Longitudinal analysis of the lung proteome reveals persistent repair months after mild to moderate COVID-19

Summary

In order to assess homeostatic mechanisms in the lung after COVID-19, changes in the protein signature of bronchoalveolar lavage from 45 patients with mild to moderate disease at three phases (acute, recovery, and convalescent) are evaluated over a year. During the acute phase, inflamed and uninflamed phenotypes are characterized by the expression of tissue repair and host defense response molecules. With recovery, inflammatory and fibrogenic mediators decline and clinical symptoms abate. However, at 9 months, quantified radiographic abnormalities resolve in the majority of patients, and yet compared to healthy persons, all showed ongoing activation of cellular repair processes and depression of the renin-kallikrein-kinin, coagulation, and complement systems. This dissociation of prolonged reparative processes from symptom and radiographic resolution suggests that occult ongoing disruption of the lung proteome is underrecognized and may be relevant to recovery from other serious viral pneumonias.

Graphical abstract

Highlights

• •Host inflammatory proteins are downregulated by 3 months after symptom onset
• •Lung matrisome expression mirrors acute chest tomography abnormalities
• •Lavage SPP1 levels are predictive of post-COVID-19 interstitial lung disease
• •Persistent lung repair is occurring at 9 months post-COVID-19

Kanth et al. evaluate the lung lavage proteome in 45 persons with mild to moderate COVID-19 after acute symptom onset (40 days, mean), recovery (82 days), and convalescence (284 days). Despite clinical resolution at convalescence, the lung proteome does not normalize, suggesting an underrecognized prolongation of lung repair.

Introduction

Since the declaration of the COVID-19 pandemic over 3 years ago, more than 750 million cases and 6.5 million deaths have been documented worldwide.^1,2 Infection from severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has resulted in a wide variety of clinical manifestations ranging from asymptomatic illness to hypoxic respiratory failure and death.^3,4,5 While the inflammatory and immunologic environment of acute COVID-19 within the lung has been well described,^6,7,8,9,10 there remains a dearth of knowledge investigating the convalescent stage of illness in patients who have recovered clinically from COVID-19, which represents the vast majority of infected individuals.

Longitudinal evaluation of the blood proteome during the acute phase of SARS-CoV-2 infection has demonstrated increased concentrations of circulating pro-inflammatory cytokines, proteins related to viral response and interferon pathway, and activation of monocytes, macrophages, and neutrophils that return to normal levels during recovery.^{11,12,13,14,15} In parallel, alterations to the adaptive immune response and persistent systemic SARS-CoV-2 immunological memory are noted in the blood at least 6 months post-infection in previously hospitalized patients with COVID-19.^6,16,17 Local cellular and proteomic composition within the lungs reveals increases in immune cells and tissue injury mediators at 6 months post-infection in patients with persistent radiographic abnormalities, but a limited assessment of five biomarkers shows resolution at 1 year.¹⁸ It is still unclear whether there are persistent alterations in the lung immunoproteomic signature after 6 months from initial infection in patients who have overt clinical and functional recovery.

In the current study, we evaluated changes in the lung proteome in patients during acute, recovery, and convalescent periods of SARS-CoV-2 infection, using three proteomic platforms applied to bronchoalveolar lavage (BAL). In contrast to healthy persons, the post-COVID-19 lung proteome revealed persistent elevations of proteins associated with tissue repair and activation of proteostasis up to 9 months after symptom onset. Notably, these processes were still occurring despite the resolution of clinical and radiographic signs of the prior infection. Serial follow-up suggests that the pulmonary proteome did not return to normal for almost 1 year after the onset of COVID-19 and that there is ongoing host repair in response to the initial infection even in patients who have clinically recovered.

Results

Discrete biological phenotypes in patients recovering from COVID-19 are associated with initial illness severity, vaccination status, and the degree of radiographic abnormalities

We analyzed samples from 45 patients with PCR-confirmed SARS-CoV-2 infection, with a variety of initial clinical presentations ranging from ambulatory to those requiring high-flow nasal cannulas. None of the patients required mechanical ventilation, and all patients survived to convalescence (Table S1). BAL and peripheral blood were collected in pre-defined acute, recovery, and convalescent phases, and proteomic analysis was performed (Figure 1; Table S2).

Figure 1.

Study timeline and methodology

Samples from healthy controls (pre-COVID-19, recruited from 2017 to 2018, n = 16) and patients with COVID-19 (n = 45) were collected. COVID-19 samples were obtained in acute (1–40 days post symptom onset, median 21 days), recovery (>40 days to 12 weeks post symptom onset, median 82.5 days), and convalescent (>12 weeks to 13 months, median 278 days) phases. Peripheral blood was obtained at the time of BAL. Proteomic analysis was performed on BAL and blood samples using data-independent acquisition mass spectrometry (DIA-MS), a proximal extension assay (PEA, Olink), and targeted immunoassays (Mesoscale and Luminex) as depicted above. Image created with BioRender (biorender.com).

Principal component analysis (PCA) of the acute phase BAL samples separated the patients into two predominant clusters, cluster 1 and cluster 2, with most separation between clusters occurring in principal component 1 in all three platforms (Figure 2A). The cluster 1 patients grouped with healthy controls by proximal extension assay (PEA) platform. However, comparison of acute cluster 1 BAL to healthy controls, using data-independent acquisition mass spectrometry (DIA-MS) and the targeted immunoassays, revealed separation by PCA, indicating an abnormal proteomic signature compared to uninfected individuals. The comparison of PCA between all three proteomic platforms revealed similar clustering of individual patients, further validating these derived phenotypes (Figure 2A). Analysis of the demographic and clinical features in both phenotypes showed cluster 2 patients were predominantly older males and presented with higher median peak severity scores and greater chest computed tomography (CT) lesion burden, and the majority were unvaccinated (Figure 2B; Table S3).

Figure 2.

Heterogeneous phenotypes of COVID-19 are identified in the lung in the acute phase

(A) Principal component analysis (PCA) of acute phase BAL. Asterisks indicates two patients who clustered differently in targeted immunoassays.

(B) Heatmap showing differentially expressed acute proteins (DEPs) in BAL using PEA between two phenotypic clusters.

(C) Volcano plot of DEPs between cluster 1 and cluster 2 (red), using PEA. Horizontal dashed line denotes a cutoff of false discovery rate (FDR) < 0.05 after age and gender adjustments.

(D) Biologically relevant common DEPs in PEA and DIA-MS platforms. Healthy controls are shown for reference comparison.

