CD8+ T cells contribute to survival in COVID-19 patients with hematologic cancers

Abstract

Cancer patients have high mortality from COVID-19, and the immune parameters that dictate clinical outcomes remain unknown. In a cohort of 100 cancer patients hospitalized for COVID-19, patients with hematologic cancers had higher mortality relative to solid cancers. In two additional cohorts, flow cytometric and serologic analyses demonstrated that solid cancer and non-cancer patients had a similar immune phenotype during acute COVID-19 whereas hematologic cancer patients had impairment of B cells and SARS-CoV-2-specific antibody responses. Despite the impaired humoral immunity and high mortality in hematologic cancer patients with COVID-19, those with a greater number of CD8 T cells had improved survival, including those treated with anti-CD20 therapy. Further, 77% of hematologic cancer patients had detectable SARS-CoV-2 specific T-cell responses. Thus, CD8 T cells may influence recovery from COVID-19 when humoral immunity is deficient. These observations suggest that CD8 T cell responses to vaccination might provide protection in hematologic cancer patients even in the setting of limited humoral responses.

Introduction

Severe infection with Coronavirus Disease 2019 (COVID-19) has been linked to immune dysregulation, including impaired or delayed production of type I and type III interferons^1–5, marked lymphopenia^6–10, and a paradoxical increase in pro-inflammatory cytokines, such as TNFα, IL-1β, and IL-6^{1,4,6,11–13}. Alteration of T cell compartments include increases in effector and activated CD4 and CD8 T cells^14–17, while changes in B-cell and humoral compartments include robust plasmablast differentiation and production of SARS-CoV-2-reactive IgM and IgG antibodies^14,18–20. More recently, distinct immunophenotypes have been associated with COVID-19 disease severity and trajectory^3,4,11,14,15. Understanding how clinical features impact the host immune response to SARS-CoV-2 will elucidate determinants of disease severity.

Cancer patients have an increased risk of severe COVID-19 ^21–24 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% amongst COVID-19 patients with hematologic cancer ^{21,24,27–34}. 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 cancer population³⁶. Further, there are limited data on the immune response to SARS-CoV-2 in cancer patients, 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 cancer patients with 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 cancer patients hospitalized with COVID-19 (COVID-19 Outcomes in Patients with Cancer, COPE, see Methods). The median age of this cohort was 68 years; 48% were female, 54% Black, and 57% were current or former smokers (Table 1). In terms of cancer-specific factors, 78% of patients had solid cancers, with prostate and breast cancers 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 subjects required ICU level care, and 38% of patients died within 30 days of admission (Supplemental Table 1). Demographics by tumor type are available in Supplemental Table 2.

Table 1 |.

COPE: Patient demographics and clinical characteristics.

	Total (N=100)
Age, median (IQR)	68 (57.5–77.5)
Gender, female	48 (48%)
Race
Black	54 (54%)
White	33 (33%)
Asian	4 (4%)
Hispanic	3 (3%)
Unknown	6 (6%)
Smoking History, Ever+	57 (57%)
Comorbidities
Cardiac	78 (78%)
Pulmonary	41 (41%)
Use of immunosuppressive drugs++	30 (30%)
BMI, median (IQR)	26.84 (23.2–31.5)
Cancer Type
Solid malignancy	78 (78%)
Genitourinary	19 (19%)
Breast	14 (14%)
Gastrointestinal	14 (14%)
Thoracic	9 (9%)
Other+++	8 (8%)
Gynecologic	7 (7%)
Head and Neck	4 (4%)
Sarcoma	3 (3%)
Heme malignancy	22 (22%)
Lymphoma	10 (10%)
Leukemia	7 (7%)
Myeloma	3 (3%)
MDS/MPN	2 (2%)
Cancer Status, Active#	46 (46%)
Cancer treatment in last 3 months
Active surveillance/surgery	53 (53%)
Cytotoxic Chemotherapy	24 (24%)
Hormone therapy	15 (15%)
Other*	8 (8%)
ECOG Performance Status	N=73
0–1	37 (50.7%)
2	13 (17.8%)
3–4	23 (31.5%)

⁺Current or prior smoker

⁺⁺Exposure to immunosuppressive medications not including cancer treatment

⁺⁺⁺Tumor types with less than 2 subjects: CNS-2, Thyroid-2, Thymus-1, Neuroendocrine-1

