Proteomic analysis of circulating immune cells identifies cellular phenotypes associated with COVID-19 severity

Abstract

Certain serum proteins, including C-reactive protein (CRP) and D-dimer, have prognostic value in patients with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Nonetheless, these factors are non-specific, providing limited mechanistic insight into the peripheral blood mononuclear cell (PBMC) populations that drive the pathogenesis of severe COVID-19. To identify cellular phenotypes associated with disease, we performed a comprehensive, unbiased analysis of total and plasma-membrane PBMC proteomes from 40 unvaccinated individuals with SARS-CoV-2, spanning the whole disease spectrum. Combined with RNA sequencing (RNA-seq) and flow cytometry from the same donors, we define a comprehensive multi-omic profile for each severity level, revealing that immune-cell dysregulation progresses with increasing disease. The cell-surface proteins CEACAMs1, 6, and 8, CD177, CD63, and CD89 are strongly associated with severe COVID-19, corresponding to the emergence of atypical CD3⁺CD4⁺CEACAM1/6/8⁺CD177⁺CD63⁺CD89⁺ and CD16⁺CEACAM1/6/8⁺ mononuclear cells. Utilization of these markers may facilitate real-time patient assessment by flow cytometry and identify immune populations that could be targeted to ameliorate immunopathology.

Graphical abstract

Potts et al. describe a multiplexed proteomic analysis of immune cells obtained from individuals with COVID-19, identifying CEACAMs 1/6/8, CD177, CD63, and CD89 as cell-surface markers upregulated in severe disease. Phenotyping identifies emergence of unusual CD4⁺ T cell and CD16⁺ monocyte populations expressing these markers in severe disease.

Introduction

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) continues to present a public health crisis due to a slow global rollout of vaccination programs and emergence of novel virus variants. Viral pathogenesis comprises an initial stage of virus replication followed by immune-cell recruitment and cytokine production. Most infected individuals generate an effective immune response that achieves viral clearance without excessive tissue damage, presenting with no or mild symptoms. However, a minority experience severe disease, driven by a dysregulated and hyperactive immune response and characterized by high levels of pro-inflammatory mediators such as interleukin (IL)-6 and tumor necrosis factor alpha (TNFα).^{1 , 2} The consequent enhanced vascular permeability, thrombosis, and tissue damage can lead to severe pneumonia, acute respiratory distress syndrome, multiple organ failure, and death.^{3 , 4}

Several studies have aimed to define perturbations in circulating immune-cell subsets and secreted factors during severe COVID-19, utilizing techniques such as bulk and single-cell transcriptomics, cytometry panels, and plasma proteomics. These have shown that severe COVID-19 is associated with profound peripheral lymphopenia^{5 , 6 , 7 , 8} characterized by recruitment of natural killer (NK) cells to the lung tissue from peripheral blood and expansion of activated yet functionally impaired^{9 , 10} inflammatory NK cells expressing cytotoxic factors and interferon-stimulated genes.^{10 , 11 , 12} Unusually for a viral infection, profound peripheral neutrophilia is also observed^{13 , 14} due to release of immature, inflammatory neutrophils via emergency myelopoiesis and a hyperinflammatory phenotype reminiscent of bacterial sepsis.^{15 , 16 , 17} Circulating CD4⁺ and CD8⁺ T cells are depleted,^{5 , 7 , 8 , 18} concurrent with expansion of activated effector cell subpopulations,^{9 , 19} and express high levels of exhaustion markers such as programmed cell death protein 1 (PD-1) and T cell immunoglobulin and mucin domain-containing protein 3 (TIM3).^{9 , 19 , 20} Significant changes are observed in the myeloid compartment, such as depletion of non-classical CD16⁺ monocytes^{17 , 21 , 22} and expansion of CD14⁺HLA-DR^low cells resembling immunosuppressive monocytes observed in sepsis.¹⁷ In addition, platelets exhibit a hyperactivated phenotype defined by expression of the activation marker CD62P (P-selectin) and readily form pro-thrombotic platelet-leukocyte aggregates.^{23 , 24 , 25}

Alongside lymphopenia, meta-analyses have identified significant association between elevated C-reactive protein (CRP), elevated D-dimer, and thrombocytopenia with poor outcome.²⁶ However, these are dysregulated in a variety of communicable and non-communicable diseases, including other viral respiratory tract infections.^{27 , 28} Further studies have identified various signatures associated with severe COVID-19,^{5 , 6 , 10 , 17 , 22 , 29 , 30 , 31 , 32 , 33 , 34} including falling levels of the plasma protein LRRC15³³ and enhanced Ki-67 and reduced COX2 expression in peripheral CD14⁺ monocytes.³⁴ Although informative, these complex signatures are not always translatable into parameters that can be measured simply and critically, and none have so far specifically targeted the cell surface in an unbiased way.

A comprehensive understanding of the immune dysregulation and immunopathology that underpins COVID-19 is therefore essential to identify patients at risk of progressing to severe disease in order to provide early therapeutic intervention. Here, we performed a detailed, unbiased proteomic analysis of the peripheral blood mononuclear cell (PBMC) plasma-membrane (PM) and cellular proteomes from seven healthy controls and 33 unvaccinated individuals with acute SARS-CoV-2 infection across the spectrum of COVID-19 disease. These data complement previous characterization of the same cohort by whole-blood transcriptomics and cytometric phenotyping⁵ alongside PBMC single-cell sequencing.²⁹ We identified progressive upregulation of a group of proteins expressed by multiple immune-cell subpopulations, including CD4⁺ T cells and non-classical monocytes, as disease severity increased. These data further our understanding of immune dysfunction in COVID-19 and define an unusual set of cellular phenotypes that could be used to identify patients at risk of progression to severe disease.

Results

Patient cohort and workflow

SARS-CoV-2 PCR-positive subjects were recruited to this study as part of a larger cohort by Bergamaschi et al.⁵ between 31^st March and 20^th July 2020. Collection dates indicated that the study subjects were infected with the original wild-type SARS-CoV-2 virus. Subjects with no or mild symptoms were recruited from routine screening of healthcare workers (HCWs). Patients with COVID-19 were recruited at or soon after admission to Cambridge University Hospitals or the Royal Papworth Hospital. Participants were divided into five categories of clinical severity (Figures 1A and 1B; Table S1): (A) asymptomatic HCWs, (B) HCWs who were either still working with mild symptoms insufficient to meet criteria for self-isolation or were symptomatic and self-isolating,^{35 , 36} (C) patients who presented to hospital but never required oxygen supplementation, (D) patients who were admitted to hospital and whose maximal respiratory support was supplemental oxygen, and (E) patients who at some point required assisted ventilation. One patient who died without admission to intensive care was also included in this severe group. Additionally, seven healthy controls (HCs), distributed across a range of age and gender, were included.

Figure 1.

Cohort outline and experimental workflow

(A) Details of study participants.

(B) Distribution of participant age across disease severity categories. Data are presented as boxplots with error bars indicating the minimum and maximum values.

(C) Schematic of experimental workflow. Patient PBMC samples were thawed and processed in parallel for TMT-based multiplexed analysis of whole-cell and PM proteomes.