(E) Ingenuity Pathway Analysis (IPA) enrichment analysis of overexpressed proteins in cluster 2 compared to cluster 1, using PEA and DIA-MS. Z score indicates a predicted activation or inhibition of a pathway, i.e., a negative z value represents pathway inhibition, and a positive z value represents pathway activation.

Statistical analysis: (C) p values are adjusted by the Benjamini-Hochberg method for an approximate control of FDR at 5% level. (D) Mann-Whitney U test. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.005. The horizontal line represents the median. For DIA-MS, proteins with >50% missing values were excluded. Missing values < 50% were imputed by minimum value/2.

After adjusting for age and gender, a comparison of BAL from cluster 1 to cluster 2 individuals identified 786 differentially expressed proteins (DEPs) using PEA, 409 DEPs using DIA-MS, and 3 DEPs (CCL20, IL2RA, and REG3A) using the targeted immunoassays (false discovery rate [FDR] 5%, absolute fold change (absFC) > 2) (Figure 2C; Table S4). Most DEPs had increased expression in cluster 2 compared to cluster 1. PEA identified decreased expression of only 4 proteins (ADGRE5, SOD1, ANXA5, and PVALB), and DIA-MS analysis identified 136 proteins with decreased differential abundance comparing cluster 2 versus cluster 1 (Figure 2C; Table S4). Decreased ANXA5A expression was detected with both platforms as well as two isoforms of superoxide dismutase (SOD1 and SOD2, in PEA and DIA-MS, respectively).

A comparison of matched BAL and plasma immunoassays from the acute phase showed a linear correlation in 21 biomarkers (Figure S1). However, PCA of the plasma immunoassays did not reveal differences in the two clusters after adjustments for age and gender (Table S4).

Functional enrichment analysis of common pathways across PEA and DIA-MS proteomic platforms comparing the cluster 2 to cluster 1 phenotypes highlighted increased expression of the acute host defense response (CORO1A, DEFA1, LBP, C4, C5, and KLK1), the inflammatory cytokine cascade (IL6, MMP9, CXCL10, CCL2, CCL8, and TNF), epithelial and extracellular matrix (ECM) (matrisome) repair and fibrogenesis (PCOLCE, TIMP1, COL6A3, COL6A1, and MMP8), and cellular apoptosis (CASP3, CASP8, BAX, and DIABLO) (Figures 2D and 2E). A sensitivity analysis adjusting for days from symptom onset to sample acquisition, in addition to age and gender, showed similar functional enrichment analysis (Table S5).

Enrichment analysis of proteins with decreased expression in cluster 2 compared to the cluster 1 phenotype via DIA-MS revealed pathways associated with phagosome maturation (ATP6V1A, CANX, human leukocyte antigen DR-A (HLA-DRA), LAMP1, and RAB5C), coordinated lysosomal expression and regulation (CLEAR) signaling pathway (ASAH1, ATP6V1A, GAA, TPP1, VPS35, and LAMP1), and iron homeostasis (ACO1 and TFRC) (Figure S2). These three pathways have all been implicated in macromolecule degradation and cellular homeostasis mechanisms.^19,20

These data identify an inflamed (cluster 2) and uninflamed (cluster 1) COVID-19 phenotype based on the acute phase proteomic signature. Clinical factors including vaccination status, higher peak severity of illness, and the degree of radiographic abnormalities, independent of age and gender, correspond with the inflamed phenotype in the early post-infectious phase.

Longitudinal evaluation of the lung proteome reveals evolution in the COVID-19 inflamed phenotype with a decrease in immune and fibrogenic mediators

Paired BAL samples from patients with COVID-19 obtained during acute, recovery, and convalescent phases (n = 12 patients, n = 6 in cluster 2 and n = 6 in cluster 1) were analyzed using PEA. A one-way blocked analysis of variance was conducted separately for cluster 1 and cluster 2 patients to discover DEPs over phases. Overall, 825 proteins were declared significant for cluster 2 patients (FDR 5%), whereas no protein was identified for cluster 1 patients. Between the two clusters of patients, differences in expression patterns over time tested using a linear mixed-effects model revealed a total of 224 DEPs (FDR 5%) (Table S6). PEA protein expression profiles in cluster 2 over time revealed three major trends: trend 1 (23 proteins had decreased expression from the acute to recovery to convalescent phases), trend 2 (199 proteins had decreased expression from the acute to the recovery phase and then increased from the recovery to the convalescent phase), and trend 3 (2 proteins had increased expression from the acute to the recovery phase and decreased from the recovery to the convalescent phase) (Figure 3A; Table S6). Most of the dynamic changes in the proteome occurred during the first 3 months, from the acute to the recovery phase, resulting in more homogeneity in the proteomic signatures over time in all cluster 2 patients (Figure 3B). The two proteins classified in trend 3, trefoil factor 3 and SOD1, function in lung injury repair and reactive oxygen species (ROS) mitigation, respectively.^21,22 Notably, while both proteins had significantly higher expression levels in recovery and convalescent phases compared to the acute phase, they were not different from healthy control BAL (Figure 3C), suggesting a return to baseline levels over time.

Figure 3.

Lung proteomic signature in the inflamed phenotype evolves rapidly within the first 3 months

(A) Parallel plots for three protein trends identified in cluster 2 phenotype patients with sequential lavage at acute, recovery, and convalescent phases using PEA. y axis represents the normalized protein expression (NPX) value for each protein.

(B) Heatmap displaying protein expression over time in paired cluster 2 patients using PEA.

(C) Two proteins in trend 3 (TFF3 and SOD1) shown in acute, recovery, and convalescent phases and compared to healthy control.

(D) Overlapping proteins with decreased expression in the recovery phase in PEA and DIA-MS platforms. Healthy controls are shown for reference comparison.

(E) Gene Ontology and Reactome enrichment analysis for trend 1 and 2 DEPs using Metascape.

(F) IPA enrichment analysis for acute versus recovery phases in paired cluster 2 proteins identified by DIA-MS, pathways with Z score cutoff of 2. Z score indicates a predicted activation or inhibition of a pathway, i.e., a negative z value represents pathway inhibition, and a positive z value represents pathway activation.

Statistical analysis: (A) testing for mean differences among phases was by blocked-ANOVA model. Testing for the interaction between clusters and phases was used to detect expression pattern differences. (C and D) Mann-Whitney U test. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.005. Horizontal line represents the median.

Parallel analysis of paired cluster 2 samples using DIA-MS comparing acute to recovery phases (n = 8 patients) was notable for 343 DEPs (FDR 5%) (Table S6), whereas recovery to convalescent phases (n = 5 patients) found no DEPs. When comparing common DEPs that decreased from acute to recovery phases in DIA-MS and PEA, four proteins were identified (PCOLCE, ITIH3, SSB, and RUVBL1) (Figure 3D).

