Assessment of genes involved in lysosomal diseases using the ClinGen Clinical Validity framework

2.1. The ClinGen Lysosomal Diseases Gene Curation Expert Panel

The ClinGen LD GCEP (https://clinicalgenome.org/affiliation/40110/) was assembled by the ClinGen Inborn Errors of Metabolism Clinical Domain Working Group in 2021 to assess the clinical validity of gene-disease relationships for genes involved in lysosomal diseases. This group currently includes 18 individuals from different institutions within the USA and Canada. Seven members of the GCEP are biocurators with training on the ClinGen gene-disease clinical validity framework, and assigned to identify, collect, and assess relevant data from published literature and public databases (such as gnomAD); several of these individuals also have clinical, research, and/or laboratory experience in LDs. Nine members of the GCEP are experts with significant clinical, research, and/or laboratory expertise in LDs who are assigned to review the data collected by the biocurators, to discuss and questions around those findings, and to come to a consensus on the clinical validity classification of a gene-disease pair. Two individuals, one with significant experience in metabolic disorders, coordinate the activities of the group.

2.2. List of genes for curation

The main workflow is shown in Fig 1. We began by reviewing the LDN list, which consisted of 56 genes (https://lysosomaldiseasenetwork.org/official-list-of-lysosomal-diseases/). Because LDs are typically multi-systemic conditions, we were aware that some of the gene-disease pairs on the list may already have been curated by other ClinGen GCEPs that focus on phenotypes that are included in LDs such as neurodevelopmental delay (Intellectual Disability-Autism GCEP). Therefore, we surveyed the ClinGen website to determine which of these genes had already been curated. The curated data and clinical validity classifications of genes already curated by other ClinGen GCEPs were reviewed by the LD GCEP to ensure that we agreed with the assessment of the data and final clinical validity classification. For genes that did not already have a curated disease entity published on the ClinGen website, we used the ClinGen Gene Tracker to determine if any other GCEPs planned to curate any of the genes and then contacted those GCEPs to discuss priorities and possible collaboration.

• Download figure
• Open in new tab

Figure 1.

LD GCEP gene curation workflow. Beginning with the LDN “Official List of Lysosomal Diseases’’ (n=56), we found 37 gene-disease pairs that had not been curated and applied the ClinGen Gene-Disease Validity Framework to assess their clinical validity, a process which included: pre-curation and Lumping/Splitting of any genes associated with multiple diseases (n=12; Table 1); full curation of the gene-disease entity; and expert panel review of the curation. Final gene-disease curations are published to the ClinGen website (https://search.clinicalgenome.org/kb/gene-validity).

2.3. Curation process

Clinical validity curations for all genes on the LDN list were performed according to the ClinGen Gene Disease Clinical Validity framework¹⁰. All data was entered into ClinGen’s online Gene Curation Interface (GCI). In brief, the ClinGen Clinical Validity framework is a semi-quantitative system that awards points for specific pieces of case-level (genetic) and experimental evidence that support a gene-disease relationship. Once curation of the evidence is complete, the scored evidence is summed to yield a total point score and corresponding clinical validity classification. Definitive and strong clinical validity classifications are achieved by 12 points or more. In addition, definitive classifications must have case-level evidence replicated over a time of three years or more. Moderate and limited classifications are met for 8-12 and 0.1-6 points respectively. No known disease relationship is used as the clinical validity descriptor if there has never been an assertion that variants in the gene are associated with human disease.

