Using whole exome sequencing to investigate the genetic bases of lysosomal storage diseases of unknown etiology

Abstract

Lysosomes are membrane-bound, acidic eukaryotic cellular organelles that play important roles in the degradation of macromolecules. Mutations that cause the loss of lysosomal protein function can lead to a group of disorders categorized as the lysosomal storage diseases (LSDs). Suspicion of LSD is frequently based on clinical and pathologic findings, but in some cases, the underlying genetic and biochemical defects remain unknown. Here, we performed whole exome sequencing (WES) on 14 suspected LSD cases to evaluate the feasibility of using WES for identifying causal mutations. By examining 2,157 candidate genes potentially associated with lysosomal function, we identified eight variants in five genes as candidate disease-causing variants in four individuals. These included both known and novel mutations. Variants were corroborated by targeted sequencing and, when possible, functional assays. In addition, we identified nonsense mutations in two individuals in genes that are not known to have lysosomal function. However, mutations in these genes could have resulted in phenotypes that were diagnosed as LSDs. This study demonstrates that WES can be used to identify causal mutations in suspected LSD cases. We also demonstrate cases where a confounding clinical phenotype may potentially reflect more than one lysosomal protein defect.

Introduction

Lysosomal Storage Diseases (LSDs) are inherited metabolic multisystemic disorders caused by mutations in genes encoding resident lysosomal enzymes and transporters, as well as proteins involved in lysosome biogenesis and other lysosome related functions. These deficiencies result in lysosomal accumulation of substrates and/or catabolites and cellular dysfunction (Segatori, 2014). Multiple tissues and organs can be affected, and clinical manifestations may include bone deformities, decline in vision and hearing, organomegaly (especially in spleen and liver), cardiac disease, and other symptoms (Scriver, 2001). Most of these diseases are neuronopathic, or have neuronopathic forms, with prominent manifestation in the central nervous system.

LSDs are mostly monogenic and more than 50 genetically distinct forms have been identified (Lubke, et al., 2009), with some individual diseases having similar clinical presentations. Although individually LSDs are rare, with incidences ranging from 1:57,000 (Gaucher disease) to 1:4,200,000 (sialidosis) (Meikle, et al., 1999), collectively they have an overall prevalence about 1 in 5,000-7,700 live births (Meikle, et al., 1999; Sanderson, et al., 2006). This is likely an underestimate as cases may be undiagnosed or misdiagnosed when they present with an atypical clinical phenotype of delayed onset and attenuated progression due to partial loss-of-function mutations (Maire, 2001). In addition, there may be defects in lysosomal proteins that are not currently associated with human disease or in proteins that are not presently assigned to the lysosome (Lubke, et al., 2009; Sleat, et al., 2007). Thus, while most LSD cases are well characterized, there are patients with disease of unknown etiology that appear to be lysosomal in origin based on clinical and/or histopathological presentation.

Biochemical and genetic analyses are the classic strategies to study the etiology of LSDs (Scriver, 2001). However, these analyses have limited application in identifying defective genes in LSDs of unknown etiology because they focus on individual proteins and can only address known lysosomal disease genes. Proteomics can provide a global and unbiased approach to identify known or novel lysosomal proteins that may be defective (Jadot, et al., 2017; Sleat, et al., 2009). However, while comparative proteomic methods can help identify candidate genes in which mutations result in loss of the encoded protein, they are not effective in cases where deleterious variants result in inactive but stable mutants.

As sequencing costs decrease, whole exome sequencing (WES) is increasingly used in human genetic studies since its successful application in identifying the genetic cause of Miller Syndrome (Ng, et al., 2010). It is a cost-effective method for identifying disease-causal variants, with WES covering nearly all protein-coding regions but only requiring ~5% of the throughput needed to sequence the whole human genome. In this study, we performed WES in 14 patients diagnosed with LSD of unknown genetic etiology and evaluated this approach to identify known and novel LSDs mutations on an exome-wide scale.

Materials and Methods

Sample Information

Cell lines from 14 unrelated patients were obtained from several sources (Table 1). These patients displayed a spectrum of phenotypes suggestive of LSD, including neurodegeneration and cellular storage, but no underlying cause had been identified. Research protocols involving human subjects were approved by the Institutional Review Board of the University of Medicine and Dentistry of New Jersey.

Table 1.