PBMCs were characterized using two parallel workflows for proteomic analysis to profile changes in the whole-cell lysate (WCL) and PM proteomes. Peptide samples were labeled with Tandem Mass Tag (TMT) reagents and analyzed in a series of multiplexed tandem mass spectrometry (MS) experiments (Figure 1C).³⁷ Aliquots of the same samples were previously analyzed using flow cytometry panels and RNA-seq.⁵

Identification of cellular phenotypes associated with disease severity by MS

A total of 7,704 proteins and 597 proteins were quantified in the WCL and PM datasets, respectively. Statistical analyses indicated that there were no significant differences in protein abundance between donors with asymptomatic or mild symptomatic disease (classes A and B), and these data were therefore combined for further analysis. In WCLs, differential protein expression was most pronounced in severe COVID-19 (classes D and E). Although 45 and 82 proteins were upregulated >2-fold relative to HCs in class AB and C donors, respectively, 278 and 392 proteins were upregulated >2-fold relative to HCs in classes D and E, respectively. These included several proteins previously identified as upregulated in severe COVID-19: CRP,¹ complement component 9 (C9),³² and lipopolysaccharide-binding protein (LBP)^{31 , 38} (Figure 2 A). Gene Ontology (GO) analysis identified significant enrichment of terms associated with antiviral defense in class AB donors, including upregulation of multiple interferon-stimulated genes such as Interferon Induced proteins with Tetratricopeptide repeats (IFITs), Myxovirus resistance (Mx) proteins, and Oligoadenylate synthetase (OAS) proteins. By comparison, proteins highly upregulated in classes D and E were enriched in functions related to antimicrobial and antibacterial defense (Figure 2B; Table S5). Data from all proteomic experiments in this study are shown in Table S6 alongside prior complementary transcriptomic analysis. Here, the worksheet “Lookup” is interactive, enabling generation of graphs of expression of any of the proteins quantified.

Figure 2.

Severe COVID is associated with substantial changes in both the whole-cell and PM PBMC proteomes

(A) Examples of data for three proteins quantified in our analysis that have been reported to be upregulated in PBMCs from individuals with severe COVID-19. Ordinary one-way ANOVA with Tukey’s multiple comparisons post hoc test on log₂-transformed data: ^∗p < 0.05, ^∗∗p < 0.005, ^∗∗∗p < 0.0005, ^∗∗∗∗p < 0.0001. Data are presented as boxplots with error bars indicating the minimum and maximum values.

(B) Hierarchical cluster analysis of fold change in expression of 5,226 whole-cell proteins quantified across all three WCL analyses for each severity class compared with healthy control. Fold changes were derived from data averaged across all donors from a given disease severity class compared with average data from HCs analyzed in the same MS experiment. Enlargements of two subclusters are shown, highlighting groups of proteins that were upregulated in classes D and E vs. AB (top panel) or in classes AB vs. E (bottom panel). DAVID enrichment terms and corresponding Benjamini-Hochberg-corrected p values are shown for each cluster.

(C) Hierarchical cluster analysis of 522 PM-enriched proteins quantified in both PM analyses. Enlargement of one subcluster is shown, identifying a group of proteins that are highly upregulated at the PBMC cell surface in severe COVID-19. Proteins that best discriminate between mild and severe disease on the basis of PCA (Figure S1) are highlighted in red.

To identify cell-surface phenotypes associated with COVID-19 severity, which could also be readily utilized in diagnostic or therapeutic settings, equivalent analyses were conducted on PM data (Figure 2C). Overall, 54 and 49 proteins were upregulated >2-fold in classes D and E versus healthy control, including a cluster of proteins that were specifically upregulated in severe, but not mild, disease (Figure 2C, right panel). Interestingly, this group of proteins included several members of the carcinoembryonic antigen-related cell adhesion molecule (CEACAM) family, CEACAMs 1, 6, and 8, in addition to immunological markers CD177, CD63, and CD89. Principal-component analysis (PCA) of WCL and PM data confirmed these markers as key proteins that were differentially expressed between severity classes, and that samples from classes D and E clustered separately from HCs and classes A, B, and C (Figures S1A–S1E). Analysis of PCA loadings revealed that a substantial proportion of the variation between classes resulted from changes in CEACAM proteins, CD177, CD63, and CD89 (Figures S1F–S1G).

Further analysis of protein profiles for each of these candidate markers identified a consistent and significant upregulation in marker abundance with increasing disease severity in both WCL and PM data (Figure 3 ). Supporting the proteomic conclusions, this trend was also observed for corresponding genes in whole-blood RNA-seq data generated in parallel by Bergamaschi et al.⁵ for 107 donors (Figures 3 and S2; Table S2) and in an independent cohort of 24 donors by Wargodsky et al.³⁹ (Figure S3A) conducted during a period in which circulation of the B.1.1.7 (Alpha) variant was predominant. It is possible that our data might be confounded by later sampling of class E donors after onset of symptoms compared with donors with mild disease (Figure S3B). To address this possibility, for each gene of interest, RNA-seq data from Bergamaschi et al.⁵ were separated into 12-day sampling bins. Expression of each gene increased significantly with COVID-19 severity, regardless of when sampling was conducted (Figure 4 ). It is also possible that data might be confounded by other factors, such as age, gender, obesity, or the presence of co-morbidities such as diabetes. CEACAM8, CD177, CD63, and CD89 were quantified in all donors, whereas CEACAM1 and CEACAM6 were quantified in a subset. For the former group of four proteins, three-way ANOVA analysis of whole-cell proteomic data determined that, although CEACAM8, CD177, CD63, and CD89 expression varied significantly according to donor class, only CD89 expression varied significantly with age, and no markers were significantly associated with gender. Similarly, for donors in classes C–E (for whom information about obesity and diabetes was available), only CD177 and CD89 exhibited weakly significant variation with obesity (p = 0.043 in each case) and no markers varied significantly in the presence or absence of diabetes (Table S6). As a result, these markers were selected for further validation and characterization.

Figure 3.

Identification of host proteins associated with severe COVID using complementary datasets

Protein and RNA expression data for candidate markers identified in our analysis as being upregulated in PBMCs from individuals with severe COVID-19. RNA expression data at the earliest available time point were extracted from previous whole-blood RNA-seq analysis of a cohort of 107 individuals, including our donors,⁵ and expressed in log₂ (Reads Per Kilobase of transcript per Million mapped reads [RPKM]). Significance of variation for CEACAM8, CD177, CD63, and CD89 in the complete WCL dataset was determined by one-way ANOVA with correction for multiple hypothesis testing using the method of Benjamini-Hochberg.⁴⁰ Significance of change between severity classes for each individual protein was then determined by ordinary one-way ANOVA with Tukey’s multiple comparisons post hoc test on log₂-transformed data: ^∗p < 0.05, ^∗∗p < 0.005, ^∗∗∗p < 0.0005, ^∗∗∗∗p < 0.0001. Data are presented as boxplots with error bars indicating the minimum and maximum values.

Figure 4.

