A unique cytotoxic CD4+ T cell-signature defines critical COVID-19

Abstract

Objectives

SARS-CoV-2 infection causes a spectrum of clinical disease presentation, ranging from asymptomatic to fatal. While neutralising antibody (NAb) responses correlate with protection against symptomatic and severe infection, the contribution of the T-cell response to disease resolution or progression is still unclear. As newly emerging variants of concern have the capacity to partially escape NAb responses, defining the contribution of individual T-cell subsets to disease outcome is imperative to inform the development of next-generation COVID-19 vaccines.

Methods

Immunophenotyping of T-cell responses in unvaccinated individuals was performed, representing the full spectrum of COVID-19 clinical presentation. Computational and manual analyses were used to identify T-cell populations associated with distinct disease states.

Results

Critical SARS-CoV-2 infection was characterised by an increase in activated and cytotoxic CD4⁺ lymphocytes (CTL). These CD4⁺ CTLs were largely absent in asymptomatic to severe disease states. In contrast, non-critical COVID-19 was associated with high frequencies of naïve T cells and lack of activation marker expression.

Conclusion

Highly activated and cytotoxic CD4⁺ T-cell responses may contribute to cell-mediated host tissue damage and progression of COVID-19. Induction of these potentially detrimental T-cell responses should be considered when developing and implementing effective COVID-19 control strategies.

Introduction

The severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) pandemic has been ongoing since March of 2020. As of February 2023, over 754 million cases of SARS-CoV-2 infection and 6.83 million fatalities from coronavirus disease 2019 (COVID-19) have been reported.¹ While several vaccines are now available for use, SARS-CoV-2 remains a leading cause of infectious disease death globally. One of the major challenges with SARS-CoV-2 infection is the spectrum of COVID-19 clinical presentation, ranging from asymptomatic to fatal. It is thought that more severe disease results from a dysregulated immune response to infection; however, variability in this immune dysfunction between individuals has limited understanding of the correlates of disease severity. Developing a more comprehensive understanding of the immune response across the spectrum of COVID-19 clinical presentation will help to differentiate protective from pathogenic immune responses. This is essential to inform the development of next-generation therapies and vaccines against SARS-CoV-2, with improved longevity and efficacy against newly emerging variants of concern (VOC).

The key correlate of protective immunity against infection and severe disease in COVID-19 is neutralising antibody responses (NAb).^2,3 As such, the factors that contribute to breakthrough infection following vaccination centre around humoral immune responses, such as waning NAb titres, and antibody escape mutations in the dominant VOC.^4–9 The T-cell response appears to have greater longevity than detectable NAbs, with sustained response to antigen stimulation demonstrated > 1-year post-infection.^10,11 Additionally, the dominant T-cell epitopes do not overlap with areas of high mutation on variant viruses, and as a result the T-cell response is preserved against antibody-escape VOC.^9,12–16 Considering the limitations of current vaccines, it has been suggested that the long-lived T-cell response against SARS-CoV-2 variants may contribute to protective immunity in the absence of a robust humoral immune response.^12,13,16 While NAb responses have been shown to tightly correlate with protection against disease, no such correlation has been shown with the T-cell response to SARS-CoV-2.¹⁷ There is evidence that polyfunctional and cross-reactive T-cell responses to seasonal coronaviruses are associated with milder disease and faster viral clearance.^18–20 However, several studies have also described an expansion of highly activated T cells in severe COVID-19 that could potentially contribute to excessive inflammatory immune responses and host-tissue damage.^21–23 As such, whether T cells play a protective or pathogenic role in COVID-19 is still unresolved.

To better define the role of T-cell subsets, we performed an explorative investigation into T-cell phenotypes across the clinical spectrum of COVID-19 presentation, utilising an unbiased analysis approach with a T-cell-centric high-dimensional cytometry panel. We report that critical COVID-19 infection is characterised by a shift from naïve T-cell phenotypes to an expansion of cytotoxic CD4⁺ T lymphocyte subsets.