Functional enrichment analysis of proteins that decreased from the acute to the recovery phase by both PEA and DIA-MS platforms reflected a decline in proinflammatory cytokines (TNFSF14, CX3CL1, CCL11, CCL13, MMP13, and S100A9), acute phase response proteins including complement factors (HP, A2M, HGF, C4B, and KLKB1), ECM organization (PCOLCE, COL1A1, COL18A1, and KRT8), and coagulation pathways (F3, F7, F13A1, FGA, FGB, FGG, and SERPINE1) consistent with an attenuation of the initial host inflammatory response to SARS-CoV-2 infection (Figures 3E and 3F).

Enrichment analysis of proteins with increased expression from the acute to the recovery phase identified by DIA-MS was associated with T cell activity and phagocytosis (HLA-DRA, HLA-DRB1, ICAM1, GRB2, ITGB2, ATP6V1A, LAMP1, MRC1, NCF2, NCF4, SFTPD, SFTPA2, TFRC, and MARCO) (Figure 3F). Additionally, there was an increase in DEPs related to actin cytoskeleton organization and epithelial repair (FLNA, KRT5, KRT6A, LGALS3, VIM, CLIC4, F11R, MYADM, CAPG, FLNA, VASP, ARPC5, IQGAP2, TWF2, COTL1, and EVL), apoptosis (BAX, CASP1, CASP14, CD44, and PYCARD), cellular homeostasis (ACO1, ATP6V1A, BAX, ACE, HK3, TFRC, TGM2, CLIC4, and VDAC1), and ROS metabolism and detoxification (AKR1B1, G6PD, CES1, GRB2, ITGB2, VDAC1, and LTA4H).

Thus, the sequential changes in the lung proteome, 3 months post-COVID-19, show a decline in the robust acute phase protein expression with a decreased expression of inflammatory, injury, and procoagulant mediators and activation of the adaptive immune response, phagolysosome, and homeostasis pathways.

Lung matrisome protein expression is acutely associated with chest CT abnormalities and decreases over time after COVID-19

An improvement in radiographic abnormalities within the first 3 months post-infection was observed in patients presenting with acute chest CT abnormalities in cluster 2 phenotype (Figures 4A and 4B; Table S7). As such, we evaluated the association of the BAL matrisome with radiographic and pulmonary function changes to investigate biological correlates to clinical presentation.

Figure 4.

Lung matrisome expression decreases after COVID-19 infection and is associated with quantitative acute chest CT abnormalities but not associated with pulmonary function tests

(A) Quantitative chest CT lesion volumes (mL) in acute phase scans in cluster 1 and cluster 2 patients.

(B) Chest CT lesion volumes (mL) over time in cluster 2 patients.

(C) Graphical summary of PEA matrisome proteins differentially expressed in cluster 2 compared to cluster 1 during the acute phase.

(D) Correlation of top 3 matrisome protein concentrations with acute phase chest CT lesion volumes. Line plots of top 3 PEA matrisome protein concentrations from the acute to the recovery phase.

(E) PEA matrisome protein levels were compared to chest CT lesion volumes and lesion burdens in the acute phase and to chest CT lesion volume and pulmonary function tests (forced expiratory volume in 1 s [FEV1%], forced vital capacity [FVC%], total lung capacity [TLC%], and diffusion capacity for carbon monoxide [DLCO%]) in the recovery phase. Black outline indicates parameters with FDR 5%.

(F) Overlapping matrisome proteins that were upregulated in cluster 2 phenotype in PEA and DIA-MS platforms. Healthy controls are shown for reference comparison.

(G) Protein abundance of secretory leukocyte peptidase inhibitor (SLPI) and typical neutrophil markers longitudinally, by DIA-MS.

Statistical analysis: (A, F, and G) Mann-Whitney U test. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.005. Horizontal line represents median. (B) Wilcoxon-matched pairs signed rank test. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.005. (D) Spearman correlation with line of best fit ±95% confidence intervals, p values calculated to account for FDR < 0.05.

As previously highlighted, evaluation of acute phase cluster 2 compared to cluster 1 by PEA and DIA-MS revealed upregulation of many ECM proteins. Matrisome classification identified 190 upregulated proteins by the PEA platform, of which 48 were part of the core matrisome (FDR 5%, absFC >2)^23,24,25 (Figure 4C). Acute phase BAL expression of COL1A1, LTBP2, and VCAN had the highest correlation with chest CT lesion volume (mL) and lesion burden (%), with the latter two (LTBP2 and VCAN) showing statistical significance after age and gender adjustments (FDR 5%) (Figures 4D and 4E). Decreasing protein expression levels paralleled the radiographic resolution of infiltrates seen on chest CT (Figures 4B and 4D). No correlation was found between the recovery phase BAL matrisome proteins with subsequent quantified radiographic abnormalities and pulmonary function testing (forced expiratory volume in 1 s, forced vital capacity, TLC% [total lung capacity], and DLCO% [diffusion capacity for carbon monoxide]) (Figure 4E). Pulmonary function testing was not obtained during the acute phase.

Additionally, 78 matrisome proteins were identified in the acute phase BAL by DIA-MS with 6 DEPs upregulated in the cluster 2 phenotype compared to cluster 1 (TIMP1, COL6A3, MMP9, ITIH3, PCOLCE, and IGFBP7). All 6 of these proteins overlapped with the PEA results, and both platforms showed a downtrend in median values from acute to recovery phases (Figure 4F). Notably, there were 10 ECM proteins (ANXA5, LGALS1, LGALS3, MUC16, SFTPA2, SFTPD, SLPI, TGM2, FLG2, and S100A4) that significantly increased from the acute to the recovery phase, composed of ECM regulators, secreted factors, and ECM-associated mediators. One such matrisome protein, secretory leukocyte peptidase inhibitor (SLPI), which functions to mitigate chronic pulmonary epithelial damage by neutrophil elastase,^26,27 did not return to baseline levels even at convalescent phase follow-up and was found to be inversely proportional to the expression of typical neutrophil markers in BAL (Figure 4G).

These data highlight an association between ECM and fibrogenic mediators in the lung with the presence of acute lung abnormalities, with expression mirroring radiographic improvement over time. However, longitudinal activation of some repair pathways remains despite clinical and radiographic resolution.