Marker upregulation during severe disease is independent of sample collection timing

Marker expression from time-normalized RNA-seq data. RNA expression data at the earliest available time point was extracted from previous whole-blood RNA-seq analysis by Bergamaschi et al.,⁵ separated into 12-day sample timing bins as described in their manuscript,⁵ and expressed in log₂(RPKM). Significance of change between severity classes for each individual protein was determined by ordinary one-way ANOVA with Tukey’s multiple comparisons post hoc test on log₂-transformed data: ^∗p < 0.05, ^∗∗p < 0.005, ^∗∗∗p < 0.0005, ^∗∗∗∗p < 0.0001. Data are presented as boxplots with error bars indicating the minimum and maximum values.

Of note, CEACAM8 and CD177 are classically regarded as neutrophil markers. As neutrophilia is a hallmark of severe COVID-19, we first determined whether observations resulted from neutrophil contamination of PBMC samples, despite exclusion of granulocytes during PBMC density gradient isolation. PBMC immunophenotyping flow cytometry data were reanalyzed for all donors in the larger cohort from which proteomic samples were drawn (Table S3).⁵ Mature neutrophil contamination was only present in a single sample, indicating an absence of confounding systemic contamination (Figures S4A and S4B). Notably, the one donor with identified mature neutrophil sample contamination (CV0144, class E) did not exhibit greater upregulation of CEACAM8 or CD177 versus other class E donors (Table S6A), suggesting that upregulation of these markers is neutrophil independent. Furthermore, upregulation of the majority of candidate markers appeared independent of secondary infection in patients with severe COVID-19 (Figure S4C).To explore further the predictive potential of these markers, we used least absolute shrinkage and selection operator (LASSO) penalized logistic regression with 10-fold cross-validation to identify the most regularized model that predicts the likelihood of an individual having severe COVID-19 (Figures S5A and S5B). The optimal model, comprising CEACAM8, CD177, and CD89, was able to identify individuals with severe COVID-19 with an accuracy of 87% (95% confidence interval [CI], 69%–96%).

Validation and phenotyping of identified markers by flow cytometry

To verify candidate markers and to phenotype cell populations expressing these proteins, multicolor flow cytometry was performed on an independent cohort of 36 donors recruited from Cambridge University Hospitals covering the spectrum of disease severity (Figures 5 A and S6A; Table S4). Corresponding to previous observations, this revealed a progressive and significant decrease in lymphocytes with worsening COVID-19, concurrent with a significant increase in platelet abundance and no overall trend in the myeloid compartment (Figure S6B). Detailed phenotyping identified further changes associated with disease severity in specific subpopulations, including depletion of CD56⁺ NK cells, CD3⁺CD4⁺ and CD8⁺ T cells, and increases in both resting CD62P⁻ and activated CD62P⁺ platelets (Figure 5B), again consistent with previously documented changes in circulating immune cells during acute COVID-19.^{6 , 7 , 8 , 17} A trend toward depletion of CD3⁺γδTCR⁺ T cells and increased abundance of CD16⁺ non-classical monocytes was also observed, although, due to sample limitations, it was not possible to collect sufficient events required for statistical significance.

Figure 5.

Validation and characterization of candidate markers of disease severity by flow cytometry

(A) Flow cytometry data for candidate markers, showing the proportion of target-positive cells within a live PBMC subpopulation (for gating strategy, see Figure S6A). Ordinary one-way ANOVA with Tukey’s multiple comparisons post hoc test: ^∗p < 0.05, ^∗∗p < 0.005, ^∗∗∗p < 0.0005, ^∗∗∗∗p < 0.0001.

(B) Heatmap of PBMC composition across disease severity classes, expressed as ratio comparing the proportion of each cell population in live PBMC relative with HCs (for gating strategy, see Figure S7B). p values as in Figure 5A.

(C) Heatmap of biomarker expression across immune-cell subsets, expressed as log₂ (fold change in geometric mean fluorescence intensity relative to HCs). p values as in Figure 5A.

Interestingly, candidate markers exhibited expression changes associated with disease severity in bulk lymphocyte, myeloid, and platelet populations. The frequency and intensity of CEACAM1, 6, and 8 expression was significantly upregulated on lymphocyte and myeloid cell populations in class E donors compared with both HCs and patients with mild disease (Figures 5A and S7A) and on platelets, significantly upregulated in severe disease versus HCs and patients with mild or moderate disease. Notably, frequency and intensity of CEACAM8 expression on myeloid cells was also significantly upregulated on patients with moderate versus mild disease, indicating progressive upregulation of this marker with increasing disease severity. The frequency and intensity of CD177 expression was upregulated in severe disease on both lymphocytes and platelets, and CD89 on lymphocytes alone, and the intensity of CD63 and CD89 expression was significantly increased on platelets (Figures 5A and S7A).

Further phenotyping was performed to define cell populations upregulating candidate markers (Figure S7B). Among lymphocytes, CD3⁺CD4⁺ T cells significantly upregulated CEACAM1/6, CD177, CD63, and CD89 (Figure 5C), in addition to strongly upregulating CEACAM8, although substantial variation between donors was observed. CD3⁺CD8⁺ T cells and γδT cells significantly upregulated CD177 and CEACAMs 1, 6, and 8, while CD56⁺ NK cells upregulated CD89 alone. Within the myeloid compartment, significant upregulation of CEACAM8 was observed on CD14⁺ classical monocytes and upregulation of CEACAMs 1, 6, and 8 on non-classical CD16⁺ monocytes, although low cell counts confounded interpretation of data for this population. Resting CD61⁺CD62P⁻ platelets significantly upregulated all identified markers in patients with severe disease, and activated CD61⁺CD62P⁺ platelets upregulated CEACAMs 1, 6, and 8 in addition to CD177. Combined t-distributed stochastic neighbour embedding (tSNE) analysis of live PBMC sampled from all class E donors (Figure 6 A) indicated that each identified biomarker was co-upregulated on distinct, singular CD4⁺ T cell (Figure 6B), CD62P⁻ platelet (Figure 6C), and CD16⁺ monocyte populations (Figure 6D).

Figure 6.

Co-upregulation of markers on CD4⁺ T cells, CD62P⁻ platelets, and CD16⁺ cells

(A) tSNE visualization of 6 × 10⁵ live PBMCs sampled from each class E donor, colored by cell population.

(B) Overlay of marker-positive CD4⁺ T cells onto tSNE analysis in (A).

(C) Overlay of marker-positive CD62P⁻ platelets onto tSNE analysis in (A).

(D) Overlay of marker-positive CD16⁺ cells onto tSNE analysis in (A).

Discussion

Understanding the complex immunobiology of COVID-19 is essential in developing predictive measures both of the severity of acute disease and to predict the development and progress of long COVID. This knowledge will also be vital to understanding efficacy of novel therapies. Here we present a searchable analysis of the PBMC cellular and PM proteomes during acute SARS-CoV-2 infection in a cohort of donors spanning the spectrum of COVID-19 disease. Our data indicate a profound shift in PBMC proteome profiles from mild to severe disease, echoing observations made in the whole-blood transcriptome⁵ and plasma proteome³² and reflecting the significant remodeling of circulating immune-cell composition during COVID-19. Notably, we observed highly significant enrichment of terms related to microbial defense among cellular proteins upregulated during severe disease. Patient metadata indicated that only a small number of patients in classes D and E had confirmed secondary infections, further corroborating the emergence of a sepsis-like state driving COVID-19 immunopathology.¹⁷ In addition, selective upregulation of canonical interferon-stimulated genes such as the IFIT and Mx families was observed in patients with mild disease. Of note, time-normalized transcriptomic analysis of the wider cohort from which our samples derived⁵ suggested that PBMCs from patients with severe disease that are collected relatively early after symptom onset can also exhibit upregulated interferon stimulated gene (ISG) expression that subsequently wanes. We observed a similar trend for donors sampled in the earlier phases of infection (Tables S1, S6).

