Main

Severe infection with COVID-19 has been linked to immune dysregulation, including impaired or delayed production of type I and type III interferons1,2,3,4,5, marked lymphopenia6,7,8,9,10 and a paradoxical increase in pro-inflammatory cytokines, such as TNF-α, IL-1β and IL-61,4,6,11,12,13. Alteration of T cell compartments include increases in effector and activated CD4 and CD8 T cells14,15,16,17, whereas changes in B cell and humoral compartments include robust plasmablast differentiation and production of SARS-CoV-2-reactive IgM and IgG antibodies14,18,19,20. More recently, distinct immunophenotypes have been associated with COVID-19 disease severity and trajectory3,4,11,14,15. Understanding how clinical features affect the host immune response to SARS-CoV-2 will elucidate determinants of disease severity.

Patients with cancer have an increased risk of severe COVID-19 (refs. 21,22,23,24), with an estimated case fatality rate of 25%25 compared to 2.7% in the general population26. Importantly, cancer is a heterogeneous disease with mortality rates as high as 55% among patients with COVID-19 who also have hematologic cancer21,24,27,28,29,30,31,32,33,34. It is less apparent whether the increased mortality by cancer subtype is independent of the confounding effects of other prognostic factors, including Eastern Cooperative Oncology Group (ECOG) performance status35, which is the most important predictor of death in the population of patients with cancer36. Furthermore, there are limited data on the immune response to SARS-CoV-2 in patients with cancer, whether it differs by cancer subtype, whether it is affected by immune-modulating therapies such as B cell-depleting therapy and, most importantly, how each of these factors influence mortality in the setting of COVID-19. We studied three cohorts of patients with cancer who also have acute COVID-19 across two hospital systems to understand the immunologic determinants of COVID-19 mortality in cancer.

Results

Hematologic cancer is a risk factor for COVID-19 mortality

We first conducted a prospective multi-center observational cohort study of patients with cancer who were hospitalized with COVID-19 (COVID-19 Outcomes in Patients with Cancer (COPE); Methods). The median age of this cohort was 68 years; 48% were female, 54% were Black and 57% were current or former smokers (Table 1). In terms of cancer-specific factors, 78% of patients had solid cancer, with prostate and breast cancer most prevalent; 46% had active cancer, defined as diagnosis or treatment within 6 months; and 49% had a recorded ECOG performance status of 2 or higher (Table 1). During follow-up, 48% of patients required intensive care unit (ICU) level care, and 38% of patients died within 30 d of admission (Supplementary Table 1). Demographics by tumor type are available in Supplementary Table 2.

Table 1 COPE: patient demographics and clinical characteristics
Full size table

We performed univariate analyses to identify factors associated with all-cause mortality in the period between hospital admission and 30 d after discharge. We included relevant covariates, including patient factors such as age, race, gender and smoking history (ever versus never)37,38,39,40; cancer-specific factors including ECOG performance status33,35 and status of cancer (for example, active versus remission)34; cancer type (for example, hematologic versus solid cancer)27,32,34,41,42; and cancer treatment33,35. Current or previous smoking (P = 0.028), poor ECOG performance status (ECOG 3–4, P = 0.001) and active cancer status (P = 0.024) (Fig. 1) were all associated with increased COVID-19 mortality. Consistent with recent data, patients with hematologic cancer appeared to have an increased risk of mortality relative to patients with solid cancer (55% versus 33% respectively; P = 0.075) (Supplementary Table 1)21,27,32,33,34,41. However, similarly to published literature, cancer treatment, including cytotoxic chemotherapy, was not significantly associated with COVID-19 mortality27,28,32,34,41.

Fig. 1: Univariate analysis of potential risk factors in COVID-19 mortality.
figure 1

Data are presented as ORs with 95% CI. ref, reference population. +BMI 18.5–24.9; ++BMI < 18.5; +++BMI > 25; #exposure to immunosuppressive medications not including cancer treatment; ^diagnosis or treatment within 6 months; *single-agent immunotherapy, targeted therapy or monoclonal antibodies.

Full size image

We then performed multivariable logistic regression to assess whether the increased mortality observed in patients with hematologic versus solid malignancy was independent of potential confounding effects from smoking history, poor ECOG performance and active cancer. In this fully adjusted analysis, hematologic malignancy was strongly associated with mortality in comparison to solid cancer (odds ratio (OR) = 3.3, 95% confidence interval (CI) 1.01–10.8, P = 0.048) (Table 2). Similar results were observed in time-to-event analyses using Kaplan–Meier methods (Fig. 2a; median overall survival (OS) not reached for patients with solid cancer versus 47 d for patients with hematologic cancer; P = 0.030) and Cox regression models (Table 2; hazard ratio (HR) = 2.56, 95% CI 1.19–5.54, P = 0.017). Moreover, patients with hematologic cancer had higher levels of some inflammatory markers on admission laboratory testing, including ferritin, IL-6 and lactate dehydrogenase (LDH) (Fig. 2b and Extended Data Fig. 1a,b). We also assessed viral persistence in 43 patients with repeat RT–PCR testing (37 solid and six hematologic). Both patients with hematologic cancer and patients with solid cancer had prolonged viral clearance, with some greater than 60 d (Extended Data Fig. 1c,d and Supplementary Table 3). Altogether, hematologic malignancy was an independent risk factor of death, with signs of a dysregulated inflammatory response.

