AUTHORS
Steven Grzegorski, MD,PhD, Christopher Bidlack, PhD2, Shelby Hemker, N/A3, Jacob Kitzman, PhD4, Karl Desch, MD5
AFFILIATIONS
1 University of Michigan, Ann Arbor, United States
2 University of Michigan, Ann Arbor, United States
3 University of Michigan, Ann Arbor, United States
4 University of Michigan, Ann Arbor, United States
5 University of Michigan, Ann Arbor, United States
ABSTRACT
Background
Rare damaging variants in SERPINC1, which encodes antithrombin (AT), are strong genetic risk factors for venous thromboembolism. Type I AT deficiency is characterized by reduced plasma antigen levels, whereas Type II deficiency involves impaired anticoagulant activity despite normal or near-normal antigen levels. Experimental functional data are available for only a small fraction of possible AT variants, limiting genotype–phenotype interpretation. Although computational prediction models, including AlphaMissense (αMis), continue to improve, prediction of AT pathogenicity remains challenging because its function depends on post-translational modification, secretion, heparin-mediated allosteric activation, metastable serpin dynamics, and precise reactive center loop–protease interactions. This work aims to bridge the gap through comprehensive mechanistic characterization of SERPINC1 single codon variants using a cell-based high throughput platform.
Methods
Plasmid DNA encoding SERPINC1 with an eGFP tag was modified to generate a comprehensive library of single-codon variants and cotransfected with Bxb1 integrase into a HEK293 landing-pad cell line. SERPINC1 variants causing secretion defects result in retention, and thus an increase in intracellular eGFP fluorescence. Cells expressing individual variants were sorted by relative eGFP expression into quartile bins. Variant frequencies were quantified by deep sequencing and used to calculate an enrichment score for each variant. A secondary functional screen of the C-terminal region (codons 385-432) was performed using a surface-displayed AT construct. Cells expressing individual variants were incubated with fluorescently labeled thrombin for 30 minutes, sorted, and processed as in the primary screen with fluorescence proportionate to AT activity.
Results
Conditioned-media ELISA confirmed that cells in the highest intracellular AT-eGFP quartile secreted significantly less AT-eGFP than cells in the lowest quartile or the unsorted population, validating intracellular retention as a proxy for secretion defects. Across three independent library screens, high-confidence data were obtained for 94% of all missense and nonsense variants (8746/9280). Secretion-defective variants from this screen showed strong agreement with known human Type I AT-deficiency. Using these known variants as a benchmark for Type I pathogenicity scores, our screen identified 1899 missense variants causing retention. Notably, substitutions in the reactive center loop and known Type II variants generally did not cause intracellular retention. Compared to ClinVar annotations, our enrichment scores demonstrate retention in 0% (0/8) of benign or likely benign variants, 25% (34/134) of variants of uncertain significance, and 36% (24/66) of pathogenic or likely pathogenic variants.
From the above secretion defects, αMis predicts 1684 (88%) as likely pathogenic, suggesting high sensitivity of αMis for pathogenicity. However, αMis predicts 3479 of missense variants as pathogenic, while our model shows that 1684 (48%) of these cause AT retention, suggesting low specificity αMis for Type I pathogenicity.
To investigate discordant predictions between our retention screen and αMis, we performed a functional screen of the 48 C-terminal amino acids. We obtained informative coverage for 814/924 (88%) of possible variants. As expected, disruption of the reactive center loop and adjacent structural motifs markedly reduced thrombin-associated functional activity. Of the variants with informative screen data, 367 variants were pathogenic by αMis. 144 (39%) showed loss of thrombin binding in the secondary screen and 167 (44%) showed a secretion defect. 162 (44%) of αMis predicted pathogenic variants did not show an experimental defect. In contrast, 95% (261/274) of αMis predicted benign variants had normal secretion and function.
Conclusions
We generated a near-comprehensive atlas of SERPINC1 missense variation, assigning likelihood scores for AT secretion defects and providing functional characterization of C-terminal variants, including the reactive center loop involved in Type II deficiency. The results provide further evidence for ClinVar annotations and demonstrate that our in vitro screens improve on the in silico predictions of αMis with improved specificity. Overall, this work helps investigators convert genetic uncertainty into a mechanism-aware dataset for interpreting AT deficiency and thrombosis risk.
DISCLOSURES
Steven Grzegorski, MD,PhD: No relevant conflicts of interest to declare.