Systems vaccinology identifies immunological correlates of SARS-CoV-2 vaccine response in solid organ transplant recipients

Abstract

Solid-organ transplant (SOT) recipients are at enhanced risk of infection and to poorly respond to vaccination due to comorbidities and immunosuppression. We performed a systems vaccinology study in 59 kidney and 31 lung transplant recipients who received 3 doses of COVID-19 mRNA BNT162b2 vaccine. We were able to characterize a baseline configuration associated with an effective humoral response to 3 doses, characterized by an innate and activated B cell profile, whereas a T cell signature was associated with a poorer response. We observed a distinct configuration associated with a detectable humoral response to 2 doses, partly mediated by double negative B cell subsets. These results suggest that, despite their immunosuppression, some SOT recipients can induce an effective humoral response to 3 doses of vaccine supported by a baseline configuration close to the healthy phenotype. Baseline immune phenotyping may help identify SOT recipients at the greatest risk of a poor vaccine response.

Fig. 1. Identification of key immune parameters associated with positive humoral response to 3 doses of COVID-19 mRNA vaccination in SOT recipients.

One hundred ten individuals were recruited to a systems vaccinology study, including 20 healthy controls, 31 LTR, and 59 KTR. Immunological profiles were assessed at baseline (day 0 of vaccination) for 444 immunological variables without any imputation. Multivariate logistic regression for each immunological variable was performed adjusting by sex, age, type of transplant and the number of years since transplantation, identifying the 10 highest magnitude parameters (based on log(OR of the z-score)) between significant (p-value < 0.05) associated with positive humoral response to 3 doses of COVID-19 mRNA vaccination. A Odds ratio (OR) of the z-score of the 10 highest difference parameters (magnitude cut-off), using the stratification of SOT recipients “positive” or “non-positive” responders according to anti-RBD binding IgG levels to 3 doses of COVID-19 mRNA vaccination. B Raw value graphs, showing healthy controls, dose 3 non-positive and dose 3 positive responders SOT recipients, for the 10 highest magnitude parameters associated with positive response to 3 doses of mRNA vaccination.

Fig. 2. Baseline immune state and immunological relationships associated with positive humoral response to 3 doses of COVID-19 mRNA vaccination and machine learning approach predicting positive humoral response to 3 doses of COVID-19 mRNA vaccination in SOT recipients.

One hundred ten individuals were recruited to a systems vaccinology study, including 20 healthy controls, 31 LTR, and 59 KTR. Immunological profiles were assessed at baseline (day 0 of vaccination) for 444 immunological variables without any imputation. Immune configurations displayed as UMAP plots for listed parameters for each individual, annotated based on patient characteristics. Distances between groups were calculated using the Calinski and Harabasz score and are shown on the right of each panel. For immunological relationships, multivariate logistic regression was performed adjusting by sex, age, type of transplant and the number of years since transplantation, identifying the 10 highest magnitude parameters (based on log(OR of the z-score)) between significant (p-value < 0.05) associated with positive humoral response to 3 doses of COVID-19 mRNA vaccination. Spearman’s correlations were calculated between each of the 10 highest magnitude parameters. For the machine learning approach, the support vector machine approach was selected to model vaccine response, based on the highest performance. A UMAP for all 444 all immune parameters, with individuals annotated as healthy controls, or dose 3 non-positive and dose 3 positive SOT recipients. B UMAP for 10 highest magnitude parameters associated with positive responses after 3 doses, with the same patient annotation. C For those immune parameters associated with positive response after 3 doses, Spearman’s correlations between each parameter pair are shown for healthy controls (left). For positive (center) and non-positive (right) after 3 doses in SOT recipients, the upper-right triangular matrix shows the Spearman’s correlations within the respective patient population, while the lower-left triangular matrix visualizes by color and size the deviations in the correlations observed between the SOT recipients and the healthy controls with green indicating high correlation within the healthy population and low correlation within the SOT population and brown low correlation within the healthy population and high correlation within the SOT population. D Values of the differential correlation for each cell population associated with a positive response to 3 doses, comparing the healthy controls and the respective SOT recipients as individual values (left) and violin plot (right). E ROC curve of the machine learning algorithm developed for predicting a positive response to 3 doses, as evaluated by nested cross-validation. Curve is mean performance in test groups during 10-fold cross-validation, with fill area the standard deviation. F Best features according to the machine learning model to identify the positive response to 3 doses of vaccine. The best features were chosen using the feature selection algorithm RFE-SVM, and the evaluation of the importance of each feature in the model was done using clustered permutation.