Table 2 COPE: event rates and point estimates of outcomes by cancer type
Full size table
Fig. 2: Hematologic cancer is an independent risk factor for COVID-19-related mortality.
figure 2

a, Kaplan–Meier curve for COVID-19 survival of patients with solid (n = 77) and hematologic (n = 22) cancer. Cox regression-computed HR for mortality in hematologic versus solid cancer, adjusted for age, gender, smoking status, active cancer status and ECOG performance status. b, Ferritin (P = 0.036), IL-6 (P = 0.034) and LDH (P = 0.001) in patients with solid (n = 62) and hematologic (n = 15) cancer hospitalized for COVID-19. All—significance determined by two-sided Mann–Whitney test: *P < 0.05 and **P < 0.01. Median and 95% CI are shown. Heme, hematologic.

Full size image

Impaired SARS-CoV-2 antibody responses in hematologic cancer

To determine whether patients with cancer who were infected with SARS-CoV-2 might exhibit an altered immune landscape, we leveraged an observational study of hospitalized patients with COVID-19 at the University of Pennsylvania Health System where blood was collected (MESSI-COVID)14. This analysis included 130 patients with flow cytometric and/or serologic analysis. Twenty-two patients had active cancer (Supplementary Tables 4 and 5), including patients undergoing cancer-directed therapies such as chemotherapy, immunotherapy or B cell-directed therapies (Supplementary Table 6). Patients with active cancer were older and predominantly female; both groups had a similar time frame of symptom onset and disease severity (Fig. 3a and Supplementary Table 4). However, patients with cancer had a higher all-cause mortality (36.4% versus 11.1%; Fig. 3a), consistent with our COPE cohort and other reported cohorts21,24,27,28.

Fig. 3: High-dimensional analyses reveal immune phenotypes associated with mortality and distinct phenotypes between solid and hematologic cancer.
figure 3

a, Demographic and mortality data for the MESSI cohort at the University of Pennsylvania. b, Relative levels of SARS-CoV-2 IgG (No Cancer versus Heme, P = 0.001; Solid versus Heme, P = 0.007) and IgM (No Cancer versus Heme, P = 0.003; No Cancer versus Solid, P = 0.03); patients with solid (n = 14) and hematologic (n = 7) cancer and patients without cancer (n = 108). c, Left, global UMAP projection of lymphocyte populations for all 45 patients pooled. Right, hierarchical clustering of EMD using Pearson correlation, calculated pairwise for lymphocyte populations. d, UMAP projection of concatenated lymphocyte populations for each EMD cluster. Yellow: high density; black: low density. e, Heat map showing expression patterns of various markers, stratified by EMD cluster. Heat scale calculated as column z-score of MFI. f, Mortality (P = 0.02), disease severity and SARS-CoV-2 antibody data, stratified by EMD cluster (Cluster 5 n = 5; Clusters 1, 2, 3 and 4 n = 40). Mortality significance was determined by the Pearson chi-square test. Severity assessed with the NIH ordinal scale for COVID-19 clinical severity (1: Death; 8: Normal Activity)14. g, UMAP projections of concatenated lymphocyte populations for solid cancer, hematologic cancer and patients without cancer. h, CD8, CD4 (No Cancer versus Heme, P= 0.003; Solid versus Heme, P = 0.01) and B cell (No Cancer versus Heme, P = 0.008; No Cancer versus Solid, P = 0.03; Solid versus Heme, P = 0.02) frequencies in healthy donors (n = 33), patients without cancer (n = 108), patients with solid cancer (n = 7) and patients with hematologic cancer (n = 4). i, UMAP projection of non-naive CD8 T cell clusters identified by FlowSOM. j, Top, UMAP projections of non-naive CD8 T cells for patients with cancer and patients without cancer. Bottom, UMAP projections indicating HLA-DR and CD38 protein expression on non-naive CD8 T cells for all patients pooled. k, Frequency of activated FlowSOM clusters in patients with HD (n = 30), patients without cancer (n = 110) and patients with cancer (n = 8) (P = 0.03). l, Representative flow plots and frequency of HLA-DR and CD38 co-expression in patients with HD (n = 30), patients without cancer (n = 110), patients with solid cancer (n = 7) and patients with hematologic cancer (n = 3) (gated on non-naive CD8) (No Cancer versus Heme, P < 0.0001; Solid versus Heme, P = 0.02). All—significance determined by two-sided Mann–Whitney test: *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001. Median and 95% CI are shown. HD, healthy donors; Heme, hematologic; MFI, mean fluorescence intensity.

Full size image

As recovery from COVID-19 results in immunologic memory in the form of antibodies and memory B cells43, we hypothesized that the observed poor outcomes in patients with active cancer might be associated with a defect in SARS-CoV-2-specific humoral immunity. Patients with cancer had significantly decreased SARS-CoV-2-specific IgG and IgM antibodies than patients without cancer (Extended Data Fig. 2a). This was largely due to patients with hematologic cancer, most (6/7) of whom had IgM and IgG levels below the cutoff of positivity of 0.48 arbitrary units (Fig. 3b and Extended Data Fig. 2b), whereas patients with solid cancer had IgG and IgM antibody responses that were more similar to patients without cancer (Fig. 3b).

