key: cord-0866831-t0l6mlm5
authors: Vijayakumar, Bavithra; Boustani, Karim; Ogger, Patricia P.; Papadaki, Artemis; Tonkin, James; Orton, Christopher M.; Ghai, Poonam; Suveizdyte, Kornelija; Hewitt, Richard J.; Desai, Sujal R.; Devaraj, Anand; Snelgrove, Robert J.; Molyneaux, Philip L.; Garner, Justin L.; Peters, James E.; Shah, Pallav L.; Lloyd, Clare M.; Harker, James A.
title: Immuno-proteomic profiling reveals aberrant immune cell regulation in the airways of individuals with ongoing post-COVD-19 respiratory disease
date: 2022-01-26
journal: Immunity
DOI: 10.1016/j.immuni.2022.01.017
sha: ade258852ad3337a3e86040d5fc94b3aaa4e2d10
doc_id: 866831
cord_uid: t0l6mlm5

Some patients hospitalized with acute COVID-19 suffer respiratory symptoms that persist for many months. We delineated the immune-proteomic landscape in the airway and peripheral blood of healthy controls and post-COVID-19 patients 3 to 6 months after hospital discharge. Post-COVID-19 patients showed abnormal airway (but not plasma) proteomes, with elevated concentration of proteins associated with apoptosis, tissue repair and epithelial injury versus healthy individuals. Increased numbers of cytotoxic lymphocytes were observed in individuals with greater airway dysfunction, while increased B cell numbers and altered monocyte subsets were associated with more widespread lung abnormalities. 1 year follow-up of some post-COVID-19 patients indicated that these abnormalities resolved over time. In summary, COVID-19 causes a prolonged change to the airway immune landscape in those with persistent lung disease, with evidence of cell death and tissue repair linked to ongoing activation of cytotoxic T cells.

Immune-proteome landscape post-COVID19

Some patients hospitalized with acute COVID-19 suffer respiratory symptoms that 25 persist for many months. We delineated the immune-proteomic landscape in the 26 airway and peripheral blood of healthy controls and post-COVID-19 patients 3 to 6 27 months after hospital discharge. Post-COVID-19 patients showed abnormal airway 28 (but not plasma) proteomes, with elevated concentration of proteins associated with 29 apoptosis, tissue repair and epithelial injury versus healthy individuals. Increased The acute immunological and inflammatory events that occur during human respiratory 63 virus infections, including SARS-CoV-2, are relatively well described (Harker and 64 Lloyd, 2021) . In contrast, the immunological landscape of the human respiratory tract 65 after recovery from acute viral infection is poorly understood. SARS-CoV-2 infection 66 results in formation of long-lasting systemic immunological memory, with virus-specific 67 antibodies and T cell responses still detectable in the majority of those infected at least 68 8 months post infection and higher titers seen in previously hospitalized individuals 69 (Dan et al., 2021) . Circulating lymphocyte counts and the function and frequency of 70 monocytes are also reduced during acute disease, but they appear to return to normal 71 shortly after resolution of acute disease (Mann et al., 2020; Scott et al., 2020) . 72

Likewise, plasma concentrations of inflammatory mediators such as IL-6 and CXCL10, 73 that are highly elevated in acute disease, reduce as individuals recover (Rodriguez et We recruited 38 patients undergoing bronchoscopy for investigation of persistent 104 respiratory abnormalities 3-6 months following acute SARS-CoV-2 infection (post-105 COVID19) (Figure 1) . All patients had ongoing respiratory symptoms and/or 106 radiological pulmonary abnormalities on computed tomography (CT). Peripheral blood 107 and bronchoalveolar lavage (BAL) was obtained. The post-COVID19 cohort was 108 stratified based on the level of respiratory support used during their initial 109 hospitalization with acute COVID19, into moderate (no/minimal oxygen administered), 110 severe (non-invasive ventilation) and very severe (invasive ventilation). We used BAL 111 fluid, plasma and historic flow cytometry analysis obtained from 29 healthy volunteers 112 recruited prior to the COVID19 pandemic as controls (demographic information in 113 Table S1 ). 114

We compared the cellular composition of BAL fluid in post-COVID19 patients to healthy 116 controls (HC) by flow cytometry ( Figure S1A ). Post-COVID19 patients had significantly 117 higher numbers of cells in their airways compared to the healthy controls (Figure 2A) . 118