Results The T-cell compartment distinguishes critical SARS-CoV-2 infection from other disease states

To assess the T-cell response in COVID-19 infection, two cohorts of patients, with PCR confirmed SARS-CoV-2 infection, were included in this study. Characteristics of the patient cohorts are shown in Supplementary tables 1 and 2. The COVIMM cohort was recruited in March 2020, when the Ancestral variant of SARS-CoV-2 was the dominant circulating variant in Sydney, Australia. The patients in the COVIMM cohort experienced asymptomatic and mild disease severity. The second cohort, COSIN, was recruited in June 2021, when the Delta variant of SARS-CoV-2 was the dominant circulating variant in Sydney, Australia. The COSIN cohort patients experience mild to critical disease severity. Thus across these two cohorts, the full spectrum of COVID-19 disease severity was represented.

To obtain a global view of the T-cell response within and between disease states, peripheral blood mononuclear cells (PBMCs) were isolated from the blood of patients and spectral cytometry was performed using a T cell-centric antibody panel. Initially, T-cell populations were manually gated (Supplementary figure 1) and differences in the proportion of each population between patients were identified by an unsupervised Principal Component Analysis (PCA), where each data point represents one patient sample (Figure 1a). When the COVID-19 severity of each patient was superimposed onto the PCA, where the ellipses indicate 95% confidence intervals, three patients with critical infection separated distinctly across the first component (dim 1, accounted for 28.3% of the variance) from most other patient samples (Figure 1a). Visualisation of the contribution of each T-cell population proportion to the principal components revealed that the expression of activation/proliferation markers HLA-DR, Granzyme B (GZMB), and Perforin (PFN) on central memory (T_CM), effector memory (T_EM) and effector memory re-expressing CD45RA (T_EMRA) contribute to the separation of samples in dimension 1 (Figure 1b). Furthermore, the proportions of these activated memory CD4⁺ and CD8⁺ T-cell subsets were negatively correlated to those of CD4⁺ and CD8⁺ naïve T cells (T_N) (Figure 1b).

Figure 1. Variance in the T-cell compartment explained by COVID-19 severity. (a) Principal component analysis (PCA) based on the relative abundance of 49 T-cell populations in patients with asymptomatic to critical COVID-19 (n = 36). Patient disease severity has been overlaid onto the PCA plot and ellipses represent 95% confidence intervals. (b) Variable contribution plot visualising the T-cell populations that contribute to the principal components. Arrow direction represents correlation, where opposing direction is negative and adjacent arrows represent positive correlation between variables. Differences between groups were determined by permutational multivariate analysis of variants.

Metaclusters of cytotoxic CD4⁺ T cells differentiate critical and severe SARS-CoV-2 infection

To fully capture the heterogeneity of activation and cytotoxic marker expression in the T-cell compartment, unbiased clustering was performed. FlowSOM clustering was set to create 25 metaclusters (Mcs) and fast interpolation-based t-SNE (Fit-SNE) was used to visualise the proportion of each Mc (Figure 2a and b), with heatmaps generated for phenotyping (Figure 2c and d). The T-cell panel included markers to define conventional CD4⁺ and CD8⁺ subsets, and so CD4⁻CD8⁻ Mcs (11 and 23) were excluded. The initial visualisation of metacluster proportions and phenotypes revealed a distinct population of GZMB⁺PFN⁺ CD4⁺ T cells which differed in proportion between severe and critical disease patients (Figure 2b and c). As the PCA identified significant variance between severe and critical disease patients, a Partial Least Squares Discriminant Analysis (PLS-DA) was performed to identify the variables (Mcs) that contribute specifically to the separation of these disease states. Like the PCA (Figure 1b), the PLS-DA revealed distinct separation of severe and critical patients, (dim 1 accounted for 39.7% of the variance; Figure 3a) and Mcs were ranked by their contribution to variance between disease states (Figure 3b). Mc13, Mc1 and Mc10 were enriched in critical infection and Mc16 and 21 in severe infection patients (Figure 3b and c). To confirm that the difference in these Mcs were not an artefact of the supervised nature of the PLS-DA, unpaired Mann–Whitney U-tests were performed to compare the proportions of each Mc between severe and critical disease patients. Mc13 (GZMB⁺PFN⁺ CD4⁺ CD45RO⁻, CCR7⁻), Mc1 (Ki-67⁺HLA-DR⁺PD-1⁺CD4⁺ CD45RO⁺CCR7⁻) and Mc10 (GZMB⁺PFN⁺Ki-67⁺ CD4⁺ CD45RO⁺CCR7⁻) were significantly enriched in critical compared to severe disease patients (Figure 3c). As visualised in the Fit-SNE plots in Figure 2, Mc13 and Mc10 accounted for the increased cytotoxic CD4⁺ T-cell populations in critical disease patients (Figure 3d). The proportion of Mc16 (CD4⁺ CD45RO⁻CCR7⁺) and Mc21 (CD8⁺ CD45RO⁻CCR7⁺) were significantly greater in severe disease patients than in critical disease patients, representing non-activated naïve (T_N) CD4⁺ and CD8⁺ T cells, respectively (Figure 3c). In addition to the expression of GZMB and PFN, markers of cytotoxic effector function, Mc13 contained cells expressing CXCR5, a feature of CD4⁺ T follicular helper (T_FH) cells (Figure 2c and d). As such, the presence of CD4⁺ T cells with a polyfunctional CXCR5⁺GZMB⁺PFN⁺ phenotype was of interest to investigate further. This analysis suggests an expansion of activated and cytotoxic CD4⁺ T-cell populations and a decrease in the proportion of T_N cell subsets is involved in progression from severe to critical SARS-CoV-2 infection.