Previously asserted gene-disease relationships can also be disputed or refuted based on the degree of conflicting evidence. In addition, as gene-disease validity classifications may change over time (e.g., due to new evidence being published since the original curation), we recommend that readers check the ClinGen website (www.clinicalgenome.org) for the most up- to-date information. Further details and the current Standard Operating Procedure (SOP) are available on the ClinGen website (https://clinicalgenome.org/curation-activities/gene-disease-validity/) and provided in Supplemental file 1)

In the LD GCEP, each curation was performed by a single biocurator with expertise in the ClinGen clinical validity framework. To begin a curation within the GCI, biocurators must first select a gene from the assigned list (using HGNC nomenclature), disease entity for that gene (denoted by a Mondo ID), and inheritance pattern for curation. The ClinGen Lumping and Splitting guidelines¹¹ were used to determine the appropriate disease entity, or entities, for curation. For genes assessed by the LD GCEP, all data used to decide whether a disease for a specific gene should be “lumped” together for curation, or “split” into different curations were discussed during the GCEP’s regular online meetings. After collecting and scoring relevant evidence, biocurators assigned a provisional classification for each gene-disease relationship that was then reviewed by experts in the group. Evidence for gene-disease relationships with a provisional clinical validity classification less than definitive or curations for which a biocurator had questions were reviewed on a group conference call during which the biocurator would present all curated data and scoring to the group. While a quorum of 2 was required, typically, at least 5 experts, with experience ranging from clinical, research, and molecular diagnostics, were present on each call and would come to a consensus on the final clinical validity classification. Definitive provisional classifications were reviewed offline by at least two experts. Once approved by two or more experts, the clinical validity classification and associated data for each curation were published to the ClinGen website (www.clinicalgenome.org), where they are openly accessible.

3.1. Genes for curation

Based upon review of the ClinGen website, 17 genes on the LDN list had already been curated by other ClinGen GCEPs. This was not surprising considering that many LDs include phenotypes under the purview of other GCEPs, such as neurodevelopmental delay (Intellectual Disability-Autism (ID-Autism) GCEP). Members of LD GCEP reviewed the curated data and clinical validity classifications for these disease-disease pairs and agreed with the final classifications. Of note, 5 of these genes had previously been curated for the General GCEP by an individual with significant experience in lysosomal diseases who is now a biocurator for the LD GCEP. Of the remaining 39 genes, 30 were on other GCEPs lists for curation; 28 of those genes were not considered to be high priority for curation by those other GCEPs and were, therefore, transferred to the LD GCEP’s list. The remaining two genes (CTNS, CLCN5) were high priority for the Tubulopathy and Glomerulopathy GCEPs, given that renal disease is the primary feature associated with these genes. These genes were therefore curated by the Tubulopathy and Glomerulopathy GCEPs (Figure 1). One of the LD GCEP biocurators is also a member of these GCEPs, thus ensuring consistency in the evaluation of the data.

3.2. Determining the disease entity for curation

Of the 37 LDN genes on our list, 12 genes were associated with more than one condition in OMIM, and one gene (SLC38A9) had no disease entity listed in OMIM. We followed the ClinGen lumping and splitting guidelines to determine the appropriate disease entity, or entities, for those genes¹¹ (Table 1). We chose to lump together any disease entities that represented different severities and ages of onset for similar phenotypic characteristics and therefore represented a clinical spectrum rather than distinct disease entities. For example, the disease entities Niemann-Pick disease, type A, and Niemann-Pick disease, type B, were lumped as acid sphingomyelinase deficiency (ASMD) because these two disorders represent a clinical spectrum with the same mechanism of disease (loss of function), and the inheritance pattern for both conditions is the same (autosomal recessive). When necessary, we requested a new Mondo disease identifier via the Mondo GitHub (https://github.com/monarch-initiative/mondo/issues) for lumped disease entities. For example, acid sphingomyelinase deficiency was not previously listed in Mondo. This term has since been added based on our request (MONDO:0100464), with the child terms, already present in Mondo, of Niemann-Pick disease type A (MONDO:0009756), and Niemann-Pick disease type B (MONDO:0011871). While there is utility in using more specific disease descriptors when discussing phenotypic features, prognosis, and treatment with patients and families, ClinGen recommends lumping disease entities across a phenotypic spectrum, for the purposes of gene-disease clinical validity assessment, when there is no clear indication to split diseases into separate entities. Lumping across a clinical spectrum is common for metabolic disorders, most of which are recessive with phenotype at least partly dependent on the impact of a combination of two alleles, and where genotype/phenotype correlation is not always known.