This increased cellularity was due to elevated numbers of airway macrophages (AM), 119 T and B cells ( Figure 2B ). CD56 + CD3 -(natural killer, NK) and CD56 + CD3 + (NKT) cells, 120 CD14 + monocytes and eosinophils were similar to those in healthy controls, while 121 neutrophils were decreased ( Figure 2B) . As a proportion of airway leukocytes, CD14 + 122 monocytes and neutrophils were decreased in post-COVID19 patients compared to 123 controls ( Figure S1B) . 124

No association between the severity of acute COVID19 in hospital and the immune 126 cell composition of the post-COVID19 BAL was observed ( Figure 2B) . In contrast to 127 the peripheral lymphopenia that is associated with acute COVID19 (Chen and Wherry, 128 2020), we found that in this post-COVID19 patient cohort the frequency of T cells, B 129 cells and CD14 + monocytes in the peripheral blood was similar to healthy controls 130 ( Figure S1C) , although the proportion of NK and NKT cells was decreased ( Figure  131 Post-COVID19 airways display a proteomic signature not reflected in blood. 136 137 We next evaluated the airway and blood (plasma) proteomes, using the Olink platform 138 to measure 435 unique proteins in BAL and plasma from 19 post-COVID19 patients 139 and 9 healthy controls. The proteins measured were highly enriched for immune-140 inflammatory processes (Table S2A-C) . Principal component analysis (PCA) of BAL 141 proteomes revealed differences between post-COVID19 patients and healthy controls 142 ( Figure 3A) , with separation of case and controls most evident along PC1. In plasma, 143 PCA also revealed differences, most evident along PC2, although the differences were 144 less marked than for BAL. However, in both BAL and plasma there was considerable 145 overlap in the spatial location of post-COVID19 and control in the PCA plots, indicating 146 heterogeneity in post-COVID19 patients, with some displaying similar proteomic 147 profiles to healthy controls. Unsupervised hierarchical clustering revealed two major 148 clusters in BAL, one consisting predominantly of post-COVID19 samples and the other 149 predominantly healthy controls ( Figure S2A ). In contrast, in plasma, there was no 150 visible structure to the clustering and lack of clear separation of cases and controls 151 ( Figure S2B) . These analyses indicates that the post-COVID19 phenotype is 152 predominantly reflected by the airway proteome rather than the peripheral blood. 153

Differential protein abundance analysis comparing post-COVID19 cases with healthy 154 controls identified 22 proteins in BAL with significantly altered concentration (5% false 155 discovery rate, FDR) ( Figure 3B -C, Table S2D ). These were all upregulated in post-156 COVID19 patients compared to healthy controls ( Figure 3C ). To provide a succinct 157 and standardised nomenclature, we report proteins by the symbols of the genes 158 encoding them (see Table S2A for mapping to full protein names). The proteins that 159 were most significantly differentially abundant between post-COVID19 and controls 160 were: SERPINA7 (thyroxine binding globulin), DPP4 (dipeptidyl peptidase 4), 161 SERPINA5 (plasma serine protease inhibitor), KLK6 (kallikrein related peptidase-6), 162 LYVE1 (Lymphatic vessel endothelial hyaluronic acid receptor 1), AREG 163 (amphiregulin), F3 (factor 3), FLT3LG (Fms-related tyrosine kinase 3 ligand), QPCT 164 (glutaminyl-peptide cyclotransferase), MMP3 (metalloproteinase-3) and SRC (Proto-165 oncogene tyrosine-protein kinase Src) (Figure 3C-D) . Pathway annotation of the 22 166 upregulated proteins using String-DB highlighted "leucocyte activation", "regulation of 167 cell death", "response to injury" and "response to wounding" ( Table S2E) In contrast to BAL, no significant differences between protein levels were detected in 176 plasma in post-COVID19 patients versus healthy controls ( Table S2F) control samples on the PCA plots for both BAL and plasma. These data suggest that 194 there are differences in both the BAL and plasma proteomes of post-COVID19 cases 195 compared to healthy controls, but that the effects are much stronger in BAL. 196 To increase power, and investigate potential protein-protein relationships, we utilized 197 a network analysis method, Weighted Coexpression Network Analysis (WGCNA) 198 (Langfelder and Horvath, 2008; Zhang and Horvath, 2005) , that leverages the 199 correlation between proteins to enable dimension reduction, thus reducing multiple 200 testing burden. We used WGCNA to identify modules of correlated proteins, and then 201 tested for association between these protein modules (represented quantitatively by 202 an eigenprotein value) and case/control status. In BAL, this revealed two modules 203 ('red' and 'blue') significantly associated with case/control status (5% FDR) (Table  204 S2G-I). 205 J o u r n a l P r e -p r o o f Immune-proteome landscape post-COVID19