T cell-depleted phenotype associated with COVID-19 mortality

Patients who recover from COVID-19 also exhibit SARS-CoV-2-specific CD4 and CD8 T cell responses43,44. We, therefore, performed exploratory high-dimensional analysis on the lymphocyte compartment of 45 patients, including 44 patients with COVID-19 (36 patients without cancer, six patients with solid cancer and two patients with hematologic cancer) and one control patient without COVID-19. Uniform manifold approximation and projection (UMAP) representation of 27-parameter flow cytometry data highlighted discrete islands of CD4 and CD8 T cells and CD19+ B cells (Fig. 3c and Extended Data Fig. 3a). We subsequently used the earth mover’s distance (EMD) metric45 to calculate the distance between the UMAP projections for every pair of patients. Clustering on EMD values identified five clusters of patients with similar lymphocyte profiles (Fig. 3c). Differences among these clusters of patients were driven by both the distribution (Fig. 3d and Extended Data Fig. 3b) and phenotype (Fig. 3e) of CD4, CD8 and B cells. EMD Cluster 1 was defined by depleted CD4 and B cells, increased CD8 T cells and increased activation and effector markers, including PD-1, CX3CR1, Ki67 and HLA-DR (Fig. 3d,e and Extended Data Fig. 3b). EMD Cluster 3 had decreased T cells and B cells, with an inactivated immune profile, and EMD Cluster 5 was depleted of both CD4 and CD8 T cells but had preserved B cells. In contrast, EMD Cluster 4 was defined by robust CCR7+CD27+ memory CD4 T cell responses and heterogenous B cell responses. EMD Cluster 2 had the most balanced responses, with CD4, CD8 and B cells represented (Fig. 3d,e and Extended Data Fig. 3b). We then correlated these five patterns of immune responses with clinical and serological variables. Patients in EMD Cluster 5 with depleted T cells had the highest mortality and disease severity, despite generating SARS-CoV-2-specific IgM and IgG antibodies (Fig. 3f and Extended Data Fig. 5d). In contrast, EMD Clusters 2 and 4, with robust CD4 and/or CD8 T cell responses, had the lowest mortality and a low disease severity (Fig. 3f and Extended Data Fig. 5d). These findings suggest a key role for T cell immunity in facilitating recovery from acute COVID-19, even in the presence of intact humoral immunity.

Distinct immune landscape in hematologic cancer

We next explored the role of cancer subtype on immune phenotype. Four of the six patients with solid cancer were in EMD Cluster 2, with a balanced immune phenotype (Fig. 3e). In contrast, both patients with hematologic cancer were in EMD Cluster 1, which had marked depletion of CD4 and B cells. Indeed, UMAP projections showed that, whereas patients with solid cancer had an immune landscape similar to patients without cancer, the two patients with hematologic cancer demonstrated loss of islands associated with CD4 and B cells (Fig. 3g). We then extended this analysis by measuring the frequency and phenotype of key lymphocyte populations in the entire MESSI-COVID cohort and healthy donor controls. Patients with COVID-19 and hematologic cancer had a significantly lower frequency of CD4 and B cells than patients with solid cancer, patients without cancer and healthy donors without COVID-19 (Fig. 3h). As T follicular helper (Tfh) cells and plasmablasts are critical in the generation of effective antibody responses, we assessed circulating Tfh and plasmablast responses. Although limited by sample size, patients with hematologic cancer had low circulating Tfh (PD1+CXCR5+) and plasmablast (CD19+CD27hiCD38hi) responses and decreased CD138 expression (Extended Data Fig. 4a). Thus, patients with hematologic malignancy appear to have quantitative defects in CD4 and B cells that might be required for effective SARS-CoV-2-specific antibody responses.

As patients with hematologic cancer had a preserved frequency of CD8 T cells, we performed FlowSOM clustering analysis on non-naive CD8 T cells from 118 patients with COVID-19 and 30 healthy donors and visualized the clusters using UMAP. UMAP clearly separated CX3CR1- and Tbet-expressing effector cells from memory CD8 T cells expressing CD27 and TCF-1 (Extended Data Fig. 4b and Fig. 3i). The effector island was composed of CD45RAloCD27lo effector memory cells (Clusters 2 and 3) and CD45RA+ TEMRA cells (Cluster 1). The memory island was composed of CCR7lo transitional memory (Cluster 5), effector memory (Clusters 7 and 8) and CCR7hi central memory (Cluster 9) cells. Activated cells, characterized by high HLA-DR, CD38 and Ki67 expression, were identified in Clusters 3, 4 and 5 (Extended Data Fig. 4c).

We then compared the landscape of CD8 T cells in patients with and without cancer. CD8 T cell subsets, including central memory, effector memory, transitional memory and TEMRA, were similar between patients with and without cancer (Extended Data Fig. 4d). However, UMAP representation of non-naive CD8 data demonstrated preferential enrichment of cells expressing HLA-DR and CD38 in patients with cancer compared to patients without cancer (Fig. 3j). Indeed, patients with cancer had higher frequencies of activated HLA-DR-, CD38- and Ki67-expressing FlowSOM clusters (Clusters 3, 4 and 5) than patients without cancer and healthy donors (Fig. 3k and Extended Data Fig. 4e). When stratified by cancer type, the increased HLA-DR and CD38 expression was restricted to the patients with hematologic cancer; patients with solid cancer and those without cancer had similar levels of activation (Fig. 3l). Altogether, patients with solid cancer and COVID-19 had an immune landscape similar to patients without cancer with COVID-19. In contrast, patients with hematologic malignancies had defects in CD4 T cells, B cells and humoral immunity but preserved and highly activated CD8 T cells. These observations raised the possibility that CD8 T cell activation might potentially compensate for blunted humoral immune responses in patients with hematologic malignancies.

