Challenging the traditional approach for interpreting genetic variants: Lessons from Fabry disease

Abstract Fabry disease (FD) is an X‐linked genetic disease due to pathogenic variants in GLA. The phenotype varies depending on the GLA variant, alpha‐galactosidase residual activity, patient's age and gender and, for females, X chromosome inactivation. Over 1000 variants have been identified, many through screening protocols more susceptible to disclose non‐pathogenic variants or variants of unknown significance (VUS). This, together with the non‐specificity of some FD symptoms, challenges physicians attempting to interpret GLA variants. The traditional way to interpreting pathogenicity is based on a combined approach using allele frequencies, genomic databases, global and disease‐specific clinical databases, and in silico tools proposed by the American College of Medical Genetics and Genomics. Here, a panel of FD specialists convened to study how expertise may compare with the traditional approach. Several GLA VUS, highly controversial in the literature (p.Ser126Gly, p.Ala143Thr, p.Asp313Tyr), were re‐analyzed through reviews of patients' charts. The same was done for pathogenic GLA variants with some specificities. Our data suggest that input of geneticists and physicians with wide expertise in disease phenotypes, prevalence, inheritance, biomarkers, alleles frequencies, disease‐specific databases, and literature greatly contribute to a more accurate interpretation of the pathogenicity of variants, bringing a significant additional value over the traditional approach.


| INTRODUCTION
A deficiency leads to progressive accumulation of its undegraded substrates globotriaosylceramide (Gb 3 ) and its deacylated derivative globotriaosylsphingosine (lyso-Gb 3 ), in tissues and body fluids. 1 FD has two main forms, "classic" and "later-onset." 1,2 "Classic" FD typically manifests during childhood, progressing to lifethreatening renal, cardiac, and/or neurological complications, with hampered quality of life and reduced life expectancy. 1,2 "Later-onset" FD typically has cardiac manifestations. 3,4 FD phenotypes primarily depend upon the disease form, nature of the GLA variant, age, gender and residual α-Gal A activity. 1 High-throughput next generation sequencing (NGS)-based screening programs in high-risk populations and newborns have identified several novel GLA variants. Understanding variant pathogenicity is the key to accurate prevalence estimation, diagnosis, and management of FD. Revised interpretation of GLA variants, previously inaptly classified "pathogenic," has decreased the estimates of FD prevalence (Table 1). [5][6][7][8] Over 1000 GLA variants have been reported, [9][10][11][12] most of them private, including some variants of unknown significance (VUS). 13 Furthermore, the non-specificity of FD symptoms presents a challenge for physicians attempting to interpret the clinical relevance of a VUS.
A panel of French physicians involved in the diagnosis and management of FD (clinical and molecular geneticists, nephrologists, and internal medicine specialists), convened to discuss the diagnostic challenges posed by GLA VUS. This paper illustrates the challenges in interpretating pathogenicity of GLA VUS through real-life cases, highlights the input of highly specialized experts and provide practical recommendations for accurate diagnosis of FD.

| TERMINOLOGY
In FD, the term "variant" had been historically used to describe the two main forms/phenotypes, creating confusion between genetic and clinical levels when the Human Genome Variation Society (HGVS) advised to replace "mutation" by the term "variant." 14 We therefore recommend keeping the term "variant" to describe GLA alleles and using "form" or "phenotype" for clinical description of the disease (i.e., classic or later-onset form/phenotype of FD). The terms "mutation" and "polymorphism" should be avoided, due to implied pathogenicity and benignity, respectively. Variants are categorized as benign (class 1), likely benign (class 2), of uncertain significance (VUS, class 3), likely pathogenic (class 4), or pathogenic (class 5). 13 3 | PHENOTYPES AND UNDERLYING BIOCHEMICAL MECHANISMS

| Clinical phenotypes
The "classic" phenotype typically manifests during childhood, 15 progressing in adulthood to life-threatening complications, hampered quality of life and reduced life expectancy ( Figure 1). 1,2 The lateronset FD phenotype mostly exhibits cardiac manifestations, which appear later in life. 1,3,4,16 Absent/very low residual α-Gal A activity is associated with the more severe classic phenotype, earlier onset of symptoms, and multiorgan involvement. 1,17,18 However, as many genetic diseases, FD has a highly variable phenotypic expressivity.

