Hereditary spastic paraplegias (HSPs) are a genetically heterogeneous group of neurodegenerative disorders with a prevalence of ~3-9/100,000 in most populations and a clinical hallmark of progressive lower limb weakness and spasticity (1, 2). HSPs result from genetic alterations resulting in dysfunction of the long axons in the corticospinal tract and posterior columns (3). Consistent with the cardinal clinical features of HSP, the primary pathological mechanism is distal axonal degeneration in a “dying-back” manner (4). HSPs are broadly categorized into pure and complicated forms based on the presence of additional clinical features such as ataxia, parkinsonism, peripheral neuropathy, cognitive dysfunction, cataracts, and icthyosis (3). HSPs segregate in several modes of inheritance, including autosomal dominant, autosomal recessive, X-linked, or mitochondrial (3). Although there is no evidence that HSP overall is more prevalent in one ethnic group over another, HSP does show ethnic differences in many of the mutated genes (Table 1).

As with many other Mendelian diseases, the introduction of next-generation sequencing (NGS) revolutionized the genetic diagnosis of HSPs with over 76 genomic loci and 58 corresponding genes (5). The autosomal dominant (AD) HSPs lead mostly to the pure form of disease and are linked to 19 spastic gait (SPG) genes. The most common AD genes are SPG4/SPAST, SPG3A/ATL1, SPG31/REEP1, and SPG10/KIF5A. Complicated HSPs more frequently occur in autosomal recessive (AR) families and are linked to 57 loci and 52 genes. The most common AR genes are SPG11/KIAA1840, SPG5A/CYP7B1, SPG7, and SPG15/ZFYVE26. Though rare, X-linked and mitochondrial inheritance are also observed. In the past 5 years alone >15 novel HSP and HSP-related genes have been reported. However, the number of families identified in the initial and follow-on papers is typically low; for example, the original REEP2 publication had 2 families with a follow-on of a single family (6, 7). Furthermore, a large number of recent SPG loci are supported by a single publication only (Table 1). This has led to the concern that we are in an asymptotic situation where, even with many new genes, the diagnostic yield may not get close to 100%. This potential gap in heritability is observed in other rare disorders as well and might be referred to as “dark matter” of clinical genomics. This review will focus on potential causes and efforts to overcome these challenges.

It is usually assumed that HSP is caused by Mendelian mechanisms and that eventually nearly all patients will receive a single-gene diagnosis. However, this is not necessarily true. Related disorders illustrate a diverse situation: inherited neuropathies (Charcot-Marie-Tooth disease, CMT) are also highly heterogeneous Mendelian disorders whereas amyotrophic lateral sclerosis (ALS) is largely not explained by Mendelian genes. The proportion of Mendelian genes is even lower in late onset neurodegenerative diseases, such as Parkinson and Alzheimer disease.

The reported diagnostic yield for exome sequencing in the general clinical setting ranges from 25 to 50% (8, 9). In hereditary spastic paraplegias, Schüle et al. identified a molecular diagnosis in 46% of families (240/519 families), despite extensive whole-exome sequencing efforts, with low success in simplex families (28%) (2). Similarly, a 20% diagnostic yield was reported in a cohort of 98 previously unsolved HSP families analyzed by a custom sequencing panel of 70 HSP and HSP-related genes (10).

While the genetics of certain neurodegenerative diseases are deemed “complex,” the field, thus far, has not applied similar models of inheritance to HSP. We believe that it is too early to tell whether non-Mendelian effects have a major contribution to HSP. There are a number of valid pro-Mendelian hypotheses as to how to close the diagnostic gap, and the coming years will allow us to test these ideas (Figure 1). These include non-coding regions of the genome, unorthodox types of mutations (such as repeat expansions) and digenic inheritance models. However, the search for risk genes and alleles will likely contribute to the understanding of HSP in multiple ways, from exploring oligogenic causation, gene/environment interactions, and phenotype modifying genes; thus, expanding the inheritance models in HSP.