CD8 T cells influence survival in patients with hematologic cancer

To more rigorously explore the role of T and B cell immunity in patients with cancer who were infected with SARS-CoV-2, we examined a cohort of patients with cancer who were hospitalized with COVID-19 at Memorial Sloan Kettering Cancer Center (MSKCC), which included a larger number of patients with hematologic malignancies, including those treated with B cell-depleting therapy. The median age of the MSKCC cohort was 65 years, and, in contrast to the MESSI cohort at the University of Pennsylvania, 81% of the cohort was White (Fig. 4a and Supplementary Tables 7 and 8). Consistent with the Penn COPE and MESSI cohorts, patients with hematologic cancer did poorly, with a mortality rate of 44.4% (Fig. 4a and Supplementary Table 7). In the MSKCC cohort, both CD4 and CD8 T cells were significantly decreased in patients with active solid and hematologic cancer compared to patients in clinical remission (Extended Data Fig. 5a). Moreover, despite the fact that a substantial number of patients with hematologic cancer from the MSKCC cohort received convalescent plasma (Supplementary Table 7), they had a significant defect in SARS-CoV-2-specific IgG and IgM responses compared to patients with solid cancer (Extended Data Fig. 5b). This was independent of disease severity and viral load (Extended Data Fig. 5c,d).

Fig. 4: CD8 T cell counts associated with survival in patients with hematologic cancer and COVID-19.
figure 4

a, Demographic and mortality data of the MSKCC cohort. b, Left, hierarchical clustering of EMD using the Pearson correlation, calculated pairwise for lymphocyte populations; right, global UMAP projection of lymphocyte populations pooled. c, UMAP projection of concatenated lymphocyte populations for each EMD cluster. Yellow: high density; black: low density. d, Mortality (Cluster 5 n = 7; Clusters 1, 2 and 4 n = 50), severity and RT–PCR cycle threshold (Cluster 1 n = 14; Cluster 2 n = 5; Cluster 4 n = 24; Cluster 5 n = 6) (lower Ct: higher viral load) stratified by EMD cluster. Mortality significance determined by the Pearson chi-square test. e, Relative levels of SARS-CoV-2 IgG and IgM of patients with recent cancer treatments (solid tx n = 9; heme αCD20 n = 7; heme other tx n = 5). f, Absolute CD8 and CD4 T cell in patients treated with B cell-depleting therapy (alive n = 7; dead n = 4). g, Absolute CD8 (P = 0.01) and CD4 T cell counts and B cell counts (P = 0.003) counts in patients with hematologic cancer (alive n = 17; dead n = 18). h, Kaplan–Meier curve for survival in patients with hematologic cancer stratified by CD8 T cell counts (threshold = 55.9; log-rank HR) (≥55.9 n = 28; <55.9 n = 13). CD8 count threshold determined by CART analysis. i, Left, representative ELISpot plates from two patients after stimulation with no peptide, EBV and SARS-CoV-2 consensus peptide pools; right, IFN-γ and IL-2 (P = 0.02) SFU per million PBMCs in patients with hematologic cancer (n = 13). Significance was determined by the two-sided Wilcoxon test. j, Proportion of patients with hematologic cancer with detectable IFN-γ SFU, after background subtraction (no peptide). Significance was determined by the chi-square test. k, Simple linear regression between IFN-γ SFU and percent activated CD8 T cells in patients with hematologic cancer who recovered from COVID-19 (n = 6). All—significance determined by two-sided Mann–Whitney test: *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001. Median and 95% CI are shown. Ct, cycle threshold; Heme, hematologic; tx, treatment.

Full size image

EMD and clustering of flow cytometry data from 20 patients with solid cancer, 31 patients with hematologic cancer and six patients in remission identified four immune phenotypes (Fig. 4b,c and Extended Data Fig. 6a,b) that corresponded to the immune phenotypes 1, 2, 4 and 5 identified in the Penn-MESSI cohort (Fig. 3c,d). The Penn phenotype 3, the only cluster that did not have patients with cancer, was not identified in the MSKCC cohort. Consistent with the Penn data, MSKCC EMD Cluster 5, with depleted of CD4 and CD8 T cells and preserved B cells, had the highest mortality of 71% and was associated with a high disease severity and viral load (Fig. 4d).

Patients in the MSKCC cohort with solid cancer were present in all four clusters, with most in Cluster 4, whereas patients with hematologic cancer were predominantly in Clusters 1 and 4 and absent from Cluster 2 (Extended Data Fig. 7a,b). Intriguingly, the clinical outcomes of patients with immune phenotype 4 was the greatest contributor to the overall mortality difference between patients with solid and hematologic cancers. Patients with hematologic cancer with phenotype 4 had a mortality of 62% versus 9% in patients with solid cancer (Extended Data Fig. 7c), with a corresponding higher viral load (Extended Data Fig. 7d). Immune phenotype 4 was characterized by robust CD4 responses and decreased, but still present, CD8 responses (Extended Data Fig. 6b). Within immune phenotype 4, patients with solid and hematologic cancers had similar CD4 and CD8 T cell counts (Extended Data Fig. 7e). However, patients with hematologic cancer had near-complete abrogation of B cells (phenotype 4A) that corresponded with a mortality rate of 62% (Extended Data Fig. 7c,f). In contrast, patients with solid cancer had intact B cell counts (phenotype 4B) with a mortality of 9% (Extended Data Fig. 7c,f). Thus, in a setting with similar CD4 and CD8 T cell numbers, B cell depletion was associated with higher mortality.

