Multiplex protein profiling of bronchial aspirates reveals disease-, mortality- and respiratory sequelae-associated signatures in critically ill patients with ARDS secondary to SARS-CoV-2 infection

Abstract

Introduction: Bronchial aspirates (BAS) obtained during invasive mechanical ventilation (IMV) constitutes a useful tool for molecular phenotyping and decision making.

Aim: To identify the proteomic determinants associated with disease pathogenesis, all-cause mortality and respiratory sequelae in BAS samples from critically ill patients with SARS-CoV-2-induced ARDS.

Methods: Multicenter study including 74 critically ill patients with COVID-19 and non-COVID-19 ARDS. BAS were obtained by bronchoaspiration after IMV initiation. Three hundred sixty-four proteins were quantified using proximity extension assay (PEA) technology. Random forest models were used to assess predictor importance.

Results: After adjusting for confounding factors, CST5, NADK, SRPK2 and TGF-α were differentially detected in COVID-19 and non-COVID-19 patients. In random forest models for COVID-19, CST5, DPP7, NADK, KYAT1 and TYMP showed the highest variable importance. In COVID-19 patients, reduced levels of ENTPD2 and PTN were observed in nonsurvivors of ICU stay, even after adjustment. AGR2, NQO2, IL-1α, OSM and TRAIL showed the strongest associations with in-ICU mortality and were used to construct a protein-based prediction model. Kaplan-Meier curves revealed a clear separation in mortality risk between subgroups of PTN, ENTPD2 and the prediction model. Cox regression models supported these findings. In survivors, the levels of FCRL1, NTF4 and THOP1 in BAS samples obtained during the ICU stay correlated with lung function (i.e., D_LCO levels) 3 months after hospital discharge. Similarly, Flt3L and THOP1 levels were correlated with radiological features (i.e., TSS). These proteins are expressed in immune and nonimmune lung cells. Poor host response to viral infectivity and an inappropriate reparative mechanism seem to be linked with the pathogenesis of the disease and fatal outcomes, respectively.

Conclusion: BAS proteomics identified novel factors associated with the pathology of SARS-CoV-2-induced ARDS and its adverse outcomes. BAS-based protein testing emerges as a novel tool for risk assessment in the ICU.

Keywords: COVID-19; ICU – intensive care unit; acute respiratory distress syndrome; bronchial aspirate; proteomics.

Figure 1

Study flowchart. First, bronchial aspirate (BAS) proteins associated with SARS-CoV-2 infection were evaluated in the whole study population, consisting of 14 critically ill non-COVID-19 and 60 COVID-19 patients, both groups assisted by IMV during their ICU stays (April-November 2020). For COVID-19 patients, BAS predictors of all-cause in-ICU mortality were analyzed in 17 nonsurvivors and 43 survivors. Then, the association between BAS proteins and functional and structural pulmonary sequelae was explored in 23 survivors of ICU stay who were subjected to a complete pulmonary evaluation 3 months after hospital discharge. Twenty survivors were unreachable or decided not to participate in follow-up (n = 17), were referred to another department (n = 2) or did not complete the pulmonary evaluation (n = 1).

Figure 2

Proteomic bioprofiles in bronchial aspirates from COVID-19 and non-COVID-19 ARDS patients. (A) Volcano plot showing the p value versus the fold change. Each point represents a detected protein. Blue dots represent the detected proteins that showed significant differences considering a p value<0.05. The association was adjusted by a propensity score (PS) compound by age, sex, obesity, chronic pulmonary disease, use of antibiotics, use of tocilizumab and use of corticoids. (B) Boxplot including bronchial aspirate proteins that showed differences between study groups. The adjusted fold change is displayed, and the significance level for each comparison is described by the p value. (C) Principal component analysis. Each point represents a patient. (D) Variable importance plot displaying the most relevant proteins according to their contribution to the random forest model. (E) Cell enrichment analysis based on single-cell RNA-seq data from Genotype-Tissue Expression (GTex). Each row represents a protein, and each column represents a cell type. The color of the point shows the expression levels in detected cells, and the size of the point indicates the percentage of cells in which the expression was detected.

Figure 3

Proteomic bioprofiles of bronchial aspirates from survivors and nonsurvivors of ARDS secondary to SARS-CoV-2 infection. (A) Volcano plot showing the p value versus the fold change. Each point represents a detected protein. Blue dots represent the detected proteins that showed significant differences considering a p value<0.05. The association was adjusted by age. (B) Boxplot including bronchial aspirate proteins that showed differences between study groups. The adjusted fold change is displayed, and the significance level for each comparison is described by the p value. (C) Variable importance plot displaying the most relevant proteins according to their contribution to the random forest model. (D) Kaplan-Meier estimations for the individual proteins and BAS protein-based prediction model. Log-rank p values are displayed. (E) Cell enrichment analysis based on single-cell RNA-seq data from Genotype-Tissue Expression (GTex). Each row represents a protein, and each column represents a cell type. The color of the point shows the expression levels in detected cells, and the size of the point indicates the percentage of cells in which the expression was detected.

Figure 4

Association between the proteomic bioprofile in bronchial aspirates with diffusion capacity and structural features in survivors of ARDS secondary to SARS-CoV-2 infection. (A) Volcano plot showing the p value versus the rho coefficient for each detected protein after comparison of D_LCO levels. Each dot represents a protein. Blue dots represent proteins that showed a biologically and clinically relevant correlation with D_LCO levels; (B) GAM modeling for D_LCO and the levels of each of the selected proteins; (C) Volcano plot showing the p value versus the rho coefficient for each detected protein after comparison of TSS levels. Each dot represents a protein. Blue dots represent proteins that showed a biologically and clinically relevant correlation with TSS levels; (D) GAM modeling for TSS and the levels of each of the selected proteins. (E) Cell enrichment analysis based on single-cell RNA-seq data from Genotype-Tissue Expression (GTex). Each row represents a protein, and each column represents a cell type. The color of the point shows the expression levels in detected cells, and the size of the point indicates the percentage of cells in which the expression was detected. Edf: estimated degrees of freedom.