A Multi-State Structural Genomics Approach Enables Large-Scale, Mechanistic, and Context-Specific Classification of ABCC6 Genetic Variants Implicated in Calcification Diseases

Abstract

Purpose: Genetic variation in ATP Binding Cassette Subfamily C Member 6 (ABCC6) can cause both pseudoxanthoma elasticum (PXE) and generalized arterial calcification of infancy (GACI). Despite both diseases being rare, there are already 930 distinct missense variants in ABCC6 reported, 87% of which are of uncertain clinical significance (VUS). New approaches are needed to interpret and classify these VUS mechanistically.

Methods: We developed 3D protein models of ABCC6 in three functionally relevant conformations to calculate the structural effects of variants and identify 3D mutational hotspots. With this and additional functional information, we categorized variants in a mechanistic ontology based on which critical functions of ABCC6 they impact. We then compared PXE and GACI-associated variants.

Results: We identified two three-dimensional hotspots of pathogenic variants and six specific functions of ABCC6 which variants impact. From this, we propose a mechanism for pathogenicity for 41% of VUS according to their impacted function, 30 of which could be reclassified as Likely Pathogenic from our non-clinical data. Finally, we found slight differences between PXE and GACI-associated variants.

Conclusion: The mechanistic information we present will guide future research to better address calcification disorders and understand genetic variants. Further, our VUS reclassification will improve the diagnosis of ABCC6-driven diseases, shortening diagnostic odysseys. We believe that computational structural genomics approaches will soon take prominence in genomics data interpretation.

Keywords: ABCC6; Generalized arterial calcification of infancy; Genomic Interpretation; Precision Medicine; Pseudoxanthoma elasticum; Variant Prioritization.

Figure 1:. The interpretation and classification of genetic variants in ABCC6 is greatly enhanced by our novel process, which uses 3 state models of the protein.

A) The domains of ABCC6 are shown linearly along the protein sequence where Transmembrane domains 1 and 2 (TMD1 and TMD2) are colored purple, Nucleotide binding domains 1 and 2 (NBD1 and NBD2) are light blue, Loop 0 and 1 (L0 and L1) are colored tan, and Transmembrane domain 0 (TMD0) is pink.B) ABCC6 was represented in three states: on the left, the inward-facing and apo conformation (also called “state 1”); in the middle, the intermediate or ligand-bound conformation (also called “state 2”); and on the right, the outward-facing or ATP-bound conformation (also called “state 3”). In the diagram, the domains are colored the same as in B. Further, the P-pocket is colored orange, the H-pocket yellow, the catalytically active nucleotide binding site is colored green and the inactive site red^,. C) The flow of variants through our process, as described in Methods, is shown as a Sankey plot. Variants are first collected and collated to those within our structures, then characterized and categorized by their structural and functional information, which was used to identify the mechanism by which variants impact ABCC6 function. This could be used as evidence for ACMG variant interpretation lines PM1 and PP3 to reclassify at least 30 variants from our non-clinical data alone.

Figure 2:. Structural modeling of ABCC6 reveals three specific mechanisms of variant effects.

A) NP_001162.5:p.G1042S (black) disrupts conformational dynamics, as shown within the state 3 model. The serine replacement will interact with the backbone of the helix opposite to it – a helix that, despite having five serines within a 9-residue range, does not normally interact with its opposite helix. As these helices are not supposed to interact but instead be free to fall back to the state 1 conformation after hydrolysis, this substitution will inhibit the dynamics of ABCC6. B) NP_001162.5:p.R1138Q (black) disrupts structural stability, as shown in the state 2 model, with many bonds connecting the TMD’s arginine to the NBD’s atoms. This arginine stabilizes the crucial joint between the two domains; when this arginine is lost for an uncharged glutamine, all these bonds are lost, and the structural stability of this joint is lost in all three states. Therefore, this variant disrupts the stability of ABCC6. C) NP_001162.5:p.G1302R (black) disrupts ATP binding, as shown in the state 3 model. This glycine is close to many sidechains and ATP at the turn of a loop. Replacing glycine with arginine causes many steric clashes, which would cause the nucleotide-binding site to disform and the nucleotide to not bind within the pocket. This will disrupt the ATP binding of ABCC6.

Figure 3:. Many pathogenic variants are spatially and sequentially near functionally significant residues across ABCC6, with the majority of PXE variants found in the nucleotide-binding domains and the majority of GACI/PXE variants in the transmembrane domains.

A) Demonstrates the location of ABCC6 variants across the protein (shown here in state 1), with the black line of distribution shown, while B) shows the location of functionally significant zones, including domains and regions with known functional impact. Both have the same scale graphically and the same model structurally for ease of comparison.

Figure 4:. Structural Stability of ABCC6 in 3 functional states reveals that pathogenic variants and, to a lesser extent, VUS are most often destabilizing.

The “largest score across states” corresponds to the highest or lowest value (whichever has the greater absolute value) for energetic stability across all three states. Hence, variants with the largest score as highly destabilizing are highly destabilizing in at least one state. Orange horizontal lines denote a significant change (in either direction) from WT, at −1.8kcal/mol and 1.8kcal/mol.

Figure 5:. Most pathogenic variants are confidently predicted, usually affecting ATP binding or structural stability.

Variant allele frequencies are shown by each structure-based function class of ABCC6 that they likely impact. Individual variants are shown within the smoothed distribution and colored by their associated phenotype from ClinVar. Total counts for variants within each effect group are shown on the right.

Figure 6:. ABCC6 pathogenic variants are highly clustered within three distinct regions.

The hotspots of ABCC6 pathogenic alleles, using only allele counts of pathogenic variants on gnomAD, where red areas have more than 30 alleles, orange areas have 17–30 alleles, and yellow regions have 12–16 alleles. Variant locations are marked with a sphere, which is pink when associated with PXE and blue when associated with GACI/PXE. Structures are of State 1 and State 3.

Figure 7:. Most VUS were predicted, with most affecting structural stability.

Variants are shown across allele frequency and separated by the function of ABCC6 that they likely impact, with the individual variants shown within the distribution colored by the phenotype associated with the variant. Total counts for variants within each effect group are shown on the right.

Figure 8:. Allele Frequencies for both Pathogenic and Uncertain Significance Variants are higher in variants that cause GACI/PXE than variants that cause PXE alone.

On the left, missense pathogenic variants are shown by their allele frequency, whereas variants found only in the ClinVar dataset with no known allele frequency are at the bottom of the plot. The density of variants across the allele frequency range reveals that most GACI/PXE variants are rarer than 1×10⁻⁴. In contrast, most PXE-only variants have lower allele frequencies or are unobserved in the healthy population. On the right, all missense VUS from the gnomAD dataset are similarly arranged by their allele frequency, revealing once again that the density of PXE variants is greatest at a lower allele frequency than variants that cause GACI/PXE.