For four genes curated by the LD GCEP (GLB1, HGSNAT, NAGLU, PSAP), there were substantial differences in the phenotypes of their respective disease entities, and thus they were split for curation(Table 1). For example, after detailed discussions with members of the Retina GCEP, we split the disease entities for HGSNAT into mucopolysaccharidosis type IIIC (MPS IIIC; MIM# 252930) and retinitis pigmentosa 73 (MIM# 616544). The LD GCEP proceeded to curate HGSNAT for MPS IIIC, and the Retina GCEP has curated HGSNAT for retinitis pigmentosa 73 (Table 1). We plan to revisit the HGSNAT data again every 2 years to ensure that no additional data has emerged that may lead us to lump these conditions together, such as reports of additional features developing in MPS IIIC in patients originally reported to have isolated retinopathy,

3.3. Clinical validity classifications: LDN list

Of the 37 LDN genes on our list, 33 were given a clinical validity classification of definitive for at least one disease entity (Figure 2, Supplemental Table 1). For the 19 LDN genes curated by other GCEPs, all were definitive for at least one disease entity (Supplemental Table 2). Therefore, in total, 52/56 (92.9%) genes on the LDN list were definitive for one or more disease entities.

• Download figure
• Open in new tab

Figure 2.

ClinGen Gene-Disease Clinical Validity evidence points for genes on the LDN list curated by the LD GCEP. 37 genes on the LDN list, that had not already been curated by other ClinGen GCEPs, underwent gene-disease clinical validity classification. 33 of those genes were curated for a single disease entity, while 4 were curated for split disease entities (Table 1, Supplemental Table 1). The additional 19 genes on the LDN list have been classified by other ClinGen GCEPs; all are definitive for at least one disease entity (Supplemental Table 2).

Including the genes with split disease entities, the LD GCEP curated 41 gene disease-relationships for the LDN list (Table 1, Supplemental Table 1). This includes 4 diseases for PSAP, and 2 for GLB1. In total, 36 of the 41 gene-disease relationships curated were definitive (Supplemental Table 1). For all but one of the definitive gene-disease relationships, the genetic evidence met or exceeded 12 points, which is the maximum points that can be counted for genetic evidence. The one definitive gene-disease relationship which did not have maximum genetic evidence was NAGA-alpha-N-acetylgalactosaminidase deficiency (MONDO:0017779). At the time of curation, nine probands had been reported in the literature; eight with biallelic missense variants and one homozygous for a nonsense variant. The total genetic evidence score was 8.85 points. The addition of experimental evidence (biochemical function and non-human animal model) brought the classification to definitive (11.85 points, rounded to 12; with approval of “definitive” by the GCEP members). Of interest, variants in NAGA have been associated with Schindler disease, type I (MIM# 609241; NAGA deficiency type I), Schindler disease, type III (MIM# 609241; NAGA deficiency type III), and Kanzaki disease (MIM# 609242; also known as Schindler disease, type II; NAGA deficiency type II). However, clinically unaffected family members have been reported with the same genotype and enzyme deficiency^12,13. The variability in clinical phenotype from asymptomatic to neurological manifestations does not diminish the definitive relationship between NAGA and “alpha-N-acetylgalactosaminidase deficiency“; all reported individuals with biallelic variants in the gene had the enzyme deficiency; other factors, such as genetic modifiers, have been suggested to influence the genotype-phenotype correlation¹³. However, while the relationship between variants in NAGA and the biochemical phenotype of alpha-N-acetylgalactosaminidase deficiency is definitive, the clinical impact of alpha-N-acetylgalactosaminidase deficiency is unclear at the current time.