| Biochemical phenotypes
Alpha-Gal A activity can be measured in leukocytes, plasma, or dried blood spot (DBS) cards. Reference values vary widely depending on pre-analytical and technical conditions. The activity of another lysosomal enzyme, e.g., β-galactosidase or hexosaminidase, should be T A B L E 1 Revised prevalence of Fabry disease in various screening populations after re-interpretation of pathogenicity simultaneously determined for quality control. Alpha-Gal A deficiency is expressed in micromole/L/h. 19,20 In males, substrate accumulation starts once α-Gal A activity drops below 20%-25% of normal levels. 1 In severe deficiency, α-Gal A levels are <1%-3% of mean control values. Although low α-Gal A activity in females is a good indicator of pathogenicity, 21 its diagnostic value in females is diminished due to normal α-Gal A activity in several heterozygous females with confirmed FD, depending on X-chromosome inactivation (XCI). 21 Globotriaosylsphingosine or lyso-Gb 3 (α-D-galactopyranosyl- Increased plasma lyso-Gb 3 levels in male FD patients were first reported in 2008. 22 This amphipathic lipid, found at high concentration in plasma of male FD patients, has been attributed to the action of acid ceramidase on Gb 3 . Lyso-Gb 3 has progressively superseded Gb 3 as the main biomarker for diagnosing and monitoring FD.

| Is lyso-Gb 3 causally involved in the pathogenesis of FD?
Increasing evidence implicates lyso-Gb 3 in the pathogenesis of FD, possibly through inflammation. 23,24 The addition of lyso-Gb 3 to cultured smooth muscle cells induces cell proliferation, suggesting its role in the vascular remodeling characteristic of FD. 20 The inhibitory effect of lyso-Gb 3 on endothelial nitric oxide synthase may also contribute to vascular dysfunction in FD patients. 25 Lyso-Gb 3 may induce glomerular damage through activation of Notch1 signaling and TGF-β1-mediated production of extracellular matrix by podocytes. 26 Recent studies suggest an activation of innate immunity and possibly of adaptive immunity by lyso-Gb 3 in target tissues including myocardium and kidney. In a recently proposed model of cardiac FD staging, myocardial inflammation can also precede the effects of Gb 3 storage on cardiac wall thickness and function. 27