Figure 2. Unbiased clustering of T-cell compartment in severe and critical patients. (a) FIt-SNE visualisation of FlowSOM automatic clustering of a subsample of T cells from each severe (n = 5) and critical (n = 3) patient with conventional T-cell population labels overlaid. (b) FIt-SNE visualisation of alterations in proportion of metaclusters making up the T-cell compartment between severe and critical disease. (c) Relative expression of cellular marker expression on FIt-SNE visualisation of FlowSOM automatic clustering of a subsample of T cells from each severe (n = 5) and critical (n = 3) patient. (d) Heatmap plot showing the relative expression of each marker on self-organised map metaclusters. The legend indicates the level of expression for each marker that is a normalised z-score between 0 and 1.

Figure 3. Comparative analysis of metacluster proportions in severe and critical patients. (a) Partial Least Squares Discriminant Analysis (PLS-DA) of relative proportions of 23 CD4+ and CD8+ T-cell population defined by automatic clustering, where ellipses represent 95% confidence intervals. (b) Christmas tree plot of the metaclusters contributing to the first component of the PLS-DA, with the greatest contributors at the bottom, and the x-axis indicates regression coefficient. (c) Non-parametric Mann–Whitney U-test of proportions of metaclusters between severe and critical infection patients; error lines represent median ± interquartile range. (d) Proportion of CD4+ T-cell metaclusters by expression of GZMB and PFN in severe and critical disease patients.

Cytotoxic CD4⁺ T lymphocytes are characteristic of disease progression

As the proportions of naïve and memory T cells were negatively correlated (Figure 1c), the distribution of naïve/memory subsets were further explored in the CD4⁺ T and CD8⁺ T-cell compartments by manual gating on CD45RO and CCR7 (Figure 4a). Critical patients exhibited increased proportions of CD8⁺ T_EMRA cells and a reduction in the proportion of CD4⁺ T_N cells, compared to mild and severe disease patients, respectively (Figure 4b). In contrast, the T-cell compartments in asymptomatic, mild, moderate and severe disease patients were composed of comparable proportions of T_N CD4⁺ and CD8⁺ T cells. While the proportion of CD4⁺ T_EMRA cells was not significantly elevated, critical patients had increased expression of the activation markers HLA-DR on CD4⁺ T_EMRA cells compared to those with asymptomatic or mild disease (Figure 4c). Expression of PD-1 was significantly elevated on CD4⁺ T_EMRA in patients with moderate disease but not in those with severe or critical COVID-19 (Figure 4c). As Mc13 exhibited intermediate expression of T_FH marker CXCR5, the frequency of T_FH cells (CD45RO⁺CXCR5⁺) between disease severities was compared. Patients with critical disease exhibited elevated frequencies of T_FH cells (Figure 4d).

Figure 4. Differentiation status of CD4+ and CD8+ T-cell compartments. (a) Summary bar plots representing proportion of CD4+ and CD8+ T-cell naïve and memory subsets between disease states defined as TN (CCR7+CD45RO−), TCM (CCR7+CD45RO+), TEM (CCR7−CD45RO+) and TEMRA (CCR7−CD45RO−). (b) Proportion of CD4+ and CD8+ TN and TEMRA subsets (c) Proportion of CD4+ TEMRA cells expressing HLA-DA and PD-1. (d) Proportion of CD4+ T cells expressing CXCR5 and CD45RO (TFH cells). Difference between groups were determined by the non-parametric Kruskal–Wallis test, with comparison of the rank mean of experimental groups by the Original FDR method; error bars represent median ± interquartile range.

