The frontiers of sequencing in undiagnosed neurodevelopmental diseases

Abstract

Characterized by impairments in brain and central nervous system development, neurodevelopmental diseases causes are highly heterogeneous. Although many of these diseases are individually rare, collectively more than 3% of the children are reported to be affected with a type of neurodevelopmental diseases worldwide, and many remain undiagnosed even with current genomic tools. Identifying the genetic causes of these diseases allows better clinical management and expands our understanding of human neurodevelopment. Over the last decade, expansion of genomic sequencing and some methodologic improvements have improved molecular diagnostic yield as well as the discovery of novel genetic causes for wide spectrum of neurodevelopmental diseases. Here we review the current diagnostic workflow and propose ways of improving the diagnostic yield.

Introduction

Neurodevelopmental diseases (NDD) are due to disruptions in the complex process of central nervous system development and maintenance. NDD commonly presents as some form of developmental delay with or without variable syndromic or clinical diagnostic features such as autistic features, epilepsy and hypotonia. Many of the NDD are chronic and progressive with reduced reproductive fitness, imposing significant medical and economic burden on affected families and society. Worldwide, it is estimated that more than three percent of children are affected with a NDD [1–3]. Although both genetic and environmental factors can play a role in NDD, most of the severe forms are expected to originate from a genetic cause with a high penetrance pathogenic variant in a single gene [4,5]. Clarifying genetic causes of these early childhood onset diseases not only ends the diagnostic odyssey for families but also permits more focused genetic counseling, may reduce unnecessary tests, and allows physicians to have a clearer understanding of disease course, organs affected, and, in some instances, insights into therapy. However, while more common genetic diseases such as Rett syndrome, Neurofibromatosis type 1 (NF1) or Duchenne muscular dystrophy (DMD) have hallmark phenotypes, the initial clinical presentation of many NDD remains difficult to link to precise genetic diagnosis due to highly overlapping phenotypes and variable expressivity of syndromic components. Furthermore, NDD are genetically highly heterogeneous. In OMIM (Online Mendelian Inheritance in Man) [6], there are close to two thousand genes already identified for a form of NDD. Hence, the era of discovery of new traits by clinical presentation is much rarer now and superseded by our ability to observe causal variants through search across all genes. Here, we review the common clinical workup adopted in genetics for NDD that comprise the largest population referred for clinical exome sequencing (CES), the current obstacles to molecular diagnosis, recent methodologic improvements, and potential means to improve diagnostic rate. We also discuss how the Medical geneticists have already been practicing precision medicine, which became a popular term recently, for decades within the domain of rare genetic diseases including NDD.

Main Text

Review of molecular diagnostic rate using currently available genomic testing

Chromosomal Microarray and Clinical Exome Sequencing

A common medical genetics approach to diagnosis of patients with NDD is to perform genomic testing in a sequential order starting with chromosomal microarray (CMA), and progressing to large gene panels and CES (Figure 1). Gene panels are often ordered instead of CES, mostly because CES is denied by insurance or in instances where repeat expansions are suspected. The diagnostic rates for CMA and CES for children with suspected genetic diseases are reported to be 9–13% [7–10] and 26–30% [11–14], respectively. Even though CES has the highest diagnostic yield for genetically heterogeneous diseases, it is clear that more than 60% of patients referred for CES remain undiagnosed after CMA and CES. Currently, reanalysis of CES can result in resolution of 10–15% of these unsolved cases, largely due to a new publication reporting a novel syndrome and genetic cause that allows clear reinterpretation of prior CES findings [15–19].

Figure 1.

Summary of available genomic tests for diagnosing rare neurodevelopmental diseases. The published diagnostic rate (DR) for each test is indicated. CMA: chromosomal microarray, CNV: copy number variant, SNV: single nucleotide variant, INDEL: small insertion and deletion, SV: structural variants including CNV.