Source, phenotype and sequencing coverage of the 14 LSD samples.

Sample ID	Clinical Information	Total Bases	Total SNVs	Exonic SNVs	Mean Coverage
00RD098	LSD cases from Netherlands - no confirmed diagnosis.	2,037,480,379	72,708	20,297	22.5
01RD492		2,179,163,363	72,137	20,641	24.1
02RD297		2,023,856,437	70,917	20,280	22.4
82RD265		2,421,322,830	73,687	20,784	26.7
95RD414		1,584,030,773	63,827	19,471	17.5
99RD299		1,941,261,757	79,121	22,162	21.4
B1278	pycnodystosis-like, Cathepsin K positive.	1,891,984,968	68,282	19,820	20.9
CABMHF1	Patient alive at 33 years. Infiltration of bone marrow by storage histiocytes (no evidence for malignancy), thrombocytopenia and splenomegaly; biochemical tests for Gaucher and Niemann-Pick diseases were negative.	1,996,568,992	69,931	20,041	22.1
CABMHF2	Patient alive at 14 years. Clinical progression and ultrastructure suggestive of late-onset/attenuated form of NPC but biochemical tests were negative.	1,652,477,376	69,397	19,882	18.3
CABMHF3	Patient died at 12 years. Neuronal storage resembling zebra bodies but sphingolipidoses, oligosaccharidoses, glycoproteinoses, and mucopolysaccharidoses excluded by undefined enzyme assays.	1,835,520,789	70,128	19,962	20.3
CABMHF4	Patient alive at 7 years. Neurodegeneration (severe central dys- or demyelination; epilepsy, tetraparesis) and dystrophy.	2,304,968,208	78,903	22,097	25.5
CABMHF5	Patient alive at 22 years. Severe neurodegeneration, mild dysmorphia and dysostosis. Storage bodies reminiscent of neuronal ceroid lipofuscinosis and mucolipidosis IV found in skin biopsy.	1,306,118,808	66,253	18,899	14.4
HL508Pa	Adult neuronal ceroid lipofuscinosis.	2,108,762,505	71,124	20,384	23.3
TC983077	Metaphyseal acroscyphodysplasia.	2,007,500,640	69,655	20,178	22.2

Exome Sequencing and Reads Mapping

Whole exome sequencing was performed on SOLiD platform (Life Technologies, Grand Island, NY) in 50x25 bp format. Exomes were enriched with SureSelect DNA - Human All Exon 50Mb Kit (Agilent, Santa Clara, CA). Raw sequences were aligned to the human reference genome (version hg 19) using LifeScope ™ Genomic Analysis Software (Thermo Fisher Scientific, Waltham, MA).

Variant Calling and Annotation

Three variant calling methods were used for variant discovery: Genome Analysis Tool Kit version 2.8-1 (GATK) (McKenna, et al., 2010), CLC Genomic Workbench (CLC) (Qiagen, Redwood City, CA) and StrandNGS (v2.3, Strand Genomics, San Francisco, CA). For the most part, variant discovery by the GATK genotyping method followed previously described recommendations (DePristo, et al., 2011). Briefly, raw sequence alignments (in binary alignment map (BAM) format) were reordered based on chromosomal coordinates, sorted, and indexed with the Picard toolset (version 1.80) (http://picard.sourceforge.net), and Samtools-0.1.19 (Li, et al., 2009). A series of GATK alignment-processing procedures were then applied to revise the alignments: indel realignment, remove duplication, and base recalibration. Multi-individual joint genotype calling was then performed on all individuals with GATK UnifiedGenotyper to generate the raw genotype calls in a single variant calling format file. Single nucleotide variants (SNVs) located within the targeted exome regions were selected based on the target-region definition provided by the SureSelect DNA - Human All Exon 50Mb Kit. Quality scores of SNVs were then recalibrated with VariantRecalibrator using recommended parameters for GATK (DePristo, et al., 2011). Variant discovery by CLC and Strand NGS were performed using default parameters as recommended by manufacturers. CLC variants with Depth of Coverage (DP) less than 10 were removed. Variant coverage was calculated by a GATK tool DepthofCoverage.