^#Diagnosis or treatment within 6 months

^*Single agent immunotherapy, targeted therapy, monoclonal antibodies

We performed univariate analyses to identify factors associated with all-cause mortality in the period between hospital admission and 30 days post-discharge. We included relevant covariates, including patient factors such as age, race, gender, and smoking history (ever versus never)^37–40; cancer-specific factors including ECOG performance status^33,35, status of cancer (e.g., active versus remission)³⁴; cancer type (e.g., heme versus solid cancer)^{27,32,34,41,42}; and cancer treatment^33,35. Current or prior smoking (p = 0.028), poor ECOG performance (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 cancers appeared to have an increased risk of mortality relative to solid cancers (55% versus 33% respectively, p=0.075) (Supplemental Table 1)^{21,27,32–34,41}. However, similar to published literature, cancer treatment, including cytotoxic chemotherapy, was not significantly associated with COVID-19 mortality^{27,28,32,34,41}.

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

Data are presented as odds ratios 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, monoclonal antibodies.

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 (OR 3.3, 95% 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 (mOS) not reached for patients with solid cancers vs 47 days for patients with hematologic cancers, p-value=0.030) and Cox regression models (Table 2, HR 2.56, 95% CI 1.19–5.54, p=0.017). Moreover, patients with hematologic cancers had higher levels of some inflammatory markers on admission laboratory testing, including ferritin, IL-6, and LDH (Fig. 2b, Extended Data Fig. 1a,b). We also assessed viral persistence in 43 patients with repeat RT-PCR testing (37 solid and 6 hematologic). Both hematologic and solid cancer patients had prolonged viral clearance, with some greater than 60 days (Extended Data Figure 1c,d, Supplemental 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.

	Heme	Solid
Death within 30 days of discharge
Event rate (%)	12 (54.6%)	26 (33.3%)
Unadjusted OR (95% CI)	2.4 (0.82–7.06)	ref
Adjusted OR (95% CI)+	3.3 (1.01–10.8)	ref
Adjusted HR (95% CI)+	2.6 (1.19–5.54)	ref

⁺Logistic regression computed odds ratio (OR) and Cox regression computed hazard ratio (HR), respectively. Adjusted for age, gender, smoking status, active cancer status, and ECOG performance status.

Fig. 2 |. Hematologic cancer is an independent risk factor for COVID-19 related mortality.

(a) Kaplan Meier curve for COVID-19 survival of patients with solid (n=77) and hematologic (n=22) cancer. Cox regression-computed hazard ratio for mortality in hematologic vs 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 solid (n=62) and hematologic (n=15) cancer hospitalized for COVID-19. (All) Significance determined by two-sided Mann Whitney test: *p<0.05, **p<0.01. Median and 95% CI shown.

Impaired SARS-CoV-2 antibody responses in hematologic cancer

To determine whether SARS-CoV-2 infected cancer patients might exhibit an altered immune landscape, we leveraged an observational study of hospitalized COVID-19 patients at the University of Pennsylvania Health System where blood was collected (MESSI-COVID)¹⁴. This analysis included 130 subjects with flow cytometric and/or serologic analysis. Twenty-two subjects had active cancer (Supplemental Tables 4, 5), including patients undergoing cancer-directed therapies such as chemotherapy, immunotherapy, or B cell directed therapies (Supplemental Table 6). Patients with active cancer were older and predominantly female; both groups had a similar timeframe of symptom onset and disease severity (Fig. 3a, Supplemental Table 4). However, cancer patients had a higher all-cause mortality (36.4% versus 11.1%, Fig. 3a), consistent with our COPE cohort and other reported cohorts^{21, 24,27,28}.

Fig. 3 |. High dimensional analyses reveal immune phenotypes associated with mortality and distinct phenotypes between solid and hematologic cancers.