The red module consisted of 37 proteins ( Figure S4A , Table S2H ), characterized by 206 proteins associated with chemotaxis, inflammation, cell death and repair. In post-207 COVID19 patients, we observed co-upregulation of groups of related red module 208 proteins such as the CXCR3 chemokines (CXCL9, CXCL10 and CXCL11), and IL1A 209 (interleukin-1A) and its antagonist IL1RN (Figure S4A and B) . We used STRING-db 210 to visualize known or predicted relationships between proteins in the module ( Figure  211 S4A and B). To highlight putative key proteins in the red and blue modules in a data-212 driven way, we identified hub proteins, defined as those that are highly interconnected 213 in the proteomic network defined by WGCNA (Table S2J ). This identified CASP3 214 these, LYVE1 and VASN were also identified in the differential abundance analysis. 232

In contrast to the BAL network analysis, no protein modules in plasma were associated 233 with case-control status. This suggests that persistent post-COVID19 respiratory 234 abnormalities have a demonstrable proteomic signature in BAL that is distinct 235 compared to that of healthy controls. In contrast, we were unable to detect changes in Concomitant with enhanced BAL neutrophilia, the major neutrophil chemokine CXCL8 353 was increased in post-COVID19 compared to HC BAL, and CXCL8 concentration 354 significantly correlated with airway neutrophils (Figure 5E ). Similarly CCL2 was 355 significantly increased in post-COVID19 BAL, and tightly correlated to BAL monocyte 356 numbers ( Figure 5G) . CXCL8 or CCL2 did not segregate with worsened CT, FVC or 357 TLCO (Figure 5G and H) . with an FVC less than 90% of that predicted (Figure 6B and C) . No other T cell 388 population or subset showed significance in individuals with reduced FVC, but similar 389 trends were present for activated CD4 T cells (Figure 6B and C) . Conversely, analysis 390 of B cells revealed that individuals with increased CT abnormality or reduced FVC or 391 TCLO had significantly increased memory B cells in their airways, while naïve B cells 392 and plasmablasts were not different ( Figure 6D) . Table S3 ). The total number of BAL cells recovered was greatly reduced 430 in all 3 patients between the initial bronchoscopy and the 1 year follow up 431 bronchoscopy, comparable to healthy control airways ( Figure 7B) . (B-E, G) was tested 762 by Mann-Whitney U test. Benjamini-Hochberg adjusted (5% FDR) *P < 0.05, **P < 763 0.01, ***P < 0.005, ****P < 0.001. A, F & G are display spearman's rho correlation. 764 Conventional flow cytometry data was analysed using FlowJo v 10.6 (Tree Star). Data 884 was pre-gated to exclude doublets and dead cells. In BAL samples CD45 + cells were 885 selected, and immune cell populations were identified using the gating strategy shown 886 in Figure S1A . Percentages of the CD45 + gate were calculated. In blood samples, 887 leukocytes were selected based on FSC and SSC and immune cell populations were 888 identified using the gating strategy shown in Figure S1A . Percentages of total 889 leukocytes were calculated. High-parameter spectral deconvolution flow cytometry 890 data from the Cytek Aurora was analysed using Cytobank (Beckman). tSNE analysis 891 was performed on 300,000 events from 11 files. Iteration number was set to 1500 with 892 a perplexity of 30 and theta of 0.5. FlowSOM analysis was performed subsequently 893 using hierarchical consensus clustering with 12 metaclusters, 100 clusters and 10 894

iterations. Manual gating of high parameter cytometry data was carried out as shown 895

in Figure S6A . Heatmaps were generated from median fluorescence values in Prism 896 9.0 (GraphPad 