Random forest machine learning algorithm suggests lung secreted phosphoprotein 1 levels may predict persistent lung abnormalities

While most patients in our cohort completely recovered from a radiographic and clinical standpoint, 4 patients had mild residual convalescent phase radiographic lesions considered to be a consequence of COVID-19. These patients were identified by screening for chest CT lesion burden greater than 1.5% on the convalescent phase chest CT, and all patients were in the cluster 2 phenotype. The median time for obtaining the convalescent CT was 272 days from symptom onset. The median abnormal CT lesion volume in these 4 patients was 174.78 mL with a median lesion burden of 4% of total lung volume. Pulmonary function tests (PFTs) at the time of the convalescent phase chest CT were variable with findings of restriction (n = 3, TLC < 80%), diffusion defects (n = 2, DLCO < 75%), and one patient with normal PFT (Figure 5A). All four of these patients were classified within the cluster 2 inflamed phenotype based on their acute phase BAL.

Figure 5.

SPP1 levels in acute phase BAL are predictive of the development of interstitial lung abnormalities

(A) Chest CT lesion quantification and pulmonary function tests of patients with post-COVID-19 residual lesions on chest CT at the convalescent phase.

(B) Variable importance plot (mean decrease in Gini index) showing acute phase BAL proteins associated with residual chest CT lesions, ranked from most likely to least. Top 30 proteins are shown.

(C) Median NPX comparing acute phase BAL proteins of patients with convalescent phase residual CT lesions versus no residual lesion using PEA.

(D) Acute phase SPP1 levels from BAL in patients with residual chest CT lesions at the convalescent phase compared to patients with normal convalescent CT and healthy controls.

(E) SPP1 levels in BAL of patients with residual chest CT lesions at the convalescent phase over time and compared to healthy controls. ILA, interstitial lung abnormalities.

Statistical analysis: (C) Wilcoxon rank-sum test. p values corrected to account for FDR < 0.05, Benjamini-Hochberg method were adjusted for top 10 proteins. (D and E) Mann-Whitney U test. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.005. Horizontal line represents median.

To determine if any biomarkers in the acute phase BAL could be predictive of the development of interstitial lung disease or fibrosis, random forest machine learning was used to identify the relative variable importance of 1,472 analytes in the PEA proteomic platform. Variable importance plots identified secreted phosphoprotein 1 (SPP1), a profibrotic macrophage cell-surface marker associated with SARS-CoV-2 infection and idiopathic pulmonary fibrosis,^28,29,30 in BAL as being the most predictive of long-term parenchymal scarring (Figure 5B). Wilcoxon rank-sum testing was also performed to compare patients with abnormal convalescent phase chest CT (n = 4) with normal scans (n = 22), and SPP1 remained the most significant biomarker (p = 0.00013) (FDR 5%) (Figures 5C and 5D). Recovery and convalescent phase BAL SPP1 in a subset of three and two patients, respectively, had decreased back to normal levels comparable to healthy controls (Figure 5E).

In summary, BAL levels of secreted SPP1 in the early stages of recovery may be predictive of the development of subsequent interstitial lung abnormalities after COVID-19 pneumonia.

Persistent overexpression of proteins related to cellular homeostasis and an altered host defense response are seen in the lungs 9 months after COVID-19 infection

To evaluate if the changes in the lung proteome completely resolve over longitudinal follow-up, convalescent phase BAL (median 278 days from symptom onset) was compared to healthy controls using all three proteomic platforms. None of the BAL samples from the convalescent phase had detectable SARS-CoV-2 virus by PCR.

The comparison of convalescent phase BAL samples (n = 16 patients) compared to healthy controls (n = 4 patients) using PEA did not reveal any significant differences. Analogous evaluation of all convalescent phase BAL (n = 13 patients) compared to healthy controls (n = 11 patients) using the targeted immunoassays revealed 22 DEPs. Two of these proteins, E-selectin and ferritin, had increased expression in the convalescent phase compared to control (Table S8). When using the DIA-MS, all convalescent phase BAL (n = 13 patients) compared to healthy control BAL (n = 6 patients) was notable for 235 DEPs (FDR 5%) (Figures 6A and 6B; Figure S3; Table S8).

Figure 6.

Convalescent phase lung proteome post-COVID-19 shows persistent activation of cellular homeostasis pathways and a depressed host defense response compared to uninfected persons

(A) Heatmap of 235 DEPs comparing the convalescent phase of COVID-19 to healthy control BAL using DIA-MS (FDR 5%).

(B) Volcano plot of DEPs. Horizontal dashed line denotes a cutoff of 0.05 for the FDR corrected p value after age and gender adjustments.

(C) IPA enrichment analysis of overexpressed proteins.

(D) Graphical summary of downregulated pathways comparing the convalescent phase of COVID-19 to healthy controls (IPA).

(E) Proteins related to phagolysosome activation, cellular repair, and complement cascades. Cluster 1 denoted by blue dots, cluster 2 by red dots. Patients enrolled during the recovery phase denoted by gray dots (no cluster).

(F) Enrichment analysis heatmap comparing overexpressed proteins in cluster 1 and cluster 2 phenotypes.

Statistical analysis: (B) p values are adjusted by the Benjamini-Hochberg method for an approximate control of FDR at 5% level. Proteins with >50% missing values were excluded. Missing values < 50% were imputed by minimum value/2. (E) Mann-Whitney U test. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.005. Horizontal line represents median. Data for Figures 6A–6C were generated using Spectronaut Quant 2.0 setting and data for Figure 6E by Spectronaut MaxLFQ settings (Table S8).

Enrichment analysis of 51 differentially abundant proteins by DIA-MS (FDR 5%, absFC>2) in both phenotypes was consistent with phagolysosome maturation (RAB5B, ATP6V1B2, ATP6V1E1, ATP6V1A, CYBB, and LAMP1), CLEAR signaling (HEXB, LAMP1, PSAP, and TPP1), epithelial-mesenchymal transition (VIM, AHNAK, and HNRNPK), inhibition of immune-mediated epithelial damage (SLPI), cellular response to oxidative stress via molecular chaperones (HSPB1, CCT3, CCT6A, PDIA6, and HSP90B1), DNA and unfolded protein repair (EIF2S3, CANX, CALR, and HNRNPK), and antiviral mechanisms (TRIM25 and MVP), all indicative of a sustained host response to the remote pulmonary insult and activation of the proteostasis network (Figure 6C). A significant increase in soluble HLA-DRA was also found in patients with COVID-19 in the convalescent phase compared to a control (absFC 2.38, adjusted q = 0.0052).

Enrichment analysis of 73 proteins with decreased expression in the convalescent phase compared to control by DIA-MS (FDR 5%, absFC>2) was integral to renin-angiotensin, complement, and kallikrein-kinin systems crosstalk (C2, C4A, C5, C6, C8G, C1R, CFB, CFH, CFI, KLKB1, AGT, and SERPINA4), coagulation system (F2, FGG, KLKB1, SERPINA1, and SERPINC1), and markers of inflammation (HP, HRG, ITIH2, ORM1, ORM2, TF, SERPINA3, CP, and AMBP) (Figure 6D). Sensitivity analysis using an additional validated peptide quantification approach (Spectronaut DIA-MS data analysis default MaxLFQ settings) resulted in similar DEPs and biological processes (Figure 6E; Table S8).