(a) Demographic and mortality data for MESSI cohort at Penn. (b) Relative levels of SARS-CoV-2 IgG (No Cancer vs. Heme, p=0.001; Solid vs. Heme, p=0.007) and IgM (No Cancer vs. Heme, p=0.003; No Cancer vs. Solid, p=0.03); solid (n=14) and hematologic (n=7) cancer patients and non-cancer patients (n=108). (c) (Left) Global UMAP projection of lymphocyte populations for all 45 patients pooled. (Right) Hierarchical clustering of Earth Mover’s Distance (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) Heatmap 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; Cluster 1,2,3,4 n=40). Mortality significance determined by Pearson Chi Square test. Severity assessed with NIH ordinal scale for COVID-19 clinical severity (1: Death; 8: Normal Activity)¹⁵. (g) UMAP projections of concatenated lymphocyte populations for solid cancer, hematologic cancer, and non-cancer patients. (h) CD8, CD4 (No Cancer vs. Heme, p=0.003; Solid vs. Heme, p=0.01) and B cell (No Cancer vs. Heme, p=0.008; No Cancer vs. Solid, p=0.03; Solid vs. Heme, p=0.02) frequencies in healthy donors (n=33), non-cancer (n=108), solid cancer (n=7), and heme cancer (n=4). (i) UMAP projection of non-naive CD8 T cell clusters identified by FlowSOM. (j) (Top) UMAP projections of non-naïve CD8 T cells for non-cancer and cancer patients. (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 HD (n=30), non-cancer (n=110), and cancer patients (n=8) (p=0.03). (l) Representative flow plots and frequency of HLA-DR and CD38 co-expression in HD (n=30), non-cancer (n=110), solid cancer (n=7), and hematologic cancer (n=3) patients (gated on non-naïve CD8) (No Cancer vs. Heme, p<0.0001; Solid vs. 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 shown.

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. Cancer patients had significantly decreased SARS-CoV-2-specific IgG and IgM antibodies compared to non-cancer patients (Extended Data Fig. 2a). This was largely due to hematologic cancer patients, the majority (6/7) of whom had IgM and IgG levels below the cutoff of positivity of 0.48 arbitrary units (Fig. 3b, Extended Data Fig. 2b), while those with solid cancers had IgG and IgM antibody responses that were more comparable 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 responses^43,44. We therefore performed exploratory high-dimensional analysis on the lymphocyte compartment of 45 patients, including 44 patients with COVID-19 (36 non-cancer, 6 solid cancer, and 2 hematologic cancer patients), and one control non-COVID-19 subject. UMAP (Uniform Manifold Approximation and Projection) 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 5 clusters of patients with similar lymphocyte profiles (Fig. 3c). Differences between 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. 3 d,e and Extended Data Fig. 3b). EMD cluster 3 had decreased T cell 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 5 patterns of immune responses with clinical and serological variables. EMD cluster 5 patients with depleted T cells had the highest mortality and disease severity, despite generating SARS-CoV-2-specific IgM and IgG antibodies (Fig. 3f, 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, 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 cancers

We next explored the role of cancer subtype on immune phenotype. Four out of the 6 solid cancer patients were in EMD cluster 2, with a balanced immune phenotype (Fig. 3e). In contrast, both hematologic cancer patients were in EMD cluster 1, which had marked depletion of CD4 and B cells. Indeed, UMAP projections showed that while solid cancer patients had an immune landscape similar to non-cancer patients, the two hematologic cancer patients 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. COVID-19 patients with hematologic cancers had a significantly lower frequency of CD4 and B cells compared to solid cancer patients, non-cancer patients, and healthy donors without COVID-19 (Fig. 3h). As T follicular helper cells (Tfh) 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 cancers had low circulating Tfh (PD1+ CXCR5+) and plasmablast responses (CD19+CD27^hiCD38^hi) and decreased CD138 expression (Extended Data Fig. 4a). Thus, patients with hematologic malignancy appear to have quantitative defects in CD4 and B cells that may be required for effective SARS-CoV-2-specific antibody responses.

As patients with hematologic cancers had a preserved frequency of CD8 T cells. We performed FlowSOM clustering analysis on non-naïve CD8 T cells from 118 COVID-19 patients 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 CD45RA^loCD27^lo effector memory cells (clusters 2 and 3) and CD45RA+ TEMRA cells (cluster 1). The memory island was composed of CCR7^lo transitional memory (cluster 5), effector memory (clusters 7 and 8), and CCR7^hi central memory cells (cluster 9). 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-naïve CD8 data demonstrated preferential enrichment of cells expressing HLA-DR and CD38 in cancer patients compared to non-cancer patients (Fig. 3j). Indeed, cancer patients had higher frequencies of activated HLA-DR, CD38, and Ki67-expressing FlowSOM clusters (clusters 3, 4, and 5) compared to non-cancer patients 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 cancers; patients with solid cancers and those without cancer had comparable levels of activation (Fig. 3l). Altogether, solid cancer patients with COVID-19 had an immune landscape similar to non-cancer COVID-19 patients. 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 may potentially compensate for blunted humoral immune responses in patients with hematologic malignancies.

