Longitudinal immune profiling reveals distinct features of COVID-19 pathogenesis

Introduction

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection results in the clinical syndrome COVID-19 which, to date, has resulted in over 5 million confirmed cases and in excess of 330,000 attributable deaths world-wide. A large number of clinical trials have been established to evaluate anti-viral and immune modulatory strategies aimed at improving clinical outcome.

SARS-CoV-2 is a single stranded, positive sense RNA virus that enters cells via human angiotensin converting enzyme 2 (ACE2)¹. Overlapping immune mechanisms exist to detect every stage of viral replication. Pattern recognition receptors of the innate immune system recognize viral antigen and virus-induced damage, increasing bone marrow hematopoiesis and immune cell mobilisation. If inflammatory mediator release is not controlled in duration and amplitude then “emergency hematopoiesis” leads to bystander tissue damage and the cytokine storm syndrome manifesting as organ dysfunction. Initial studies suggest cytokine storm syndrome occurs in COVID-19². Indeed, neutrophilia and lymphopenia (resulting in an increased neutrophil:lymphocyte ratio), increased systemic interleukin-6 (IL-6) and C-reactive protein correlate with incidence of ICU admission and mortality³.

The Coronavirus Immune Response and Clinical Outcomes (CIRCO) study was conducted at four hospitals in Greater Manchester, UK. The aim was to identify demographic, clinical and immunological factors associated with different immune phenotypes in response to SARS-Cov-2 infection in hospitalised patients. Single time point analysis of immunity (e.g at admission) usually involves thousands of patients. However, we felt it was important to study the kinetics of the immune response at depth in a longitudinal well-defined patient cohort. Understanding the specific elements and kinetics of immune excess is critical to gain insight into immune phenotypes associated with disease progression, identify potential biomarkers that predict clinical outcomes and determine at which stage of the disease immune modulation may be effective².

Methods

Results

In total, 73 patients were recruited and 49 were stratified for disease severity. Six patients were excluded due to an alternative diagnosis (2 patients), indeterminate imaging findings with negative for SARS-CoV-2 NP test (2 patients) or a confounding acute illness was diagnosed (2 patients). Two patients could not be stratified for disease severity due to insufficient clinical observation data and a further 16 because recruitment occurred more than 7 days after admission. The median time from symptom onset to hospital admission was 7 days (IQR - 4-10). The overall median age was 61 (IQR 51-71) and 63% were male. The most frequent co-morbidities were diabetes, ischaemic heart disease, hypertension, asthma and chronic obstructive pulmonary disease (table 1). The majority (86%) of patients tested positive for SARS-CoV-2 via nasopharyngeal RT-PCR. In 14% of patients, symptoms and radiographic features were highly suggestive of COVID-19, but nasopharyngeal tested negative for the virus and thus a clinical diagnosis was made. Complications occurred in 8 out of 49 patients and included pulmonary embolism and acute kidney injury.

Patient disease severity was defined as mild (less than 28% FiO2), moderate (28-60% FiO2) or severe (above 60% FiO2, or admission to intensive care) (appendix 1). All severe patients were recruited from ward level care. Only 1 patient met the criteria for direct ICU admission. Out of the 10 severe patients, 7 were treated in intensive care, 4 were intubated and ventilated and 2 were treated by CPAP alone (data not available for 1 patient). Three severe patients had ceilings of care in place that specified ward level care due to frailty. Death occurred in 50% of severe cases of COVID-19.

On hospital admission little difference was observed in total white blood cell counts, neutrophils or lymphocytes (figure 1A). Although, as reported previously⁴, a higher neutrophil to lymphocyte ratio (NLR) on hospital admission was observed in those patients whose disease trajectory was ultimately severe, whereas there were no appreciable differences observed in monocytes (figure 1A, 1B and table 1). By only analysing differential white cell counts, important correlations may be missed. We, therefore, performed multidimensional flow cytometry covering over 80 immune markers across 5 panels on whole blood cells and separated PBMCs.

• Download figure
• Open in new tab

Figure 1. Whole blood immune profile of COVID-19 patients.