Our results confirmed the increase across disease classes of existing markers of COVID-19 severity such as CRP and complement C9. Unbiased profiling of the PBMC PM proteome identified a signature of severe COVID-19, marked by the upregulated expression of a group of proteins with immunoregulatory functions: CEACAM1, 6, and 8, and CD177, CD63, and CD89. A predictive model trained on a subset of these features highlighted the high sensitivity and specificity with which severe and mild disease can simply be distinguished on the basis of these markers. Notably, this upregulation at the level of transcript was corroborated by two further whole-blood RNA-seq studies, and time-normalized transcriptomic data indicated that the trend of increasing expression was independent of sample collection timing. CD177 is a glycosyl-phosphatidylinositol (GPI)-linked surface glycoprotein that is canonically regarded as a neutrophil marker,⁴⁰ although expression on monocytes has also been reported.⁴¹ CD177 plays characterized roles in mediating neutrophil endothelial transmigration via binding to PECAM1⁴² in addition to activation and degranulation as part of a CD177/PR3-CD11b/CD18 complex.^{43 , 44} CD63 is a ubiquitously expressed member of the tetraspanin membrane protein family involved in cell adhesion and intracellular trafficking.⁴⁵ Typically localized to intracellular compartments, cell-surface CD63 is used as a marker of platelet⁴⁶ and T cell⁴⁷ activation in addition to granulocyte degranulation.^{48 , 49} CD89 is an Fc receptor expressed on neutrophils and monocytes⁴¹ that binds to immunoglobulin A (IgA) immune complexes⁵⁰ and CRP,⁵¹ initiating cell activation and cytokine release,⁵² and it has a role in protecting against bacterial sepsis.⁵³

CEACAMs 1, 6, and 8 belong to a family of immunoglobulin-like surface glycoproteins that can form homophilic and heterophilic interactions in conjunction with an array of binding partners that participate in a diverse range of processes, including cell adhesion, signaling, and immunoregulation.⁵⁴ Members of the CEACAM family are expressed on a broad range of cell types: CEACAM1 on epithelial cells, endothelial cells,⁵⁴ and activated T cells⁵⁵; CEACAM6 on epithelial cells, neutrophils, and monocytes⁴¹; CEACAM8 on granulocytes and monocytes.⁴¹ Notably, members of the family have established roles in modulating immune-cell function. CEACAM1 has an established inhibitory role in T cells,⁵⁶ suppressing signaling through the TCR-CD3 complex via phosphatase recruitment⁵⁷ in response to stimuli such as homophilic interaction⁵⁸ with CEACAM1 on antigen-presenting cells or heterophilic interaction with CEACAM6.⁵⁹ In addition, CEACAM1 forms a complex with and mediates the inhibitory function of the exhaustion marker TIM3⁶⁰ on T cells and negatively regulates NK cell activity.⁶¹ CEACAM8 is upregulated on granulocytes following activation,⁶² mediates adhesion to endothelial cells⁶³ and cytokine release,⁶⁴ and is highly upregulated on neutrophils in bacterial sepsis.⁶⁵

Interestingly, CD177 levels have previously been associated with COVID-19 severity and intensive care unit (ICU) admission in an analysis of a French cohort by whole-blood transcriptomics and serum profiling.⁶⁶ The authors proposed that increased circulating CD177 reflects the dysregulated neutrophil activation observed in severe COVID-19. We also observed an equivalent progressive and significant upregulation of CD177 in PBMC preparations. Isolation of PBMCs by density gradient centrifugation typically excludes granulocytes, and an absence of mature neutrophil contamination was verified by flow cytometry. Phenotyping indicated the progressive emergence of a CD3⁺CD4⁺CD177⁺ T cell population as disease severity increased in the context of a broader depletion of both CD4⁺ and CD8⁺ T cells. This population also co-upregulated CEACAMs 1 and 6, CD63, CD89, and likely CEACAM8, although the latter marker exhibited substantial inter-donor variation.

Simultaneous expression of CEACAMs 6 and 8, CD177, and CD89 on CD4⁺ T cells is intriguing, as expression of these markers is regarded as restricted to granulocytes and monocytes. CD177 expression may plausibly facilitate migration into critical tissues and cell activation in the context of infection. Upregulation of CD89 is particularly interesting due to its role in bacterial sepsis⁵³ and the relationship between severe COVID-19 and a sepsis-like syndrome. Recognition of IgA,⁵⁰ the most prevalent antibody isotype in the respiratory tract, and CRP,⁵¹ a widely used biomarker of severe COVID-19, by CD89 may have functional implications in cytokine release from this cell population. Concurrent upregulation of the activation factors CEACAM8 and CD89 and the inhibitory factors CEACAM1 and 6 is unusual and resembles observations of SARS-CoV-2-specific CD8⁺ T cells in patients with severe disease expressing both exhaustion markers and transcriptional signatures of inflammatory cytokine production and cytotoxicity.^{67 , 68} Our data suggest that a non-canonical but similar phenotype may therefore also develop in the CD4⁺ T cell population in the context of severe COVID-19.

It is presently unclear whether this CD4⁺ population contains SARS-CoV-2-specific cells, represents activated bystander T cells, or cells with other specificities. SARS-CoV-2-specific CD4⁺ T cells have been observed from 2 to 4 days after symptom onset,⁶⁹ consistent with the timescale in our cohort. Antigen-independent activation of CD8⁺ T cell populations by inflammatory cytokines has been reported in several viral infections, including hepatitis A,⁷⁰ influenza A,⁷¹ and SARS-CoV-2⁵ and are proposed to contribute to immunopathology in these settings. Bystander activation of CD4⁺ T cells is also proposed to contribute to pathogenesis during herpes simplex virus^{72 , 73} and dengue virus infection.⁷⁴ Interestingly, others have previously described the enrichment of activated cytotoxic CD16⁺ T cells in patients with severe COVID-19 that do not exhibit specificity for SARS-CoV-2 and are associated with poor outcome.⁷⁵ The CD4⁺ T cell population described here may plausibly constitute a further example of immune dysregulation in the T cell compartment during severe COVID-19 and contribute to immunopathology.

Previous studies have identified a hyperactive platelet phenotype in critically ill patients with COVID-19 that is proposed to contribute to hypercoagulability and tissue damage observed during acute respiratory distress syndrome. This phenotype is characterized by upregulated expression of the activation markers CD62P, LAMP3, and CD63,^{24 , 76} increased formation of platelet-leukocyte aggregates,^{24 , 25} and indications of degranulation.⁷⁷ Supporting these observations, we observed a clear shift in platelet phenotype in patients with severe COVID-19, with a profound increase in both activated CD61⁺CD62P⁺ and non-activated CD61⁺CD62P⁻ platelets and upregulated expression of CD63. Interestingly, we also identified a CD62P⁻ platelet population that co-upregulated CD177, CD89, and CEACAMs 1, 6, and 8 to high levels. Expression of CD89, for example, may facilitate platelet degranulation. Upregulation of CD177 and CEACAM family members is particularly interesting in this context, given their roles in cell adhesion through homophilic or heterophilic interactions. Expression of these factors on platelets, monocytes, and neutrophils may represent a mechanism that facilitates cell-cell contacts and formation of thrombotic aggregates.

