Genetic heterogeneity of heritable ectopic mineralization disorders in a large international cohort

Abstract

Purpose:

Heritable ectopic mineralization disorders comprise a group of conditions with a broad range of clinical manifestations in non-skeletal connective tissues. We report the genetic findings from a large international cohort of 478 patients afflicted with ectopic mineralization.

Methods:

Sequence variations were identified using a next-generation sequencing panel consisting of 29 genes reported in association with ectopic mineralization. The pathogenicity of select splicing and missense variants was analyzed in experimental systems in vitro and in vivo.

Results:

A total of 872 variants of unknown significance as well as likely pathogenic and pathogenic variants were disclosed in 25 genes. A total of 159 distinct variants were identified in 425 patients in ABCC6, the gene responsible for pseudoxanthoma elasticum, a heritable multi-system ectopic mineralization disorder. The interpretation of variant pathogenicity relying on bioinformatic predictions did not provide a consensus. Our in vitro and in vivo functional assessment of fourteen ABCC6 variants highlighted this dilemma and provided unambiguous interpretations to their pathogenicity.

Conclusions:

The results expand the ABCC6 variant repertoire, shed new light on the genetic heterogeneity of heritable ectopic mineralization disorders, and provide evidence that functional characterization in appropriate experimental systems is necessary to determine the pathogenicity of genetic variants.

INTRODUCTION

Ectopic mineralization, the deposition of hydroxyapatite in non-skeletal connective tissues, is a major cause of morbidity and mortality worldwide, particularly when it affects the cardiovascular system.¹ Abnormal calcification of blood vessels occurs in cardiovascular diseases, cancer, diabetes, chronic kidney disease, and autoimmune inflammatory conditions. Although ectopic mineralization usually occurs via acquired dysplastic, metabolic, or inflammatory mechanisms complicating the identification of the critical pathophysiological cause in each disorder, it also occurs due to variants in genes encoding the cellular machinery that physiologically maintains mineral homeostasis and protects against calcium phosphate precipitation. To this end, recent studies of genetic disorders of ectopic mineralization have disclosed several causal genes that have provided new insights into the mechanisms of ectopic mineralization processes in general.^2,3 Among the hereditary ectopic mineralization disorders, pseudoxanthoma elasticum (PXE, OMIM 264800) is the most common orphan disease with prevalence of ~ 1:50,000.

PXE, a multi-system disorder with late-onset clinical manifestations in the skin, eyes, and arterial blood vessels,⁴ is caused in most cases by pathogenic vaiants in the ABCC6 gene.⁵ The encoded protein, ABCC6, is a transmembrane transporter expressed primarily in hepatocytes. Although the substrate of ABCC6 remains unknown, reduced plasma levels of inorganic pyrophosphate (PPi), a potent mineralization inhibitor, in an ABCC6-dependent manner, starts to explain the metabolic nature and the underlying pathomechanism of ectopic mineralization in PXE.² Other clinical entities, both genetic and acquired, share overlapping and diverging clinical manifestations with PXE, making it extremely difficult to classify patients based on their pathogenic variants in distinct genes.

The advent of next-generation sequencing (NGS) has enabled progress in the detection of pathogenic variants in various genetic disorders. While sequence variants in distinct genes for ectopic mineralization disorders have been identified, they remain mostly uncharacterized, thus hampering their association with clinical diagnosis. In this study utilizing a targeted NGS gene panel, we report the mutational landscape in a large international cohort of patients afflicted with ectopic mineralization. We also experimentally assessed the outcomes of several synonymous, splicing, and missense variants, which among all types of variants are under-recognized, often overlooked, or erroneously annotated. We focused on ABCC6, the gene implicated in PXE, a prototype of heritable ectopic mineralization disorders. Due to the metabolic nature of PXE, the establishment of the causal link between ABCC6 transporter functionality in the liver and mineralization in connective tissues requires specialized assays to delineate the pathogenenic effects of ABCC6 variants. We developed in vitro and in vivo functional assays which provided critical information as to whether a variant in ABCC6 is disease-associated.