All the definitive gene-disease relationships on the LDN list that were curated by the LD GCEP (n=36) had supportive experimental evidence. This included evidence indicating that the biochemical function of the gene product is consistent with phenotypic features in patients with the disease, and all had non-human animal models that recapitulate features of the disease. Additional evidence curated for some genes included protein interaction, for those genes encoding subunits of a multi-subunit complex or enzyme; rescue in humans for genes causing conditions for which enzyme replacement therapy is available; and rescue experiments in animal models or cultured cells. In total 34 (94%) of the definitive gene disease relationship met or exceeded the 6-point maximum for experimental evidence and the remaining 2 (6%) were given 4 points.

Four of the 37 LDN genes on our list were found to not be definitively associated with their respective LD: ATG7 (strong), HYAL1 (moderate), ATG5 (limited), and SLC38A9 (no known disease relationship) (Figure 2; Supplemental Table 1). ATG7 had a strong classification for the associated disease entity, spinocerebellar ataxia, autosomal recessive 31, a neurodevelopmental disorder with multisystem involvement and impaired autophagy¹⁴. While ATG7 was described in the LDN list as a “non-disease protein,” biallelic ATG7 variants were reported in 5 families with features of complex neurodevelopmental disorder with brain, muscle, and endocrine involvement¹⁴, yielding 8.2 points for genetic evidence. This genetic evidence, along with experimental evidence including the biochemical function of the protein¹⁵ and a mouse model recapitulating the human disease phenotype¹⁶, led to a classification of strong (total 14.2 points). Although the total score exceeded the 12-point cutoff for a definitive gene-disease association, the ClinGen curation guidelines¹⁰ require replication of the genetic evidence (i.e., further reported cases) over 3 or more years in order for the classification to be definitive. This requirement was not met for ATG7. HYAL1 was curated for mucopolysaccharidosis type IX (MONDO:0011093). At the time of curation, there were only 2 published probands, one of whom had two affected siblings^17–19, yielding 4.6 points of genetic evidence. Experimental evidence (2.5 points) included the expression pattern of the gene¹⁸ and an animal model²⁰. A total of 7.1 points was thus given, yielding a moderate classification. ATG5 was curated for spinocerebellar ataxia, autosomal recessive 25. Only one family, with two affected siblings, had been published at the time of the curation, leading to 0.5 points being given for genetic evidence²¹. The gene-disease relationship was supported by the biochemical function of the gene product²² and a drosophila model in which the phenotype was rescued by the wild type gene²¹ (6 points experimental evidence). Based on the level of supportive evidence, the gene disease relationship was classified as limited. While the points awarded summed to 6.5, borderline between limited (0.1-6 points) and moderate (7-11 points), the GCEP members agreed on a limited classification for this gene-disease relationship based on a limited number of reported cases. SLC38A9 was given a classification of “no known disease relationship” because, based on our literature review, there were no reports describing individuals with a clinical phenotype caused by variants in the gene. Based on published experimental data, there is evidence suggesting a role for SLC38A9 in lysosomal disease. SLC38A9 encodes “solute carrier family 38, member 9”, a lysosomal amino acid transporter protein involved in regulating the mechanistic target of rapamycin complex 1 (mTORC1)^23–25. SLC38A9 physically interacts with NPC intracellular cholesterol transporter 1²⁵, a protein encoded by the NPC1 gene, which is definitively associated with Niemann-Pick disease type C1. HEK293T cells harboring SLC38A9 variants lacking key amino acid residues have impaired amino acid-mediated signaling of mTORC1²⁴; and abnormal amino acid metabolism was observed in a SLC38A9 knockout zebrafish model²⁶. While there is currently “no known disease relationship” for SLC38A9, ClinGen adds the additional tag of “animal model only” to reflect this evidence.