As in other studies,^{17 , 21} our data indicated perturbations in the myeloid compartment. Phenotyping indicated that CEACAMs 1, 6, and 8 were prominently co-upregulated on a small population of CD16⁺CD14⁻ cells and CEACAM8 on CD14⁺ CD16⁻ cells. One possible explanation for the former observation is that the identified proteins are upregulated on non-classical CD16⁺ monocytes. Several studies have described depletion of non-classical monocytes in severe COVID-19,^{17 , 21 , 22} although this was not observed in single-cell sequencing of our cohort²⁹ and does not preclude expansion of a subpopulation within a broader population contraction. Alternatively, our data may represent increased abundance of immature low-density pro- or pre-neutrophils (LDNs) during severe disease. LDNs migrate with PBMCs during density gradient isolation, unlike their mature counterparts, and are typically released via emergency myelopoiesis⁷⁸ during infection,^{79 , 80} sepsis,^{81 , 82} and in autoimmune conditions.⁸³ LDNs are associated with a dysfunctional immunosuppressive environment.^{84 , 85} Emergence of immature neutrophils has previously been reported in severe COVID-19,^{15 , 17 , 21} including a CEACAM8⁺ subpopulation,¹⁷ although this population exhibited low to intermediate CD16 expression.¹⁵ Prior characterization of PBMCs from this cohort by single-cell RNA-seq documented the appearance of a rare C1QA/B/C⁺CD16⁺ monocyte population in patients with severe disease that is predicted to interact with and contribute to platelet activation²⁹ but did not show the emergence of any mature or immature neutrophil population. Our data may therefore suggest the appearance of a CD16⁺CEACAM1/6/8⁺ monocyte subset in the context of severe COVID-19.²⁶ Taken together, our observations of expression of CEACAM family members and CD177 on platelets and CEACAM expression on both classical and non-classical monocytes may represent another mechanism for formation of pathological platelet-monocyte aggregates.

In summary, we have identified immune-cell surface phenotypes associated with COVID-19 severity and have characterized unusual CD4⁺ T cell, resting platelet, and monocyte populations that co-express individual marker proteins in PBMCs from individuals with severe COVID-19. These cellular phenotypes could be readily utilized in identifying individuals with or developing severe disease in addition to providing future avenues for expanding understanding of mechanisms of immunopathology. Further longitudinal studies are required to determine the prognostic value of these markers and their potential involvement with persistent long COVID symptoms.

Limitations of the study

This study assessed a relatively small cohort of individuals at a single time point during acute SARS-CoV-2 infection. The phenotypes identified here may plausibly be shared with other severe respiratory infections and it will be important to address similarities in future work, since these might provide more global prognostic markers. Expansion of our analyses to a greater number of donors would improve the power of our study, and prospective studies will enable assessment of the prognostic and diagnostic value of the identified phenotypes. Finally, as novel SARS-CoV-2 variants have arisen following the conclusion of this study and vaccination coverage has increased, it would be of particular interest to examine the impact of these parameters on our observations.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-human CD61 BUV395 (RUU-PL7F12)	BD	RRID: AB_2871633
Anti-human CD16 BUV496 (3G8)	BD	RRID: AB_2870224
Anti-human CD4 BUV563 (RPA-T4)	BD	RRID: AB_2870854
Anti-human CD62P BUV737 (AK-4)	BD	RRID: AB_2873476
Anti-human CD8 BUV805 (RPA-T8)	BD	RRID: AB_2873737
Anti-human gdTCR BV421 (B1)	BioLegend	RRID: AB_2562317
Anti-human CD66a/c/e BV605 (ASL-32)	BioLegend	RRID: AB_2800910
Anti-human CD3 BV785 (OKT3)	BioLegend	RRID: AB_2563507
Anti-human CD177 FITC (MEM-166)	BioLegend	RRID: AB_2072603
Anti-human CD66 b PE (G10F5)	BioLegend	RRID: AB_2077857
Anti-human CD56 PE-Dazzle (HCD56)	BioLegend	RRID: AB_2563564
Anti-human CD63 APC (H5C6)	BioLegend	RRID: AB_10916521
Anti-human CD89 Alexa Fluor 700 (A59)	BioLegend	RRID: AB_2750067
Anti-human CD14 APC-Fire (M5E2)	BioLegend	RRID: AB_2632660
Chemicals, peptides, and recombinant proteins
TMTpro 16plex Label Reagent Set	Thermo Fisher Scientific	Cat# A44522
HPLC water	VWR	Cat# 23595.328
LC-MS grade acetonitrile	Merck	Cat# 1.00029.2500
Formic acid	Thermo Fisher Scientific	Cat# 85178
Guanidine hydrochloride (8M)	Thermo Fisher Scientific	Cat# 24115
DL-Dithiothreitol	Sigma-Aldrich	Cat# 43815-1G
Iodoacetamide	Sigma	Cat# I1149-5G
Lysyl Endopeptidase	Wako	Cat# 125-02543
Trypsin Protease	Pierce	Cat# 90058
Hydroxylamine	Sigma	Cat# 438227
Acetonitrile, Extra Dry	Acros Organics	Cat# AC364311000
HEPES (1M, pH 7.0–7.6)	Sigma	Cat# H0887
LIVE/DEAD Fixable Aqua Dead Cell Stain Kit	Thermo Fisher Scientific	Cat# L34966
Human TruStain FcX	Biolegend	Cat# 422302
BD Horizon Brilliant Stain Buffer	BD	Cat# 563794
FluoroFix Buffer	BioLegend	Cat# 422101
Critical commercial assays
Micro BCA Protein Assay Kit	Thermo Fisher Scientific	Cat# 23235
AbC Total Antibody Compensation Bead Kit	Thermo Fisher Scientific	Cat# A10497
ArC Amine Reactive Compensation Bead Kit	Thermo Fisher Scientific	Cat# A10346
Deposited data
Mass spectrometry data	This paper	PRIDE repository: PXD040703
Flow cytometry data (1)	This paper	Mendeley Data: 10.17632/yj5rm3k3py.1
Flow cytometry data (2)	This paper	Mendeley Data: 10.17632/jv9jvm3r98.1
Flow cytometry data (3)	This paper	Mendeley Data: 10.17632/v65cjwvsjz.1
Software and algorithms
“MassPike”, a Sequest-based software pipeline for quantitative proteomics	Professor Steven Gygi’s lab, Harvard Medical School, Boston, USA	N/A
Cluster 3.0	Stanford University, University of Tokyo	http://bonsai.hgc.jp/%7Emdehoon/software/cluster/software.htm
Java Treeview	SourceForge.net	http://jtreeview.sourceforge.net/
Perseus	Max Planck Institute of Biochemistry, Martinsried, Germany	https://maxquant.net/perseus/
DAVID software	Huang da et al., 2009	https://david.ncifcrf.gov/
Graphpad Prism 9	GraphPad Software	https://www.graphpad.com/scientific-software/prism/
R	R Core Team 2015	N/A
Rstudio Desktop 2022.02.2 + 485	Rstudio	https://www.rstudio.com/
FlowJo	BD	v10, https://www.flowjo.com/solutions/flowjo
Other
Orbitrap Fusion Lumos Mass Spectrometer	Thermo Fisher Scientific	Cat# IQLAAEGAAP FADBMBHQ
Dionex UltiMate 3000 UHPLC	Thermo Fisher	N/A
Cytek Aurora Flow Cytometer	Cytek Biosciences	N/A

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Prof. Michael Weekes (mpw1001@cam.ac.uk).