MATERIAL AND METHODS

Patient cohort

The international cohort included 478 patients enrolled in the registry of PXE International, a patient organization advocating on behalf of patients and families with PXE.⁶ These patients have a clinical diagnosis of PXE by a dermatologist and/or an ophthalmologist, based on the diagnostic criteria as proposed by the PXE International Research Consortium at the 2014 Annual Meeting.⁷ They had either a skin biopsy, which by histopathological stains revealed fragmented elastic fibers and calcium phosphate deposition in the lesional skin, and/or retinal changes including peau d’orange and angioid streaks. The diagnosis was primarily made on consensus-based clinical practice guidelines for this rare disease. These clinical criteria are not validated and measures of sensitivity and specificity are not applicable.

Targeted-multigene NGS sequencing and bioinformatics

Genomic DNA was obtained from PXE International Biobank and extracted from saliva or peripheral blood samples of affected individuals (DNA Genotek Inc., Ontario, Canada; Qiagen, Valencia, CA). The NGS panel contained 21 ectopic mineralization-associated genes (ABCC6, ADIPOQ, AHSG, ANKH, APOE, ATF4, CASR, ENPP1, FAM20A, FGF23, GALNT3, GGCX, KL, MGP, NT5E, SAMD9, SLC20A2, SLC29A1, SPP1, TRIM24, and TNFRSF11B)⁸ as well as 8 genes (A2AP, ALPL, ENTPD1, PAI-1, PLAT, PLAU, PLAUR, and PLG) involved in the fibrinolysis pathway and regulation of ectopic mineralization in soft tissue following injury.^9–12 We performed variant detection and bioinformatic analyses according to previously published approaches.^13,14 Final prioritization of variants employs a two-tiered procedure following the latest American College of Medical Genetics and Genomics/Association for Molecular Pathology (ACMG/AMP)¹⁵ and Sherloc guidelines.¹⁶ First, we included sequence variants that were classified by ACMG/AMP as variants of unknown significance (VUS), likely pathogenic (LP), and pathogenic (P). Secondly, these variants were refined by Sherloc using a semi-quantitative 0–5 classification system. Variants classified as 3-U, VUS and above were included in the final variant list.

The ABCC6 gene-centric prediction of selected variants was performed by plotting the combined annotation-dependent depletion (CADD) score versus minor allele frequency (MAF) for each variant using PopViz server.¹⁷ The CADD damaging scores were used to assess the deleteriousness of each variant.¹⁸ The ABCC6-gene specific CADD score within the 95% confidence interval and the recommended MAF threshold were calculated using the mutation significance cutoff (MCS) method.¹⁹

Genotype-phenotype correlation analysis

Of 478 patients with ectopic mineralization, 172 PXE patients with biallelic ABCC6 variants had complete Phenodex scores, the internationally standardized scoring system for evaluation of the severity of PXE in five main clinical areas: skin (S), eyes (E), gastrointestinal system (G), heart (C), and vasculature (V).²⁰ The composite of the Phenodex scores reflects the severity of PXE. The genotype-phenotype analysis in these patients was performed according to previously published method.²¹

In vitro mini-gene splicing assay for ABCC6 synonymous and intronic variants

WT ABCC6 mini-gene segments were cloned into pCMV-3Tag-8 or pCMV-3Tag-3a vector. Individual sequence variants were generated using the Quick-Change XL site-directed mutagenesis kit (Stratagene, La Jolla, CA). HepG2 cells were transfected with WT or mutant constructs using FuGENE HD transfection reagent (Promega, Madison, WI). Cells were collected 48 hours post-transfection for RNA extraction. The splicing of ABCC6 mini-gene pre-mRNA was assessed by RT-PCR using a forward primer specific to the mini-gene and a reverse primer targeting the vector’s Flag-tag sequence, which allows splicing of the mini-gene-produced transcripts to be studied. The different transcript isoforms were Sanger sequenced.

In vivo assessment of human ABCC6 missense variants

The WT human ABCC6 cDNA was cloned into a pENTR™/D-TOPO vector (Life Technologies, Carlsbad, CA).²² Constructs containing each variant were generated using the Quick-Change XL site-directed mutagenesis kit (Stratagene). The WT or mutant ABCC6 cDNA was recombined into the pAd/CMV/V5-DEST adenovirus vector and the replication-deficient recombinant adenoviruses were produced in HEK293A cells as previously described.²² The Abcc6^tm1JfK mice²³ on immunodeficient Rag1^tm1Mom/J mouse background (referred to as Abcc6^−/−Rag1^−/−) were used to test the pathogenicity of ABCC6 missense variants, with appropriate age-matched wild-type (WT) C57BL/6J mice. In all groups, 6-week-old Abcc6^−/−Rag1^−/− mice were intravenously injected with recombinant adenoviruses at 4×10⁸ infectious unit (IFU) per mouse. Mice were euthanized at one or four weeks after injection followed by immunostaining of the ABCC6 protein in the liver, determination of PPi concentrations in plasma, and analysis of ectopic mineralization in the muzzle skin, as previously described for the WT human ABCC6 protein.²²