Proteomic data was normalised using standard Olink workflows to produce relative 965 protein abundance on a log2 scale ('NPX'). BAL and plasma proteomic data were 966 normalised separately. Quality assessment was performed by (1) examination of Olink 967 internal controls and (2) inspection of boxplots, relative log expression plots, and PCA. 968 PCA was performed using singular value decomposition. Following these steps, 2 969 clear outlying samples were removed from the BAL dataset. To identify proteins that 970 were differentially abundant between case and controls, for each protein we performed 971 linear regression (lm function in R) with case/control status as the independent variable 972 and protein concentration (NPX/ml) as the dependent variable. P-values were adjusted 973 for multiple testing using the Benjamini-Hochberg procedure (p.adjust function in R). A 974 5% false discovery rate was used to define statistical significance. We used the 975 WGCNA R package (Langfelder and Horvath, 2008; Zhang and Horvath, 2005) to 976 create a weighted protein correlation network. Prior to WGCNA analysis, protein data 977 were scaled and centred, and missing data were imputed using the R caret package. 978

We used the WCGNA adjacency function to produce a weighed network adjacency 979 matrix, using parameters "type=signed" and "power=13". This soft-thresholding power 980 was selected as the lowest power to achieve approximate scale-free topology. We next 981 defined a topological overlap matrix of dissimilarity using the TOMdist function. 982

Clusters ('modules') of interconnected proteins were identified using hierarchical 983 clustering and the cutreeDynamic function with parameters: method="hybrid", 984 deepSplit=2, minClusterSize=15. We then tested association of these modules with 985 case/control status. Multiple testing correction was performed to account for the 986 number of modules. We report both Benjamini-Hochberg and Bonferroni adjusted p-987 values to provide two levels of stringency. To assess the distribution of p-values from 988 the differential protein abundance analyses, we plotted histograms and constructed 989 QQ plots. QQ plots were made by comparing the expected distribution of -log10 P 990 values under the null hypothesis of no proteomic differences between post-COVID19 991 patients and controls to the observed p-values for the 435 proteins. We performed 992 J o u r n a l P r e -p r o o f 

Antifibrotics Modify 1028 B-Cell-induced Fibroblast Migration and Activation in Patients with Idiopathic 1029 Pulmonary Fibrosis

The Transferrin Receptor 1032 CD71 Delineates Functionally Distinct Airway Macrophage Subsets during Idiopathic 1033 Pulmonary Fibrosis

Systems biological 1036 assessment of immunity to mild versus severe COVID-19 infection in humans

Aberrant populations of 1040 circulating T follicular helper cells and regulatory B cells underlying idiopathic 1041 pulmonary fibrosis

Exuberant fibroblast 1044 activity compromises lung function via ADAMTS4

Dynamics of human monocytes and 1047 airway macrophages during healthy aging and after transplant

T cell responses in patients with COVID-19

Immune signatures underlying post-acute COVID-19 lung 1052 sequelae

Immunological memory to SARS-CoV-2 1055 assessed for up to 8 months after infection

The Role of Immune 1057 and Inflammatory Cells in Idiopathic Pulmonary Fibrosis

Features of 20 1060 133 UK patients in hospital with covid-19 using the ISARIC WHO Clinical 1061 Characterisation Protocol: prospective observational cohort study

Characterisation of in-1064 hospital complications associated with COVID-19 using the ISARIC WHO Clinical 1065 Characterisation Protocol UK: a prospective, multicentre cohort study

Balancing Immune Protection and Immune 1068 Pathology by CD8(+) T-Cell Responses to Influenza Infection

Distinct developmental pathways from 1071 blood monocytes generate human lung macrophage diversity

proteomic analysis of severe COVID-19 reveals survival-associated signatures, tissue-1080 specific cell death, and cell-cell interactions

Longitudinal 1083 proteomic profiling of dialysis patients with COVID-19 reveals markers of severity and 1084 predictors of death

Peripheral 1087 and lung resident memory T cell responses against SARS-CoV-2

Pulmonary function and radiological features 4 months after COVID-19: first results 1092 from the national prospective observational Swiss COVID-19 lung study

Six-month Follow-up Chest CT Findings after Severe

Overlapping and distinct features of viral and 1098 allergen immunity in the human lung

The Respiratory Microbiome in Chronic 1101

Hypersensitivity Pneumonitis Is Distinct from That of Idiopathic Pulmonary Fibrosis

RSV-specific airway resident 1105 memory CD8+ T cells and differential disease severity after experimental human 1106 infection

