Comparison of in silico prediction and experimental assessment of ABCB4 variants identified in patients with biliary diseases

Abstract

Genetic variations of the phosphatidylcholine transporter, ABCB4 cause several biliary diseases. The large number of reported variations makes it difficult to foresee a comprehensive study of each variation. To appreciate the reliability of in silico prediction programs, 1) we confronted them with the assessment in cell models of two ABCB4 variations (E528D and P1161S) identified in patients with low phospholipid-associated cholelithiasis (LPAC); 2) we extended the confrontation to 19 variations that we had previously characterized in cellulo. Four programs (Provean, Polyphen-2, PhD-SNP and MutPred) were used to predict the degree of pathogenicity. The E528D and P1161S variants were studied in transfected HEK293 and HepG2 cells by immunofluorescence, immunoblotting and measurement of phosphatidylcholine secretion. All prediction tools qualified the P1161S variation as deleterious, but provided conflicting results for E528D. In cell models, both mutants were expressed and localized as the wild type but their activity was significantly reduced, by 48% (P1161S) and 33% (E528D). These functional defects best correlated with MutPred predictions. MutPred program also proved the most accurate to predict the pathogenicity of the 19 ABCB4 variants that we previously characterized in cell models, and the most sensitive to predict the pathogenicity of 65 additional mutations of the Human Gene Mutation Database. These results confirm the pathogenicity of E528D and P1161S variations and suggest that even a moderate decrease (by less than 50%) of phosphatidylcholine secretion can cause LPAC syndrome. They highlight the reliability of in silico prediction tools, most notably MutPred, as a first approach to predict the pathogenicity of ABCB4 variants.

Vasor: Accurate prediction of variant effects for amino acid substitutions in MDR3

“…As such, Vasor should outperform each predictor due to the additional information gathered from the other features. Additionally, we compared Vasor to MutPred2 as an external prediction tool; the predecessor tool MutPred was indicated to perform well on MDR3 classification problems (16). Vasor outperformed EVE, PolyPhen-2, PON-P2, and MutPred2 according to ROC (Fig.…”

“…We excluded variants with unclear information on disease association (i.e., no in vitro verification analysis and no information on clinical indications for disease association) to eliminate False Positives or False Negatives. As studied benign variants for MDR3 are rare (13,16), further missense variants were obtained from gnomAD v2.1.1 (58) to increase the number of benign variants. During the generation of the gnomAD database, individuals with severe pediatric diseases are removed; however, it is possible that pathogenic variants exist in the gnomAD dataset.…”

“…MDR3 variants were obtained from a literature search for variants causative of MDR3 dysfunction or known variants with no effect in any MDR3-associated disease (4,6,10,12,13,16,. We excluded variants with unclear information on disease association (i.e., no in vitro verification analysis and no information on clinical indications for disease association) to eliminate False Positives or False Negatives.…”

“…To date, there is no MDR3-specific predictor available for classifying amino acid substitutions despite MDR3’s vital role in bile homeostasis. An initial evaluation of general predictor performances on MDR3 variants suggested MutPred as a well-performing tool (16); however, generalization is difficult due to only 21 tested variants with established cellular effects. Additionally, the tested variants presented a clear bias towards pathogenic effects.…”

“…As such, Vasor should outperform each predictor due to the additional information gathered from the other features. Additionally, we compared Vasor to MutPred2 as an external prediction tool; the predecessor tool MutPred was indicated to perform well on MDR3 classification problems(16). Vasor outperformed EVE, PolyPhen-2, PON-P2, and MutPred2 according to ROC (Fig.4 A) and precision-recall-curves (Fig.4 C), with an area under the curve (AUC) of 0.98 for Vasor against 0.90 for EVE, 0.89 for MutPred2, 0.87 for PolyPhen2, and 0.80 for PON-P2 for the ROC and an AUC of 0.94 for Vasor against an AUC of 0.86 for EVE, 0.74 for MutPred2, 0.72 for PolyPhen2, and 0.51 for PON-P2…”