Whole Genome Sequencing and Transcriptome Sequencing

The most plausible next step for NDD patients who remain undiagnosed after CES reanalysis is to have whole genome sequencing (WGS) performed. However, only a small portion of patients will undergo WGS largely because approval by insurance is low. Thus, much of this experience is derived from large research studies such as Undiagnosed Diseases Network (UDN) [20,21], Genomics England [22,23] and Deciphering Developmental Disorders (DDD) [24,25] studies where individual unsolved cases are sequenced more thoroughly. Although many WGS studies have reported a diagnostic rate above 40%, this includes variants that could have been identified by CMA or CES alone. Augmentation of the diagnostic rate by WGS is rather modest at 7–8% [26,27] with most of the added value derived from structural variants (SVs) not identifiable by CMA or CES such as inversions, translocations and copy number variants (CNVs) that are smaller than what CMA can detect and larger than what CES can detect. Many WGS variants are within intronic, untranslated or intergenic regions that remain difficult to interpret. Some clinical laboratories offer WGS as a diagnostic test, but the test is nearly equivalent to CES test with even SV calling not included as part of the standard clinical pipeline.

Recently, several reports described how transcriptome sequencing (RNAseq) can be used to interpret some of these non-coding variants for diagnosing exome or genome negative patients. Earlier studies [28–30] have focused on diseases for which relevant tissues were accessible such as muscle for muscle diseases and patient-derived skin fibroblast for mitochondrial diseases. The diagnostic rate augmented by these studies ranged from 2–24% (adjusted to only include cases for which RNAseq was essential for the diagnosis). Two more recent studies [27,31] have used blood, patient-derived skin fibroblast or muscle for wide spectrum of rare diseases including NDD and showed diagnostic rates of 2% [31] and 15% [27] (adjusted to only include cases for which RNAseq was essential for the diagnosis). Only one of these studies [27] fully incorporated the WGS data with RNAseq data to interpret the DNA variants not identifiable by CES or interpretable without the transcriptome data. This study reported multiple cryptic deep intronic pathogenic variants altering splicing as opposed to variants near known splice junctions, suggesting integrating the two omics datasets yields higher diagnostic rate than using RNAseq data alone. However, even with this integration, only about 15% of the cases could be resolved.

Approaches to increase the molecular diagnostic rate

The overall diagnostic rate achieved for NDD after going through series of diagnostic workup from CMA to WGS and RNAseq is estimated to be approximately 50% (Figure 1). Taking examples from a few well-characterized and comprehensively studied genetic diseases with close to 100% diagnostic rate such as cystic fibrosis (CF), NF1, and DMD, we predict that the diagnostic rate of this large cohort of undiagnosed patients with NDD are likely to have substantial improvements in diagnostic rate. However, because some NDD may be environmentally caused or oligogenic, it is unlikely that we can ever reach a diagnostic rate of 100%. The biggest difference between these well-characterized diseases with high diagnostic rate and NDD is that the well-characterized diseases have clinical phenotypes distinct enough for the medical geneticists to focus on a single gene and exhaustively search for a pathogenic variant while most of the NDD have such a substantial genetic heterogeneity with significant overlapping and variable phenotypes, the search space has to vastly expand to thousands of genes. It is also reasonable to extrapolate that thousands of additional genes are yet to be identified as causal for Mendelian genetic forms of NDD. Thus, methods that provide a more comprehensive search for mutation types or locations within the genome that are relatively unexplored provides an attractive route forward for novel gene discovery.