CD8 T cells influence survival in patients with hematologic cancers

To more rigorously explore the role of T and B cell immunity in SARS-COV-2 infected cancer patients, we examined a cohort of cancer patients hospitalized with COVID-19 at the 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 Penn, 81% of the cohort was white (Fig. 4a, Supplemental Table 7,8). Consistent with the Penn COPE and MESSI cohorts, patients with hematologic cancers did poorly, with a mortality rate of 44.4% (Fig. 4a, Supplemental Table 7). In the MSKCC cohort, both CD4 and CD8 T cells were significantly decreased in patients with active solid and hematologic cancers, compared with patients in clinical remission (Extended Data Fig. 5a). Moreover, despite the fact that a substantial number of patients with hematologic cancers from the MSKCC cohort received convalescent plasma (Supplemental Table 7), they had a significant defect in SARS-CoV-2-specific IgG and IgM responses as compared to solid cancers (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 hematologic cancer patients with COVID-19.

(a) Demographic and mortality data of MSKCC cohort. (b) (Left) Hierarchical clustering of Earth Mover’s Distance (EMD) using 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; Cluster 1,2,4 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 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 aCD20 n=7; heme other tx n=5). (f) Absolute CD8 and CD4 T cell counts 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 (p=0.003) counts in hematologic cancer patients (alive n=17; dead n=18). (h) Kaplan-Meier curve for survival in hematologic cancer patients stratified by CD8 T cell counts (threshold = 55.9; log-rank hazard ratio) (>=55.9 n=28; <55.9 n=13). CD8 count threshold determined by Classification and Regression Tree (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) spot forming units (SFU) per million PBMCs in heme cancer patients (n=13). Significance determined by two-sided Wilcoxon test. (j) Proportion of heme cancer patients with detectable IFN-γ SFU, after background subtraction (no peptide). Significance determined by Chi Square test. (k) Simple linear regression between IFN-γ SFU and percent activated CD8 T cells in heme cancer patients that 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 shown.

EMD and clustering of flow cytometry data from 20 solid cancer, 31 hematologic cancer, and 6 remission patients identified 4 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 cancer patients, was not identified in the MSKCC cancer 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).

Solid cancer patients in the MSKCC cohort were present in all 4 clusters, with a majority in cluster 4, while hematologic cancer patients were predominantly in clusters 1 and 4 and absent from cluster 2 (Extended Data Figure 7a and 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. Hematologic cancer patients with phenotype 4 had a mortality of 62% versus 9% in patients with solid cancers (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 Figure 7e). However, patients with hematologic cancers had near-complete abrogation of B cells (phenotype 4A) that corresponded with a mortality rate of 62% (Extended Fig 7c and f). In contrast, patients with solid cancers had intact B cells counts (phenotype 4B) with a mortality of 9% (Extended Data Fig 7c and f). Thus, in a setting with similar CD4 and CD8 T cell numbers, B cell depletion was associated with higher mortality.

Anti-CD20 therapy (αCD20) with rituximab or obinutuzumab-containing regimens depleted circulating B cells and significantly impaired SARS-CoV-2-specific IgG and IgM responses (Fig. 4e, Extended Data Fig. 8a). αCD20 was not associated with quantitative changes in CD4 and CD8 T cells. However, patients treated with anti-CD20 therapy exhibited significant reductions in CD4 and CD8 naïve and memory T cells, with a relative skewing towards effector differentiation and an activated HLA-DR+CD38+ phenotype (Extended Data Fig. 8b,c). These changes were not seen in αCD20-treated hematologic patients without COVID-19 (Extended Data Fig. 8d,e). Importantly, 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, αCD20 treated patients 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 cancers, including those on chemotherapy who also had quantitative (Extended Data Fig. 8a), and possibly qualitative B cell defects. Hematologic cancer patients who survived had higher CD8 T cell count (Fig. 4g), which was not observed in solid cancer patients (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, Fig 4g). Classification and Regression Tree Analysis (CART) identified a CD8 T cell level that was predictive of survival after COVID-19 in patients with hematologic cancers (Fig. 4h). Thus, patients with hematologic cancers, in the setting of defective humoral immunity, were more highly dependent on adequate CD8 T cell counts than patients with solid cancers.