Anti-CD20 (αCD20) therapy with rituximab or obinutuzumab-containing regimens depleted circulating B cells and significantly impaired SARS-CoV-2-specific IgG and IgM responses (Fig. 4e and Extended Data Fig. 8a). αCD20 was not associated with quantitative changes in CD4 and CD8 T cells. However, patients treated with αCD20 therapy exhibited significant reductions in CD4 and CD8 naive and memory T cells, with a relative skewing toward effector differentiation and an activated HLA-DR+CD38+ phenotype (Extended Data Fig. 8b,c). These changes were not seen in patients with hematologic cancer without COVID-19 who received αCD20 (Extended Data Fig. 8d,e). Notably, despite the loss of B cells and humoral immunity, αCD20 therapy was not associated with increased mortality, disease severity or viral load when compared to chemotherapy or observation (Extended Data Fig. 9a).

We sought to understand why αCD20 therapy was not associated with greater mortality in these patients. Patients treated with αCD20 therapy were restricted to immune phenotypes 1 and 4, characterized by depleted B cells (Extended Data Fig. 9b). However, phenotype 1, characterized by preserved CD8 T cells, was associated with a lower mortality (Extended Data Fig. 9c). Indeed, patients treated with αCD20 who survived their COVID-19 hospitalization had higher CD8 T cell counts (Fig. 4f) and lower viral load (Extended Data Fig. 9d). We extended these analyses to other patients with hematologic cancer, including those on chemotherapy who also had quantitative (Extended Data Fig. 8a), and possibly qualitative, B cell defects. Patients with hematologic cancer who survived had higher CD8 T cell counts (Fig. 4g), which was not observed in patients with solid cancer (Extended Data Fig. 9e). Conversely, CD4 T cell counts were not associated with mortality, and higher B cell counts were associated with increased mortality (Extended Data Fig. 9e and Fig. 4g). Classification and regression tree (CART) analysis identified a CD8 T cell level that was predictive of survival after COVID-19 in patients with hematologic cancer (Fig. 4h). Thus, patients with hematologic cancer, in the setting of defective humoral immunity, were more highly dependent on adequate CD8 T cell counts than patients with solid cancer.

Finally, to assess the antigen-specificity of T cell responses in patients with cancer who were infected with SARS-CoV-2, we stimulated peripheral blood mononuclear cells (PBMCs) from 13 patients with hematologic cancer and ten patients with solid cancer using major histocompatibility complex (MHC)-I- and MHC-II-restricted SARS-CoV-2 and Epstein–Barr virus (EBV) peptide pools and measured IFN-γ and IL-2 responses using enzyme-linked immunosorbent spot (ELISpot) assays (Extended Data Fig. 10a). After background (no-peptide) subtraction, ten of 13 (77%) patients with hematologic cancer had SARS-CoV-2-specific IL-2 and/or IFN-γ responses (Fig. 4i), including eight of eight (100%) patients treated with αCD20 (Fig. 4j and Extended Data Fig. 10b). In general, these SARS-CoV-2-specific responses were greater in magnitude than EBV-specific responses (Fig. 4i), and patients with hematologic cancer had greater SARS-CoV-2 antigen-specific T cell responses than patients with solid cancer (Extended Data Fig. 10c). Moreover, in patients with hematologic cancer who recovered from COVID-19, SARS-CoV-2-specific IFN-γ responses correlated with the frequency of HLA-DR+CD38+ CD8 T cells (Fig. 4k) but not with HLA-DR+CD38+ CD4 T cells (Extended Data Fig. 10d). This correlation was not seen in patients who died (Extended Data Fig. 10e). Taken together, these findings suggest that, for patients with hematologic cancer who recover from COVID-19, increased peripheral CD8 T cell activation might reflect an appropriate induction of SARS-CoV-2-specific T cell responses. In contrast, peripheral CD8 T cell activation might be uncoupled from SARS-CoV-2 specific T cell responses in patients who ultimately die from disease. These data do not exclude a role for SARS-CoV-2-specific CD4 T cells, which has been associated with control of SARS-CoV-2 infection in patients without cancer and in mouse models46,47.

Discussion

A notable feature of the COVID-19 pandemic has been the dramatic heterogeneity in clinical presentations and outcomes, yet mechanistic explanations for the wide variance in disease severity have remained elusive. We speculated that investigating both the clinical outcomes and immunologic profile of patients with cancer might provide valuable insight into how the arms of the immune system contribute to viral control and mortality during acute COVID-19. Our work reveals several key insights. First, we established, in a prospective clinical cohort, that hematologic malignancy is an independent predictor of COVID-19 mortality after adjusting for ECOG performance and disease status. Given that patients with a poor ECOG performance status are known to have higher COVID-19 mortality35, this work extends the findings of a recent meta-analysis that reported an increased COVID-19 mortality rate in patients with hematologic cancer48 by demonstrating that the increased mortality risk seen in hematologic cancer was, in fact, driven by cancer subtype rather than differences in patient characteristics.

