Death-1 receptor (PD-1), T cell immunoglobulin and mucin-domain containing-3 receptor (TIM-3), LAG-3 (Lymphocyte Activating 3), compared to fractions I and II. No significant difference was found in TIGIT expression of these three fractions. To investigate the intrinsic regulation of HLA-DR⁺CD38^hi CD8⁺ T cells, we examined the expression of T-bet and Eomesodermin (Eomes), two key transcription factors governing CD8⁺ T cell exhaustion. We found that HLA-DR⁺CD38^hi CD8⁺ T cells contained higher percentage of T-bet^dimEomes^hi cells, which represented a terminal exhausted status, than fraction I and II. Meanwhile, these three fractions had comparable T-bet^hiEomes^dim cells ( Figure 4 ).

Consistent with higher frequency of terminally differentiated cells, HLA-DR⁺CD38^hi CD8⁺ T cells showed elevated BAX expression and decreased Bcl-2 expression compared with fraction I and II, indicative of high susceptibility to apoptosis. We next investigated killing potential of HLA-DR⁺CD38^hi CD8⁺ T cells using intracellular staining of granzyme B and perforin, which are responsible for cytotoxic T lymphocytes to exert their killing function. Distinct from classical exhausted T cells, HLA-DR⁺CD38^hi CD8⁺ T cells showed no significant changes of granzyme B and perforin intracellular staining compared with fraction I. Meanwhile, HLA-DR⁺CD38^dim CD8⁺ T cells exhibited the highest levels of granzyme B and perforin. Subsequently, we further confirmed that HLA-DR⁺CD38^hi CD8⁺ T cells are highly proliferative by expressing higher levels ki67 and CD71 ( Figure 5 ). In all, these data suggested an overactivated and consequently disordered immune status of HLA-DR⁺CD38^hi CD8⁺ T cells during acute COVID-19 infection.

In our previous study, high levels of cytokines and soluble checkpoint molecules were reported to correlate with S/C illness of COVID-19 (25, 26). Due to the disordered immune status of HLA-DR⁺CD38^hi CD8⁺ T cells, we wondered the effects of these cells in the storms of cytokines and soluble checkpoint molecules. We evaluated serum levels of 45 cytokines/chemokines/growth factors and 14 soluble checkpoints using Luminex multiplex assay from 27 COVID-19 patients at different time points during hospitalization and collected longitudinal data. Levels of 17 factors such as HGF, IL-18, IL-1RA, MCP-1, RANTES, IL-10, and SCF showed significantly positive correlations with percentage of HLA-DR⁺CD38^hi CD8⁺ T cells. Moreover, 10 serum soluble checkpoint molecules, such as TIM3, CD27, IDO, and LAG3, were positively correlated with percentage of HLA-DR⁺CD38^hi CD8⁺ T cells. Moreover, two other populations HLA-DR⁺CD38⁻ and HLA-DR⁺CD38^dim showed no correlations to storm of cytokines and soluble checkpoint molecules ( Table 2 and Figure S4 ). Thus, we hypothesized that elevated HLA-DR⁺CD38^hi CD8⁺ T group might potentially contribute to the storm of cytokine and soluble checkpoint molecules occurring in COVID-19 patients.