(A) Hospital assessed white blood cell count (WCC), lymphocyte count, monocyte count, neutrophil count and (B) Neutrophil to lymphocyte ratio (NLR) calculated using hospital assessed neutrophil and lymphocyte counts. Grey region represents normal range. Patients stratified into mild (n=17), moderate (n=20) and severe (n=10) disease groups. (C) Uniform Manifold Approximation and Projection (UMAP) of flow cytometry panel broadly visualising white cells in whole blood. Representative images for healthy individuals, mild, moderate and severe patients are shown. (D) Neutrophil, monocyte and lymphocyte frequencies for healthy individuals and recruitment samples for COVID-19 patients, according to disease severity. Mild (n=12), moderate (n=13) and severe (n=5). (E) Longitudinal time course of CD3⁺ T cells and CD16⁺ neutrophils segregated by disease severity. Individual patients are shown as different colours and shapes. Crossed squares for severe patients are time points in ITU. For comparisons of disease subtypes and healthy individuals: Kruskal-Wallis test with multiple comparisons was carried out. (*P<0.05, **P<0.01, ***P<0.001).

During a pandemic, it is impossible to recruit healthy individuals for blood sampling from the community. As the data emerged, it became clear that the patients in the mild disease category were indistinguishable in the majority of analyses from our frontline volunteers. Furthermore, comparing the results across the different disease severities provided additional validation of our control group that tracked most closely with the patients experiencing mild disease. Typically, all parameters returned to within control levels on recovery, irrespective of disease severity.

Using Uniform Manifold Approximation and Projection (UMAP) we were able to visualise global immune cell differences between COVID-19 patients and healthy individuals (figure 1C). Broadly, this included alterations in the characteristics of neutrophils and monocytes and decreases in B cells, T cells, basophils and NK cells in the initial recruitment blood sample, taken within 7 days of hospital admission (figure 1D and appendix 3). These differences were exaggerated in patients with severe disease. This global picture was confirmed by manual flow cytometric gating (figure 1D). In addition, the frequency of basophils in whole blood and plasmacytoid dendritic cells (pDCs) within PBMCs were significantly decreased in patients compared to healthy individuals (appendix 3A-B). Therefore, hospital differential count (on day of admission) and laboratory (on day of recruitment to study) analysis of overall immune cell populations concur. Longitudinal analysis revealed that in the majority of patients (70%) (irrespective of severity) T cell frequencies in whole blood increased prior to hospital discharge, while neutrophil frequencies reciprocally decreased (figure 1E). In two severe patients who did not recover, T cell frequencies were extremely low and neutrophil frequencies high (figure 1E, white and pink squares). This suggests that recovery is associated with a reduction of neutrophils and an increase in T cells.

Within the reduced T cell population there were no dramatic alterations in CD4+ or CD8+ T cell frequencies, other than a slight decrease in CD4+ T cells in severe COVID-19 patients (figure 2A and 2B). Both T cell subsets showed signs of activation in COVID-19 patients; however, this was more striking in CD8+ T cells, did not track with disease severity, and was highly variable amongst patients (appendix 4A-C). No differences were observed in the expression of the regulatory molecule PD-1 (appendix 4D). Interestingly, in 34/43 COVID-19 patients higher perforin expression was observed in CD8+ T cells compared to healthy individuals (appendix figure 4E), implying these cells have activated a cytotoxic programme. Perforin expression in CD8+ T cells did not significantly track with disease severity (figure 2C,D), although we did observe a correlation with CRP (figure 2E). Therefore, only minor differences were observed in the CD8+ T cell compartment.

• Download figure
• Open in new tab

Appendix Figure 1

• Download figure
• Open in new tab

Appendix Figure 2

• Download figure
• Open in new tab

Appendix Figure 3

• Download figure
• Open in new tab

Appendix Figure 4

• Download figure
• Open in new tab

Figure 2. Altered phenotype of T and B cells in COVID-19 patients.