Antigen 1108 persistence and the control of local T cell memory by migrant respiratory dendritic cells 1109 after acute virus infection

SARS-CoV-2 organising pneumonia: 'Has there 1111 been a widespread failure to identify and treat this prevalent condition in COVID-19?'. 1112

A dynamic 1115 COVID-19 immune signature includes associations with poor prognosis

WGCNA: an R package for weighted 1118 correlation network analysis

The MERS-CoV Receptor DPP4 as a Candidate Binding Target of the SARS-CoV-2

Single-cell landscape of bronchoalveolar immune cells in patients with 1124 COVID-19

Longitudinal analyses reveal 1127 immunological misfiring in severe COVID-19

Long-COVID': a cross-sectional 1130 study of persisting symptoms, biomarker and imaging abnormalities following 1131 hospitalisation for COVID-19

Longitudinal immune 1134 profiling reveals key myeloid signatures associated with COVID-19

Monocyte-derived alveolar macrophages drive lung fibrosis and persist in the 1138 lung over the life span

Amphiregulin-Producing 1141

Pathogenic Memory T Helper 2 Cells Instruct Eosinophils to Secrete Osteopontin and 1142 Facilitate Airway Fibrosis

Persistent 1145 Post-COVID-19 Interstitial Lung Disease. An Observational Study of Corticosteroid 1146 Treatment

Post-acute COVID-1149 19 syndrome

B cells in chronic 1151 obstructive pulmonary disease: moving to center stage

SARS-CoV-2 infection 1155 generates tissue-localized immunological memory in humans

Alveolar macrophages are a major determinant of early 1158 responses to viral lung infection but do not influence subsequent disease development

Dipeptidyl peptidase 4 is a 1162 functional receptor for the emerging human coronavirus-EMC

Systems-Level 1165 Immunomonitoring from Acute to Recovery Phase of Severe COVID-19

Distinct cellular immune 1169 profiles in the airways and blood of critically ill patients with COVID-19

The CD8 T Cell Response to Respiratory Virus 1171 Infections

Recovery of monocyte exhaustion is 1174 associated with resolution of lung injury in COVID-19 convalescence. medRxiv

Long Covid in adults discharged 1178 from UK hospitals after Covid-19: A prospective, multicentre cohort study using the 1179 ISARIC WHO Clinical Characterisation Protocol. medRxiv

Lung airway-surveilling 1182 CXCR3(hi) memory CD8(+) T cells are critical for protection against influenza A virus

Dynamics of influenza-induced lung-resident memory T cells 1186 underlie waning heterosubtypic immunity

Longitudinal profiling of 1189 respiratory and systemic immune responses reveals myeloid cell-driven lung 1190 inflammation in severe COVID-19

Location, location, location: Tissue 1192 resident memory T cells in mice and humans

Inflammatory profiles across the spectrum of disease reveal a distinct role for GM-CSF 1196 in severe COVID-19

CT Lung Abnormalities after COVID-19 at 3 Months and 1 Year 1199 after Hospital Discharge

Diverse Functional Autoantibodies in Patients 1202 with COVID-19. medRxiv

Lung-resident memory CD8 T cells (TRM) are indispensable for optimal cross-1205 protection against pulmonary virus infection

Matrix 1208 metalloproteinase 3 is a mediator of pulmonary fibrosis

A general framework for weighted gene co-1210 expression network analysis

pathway enrichment analysis for the 435 proteins measured. This was performed using 993 terms from KEGG database (Supplementary File 1B) and the Reactome database 994 (Supplementary File 1C) . Protein modules were visualised using STRING 995 (https://string-db.org/), with known or suspected interconnections between module 996 members displayed as edges in a network diagram. An edge represents a protein-to-997 protein relationship defined as shared contributions to a particular function, and not 998 necessarily implying physical binding. In Figure 3C , edge colour indicates the type of 999 evidence for the relationship: turquoise represents known interactions from curated 1000 databases; magenta represents experimentally determined interactions; green 1001 represents predicted Interactions from gene neighbourhood analysis; red represent 1002 predicted interactions from gene fusions, blue represent predicted Interactions from 1003 gene co-occurrence; light green represents interaction from text-mining; black 1004 represents interaction from co-expression data, and violet represents information from 1005 protein homology.