Differences between COVID-19 phenotypes were also evaluated by comparing convalescent phase BAL from cluster 2 (n = 5) and cluster 1 (n = 5) patients to healthy controls; 3 patients were not assigned to a cluster phenotype as they entered the study during the recovery phase. Functional enrichment analysis of cluster 2 compared to cluster 1 showed activation of the similar pathways, with more robust enrichment in cluster 2 (Figure 6F). Further analysis showed 28 overlapping proteins between PEA and significant proteins queried by DIA-MS in the convalescent phase compared to control; of these, 18 proteins showed similar trends to DIA-MS (Figure S4). The two proteins that were overexpressed in the targeted immunoassays, E-selectin and ferritin, showed similar trends in DIA-MS.

In summary, proteomic analysis of BAL obtained at 9–12 months after SARS-CoV-2 infection suggests the presence of persistent cellular repair, ongoing proteostasis, and depressed complement expression when compared to uninfected persons. Notably, the activation of homeostatic pathways is evident despite near-complete symptomatic and radiographic resolution, suggesting prolonged reparative processes following the initial infection.

Discussion

Longitudinal evaluation of patients after SARS-CoV-2 infection has provided detailed analysis of clinical symptomatology, non-invasive diagnostic testing, and immunophenotyping of blood and upper respiratory samples,^{12,31,32,33,34,35,36} with limited exploration of the lower respiratory tract in patients.^18,37,38 To the best of our knowledge, this is the first comprehensive study investigating the evolution of the lung proteome over the course of 9 months in patients who have recovered from mild to moderate COVID-19. We utilized three proteomic platforms to identify proteins contributing to lung processes that occur after SARS-CoV-2 infection and the subsequent reparative processes that result from this injury. We identified two distinct biological phenotypes in the initial post-COVID-19 period, but these differences in the proteomic signature became more homogeneous in both groups after 3 months. Despite this progressive resolution, we observed a persistent alteration in the host defense response proteins and continued cellular repair and proteostasis in the lung as late as 9 months after symptom onset.

The early phase of post-acute COVID-19 infection demonstrated clustering of the lung proteomic signature into uninflamed (cluster 1) and inflamed (cluster 2) phenotypes. This clustering of patients was nearly identical across all three proteomic platforms. Over 750 and 550 proteins were independently linked with the inflamed phenotype using DIA-MS and PEA, respectively, and correlated with peak illness severity, chest CT abnormalities, and vaccination status. Proteins significantly overexpressed in the inflamed phenotype included coagulation markers, epithelial repair mediators, as well as inflammatory cytokines and chemokines, several of which have been previously associated with illness severity, mortality, and persistent symptoms in COVID-19.^{11,15,39,40,41,42,43,44,45,46,47,48} While the uninflamed subgroup showed a less robust inflammatory and repair response, the protein signature was still different from uninfected healthy controls with relative activation of pathways related to the host defense and repair response. Previous studies have shown that even in asymptomatic infected persons with SARS-CoV-2, long-lasting T cell immunity and cross-reactivity to prior infection with seasonal coronaviruses result in antiviral mechanisms.^49,50,51,52 Although we demonstrated a correlation between a number of BAL and plasma biomarkers in the acute phase, these data are not sufficient to conclude that the two compartments are predictive of one another. The data are derived from one of the three queried proteomic platforms and represent a single time point. Furthermore, the time from symptom onset to sample collection varied among the patients.

In the inflamed subgroup, evolution of the proteome from the acute to the recovery phase revealed a decrease in inflammatory cytokines and proteinases and acute phase response proteins, reflective of a transition from the initial activation of innate immune response to a resolution phase. Additionally, decreased expression of proteins associated with complement, coagulation, and the kinin-kallikrein system was identified; the interaction of these three pathways is a hallmark of thrombo-inflammation, which has been uniquely characterized in COVID-19 pathogenesis.^{53,54,55,56,57,58} Downregulation of these pathways over time is reflective of the absence of severe end-organ damage after COVID-19 infection in our cohort.

Sequential assessment of proteins with increased expression from the acute to the recovery phase was reflective of multiple processes: antibody-mediated immune response, phagocytosis, actin cytoskeletal reorganization and epithelial repair, apoptosis, and ROS mitigation. All reflect the continuum of the host defense and repair response over time. ROS have been linked to neutrophil extracellular trap (NET) formation and associated cell death, leading to DNA damage.⁵⁹ Prior investigations have shown significantly elevated blood and lung tissue levels of NETs after COVID-19 contribute to the pathogenesis of immunothrombosis.^60,61,62 Glucose-6-phosphate dehydrogenase (G6PD), which was upregulated in the recovery phase in our patients, protects alveoli from the harmful effects of oxidative damage from excess ROS and NETosis. Patients with G6PD deficiency have shown worse outcomes after COVID-19 infection. These findings highlight key homeostatic mechanisms in our cohort that are lung protective and that continue to occur even 3 months from the initial infection.^63,64,65

In addition to mitigation of the initial inflammatory host response, restoration of the pulmonary architecture requires a coordinated interplay of mediators to repair epithelial damage. This process, however, necessitates strict regulation to prevent overactive remodeling that can result in pulmonary fibrosis. Prospective studies have demonstrated a significant improvement in COVID-19-related CT abnormalities on 6-month follow-up in patients who presented with mild to moderate disease.^66,67,68 These observations are reflective of our cohort, none of whom required invasive mechanical ventilation at peak illness. In line with this, we see an association between ECM mediators in the lung with the presence of acute lung abnormalities quantified on chest CT in the inflamed phenotype. Decreased sequential expression of core matrisome proteins mirrors the rapid improvement in radiographic abnormalities over time, as confirmed by the paucity of patients in our cohort who had significant residual interstitial lung disease. We also observed an increase in other ECM regulators, including SLPI, a potent inhibitor of neutrophil elastase that functions to protect the lung from ongoing inflammatory injury. Recent evaluation of autopsy SARS-CoV-2 lung at the time of autopsy showed decreased SLPI expression in smooth muscle glands and epithelium, suggestive of a virus-induced suppression of SLPI.²⁷ The chronically elevated SLPI levels in our patients shed light onto intrapulmonary homeostatic mechanisms required to prevent post-viral interstitial lung disease. Of interest, these processes are activated for several months after the initial viral infection, suggesting a delayed or prolonged recovery phase after even mild to moderate viral-induced inflammation.