Second, using high-dimensional analyses, we define immune phenotypes associated with mortality in SARS-CoV-2-infected patients wth cancer. In particular, patients with depleted T cell responses had the highest mortality, regardless of the presence of B cell responses. Thus, humoral immunity alone is often insufficient in acute COVID-19. This is consistent with recent data demonstrating the importance of SARS-CoV-2-specific CD4 and CD8 T cells in viral clearance and limiting disease severity46,47. In fact, greater B cell responses were associated with higher mortality in both solid and liquid cancers. B cell responses might be a marker of disease severity, as seen with plasmablasts14,18 and neutrophils18,49,50 in severe COVID-19. Alternatively, some components of the B cell and humoral responses might be aberrant and pathogenic, as might be the case with autoantibodies targeting type I interferons in severe COVID-19 (ref. 51). CD8 T cells are known to be critical for viral clearance, particularly in response to higher viral inocula52. Our data suggest that CD8 T cells play a key role in limiting SARS-CoV-2, even in the absence of humoral immunity. This is consistent with recent data demonstrating the presence of SARS-CoV-2-specific CD8 T cell responses in acute and convalescent individuals46,53,54,55,56, and that CD8 T cells contribute to protection from SARS-CoV-2 rechallenge in the setting of waning antibody responses in convalescent macaques57.

The compensatory role of T cells was restricted to patients with hematologic, but not solid, malignancies. Thus, CD8 T cells likely play an important role in the setting of quantitative and qualitative B cell dysfunction in patients with lymphoma, multiple myeloma and leukemia undergoing αCD20 therapy, chemotherapy or Bruton tyrosine kinase inhibition. Indeed, we could identify SARS-COV-2-specific T cell responses in most patients with hematologic cancer, including all patients treated with αCD20, and they were generally greater in magnitude than in patients with solid cancer. These SARS-CoV-2-specific T cell responses correlated with peripheral activated CD8 T cells, but not CD4 T cells, in patients who recovered from infection. These data suggest a role for SARS-CoV-2-specific CD8 T cells in controlling acute infection, although more definitive studies are needed. These observations also have important translational implications. CD8 T cell counts might inform on the need for closer monitoring and a lower threshold for hospitalization in patients with COVID-19 and hematologic malignancy. Furthermore, the clinical benefit of dexamethasone, which demonstrated an overall mortality benefit in hospitalized patients with COVID-19, but is known to suppress CD8 T cell responses58, should be investigated further in patients who recently received αCD20 therapy.

Our findings do not exclude a key role for humoral immunity in protection from COVID-19 mortality. Indeed, although patients with solid cancer had a cellular and humoral immune landscape that was similar to patients without cancer59, patients with hematologic cancer had substantial defects in B cells and humoral immunity. Our analysis revealed that blunted humoral immune responses resulted in an increased mortality rate in patients with diminished, but not absent, CD8 T cell responses. Thus, B cells and associated antibody responses play an important role in acute SARS-CoV-2 infection, and CD8 T cell responses that are normally sufficient might no longer be adequate in the setting of compromised humoral immunity. This is consistent with published data demonstrating that uncoordinated immune responses in the elderly was associated with severe disease and poor outcomes46.

Importantly, both B cell-depleting therapies and cytotoxic chemotherapy agents, which can compromise the T cell compartment, are mainstays of lymphoma therapy. Both are administered, often in combination, with curative intent for patients with aggressive lymphoma but also for debulking or palliation in patients with indolent lymphoma. Based on our data, we suggest that oncologists and patients considering treatment regimens that combine B cell depletion with cytotoxic agents carefully weigh the associated increased risk of immune dysregulation against the benefit of disease control when making an educated decision on whether to initiate such treatments, particularly in non-curative settings.

Finally, our finding that CD8 T cell immunity is critical for survival in patients with hematologic malignancy and COVID-19 has implications for the vaccination of these patients. The current Food and Drug Administration-approved COVID-19 mRNA vaccines induce robust CD8 T cell responses in addition to humoral responses60,61,62,63. Our findings suggest that vaccination of patients with hematologic cancer might provide protection through T cell immunity, despite the likely absence of humoral responses. Ultimately, understanding how the immune response relates to disease severity, cancer type and cancer treatment will provide important insight into the pathogenesis of and protective immunity from SARS-CoV-2, which might have implications for the development and prioritization of therapeutics and vaccines in subpopulations of patients with cancer.

Methods

General design/patient selection

We conducted a prospective observational cohort study of patients with cancer hospitalized with COVID-19 (UPCC 06920). The University of Pennsylvania and Lancaster General Health institutional review boards (IRBs) approved this project, and informed consent was obtained from all patients. Adult patients with a current or previous diagnosis of cancer and hospitalized with a probable or confirmed diagnosis of COVID-19, as defined by World Health Organization (WHO) criteria64, within the University of Pennsylvania Health System between April 28, 2020, and September 15, 2020, were approached for consent. Participating hospitals included the Hospital of the University of Pennsylvania, Presbyterian Hospital, Pennsylvania Hospital and Lancaster General Hospital. The WHO’s definition of a probable case of SARS-CoV-2 is based on patients having a combination of high-risk symptoms, suspect chest imaging or death in the setting of respiratory distress and confirmed or probable contact to COVID-19, whereas a confirmed case is defined by someone with positive nucleic acid amplification testing or SARS-CoV-2 antigen testing in the setting of symptoms or probable COVID-19 contact64. We enrolled 114 patients across all four hospitals; 14 patients were excluded from the analyses owing to either low suspicion for COVID-19 infection or benign tumor diagnosis. Our final cohort of 100 patients included 48 females with a median age of 68 years (interquartile range (IQR) 57.5–77.5 years). The index date was defined as the first date of hospitalization within the health system for probable or confirmed COVID-19. Repeat hospitalizations within 7 d of discharge were considered within the index admission. Patients who died before being approached for consent were retrospectively enrolled. Patients were followed from the index date to 30 d after their discharge or until death by any cause. This study was approved by the IRBs of all participating sites.