| Can lyso-Gb 3 be used as a biomarker in FD?
Plasma lyso-Gb 3 levels are dramatically increased in male patients with classic FD (up to 100-fold control values). 22,[28][29][30][31][32] Concentrations in males with a mild phenotype and in heterozygous classic females are lower, but nonetheless abnormal. [28][29][30][31][32] In male FD patients, lyso-Gb 3 levels are higher in those with frameshift and nonsense variants than in those with missense variants. 33  patients with an uncertain diagnosis of FD carrying a VUS, while normal plasma lyso-Gb 3 levels were observed in individuals with negative tissue biopsies. 30 Multiple studies have investigated the association between plasma lyso-Gb 3 levels and clinical FD manifestations. In both male and female FD patients, a correlation between plasma lyso-Gb 3 exposure and disease severity was reported. 29 Plasma lyso-Gb 3 concentration was identified as an independent risk factor for cerebrovascular white matter lesions in male patients and left ventricular hypertrophy in female patients. 29 A study of patients with genetic variants associated with classic FD found a correlation between lyso-Gb 3 concentrations and left ventricular mass index, but not kidney function. 36 In later-onset FD patients carrying the c.644A>G; p.(Asn215Ser)/p.
N215S variant, lyso-Gb 3 levels correlated with left ventricular mass, glomerular filtration rate, and overall disease severity. 16 Finally, a correlation between highly elevated plasma lyso-Gb 3 levels and severe clinical events has been recently reported. 37 Urinary lyso-Gb 3 levels are also increased in FD patients, and correlate with Gb 3 levels. 38 Like urinary Gb 3 concentration, total urinary concentration of lyso-Gb 3 and its structural analogs represent a specific diagnostic biomarker, and is elevated in both classical and non-classical FD patients. 39 3.2.3 | Does lyso-Gb3 have a prognostic value in therapeutic drugs monitoring in FD patients?
In hemizygous FD patients, enzyme replacement therapy (ERT) markedly reduces (though does not always normalize) plasma lyso-Gb 3 levels within 3 months, thereafter, remaining relatively stable. 40,41 Similar reduction or stabilization of lyso-Gb 3 levels was reported for female FD patients. 41 Later treatment initiation results in less effective lyso-Gb 3 clearance. In male patients with classic FD, lyso-Gb 3 levels a year after ERT initiation were lower in patients who began ERT before age 25, versus those who started later in life. 42 An early initiation of ERT in two yet clinically asymptomatic children with classic FD completely normalized lyso-Gb 3 . 43 In FD patients, treatment with oral chaperone migalastat for 1-2 years led to reduced, but not normalized, plasma lyso-Gb 3 levels. 44,45 Regular monitoring of biochemical response to chaperone therapy (e.g., plasma lyso-Gb 3 and α-Gal A activity) is crucial to adjust treatment approach, and patient clinical control at least every 6-12 months. 46 In summary, state-ofthe-art knowledge indicates that lyso-Gb 3 is a useful biomarker both for diagnosing and monitoring FD patients.

| Additional factors influencing phenotype in Fabry disease
Multiple factors may contribute to the phenotypic heterogeneity associated with a given GLA variant. 47 Theoretically, modifier genes may alter the expression of GLA or another disease-causing gene influencing FD phenotype. 48 As in other genetic diseases, environmental factors (e.g., hypertension, smoking) may also contribute to phenotypic variability.
The clinical presentation is more variable in heterozygous females than hemizygous males, likely due to epigenetics. 1,21,49 The inactivation of one of the two X-chromosomes is generally random in females, though some females have preferential inactivation of the same chromosome in >80% of cells (skewed XCI), affecting the clinical phenotype and prognosis of FD. 21  Some GLA variants have been classified as VUS, partly because many of them are private to one or few families and because ACMG-AMP classification is difficult to implement in routine practice. Interpreting variants is particularly complex in the later-onset phenotype, which may mimic isolated HCM and lack other FD features. 47 The pathogenicity of several missense GLA variants has been extensively discussed in the literature. However, for some of them it remains unclear, requiring insights from the experts for an accurate interpretation.
Hereafter, eight cases illustrating the challenges in interpretation of GLA variants' pathogenicity, and the insights from field experts, are presented. Written informed consent was obtained for all cases.  The high residual enzyme activity associated with p.(Ala143Thr) is in favor of a pseudo-deficiency rather than a disease-causing allele per se. 35,56 Of note, the dbFGP database, which includes inputs from FD experts and researchers, clearly states p.(Ala143Thr) as benign (Table 3)  GLA carrier, was already considered for FD-specific treatment in Italy.
Extensive investigations did not reveal any sign or symptom of FD; α-Gal A levels were 57% of normal in two independent assays. Brain MRI, glomerular filtration rate and lyso-Gb 3 were normal and FD was consequently excluded.
p.(Ser126Gly) has been found in 74 cases (exomes frequency of 0.033%, n = 60; genomes frequency of 0.063%, n = 14) in GnomAD, above the global frequency of FD (Table 3). While several publications claimed p.(Ser126Gly) to be pathogenic when associated with a given haplotype (though no convincing clinical evidence was provided), 63