Finally, to assess the antigen-specificity of T cell responses in SARS-CoV-2 infected cancer patients, we stimulated PBMC from 13 hematologic cancer patients and 10 solid cancer patients using MHC-I and MHC-II restricted SARS-CoV-2 and EBV peptide pools and measured IFN-γ and IL-2 responses using ELISpot (Extended Data Figure 10a). After background (no-peptide) subtraction, 10 out of 13 (77%) of hematologic cancer patients had SARS-CoV-2- specific IL-2 and/or IFN-γ responses (Figure 4i), including 8 out of 8 (100%) of patients treated with αCD20 (Figure 4j, Extended Data Fig 10b). In general, these SARS-CoV-2-specific responses were greater in magnitude than EBV-specific responses (Figure 4i), and hematologic cancer patients had greater SARS-CoV-2 antigen-specific T cell responses than solid cancer patients (Extended Data Fig 10c). Moreover, in hematologic cancer patients 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 10d). This correlation was not seen in patients who died (Extended Data Fig 10e). Taken together, these findings suggest that for hematologic cancer patients who recover from COVID-19, increased peripheral CD8 T cell activation may reflect an appropriate induction of SARS-CoV-2-specific T cell responses. In contrast, peripheral CD8 T cell activation may be uncoupled from SARS-CoV-2 specific T cell responses in patients who ultimately succumb to 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 non-cancer patients and in mouse models^46,47.

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 cancer patients might shed 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 establish, 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 cancer patients. 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 severity^46,47. In fact, greater B cell responses were associated with higher mortality in both solid and liquid cancers. B cell responses may be a marker of disease severity, as seen with plasmablasts^14,18 and neutrophils^18,49,50 in severe COVID-19. Alternatively, some components of the B cell and humoral responses may be aberrant and pathogenic, as may be the case with autoantibodies targeting type I interferons in severe COVID-19⁵¹. 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 individuals^46,53–56, 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 anti-CD20, chemotherapy, or Bruton tyrosine kinase (BTK) inhibition. Indeed, we could identify SARS-COV-2-specific T cell responses in the majority of hematologic cancer patients, including all patients treated with αCD20, and they were generally greater in magnitude than solid cancer patients. 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 may inform on the need for closer monitoring and a lower threshold for hospitalization in COVID-19 patients with hematologic malignancies. Furthermore, the clinical benefit of dexamethasone, which demonstrated an overall mortality benefit in hospitalized COVID-19 patients but is known to suppress CD8 T cell responses⁵⁸, should be investigated further in patients who recently received anti-CD20 therapy.

Our findings do not exclude a key role for humoral immunity in protection from COVID-19 mortality. Indeed, while patients with solid cancers had a cellular and humoral immune landscape that was similar to patients without cancer⁵⁹, patients with hematologic cancers 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 may 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 lymphomas, but also for debulking or palliation in patients with indolent lymphomas. Based on our data, we would 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 hematologic malignancy patients with COVID-19 has implications for the vaccination of these patients. The current FDA-approved COVID-19 mRNA vaccines induce robust CD8 T cell responses in addition to humoral responses^60–63. Our findings suggest that vaccination of hematologic patients 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 may have implications for the development and prioritization of therapeutics and vaccines in cancer subpopulations.

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 IRB approved this project and informed consent was obtained from all patients. Adult patients with a current or prior diagnosis of cancer and hospitalized with a probable or confirmed diagnosis of COVID-19, as defined by the WHO criteria⁶⁴, within the University of Pennsylvania Health System (UPHS) 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 while 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 4 hospitals, 14 patients were excluded from the analyses due to either low suspicion for COVID-19 infection, or benign tumor diagnosis. Our final cohort of 100 subjects included 48 females with a median age of 68 (IQR 57.5–77.5). 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 days of discharge were considered within the index admission. Patients who died prior to being approached for consent were retrospectively enrolled. Patients were followed from the index date to 30-days following their discharge or until death by any cause. This study was approved by the institutional review boards of all participating sites.