Variants were annotated using ANNOVAR (Wang, et al., 2010) with the following command: table_annovar.pl -buildver hg19 inputfile annovar/humandb/ -out outfile-remove -protocol refGene,cytoBand,genomicSuperDups,esp6500siv2_all,snp138,ljb26_all, 1000g2015aug_all,exac03,exac03nontcga,exac03nonpsych,clinvar_20160302 -operation g,r,r,f,f,f,f,f,f,f,f -nastring .-csvout

Candidate Gene Selection

A list of 2,157 candidate genes that are potentially associated with lysosome function and/or disease was compiled (Supp. Table S1). This list was based on multiple sources including: a) proteomic analyses of lysosomes and lysosomal proteins (Bagshaw, et al., 2005; Chapel, et al., 2013; Schroder, et al., 2007), b) genes involved in lysosomal related pathways (Di Fruscio, et al., 2015), c) proteins transcribed by TFEB, a master gene for lysosomal biogenesis (Palmieri, et al., 2011), d) glycoproteins that contain the mannose 6-phosphate lysosomal targeting modification (Sleat, et al., 2008; Sleat, et al., 2013), and e) global protein subcellular localization studies (Jadot, et al., 2017).

Candidate Variant Selection

Based on the ANNOVAR annotation, nonsynonymous variants and loss-of-function (LoF) variants, including stop-gain variants and canonical splicing site variants, were selected from the total SNV call sets for the 2,157 candidate genes (Supp. Table S1). For all other genes, only LoF variants were selected. The selected variants were then filtered based on their allele frequency in the population. Variants with allele frequencies ≥1% in Non-Finnish European (NFE) in the ExAC database (http://exac.broadinstitute.org/) and/or ≥5% in the 1000 Genomes Project Phase 3 dataset (1000 Genomes Project Consortium, et al., 2015) were removed. An aggregate score system (Sleat, et al., 2016) based on the results from eight functional prediction programs provided by ANNOVAR was applied to prioritize remaining nonsynonymous variants. All candidate variants in each individual are listed in the Supp. Data File.

Sanger Sequencing Validation

Candidate variants were evaluated by Sanger sequencing. Primers (Supp. Table S2) were designed with Primer3 (Steve Rozen, 1998) for PCR amplification of the genomic region covering the variants sites. The PCR amplicon containing the candidate variant was sequenced directly for amplicons less than 700 bp in size. Amplicons larger than 700 bp were first subcloned with Zero Blunt TOPO kit (Thermo Fisher Scientific, Waltham, MA) and individual clones were sequenced to determine the haplotype of heterozygous variants. Purified PCR products or plasmid DNAs were sequenced by ABI 3730 DNA Sequencer at GenScript (Piscataway, NJ). Electrophoregrams were examined with BioEdit Sequence Alignment Editor (Hall, 1999). Primers and PCR methods used for gender determination were as described previously (Hedges, et al., 2003). All novel variants that were validated by Sanger sequencing have been submitted to dbSNP and their Submitted SNP numbers (ss#) are listed in Table 3.

Table 3.

Candidate pathogenic variants.

Case	Gene	Chr	Position	Nucleotide change	Amino acid change	Methods	ExAC NFE AF	Score	Genotype	Allele pathogenicity
LSD Candidate Genes
00RD098	GLB1	3	33099692	rs72555366:G>A	NP_000395:p.Arg208Cys	G, C, S	4.52E-05	6	Het	known pathogenic
82RD265	SLC31A1	9	116021039	ss2137536899:C>T	NP_001850:p.Arg90Ter	G, S	na	1	Het	not known
82RD265	SLC31A1	9	116022721	ss3023056067:G>T	NP_001850:p.Val181Leu	C, S	na	−4	Het	not known
95RD414	GLA	X	100653420	rs28935490:C>A	NP_000160:p.Asp313Tyr	G, C, S	0.0044	5.5	Het	known pathogenic
95RD414	SMPD1	11	6413175	rs120074128:C>A	NP_000534:p.Gln294Lys	G, C	3.00E-05	8	Homo	known pathogenic
CABMHF1	NPC1	18	21136367	rs373751051:C>T	NP_000262:p.Arg389His	G, C	1.50E-05	5.5	Het	possible pathogenic
CABMHF1	NPC1	18	21140411	rs55680026:T>C	NP_000262:p.Asn222Ser	S	0.0048	0	Het	possible polymorphism or late-onset
CABMHF1	SMPD1	11	6415259	rs144873307:G>S	NP_000534:p.Gly492Ser	G, C, S	0.0013	−6	Het	possible pathogenic
LoF variants in other genes
00RD098	KRIT1	7	91870373	ss3023056068:G>A	NP_001013424:p.Gln66Ter	G, C, S	8.24E-06	na	Het	known pathogenic
CABMHF3	TBCK	4	107183332	ss3023056069:G>A	NP_001156907:p.Gln102Ter	G, C, S	na	na	Homo	not known

Methods: G: GATK, C: CLC workbench, S: Strand NGS; Score: na: not available. Validation: Het: heterozygous; Homo: homozygous.