The computational analyses in Figures 1 and 2 revealed distinct populations of cytotoxic (GZMB⁺PFN⁺) CD4⁺ T-cell subsets enriched in critical disease patients. As cytotoxic CD8⁺ T cells have been correlated with disease severity and mortality in COVID-19,²² the proportion of GZMB⁺PFN⁺ CD4⁺ and CD8⁺ subsets were of interest to investigate further. As a proportion of lymphocytes, critical patients had expanded populations of cytotoxic CD4⁺, but not CD8⁺ T cells. Cytotoxic CD4⁺ T cells made up a median of 2.08% of the lymphocyte compartment in critical infection patients, and between 0.29 and 1.25% in all other disease states (Figure 5a). To delineate the subsets of cytotoxic CD4⁺ T cells contributing to this, the proportions of GZMB⁺PFN⁺ T_FH and GZMB⁺PFN⁺ CD4⁺ T cells were compared across disease states. In both these CD4⁺ subsets there was an elevated proportion of cells co-expressing GZMB and PFN in critical disease patients (Figure 5b). As these cytotoxic CD4⁺ subsets were predominantly present in critical disease patients, these patients were selected to provide a qualitative breakdown of the lineage identity of cytotoxic T_FH and CD4⁺ T cells. As shown in Figure 5c, cytotoxic CD4⁺ T cells were primarily composed of T_EM and T_EMRA populations, whereas total CD4⁺ T cells  were largely of a T_N and T_CM phenotype (Figure 5c). Similarly, cytotoxic CD4⁺ T_FH cells were almost exclusively of a CCR7⁻PD-1⁺ phenotype and total T_FH cells were predominantly of a CCR7⁺PD-1⁻ phenotype (Figure 5d). These data suggest that there is an expansion of cytotoxic CD4⁺ T cells, of an effector memory and follicular helper cell phenotype, in critical SARS-CoV-2 infection.

Figure 5. Cytotoxic CD4+ and CD8+ T cells are expanded in critical infection. (a) Proportion of GZMB+PFN+ CD4+ and CD8+ T cells across disease states. (b) Proportion of CD4+ T cells and TFH cells (CXCR5+CD45RO+) expressing GZMB and PFN. (c) Summary pie chart plot of the mean proportion of total CD4+ T cells and cytotoxic (GZMB+PFN+) CD4+ T cells defined as TN (CCR7+CD45RO−), TCM (CCR7+CD45RO+), TEM (CCR7−CD45RO+) and TEMRA (CCR7−CD45RO−) in critical infection patients (n = 3). (c) Summary pie chart plots of the mean proportion of total CD4+ TFH (CXCR5+CD45RO+) cells and cytotoxic (GZMB+PFN+) CD4+ TFH cells by CCR7+PD-1− and CCR7−PD-1+ phenotype in critical infection patients (n = 3). Differences between groups were determined by the non-parametric Kruskal–Wallis test, with comparison of the rank mean of experimental groups by the Original FDR method; error bars represent median ± interquartile range.

Discussion

Characterising immune responses that associate with different disease severities of COVID-19 may help to define protective immune responses to SARS-CoV-2 and guide the rational development of next-generation vaccines. This study provides a phenotypic analysis of the T-cell response in patients experiencing asymptomatic to critical SARS-CoV-2 infection. We confirm that the T-cell compartment is distinctly altered in critical SARS-CoV-2 infection, defined by an expansion of effector memory subsets, and increased expression of activation and cytotoxic functional markers on CD4⁺ T cells. These data suggest a potentially pathogenic role of cytotoxic CD4⁺ CTLs in the progression of COVID-19.