Systematic reanalysis of unresolved clinical exomes can reveal causative variants that were not prioritized in the initial analysis and increase the diagnostic yield; however, the search should ideally be expanded beyond mutations in the protein-coding regions (11). Full exploration of the non-coding space will require whole-genome sequencing; however, fortunately, the untranslated regions and regions adjacent to exons are typically covered by whole-exome sequencing. Though sparse annotation poses considerable challenges in interpreting these variants, Minnerop et al. found deep intronic mutations in POLR3A to be a frequent cause of HSP and cerebellar ataxia. After identifying intronic variants in a single recessive spastic ataxia family, the authors screened a cohort of 618 cases. They found that compound heterozygous POLR3A mutations accounted for ~3.1% of genetically unclassified autosomal recessive and sporadic index cases. Over 80% of these cases shared the same deep-intronic mutation which activates a cryptic splice site. This study nicely demonstrates the potential held within the non-coding genome.

As the limitation of non-coding coverage is overcome by whole-genome sequencing, the clinical and functional interpretation of variants will remain a challenge. Recently, the complementation of genetic sequencing with transcriptome sequencing (RNA-seq) has successfully improved the diagnostic yield in Mendelian disorders (12, 13). RNA-seq adds a functional layer to help prioritize genes and variants by providing information about aberrant splicing, abnormal gene expression levels, and extreme cases of mono-allelic expression (13). Since gene expression and mRNA isoforms vary widely across tissues, sequencing disease-relevant tissue is critical to the potential value with this approach (12). Due to the difficulties in obtaining the disease-relevant tissue, transcriptome sequencing may have limited potential in HSP. However, the recent mapping of the axonal transcriptome from induced pluripotent stem cell derived motor neurons may provide an avenue to overcoming this challenge (14).

Another contributor of genome variability that could help resolve the diagnostic gap is structural variation, including copy number variations, translocations, and inversions (15). Copy number variations (CNVs), in particular, are known to play an important role in HSPs, causing 8%-41% of SPG4 due to the Alu genomic architecture of SPAST (16–19). Screening of approximately 600 independent HSP cases over a 5-year period in ATL1, SPAST, NIPA1, SPG7, and REEP1 has detected numerous deletions (20). Recently, a complex homozygous 4-kb deletion/20-bp insertion that removes the last two exons and part of the 3′UTR was identified in DSTYK (21). These studies exemplify the relevance of structural variation in HSPs. However, major challenges still exist that hinder full exploration of these mutations. Since CNVs are also a large source of normal variation and their pathogenic potential can depend on genetic ancestry and environmental factors, determining whether a CNV is benign or pathogenic remains a considerable clinical challenge (15, 22). Additionally, CNV detection from whole-exome sequencing remains immature and unstable, as indicated by the low concordance between CNV variant callers (23). Another popular method for CNV detection is high-resolution microarray, which is commonly used in clinical cytogenetics where sequencing based analysis can overwhelm a typical laboratory's workflow (24). Though research laboratories largely rely on NGS, the additional high-resolution microarrays may lead to diagnoses in cases with suspected CNV. Improved sequencing technology and structural variation interpretation and detection will resolve more HSP families in the future.

Standard clinical genomic analysis focuses on typical modes of inheritance, such as autosomal dominant, autosomal recessive, and X-linked, while unusual inheritance modes are often ignored. Inclusion of these “genomic” mechanisms in analysis pipelines, for example genomic imprinting, repeat expansions, and uniparental disomy, can lead to successful identification of overlooked molecular diagnoses.

Uniparental isodisomy is particularly relevant to HSPs as this mechanism has been observed in two autosomal recessive spastic ataxia of Charlevoix-Saguenay (ARSACS) cases, one SPG18 case, and four SPG35 cases (25–27). Uniparental isodisomy is the inheritance of both chromosomes from the same parent, resulting from errors in meiosis and/or mitosis (28). Isodisomic events generate regions of homozygosity in the offspring, allowing for the inheritance of a homozygous variant from one heterozygous carrier-parent (29). During trios analysis, these homozygous variants are usually discarded as sequencing errors since only one parent is a carrier. The occurrence of uniparental isodisomy can be inferred from very long regions of homozygosity that are isolated to a single chromosome; therefore, uniparental isodisomy detection can be incorporated into a standard whole-exome pipeline by modifying existing methods for long regions of homozygosity (30, 31). The high number of SPG35 cases resulting from UPD suggests that this may be a frequent genetic mechanism for FA2H homozygous mutations in non-consanguineous families (27). Uniparental isodisomy is important to detect because of its impact on molecular diagnosis and recurrence risk in families.