Paths to novel gene discovery

Positional cloning of novel disease genes using linkage and mutation screening resulted in a wave of steadily increasing novel genes discoveries since 1986 when the first positionally cloned disease genes were identified. However, this type of discovery was largely supplanted by exome sequencing, with or without positional information, starting in 2011 when exome sequencing could be applied to research cohorts and was implemented as a clinical diagnostic test. Novel discovery of genetic causes of NDD are still being made, but more recently, the discovery rate has decreased, showing a similar trajectory to the discovery rate of all Mendelian diseases (Figure 2) [32,33]. While large families and specifically collected individuals with similar phenotypes have led to substantial gene discovery [33,34], many additional discoveries have also occurred by post-hoc aggregation of individuals with rare genetic variants and similar phenotypes [35–37]. Tools to assist clinicians like GeneMatcher [38,39] are now commonly used to search for additional individuals with similar DNA variants, but can lead to artificially high ‘match rates’ and even with this accepted mechanism, the rate of detection/publication of genes associated with Mendelian diseases is decreasing. We infer that many of the remaining disease causing genes to be discovered are likely to be for ultra-rare diseases or lethal conditions that only present under rare circumstances such as somatic mosaicism or potentially with oligogenic causes. For these diseases, finding the second and third cases with the same molecular basis becomes more challenging, requiring significant functional analysis as supporting evidence, which is not always achievable. With prenatal CES more routinely performed, some of the genes causing lethal conditions may be detected. As shown in figure 2, there are approximately twice as many autosomal recessive genes as autosomal dominant genes for both NDD and all Mendelian diseases and the ratio has been consistent over the last 8 years for newly discovered genes except for 2012 when CES launched. This suggests that many of the to-be-discovered causal genes are likely to be recessive.

Figure 2.

Approximate number of novel genes discovered to be associated with an NDD (stacked bar) and Mendelian disease (blue line) per year (primary Y-axis) and cumulatively (secondary Y-axis and yellow and green curved line graph). For NDD, the genes were separately counted for autosomal dominant (NDD-AD), autosomal recessive (NDD-AR) and X-linked (NDD-X) disease genes. Estimation was made by counting the number of genes deposited in the Human Gene Mutation Database (HGMD) for the first time for each year. Only the disease-causing mutations (DM) were considered. (A) Number of autosomal dominant (AD), autosomal recessive (AR) and X-linked genes for NDD and all Mendelian disease genes cumulatively.

Increasing diagnostic rate from RNAseq

Augmentation of WGS with RNAseq is primarily limited by ready access to relevant tissues likely to express the mutant allele. Available tissues, such as blood or skin fibroblasts, for NDD will be inadequate if the mutant gene is mainly expressed in brain. The broad search space of known and unknown novel genes is in stark contrast to diseases like CF, NF1 or DMD, where the clinical phenotype is widely recognized. Of ~2,000 known potential genes for NDD, a portion are inadequately expressed in accessible tissues [27] but coverage improvement is likely possible by generating deeper RNAseq data. However, this will not improve observations of genes that are not expressed in the accessible tissues. Generating patient specific neuronal cells by reprograming patient-derived cells into induced pluripotent stem cells (iPSC) and differentiating into neurons [40,41] or directly converting patient-derived fibroblast into neuronal cells [42–44] are reasonable approaches to address this matter, but remain too inefficient to implement broadly on patient derived cells. Further, only a limited set of neuronal cell types are likely to be generated and thus one can reasonably project that some neuronal genes will still not be expressed. Alternate methods such as deregulating gene expression broadly may be helpful in order to generate spliced mRNA that may reveal splicing abnormalities caused by a DNA variant. These methods may be coupled with mechanisms to suppress nonsense-mediated decay (NMD) in patient-derived cells [45,46]. NMD can greatly challenge our ability to observe the mutant allele with a premature termination codon as the coverage of the mutant transcript is lower and thus less observed by RNA sequencing.

The most commonly used sequencing library preparation protocol is selection of polyadenylated (polyA) transcripts, which excludes non-coding RNAs and small RNAs. These provide an excellent assessment of the protein coding portion of the human genome, but may miss important functional RNA elements that may be observed with other methods. To date there has been little identification of causal mutations within these non-coding RNAs, and there remains no good estimate of how much variants in these RNAs contribute to the pathogenesis of rare NDD, but there is a growing literature of the importance of some long non-coding RNAs in the mechanism of pathogenesis of some of the rare diseases [47].