Statistical analysis

The data were analyzed by two-way ANOVA. Genotype-phenotype correlation analysis was performed using Fisher’s exact test across each of the five Phenodex clinical areas. Statistical significance was considered with P < 0.05 and all statistical analyses were conducted using Prism 8 (GraphPad, San Diego, CA).

RESULTS

Genetic landscape in 478 patients with ectopic mineralization

Our cohort included 478 PXE patients from 413 distinct families. The highly variable clinical presentation in our cohort prompted an effort to confirm the diagnosis and classify the patients based on their variants in distinct genes. To identify sequence variants and genes that underlie ectopic mineralization in these individuals, we developed a 29 gene-targeted NGS panel. After classifying the variants according to ACMG/AMP criteria¹⁵ followed by further refinement by Sherloc,¹⁶ a total of 872 variants of unknown significance as well as likely pathogenic and pathogenic variants were identified in 438 patients in all genes except ATF4, MGP, SPP1 and TNFRSF11B, expanding the molecular genetic bases of this group of heterogeneous disorders (Table S1).

Not surprisingly, most of the patients (88.9%, 425/478) had variants in ABCC6, the candidate gene for the classic form of PXE. When stratified by the number of variants, 327 patients (68.4%, 327/478) had biallelic variants in ABCC6 among whom 60 had also monoallelic variants in other genes. Among all patients, 96 (20.2%, 96/478) had a monoallelic variant in ABCC6 among whom 19 also had monoallelic variants in other genes (Fig. 1a). Among the 159 distinct variants identified in ABCC6, missense variants were the most frequent (58.5%, 93/159), followed by indel (18.9%, 30/159), nonsense (12.5%, 19/159), and putative splicing variants (10.7%, 17/159). A total of 73 distinct ABCC6 variants were not previously reported (45.9%, 73/159); these included 40 missense, 8 nonsense, 18 indels, and 7 splicing variants (Fig. 1b, Fig. 2). In our cohort, variant p.R1141* is the most common, with a prevalence of 37.6% (180/478). AluI-mediated deletion of exons 23–29 (del23–29), the second most common variant, accounted for 17.4% (83/478) of all patients (Fig. 1c). No statistically significant genotype-phenotype correlation was observed in 172 patients who had both biallelic ABCC6 variants and a complete Phenodex score (Fig. S1).

Figure 1. Genetic spectrum of 478 patients with ectopic mineralization.

(a) Left panel: Among 478 patients, 327 had biallelic variants in ABCC6 (68.6%, 327/478), 96 had monoallelic ABCC6 variant (20.1%, 96/478), 40 did not have variants in any of the 29 genes (8.4%, 40/478), 15 had variants in other genes (3.1%, 15/478). In addition to ABCC6, variants were also identified in other genes in 60 and 19 patients (numbers are boxed), accounting for 18.5% and 19.8% of patients with biallelic and monoallelic ABCC6 variants, respectively. Right panel: Variant distribution in the 15 patients who carry biallelic variants in GGCX, ENPP1, SAMD9, ENTPD1, GALNT3, and PLG, or monoallelic variant in GGCX or ENPP1. Among 106 previously unreported variants in genes other than ABCC6, 5 were Class 4 (4 LP and 1 P), 10 were Class 5 (10 P), and 93 were Class 3 (1 LB, 12 LP, and 80 VUS) (for details, see Table S1). (b) Signature of distinct variants in ABCC6. Among 159 distinct variants in ABCC6, missense variants were the most frequent (58.5%, 93/159), followed by indel (18.8%, 30/159), nonsense (11.3%, 18/159), and splicing variants (10.7%, 17/159). A total of 73 distinct variants were not previously reported (45.9%, 73/159); these included 40 missense, 8 nonsense, 18 indels, and 7 putative splicing variants. Among the 73 novel variants, 43 were Class 4 (10 P and 33 LP), 27 were Class 5 (27 P), and 3 were Class 3 (2 LP and 1 VUS) (for details, see Table S1). (c) Recurrent ABCC6 variants identified in 10 or more patients.

