Nucleocapsid-specific antibody function is associated with therapeutic benefits from COVID-19 convalescent plasma therapy

Abstract

Coronavirus disease 2019 (COVID-19) convalescent plasma (CCP), a passive polyclonal antibody therapeutic agent, has had mixed clinical results. Although antibody neutralization is the predominant approach to benchmarking CCP efficacy, CCP may also influence the evolution of the endogenous antibody response. Using systems serology to comprehensively profile severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) functional antibodies of hospitalized people with COVID-19 enrolled in a randomized controlled trial of CCP (ClinicalTrials.gov: NCT04397757), we find that the clinical benefits of CCP are associated with a shift toward reduced inflammatory Spike (S) responses and enhanced nucleocapsid (N) humoral responses. We find that CCP has the greatest clinical benefit in participants with low pre-existing anti-SARS-CoV-2 antibody function and that CCP-induced immunomodulatory Fc glycan profiles and N immunodominant profiles persist for at least 2 months. We highlight a potential mechanism of action of CCP associated with durable immunomodulation, outline optimal patient characteristics for CCP treatment, and provide guidance for development of a different class of COVID-19 hyperinflammation-targeting antibody therapeutic agents.

Keywords: COVID immunomodulation; COVID-19; Fc effector functions; SARS-CoV-2; antibody Fc glycosylation; convalescent plasma; functional antibodies; immunodominance shift; nucleocapsid; systems serology.

Conflict of interest statement

Declaration of interests G.A. is a founder of SeromYx Systems, Inc., an equity holder in Leyden Labs, and a member of the scientific advisory board of Sanofi Pasteur.

Copyright © 2022 The Authors. Published by Elsevier Inc. All rights reserved.

Figures

Graphical abstract

Figure 1

Global anti-SARS-CoV2 response in CCP-treated and control individuals (A) Schematic of the UPenn CCP2 randomized clinical trial of CCP and the Ab profiling performed in this paper. In total, we profiled 302 samples from 79 patients. Patients were randomly assigned to CP treatment (n = 40) or standard of care treatment (n = 39). Patient serum samples were collected on day 1 (n = 79), day 3 (n = 59), day 8 (n = 37), day 15 (n = 44), day 29 (n = 38), and day 60 (n = 45). Because patients experienced symptomatic COVID-19 for a variable number of days prior to presenting to the hospital, we organized patient serum samples by day of the trial (day = 1 enrollment in clinical trial) and by day of symptom onset (day 1 = first day of COVID-19-associated symptoms). (B) Clinical severity score in the CCP-treated and control groups. Significance corresponds to two-sided Wilcoxon test p values (p = 0.0333; ∗p 

Figure 2

CCP provides clinical benefits by limiting development of the inflammatory S Ab trajectory (A) Investigation of pre-existing (day 1) S Ab profiles by UMAP visualization of the samples on day 1 (pre-existing). (B) The polar plots depict the mean percentile of each Ab feature at each week since onset of symptoms across the control arm (top) and CCP treatment arm (bottom). Numbers of patient samples included per time point are as follows (week [# control, # CCP treated]): week 1 (15, 15), week 2 (31, 32), week 3 (24, 20), week 4 (14, 15), week 5 (7, 10), and week 6 (10, 12). (C–E) We employed a four-parameter logistic regression to fit the Ab growth trajectories to dissect the time-specific differences between CCP-treated and control patients for each Ab feature. (C) A visual representation of the logistic regression model and the effect of each parameter on the curve of the model. (D) This heatmap shows the Akaike weighted average parameter differences between the two groups. Each column shows a parameter, which is normalized across the features. The color intensity indicates whether the parameter is higher in the CCP-treated (blue) or control (orange) model. (E) The bar plot depicts the delta-AIC of the best model compared with the model without differences. The higher the delta-AIC, the better the model can explain the trajectory difference. The sign of delta-AIC represents the AUC difference between the CCP-treated and control curves, showing whether the Ab feature is enriched in the CCP-treated model (negative) or the control model (positive). The bars are colored according to whether the feature was enriched in the CCP-treated model (pink) or control model (blue). (F–I) Partial Least Squares regression [PLS-R] model that predicts clinical severity in CCP-treated and control patients based on the top 30 features suggested by the four-parameter logistic model. The PLS-R model uses SARS-CoV-2 humoral profiles from week 3 since onset of symptomatic COVID-19. (F) The score plot of PLS-R regression shows the separation of week 3 samples along the continuum of clinical severity. Each dot represents a patient. (G) The bar graph shows the variable importance in projection (VIP) score in the PLS-R model of the LASSO-selected features. Features are colored by disease severity: features enriched in patients with higher severity (CSC > 20, red) or lower severity (CSC ≤ 20, blue) of COVID-19. (H) The network diagram illustrates the co-correlated features that are significantly correlated with the LASSO-selected features (larger nodes) (p  0.7). Nodes enriched in patients with higher severity or lower severity of COVID-19 are colored red and blue, respectively. (I) The receiver operating characteristic [ROC] curve represents the predictive ability of the LASSO-selected S features (S1 FcR2AH, RBD IgG1, S ADNP, RBD FcAR, S C1q, and S1 FcRn) to distinguish a higher severity score (CSC > 20) from a low severity score (CSC ≤ 20) using the PLS-DA model. Five-fold cross-validations were run 100 times, achieving a mean area under the curve (AUC) of 76.6. The blue line represents the mean ROC curve, and the dotted lines represent individual cross-validation ROC curves. All Ab measurements were taken in technical duplicates (Ab level and FcR binding assays) and biological duplicates (ADCP, ADNP, and ADNKA) and used as an average of the two for the analysis in this figure.