CD4⁺ T cells are well-established critical responders in viral infection; however, cytotoxic function is more commonly associated with CD8⁺ T-cell populations. As such, it is of interest that unbiased clustering and discriminant analysis identified CD4⁺ T-cell Mcs with a cytotoxic phenotype (Mc13 and Mc10) as associated with the progression of disease from severe to critical (Figures 2c and 3d). CD4⁺ T cells can mediate host-cell death through secretion of GZMB and PFN,^24,25 and CD4⁺ CTLs have been identified during viral infections including human immunodeficiency virus, human cytomegalovirus, Epstein Barr virus, influenza, dengue virus and more recently SARS-CoV-2.²⁶ In COVID-19, CD4⁺ T cells expressing high levels of PFN1, GZMB and GZMH transcripts have been identified by scRNA-seq to be enriched in hospitalised compared to non-hospitalised patients.²⁷ The data provided in the current study describe an increase in CD4⁺ CTLs, and a distinctive expansion of GZMB⁺PFN⁺ CD4⁺ T cells during critical infection at the level of protein expression. While the circulating T-cell populations have been analysed in this cohort, high infiltration of CD4⁺ CTLs, as well as CD8⁺ CTLs, have also been reported in the lung parenchyma of severely ill COVID-19 patients.^24,28 As CD4⁺ CTLs have been shown to induce host cell death in an HLA class II restricted manner, the elevated expression of HLA class II in the respiratory epithelium of COVID-19 decedents suggests that CD4⁺ CTLs may play a role in the profound host tissue damage associated with SARS-CoV-2 acute respiratory distress syndrome (ARDS).²⁴

Mc13, shown to be enriched in critical disease patients, also exhibited expression of CXCR5, a feature of CD4 T_FH cells. Further investigation of CD4⁺ T cells expressing CXCR5, a chemokine receptor responsible for T-cell homing to B-cell follicles in secondary lymphoid organs, demonstrated a unique enrichment of cytotoxic CD4⁺ T_FH cells in critical disease patients. These cytotoxic CD4⁺ T_FH cells were predominantly of a CCR7⁻PD-1⁺ phenotype. This phenotype has been previously described as circulating T_FH cells.²⁹ While a rarely described population, cytotoxic T_FH cells have been shown to induce B cell death, correlate negatively with antibody titres, and were associated with smaller germinal centres in recurrent Strep A infection in children.³⁰ In COVID-19, post-mortem investigations have shown loss of germinal centre B cells and absence of germinal centres in the lymph nodes of decedents.³¹ Furthermore, elevated proportions of cytotoxic T_FH cells have been found to correlated negatively with anti-S1/S2 SARS-CoV-2 antibodies in hospitalised patients. However, this correlation was not observed in non-hospitalised patients.²⁷ While these data may suggest a relationship between cytotoxic T_FH cells and disturbed B cell and antibody responses in severely ill COVID-19 patients, antibody-secreting plasmablasts have been correlated with mortality in COVID-19,²² and higher antibody responses are associated with more severe disease.³² As there are limited studies investigating cytotoxic T_FH in COVID-19, the implications of cytotoxicity in T_FH cells require further investigation to determine their function and any potential detrimental effect this cell type may have on antibody responses to SARS-CoV-2.

CD4⁺ T-cell cytotoxicity has been proposed to be a compensatory mechanism to combat exhaustion of CD8⁺ T cells, in which CD8⁺ CTL expression of GZMB and PFN decrease, and PD-1 increases.^33,34 While no change in the proportion of CD8⁺ GZMB⁺PFN⁺ subsets were observed here across disease severities, GZMB, PFN and HLA-DR expression levels on CD8⁺ effector memory subsets were highlighted as contributing to separation of critical disease patients in the PCA. On the other hand, PD-1 expression was not highlighted as a key variable in the separation of disease states in the PCA (Figure 1b). This suggests that if CD8⁺ CTL exhaustion was present it was not a differentiating feature of disease progression. Previous studies have reported elevated proportions of hyperactivated CD8⁺ cells, defined by expression of HLA-DR⁺CD38⁺PD-1⁺TIM-3⁺, in severe and critical SARS-CoV-2 infection.^21,22 GZMB and PFN expression has been correlated with critical disease and mortality in COVID-19.^22,35–39 Rha et al. (2021) reported GZMB and PFN were expressed by almost all SARS-CoV-2-specific multimer⁺ CD8⁺ T cells.⁴⁰ As such, enrichment for SARS-CoV-2 specific T cells may provide more insight into the changes in proportion and phenotype of cytotoxic CD8⁺ T cells between disease states. In combination with these studies, the data presented here suggest that hyperactivated effector memory subsets of CD8⁺ T cells may contribute to COVID-19 progression.