Figure 2. ABCC6 gene/protein structure and sequence variants found in this study.

The 3D structure of human ABCC6 protein was visualized by PyMol v.2 (Schrödinger, Inc., New York, NY). Variants in red represent those that were not previously reported. Variants in blue boxes were analyzed in functional assays. Variants with a red asterisk are those that do not meet our variant classification criteria but nonetheless included in our functional study. (a) Missense variants in the TMD0 (light blue), TMD1 (light blue), L1 (light purple), and TMD2 (light blue) domains as well as N’ and C’ terminus. Missense variants in the L0 (green), NBD1 (purple), and NBD2 (blue) domains were shown in panel b. (b) Missense variants in L0, NBD1, and NBD2 domains. (c) Different types of variants, including insertion, deletion, splicing variants, and premature termination variants, were shown in the ABCC6 gene structure consisting of 31 exons.

There were 40 patients (8.5%, 40/478) in whom no variants in any of the 29 genes could be detected (Fig. 1a). Interestingly, 14 patients (2.9%, 14/478) were found not to harbor variants in ABCC6, but instead they harbored biallelic variants in the GGCX, ENPP1, SAMD9, ENTPD1, GALNT3, and PLG genes, expanding the genetic heterogeneity in our cohort (Fig. 1a).

Functional assessment of the effect of ABCC6 synonymous and intronic variants on splicing in HepG2 cells

In silico variant predictions

The predictions for one synonymous and six intronic, putative splicing variants in ABCC6 were largely consistent between MutationTaster and Human Splicing Finder, except for c.3883–6G>A (Table S2). All variants had CADD scores less than 20, predicted not to be among the top 1% most deleterious to the human genome. ACMG/AMP and Sherloc’s classifications were not always concordant. These variants were further analyzed to allow a gene-centric visualization of the pathogenicity of variants.¹⁷ Using the ABCC6-specific CADD score of 4.327 with 95% confidence interval¹⁹ and MAF of 0.00418 as thresholds, the functionally assessed variants were classified to different prediction categories (Fig. S2). While c.999–52C>T and c.3883–24G>A variants were classified as 1-B by Sherloc, implying that they were not pathogenic, and thus initially removed from our analysis, their frequencies in our patient cohort were found to be 1,025- and 17.5-fold enriched, respectively, in comparison to the control population (Table S2). Consequently, they were included in our functional assay.

In vitro mini-gene splicing assay in HepG2 cells

The lack of consensus predictions prompted us to assess these variants in functional assays. The optimal method for determining whether a suspected variant impairs splicing is transcriptome analysis of the relevant tissues from individuals carrying the variant allele. However, this approach has significant limitations for ABCC6 due to a shortage of liver biopsies from individuals carrying these variants, as liver is the organ that predominantly expresses ABCC6.^24,25 To overcome this limitation, we analyzed the pathogenicity effects of these variants in a mini-gene splicing assay.

The splicing pattern of each variant in transfected HepG2 cells was compared with normal splicing of their corresponding WT mini-gene constructs (Fig. 3a). Specifically, RT-PCR showed products that were indistinguishable between the WT and the c.999–52C>T containing mini-genes, and sequencing results revealed the same normally spliced transcripts (Fig. 3b). In contrast to the predominant RT-PCR product from the WT allele, smaller bands were observed from the c.1868–5T>G and c.2070+5G>A harboring mini-gene constructs (Fig. 3c). Sanger sequencing revealed that these variants resulted in complete loss of exons 15 and 16 in the transcript, respectively (Fig. 3c). A synonymous variant consisting of substitution of the last nucleotide in exon 26, c.3735G>A (p.E1245E), resulted in skipping of exon 26 (Fig. 3d).

Figure 3. Illustration of mini-gene constructs and evaluation of splicing events of seven ABCC6 variants.

