Dysregulated proteasome activity and steroid hormone biosynthesis are associated with mortality among patients with acute COVID-19

Abstract

The persistence of coronavirus disease 2019 (COVID-19)-related hospitalization severely threatens medical systems worldwide and has increased the need for reliable detection of acute status and prediction of mortality. We applied a systems biology approach to discover acute-stage biomarkers that could predict mortality. A total 247 plasma samples were collected from 103 COVID-19 (52 surviving COVID-19 patients and 51 COVID-19 patients with mortality), 51 patients with other infectious diseases (IDCs) and 41 healthy controls (HCs). Paired plasma samples were obtained from survival COVID-19 patients within 1 day after hospital admission and 1-3 days before discharge. There were clear differences between COVID-19 patients and controls, as well as substantial differences between the acute and recovery phases of COVID-19. Samples from patients in the acute phase showed suppressed immunity and decreased steroid hormone biosynthesis, as well as elevated inflammation and proteasome activation. These findings were validated by enzyme-linked immunosorbent assays and metabolomic analyses in a larger cohort. Moreover, excessive proteasome activity was a prominent signature in the acute phase among patients with mortality, indicating that it may be a key cause of poor prognosis. Based on these features, we constructed a machine learning panel, including four proteins [C-reactive protein (CRP), proteasome subunit alpha type (PSMA)1, PSMA7, and proteasome subunit beta type (PSMB)1)] and one metabolite (urocortisone), to predict mortality among COVID-19 patients (area under the receiver operating characteristic curve: 0.976) on the first day of hospitalization. Our systematic analysis provides a novel method for the early prediction of mortality in hospitalized COVID-19 patients.

Keywords: Acute phase; COVID-19; Mortality; Omics; Prediction.

Fig. 1

Study Overview. (A) Overview of assay modalities and validation methods. (B) Summary of COVID-19 COVID-19-A patients (n = 52) and COVID-19-M patients (n = 51). The y-axis displays patient identification numbers; the x-axis shows days since disease onset

Fig. 2

Plasma Proteome Analyses Reveal the Landscape of Host Responses in Patients with Acute COVID-19. (A) PLS-DA score plots for COVID-19-A, COVID-19-R, IDC, and HC groups. (B) Venn diagram of the numbers of DEPs among COVID-19-A, COVID-19-R, IDC, and HC groups. (C) Heatmap of 262 DEPs clustered using Mfuzz into five discrete significant clusters. (D) GO-BP enrichment analysis of all DEPs in each cluster, showing the top 5 GO terms. Green box highlights suppressed immunity in clusters 2 and 5. Blue box highlights enhanced inflammation in cluster 3. (E) Heatmap showing expression levels of DEPs related to suppressed immunity. Correlation analysis of immunity-associated proteins and clinical indexes. (F) KEGG terms for all DEPs in each cluster, showing the top 5 GO terms. Red box highlights metabolic suppression in cluster 4. (G) Heatmap showing expression levels of DEPs related to enhanced inflammation. Correlation analysis of inflammation-related proteins and clinical indexes. (H) Expression levels of altered proteasome subunits across the four groups. Statistical significance was determined by one-way ANOVA and Tukey’s HSD. *P < 0.05; **P < 0.01; ***P < 0.001. (I) GSEA to assess the enrichment of acute phase and adaptive immunity proteins during the acute phase of disease in COVID-19-A patients, compared with HCs

Fig. 3

Plasma Metabolome Analyses Reveal Suppressed Steroid Hormone Biosynthesis in Patients with Acute COVID-19. (A) Venn diagram of DEMs among COVID-19-A, COVID-19-R, IDC, and HC groups. (B) PLS-DA score plots for COVID-19-A, COVID-19-R, IDC, and HC groups. (C) Cluster of DEMs. (D) KEGG terms enriched in clusters 1 and 4. (E) Many intermediates in the steroid hormone biosynthesis pathway were significantly decreased. Decreased metabolites are labeled in purple. Statistical significance was determined by one-way ANOVA and Tukey’s HSD. *P < 0.05; **P < 0.01; ***P < 0.001