| Databases and in-silico prediction softwares
Genetic databases and in silico prediction softwares are useful resources (  The Japanese Fabry Database LO [5]; classic [4] Table 3).
While in silico prediction softwares (e.g., PolyPhen-2, PROVEAN, SIFT, Mutation Taster) can help in determining the pathogenicity of missense variants, their partial reliability is well-known as confirmed with/for the five most common (likely) benign GLA variants (Table 3).
With respect to databases, those specific for FD and curated by FD experts, appear to be most valuable for classifying GLA variant pathogenicity ( Table 2).   Figure 3), which may be extended to some other genetic conditions.

| Insights from disease specific biomarkers
In In the case of FD, where a "pharmacogenetic" specific treatment exists, it should be noted that "amenability" of a given GLA variant does not necessarily imply pathogenicity, since the available amenability (Asp313Tyr)) do not meet the >1% frequency threshold.
When considering the pathogenicity of a GLA variant, if the variant's AF is higher than 0.0125% (FD prevalence: 1/8000) and higher than the frequency of the most common pathogenic allele (indicated as the "Popmax Filtering AF" in GnomAD), the variant should be considered as "likely benign." 13 Once pathogenicity of a missense variant has been verified, the next step is to determine its association with a specific phenotype.
Whether the "classic" or "later-onset" phenotype of FD is more prevalent is still unclear; several arguments exist for both hypotheses. The total number of genetic variants responsible for classic FD overwhelms the number of genetic variants associated with later-onset FD ( Figure 2). 69 However, a few later-onset variants have been found at relatively higher frequency. Later-onset phenotype with single-organ impact and lower morbidity has minimal effects on genetic fitness, with undetected but higher segregation within populations and consequent higher prevalence. 51 Familial segregation analysis of a GLA VUS can be performed for further information. Absence of co-segregation strongly suggests benignity while presence of co-segregation is in favor pathogenicity. 70 However, identifying relatives who can effectively contribute to pathogenicity classification of a VUS, is not always feasible, especially when no male relative exists, thereby limiting the relevance of α-Gal A assay. 71

| Periodically review the literature and specific databases
It is important to periodically review the literature and perform a specific database search for additional evidence on pathogenicity/ benignity of variants, since with growing evidence over time, some variants originally designated VUS may be reclassified as either likely pathogenic/pathogenic or in contrast likely benign/benign and in the latter case, no longer considered to be FD-causing variants. 35,54,58,59,62,72 However, careful consideration is necessary when using the variants classification from these databases, as they allow

| CONCLUSION
Since the publication of ACMG-AMP, 13 the approach to interpreting unknown/novel GLA variants integrates information from in vitro prediction softwares, AF and gene databases. 13 However, this traditional approach is not universally applicable for rare diseases. We recommend that physicians should be aware of potential errors in interpreting GLA variants. Misdiagnosis prolongs patient diagnostic odyssey, resulting in higher morbidity and adding unnecessary financial burden to the healthcare system. 69 Accurate interpretation benefits from the input of rare disease expert clinicians and geneticists (Table 4; Figure 3). We call for more data sharing by referral centers worldwide to increase the robustness of disease specific databases.
These lessons obtained from Fabry disease contribute, with practical concepts, to better interpret the pathogenicity of allelic variants in genetic diseases for a more accurate diagnosis and effective management.

ACKNOWLEDGMENTS
The authors received editorial/writing support in the preparation of this manuscript that was founded by Sanofi Genzyme, but no payment for writing this publication. The authors are responsible for the con- speaker's honoraria from Amgen.

AUTHOR CONTRIBUTIONS
Each author participated in the meetings, analyzing literature, revising, and interpreting the presented cases with current data, establishing recommendations and approved the manuscript. Dominique P. Germain designed and supervised the study, substantially contributed to developing the manuscript and wrote the revised version.

DATA AVAILABILITY STATEMENT
Data are not shared due to patient confidentiality and ethical restrictions.