High-resolution multiplexed antibody-omics and interpretable machine learning unveil novel pathogenic mechanisms in kidney transplant rejection

Antibody-mediated rejection (AbMR), driven by donor-specific alloantibodies (DSAs), is a major cause of late-stage kidney allograft failure, leading to premature graft loss in over half of affected patients. Despite efforts to link DSA features (e.g., HLA-specific IgG titers) to rejection risk, the immune mechanisms distinguishing DSA+ patients who develop AbMR remain unclear. In this first-in-class study, we develop a sample-sparing and cost-effective technique to generate the most comprehensive biophysical profile of DSAs reported to date. Further, given the complex pathological context and heterogeneity of samples we use a novel interpretable machine learning algorithm to learn signatures reflecting putative causal mechanisms of transplant rejection. We identify distinct mechanistically informative signatures at early and late times post-transplant. These antibody signatures, reflecting both quality and quantity of the humoral response, successfully discriminate DSA+ patients with and without AbMR. In addition to recapitulating known features of AbMR, our analyses reveal a significant and previously underappreciated role for IgM responses and glycosylation patterns, including sialylation and galactosylation, in both early and late rejection. Our identified signatures hold across two independent and geographically distinct cohorts. Leveraging biomedical and computational innovation, we resolve prior inconsistencies in the field by implementing an unbiased systems framework identifying biophysical trends. These trends include selective enrichment of class I HLA-specific IgM and class II HLA-specific IgG responses in late and early rejection, respectively, which were overlooked earlier due to assay and methodological limitations. Corresponding functional relevance of putative causal signatures is further supported by observations from a murine model of chronic rejection, where we observe a significant increase in serum IgM-DSA associated with high risk of rejection as compared to serum IgG-DSA, warranting further exploration into the role of IgM in AbMR. Finally, addressing the lack of a comprehensive approach for pre-diagnosis of late AbMR patients reflecting the complex pathology of late AbMR and heterogeneity of samples (with time post-transplant ranging from 1-10 years), we formulate a risk score from our signatures. This composite risk score, combining IgM and sialylation metrics robustly predicts late AbMR with high sensitivity and specificity, offering a clinically actionable tool for early risk stratification. Together, leveraging our innovative pipeline we show the distinct roles of antibody isotypes/subclasses and glycosylation in disease progression, with IgM and glycosylation signatures showing strong diagnostic and prognostic value. Ultimately, the modularity of approach establishes a generalizable framework for understanding a plethora of complex immune-mediated tissue injury contexts beyond kidney transplantation.