Fig. 4

Validation of Typical Features Related to Acute COVID-19. (A) Validation of DEPs related to enhanced inflammation, suppressed immunity, and proteasomal activation by ELISA in the training and test cohorts, respectively. (B) KEGG terms for DEMs among patients in the test cohort. (C) Validation of DEMs related to steroid hormone biosynthesis. Statistical significance was determined by one-way ANOVA and Tukey’s HSD. *P < 0.05; **P < 0.01; ***P < 0.001

Fig. 5

Proteomic Features of COVID-19-M Patients in the Acute Phase: Proteasomal Activation. (A) PLS-DA score plots for COVID-19-A, COVID-19-M, and HC groups. (B) Venn diagram of the number of DEPs among COVID-19-A, COVID-19-M, and HC groups. (C) Heatmap of 367 DEPs clustered using Mfuzz into four discrete significant clusters. (D) GO-BP enrichment analysis of all DEPs in cluster 1, clusters 2 and 4, and cluster 3, respectively. The top 5 GO terms are shown. (E) KEGG analysis of all DEPs in cluster 1, clusters 2 and 4, and cluster 3, respectively. The top 5 GO terms are shown. (F) Expression levels of proteasome subunits among COVID-19-A, COVID-19-M, and HC groups. Statistical significance was determined by one-way ANOVA and Tukey’s HSD. *P < 0.05; **P < 0.01; ***P < 0.001. (G) GSEA to assess the enrichment of proteasome signatures during the acute phase of disease in COVID-19-M patients, compared with COVID-19-A patients. ES, enrichment score; P-values were calculated via permutation test. (H) Correlation analysis of proteasome-associated proteins and clinical indexes. Red and blue numbers represent positive and negative correlations, respectively. (I) Correlation analysis of proteasome-associated proteins and Lac level in COVID-19 patients. *correlation P < 0.05. **correlation P < 0.01

Fig. 6

Metabolomic Features of COVID-19-M Patients in the Acute Phase: Suppressed Steroid Hormone Biosynthesis. (A) Venn diagram of the number of DEMs among COVID-19-A, COVID-19-M, and HC groups. (B) PLS-DA score plots for COVID-19-A, COVID-19-M, and HC groups. (C) Hierarchical clustering illustrating four DEP patterns across the three groups. (D) KEGG terms enriched in decreased clusters (1 and 4). (E) Expression of DEMs in the steroid hormone biosynthesis pathway. Statistical significance was determined by one-way ANOVA and Tukey’s HSD. *P < 0.05; **P < 0.01; ***P < 0.001

Fig. 7

Identification and Validation of Potential Biomarkers for Prediction of Mortality Risk in COVID-19 Patients. (A) Workflow for predictive marker selection. (B) ROC curve illustrating the performance of classifiers based on the combination panel. The model was trained with 30 samples and evaluated by patient-based five-fold cross-validation. (C) Biomarker panel confusion matrix among different plasma samples. (D) AUC values for five biomarkers and the combined panel in distinguishing COVID-19-M patients from COVID-19-A patients and HCs in the validation cohort. The model was tested with 114 samples collected from COVID-19 patients and HCs, then evaluated by patient-based five-fold cross-validation. (E) Kaplan–Meier survival curves were established according to mortality risk score; optimal cutoff values were derived from X-tile (all P < 0.0001, log-rank test). Patients were divided into two groups based on the median expression levels of PSMA1, PSMA1, PSMA7, and PSMB1. P-values were calculated by two-tailed log-rank tests. (F) AUC values for clinical indexes in distinguishing COVID-19-M patients from COVID-19-A patients and HCs. The model was trained and tested using 144 samples collected from both training and test cohorts, then evaluated by patient-based five-fold cross-validation