In four patients who developed mild interstitial lung abnormalities in the convalescent phase, we found that acute phase BAL levels of SPP1 were highly associated with chronic post-COVID-19 interstitial lung disease. None of these patients had a documented past medical history of interstitial lung disease or fibrosis, and the pattern of residual lung injury was not consistent with usual interstitial pneumonia or nonspecific interstitial pneumonia, but rather of residual post-infectious abnormalities. SPP1 is a macrophage cell-surface marker linked to a reparative profibrotic subset of M2-like macrophages identified by single-cell sequencing of BAL cells after severe COVID-19^30,69,70,71. While none of the four patients had severe illness presentation or advanced downstream fibrosis, lung SPP1 levels were significantly higher in the acute phase but trended toward normalcy by recovery and convalescent phases. This suggests that the initial parenchymal insult and early dysregulation of repair processes are extensive enough to lead to end-organ dysfunction. It is also likely that there is continued alteration in other pro-fibrotic mediators, as shown by elevation of ECM-associated proteins in the recovery phase. Overall, our findings suggest that SPP1 levels in BAL early in the post-acute illness phase may serve as a potential marker for the subsequent risk of developing fibrosis. Larger prospective cohort studies are needed to confirm these observations.

While there was evidence of substantial proteomic differences between the inflamed and uninflamed phenotypes in the first 3 months post-symptom onset, the lung proteome became more uniform and less dynamic, as evidenced by the lack of DEPs comparing recovery to convalescent phases. Whether or not this proteomic stability reflected a return to a healthy pre-COVID-19 protein signature was of interest. Previous proteomic investigation of the lung revealed persistent alterations in epithelial damage and repair mediators (LDH, albumin, AREG, KRT19, and MMP3) 3–6 months after illness onset that correlates with airway T cell frequency in patients with chronic respiratory disease. However, in a subset of three patients, these changes resolve at 1 year.¹⁸ When comparing all patients in the convalescent phase to healthy controls, we found evidence of sustained host response to the remote viral insult, including phagolysosome activation, DNA and unfolding protein repair, iron homeostasis, and restoration of the epithelium. The elevation of several proteins associated with phagosome maturation and lysosomal fusion is indicative of ongoing degradation and clearance of internalized microorganismal remnants and debris.⁷² Though there was an absence of detectable virus by reverse-transcription PCR in convalescent phase BAL, this does not exclude a contribution of tissue-associated virus or viral remnants leading to prolonged antigen presentation.^38,73 The molecular processes that result in the phagolysosome complex are necessary for cellular homeostasis and prevention of dysregulated cell death.^74,75,76 While the timeline of these processes in vivo has not been well described, in vitro studies show that various immune signals, including IFN-γ, can delay phagosome maturation to enhance microbial killing or antigen presentation.^77,78,79 Related to this, we observed an overexpression of soluble HLA-DRA, which has previously been found to be elevated in the blood of convalescent patients with COVID-19 compared to a control.⁸⁰ While the authors demonstrated a correlation between soluble human leukocyte antigen and anti-SARS-CoV-2-specific T cell activation, other reports established T cell inhibition and anergy in the absence of co-stimulation.^81,82,83 As such, whether these processes reflect ongoing cellular homeostasis or continued antigen presentation with T cell activation is unknown.

The findings of low levels of complement and acute phase mediators in convalescent phase samples relative to control were unexpected. Of note, this decreased expression was present in all three phases for many proteins, including C2. The sustained complement and acute phase mediator deficiency may be due to several proposed mechanisms including continued negative feedback loop of proteins implicated in the acute inflammatory response, sequestration of complement fragments into phagosomes after opsonization, or reduced secretion of proteins secondary to an incompletely restored epithelium. Both complement and acute phase proteins are produced and secreted by extrahepatic cells including airway epithelium, macrophages, and endothelial cells.^84,85,86 The concurrent upregulation of markers of epithelial-mesenchymal transition in convalescence supports the concept of dysregulated extrahepatic secretion of these acute phase mediators.

Overall, with the resolution of clinical signs and symptoms, the lung proteome after SARS-CoV-2 infection still shows evidence of ongoing repair and resolution of the host defense response despite differences in the initial host clinical presentation. Given the absence of protracted pulmonary dysfunction in our patients, these findings reflect a timeline of biological processes occurring in a coordinated post-COVID-19 response. However, this persistent alteration found in all post-COVID-19 phenotypes suggests that there may be more contribution of pathogen-specific factors rather than host risk factors. Whether or not this has later clinical implications is unknown. Similarly, a detailed proteomic analysis of other viral pulmonary infections after recovery is needed.

Limitations of the study

BAL has inherent limitations regarding protein concentrations that may vary among patient samples. Given that most proteomic platforms are standardized against plasma samples, there are inherent differences in sensitivity. Thus, studies describe relative abundance of biomarkers. However, using different proteomic platforms in our cohort, we found similar hierarchical clustering and biological themes among the groups using these different methods. These data suggest that protein concentration did not detract from the current observations. Our control patients were generally younger, and we adjusted all our results for age and gender. We did not include severely ill patients receiving mechanical ventilation in our cohort. It is likely that the magnitude of their initial host responses would be similar but of greater intensity. We did not adjust for COVID-19 therapeutics in the smaller fraction of patients who received them in our study. We see the same processes occurring in both phenotypic clusters at the convalescent phases, suggesting these pathways are occurring independent of prior therapeutic administration. Finally, we did not have access to post-acute BAL samples from other respiratory viruses (e.g., influenza) to compare our findings to. This highlights the need for further investigation into the long-term pulmonary sequelae with other respiratory viruses to determine if our findings are unique to SARS-CoV-2.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Chemicals, peptides, and recombinant proteins
SDS, 10% Solution, RNase-free	Invitrogen	AM9822
Trypsin/Lys-C Mix, Mass Spec Grade	Promega	V5071
Critical commercial assays
V-PLEX Human Biomarker 54-Plex Kits	Meso Scale Discovery	K15248D
Human Luminex® Discovery assays	R&D Systems	LXSAHM
Deposited data
Proximal extension assay proteomic dataset	This manuscript	Figshare: https://doi.org/10.25444/nhlbi.25666668
Targeted Immunoassays proteomic dataset	This manuscript	Figshare: https://doi.org/10.25444/nhlbi.25666668
Data independent acquisition mass spectrometry proteomic dataset	This manuscript	MassIVE Repository: MSV000094898 Figshare: https://doi.org/10.25444/nhlbi.25666668
Data independent acquisition mass spectrometry proteomic data analyses	This manuscript	MassIVE Repository: MSV000094898 Figshare: https://doi.org/10.25444/nhlbi.25666701
Software and algorithms
RStudio	https://posit.co	v2023.12.1
R	https://www.r-project.org	v4.3.1
SAS	https://www.sas.com/en_us/home.html	v9.4
JMP	https://www.jmp.com/en_us/home.html	v16.1.0
Spectronaut	Biognosys: https://biognosys.com/software/spectronaut/	v15

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Shreya M. Kanth (shreya.kanth@nih.gov).

