Abstract
Introduction
B-cells have been strongly implicated in cardiac allograft rejection (CAR). Recently, however, the CTOT-11 trial showed that depleting mature CD20+ B-cells did not reduce rates of rejection in cardiac allograft recipients and unexpectedly increased the severity of allograft vasculopathy. Therefore, it can be hypothesized that differing phenotypic subtypes of B-cells correspond with different biological mechanisms relating to CAR. Though, current applications to quantify these subtypes of immune cells, i.e with immunohistochemistry or flow cytometry, are often restricted by limited cell markers and cost-burden; therefore, we demonstrate a novel deconvolution method, FARDEEP, that has been validated to accurately enumerate peripheral blood mononuclear cell-subtypes (PBMCs) in a quicker and more cost-effective manner.
Purpose
To better understand the association of different B-cell subtypes in CAR by identifying the B-cell subtype most predictive for pathologically defined rejection.
Methods
The machine learning tool, FARDEEP, was trained with the transcriptomic signatures of 29 PBMC subtypes, characterized by previous single-cell RNA experiments. FARDEEP then was used to deconvolute data-mined RNA from 259 blood samples from 98 cardiac allograft recipients enrolled in the CARGO study (GSE2445). Random forest tree (RF) was then used to analyze the levels of deconvoluted subtypes to predict the severity of rejection assessed by endomyocardial biopsy. Finally, RF was used to identify the subtypes of PBMCs most valuable in predicting rejection.
Results
Out of the 259 samples with consensus pathological readings, 140 had a consensus International Society of Heart and Lung Transplant grade of 0, 63 with grade 1a, 31 with grade 1b, and 25 with grade 3a or higher. We grouped biopsy samples with grade 0, 1a, and 1b as “low-risk” rejection (n=234). 3a or higher samples were grouped as “high-risk” (n=25). There were no grade 2s in the dataset. According to the dataset, blood was extracted from patients on average 72.5 days post-transplant. The RF had good performance in predicting rejection severity. (Figure 1a) CD20- plasmablast cells were stronger predictors for differentiating high-risk from low-risk compared to CD20+ B-cell populations (i.e B Naive and B Memory cells). (Figure 1b) Overall, however, dendritic cells (DCs), neutrophils, monocytes, and basophils were the strongest predictors for rejection.
Conclusion
Our findings support the results from the CTOT-11 trial showing that CD20+ B-cells may not contribute to CAR as significantly as seen with other PBMC subtypes. Instead, we showed that among B-cells, CD20- plasmablasts were more likely associated with CAR, possibly explaining why targeting CD20 was ineffective in preventing rejection. Thus, targeting plasmablast-associated markers could potentially be more useful to prevent CAR.
Model Performance with Variables
Funding Acknowledgement
Type of funding source: Private grant(s) and/or Sponsorship. Main funding source(s): 1) Society of Academic Emergency Medicine Foundation; 2) The Jewish Fund
Chemokines and B cells in renal inflammation and allograft rejection