Data collection

Baseline characteristics, including patient factors (age, gender, race/ethnicity, comorbidities, smoking history and body mass index (BMI)) and cancer factors (tumor type, most recent treatment, ECOG performance status and active cancer status), as well as COVID-19-related clinical factors, including change in levels of care, complications, treatments such as need for mechanical ventilation, laboratory values (complete blood counts with differentials and inflammatory markers including LDH, C-reactive protein, ferritin and IL-6), and final disposition were extracted by trained research personnel using standardized abstraction protocols. Active cancer status was defined by diagnosis or treatment within 6 months of admission date. Cancer treatment status was determined by the most recent treatment within 3 months before admission date.

The primary study endpoint was all-cause mortality from hospital admission until 30 d of hospital discharge. Disease severity was categorized using a modified version of the National Institutes of Health (NIH) ordinal scale including all post-hospitalization categories: 1: hospitalized, not requiring supplemental oxygen but requiring ongoing medical care; 2: hospitalized requiring any supplemental oxygen; 3: hospitalized requiring non-invasive mechanical ventilation or use of high-flow oxygen devices; 4: hospitalized receiving invasive mechanical ventilation or extracorporeal membrane oxygenation; and 5: death65. Disease severity was assessed every 7 d throughout a patient’s admission.

Statistical analysis

Cohort characteristics were compared using standard descriptive statistics. One-time imputation of missing values for ECOG performance status was done using the predicted mean value from an ordinal logistic model (proportional odds) of complete data. The ordinal model was fitted with forward stepwise selection, with entry at P = 0.1 and removal at P = 0.2, using clinical variables expected to be correlated with ECOG performance status. Those variables included several items in the Charlson Comorbidity Index and other clinical variables.

Univariate analyses examined demographic and clinical variables and cancer subtype (hematologic versus solid cancer) as predictors of death within 30 d of discharge and of ICU admission. ORs and 95% CIs were used to generate the forest plot illustration. Baseline laboratory tests were compared by cancer type using Mann–Whitney tests, and available RT–PCR data were used to determine length of RT–PCR positivity by cancer type.

Rates of ICU admission and death were calculated for the overall cohort and stratified by cancer subtype. A multivariate logistic model was used to examine the adjusted effect of solid versus hematologic designation. Covariates included demographic variables of age and sex (race was omitted for missing data). Covariates also included clinical variables that attained a P value of 0.1 in the univariate analyses. The final model included age, sex, smoking status, active disease status and ECOG performance status. A Cox proportional hazards regression model was also performed to determine the association between cancer type and mortality and identically adjusted for age, sex, smoking status, active cancer status and ECOG performance status. OS was measured from date of hospitalization to last follow-up or death; median OS was estimated using the Kaplan–Meier method; and differences by cancer subtype were compared using the log-rank test.

Immune profiling of patients hospitalized for COVID-19—MESSI

Information on clinical cohort, sample processing and flow cytometry is described in Mathew et al.14. Briefly, patients admitted to the Hospital of the University of Pennsylvania with a positive SARS-CoV-2 PCR test were screened and approached for informed consent within 3 d of hospitalization. Peripheral blood was collected from all patients, and clinical data were abstracted from the electronic medical record into standardized case report forms. All participants or their surrogates provided informed consent in accordance with protocols approved by the regional ethical research boards and the Declaration of Helsinki. Methods for PBMC processing, flow cytometry and antibodies used were previously described14. Missing data, including antibody and flow cytometry data, are largely driven by sample availability and are assumed to be unrelated to the immunologic endpoint of interest and other variables.

Serologic enzyme-linked immunosorbent assay

Enzyme-linked immunosorbent assays (ELISAs) were completed using plates coated with the receptor-binding domain of the SARS-CoV-2 spike protein, as previously described66. Briefly, before testing, plasma and serum samples were heat-inactivated at 56 °C for 1 h. Plates were read at an optical density (OD) of 450 nm using the SpectraMax 190 microplate reader (Molecular Devices). Background OD values from the plates coated with PBS were subtracted from the OD values from plates coated with recombinant protein. Each plate included serial dilutions of the IgG monoclonal antibody CR3022, which is reactive to the SARS-CoV-2 spike protein, as a positive control to adjust for inter-assay variability. Plasma and serum antibody concentrations were reported as arbitrary units relative to the CR3022 monoclonal antibody. A cutoff of 0.48 arbitrary units was established from a 2019 cohort of pre-pandemic individuals and used for defining seropositivity.