The awareness of the genetic overlap between HSPs and spinocerebellar ataxias (SCAs) has increased as the number of loci causing both diseases expands; however, the overlap of HSP and SCA caused by triplet-repeat expansions is less emphasized (32). Bettencourt et al. reviewed the literature of triplet-repeat expansions mimicking spastic paraplegias: expansions have been observed at the ATXN1, ATXN2, ATXN3, ATXN7, ATN1, and FXN loci (32). The authors recommend incorporating triplet-repeat ataxia analysis into diagnostic algorithms, especially ATXN3 in autosomal dominant complicated HSP and FXN in autosomal recessive or sporadic spastic paraplegias (32). It is conceivable that additional, yet to be identified, coding and non-coding repeat expansion loci cause HSP.

Based on the assumption of fully penetrant alleles, traditional Mendelian disease analysis focuses on the rare DNA variation that segregates within a family. However, these locus-specific family studies treat Mendelian traits as distinct entities and disregard a more comprehensive genetic model for human disease in which variants of varying effect size as well as environmental influences contribute to disease (33). The challenge is the unexpected large amount of variation in the human genome on a population level, where >99% of all variants show a minor allele frequency of < 1% (34). Since many of these variants are without phenotypic consequence, some certainly are very harmful, and a considerable number must have effect sizes that are below the threshold of a Mendelian gene but contribute significantly to phenotypic expression. Identification of strong effect sizes in the background of mostly minor effects is the next big challenge in human genetics. Recent method developments in statistical genetics allow for unbiased genome-wide screens for non-Mendelian alleles, and surprisingly, are able to re-identify bona fide Mendelian genes as well (35). The application to HSP genomics will eventually generate a more complete genetic architecture of the disease.

Though the high amount of locus heterogeneity present in HSPs complicates clinical diagnosis, it does provide an opportunity to study the overarching biological pathways through analysis of molecular networks. Genes and their products form complex networks within cells that are governed by specific laws and principles (66). These complex networks model the non-linear genotype-phenotype relationships observed in HSPs, such as incomplete penetrance and epistasis, that deviate from the “one gene, one phenotype” principle (66, 67). Functionally related proteins interact with each other to accomplish similar biological mechanisms, thus forming cellular pathways (66). Network medicine capitalizes on these interactions and hypothesizes that perturbation of a single gene product will propagate along the entire network (68). The interactions between a set of disease-causing genes can be summarized into a disease module for further study (69). A disease module can be analyzed for differences in complete loss of gene products vs. interaction-specific perturbations, global relationships to other human diseases, novel candidate disease genes, and emerging biological pathways (68, 70–73). Novarino et al. combined exome sequencing with network analysis to summarize a global view of HSP (74). A HSPome was created from previously published HSP seed genes and candidate genes from whole exome sequencing. From the HSPome, they extracted subnetworks of functionally related proteins that form pathological modules, including ER-associated degradation, endosomal and membrane-trafficking, and purine nucleotide metabolism. Three candidate genes arose from the HSPome that were found to be mutated in HSP patients. Lastly, the authors discovered the HSP seed genes were significantly overlapping with amyotrophic lateral sclerosis, Alzheimer's disease, and Parkinson's disease, while no overlap was observed between HSP and neurodevelopmental disorders nor non-neurological disorders (74). Recently, the relationship between the inherited axonopathies, HSP and Charcot-Marie-Tooth 2 (CMT2), was explored through a network analysis of protein-protein interactions. The HSP disease module was found to significantly overlap both the CMT2 and hereditary ataxia modules. Pathway analysis revealed ribosomal protein and viral infection response pathways (75). With the rapid pace of gene discovery in HSPs, network analysis will continue to be a powerful approach for deciphering the complex interactions underlying the phenotype.