NTC, no template control. (a) The pCMV-3Tag-8 or pCMV-3Tag-3a was used as expression vector for mini-gene splicing assay in HepG2 cells. Amplification of the cDNA utilizes a forward primer in the first exon in the mini-gene and the reverse primer in the Flag-tag region in the vector, thus abrogates concerns of amplification of endogenous ABCC6 transcript in HepG2 cells. (b) Schematic illustration of the mini-gene structure consisting of the c.999–52C>T variant, gel electrophoresis of the RT-PCR products (arrows) and normal splicing. (c) Schematic illustration of the mini-gene structure consisting of c.1868–5T>G and c.2070+5G>A variants and gel electrophoresis of the RT-PCR products (arrows). Sanger sequencing of the products revealed that c.1868–5T>G and c.2070+5G>A caused skipping of exon 15 and 16, respectively. (d) Schematic illustration of the mini-gene structure consisting of the c.3735G>A variant (arrow, the last nucleotide in exon 26). RT-PCR showed that this variant resulted in a smaller product (arrowhead) than that of the WT mini-gene (arrow). Sanger sequencing of the products revealed that c.3735G>A caused skipping of exon 26. (e) Schematic illustration of the mini-gene structure consisting of c.3883–6G>A, c.3883–24G>A, and c.3883–46A>G variants. Two separate RT-PCR reactions were performed due to the difficulty of amplification of both large and small products in the same reaction. Small products up to 1 Kb (possibility of exon skipping) and large products up to 4 Kb (possibility of intron retention) were separated by gel electrophoresis. Sanger sequencing of the products (indicated by arrows) revealed that the transcript of the WT and c.3883–24G>A mini-gene resulted in normal splicing. In contrast, c.3883–6G>A introduced a new splice acceptor site “AG” two nucleotides before the canonical 3’ splicing acceptor site in intron 27, therefore, resulting in the addition of four nucleotides “gcAG” into the mRNA. The c.3883–46A>G variant resulted in skipping of exon 28.

The amplification products from c.3883–6G>A and c.3883–24G>A variants were indistinguishable from the WT mini-gene (Fig. 3e). However, c.3883–46A>G variant resulted in a shorter product, in addition to the customarily spliced transcript (Fig. 3e). The c.3883–6G>A variant synthesized an RNA transcript consistent with the use of an alternative 3’ splice site that led to the retention of four nucleotides “gcAG” at the intron 27-exon 28 boundary. Sequencing of the shorter amplification product from the c.3883–46A>G variant revealed skipping of exon 28 (Fig. 3e). Neither exon skipping nor intron retention was found for the c.3883–24G>A variant (Fig. 3e).

Skipping of exon 15, 16, and addition of four nucleotides at the beginning of exon 28 in the ABCC6 pre-mRNA, caused by the c.1868–5T>G, c.2070+5G>A, and c.3883–6G>A variants, respectively, are predicted to result in out-of-frame translation and generation of truncated and non-functional protein. While skipping of exon 26 and exon 28 due to c.3735G>A and c.3883–46A>G variants is predicted to result in in-frame deletion of 34 and 53 amino acids in their corresponding proteins, respectively, the proteins are predicted to be non-functional due to loss of amino acids at the essential second nucleotide-binding fold critical for the efflux function of the ABCC6 protein.⁵

Comparison of mini-gene splicing results with in silico variant predictions

Notable discrepancies were observed between in silico predictions and experimental data. Specifically, predictions were concordant with experimental results for four out of seven variants using MutationTaster and five out of seven variants using Human Splicing Finder (Table S2). MutationTaster incorrectly predicted normal splicing for c.1868–5T>G, c.3883–6G>A, and c.3883–46A>G. Similarly, Human Splicing Finder incorrectly predicted normal splicing for c.1868–5T>G and c.3883–46A>G. Spliceman predicted a modest probability, 53%, of mis-splicing for c.3735G>A, but experimental results demonstrated skipping of exon 26. The CADD scores of c.1868–5T>G, c.3883–6G>A, and c.3883–46A>G are less than 10, but experimental results revealed aberrant splicing. Variants c.999–52C>T and c.3883–24G>A, enriched in our patient cohort but classified as 1-B by Sherloc and in the MAF < 0.00418 but CADD < 4.327 group, were found not to alter splicing.