Figure 3

CCP also provides clinical benefits by enhancing N-focused humoral response A nested mixed linear model was created for each Ab feature with and without a variable accounting for patient’s treatment group (CCP treatment versus control) to assess CCP effects on host anti-SARS-CoV-2 humoral development. (A) Volcano plot showing the T value (normalized coefficient) of the patient treatment group variable incorporated in the mixed linear model (x axis) and p value of the LRT for the model fit difference between the two nested models (y axis). A positive T value represents a feature enriched in CCP-treated individuals, and a negative T-value represents a feature enriched in control individuals. (B–D) (B and C) N-specific humoral profile of CCP-treated and control patients. (B) Boxplots of selected N-specific features prior to treatment (day 1). Each box represents the median (central line) and IQR (25th and 75th percentiles), and the two whiskers represent 1.5 × IQR. (C and D) A Linear regression model was used to assess whether N features could predict COVID-19 clinical severity of CCP-treated and control patients as measured by the clinical severity score. (C) The bar plot shows the percentage of explained variance by clinical data (clinical characteristics, severity risk factors, and concurrent medications) and N Ab features. (D) The bar plot shows the contribution of each N Ab feature to COVID-19 clinical severity. The magnitude represents the percentage of variation in the clinical severity score explained by each feature, and the directions represents whether the Ab feature was associated with better (i.e., negative explained variance of clinical severity score [%]) or worse (i.e., positive explained variance of clinical severity score [%]) clinical outcomes. All Ab measurements were taken in technical duplicates (Ab level and FcR binding assays) and biological duplicates (ADCP, ADNP, and ADNKA) and used as an average of the two for the analysis in this figure.