Historically, many movement disorders, including HSPs, have been clinicogenetically classified based on the predominant phenotype of the first gene locus (76). These classification systems have similar shortcomings, including erroneously assigned loci, duplicated loci, missing loci, and unconfirmed loci (76, 77). Furthermore, these classification systems suggest that HSPs are a distinct and isolated disorder, when in fact HSPs exist on a spectrum between inherited ataxias and axonal Charcot-Marie-Tooth disease (CMT2) (76, 78). Not only do these disorders share clinical symptoms, such as prominent lower extremity spasticity, but they can also be caused by mutations within the same genes (76, 78, 79). Next-generation sequencing greatly facilitated the appreciation of these genetic overlaps by providing an unbiased approach that broke through the prior clinical and diagnostic preconceptions (76). Phenotypic expansions continue to blur the lines drawn in neurological disorders; for example, HSP was recently associated to PLA2G6, the causative gene underlying heterogeneous PLA2G6-associated neurodegeneration (PLAN) (80). Additionally, KIF5A, known to cause both HSP and CMT2, contains a C-terminal hotspot of mutations that can cause a classical amyotrophic lateral sclerosis phenotype (81). Furthermore, mutations in ATP13A2, originally linked to a rare form of juvenile-onset atypical Parkinson disease (Kufor-Rakeb syndrome), are now also associated with neurodegeneration with brain iron accumulation, neuronal ceroid lipofuscinosis, and most recently, complicated HSP (82–84). Awareness and consideration of phenotypic expansions will be essential for both individual genetic diagnoses as well as revealing common pathways underlying neurodegeneration (80). As the phenotypic spectrum broadens across the neurologic community, these historical classifications are being reconsidered (78). To address this issue, Synofzik and Schüle have proposed a mechanism based classification system for the ataxia-spasticity spectrum, based on unbiased modular phenotyping, that captures nuanced phenotypic expression, opens ataxia and spasticity to a multisystem neuronal dysfunction, and help to prioritize research on shared pathways (76).

The above-mentioned approaches increasingly require larger datasets which contradicts, of course, the low prevalence of rare disease. This requirement exceeds what single labs have traditionally been able to collect from local clinics. It will be insufficient to exchange candidate gene information or candidate alleles. To apply statistical approaches, one needs to gather hundred if not thousands of HSP samples and adequate controls. This led to the notion of raw genetic data aggregation as the next frontier for HSP gene discovery. The most prominent example of systematic data aggregate in HSP is the GENESIS Project database (tgp-foundation.org). Over 600 HSP, 500 ataxia, and 890 CMT exomes or genomes have been aggregated from several dozen laboratories in 22 different countries(85). We have begun to complement Mendelian gene discovery efforts with modifier gene studies and rare variant burden analyses. It is still early, but it appears that significant results can be achieved beginning with 200+ exomes and sufficient control samples. For example, an exome-wide association study with 202 cases and 6,905 controls successfully found a signal in GREB1L (joint p-value = 2.3 × 10⁻⁷) in renal agenesis and hypodysplasia (86). Another exome-wide association study in CMT, with 343 cases and 935 controls, identified a significant association in EXOC4 (p-value = 6.9 × 10⁻⁶, OR = 2.1) and nominal associations with other known CMT genes (87).

The current period is a remarkable time for HSP research. Studies from many countries are reporting a steady pace of novel Mendelian genes, complementing existing multigene clinical panels. The diagnostic yield has never been higher; albeit it is hindered by an increasing burden of Variants of Uncertain Significance. In addition, gene therapy approaches are maturing in related disorders, and it appears only a matter of time before they are applied to specific HSP genes. These novel therapeutic approaches include gene replacement, antisense oligonucleotides (ASO), and soon gene editing. Most of them require a specific genetic diagnosis. This emphasizes the need to fill the diagnostic gap currently estimated at 30-40%, but larger for sporadic HSP cases.

Contemplating the details of recently discovered HSP genes suggests that traditional Mendelian mutations may not be able to account for the majority of yet to be diagnosed patients. We have outlined the various possibilities of non-protein coding Mendelian variation but also an increasing interest in multigene causation or phenotypic modification. It appears to soon to tell how the genetic architecture of HSP will look like in 10 years' time. However, we are certain that non-Mendelian elements will play a role, if only as secondary protective/worsening factors. Still, cohort approaches now enabled through extensive collaborations and data aggregations will likely also hold surprises for non-traditional disease causation in HSP.

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.