Functional assays

Enzyme assays were conducted essentially as described previously using the following substrates: beta-galactosidase and methylumbelliferyl-β-galactoside (Sleat, et al., 1996); tripeptidyl peptidase 1 and Ala-Ala-Phe-aminomethylcoumarin using the endpoint assay (Sohar, et al., 2000). Acid sphingomyelinase was analyzed using BODIPY-labeled C12 sphingomyelin in triplicates as previously described (He, et al., 2001). Cholesterol esterification measuring formation of [3H]cholesterol oleate from [3H]oleate and analysis of lysosomal cholesterol accumulation using filipin staining were conducted as described (Kruth, et al., 1986; Pentchev, et al., 1986).

Results

Exome Sequencing and Variant Discovery Pipelines

We performed WES on 14 unrelated patients suspected to have potential LSDs (Table 1). The average exome coverage for each individual varied from 14.4x to 26.7x, with a mean of 21.5x (Table 1). We implemented three different variant calling methods (Materials and Methods). Given that there are differences in the variants identified by each method, we performed downstream analyses on all three data sets to reduce method bias and make full use of all variant calling information. Because the 14 patients are genetically unrelated and their clinical phenotypes are diverse, it is unlikely that they share the same disease-causing mutation. Therefore, we examined each sample independently using the screening process outlined in Figure 1.

Figure 1:

Overall analysis flow for identifying pathogenic LSD variants.

Candidate Pathogenic Variants in LSD Candidate Genes

First, we focused on a curated set of 2,157 candidate genes that are potentially associated with lysosome function (Supp. Table S1, see Materials and Methods for candidate gene selection details). Variants in these genes were extracted from the total SNV call sets and filtered as described in Figure 1. Briefly, nonsynonymous and LoF SNVs within the candidate genes were first extracted. Selected variants were then filtered based on the population minor allele frequency. Lastly, an aggregate score system (Sleat, et al., 2016) was applied to identify variants with potential functional impact. After filtering, ~ 90 variants were retained for each sample (Table 2). We then compared these variants to known pathogenic LSD variants to identify candidate variants.

Table 2.

Screening statistics for LSD candidate variants.

Individual ID	Nonsynonymous variants in 2157 candidate genes	Variants after AF filtering *	Variants after aggregate score filtering
00RD098	980	179	63
01RD492	993	220	102
02RD297	1023	257	101
82RD265	993	176	78
95RD414	912	130	57
99RD299	1041	249	102
B1278	1000	228	104
CABMHF1	963	175	90
CABMHF2	974	185	92
CABMHF3	979	195	100
CABMHF4	1148	304	164
CABMHF5	890	136	53
HL508Pa	1005	188	74
TC983077	905	167	73
Average	986	199	90

^*ExAC Non-Finnish European: AF < 0.01; 1000 Genomes: AF < 0.05

From the filtered variant sets, we identified eight candidate pathogenic variants in five genes in four individuals (Table 3). Sanger sequencing validated seven as heterozygous and one as homozygous. Two examples are shown in Supp. Figure S1. To investigate whether these variants had been previously reported to be pathogenic, we searched the Human Gene Mutation Database (HGMD) (Stenson, et al., 2003). Three variants were previously reported to be pathogenic and three variants were predicted to be possible pathogenic. We also identified two novel mutations in the gene SLC31A1 (MIM# 603085), which encodes a protein that is potentially related to lysosomal function but which has not previously been associated with human disease (Table 3).