Investigating the immune response in asymptomatic and mild disease patients can shed light on the immune response that effectively controls viral replication and disease progression. Control of viral load by T cells was demonstrated in B cell depleted Rhesus Macaques,⁴¹ and CD8⁺ T-cell responses correlated with better clinical outcome in patients with inborn errors in humoral immune responses and B cell impairment in SARS-CoV-2 infection.^42,43 In the current study, asymptomatic and mild patients exhibited low proportions of activated or cytotoxic cells, and a predominantly naïve phenotype in both the CD4⁺ and CD8⁺ compartments. However, there is an inherent limitation to investigating the peripheral immune response in a respiratory infection where localised inflammation and resident immune cell responses may not be reflected in circulation.⁴⁴ Furthermore, this study did not investigate SARS-CoV-2-specific T cells, and as such a proportionally smaller T-cell response may be missed in these less severe disease states. It has been shown previously that individuals with mild infection are less likely to have detectable SARS-CoV-2-specific T-cell responses.⁴⁵ In combination with the relative lack of a robust T-cell response in asymptomatic and mild disease patients here, this may question the necessity of a strong T-cell response to prevent the progression of COVID-19. This is consistent with the diverse polyclonal, less differentiated SARS-CoV-2 specific T-cell response observed in children with mild and asymptomatic T-cell response than the clonally expanded memory T-cell response with markers of cytotoxicity and exhaustion present in adults.⁴⁶ The protective capacity of T cells should be investigated further to assess whether preserved T-cell response against variant SARS-CoV-2 viruses provide protection against severe disease, as has been suggested previously.^12,13,16

There are several clinical variables that may confound this analysis, including the impact of the SARS-CoV-2 variant, age, sex and time of sampling on disease onset. Firstly, the two cohorts included in this study were infected with the Ancestral or Delta variant of SARS-CoV-2 (Supplementary tables 1 and 2). However, as two cohorts had asymmetric disease severity, where COVIMM patients experienced asymptomatic and mild disease, and COSIN patients experienced mild to critical disease, a direct comparison of the impact of the viral variant on T-cell response in different disease states could not be performed. Secondly, the critical disease patients were older than asymptomatic and mild patients which is reflective of the high median age of COVID-19 patients admitted to hospital and ICU.⁴⁷ Age over 65 years is associated with a progressive decline in immune function characterised by decreased thymic function, contraction of naive T-cell populations and perturbed T-cell function, such as reduced cytokine production and proliferative capacity.^48,49 It has been shown previously that SARS-CoV-2 immune dysregulation mimics that seen in age-related immunosenescence.⁵⁰ The time of sample collection after COVID-19 diagnosis is also a possible limitation in this study. However, when the PERMANOVA for the PCA in Figure 1 was replicated with clinical variables including time of sampling, sex and age, no significant difference between groups was observed (Supplementary table 3). Finally, we recognise that the analyses performed are limited by sample size, which is particularly relevant to the focus on patients with critical infection. As well as reducing the power of analysis, this prevents stratification of patients by clinical variables such as age and sex. Additionally, it was noted that three mild disease patients localised on the PCA with the critical patients; however, the difference between critical disease and all other disease states was confirmed by PERMANOVA (Supplementary table 3). As clinical characteristics such as age, sex, time of sampling did not explain this (Supplementary tables 1 and 2), this clustering is likely a reflection of natural inter-individual immune variability.

The data presented in our study add to our understanding of the contribution of T-cell responses to disease progression in SARS-CoV-2 infection. The expansion of cytotoxic CD4⁺ T-cell subset in critical disease patients suggests that CTLs may be contributing to the host tissue damage and systemic inflammatory disease that is associated with fatal COVID-19. As such, the potentially detrimental role of T-cell responses in COVID-19 should be considered in the development of next-generation therapies and vaccines against SARS-CoV-2.

Methods Experimental design

This study aimed to characterise the T-cell compartment of patients across the clinical spectrum of COVID-19. To achieve this, PBMCs were collected from patients with asymptomatic to critical disease and a high parameter spectral cytometry panel was developed to assess the proportions and functions of T-cell subsets. Differences in T-cell phenotype across COVID-19 severity were assessed by PCA. Populations contributing to the distinction between severe and critical patients were identified by FlowSOM clustering and PLS-DA, then validated by manual gating.