Materials availability

This study did not generate new unique reagents.

Data and code availability

The proximal extension assay data and targeted immunoassay data have been deposited to FigShare (https://doi.org/10.25444/nhlbi.25666668). The data independent acquisition mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the MassIVE partner repository and are publicly available as of the date of publication (MassIVE: ftp://MSV000094898@ucsd.edu). The paper does not report original code. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Experimental model and subject details

Patient cohort and clinical data collection

The COVID ARC-19 (Cardiopulmonary Inflammation and Multi-System Imaging During the Clinical Course of COVID-19 Infection in Asymptomatic and Symptomatic Persons, NCT04401449) is a natural history study designed to comprehensively characterize the clinical and biologic effects of SARS-CoV-2 infection longitudinally. This study prospectively enrolled patients 18 years or older to the National Institutes of Health Clinical Center with COVID-19, confirmed by positive polymerase chain reaction. The patients described in the current study were enrolled from May 21, 2020 through October 20th, 2021. Participants (n = 45) underwent sequential imaging, pulmonary function testing, and sampling of blood and bronchoalveolar lavage during acute (1–40 days from symptom onset), recovery (41 days-3 months from symptom onset) and convalescent (>3 months–12 months from symptom onset) phases of COVID-19 illness. Clinical data including age, gender, race and past medical history were collected for all patients. Demographics are summarized in Table S1, using medians with interquartile ranges and percent of total, where appropriate. For all participants in this cohort, gender identity aligned with sexual orientation assigned at birth. Clinical severity of illness at the time of initial symptom onset was defined by the NIAID ordinal scale.⁸⁷ All participants provided informed consent in written or electronic forms. Data was managed using the REDCap secure data collection tool. Blood and bronchoalveolar lavage samples were obtained from archived samples from healthy volunteers enrolled in NCT02392442 (Effects of Bronchial Segmental Endotoxin Instillation in Humans). The study was approved by the National Institute of Health institutional review board. To ensure confidentiality of individual subjects, publishable patient-level demographic information is limited as per the protocol of the National Institutes of Health. Patient-specific data acquired for all analyses are referenced throughout the manuscript in main figures and supplemental tables.

Method details

Quantification of imaging abnormalities

A lung segmentation algorithm that was developed to identify and localize whole lung and lung lesion regions involved with SARS-CoV-2 infection was applied to chest imaging of the enrolled patients.⁸⁸ Output data included total lung volume (mL), total lesion volume (mL) and lesion burden (lung volume/lesion volume, %). Output values from the soft-tissue series were used for data analysis.

Plasma sample collection and processing

Blood samples were collected in EDTA tubes and processed within 2 h after the blood draw. The plasma was processed in a BSL2 lab with level 3 practices in place. Whole blood was centrifuged at 1200 g for 10 min at 4°C. After centrifugation, plasma was pipetted into 1 mL Eppendorf tubes and stored at −80°C.

Bronchoalveolar lavage sample collection and processing

Patients undergoing bronchoscopy had a pre-procedure high resolution chest computed tomography (CT). Flexible fiberoptic bronchoscopy was performed in a monitored procedure unit using local anesthesia and conscious sedation. The bronchoscope was introduced orally and passed to lung sub-segment of interest based and lavage was performed with instillation of 30 mL aliquots of saline, with a maximum instillation of 180mL. BAL samples were collected in sterile traps and processed within 2 h post procedure in a BSL2 lab with level 3 practices. BAL was tested by reverse transcriptase–polymerase chain reaction (RT-PCR) for SARS-CoV-2. BAL was centrifuged at 1200 g for 10 min at 4°C. After centrifugation, the supernatant was pipetted into 1 mL tubes and stored at −80°C. Viral inactivation of the BAL for PEA and immunoassays was achieved by exposure to a cobalt-60 source as previously described.⁸⁹ Samples for mass spectrometry were inactivated by preparing them in final volume of 5% sodium dodecyl sulfate in preparation for denaturing and digestion.⁹⁰ BAL from historical pre-COVID-19 controls (NCT02392442) was stored at −80°C.

Proximal extension assay (PEA) proteomics assay

BAL samples from patients with COVID-19 (n = 31) and a single time point from healthy controls (n = 16) were quantified using a proximal extension assay (Olink Proteomics, Uppsala, Sweden), allowing for simultaneous analysis of 1536 total protein biomarkers across 4 panels (Olink Explore 1536, https://olink.com/kit-descriptions/). The resulting output is the Normalized Protein Expression (NPX) value in log2 scale, which is derived from the inverse of the cycle threshold value and normalized using standard the manufacturer’s instructions. The PEA data for each deidentified patient is uploaded to FigShare (FigShare: https://doi.org/10.25444/nhlbi.25666668).