Functional assessment of ABCC6 missense variants in an Abcc6^−/− mouse model of PXE

In silico variant predictions

While the selected ABCC6 missense variants are rare with allele frequency < 0.009% in the gnomAD population genetics database, except p.R391G (0.55%), their frequencies were significantly enriched in our cohort (Table S3). Among all variants examined, p.R1138W and p.R391G had the highest allele frequency of 1.94% and 1.94%, corresponding to 451- and 3.5-fold enrichment, respectively, in comparison to their frequency in gnomAD. The aggregated predictions for all seven variants were deleterious, although discrepancies were observed when various prediction tools were used. The CADD scores of all variants were above 20, however, Sherloc classified one of them, p.R391G, as 3-U (Table S3). When variant-level predictions were made using ABCC6 gene-specific thresholds, all variants except p.T364R and p.R391G were grouped in the MAF < 0.00418 and CADD > 4.327 category (Fig. S2).

Functional characterization in an adenovirus-mediated liver-specific expression system

As ABCC6 has a specialized efflux function in the liver contributing to PPi in plasma, its concentration in the Abcc6^−/− mouse model of PXE was reduced to approximately 35% of the normal levels of WT mice.²⁶ Consistent with our prior studies,²² a single intravenous administration of 4×10⁸ IFU of recombinant adenovirus carrying WT human ABCC6 cDNA in Abcc6^−/−Rag1^−/− mice demonstrated sustained hepatic expression of human ABCC6 protein up to four weeks, in the basolateral plasma membrane of hepatocytes, the physiologic location for ABCC6 (Fig. 4). Reconstitution in the liver of human ABCC6 restored plasma PPi levels and prevented ectopic mineralization of the dermal sheath of vibrissae in the muzzle skin, an early and reliable biomarker in the overall mineralization process in these mice (Fig. 5). Utilizing this adenovirus-mediated liver-specific ABCC6 transgene expression system, we evaluated the effects of human ABCC6 missense variants on the mutant ABCC6 protein’s expression, stability, and subcellular localization in hepatocytes, followed by determination of PPi concentrations in plasma and the degree of ectopic mineralization in the muzzle skin.

Figure 4. ABCC6 protein expression in the liver of mice administered with recombinant adenoviruses containing human ABCC6 transgenes.

The Abcc6^−/−Rag1^−/− mice, at six weeks of age, were administered a single injection of 4×10⁸ IFU of adenovirus carrying either WT human ABCC6 cDNA or a missense variant. ABCC6 expression in the mouse liver was analyzed one and four weeks after injection. Immunofluorescent labeling of human ABCC6 (Green) revealed that the WT protein was expressed at one week and its level was maintained at four weeks. Dual labeling with Na,K-ATPase (Red) revealed their co-localization on the basolateral side of the plasma membrane of hepatocytes. Compared with the WT protein, p.R518Q, p.R760W, and p.R807Q resulted in reduced ABCC6 abundance and stability, with mixed plasma membrane and intracellular localization. Mutants p.T364R and p.R1138Q had exclusive intracellular expression with reduced protein expression. In contrast, p.R391G and p.G1302R had ABCC6 abundance and expression pattern similar to the WT protein. Scale bar, 200 μm (solid line) and 24 μm (dashed line). Blue, DAPI staining of nuclei. n = 6 – 8 mice per group.

Figure 5. Histopathology, plasma PPi levels, and the degree of ectopic mineralization in the Abcc6^−/−Rag1^−/− mice.

The Abcc6^−/−Rag1^−/− mice, at 6 weeks of age, were administered a single injection of 4×10⁸ IFU of adenovirus carrying either WT human ABCC6 cDNA or a missense variant. The mice were analyzed four weeks after injection, at ten weeks of age. (a) Ectopic mineralization in the dermal sheath of vibrissae was analyzed by von Kossa stains (arrows). Scale bar, 400 μm. (b) Plasma PPi levels in Abcc6^−/−Rag1^−/− mice administered with recombinant adenoviruses. (c) The calcium content in the muzzle skin containing vibrissae. The data were presented as mean ± SD. n = 6 – 8 mice per group. *P < 0.01, **P < 0.001 compared with C57BL/6J WT mice; ^#P < 0.01, ^##P < 0.001 compared with Abcc6^−/−Rag1^−/− control mice; ns, not significant. WT, wild-type; KO, Abcc6^−/− knockout.