(A,B) Cumulative data show frequency of (A) CD8⁺ and (B) CD4⁺ T cells in PBMC of healthy individuals (n=24) and COVID-19 patients with mild (n=17), moderate (n=18) and severe (n=9-10) disease. (C,D) Representative flow cytometry plots and cumulative data show frequencies of CD8⁺ T cells which are positive for Perforin in healthy individuals (n=16) and COVID-19 patients with mild (n=16), moderate (n=12) and severe (n=7) disease. (E) Graph showing correlation of Perforin⁺ CD8⁺ T cell frequency with C-reactive protein (CRP) in COVID-19 patients. (F) Cumulative data show ex vivo frequency of CD19⁺ B cells in healthy individuals (n=31) and COVID-19 patients with mild (n=14), moderate (n=19) and severe (n=9) disease. (G) Representative flow cytometry plots and cumulative data show Ki67 expression by B cells in healthy individuals (n=16) and COVID-19 patients (n=45). Cumulative graph shows correlation of Ki67⁺B cells with C-reactive protein (CRP). (H) Representative flow cytometry plots and cumulative data show frequency of CD27^hiCD38^hi plasmablasts in healthy individuals (n=31) and COVID-19 patients (n=66). (I) Cumulative data show frequency of double negative (CD27^-IgD^-) B cells in healthy individuals (n=31) and COVID-19 patients with mild (n=14), moderate (n=19) and severe (n=9) disease. For healthy individuals versus COVID-19: Mann-Whitney test was carried out. For comparisons of disease subtypes and healthy individuals: Kruskal-Wallis test with multiple comparisons was carried out. For correlations: Spearman ranked coefficient correlation test was carried out (*P<0.05; **P<0.01; P<0.001; ns, not significant).

The frequency of B cells was also reduced in COVID-19 patients with severe disease compared to patients with mild and moderate disease (figure 2F). Although reduced, B cells displayed an increased expression of Ki67 (indicative of activation), which positively correlated with CRP levels (figure 2G). When examining B cell subsets no major differences in naive and switched memory B cell subsets were observed in COVID-19 patients (appendix 5A and 5B). However, we observed an expansion of antibody-secreting plasmablasts (CD27^hiCD38^hiCD24^-) in COVID-19 patients, which did not track with disease severity (figure 2H and appendix 5C). The only other population of B cells raised in patients with severe COVID-19 compared to patients with mild and moderate disease was a significant expansion of double negative (DN) B cells (CD27-IgD-) (figure 2I and appendix 5A). DN B cells have previously been shown to exhibit an exhausted phenotype in patients with HIV. This suggests that patients with severe COVID-19 may generate less effective B cell responses.

• Download figure
• Open in new tab

Appendix Figure 5

By far the most significant changes were observed in monocytes. In mild COVID-19 there was an expansion of the inflammatory CD14⁺CD16⁺ (intermediate) sub-population (appendix 6A), and enhanced CD64, (high affinity Fc receptor for monomeric IgG (FcγRI)) on CD14⁺ classical monocytes (figure 3A). Upon stimulation with LPS, monocytes from mild COVID-19 also displayed increased intracellular TNF, which significantly correlated with CD64 (figure 3B, C) whilst IL-1β levels were lower in monocytes from COVID-19 patients (appendix 6D). Severe COVID-19, on the other hand, was associated with monocytes displaying increased expression of the cell cycle marker, Ki67 (normally <5% in healthy peripheral blood), irrespective of whether monocytes were stimulated or not (figure 3C and appendix 6C), which strongly correlated with hospital data for CRP (figure 3C). We next examined cyclooxygenase-2 (COX-2) expression in monocytes, which is an enzyme involved in the production of prostaglandins. Remarkably, an early reduction of COX-2 was evident in patients that went on to develop severe COVID-19 (figure 3D).

• Download figure
• Open in new tab

Appendix Figure 6

• Download figure
• Open in new tab

Appendix Figure 7

• Download figure
• Open in new tab

Appendix Figure 8

• Download figure
• Open in new tab

Figure 3. Dysregulation of circulating monocytes in COVID-19.

