Integrating Functional, Structural, and Predictive Data in the Analysis of the Recombination Activating Gene (RAG) Genomic Variants

The RAG1 and RAG2 genes encode proteins essential for initiating V(D)J recombination in T cell receptor and immunoglobulin gene loci. Biallelic pathogenic variants in these genes are implicated in a spectrum of immunodeficiency disorders, characterized by clinical heterogeneity. Previous studies from our laboratory have demonstrated that the severity of both clinical and immunological manifestations correlates with the residual recombination activity of the mutant RAG proteins, as assessed by an in vitro functional recombination assay.In this study, we tested the recombination activity of 58 RAG1 and RAG2 variants whose functional activity had never been previously reported. Subsequently, we compiled a list of almost 250 RAG1 and RAG2 missense variants that had been functionally tested in our laboratory throughout the years. We confirmed that in RAG1, the most severe variants (<10% of residual activity) predominantly localize at the catalytic core, especially at the nonamer-binding domain (NBD), responsible for DNA recognition and binding. In RAG2, the most severe variants cluster within the catalytic beta-propeller domain. We then compared the functional data to the in silico AlphaMissense predicted pathogenicity score to evaluate how this tool correlates with the empirical values. Predictions for both RAG1 and RAG2 variants were concordant with our functional assay in more than 80% of the cases to define likely pathogenic variants. However, AlphaMissense predicted only around 67% of the functional variants as benign, reflecting a lower negative predictive value. Interestingly, the variants whose activity was discordant between the two methods do not seem to be clustered in specific domains. To address these discrepancies, we performed computational modelling to predict variant residual activity using models based on different RAG1/RAG2 complex functional states. The predicted recombinase activity scores leveraging computational structural genomic measures correlated very well with our laboratory measures (r = 0.93, p < 1 × 10–16).Overall, with this work, we aim to elucidate how specific RAG gene variants impact protein structure and recombination, and how this correlates with the clinical phenotypes. The final goal is to help to predict the severity of these mutations and to contribute to the interpretation of variants of uncertain significance (VUS).