Materials availability

This study did not generate new unique reagents.

Experimental model and subject details

Human subjects

25 patients with confirmed or suspected SARS-CoV-2 infection were recruited between 31/3/2020 and 20/7/2020 from Cambridge University Hospitals (CUH) or the Royal Papworth Hospital (Cambridge, UK).^{5 , 88 , 89} Infection was confirmed in 24 participants via a nucleic acid amplification tests. One donor presented with relevant symptoms and required low flow oxygen support but tested negative by PCR. 15 individuals were recruited through the Cambridge University Hospitals healthcare worker (HCW) PCR screening program,^{35 , 36} including eight with no or mild symptoms and a positive test for SARS-CoV-2, and seven who tested negative for SARS-CoV-2 (Figure 1). These negative control samples were selected to represent the age and sex distribution of SARS-CoV-2-positive participants. Recruitment of inpatients and HCWs at Cambridge University Hospitals was undertaken by the NIHR Cambridge Clinical Research Facility outreach team and the NIHR BioResource research team. Ethical approval was obtained from the East of England – Cambridge Central Research Ethics Committee (‘‘NIHR BioResource’’ REC ref. 17/EE/0025, and ‘‘Genetic variation AND Altered Leukocyte Function in health and disease – GANDALF’’ REC ref. 08/H0308/176). One healthy control donor was obtained for flow cytometry analysis under ethical approval from the Health Research Authority (HRA) Cambridge Central Research Ethics Committee (97/092). All participants provided informed consent. Detailed information on the patients can be found in Table S1. Collection dates for this cohort indicate that participants were infected with original wild-type SARS-CoV-2.

Method details

Clinical data collection

Clinical data were collected from medical charts and entered into spreadsheets. Available laboratory test results were extracted from Epic electronic health records (Cambridge University Hospitals) and from MetaVision ICU (Royal Papworth hospital ITU). SARS-CoV-2-positive HCW participants were categorised into two groups according to whether they were asymptomatic (group A) or had COVID-19 symptoms at the time of PCR testing (group B). Symptoms considered to be possible manifestations of COVID-19 were new onset fever (>37.8°C), cough, loss of sense of small, hoarseness, nasal discharge or congestion, shortness of breath, wheeze, headache, muscle aches, nausea, vomiting and diarrhea.³⁶

Hospital patients were assigned to one of three groups, reflecting the maximum level of respiratory support received during their hospital stay. Group C patients did not receive any supplemental oxygen. Group D patients received supplemental oxygen using low flow nasal prongs, a simple face mask, a Venturi mask or a non re-breathe face mask. Patients who received any of non-invasive ventilation (NIV), mechanical ventilation or extracorporeal membrane oxygenation (ECMO) were assigned to group E. Patients in group D who died in hospital during the study were also assigned to group E. In patients who were already established on home NIV for chronic respiratory failure, NIV delivered as per the home prescription (e.g. nocturnal) was not considered for the purpose of classification. Moreover, oxygen requirements that were clearly not related to COVID-19 were also not considered for classification purposes.

Peripheral blood mononuclear cell preparation and flow immunophenotyping

Each participant provided 27 mL of peripheral venous blood collected into 9 mL sodium citrate tubes. Peripheral blood mononuclear cells (PBMCs) were isolated using Leucosep tubes (Greiner Bio-One) with Histopaque 1077 (Sigma) by centrifugation at 800 × g for 15 min at room temperature. PBMCs at the interface were collected, rinsed twice with autoMACS running buffer (Miltenyi Biotech) and cryopreserved in FBS with 10% DMSO. All samples were processed within 4 h of venepuncture.

For proteomic and flow cytometry analysis, frozen PBMC samples were thawed in a water bath at 37°C and immediately diluted in TexMACS media (Miltenyi Biotech), centrifuged, resuspended in TexMACS supplemented with 10U/ml DNAse (Benzonase, Merck-Millipore) and rested at 37°C for 1h. PBMCs were then centrifuged, resuspended in fresh media and counted. For proteomic analysis, PMBCs were washed twice in ice-cold PBS pH 7.4 (Sigma). 75% of cells were used for plasma membrane profiling, while the remaining 25% was used for whole cell lysate proteomic analysis. For flow cytometry, cells were directly processed as described below.

Plasma membrane profiling

Following washing, cellular surface sialic acid residues were oxidised then biotinylated with 1 mM sodium meta-periodate (Thermo), 100mM aminooxy-biotin (Biotium) and 10 mM aniline (Sigma) in ice-cold PBS pH 6.7, by rocking the cells with 3 mL of the mix at 4°C for 30 min. The reaction was quenched by adding glycerol (Sigma) to a final concentration of 1 mM. The cells were then washed twice with ice-cold PBS pH 7.4 containing CaCl₂ and MgCl₂, and then lysed with 1.6% Triton X-100 (Thermo), 150 mM NaCl (Sigma), 1× protease inhibitor (complete, without EDTA (Roche)), 5 mM iodoacetamide (Sigma) and 10 mM Tris-HCl pH 7.6 (Sigma) for 30 min at 4°C. Nuclei and debris were removed by centrifugation at 4°C, once at 4,000 × g for 5 min then twice at 13,000 × g for 5 min. Samples were then snap-frozen in liquid nitrogen and stored at – 80°C prior to immunoprecipitation and protein digestion.

The following washes were performed using Poly-Prep columns (Bio-Rad) attached to a vacuum manifold. Biotinylated proteins were enriched by incubating with high affinity streptavidin agarose beads (Pierce) at 4°C for 75 min on a rotor, then were washed extensively with lysis buffer (1% Triton X-100, 150 mM NaCl, 10 mM Tris-HCl pH 7.6) and PBS (with CaCl₂ and MgCl₂)/0.5% SDS (Invitrogen). Beads were next incubated with PBS/0.5% SDS/100 mM dithiothreitol (DTT) for 20 min at room temperature. Further washes were performed with UC buffer (6 M urea in 0.1 M Tris-HCl pH 7.6), before alkylation with UC buffer containing 50 mM iodoacetamide for 20 min at room temperature in the dark. Beads were washed again with UC buffer and HPLC-grade H₂O, transferred to a screw cap column (Pierce) and then proteins were digested on-bead with 35 μL of 8 ng/μL trypsin (Thermo) in 200 mM HEPES pH 8.5 (Sigma) for 3 h in a shaking 37°C incubator. The digested peptides were eluted and stored at −80°C before TMT labelling.