1). 00RD098

We found one previously reported mutation in the GLB1 (MIM# 611458) gene encoding lysosomal β-galactosidase (GLB1). Defects in GLB1 cause GM1 Gangliosidosis. The patient was heterozygous for an allele encoding NP_000395:p.Arg208Cys , which previously has been reported to be pathogenic (Boustany, et al., 1993). However, we found reduced but considerable β-galactosidase activity in 00RD098 fibroblasts (~13,000 emission/ug protein in 00RD098 vs ~21,000 -35,000 emission/ug protein in controls, Figure 2). p.Arg208Cys was previously shown to result in loss of β-galactosidase activity against synthetic substrates (Boustany, et al., 1993). One possibility is that this patient is compound heterozygous for p.Arg208Cys and another mutation which retains activity against synthetic but not natural substrates. However, we observed no other missense variant to support this hypothesis. Therefore, we conclude that this patient is a carrier for the p.Arg208Cys GLB1 mutation but diagnosis of GM1 gangliosidosis is uncertain.

Figure 2: Enzyme assays.

Panel A, β-galactosidase was measured as a candidate defect in 00RD098 and in two controls (UNC2 and WJ5253A). Panel B, Tripeptidyl peptidase 1 (another lysosomal protein) was measured as a positive control. Error bars indicate the standard error of the mean (SEM).

2). 95RD414

We found a known pathogenic variant that results in NP_000534:p.Gln294Lys (Pavlu and Elleder, 1997) in the gene encoding acid sphingomyelinase (SMPD1, MIM# 607608) in 95RD414 and it was confirmed to be homozygous. Note that this allele has been previously referred to as p.Gln292Lys(Pavlu and Elleder, 1997). Acid sphingomyelinase converts sphingomyelin to ceramide and mutations in the SMPD1 result in Niemann-Pick type A and B diseases. Enzyme assay showed the acid sphingomyelinase activity was greatly reduced in this patient, supporting the genetic analysis (Figure 3). Filipin staining and cholesterol esterification assays ruled out NPC disease.

Figure 3: Acid sphingomyelinase assay.

Acid sphingomyelinase was measured as candidate defect in 95RD414 and in three controls (CABMHF3, a case with no detected pathogenic changes in NPC1 and GM23248 and GM23249, fibroblasts from two apparently normal individuals). Cholesterol esterification assays and filipin staining on all four lines were within control values. Error bars indicate SEM.

Of note, this female patient (Supp. Figure S2) was also heterozygous for a previously reported NP_000160:p.Asp313Tyr disease allele (Eng, et al., 1993) in gene GLA (MIM# 300644) that encodes lysosomal alpha-galactosidase (GLA). GLA defects are associated with Fabry disease. While some heterozygotes have clinical manifestations of this X-linked disease, there is some controversy regarding the clinical relevance of p.Asp313Tyr (Niemann, et al., 2013). However, it is possible that this GLA variant contributed to the clinical phenotype, and the patient manifests a blended phenotype based on mutations in both SMPD1 and GLA. The blended phenotype can obscure the original diagnosis in this case. Increased urine globotriaosylceramide level would lend more evidence to the pathogenicity of this mutation.

3). CABMHF1

CABMHF1 was heterozygous for NP_000262:p.Asn222Ser in NPC1 (MIM# 607623), the Niemann-Pick type C disease gene (Park, et al., 2003), and a novel variant in NPC1 resulting in NP_000262:p.Arg389His. The pathogenicity of the p.Asn222Ser variant is unclear – it has been reported to be a polymorphism, with an allele frequency inconsistent with NPC1 but there also exists evidence that it may be associated with late onset forms of disease (Wassif, et al., 2016). p.Arg389His has not previously been reported in Niemann-Pick type C, but it is a very promising candidate as there are two known pathogenic mutations at this position resulting in different amino acid substitutions: p.Arg389Leu (Fancello, et al., 2009) and p.Arg389Cys (Park, et al., 2003). Sequencing of cloned NPC1 gene DNA fragments containing the two variant loci showed that the two mutant variants are located on different copies of chromosome 18 (Supp. Figure S3). Thus, the patient is compound heterozygous for these variants. We also identified a heterozygous variant in the SMPD1 gene: NP_000534:p.Gly492Ser (Table 3). Two patients with mild Niemann–Pick B disease phenotype and no acid sphingomyelinase activity were compound heterozygous for p.Gly492Ser and another SMPD1 mutation (Irun, et al., 2013). In addition, there are known pathogenic mutations very close to it: p.Thr488Ala (Rodriguez-Pascau, et al., 2009), and p.Tyr490Asn (Simonaro, et al., 2002). Similar to 95RD414, this patient could have a disease phenotype that resulted from synergistic heterozygosity of both NPC1 and SMPD1 mutations. Unfortunately, cell stocks from this case were no longer viable at the time of this study and thus we could not conduct functional validation.

