AFFILIATIONS
1 Beth Israel Deaconess Medical Center, Boston, Massachusetts, United States, Harvard Medical School, Boston, Massachusetts, United States
2 Beth Israel Deaconess Medical Center, Boston, Massachusetts, United States, Harvard Medical School, Boston, MA, United States
3 Beth Israel Deaconess Medical Center, Boston, Massachusetts, United States, Broad Institute, Cambridge, Massachusetts, United States, Harvard Medical School, Boston, MA, United States
4 Beth Israel Deaconess Medical Center, Boston, Massachusetts, United States, Harvard Medical School, Boston, Massachusetts, United States
5 Beth Israel Deaconess Medical Center, Boston, Massachusetts, United States, Harvard Medical School, Boston, Massachusetts, United States
6 Beth Israel Deaconess Medical Center, Boston, Massachusetts, United States, Broad Institute, Cambridge, Massachusetts, United States, Harvard Medical School, Boston, Massachusetts, United States, Massachussetts General Hospital, Boston, Massachusetts, United States
7 Harvard Medical School, Boston, Massachusetts, United States, Massachusetts General Hospital, Boston, Massachusetts, United States
8 Beth Israel Deaconess Medical Center, Boston, Massachusetts, United States, Beth Israel Deaconess Medical Center, Boston, Massachusetts, United States, Harvard Medical School, Boston, Massachusetts, United States
ABSTRACT
Introduction. Tissue factor (TF, encoded by F3) is a type I transmembrane protein whose extracellular domain binds factor VII/VIIa to initiate blood coagulation. Quantitative and qualitative changes in F3 impact the initiation of blood coagulation, but the functional impact of human F3 coding variants is largely unknown. Because stability, trafficking and biogenesis govern surface TF exposure, we generated a random mutagenesis stability map of TF, benchmarked it against in silico predictors, and validated top gnomAD-present hits in vitro and in plasma from human UK biobank participants.
Methods. Random-mutagenesis was used to construct an F3 variant library (depth 2.03×107 clones; mean 1.66 missense mutations/protein) in a lentiviral protein stability reporter such that TF was expressed with a C-terminal GFP followed by an internal ribosome entry site and mCherry to ensure TF-GFP and mCherry synthesis in a 1:1 stoichiometric ratio. U937 monocytic cells were transduced with the library, sorted by GFP:mCherry to isolate the most and least stable variants, and the F3 coding sequence was amplified from genomic DNA for deep sequencing. Reads were mapped to the F3 cDNA reference sequence using BWA-MEM and analyzed using Analyze Saturation Mutagenesis in GATK. Variant effects were scored using DiMSum (n=3 independent screens) and variants with standard deviation <2 evaluated against variant effect prediction (ie AlphaMissense), human genetic (ie gnomAD, ClinVar), and structural modeling (ie AlphaFold, ProteinMPNN) tools. Seven strongly destabilizing human variants were individually reconstructed for validation including TF stability, immunoblot, and a cellular two-stage factor Xa generation assay. Olink proteomic data for UK biobank participants was used to compare plasma TF in carriers of unstable variants and reference sequence controls.
Results. Our library captured all possible single-nucleotide missense variants and all 372 human missense variants present in gnomAD. As expected, truncating variants were significantly more destabilizing compared to missense variants (p=0.02). Buried residues as determined by relative solvent accessibility <0.25 were less tolerant to mutation than surface exposed amino acids (p=0.0003), providing biophysical validation of the screen. Position level mutational tolerance also correlated with evolutionary conservation (ρ=−0.33, p=6×10-8). The strongest determinant of TF abundance was biogenesis rather than stability, with the signal peptide ~6-fold more sensitive to mutation than the rest of the protein (9.0% vs 1.6% destabilizing; p=6×10-18). Against the screen's high-confidence, destabilizing single nucleotide missense variants, no in silico predictor exceeded AUROC 0.63 (ESM-2 0.63, AlphaMissense 0.62, REVEL 0.61). AlphaMissense classified 59% of these experimentally destabilizing variants as likely benign, underscoring the limitations of even the best variant pathogenicity predictors. To explore these discrepancies, several high-confidence gnomAD variants were validated by GFP:mCherry flow cytometry and their ability to support factor Xa generation. All variants tested exhibited reduced TF stability (0.7–66% of WT; six at p<0.05) with a concordant decline in factor X activation. These variants include reclassified variants L21P and N114I, predicted benign by AlphaMissense, yet confirmed destabilizing in cells. In patient plasma, carriers of screen-destabilizing F3 variants (n=12) showed reduced TF levels (~61% of wild-type; p=9.1×10-7), supporting our screen’s capacity to identify type I (quantitative) variants in the human population.
Conclusions. We demonstrate a generalizable, scalable functional genetic approach to identifying unstable human missense variants in membrane proteins. Our experimental TF stability map reproduces structural and evolutionary constraints and highlights limitations of existing variant effect prediction tools. We predict all human F3 missense variants likely to result in partial quantitative TF deficiency and validate these findings in the human population. As routine clinical assays do not capture the contribution of endogenous TF to coagulation initiation, our data will support interpretation of F3 missense variants and their potential contribution to human bleeding and thrombotic phenotypes.
DISCLOSURES
Jacob G Ludington: No relevant conflicts of interest to declare.