Data Collection

Baseline characteristics including patient (age, gender, race/ethnicity, comorbidities, smoking history, body mass index) and cancer (tumor type, most recent treatment, ECOG performance status, active cancer status) factors 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, CRP, 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 prior to admission date.

The primary study endpoint was all-cause mortality from hospital admission until 30-days of hospital discharge. Disease severity was categorized using the a modified version of the 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 noninvasive mechanical ventilation or use of high-flow oxygen devices; 4, hospitalized receiving invasive mechanical ventilation or extracorporeal membrane oxygenation (ECMO); 5, death⁶⁵, and was assessed every 7 days throughout a patients admission.

Statistical Analysis

Cohort characteristics were compared using standard descriptive statistics. One-time imputation of missing values for ECOG 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 0.2, using clinical variables expected to be correlated with ECOG performance status. Those variables included several items in the Charlson and severity score, and other clinical variables.

Univariate analyses examined demographic and clinical variables and cancer subtype (hematologic versus solid cancer) as predictors of death within 30 days of discharge and of ICU admission. Odds ratios 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 was 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. Overall survival (OS) was measured from date of hospitalization to last follow up or death and the median OS was estimated using Kaplan-Meier method and differences by cancer subtype compared using 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, Science 2020. 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 days of hospitalization. Peripheral blood was collected from all subjects 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 (ELISA)

ELISAs were completed using plates coated with the receptor binding domain (RBD) of the SARS-CoV-2 spike protein as previously described⁶⁶. Briefly, prior to testing, plasma and serum samples were heat-inactivated at 56°C for 1 hour. Plates were read at an optical density (OD) of 450nm 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, Science 2020¹⁴ was analyzed. Briefly, samples were acquired on a 5 laser BD FACS Symphony A5. Up to 2 × 10^6 live PBMC were acquired per each sample. During the early sample acquisition period, three antibodies in the flow panel were changed. Three cancer patients and twelve non-cancer patients were stained using this earlier flow panel. Flow features of these patients were visually assessed for batch variations against data from the later flow panel. The three cancer patients were included with the rest of the cohort when batch effects were determined to have little impact on confidence in gated populations. These three cancer patients were excluded in analysis of cell populations defined by proteins associated with the three changed antibodies.Due 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 Mann-Whitney test. Tests of association between paired continuous variables were performed by Wilcoxon matched-pairs test. Tests of association between binary variables across two groups were performed using 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 Office Suite). Classification and Regression Tree analysis (CART) was performed using R package ‘rpart’.

High dimensional data analysis of flow cytometry data

UMAP analyses were conducted using 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 down sampling of 10000 cells from each FCS file in lymphocytes and 1500 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 10 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, T-bet, 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, T-bet, and HLA-DR. Clusters with less than 0.5% frequency were excluded from downstream analysis. Heatmaps were created using R package pheatmap. To group individuals based on lymphocyte landscape, pairwise Earth Mover’s Distance (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, MSKCC

Patients admitted to Memorial Sloan Kettering Cancer Center with a positive SARS-CoV-2 PCR test were eligible for inclusion. Project was approved by the IRB of the Memorial Sloan Kettering Cancer Center, 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#625642). Immunophenotyping of peripheral blood mononuclear cells 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 Facs Canto using FACSDiva software.

Enzyme-linked immunosorbent (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) and with or without peptide (Miltenyi Peptivator EBV consensus or SARS-CoV-2 Select) at a final concentration of 0.3 micromolar. The EBV consensus pool (130–099-764) contains 43 MHC class I and class II peptides derived from 13 EBV proteins while the SARS-CoV-2 select peptide pool (130–127-309) contains 88 lyophilized MHC class I and class II restricted peptides derived from the whole proteome of SARS-CoV-2. Plates were incubated in a 37C humidified incubator for 18 hours, after which plates were stained as per manufacturer’s instructions and quantified using an automated ImmunoSpot S6 Analyzer. Results are expressed in Spot Forming Units (SFU) per 10^6 PBMCs.

Extended Data

Extended Data Fig. 1 |. Inflammatory markers, blood cell counts, and viral load in cancer patients with COVID-19.