Targeted immunoassays proteomics assays

Soluble biomarkers were analyzed on paired BAL and EDTA plasma obtained from COVID-19 patients (n = 32 BAL, n = 24 plasma) and pre-COVID19-era healthy subjects (n = 12 BAL, n = 7 plasma) using multiplex platforms. V-PLEX Human Biomarker 54-Plex Kits (Meso Scale Discovery, Rockville, MD [IL-1α, IL-1β, IL-2, IL-4, IL-5, IL-6, IL-7, IL-8, IL-10, IL-12p70, IL-12p40, IL-13,IL-15, IL-16, IL-17, IFN-γ, TNF-α, TNF-β, GM-CSF, VEGF, CCL-11/Eotaxin-1,CCL26/Eotaxin-3, CXCL10/IP-10, MCP-1/CCL2, MCP-4/CCL13, CCL22/MDC, MIP-1α/CCL3, MIP-1β/CCL4, CCL17/TARC, bFGF, Flt-1, PIGF, Tie2, VEGF-A, VEGF-C, VEGF-D, IL17 A/F, IL-17, IL-17C, IL-17D, IL-1RA, IL-3, IL-9, TSLP, IL-21, IL-22, IL-23, IL-27, IL-31, and MIP-3α]) were analyzed on a QuickPlex SQ 120 reader (Meso Scale Discovery, Rockville, MD) according to the manufacturer’s instructions as previously described.¹¹ Customized, bead-based, multiplex Human Luminex Discovery assays (R&D Systems, Minneapolis, MN [CRP, SAA, sICAM-1, sVCAM-1, IL-33, S100A9, sTNFRSF1A/sTNF RI, sTNFRSF1B/sTNF RII I, sL-selectin/sCD62L, S100A8, SCF/c kit ligand, TNFSF14/LIGHT, sVEGFR1/Flt-1, sCD31/sPECAM, CX3CL1/Fractalkine, IL-23, IL-18/IL-1F4, IL-3, sST2/sIL-33R, sCD163, IL-1RA, soluble FAS ligand/TNFSF6, RAGE, sCD25/sIL-2Rα, M-CSF, REG3A, G-CSF, sVCAM-1/sCD106, sCD40 ligand/TNFSF5, vWF-A2, uPAR, Endocam/ESM-1, Angiopoietin-2, D-dimer, SP-D, TREM-1, HGF, sL-selectin/sCD62L, CA15/MUC, lactoferrin, MPO, lipocalin-2/NGAL, LBP and ferritin]) were analyzed on a Bio-Plex 3D instrumentation (Bio-Rad, Hercules, CA) according to the manufacturers specifications for standards and dilutions. Standard curves were analyzed using nonlinear curve fitting and unknowns were calculated based on the derived equation. Samples that exceeded the highest standards were reanalyzed after dilution until the values fell within the range of the known standards. Two control plasma samples and a control sample spiked with a known quantity of each analyte were analyzed on each plate to assess the inter-plate variation and to determine the effect of the biological matrix on the measurement of each analyte. For most analytes, the control samples had <25% variation from plate to plate, and the recoveries were generally >70%. The targeted immunoassay data for each deidentified patient in BAL and plasma are uploaded to FigShare (FigShare: https://doi.org/10.25444/nhlbi.25666668).

Data independent acquisition mass spectrometry (DIA-MS) proteomics assay

Protein concentration in each sample was measured with BCA assay (Fisher Scientific). Equal amounts of proteins were processed for quantitative proteomics, by using a procedure published previously.⁹¹ In brief, proteins were treated with dithiothreitol and iodoacetamide, loaded onto an S-Trap column (ProtiFi, LLC), and then digested with sequencing-grade Lys-C/trypsin (Promega) by incubation at 37°C overnight. Resulting peptides were analyzed by a nanoAcquity UPLC system (Waters) coupled with Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher), by using similar settings as described previously.⁹¹ Peptides captured onto the C18 Trap column (Waters Acquity UPLC M-Class Trap, Symmetry C18, 100 Å, 5 μm, 180 μm × 20 mm) were separated with an analytical column (Waters Acquity UPLC M-Class, peptide BEH C18 column, 300 Å, 1.7 μm, 75 μm × 150 mm) with a flow rate of 400 nL/min at 40°C. A 135-min gradient of buffer A (2% ACN, 0.1% formic acid) and buffer B (0.1% formic acid in ACN) was used for separation: 1% buffer B at 0 min, 5% buffer B at 1 min, 22% buffer B at 90 min, 50% buffer B at 100min, 98% buffer B at 120 min, 98% buffer B at 130 min, 1% buffer B at 130.1 min, and 1% buffer B at 135 min. Data files were acquired on the Orbitap Fusion Lumos mass spectrometer in DIA mode. Mass spectra were recorded with Xcalibur 4.0, by using parameters described previously.⁹¹ Analysis of DIA raw files was done by using Spectronaut software (Biognosys) v15) with a hybrid library. Both MaxLFQ and Quant2.0 approaches were used to quantify proteins with default settings in Spectronaut. In brief, dynamic retention time prediction with local regression calibration was selected. Interference correction on MS and MS2 level was enabled. The Qvalue Cutoff was set to 1% at peptide precursor and protein level using scrambled decoy generation and dynamic size at 0.1 fraction of library size. MS2-based quantification by area was used, enabling local cross-run normalization. The DIA-MS data for each analysis is uploaded to FigShare (FigShare: https://doi.org/10.25444/nhlbi.25666701). Partial data (.sne files) are available for download on the MassIVE Proteome Exchange (MassIVE Repository: ftp://MSV000094898@massive.ucsd.edu).

Quantification and statistical analysis

Data analysis and visualization

All statistical analyses were performed using the R statistical Software, version 4.3.1 (R Foundation for Statistical Computing, Vienna, Austria). Data were summarized by median (interquartile range) for continuous variables and frequency (percentage) for categorical variables. Mann-Whitney U test was used to compare between different patient subgroups and Wilcoxon signed-rank test was used to compare differences in the paired data. To control for multiple testing, the p-values for the individual biomarkers were adjusted using the Benjamini-Hochberg false discovery rate procedure. In each proteomic platform, principal component analysis (PCA) and hierarchical clustering analysis using the Ward’s method were applied to identify clusters of patients that were similar to one another based on their biomarker values. Heatmaps generated by the R statistical Software v 4.3.1 were used to display and compare the biomarker levels between different patient clusters and pre-defined acute, recovery, and convalescent phases, where each biomarker was standardized by first subtracting its mean and then divided by its standard deviation (sd). Random forest machine learning method was used to identify the biomarkers in the acute phase BAL that were most predictive of persistent lung abnormalities and provide variable importance ranking to eliminate biomarkers with little predictive values.

PEA and targeted immunoassays data analysis

In analysis of biomarkers from the PEA and targeted immunoassays platform, the analysis of covariance (ANCOVA) model was used to compare the mean levels of biomarkers between patient clusters, adjusted for patients’ age and gender. For each protein in the PEA, a linear mixed-effects model was used to identify differences in biomarker values over phases between groups. Additionally, a linear fixed-effects model analysis was conducted to discover any biomarker values across all phases. The biomarker values in targeted immunoassays platform were transformed with the rank-based inverse normal transformation in the ANCOVA model.

DIA-MS data analysis

For DIA-MS analysis, the high quality MS1 and MS2 signals in the DIA profiles were jointly analyzed using an ANCOVA model. First, proteins with >50% missing values were excluded and any protein with missing values ≤ 50% were imputed using the half of the minimum value among the non-missing values for the protein. Second, the MS1 and M2 intensities of each biomarker were normalized by dividing by their mean intensities across all the subjects. Then the log₂ of the normalized intensities was treated as the response variable to assess the difference between the patient subgroups, adjusted for the signal source (MS1 or MS2), age and gender.

Pathway analysis

Enrichment analysis of differentially expressed proteins was performed using Ingenuity Pathway Analysis (Qiagen) and Metascape.⁹² The Z score was used to predict activation or inhibition of a biological pathway.

Published: July 8, 2024