Patient demographics

SARS-CoV-2 PCR-positive patients were enrolled through the Royal Prince Alfred (RPA) hospital COVID-19 clinic or virtual care system in March of 2020, at which time the Ancestral SARS-CoV-2 virus was the dominant circulating variant in Australia (COVIMM cohort). Ethics approval was granted by the RPA ethics committee, human ethics number X20-0117 and 2020/ETH00770. Verbal consent was given by all participants. Additional patients were enrolled in the COSIN study in June 2021, when Delta was the dominant circulating variant in Sydney, Australia, through seven hospital microbiology laboratories, and out-patient care units as described by Balachandran et al. (2022), are also included.⁵¹ Ethics approval was granted by Human Research Ethics Committees of the Northern Sydney Local Health District and the University of New South Wales, NSW Australia (ETH00520), and written consent was obtained from all patients.

SARS-CoV-2 infection was defined to be a positive nasopharyngeal RT-PCR performed by accredited laboratories within the Sydney Health District. Patients were classified as asymptomatic (n = 5), mild (n = 18), moderate (n = 4), severe (n = 5) or critical (n = 3) COVID-19 disease severity as defined by the NIH guidelines (https://www.covid19treatmentguidelines.nih.gov/overview/clinical-spectrum/). Patients were classified as mild if their symptoms were self-manageable. While severity of COVID-19 was determined by the NIH guidelines, it is of note that one critical patient was reported to have an arterial blood oxygen saturation (SaO₂; Supplementary table 2) of 15%, significantly lower than all other hospitalised patients.

Patient PBMCs were isolated from whole blood by Ficoll-density gradient separation, and cryopreserved in heat-inactivated FBS (Sigma-Aldrich) with 10% DMSO at −80°C. There was no significant difference in the median age, time of sampling from a positive PCR test or time of sampling from symptom onset between the two cohorts (Supplementary tables 1 and 2).

Immunophenotyping by spectral cytometry

Cryopreserved PMBC samples were thawed and diluted to a concentration of 1 × 10⁶ live PMBCs in RMPI 1640 (Gibco) supplemented with 10% FBS and 50 μg mL⁻¹ streptomycin and 50 μg mL⁻¹ penicillin (Gibco) in a U-bottom 96-well plate. Cells were washed in FACS buffer (2% FBS, 2 mM EDTA, in PBS), dead cells labelled with LIVE/DEAD™ Fixable Blue dye (Invitrogen) for 20 min at 4°C and washed twice in FACS buffer. Cells were stained with extracellular markers (Supplementary table 4; CD3-BUV395 (BD Horizon), CD4-PerCPCy5.5 (BioLegend), CD8-BUV496 (BD Horizon), CD19-BV510 (BioLegend), CD56-BV605 (BD Biosciences), HLA-DR-BV650 (BD Biosciences), CD45RO-BV786 (BioLegend), CD197-PECy7 (BioLegend) and CXCR5-AF647 (BD Biosciences)) in FACS buffer for 20 min at 4°C. Cells were then permeabilised and fixed with Transcription Factor Buffer Set (BD Pharmingen™) for 40 min at 4°C. Cells were washed then stained with intracellular markers (Supplementary table 4) PD-1-BV737 (BD Horizon), Ki-67-BV711 (BioLegend), Perforin-PE (BioLegend) and Granzyme B-APC (BioLegend) in Transcription Factor wash buffer (BD Pharmingen™). Anti-Mouse Ig, κ/Negative Control Particles Set (BD™ CompBeads) were used for single stain controls and stained with corresponding extracellular and intracellular stains. Fluorescent minus one controls were included for intracellular markers and stained at the concentration of patient samples at corresponding staining times. The samples underwent a second fixation step with 4% paraformaldehyde for 30 min, washed and resuspended in 200 mL FACS buffer for acquisition. Patient samples and controls were acquired using the 5-laser Cytek Aurora® Spectral Cytometer. Spectral unmixing was performed using the inbuilt SpectroFlo® software after acquisition of unstained and single-stained controls and before patient sample acquisition.

Data analysis

Data analysis was performed in FlowJo v10.8.1, R v4.2.1 and GraphPad Prism v9. Dead cells, doublets and debris were excluded in FlowJo v10.8.1 by FSC-A, FSC-H, SCC-A and LIVE/DEAD dye (Supplementary figure 1). B cells were excluded by CD3⁻CD19⁺ phenotype. Compensation for unmixing errors was performed in FlowJo v10.8.1. Irregularities in staining and acquisition between batches were controlled by matching gates on control samples and applying identical gate lineage and functional marker expression within each staining batch.