4). 82RD265

We found two novel variants in case 82RD265 in the lysosome-functional related gene SLC31A1, which encodes solute carrier family 31 (copper transporters) member 1: nonsense mutation NP_001850:p.Arg90Ter and a non-synonymous NP_001850:p.Val181Leu mutation. SLC31A1 was targeted for analysis based on its dual localization to both the lysosome and to the plasma membrane, and we have reported previously on the variants in this patient (Jadot, et al., 2017). While p.Arg90Ter is a clear null, further functional analysis to test the cellular copper uptake are necessary to demonstrate the effect of these two variants together.

Loss-of-function Variants Outside of LSD Candidate Genes

We only identified potential disease-causing mutations in our lysosomal candidate list in four cases. Several reasons could account for the low identification rate, and one possibility is that these cases were misdiagnosed as LSDs. While LSDs have characteristic clinical phenotypes, many are not unique to this class of disorders. To test the possibility that disease may result from mutations encoding non-lysosomal proteins, we further examined all LoF variants (i.e., stop-gain and splicing variants) in the whole exome (Supp. Data File). In two cases, we identified potential pathogenic variants.

1). 00RD098

While we identified promising candidates in this case (see above), functional validation experiments failed to verify lysosomal defects. However, we identified and validated a heterozygous stop-gain variant NP_001013424:p.Gln66Ter in the Krev interaction trapped protein 1 (KRIT1) gene (MIM# 604214). The variant was detected in one individual as heterozygous among 60,660 sequenced individuals in the ExAC database (http://exac.broadinstitute.org/variant/7-91870373-G-A). Homozygous mouse KRIT1 mutants die prenatally due to cerebral cavernous malformations (Whitehead, et al., 2004). A previous study also reported a human family with this p.Gln66Ter variant (Gianfrancesco, et al., 2007). Family members carrying this variant exhibited a high variability in the penetrance and phenotypes (Gianfrancesco, et al., 2007). Given 00RD098 carries one GLB1 variant and showed reduced β-galactosidase activity, this KRIT1 mutation and associated nervous system angiomas may have contributed to a blend phenotype that leads to the LSD diagnosis.

2). CABMHF3

In case CABMHF3, we identified a homozygous stop-gain variant NP_001156907:p.Gln102Ter in the TBC1 Domain Containing Kinase (TBCK) gene (MIM# 616899). The mutation is not observed in the ExAC database. CABMHF3 was diagnosed as a potential LSD case by the presence of neuronal storage. Interestingly, in a previous study of patients who are homozygous for a 4-bp deletion in TBCK at amino acid 205, the affected siblings (two boys and one girl) presented with profound hypotonia, global developmental delay, and slow motor development with no progress beyond the ability to sit independently (Guerreiro, et al., 2016). They also had epilepsy and similar distinctive facial features, and the two youngest siblings had signs of precocious puberty. The older sibling died at 9 years of age, and the youngest died at 12 years (Guerreiro, et al., 2016). There was no report of analysis of neuronal pathology in these cases so it is unclear whether TBCK defects were associated with neuronal storage, but the clinical presentation appears to be consistent with the diagnosis of LSD. In another recent study of nine unrelated families with LoF mutations in TBCK, the patients presented with intellectual disability and hypotonia, and their suspected diagnoses included LSD (Bhoj, et al., 2016). In addition, TBCK was implicated in the mTOR pathway (Liu, et al., 2013), which regulates lysosomal biogenesis and function (Kinghorn, et al., 2017). Therefore, although subcellular localization analysis (Jadot, et al., 2017) indicates that TBCK does not reside within the lysosome, these studies suggest that TBCK may control some aspects of lysosomal function and could result in phenotypes in CABMHF3.

Discussion

In this study, we explored the feasibility of identifying pathogenic alleles using WES in 14 patients with disease that was suspected to be lysosomal in origin. In an initial targeted approach that focused on proteins that have lysosomal location or function, we identified probable disease-causing mutations in three cases in genes encoding proteins that are known to be associated with lysosomal disease. In another case, we found novel mutations in the gene encoding SLC31A1 – a protein that may have lysosomal function but has not previously been associated with human disease.