Whole cell lysate digestion

After washing cells were lysed in 50 μL of 6 M guanidine (Thermo)/50 mM HEPES pH 8.5, vortexed extensively and sonicated. Cell debris was removed by centrifuging twice at 13,000 × g for 10 min at 4°C.

DTT was added to a final concentration of 5 mM and incubated at room temperature for 20 min. Cysteine residues were alkylated with 15 mM iodoacetamide and incubated for 20 min at room temperature in the dark. Excess iodoacetamide was quenched with DTT for 15 min. Samples were diluted with 200 mM HEPES pH 8.5 to a final concentration of 1.5 M guanidine, followed by digestion at room temperature for 3 h with LysC protease (Wako) at a 1:100 protease-to-protein ratio. Samples were further diluted with 200 mM HEPES pH 8.5 to a concentration of 0.5 M guanidine. Trypsin was then added at a 1:100 protease-to-protein ratio followed by overnight incubation at 37°C. The reaction was quenched with 5% formic acid and centrifuged at 21,000 × g for 10 min to remove undigested protein. Peptides were subjected to C18 solid-phase extraction (SPE, Sep-Pak, Waters) and vacuum-centrifuged to near-dryness.

Peptide labelling with tandem mass tags

TMTpro reagents (0.8 mg, Thermo) were dissolved in 45 μL anhydrous acetonitrile. For PM samples, 35 μL of digested peptide was labelled with 10 μL of TMT reagent in a final volume of 50 μL with a final acetonitrile concentration of 30% (v/v). For WCL samples, desalted peptides were dissolved in 200 mM HEPES pH 8.5 and peptide concentration measured by microBCA (Thermo). 30 μg of peptide was labelled with 10 μL of TMT reagent at a final acetonitrile concentration of 30% (v/v).

PM and WCL samples were then processed in parallel. Following incubation with TMT reagents at room temperature for 1 h, the reactions were quenched with hydroxylamine to a final concentration of 0.5% (v/v). Equivalent amounts of each sample were combined in batches. An unfractionated 1h analysis of each batch was carried out initially to ensure equivalent peptide loading across each TMT channel, thus avoiding the need for excessive electronic normalisation. Samples were then subjected to high pH reversed-phase (HpRP) fractionation (below).

Offline HpRp fractionation

TMT-labelled tryptic peptides derived from WCL samples were fractionated using an Ultimate 3000 RSLC UHPLC system (Thermo) equipped with a 2.1 mm internal diameter (ID) x 25 cm long, 1.7 μm particle Kinetix Evo C18 column (Phenomenex). Mobile phase consisted of A: 3% acetonitrile (MeCN), B: MeCN, and C: 200 mM ammonium formate pH 10. Isocratic conditions were 90% A/10% C, and C was maintained at 10% throughout the gradient elution. Separations were conducted at 45°C. Samples were loaded at 200 μL/minute for 5 min. The flow rate was then increased to 400 μL/minute over 5 min, after which the elution gradient proceeded as follows: 0–19% B over 10 min, 19–34% B over 14.25 min, 34–50% B over 8.75 min, followed by a 10 min wash at 90% B. UV absorbance was monitored at 280 nm and 15 s fractions were collected into 96 well microplates using the integrated fraction collector. For WCL samples, fractions were recombined orthogonally in a checkerboard fashion, combining alternate wells from each column of the plate into a single fraction, and commencing combination of adjacent fractions in alternating rows. This yielded two sets of 12 combined fractions, A and B. For PM samples, all wells in sets of two adjacent columns were recombined to yield six combined fractions. Wells were excluded prior to the start or after the cessation of elution of peptide-rich fractions, as identified from the UV trace. Samples were dried in a vacuum centrifuge and resuspended in 10 μL MS solvent (4% MeCN/5% formic acid) prior to LC-MS3.

For WCL samples, a series of 12 or 24 LC-MS3 analyses were next performed (either all of set A or all of sets A and B). For PM samples, either six fractions and an unfractionated singleshot sample were analyzed (batch 1) or an unfractionated singleshot sample was analyzed (batch 2).

LC-MS3

Mass spectrometry data were acquired using an Orbitrap Lumos (Thermo Fisher Scientific, San Jose, CA). An Ultimate 3000 RSLC nano UHPLC equipped with a 300 μm ID x 5 mm Acclaim PepMap μ-Precolumn (Thermo) and a 75 μm ID x 50 cm 2.1 μm particle Acclaim PepMap RSLC analytical column was used. Loading solvent was 0.1% formic acid (FA), analytical solvent A: 0.1% FA and B: 80% MeCN +0.1% FA. All separations were carried out at 55°C. Samples were loaded at 5 μL/min for 5 min in loading solvent before beginning the analytical gradient. The following gradient was used: 3–7% B over 3 min, 7–37% B over 173 min, followed by a 4 min wash at 95% B and equilibration at 3% B for 15 min. Each analysis used a MultiNotch MS3-based TMT method (4, 5). The following settings were used: MS1: 380–1500 Th, 120,000 Resolution, 2×10⁵ automatic gain control (AGC) target, 50 ms maximum injection time. MS2: Quadrupole isolation at an isolation width of m/z 0.7, CID fragmentation (normalised collision energy (NCE) 35) with ion trap scanning in turbo mode from m/z 120, 1.5×10⁴ AGC target, 120 ms maximum injection time. MS3: In Synchronous Precursor Selection mode the top 6 MS2 ions were selected for HCD fragmentation (NCE 65) and scanned in the Orbitrap at 60,000 resolution with an AGC target of 1×10⁵ and a maximum accumulation time of 150 ms. Ions were not accumulated for all parallelisable time. The entire MS/MS/MS cycle had a target time of 3 s. Dynamic exclusion was set to ±10 ppm for 70 s. MS2 fragmentation was trigged on precursors 5×10³ counts and above.

Flow cytometry

1x10⁶ cells per donor were resuspended in 100μL Horizon Brilliant Stain Buffer (BD) and blocked with Human TruStain FcX (BioLegend) at room temperature for 5 min. Cells were then stained with a 15-colour panel (see key resources table) for 30 min at 4°C, before washing in an excess of PBS, centrifugation and resuspension in 100μL FluoroFix Buffer (BioLegend). Single color compensation controls were prepared for the panel using AbC Compensation beads (Thermo) and ArC Compensation beads (Thermo) for antibody stains and amine-reactive viability stains respectively, or healthy control PBMCs. Samples were then analyzed on a Cytek Aurora flow cytometer (Cytek Biosciences) and data analyzed in FlowJo (BD).

Quantification and statistical analysis

Data analysis

Mass spectra were processed using a Sequest-based software pipeline for quantitative proteomics, ‘‘MassPike’’, through a collaborative arrangement with Professor Steven Gygi’s laboratory at Harvard Medical School. MS spectra were converted to mzXML using an extractor built upon Thermo Fisher’s RAW File Reader library (version 4.0.26). In this extractor, the standard mzxml format has been augmented with additional custom fields that are specific to ion trap and Orbitrap mass spectrometry and essential for TMT quantitation. These additional fields include ion injection times for each scan, Fourier Transform-derived baseline and noise values calculated for every Orbitrap scan, isolation widths for each scan type, scan event numbers and elapsed scan times. This software is a component of the MassPike software platform and is licensed by Harvard Medical School.

