Serological Responses to Human Virome Define Clinical Outcomes of Italian Patients Infected with SARS-CoV-2

Abstract

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is responsible for the pandemic respiratory infectious disease COVID-19. However, clinical manifestations and outcomes differ significantly among COVID-19 patients, ranging from asymptomatic to extremely severe, and it remains unclear what drives these disparities. Here, we studied 159 hospitalized Italian patients with pneumonia from the NIAID-NCI COVID-19 Consortium using a phage-display method to characterize circulating antibodies binding to 93,904 viral peptides encoded by 1,276 strains of human viruses. SARS-CoV-2 infection was associated with a marked increase in individual's immune memory antibody repertoires linked to trajectories of disease severity from the longitudinal analysis also including anti-spike protein antibodies. By applying a machine-learning-based strategy, we developed a viral exposure signature predictive of COVID-19-related disease severity linked to patient survival. These results provide a basis for understanding the roles of memory B-cell repertoires in COVID-19-related symptoms as well as a predictive tool for monitoring its clinical severity.

Figure 1.

Overview of viral exposure of the Brescia cohort. (A) Workflow of VirScan. (B) The total number of unique epitopes in non-COVID and COVID-19 cases. (C) Prevalence of viruses of 156 COVID-19 cases (left), the number of unique epitopes of moderate, severe and critical groups of patients (middle 3 panels) as well as the composition of prevalence of the three groups of patients (right). (D) Antibody reactivity of all the epitopes presented in at least two cases. Each row represents an epitope while each column stands for one CODIV-19 case. −log10 (p-value) was used to measure the enrichment of peptide. (E) The total enrichment of all epitopes in the moderate, severe, and critical groups. Log transformation was applied. In violin plots, boxes span the interquartile range; lines within boxes represent the median; the width of violin plots indicates the kernel density of values.

Figure 2.

SARS-CoV epitope reactivity in moderate, severe, and critical cases. (A) Total reactivity of all SARS-CoV epitopes in non-COVID and COVID-19 cases. Log transformation was applied. (B) Antibody reactivity of 85 SARS-CoV epitopes. Each row represents the significant peptide tiling corresponding to spike protein (1–1255) and nucleocapsid protein (1–422). The color intensity of each cell corresponds to the scaled −log10(p value) measure of significance of enrichment for a peptide in a sample. (C) Organization of SARS-CoV-2 genome encoding various viral proteins. (D) B-cell epitope prediction score for spike and nucleocapsid based on the Immune Epitope Database and Analysis Resource (IEDB). (E-F) Sequence alignment of reactive peptides corresponding to spike (E) and nucleocapsid protein (F) of SARS-CoV and SARS-CoV-2. Only peptide sequences in the phage library are shown. Residues with perfect match are capitalized. Predicted epitopes by IEDB are highlighted. (G) Normalized EBS of spike (785–840, left) and nucleocapsid (141–196 right) proteins in moderate, severe and critical patients. (H) Normalized EBS of spike (785–840, left) and nucleocapsid (141–196 right) proteins in patients at different time points. In violin plots, boxes span the interquartile range; lines within boxes represent the median; the width of violin plots indicates the kernel density of values.

Figure 3.

Longitudinal progression of the normalized EBS across individuals. (A) Individual trajectories over time for patients grouped by disease severity (gray lines), which were averaged (solid blue line) and fitted by linear regression (dashed blue line; slope and standard error shown in the legend). Baseline refers to the first sample obtained after admission to the hospital. (B) Normalized EBS in SARS-CoV spike, nucleocapsid, and all SARS-CoV reactive proteins compared against the normalized EBS across all VirScan epitopes. (C) Heatmap showing the longitudinal progression of individual patients integrated across all three patient groups by disease severity.

Figure 4.

The development of a viral exposure signature predictive of disease severity. (A) XGboost with 10-fold cross validation for 100 iterations of balanced input data generated by ROSE. Each iteration showing the AUC value of training and cross validation sets. (B) COVID-VES signature consisted of 28 viruses that were selected in at least 50 of the 100 iterations predicted by XGboost. (C) Survival risk prediction based on COVID-VES signature viruses in low and high-risk patient groups. Survival time was based on days since admission. (D) The results from Cox proportional hazards regression analysis. ND, not determined.