Fig. 3. Identification of key immune parameters associated with detectable humoral response to 2 doses of COVID-19 mRNA vaccination in SOT recipients.

One hundred ten individuals were recruited to a systems vaccinology study, including 20 healthy controls, 31 LTR and 59 KTR. Immunological profiles were assessed at baseline (day 0 of vaccination) for 444 immunological variables without any imputation. Multivariate logistic regression for each immunological variable was performed adjusting by sex, age, type of transplant and the number of years since transplantation, identifying the 10 highest magnitude parameters (based on log (OR of the z-score)) between significant (p-value < 0.05) associated with detectable humoral response to 2 doses of COVID-19 mRNA vaccination. A Odds ratio (OR) of the z-score of the 10 highest difference parameters (magnitude cut-off), using the stratification of SOT recipients “detectable” or “undetectable” responders according to anti-RBD binding IgG levels to 2 doses of COVID-19 mRNA vaccination. B Raw value graphs, showing healthy controls, dose 2 undetectable and dose 2 detectable responders SOT recipients, for the 10 highest magnitude parameters associated with detectable response to 2 doses of mRNA vaccination.

Fig. 4. Baseline immune state and immunological relationships associated with detectable humoral response to 2 doses of COVID-19 mRNA vaccination and machine learning approach predicting detectable humoral response to 2 doses of COVID-19 mRNA vaccination in SOT recipients.

One hundred ten individuals were recruited to a systems vaccinology study, including 20 healthy controls, 31 LTR and 59 KTR. Immunological profiles were assessed at baseline (day 0 of vaccination) for 444 immunological variables without any imputation. Immune configurations displayed as UMAP plots for listed parameters for each individual, annotated based on patient characteristics. Distances between groups were calculated using the Calinski and Harabasz score and are shown on the right of each panel. For immunological relationships, multivariate logistic regression was performed adjusting by sex, age, type of transplant and the number of years since transplantation, identifying the 10 highest magnitude parameters (based on log (OR of the z-score) between significant (p-value < 0.05) associated with detectable humoral response to 2 doses of COVID-19 mRNA vaccination. Spearman’s correlations were calculated between each of the 10 highest magnitude parameters. For machine learning approach, the support vector machine approach was selected to model vaccine response, based on the highest performance. A UMAP for all 444 all immune parameters, with individuals annotated as healthy controls, or dose 2 undetectable and dose 2 detectable SOT recipients. B UMAP for 10 highest magnitude parameters associated with detectable responses after 2 doses, with the same patient annotation. C For those immune parameters associated with detectable response after 2 doses, Spearman’s correlations between each parameter pair are shown for healthy controls (left). For detectable (center) and undetectable (right) after 2 doses in SOT recipients, the upper-right triangular matrix shows the Spearman’s correlations within the respective patient population, while the lower-left triangular matrix visualizes by color and size the deviations in the correlations observed between the SOT recipients and the healthy controls with green indicating high correlation within the healthy population and low correlation within the SOT population and brown low correlation within the healthy population and high correlation within the SOT population. D Values of the differential correlation for each cell population associated with a detectable response to 2 doses, comparing the healthy controls and the respective SOT recipients as individual values (left) and violin plot (right). E ROC curve of the machine learning algorithm developed for predicting a detectable response to 2 doses, as evaluated by nested cross-validation. Curve is mean performance in test groups during 10-fold cross-validation, with the fill area the standard deviation. F Best features according to the machine learning model to identify the detectable response to 2 doses of vaccine. The best features were chosen using the feature selection algorithm RFE-SVM, and the evaluation of the importance of each feature in the model was done using clustered permutation.