(A) Levels of CD64 as assessed by mean fluorescence intensity (MFI) on CD14⁺classical monocytes in healthy individuals (n=25) and total COVID-19 patients (n=58). COVID-19 patients were stratified into mild (n=12), moderate (n=10) and severe (n=8) disease. (B) Production of TNFα by CD14⁺ monocytes following LPS stimulation and correlation of CD64 levels (MFI) with TNFa production, in healthy individuals (n=29) and total COVID-19 patients (n=59). COVID-19 patients were stratified into mild (n=14), moderate (n=15) and severe (n=7) disease. (C) Representative FACS plots demonstrating intracellular Ki67 staining by CD14⁺monocytes, summary graphs demonstrating expression of Ki67 following LPS stimulation and correlation of Ki67 (% of monocytes expressing Ki67) with C-reactive protein (CRP) in healthy individuals (n=25) and total COVID-19 patients (n=60). COVID-19 patients were stratified into mild (n=14), moderate (n=14) and severe (n=8) disease. (D) Representative FACS plots demonstrating intracellular COX2 expression by CD14⁺monocytes and summary graphs demonstrating proportion of CD14⁺ monocytes expressing COX2 following LPS stimulation, and levels of expression of COX2 as assessed by MFI, in healthy individuals (n=21) and total COVID-19 patients (n=51). COVID-19 patients were stratified into mild (n=12), moderate (n=11) and severe (n=6) disease. Individual data points shown with median for healthy versus COVID-19 and disease subtypes. For healthy versus COVID-19: Mann-Whitney test was carried out. For comparisons of disease subtypes and healthy individuals: Kruskal-Wallis test with multiple comparisons was carried out. For correlations: Spearman ranked coefficient correlation test was carried out (*P<0.05; **P<0.01; P<0.001).

We next examined sequential samples per patient and plotted the results based on the time since reported onset of symptoms. CD8+ T cells remained remarkably stable over the disease course (appendix 7A and 7B). However, we noticed that in mild and moderate COVID-19 perforin expressing CD8+ T cells increased with time, with the highest level immediately prior to discharge (appendix 7C,D). This suggests development of functional CD8+ T cells is beneficial in COVID-19. Reduced B cells was associated with a severe disease trajectory (notably in one of the deceased patients) (appendix 7E), particularly the switched memory subset (appendix 7F). Collectively, this suggests that low B cell responses associate with severe disease. This reduction may be due to exhaustion that is supported by an expansion of double negative B cells in severe patients (appendix 7G). The most striking trajectories that correlated with development of severe COVID-19 were again in the monocyte population, where low COX-2 was present throughout (figure 4A) and higher Ki67 was observed prior to ITU admission (figure 4B). TNF expression following monocyte stimulation remained higher in patients with mild disease throughout the disease course (figure 4C).

• Download figure
• Open in new tab

Figure 4. Longitudinal assessment of immune parameters during hospital admission.

(A) Graph shows frequencies of CD14⁺ monocytes expressing COX2 following LPS stimulation, and levels of expression of COX2 as assessed by MFI in admitted patients over all measured time points, in mild (grey shapes, n=7), and severe (red shapes, n=4) COVID-19 patients. (B) Graph shows frequencies of CD14⁺ monocytes expressing Ki67 following LPS stimulation in admitted patients over all measured time points, in mild (grey shapes, n=6), and severe (red shapes, n=5) COVID-19 patients. (C) Graph shows frequencies of CD14⁺ monocytes producing TNFa following LPS stimulation in admitted patients over all measured time points, in mild (grey shapes, n=7) and severe (red shapes, n=6) COVID-19 patients. On all graphs x axis values represent the number of days since onset of symptoms and the dotted line the median value from healthy individuals. (D) Levels of systemic IL-6, MCP-1, IP-10, TNFα, and IFNγ were measured in serum from healthy individuals (n=7) and COVID-19 patients with mild (n=12), moderate (n=7) and severe (n=7) disease using LEGENDplex assay. (E) Graphs show levels of systemic IL-6, MCP-1 and IP-10 in admitted patients over all measured time points, in mild (green shapes; n=12) and severe (red shapes; n=7) COVID-19 patients. Deceased patient represented by triangles. On all graphs x axis values represent the number of days since reported onset of symptoms and the dotted line the median value from healthy individuals. (F) Serum levels of IL-6, IP-10 and MCP-1 at last time-point Pre-ITU and first time-point after transfer to ITU. For comparisons of disease subtypes and healthy individuals: Kruskal-Wallis test with multiple comparisons was carried out. (*P<0.05; **P<0.01; P<0.001).