Adenovirus injection was initiated at six weeks of age, a time point at the earliest stages of ectopic mineralization in the Abcc6^−/−Rag1^−/− mice. Mice were analyzed at one and four weeks after injection. Immunostaining of liver sections with antibodies against human ABCC6 and Na,K-ATPase, a plasma membrane marker at the basolateral side of hepatocytes, demonstrated that p.R391G and p.G1302R did not affect the targeting of the mutant proteins to the plasma membrane. Both were expressed on the basolateral side of hepatocytes with abundance similar to the WT protein (Fig. 4). The p.R391G variant restored plasma PPi levels and prevented ectopic mineralization when analyzed by von Kossa staining and by quantification of the calcium content in the muzzle skin (Fig. 5). Unlike p.R391G, p.G1302R failed to normalize plasma PPi levels and had no effects on ectopic mineralization. The five remaining variants, p.T364R, p.R1138W, p.R518Q, p.R760W, and p.R807Q, resulted in a dramatic reduction of ABCC6 abundance and stability, two fundamental properties that underlie protein function. The p.T364R and p.R1138W mutant proteins were localized exclusively into the intracellular compartment of hepatocytes, while the p.R518Q, p.R760W, and p.R807Q mutant proteins showed both plasma membrane and intracellular localization (Fig. 4). The latter five variants were barely detectable at four weeks, failed to normalize plasma PPi levels, and did not prevent ectopic mineralization in the Abcc6^−/−Rag1^−/− mice (Fig. 5). Collectively, our results showed that six of the missense variants had deleterious effects on the ABCC6 protein, including changes in its abundance, cellular trafficking, conformational stability, ability to contribute to plasma PPi levels, and their combination thereof, hence providing no protective effects on ectopic mineralization.

Comparison of in vivo experimental results with in silico predictions

Different ABCC6 missense variants have different outcomes on the functionality of the ABCC6 protein (Table S3). Variants p.R391G and p.G1302R were found to be benign and pathogenic, respectively, albeit both mutant proteins were abundantly expressed with proper plasma membrane targeting in hepatocytes. Variants p.T364R and p.R1138W were pathogenic due to defective intracellular trafficking and apparent instability of the mutant proteins. Mutants p.R518Q, p.R760W, and p.R807Q had mixed plasma membrane and cytoplasmic expression, and were not able to compensate for the loss of the endogenous mouse protein to contribute enough PPi from the liver to circulation; hence showed no protective effects on ectopic mineralization. Although p.R391G was enriched in our cohort with a high CADD score of 23.5 and classification of 3-U by Sherloc, the experimental results suggest it being a benign variant. While both p.T364R and p.R391G were classified in the same MAF > 0.00418 and CADD > 4.327 group, they were found to be pathogenic and benign variants, respectively.

DISCUSSION

The heterogeneous clinical presentation of ectopic mineralization without clearly defined molecular genetics poses a major hurdle to clinical diagnosis and management. The rise of molecular genetics has unlocked doors that continue to advance understanding of the molecular pathogenesis of this clinically heterogeneous group of disorders. Gene-targeted NGS-based panels have enabled the identification of sequence variants in clinically relevant genes. Instead of testing the candidate genes one at a time by Sanger sequencing, a group of target genes can be sequenced at the same time using the NGS method. In this study, we describe the genetics of a large cohort of PXE patients with aberrant mineralization in connective tissues. Our 29-gene panel revealed a striking degree of genetic heterogeneity in a PXE patient group, enabling us to categorize them into more precisely characterized sub-groups based on their mutated genes.

Most of the patients (88.9%, 425/478) in our cohort had variants in ABCC6. Our study disclosed a total of 159 distinct variants in ABCC6, and 73 of them were unique, expanding the ABCC6 variant repertoire. Several attempts have been made but failed to identify a clear genotype-phenotype correlation in patients with PXE.^20,21,27 No correlation could be established in our large cohort. Complicating the genetics in this group of patients is the presence of monoallelic variants in other genes complementing biallelic ABCC6 variants. In fact, dysregulated interactions between genes known to cause ectopic mineralization, and cumulative effects of ectopic mineralization-associated genes, have been encountered in patients with ectopic mineralization.^28–31 The potential interaction and possible synergistic effects of these genes may contribute to the varying degrees of ectopic mineralization in these patients and explain the lack of genotype-phenotype correlation.⁵