Clinical laboratory values for (a) inflammatory markers and (b) cell counts in solid (n=62) and hematologic (n=21) cancer patients. Repeat SARS-CoV-2 RT-PCR testing results (c) from first positive test to last performed test and (d) from first positive test to last positive test. (All) Significance determined by two-sided Mann Whitney test.

Extended Data Fig. 2 |. SARS-CoV-2 antibody levels in cancer patients with COVID-19.

(a) Relative levels of SARS-CoV-2 IgG (p=0.02) and IgM (p=0.0008) in non-cancer (n=108) and cancer (n=21) patients. (b) Relative IgG levels in cancer patients. Each dot represents a cancer patient (Heme: Red; Solid: Yellow). (All) Significance determined by two-sided Mann Whitney test: *p<0.05, ***p<0.001. Median and 95% CI shown.

Extended Data Fig. 3 |. Dimensionality reduction and EMD clustering of MESSI cohort.

(a) UMAP projections of lymphocytes with indicated protein expression. (b) Frequencies of CD19+, CD3+, CD3+CD8+, and CD3+CD4+ cells of patients in each EMD cluster (Cluster 1 n=7; Cluster 2 n=16; Cluster 3 n=6; Cluster 4 n=10; Cluster 5 n=5). (All) Median and 95% CI shown.

Extended Data Fig. 4 |. Cellular phenotyping of COVID-19 patients with cancer.

(a) Frequencies of circulating T follicular helper cells (cTfh), plasmablasts (No Cancer vs. Heme, p=0.0001; Solid vs. Heme, p=0.006), and CD138 expression on plasmablasts (HD n=33; non-cancer n=108; solid cancer n=7; heme cancer n=3). (b) UMAP projection of non-naïve CD8 T cells with indicated protein expression. (c) Heatmap showing expression patterns of various markers, stratified by FlowSOM clusters. Heat scale calculated as column z-score of MFI. (d) Frequencies of CD8 subsets: naive (CD45RA+CD27+CCR7+), central memory (CD45RACD27+CCR7+), transition memory (CD45RA-CD27+CCR7-) (p<0.0001), effector memory (CD45RA-CD27-CCR7-), and TEMRA (CD45RA+CD27-CCR7-) (p=0.002) (HD n=33; non-cancer n=108; cancer n=9). (e) (Left) HLA-DR and CD38 co-expression in concatenated activated clusters (3, 4, and 5) and associated UMAP localization. (Right) Frequency of clusters 3 (p=0.03) and 5 (HD n=30; non-cancer n=110; solid-cancer n=8). (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 shown.

Extended Data Fig. 5 |. Cellular, serologic, and clinical features in solid and hematologic cancer patients with COVID-19.

(a) Absolute counts of CD4 (Remission vs. Heme, p=0.01; Remission vs. Solid, p=0.02), CD8 (p=0.02), and CD19 (Remission vs. Heme, p=0.008; Solid vs. Heme, p=0.0003) expression in remission (n=11), solid cancer (n=23), and hematologic cancer (n=41) patients. (b) Relative levels of SARS-CoV-2 IgG (p=0.003) and IgM (p=0.0007) in solid (n=11) and hematologic cancer (n=14) patients. (c) Severity (NIH ordinal scale for COVID-19 clinical severity) and RT-PCR cycle threshold (remission n=9; solid n=25; heme n=28) (Lower Ct: Higher viral load). (d) NIH ordinal scale for COVID-19 clinical severity. (All) Significance determined by two-sided Mann Whitney test: *p<0.05, **p<0.01, ***p<0.001. Median and 95% CI shown.

Extended Data Fig. 6 |. Dimensionality reduction and EMD clustering of MSKCC cohort.

(a) UMAP projections of lymphocytes with indicated protein expression. (b) Absolute counts of CD19+, CD3+, CD3+CD8+, and CD3+CD4+ cells of patients in each EMD cluster (Cluster 1 n=18; Cluster 2 n=6; Cluster 4 n=26; Cluster 5 n=7). (All) Median and 95% CI shown.

Extended Data Fig. 7 |. EMD Cluster 4 drives differences in mortality between hematologic and solid cancer patients.