We next assessed soluble inflammatory mediators on admission to hospital and across the length of hospital stay. Of the many mediators analysed in serum the ones significantly different between patients and healthy controls were IL-6, IL-10, monocyte-chemoattractant protein-1 (MCP-1) and interferon gamma-induced protein 10 (IP-10) (figure 4D and appendix 8). The increased levels of these cytokines were specific to COVID-19 patients with severe disease, and strongly indicate a role for myeloid cells in disease pathogenesis of COVID-19. Longitudinal analysis revealed that the highest levels of inflammatory mediators occurred early in disease trajectory following hospital admission, and critically, declined rapidly following admission to ITU despite no indications of recovery (figure 4E and 4F).

Discussion

This detailed longitudinal analysis of COVID-19 patients of varying severity and outcome, the first reported here, has revealed common immunological features and aspects that track with severity. We provide information concerning pathogenesis that should influence clinical trials and therapeutics. An aspect not expected included the presence of proliferating (Ki67⁺) monocytes in the blood, with low COX-2 and high Fc receptor γ (FcRI, CD64). These properties all reverted in patients with good outcome. Also, not anticipated, was that multiple aspects of inflammation were diminished as patients progressed in severity and time. Thus, production of IL-6, IP-10, MCP-1 typically reduced late in disease, even in patients who did not recover.

The cellular analysis indicates considerable abnormalities in the myeloid compartment, much more so than the lymphopenia reported previously⁵. In the myeloid compartment, the profound neutrophilia, which reduces as patients improve, suggests that neutrophils and their products may contribute to the pathology. Activated neutrophils produce large quantities of S100A1 and A2, HMGB1 and free radicals. Furthermore, neutrophils release fibrous scaffolds containing DNA, histones, neutrophil elastase and myeloperoxidase that trap and kill pathogens, but also contribute to the development of acute respiratory distress syndrome^6,7. To support their role, elevated neutrophil products have been identified in the sera of COVID-19 patients that correlate with clinical parameters such as C-reactive protein, D-dimer, and lactate dehydrogenase⁸.

The most unusual cell lineage in COVID-19 is the monocytes. In health, monocytes proliferate in the bone marrow and do not proliferate in blood, and may do so again in tissues as they differentiate⁹. Monocytes expressing proliferation markers in the blood have not previously been reported in COVID-19 patients. In these patients, up to 80% of monocytes were expressing proliferation marker Ki67, the highest consistently in patients with severe disease. The most likely explanation for this phenomenon is premature release of monocytes from the marrow, due to inflammatory signals. This has been previously reported in pandemic H1N1 influenza^10,11.

Equally remarkable was the reduced expression of COX-2 in monocytes, from an early stage in disease in patients with a severe disease trajectory. COX-2 facilitates the production of prostanoids including prostaglandin E2 (PGE2) and increases lipid mediators involved in vasodilation, vascular permeability and immune cell chemotaxis¹². It may also hinder the production of anti-inflammatory/pro-resolution prostanoids, the resolvins¹³. Some viruses (e.g avian H1N1 influenza the 2003 SARS coronavirus¹⁴) up-regulate the cyclo-oxygenase pathway to regulate viral replication. However, its reduction in monocytes in severe lung viral infection has not been reported previously. Reduced COX-2 alongside high IP-10, as seen here in severe COVID-19 patients, is an immune profile strongly associated with pathology in idiopathic pulmonary fibrosis (IPF)¹⁵. Therefore our data indicate a strong fibrotic signature in patients with severe disease, supporting studies observing an unusual pattern of fibrosis in COVID-19.

