Logo image
Back
Integration of structural modeling and CADD scores improve variant classification for CFH SCRs 5-18
Abstract   Open access   Peer reviewed

Integration of structural modeling and CADD scores improve variant classification for CFH SCRs 5-18

Cobey Donelson, Hector Martín Merinero, Seth Welsh, Santiago Rodríguez de Cordoba, Richard Smith and Yuzhou Zhang
Immunobiology (1979), Vol.230(4), p.152953
07/2025
DOI: 10.1016/j.imbio.2025.152953
url
https://doi.org/10.1016/j.imbio.2025.152953View
Published (Version of record) Open Access

Abstract

Complement Factor H (FH) is a soluble glycoprotein consisting of 20 short consensus repeats (SCRs) organized into three functional regions: N-terminal SCRs 1-4 accelerate the decay of the C3-convertase of the alternative pathway (AP) and act as a cofactor for Factor I-mediated inactivation of C3b; C-terminal SCRs 19-20 recognize cell surfaces by binding sialic acids and deposited C3b; and mid-region SCRs 5-18, the function of which remain poorly characterized. Here, we used a structural prediction algorithm in conjunction with Combined Annotation Dependent Depletion (CADD) scores to provide additional evidence for variant classification in SCRs5-18. As no experimentally determined structure is available for FH SCRs 5-18, we generated a model using AlphaFold3 (AF3) and optimized it with ForceFieldX (FFX) to reduce steric clashes and improve side-chain conformations. We next experimentally validated known pathogenic and benign variants as training sets to determine the impact of missense mutations on protein stability as predicted by ddGun. We used classification thresholds for likely pathogenic (CADD score > 20 and |ddGun| > 1.0 kcal/mol) and likely benign variants (CADD score < 10 and |ddGun| < 0.5 kcal/mol) based on previous studies (1, 2). These thresholds were subsequently applied to evaluate 994 VUSs identified in gnomAD. Our pathogenic training set consisted of 9 variants while the benign had 41 variants. Our combined approach demonstrated a positive predictive value (PPV) of 89% with an error rate of 11%, and a negative predictive value (NPV) of 59% with an error rate of 41% for the test sets. When our methods were applied to the 994 VUSs, we identified 77 (8%) variants supporting a likely pathogenic classification; 8 replace highly conserved residues and 16 are within five positions of such residues. Additionally, we identified 340 variants (34%) with evidence supporting their reclassification as likely benign. Our study demonstrates the feasibility of combining structural modeling and stability predictions for variant interpretation in CFH SCRs 5-18. Future research should focus on expanding the training set and experimental validations. These findings highlight the importance of domain-specific approaches in evaluating CFH variants to improve diagnosis for complement-mediated diseases. (1)Martin Merinero, et al.(2021). "Functional characterization of 105 factor H variants associated with aHUS: lessons for variant classification." Blood 138(22): 2185–2201. (2)Tollefson, M. R., et al. (2023). "Assessing variants of uncertain significance implicated in hearing loss using a comprehensive deafness proteome." Hum Genet 142(6): 819–834.

Details

Metrics

5 Record Views
Logo image