As samples were not all stained and acquired concurrently, a control patient was stained and acquired with each set of samples (batch control). Manual gating in FlowJo v10.8.1 was performed in all analyses that compared samples between these batches, and the batch control was used to ensure gates were set at optimal cut-off points for each batch of samples. For this reason, the PCA (Figure 1) was calculated using 49 overlapping CD3⁺ T-cell populations that were manually gated in FlowJo v10.8.1. After exclusion of CD19⁺ B cells, T cells were gated as CD56⁺ and CD56⁻. Both CD56⁺ and CD56⁻ populations were subsequently gated by CD4⁺ and CD8⁺ T cells. The CD56⁻ CD4⁺ and CD8⁺ T cells were divided into T_N (CCR7⁺CD45RO⁻), T_CM (CCR7⁺CD45RO⁺), T_EM (CCR7⁻CD45RO⁺), T_EMRA (CCR7⁻CD45RO⁻) subsets. To perform the PCA, the proportion of populations of interest were exported into Excel v16.54, to be imported into R v4.2.1. The populations of interest were CD4⁺ and CD8⁺ T_N, T_CM, T_EM, T_EMRA cells, and CD56⁺CD4⁺, CD56⁺CD8⁺ and CD4⁺/CD8⁺ memory subsets expressing HLA-DR, PD-1, Ki-67, PFN and GZMB (Supplementary figure 2). Each population was calculated as a proportion of its respective CD4⁺ or CD8⁺ T-cell compartment, and these proportions were used for PCA analysis in R v4.2.1 using the Spectre package.⁵²

For the comparison between severe and critical infection, the R package Spectre was utilised for computational analyses.⁵² CD3⁺ T cells were exported for each patient sample and arcsine transformation was applied to the data. FlowSOM clustering created 25 Mcs. Fit-SNE dimensionality reduction⁵³ was run on subsampled data of 50 000 cells from severe and critical groups to create a representative plot between disease states and a heatmap plot of cellular marker expression on each metacluster was created using the pheatmap function. To identify contributions to differences between severe and critical infection, a PLS-DA was done.⁵⁴ To validate the phenotype of Mcs, an FCS file of marker expression on each metacluster was created using the write.files function in Spectre. This allowed for clear analysis of marker expression for each metacluster in FlowJo v10.8. After exclusion of CD4⁻CD8⁻ FlowSOM Mcs, PLS-DA was performed to identify the variability in proportion of Mcs between severe and critical disease. Validation of the statistical differences in proportion of Mcs between groups was performed in GraphPrism v9.

The difference between groups were analysed by the two-sided paired Mann–Whitney U-test and the non-parametric Kruskal–Wallis test, with comparison of the rank mean of experimental groups by the Original False Discovery Rate (FDR) method with correction for multiple comparisons. Statistical significance between groups in computational analyses were calculated with a permutation ANOVA (PERMANOVA), with correction for multiple comparisons by the FDR method. Statistical significance was set as P ≤ 0.05.

Acknowledgments

FMW and TMA are supported by the International Society for the Advancement of Cytometry (ISAC) Marylou Ingram Scholars program. The COSIN cohort was supported by Snow Medical Foundation as an investigator initiated study. We acknowledge the support of the University of Sydney Advanced Cytometry Facility. The authors thank the study participants for their contribution to this research and the clinical staff who collected the samples. Open access publishing facilitated by The University of Sydney, as part of the Wiley - The University of Sydney agreement via the Council of Australian University Librarians.

Author contributions

Conceptualisation: SB, CA, FMW, TA, JAT, MS, UP, ALF. Methodology: SB, CA, FMW, TA, ALF, UP, JAT, MS. Investigation: SB, CA, HL, SA. Visualisation: SB, FMW, TA. Resources: JJP, PK, AB, MM, RAB, AL, AG, OH, WJB. Funding acquisition: JAT, CC, MS, ALF, WJB, RAB, AL. Supervision: CA, MS, JAT. Writing—original draft: SB, CA, MS, JAT. Writing—review & editing: all authors.

Conflict of interest

The authors declare no conflict of interest.

Data availability statement

All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supporting Information. The data sets generated during and/or analysed during the current study are available from the corresponding author on request.