Several studies have also applied next generation sequencing technology to identify LSD pathogenic variants, although these studies have focused on lysosomal proteins in their sequencing surveys. For example, one study used a panel of 57 lysosomal genes to identify potential causal mutations in ~40% (26/66) of the cases (Fernandez-Marmiesse, et al., 2014). Another recent study used a gene panel that includes 891 genes involved in autophagy-lysosomal pathways (Di Fruscio, et al., 2015). A diagnosis rate of ~60% (29/48) was reported for patients with the clinical phenotype of neuronal ceroid lipofuscinosis (Di Fruscio, et al., 2015). Because we started from WES, we were able to examine LoF variants outside of the lysosomal gene list. Using this unbiased approach, we identified potentially pathogenic variants in two cases in genes that encode proteins that do not appear to have lysosomal function. These results support the use of WES or whole genome sequencing as a genetic diagnostic method for suspected LSD cases. Under a clinical setting, pedigree analysis would provide further weight to assignment of potential pathogenic alleles.

We only identified potential causal mutations in gene encoding lysosomal proteins in four cases and several reasons may account for this. First, our study was based on the hypothesis that the diseases were caused by monogenic coding variants. If the diseases were caused by non-coding variants (e.g., intronic or promoter variants), WES will be likely to miss the causal variant. Second, although we applied several bioinformatic methods for variant identification, sensitivity is limited by the technology: with variability in exome coverage, we do not have 100% sensitivity for all coding variants. Thirdly, the diagnosis of a likely LSD case based on clinical criteria may not be correct. This may be illustrated by the two cases that appear to have pathogenic variants in genes encoding non-lysosomal proteins, KRIT1 and TBCK. Another limitation of our study is that because we analyzed historical archived cases, we cannot conduct additional phenotyping on the probands or obtain more detailed clinical and family histories. Nevertheless, our results demonstrated the feasibility of using WES to identify potential causal mutations and can potentially benefit LSD diagnosis.

It is worth noting that in several cases we not only identified probable mutations (homozygous or compound heterozygous) in genes encoding lysosomal proteins that are known to be associated with human diseases, but we also find heterozygosity for pathogenic variants in genes encoding other lysosomal proteins. These observations raise the intriguing possibility that clinical presentation in these cases may reflect synergistic effects of multigenic defects, which is possibly why initial diagnosis was complicated. For example, case 95RD414 was homozygous for a known pathogenic allele in SMPD1 (p.Gln294Lys), and this was validated by functional assay, but we also identified a known pathogenic allele in GLA (p.Asp313Tyr), which may produce disease manifestations in this female patient, depending on the pattern of lyonization of this X-linked gene. Case CABMHF1 was compound heterozygous for a possible pathogenic allele in NPC1 (p.Arg389His), as well as another allele of unclear significance (p.Asn222Ser), but was also heterozygous for a known mutation in SMPD1 (p.Gly492Ser). It is possible that the combination of mutations in NPC1 and SMPD1 deficiency manifested in a complex presentation. These cases may be examples of synergistic heterozygosity and blended phenotypes, where disease was resulted from multiple partial defects in one or more metabolic pathways (Li, et al., 2016; Vockley, et al., 2000). Case 00RD098 was heterozygous for a known pathogenic allele in GLB1 and while we did not identify a marked deficiency of this enzyme, it is possible that this complicated the phenotype of an as yet undiscovered mutation in another gene. The concept that haploinsufficiency for a lysosomal protein may influence the phenotype of a genetically distinct defect is not without precedent: for example, glucocerebrosidase defects as a risk factor for Parkinson disease are well established (reviewed in (Aflaki, et al., 2017)).

Conclusion

Using WES data from 14 patients suspected to have LSDs, we identified both known pathogenic variants and novel potential pathogenic variants in five genes in four cases. In addition, in two cases we identified LoF variants in genes that are not thought to be associated with LSDs, although mutations in these genes could have resulted in phenotypes that were diagnosed as LSDs. This study demonstrates that WES can be used to study the genetic bases of LSDs with unknown etiology and has potential as an approach for genetic diagnosis of LSDs. In addition, we identify cases where a confounding clinical phenotype may potentially reflect more than one lysosomal protein defect. This possibility should be borne in mind in apparent cases of lysosomal disease that defy traditional methods of diagnosis.