3.4. ClinVar classifications based on clinical validity

The importance of the clinical validity classification is reflected in the classifications of variants submitted to ClinVar (data accessed March 2, 2024) (Supplemental Table 3). This is consistent with current recommendations²⁸ that genes with a clinical validity classification of limited or below, in general, should not be included on diagnostic panels or reported from exome or genome analyses. ATG7, has relatively few variants reported in ClinVar (48 total variants, 6 pathogenic), which may reflect that the first cases were in 2021 (spinocerebellar ataxia, autosomal recessive 31). While HYAL1 was found to have moderate clinical validity, cases were first reported in 1999¹⁸; accordingly, there are 349 total variants, including 34 pathogenic and 2 likely pathogenic, in ClinVar. ATG5 and SLC38A9, which were respectively curated as limited and no human disease relationship, have no pathogenic or likely pathogenic variants submitted to ClinVar by clinical diagnostic laboratories, indicating the uncertainty of the clinical significance of variants in these genes based on limited or no clinical validity evidence (Supplementary Table 3). Based on ClinGen’s recommendations for recuration (https://clinicalgenome.org/site/assets/files/2164/clingen_standard_gene-disease_validity_recuration_procedures_v1.pdf), the gene-disease relationships that do not have a definitive classification will be re-evaluated in the future in order to look for and evaluate any new evidence that could impact the classification.

3.5. Challenges and key takeaways

Molecular genetic testing has the capacity to expedite diagnosis of LDs, supporting early therapeutic interventions and improved outcomes for individuals and their families^3,5. With NGS-based multigene panels or genome-wide sequencing increasingly used as first-line techniques, genetic testing for LD involves analyzing many genes, which may vary in the strength to which they are associated with their respective LD. To accurately interpret the pathogenicity of the genetic variants identified, one must first assess the clinical validity of the relationship between the given gene with its disease entity⁸.

Here, we applied a systematic framework integrating genetic and experimental evidence¹⁰ to classify the clinical validity of the 56 proposed LD genes from the LDN. 52 of these 56 (92.9%) genes were definitively associated with at least one LD, supporting their inclusion in the genetic evaluation of suspected LD. Of the remaining 4 genes, two had limited (ATG5) or no evidence (SLC38A9) for association with LD; in ClinVar, these genes had a high burden of VUS, highlighting the analytic challenges posed by genes with uncertain clinical validity and the need for cautious interpretation if including them in diagnostic genetic evaluation.

Our experience highlights several major themes and associated challenges for classifying the clinical validity of proposed LD genes. Twelve of the 37 LDN genes curated by the LD GCEP were associated with multiple disease entities, and thereby required precuration to determine whether these disease entities should be lumped together within a single disease spectrum, or split as separate conditions (Table 1). As LDs frequently have clinically heterogeneous and variably severe presentations³, it can be difficult to distinguish whether a given set of disease entities fall within a single disease spectrum or instead are distinct disorders. To address this challenge, we applied the ClinGen Lumping/Splitting heuristic to assess the overlap in disease mechanism and clinical manifestations; for eight genes, there was sufficient overlap to lump the disease entities, whereas for the remaining four genes, the molecular genetics and/or clinical features of the disease entities differed enough to merit them being split. As illustrated by the example of acid sphingomyelinase deficiency, our work on lumping and splitting altered the categorization of the associated disorders in MONDO, emphasizing how systematic curation can provide new insight regarding gene-disease relationships. With the advent of LD patient registries²⁷, we anticipate that the understanding of the phenotypic spectrum of LDs will continue to grow, thereby informing lumping/splitting for LDs.

Classifying the clinical validity of a proposed LD gene requires evaluating both the genetic and experimental evidence supporting their association. As the majority of LDN genes had been first reported in association with their LD several decades ago, there was robust genetic and experimental evidence available, conferring a definitive classification. However, for several of the more recently reported genes, there were fewer published cases available, contributing to non-definitive classifications with their LDs. Given the rarity of LDs, it can be difficult to interpret whether a relative dearth of published cases argues against the gene-disease association or instead reflects the fact that individuals with the given genetic disease are exceedingly rare. In addition to continued reexamination of the published literature to identify newly reported cases, data-sharing initiatives such as Matchmaker Exchange²⁸ can help address this issue via connecting individuals with similar phenotypes who have variants in a given candidate gene. Moreover, for our initial curation efforts, we focused on genes that were on the LDN list. This approach may have biased our curations towards definitive and strong gene-disease relationships.