A total of 40 patients, among whom 29 had Phenodex scores, were negative for variants in any of the 29 genes. It is possible that these patients harbor variants in deep intronic regions, in the 5’- and 3’- untranslated regions, large deletions or gene arrangement, that are not captured by our gene panel. In this context, exome- and genome-wide NGS approaches can be used in an unbiased manner to uncover unrecognized variants and genes for ectopic mineralization. Furthermore, 14 patients were found not to have ABCC6 variants, but instead, they carried biallelic variants in GGCX (5 patients), ENPP1 (5 patients), ENTPD1 (1 patient), GALNT3 (1 patient), SAMD9 (2 patients), and PLG (1 patient). Loss-of-function variants in GGCX, ENPP1, GALNT3, and SAMD9 cause autosomal recessive PXE-like disorder with multiple coagulation factor deficiency (PXE/VKCFD1, OMIM 610842), generalized arterial calcification of infancy (GACI, OMIM 208000), hyperphosphatemic familial tumoral calcinosis (HFTC, OMIM 211900), and normophosphatemic familial tumoral calcinosis (NFTC, OMIM 610455), respectively.^32–35 Pathogenic variants in ENTPD1 and PLG cause autosomal recessive spastic paraplegia 64 and type I plasminogen deficiency, respectively, however, ectopic mineralization has not been described in these diseases.^36,37 Further studies are needed to follow up these patients for gene-specific ectopic mineralization phenotypes.

Although NGS has been revolutionary for research and clinical diagnostics, providing high throughput sequencing power at a plummeting cost, the interpretation of sequence variants remains a significant hurdle. The ambiguity and discrepancy between various bioinformatic predictions supports the recommendation that functional assessment of genetic variants should be conducted in relevant experimental systems.³⁸ In this study, we chose ABCC6 as a prime example to interrogate the reliability of bioinformatic predictions for selected variants, especially VUS. Using a mini-gene splicing assay, five out of seven ABCC6 variants analyzed, including one synonymous variant, were found to alter splicing. These results provided evidence that intronic variants away from the canonical splice sites can disrupt splicing. One limitation of the mini-gene assay was that the splicing patterns were analyzed in vitro under conditions which may not recapitulate biologically important splicing events in vivo. Moreover, the mini-genes contain only a portion of the ABCC6 pre-mRNA so that the variant being tested is not in its native context. The possibility remains that proper splicing requires signals far away from the mini-gene sequence segments, and this possibility cannot be tested in the current system.

In contrast to variants affecting primarily splicing, the consequences of missense variants can be manifold. They include alterations in multiple layers of biological complexity that are almost impossible to capture by bioinformatic predictions. The adenovirus-mediated delivery of ABCC6 transgenes to Abcc6^−/− mouse provides a functional assay to study consequences of the missense substitutions on the protein in the liver where ABCC6 is endogenously expressed. These changes can be complemented with measurements of PPi concentrations in plasma, which is dependent on functional hepatic ABCC6, and the extent of ectopic mineralization in the muzzle skin, a phenotypic hallmark of PXE in the mouse model. We found that amino acid substitutions in ABCC6 can result in the loss of function through reduced protein abundance, perturbed cellular trafficking, or putative loss of conformational stability, or any combination thereof, collectively contributing to reduced PPi plasma levels. As missense variants account for up to 60% of all pathogenic variants in ABCC6, their characterization is of paramount importance to genetic annotation. Compared to prior studies with evaluation of ABCC6 missense variants via transient expression of the ABCC6 transgene in the mouse liver through hydrodynamic tail vein injection of plasmid DNA,^39,40 the liver tropism of adenovirus vectors enables sustained high-level ABCC6 transgene expression in hepatocytes, allowing determination of steady-state plasma PPi levels and of the extent of ectopic mineralization in the mouse model of PXE with great advantage.

In conclusion, we evaluated molecular genetics and genotype-phenotype correlation in a large cohort of PXE patients referred by a patient advocacy organization, PXE International. Our findings show that bioinformatic predictions do not accurately estimate the likelihood that mis-splicing will occur or capture the multifaceted aspects of the functional changes at the protein level. Therefore, researchers should be vigilant to properly annotate these variants. Misclassification of the variants increases the risk that other variants may be erroneously regarded as pathogenic or overlooked as pathogenic. Functional characterization of variants, especially the recurring ones and VUS, in appropriate model systems, should be recognized as an essential step.

DATA AVAILABILITY

All data associated with this study are presented in the paper or Supplemental data.