The function of monocytes in COVID-19 was also assessed by cytokine immunostaining. It was noteworthy that intracellular TNF was augmented in COVID-19 monocytes at hospital admission, supporting the hypothesis that early anti-TNF should be tested in COVID-19¹⁶.

Our data concurs with several features of the disease in Wuhan, China, and is corroborated in single cell RNA sequencing of bronchoalveolar lavage cells at a single time point¹⁷, but importantly we extend this into a longitudinal analysis that gives greater insight. Similarities include elevated CRP and IL-6 in patients at the time of hospitalization who eventually died¹⁸ and increased IP-10 (CXCL-10) in those who later developed severe disease¹⁹. IP-10 is an interferon-inducible chemokine receptor that facilitates directed migration of many immune cells²⁰ and is elevated in other coronavirus infections including MERS-CoV and SARS-CoV²¹ and in Influenza virus of swine origin (H1N1)^22,23. The heightened levels of monocyte-chemoattractant protein 1 (MCP-1) upon admission further indicate dysregulation of monocyte function and migration in patients with severe disease. Importantly, IL-6, IP-10 and MCP-1 levels are highest around the time of hospital admission, but are reduced rapidly as patients are admitted to ITU, which likely signifies exhaustion of immune cells and the lineages producing them.

There are, of course, limitations to any study of samples during a viral pandemic for which there is no vaccine. However, we believe that these do not reduce the importance of the messages derived from our study. A longitudinal analysis in real time for phenotypic, functional and soluble markers naturally limits the number of patients interrogated. In-depth analysis of smaller cohorts however, is necessary to gain insight into mechanism and is of interest to the pharmaceutical industry. In addition to the limitations of control subjects addressed above, the only other is that patients may not accurately define the onset of symptoms. As data is plotted per patient, however, this does not affect the interpretation of the results.

There are clinical implications of our data, as far as it has been analysed. Using non-steroidal anti-inflammatory drugs (NSAIDs) remains controversial²⁴ and our study would suggest it is not desirable, as it would compound the already low COX-2²⁵. Since most of the pathogenic mechanisms involve myeloid cells, neutrophils and monocytes, it would be advantageous to reduce their influx to the lung once lung pathology is established. Relevant strategies include inhibition of the complement anaphylatoxin C5a²⁶ or IL-8 (CXCL8), which are strong chemoattractants for many immune cells, including neutrophils²⁷. Antagonism of CXCR2 that mobilises neutrophil and monocyte from the bone marrow, neutrophil elastase inhibitors and inhibition of G-CSF, IL-23 and IL-17 that promote neutrophil survival, are also options²⁸. Anti-IL-6 and anti-TNF agents are already being investigated for COVID-19 treatment and are relevant to neutrophils, which express the requisite cytokine receptors. Furthermore, JAK inhibitors are currently in clinical trials and may also reduce neutrophil levels²⁹. Targeting toxic products of neutrophils such as S100A1/A2, HMGB1 and free radicals³⁰, but also the formation of NETS could be beneficial³⁰.

Finally, one of the clearest messages is that treating patients early after hospitalization is likely to be most beneficial - cytokines and cell levels are high and there is the potential for recovery. There are also potential predictors of severity which could stratify patients for treatment at admission, including high neutrophil/Tcell ratio, high Ki67 and low COX2 in monocytes.

Contributors

ERM, MM, SBK, JEK, CJ, TS, JRG and TH performed the study design, tissue processing, data collection, data analysis, data interpretation and manuscript writing.

SK and MR performed data analysis and generated the figures.

AU, NDB, PD, AS, TF performed sample collection and biobanking, AU, AS, TF contributed to data interpretation, presentation of clinical data and paper construction L-PH, RTRC, GL, CIRCO and MF contributed to study design, patient consent and sample collection and manuscript preparation

Declaration of interests

GL is Co-founder and Scientific Advisory Board Member of Gritstone Oncology Inc., which is a public company that develops therapeutic vaccines (primarily for the treatment of cancer).

Data sharing statement

Individual participant data that underlie the results reported will be shared immediately following publication to anyone for any purpose. Data is available by contacting TH or JG and requestors will be required to sign a data access agreement.