A combined database was constructed from the human Uniprot database (11/11/2020), the SARS-CoV-2 Uniprot reference proteome and common contaminants such as porcine trypsin and endoproteinase LysC. The combined database was concatenated with a reverse database composed of all protein sequences in reversed order. Searches were performed using a 20 ppm precursor ion tolerance. Fragment ion tolerance was set to 1.0 Th. TMT tags on lysine residues and peptide N termini and carbamidomethylation of cysteine residues (57.02146 Da) were set as static modifications, while oxidation of methionine residues (15.99492 Da) was set as a variable modification.

To control the fraction of erroneous protein identifications, a target-decoy strategy was employed.⁹⁰ Peptide spectral matches (PSMs) were filtered to an initial peptide-level false discovery rate (FDR) of 1% with subsequent filtering to attain a final protein-level FDR of 1%. PSM filtering was performed using a linear discriminant analysis, as described previously.⁹⁰ This distinguishes correct from incorrect peptide IDs in a manner analogous to the widely used Percolator algorithm (https://noble.gs.washington.edu/proj/percolator/), though employing a distinct machine-learning algorithm. The following parameters were considered: Xcorr, DCn, missed cleavages, peptide length, charge state, and precursor mass accuracy. Protein assembly was guided by principles of parsimony to produce the smallest set of proteins necessary to account for all observed peptides (algorithm described in⁹⁰).

Proteins were quantified by summing TMT reporter ion counts across all matching peptide-spectral matches using’’MassPike’’, as described previously.⁹¹ Briefly, a 0.003 Th window around the theoretical m/z of each reporter ion was scanned for ions and the maximum intensity nearest to the theoretical m/z was used. The primary determinant of quantitation quality is the number of TMT reporter ions detected in each MS3 spectrum, which is directly proportional to the signal-to-noise (S:N) ratio observed for each ion. Conservatively, every individual peptide used for quantitation was required to contribute sufficient TMT reporter ions so that each on its own could be expected to provide a representative picture of relative protein abundance.⁹¹ A per-sample S:N ratio of >15 was required such that, for example, for a 16-plex experiment a combined S:N ratio of >240 across all TMT reporter ions would be needed for a peptide to pass filtering. An isolation specificity filter with a cut-off of 50% was additionally employed to minimise peptide co-isolation.⁹¹ Peptides meeting the stated criteria for reliable quantitation were then summed by parent protein, in effect weighting the contributions of individual peptides to the total protein signal based on their individual TMT reporter ion yields. Protein quantitation values were exported for further analysis in Excel.

For protein quantitation, reverse and contaminant proteins were removed, then each reporter ion channel was summed across all quantified proteins and normalised assuming equal protein loading across all channels. Missing values in the mass spectrometry data were imputed for a small number of donor-protein datapoints (11 datapoints in total for all WCL analyses, two datapoints in total for both PM analyses) by setting missing values to the minimum intensity observed for the protein within each multiplexed experiment. Data for all HLA isoforms were removed, due to variation in HLA allele expression between donors. Proteins originating from red blood cell (RBC) contaminants were removed if they met two criteria: (1) they were identified as one of the top 15% most abundant RBC proteins from Ravenhill et al.⁹² and (2) they had a coefficient of variation greater than 0.75 in any given disease severity class when comparing protein abundance fold-change versus healthy control across donors.

Hierarchical centroid clustering of proteomic data was carried out in Cluster 3.0 (Stanford University) using an uncentred Pearson correlation similarity metric.

Statistical analysis

For analysis of the complete WCL dataset (Figure 3), ordinary one-way ANOVA tests were carried out in Python on log₂-transformed fold-change values for each donor versus healthy controls, omitting missing values. Generated p values were then corrected for multiple comparisons using the Benjamini-Hochberg method in Excel. For displayed individual proteins or cell populations, ordinary one-way ANOVA tests with Tukey’s adjustment for multiple comparisons were carried out in GraphPad Prism 9 on log₂-transformed fold-change values for each donor versus healthy controls (Figures 2A, 3, 4, 5C, and S3A) or on untransformed cell population proportion values (Figures 5A and 5B, S7A). Adjusted p values <0.05 were considered significant. Table S6D: one- and three-way ANOVAs with Tukey’s multiple comparisons post-hoc test for WCL MS data or Kruskal-Wallis tests for PM MS data were carried out in Perseus.

Pathway analysis

The Database for Annotation, Visualization and Integrated Discovery (DAVID) was used to identify enrichment of pathways in upregulated gene clusters as specified in the text. In each case, the cluster was searched against the background of all proteins quantified in our proteomics data using default fsettings.

tSNE analysis

Live PBMC (gated as in S2A) from class E donors CV0191, CV0197, CV0203, CV0212, CV0272 and CV0278 were concatenated, downsampled and tSNE analysis carried out in FlowJo v10. The tSNE parameters utilised were as follows: learning configuration: opt-SNE, iterations: 1000, perplexity: 30, learning rate: 438,000, KNN algorithm: Exact (vantage point tree) and gradient algorithm: Barnes-Hut.

Predictive model building

LASSO penalised logistic regression was performed using the cv.glmnet function implemented in the glmnet package in R using 10-fold cross-validation to identify the most regularised model. The most regularised model was defined as that having an error rate within 1 SE of the minimum mean cross-validated error.⁹³

Published: May 29, 2023

Supplemental information

Table S5. Functional enrichment analysis of proteomic data, related to Figure 2

Enrichment of functional pathways in clusters of cellular proteins upregulated during COVID. Database for Annotation, Visualization and Integrated Discovery (DAVID) enrichment terms and corresponding Benjamini-Hochberg-corrected p values are shown for each cluster in Fig. 2B.

Table S6. Interactive spreadsheet of all proteomic and transcriptomic data in the manuscript, related to Figures 2, 3, 4

(A) Interactive searchable spreadsheet containing all data and statistics from WCL, PM and RNA-seq analyses. (B) Proteomic data from all WCL analyses. (C) Proteomic data from WCL analyses for proteins quantified across all three WCL experiments. (D) Results of statistical tests comparing relative abundance of each protein quantified in WCL analyses. (E) Proteomic data from second PM analysis. (F) Proteomic data from all PM analyses. (G) Results of statistical tests comparing relative abundance of each protein quantified in second PM analysis. (H) Transcriptomic data from all donors generated in Bergamaschi et al.⁵ at day 0 time point. Data expressed as log₂(RPKM). (I) Transcriptomic data from donors also analyzed in proteomic analyses, generated in Bergamaschi et al.⁵ at day 0 time point. Data expressed as log₂(RPKM).

Data and code availability

• •The mass spectrometry raw files and associated unmodified peptide and protein quantitation data have been deposited in the ProteomeXchange Consortium via the PRIDE partner repository with identifier PXD040703.^{86 , 87} Unprocessed flow cytometry data have been deposited as three separate files in the Mendeley Data repository and are available at: https://doi.org/10.17632/yj5rm3k3py.1, https://doi.org/10.17632/jv9jvm3r98.1 and https://doi.org/10.17632/v65cjwvsjz.1.
• •This paper does not report original code.
• •Any additional information required to reanalyse the data reported in this paper is available from the lead contact upon request.