Flow cytometry and statistical analysis

Flow cytometry data from Mathew et al.14 were analyzed. Briefly, samples were acquired on a five-laser BD FACSymphony A5. Up to 2 × 106 live PBMCs were acquired per each sample. During the early sample acquisition period, three antibodies in the flow panel were changed. Samples of three patients with cancer and 12 patients without cancer were stained using this earlier flow panel. Flow features were visually assessed for batch variations against data from the later flow panel. The three patients with cancer were included with the rest of the cohort when batch effects were determined to have little effect on confidence in gated populations. These three patients with cancer were excluded in analysis of cell populations defined by proteins associated with the three changed antibodies. Owing to the heterogeneity of clinical and flow cytometric data, non-parametric tests of association were used throughout the study. Tests of association between unpaired continuous variables were performed by the Mann–Whitney test. Tests of association between paired continuous variables were performed by the Wilcoxon matched-pairs test. Tests of association between binary variables across two groups were performed using the Pearson chi-square test. All tests were performed using a nominal significance threshold of P < 0.05 with Prism version 9 (GraphPad Software) and Excel (Microsoft). CART analysis was performed using the R package rpart.

High-dimensional data analysis of flow cytometry data

UMAP analyses were conducted using the R package uwot. FlowSOM analyses were performed on Cytobank (https://cytobank.org). Lymphocytes and non-naive CD8 T cells were analyzed separately. UMAP analysis was performed using equal downsampling of 10,000 cells from each FCS file in lymphocytes and 1,500 cells in non-naive CD8 T cells, with a nearest neighbors of 15, minimum distance of 0.01, number of components of 2 and a Euclidean metric. The FCS files were then fed into the FlowSOM clustering algorithm. A new self-organizing map (SOM) was generated for both lymphocytes and non-naive CD8 using hierarchical consensus clustering. For each SOM, 225 clusters and ten metaclusters were identified. For lymphocytes, the following markers were used in the UMAP and FlowSOM analyses: CD45RA, PD-1, IgD, CXCR5, CD8, CD19, CD3, CD16, CD138, Eomes, TCF-1, CD38, CD95, CCR7, CD21, Ki-67, CD27, CD4, CX3CR1, CD39, Tbet, HLA-DR and CD20. For non-naive CD8 T cells, the following markers were used: CD45RA, PD-1, CXCR5, CD16, Eomes, TCF-1, CD38, CCR7, Ki-67, CD27, CX3CR1, CD39, Tbet and HLA-DR. Clusters with less than 0.5% frequency were excluded from downstream analysis. Heat maps were created using R package pheatmap. To group individuals based on lymphocyte landscape, pairwise EMD values were calculated on the lymphocyte UMAP axes using the emdist package in R. Resulting scores were hierarchically clustered using the hclust function in the stats package in R.

Immune profiling of patients hospitalized for COVID-19 at MSKCC

Patients admitted to MSKCC with a positive SARS-CoV-2 PCR test were eligible for inclusion. The project was approved by the IRB of MSKCC, and informed consent was obtained from all patients. For inpatients, clinical data were abstracted from the electronic medical record into standardized case report forms. Clinical laboratory data were abstracted from the date closest to research blood collection. Peripheral blood was collected into BD Horizon Dri tubes (BD, cat. no. 625642). Immunophenotyping of PBMCs via flow cytometry was performed in the MSKCC clinical laboratory. The lymphocyte panel included CD45 FITC (BD, 340664, clone 2D1, 1:40), CD56+ 16 PE (BD 340705, clone B73.1, 1:40; BD 340724, clone NCAM 16.2, 1:40), CD4 PerCP-Cy5.5 (BD 341653, clone SK3, 1:200), CD45RA PE-Cy7 (BD 649457, clone L48, 1:80), CD19 APC (BD 340722, clone SJ25C1, 1:80), CD8 APC-H7 (BD 641409, clone SK1, 1:80) and CD3 BV 421 (BD 562426, clone UCHT1, 1:80). The naive/effector T panel included CD45 FITC (BD 340664, clone 2D1, 1:40), CCR7 PE (BD 560765, clone 150503, 1:80), CD4 PerCP-Cy5.5 (BD 341653, clone SK3, 1:200), CD38 APC (BioLegend, 303510, clone HIT2, 1:20), HLA-DR V500 (BD 561224, clone G46-6, 1:80), CD45RA PE-Cy7 (BD 649457, clone L48, 1:80), CD8 APC-H7 (BD 641409, clone SK1, 1:80) and CD3 BV 421 (BD 562426, clone UCHT1, 1:80). The immune phenotypes were based on NIH vaccine consensus panels and the Human Immunology Project67. Samples were acquired on a BD FACSCanto using FACSDiva software.

ELISpot

ELIspot assays: 200,000 PBMCs per well were plated on Human IFN-γ/IL-2 Double-Color ELISpot plates (ImmunoSpot) in the presence of anti-human CD28 (0.2 ug ml−1) and with or without peptide (Miltenyi PepTivator EBV consensus or SARS-CoV-2 Select) at a final concentration of 0.3 µM. The EBV consensus pool (130-099-764) contains 43 MHC-I and MHC-II peptides derived from 13 EBV proteins, whereas the SARS-CoV-2 Select peptide pool (130-127-309) contains 88 lyophilized MHC-I- and MHC-II-restricted peptides derived from the whole proteome of SARS-CoV-2. Plates were incubated in a 37 °C humidified incubator for 18 h, after which plates were stained as per manufacturer instructions and quantified using an automated ImmunoSpot S6 Analyzer. Results are expressed in spot-forming units (SFU) per 106 PBMCs.

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.