Figure 4

Patients with fewer functional pre-existing Abs benefit the most from CCP (A–F) Seventy-nine samples collected before treatment (day 1) were used to evaluate the association between Ab profiles and clinical severity, as measured by the clinical severity score on day 28. Spearman correlation-based clustering was used to identify the population benefitting from CCP. (A) The heatmap represents the normalized day 1 Ab profiles. Patient samples were clustered into four groups based on the similarity of Spearman correlation coefficients of day 1 SARS-CoV-2 Ab profiles between samples. (B) Principal-component analysis (PCA) plot (bottom) shows the relatedness of patients in each of the identified four clusters, and the density plot (top) displays the organization of the patient samples from each of the four clusters along principal component 1 (PC1). (C) The boxplots show the clinical severity scores of CCP-treated and control patients in each of the four clusters. A Wilcoxon rank test was used to test for differences in clinical severity scores between the two groups in each cluster (two-sided p value: 0.365, 0.799, 1, 0.00415). (D) The volcano plots show the Ab function, titers, and FcR binding features that were most different between cluster 4 and the rest of the population (clusters 1, 2, and 3). The pop-out highlights features with a log fold change (logFC) between −2 and 2 as well as p  2 are colored according to the population in which they were enriched; i.e., in cluster 4 (brown) or in clusters 1, 2, and 3 (cyan). (E) Boxplots of representative Ab functions enriched in non-cluster 4 patients. (F) Boxplot of the clinical severity score of CCP-treated and control patients in cluster 4 and clusters 1, 2, and 3. A two-sided Wilcoxon test was performed to compare age between treatment arms. (G) Boxplot of the age of CCP-treated and control patient in cluster 4 and clusters 1, 2, and 3. A two-sided Wilcoxon test was performed to compare age between treatment arms. (H–J) Re-clustering patients based on the benefit signature identified in (D) and (E) on the whole population. (H) The heatmap shows the two clusters (clusters A and B) identified by the benefit signature—the features that most distinguished cluster 4 from clusters 1, 2, and 3 patients (the top 12 features). (I) Boxplots of clinical severity of CCP-treated and control groups in clusters A and B. The difference between CCP-treated and control patients’ clinical severity was tested by two-sided Wilcoxon test. (J) Boxplots of the age of CCP-treated and control groups in clusters A and B. (K) Three separate linear regression models were used to assess which type of pre-existing Ab features best predicted clinical severity in CCP-treated individuals. The bar plots show the percentage of explained variance by Ab titers, Ab functions, or IgG1 titer-corrected Ab functions in the separate models. We used the top 12 features that differed between cluster 4 and clusters 1, 2, and 3 for the linear regression model of each Ab feature category. For the boxplots in (C), (E)–(G), (I), and (J), each box represents the median (central line) and IQR (25th and 75th percentiles), and the two whiskers represent 1.5 × IQR. ∗p 

Figure 5

CCP recipients have highly sialylated and galactosylated S-specific Fc modifications long after treatment (A–C) (A) Bar graphs of S-specific IgG1 levels in CCP-treated and control participants 60 days after randomization. S-specific Fc glycosylation patterns were measured by capillary electrophoresis in all participants with day 60 samples collected from CCP-treated (n = 19) and control (n = 16) participants. Shown are representative chromatographs of CCP-treated (B) and control (C) participants. (D and E) LASSO PLS-DA was performed to identify the Fc glycan features that separated the two groups. The PLS-DA score plot (D) shows that the S-specific Fc glycans can separate CCP-treated from control participants, with LV1 explaining 41% of variation that separates the two groups along the x axis. Each dot shows an Fc glycan measurement. The LV1 loading plot (E) shows the LASSO-selected features. Pink represents features enriched in CCP-treated participants, and blue represents features enriched in control participants. (F) The Spearman correlation network shows the co-correlated features (small nodes) that are significantly correlated (p  0.5) with the model-selected features (large nodes). Large nodes are colored according to the treatment arm in which they are enriched. Edges are colored by magnitude and sign of correlation, with dark red and dark blue representing strong correlation and anti-correlation, respectively. (G and H) Univariate plots for G2S2FB (G) and disialylated (H) Fc glycans in CCP-treated (pink) and control (blue) participants. ∗p 

Figure 6

Long-lasting N immunodominance in CCP recipients Shown are SARS-CoV-2 functional Ab profiles of 45 patients (CCP-treated, n = 25; control, n = 20) on day 60. (A) Polar plots of the mean percentile of each Ab feature across the control (left) and CCP-treated study arms (right). The features were grouped by the antigen detectors and are depicted in a key. (B) Volcano plot showing the difference between the humoral profile of the CCP-treated group and control group by the FC of mean value (x axis) and two-sided p value from Wilcoxon rank test (y axis). (C–F) LASSO PLS-DA model identified the Ab features that distinguish CCP-treated from control patients on day 60. (C) The PLS-DA score plot demonstrates that CCP-treated and control day 60 patients can be discriminated by the LASSO-selected features. Each dot represents an individual patient. (D) VIP score of the selected features. The magnitude indicates the importance of the features in driving separation in the model. Pink represents a feature enriched in CCP-treated patients, and blue represents a feature enriched in control patients. (E) The performance and robustness of the model was validated with permutation testing. The violin plot shows the distributions of repeated classification accuracy testing using label permutation. The p value from the permutation testing is two sided. Black squares indicate the median accuracy and black lines represent 1 SD. (F) The correlation network shows the co-correlated features (small nodes) that are significantly correlated (p  0.3) with the model-selected features (large nodes). Large nodes are colored according to the treatment arm in which they are enriched. All Ab measurements were taken in technical duplicates (Ab level and FcR binding assays) and biologic duplicates (ADCP, ADNP, and ADNKA) and used as an average of the two for the analysis in this figure.