(a) EMD cluster distributions of heme and solid cancer patients. (b) Number of patients with hematologic, solid, and remission cancer status within each EMD cluster. (c) Mortality of patients within each EMD cluster for hematologic and solid cancers. (d) RT-PCR cycle threshold of solid and heme cancer patients in EMD cluster 4 (solid n=11; heme n=11) (p=0.02). (e) Absolute CD8 and CD4 T cell counts for subjects in EMD cluster 4 stratified by solid (n=11) and heme (n=13) cancer. (f) Global UMAP projections of lymphocytes for subjects in EMD cluster 4: (Left) Hematologic cancer; (Middle) Solid cancer. (Right) Absolute B cell counts for subjects in EMD cluster 4 stratified by solid (n=11) and heme (n=13) cancer (p=0.004). (All) Significance determined by two-sided Mann Whitney test: *p<0.05, **p<0.01. Median and 95% CI shown.

Extended Data Fig. 8 |. Effect of cancer treatment on T cell differentiation in COVID-19.

(a) Absolute counts of CD4, CD8, and CD19 expressing cells. Frequencies of (b) CD4 and (c) CD8 T cell subsets in cancer patients treated with immune checkpoint blockade therapies, chemotherapies, and B cell depleting therapies. Frequencies of (d) CD4 and (e) CD8 T cell subsets in heme cancer patients with and without COVID-19, and with and without αCD20. Naive (CD45RA+CCR7+), CM (CD45RA-CCR7+), EM (CD45RA-CCR7-), TEMRA (CD45RA+CCR7-). (All) Remission n=11, obs n=12, chemo only n=9, solid ICB n=7, heme αCD20 n=10, non-COVID heme no tx n=5, and non-COVID heme αCD20 n=5. Significance determined by two-sided Mann Whitney test. Median and 95% CI shown.

Extended Data Fig. 9 |. Association of mortality with cell counts and viral load.

(a) Mortality, severity, and RT-PCR cycle threshold stratified by cancer treatment (remission n=9; solid obs n=6; solid tx n=19; heme obs n=5; heme chemo n=4; heme αCD20 n=10). Severity assessed with NIH ordinal scale for COVID-19 clinical severity. (b) Recent cancer treatment of patients in each EMD cluster. (c) Mortality of patients treated with B cell depleting therapy in EMD cluster 1 (red) and EMD cluster 4 (blue). (d) RT-PCR cycle threshold of patients treated with αCD20 therapy (alive n=7; dead n=3). (e) Absolute counts of CD8, CD4, and CD19 (p=0.004) cells in solid cancer patients (alive n=16; dead n=7). (All) Significance determined by two-sided Mann Whitney test: **p<0.01. Median and 95% CI shown.

Extended Data Fig. 10 |. EBV and SARS-CoV-2 ELISpot in COVID-19 cancer patients.

ELISpot was performed after stimulation of PBMC with no peptide, EBV, and SARS-CoV-2 peptide pools. (a) IFN-γ (No Peptide vs. COVID, p=0.003) and IL-2 (No Peptide vs. COVID, p=0.03; EBV vs. COVID, p=0.02) spot forming units (SFU) per million PBMC in heme cancer patients (n=13). Significance determined by Wilcoxon test. (b) IFN-γ SFU per million PBMC in heme cancer patients treated with (n=8) and without (n=5) αCD20 with no-peptide background condition subtracted. (c) IFN-γ and IL-2 SFU between solid (n=10) and heme (n=13) cancer patients. Simple linear regression between IFN-γ SFU per million cells and percent activated CD4 (d) and CD8 (e) T cells in alive (n=8) and dead (n=5) heme cancer patients. (All) Significance determined by two-sided Mann Whitney test. *p<0.05, **p<0.01. Median and 95% CI shown.

Data Availability Statement:

Flow Cytometry data collected for the MESSI:COVID cohort was deposited in FlowRepository. http://flowrepository.org/id/FR-FCM-Z3XT. Raw data are included in Supplementary Table 10. External data requests can be directed to the corresponding authors, who will respond within one week and help facilitate the request. Access to clinical datasets for the COPE study, clinical flow cytometry and clinical metadata from the MSKCC cohort will be available based on approval through the IRB of the University of Pennsylvania and Memorial Sloan Kettering Cancer Center and may be subject to patient privacy.