Identifying variants challenging to detect or interpret by current technologies

There will be pathogenic variants that are not detectable by short-read sequencing technologies that are most commonly used, although based on the examples of the diseases with near 100% diagnostic rate, we can predict that not a significant number of cases will have these types of variants. For example, the molecular diagnostic rate for patients clinically diagnosed with CF is 97–98% by conventional DNA sequencing test alone [48,49]. Significant portion of the 2–3% undiagnosed cases most likely have deep intronic variants affecting splicing but CFTR gene is not expressed in accessible tissues such as blood so it is not feasible to further explore. RNA or cDNA sequencing has been utilized more routinely for NF1. When the combined DNA and RNA (cDNA) sequencing analysis approach is used, ~97% of the cases receive a molecular diagnosis and RNA sequencing is estimated to be essential for ~10% of the diagnosed cases [50]. DMD has ~99% molecular diagnostic rate with most of the patients identified with pathogenic variants detectable by DNA sequencing test alone but ~80% of the variants are large deletions or duplication and only ~1% of the cases are solved by RNA sequencing with deep intronic variants altering splicing [51,52]. These examples suggest that the current technologies are capturing most of the pathogenic variants but they also show that the type of variants found to be pathogenic could vary by gene and a subset of genes can be enriched with pathogenic variants not readily detectable or interpretable by current technologies. Generating longer read genome and transcriptome data by newer technologies such as PacBio [53,54], Oxford Nanopore [55,56], BioNano [57] and 10X genomics [58] are some of the available alternate technologies. However, rare Mendelian disease patients diagnosed by pathogenic variants identified by these technologies, which could not have been detected by more routinely used technologies, are yet to be reported.

Improving variant calling and variant interpretation

Most genomic testing is performed on peripheral blood DNA and most of the variant callers are highly optimized to detect homozygous, hemizygous or heterozygous variants and not mosaic variants which can be an important contributor, especially for lethal diseases [59]. An example NDD caused by somatic variants is hemimegalencephaly for which only one hemisphere of the brain overgrows, often causing severe epilepsy and developmental delay [60]. Complementing the variant calling with somatic variant caller can be applied to resolve this limitation. However, collecting the appropriate tissue will be essential to detect some of these somatic mosaic variants and that could be challenging.

An important component of variant interpretation from exome, genome and transcriptome sequencing is filtering out common variants that cannot be causal for ultra-rare NDD. The establishment of ExAC (Exome Aggregation Consortium) [61] that merged into gnomAD (Genome Aggregation Database) browsers [62] made filtering out the common variants significantly more effective. As these dataset grows, especially with more data added from the under-sequenced minority populations and non-coding regions, it is expected that parental comparator sequencing may not even be necessary to identify de novo variants for autosomal dominant NDD as causal heterozygous variants for such a rare, severe diseases are not expected to be observed in the population although there could be exceptions due to factors such as incomplete penetrance, imprinting or mosaicism in the population. For recessive diseases, variants can be observed in the population but not at high frequency. ACMG-AMP (American College of Medical Genetics-Association for Molecular Pathology) joint consensus recommendation for the interpretation of sequence variants uses >5% minor allele frequency (MAF) threshold for classifying variants benign although in practice, a 1% threshold is more commonly used for most of the rare diseases [63–65]. The MAF threshold can be further lowered from 1% because as plotted in figure 3, with the exception of few genes, MAF of the pathogenic or likely pathogenic variants for recessive diseases are well under 0.1% with ~97% of the variants being under 0.01%, suggesting that MAF of 0.1% or greater are unlikely to be causal for most recessive NDD.

Figure 3.

Minor allele frequency distribution of pathogenic/likely pathogenic variants in autosomal recessive neurodevelopmental diseases. Neurodevelopmental disease gene list was created by querying OMIM for genes associated with autosomal recessive diseases involving developmental delay, intellectual disability, mental retardation, seizures or epilepsy. Variants that were in HGMD as disease causing and classified as pathogenic or likely pathogenic in ClinVar were selected. gnomAD was used to extract MAF information on these variants.

Lastly, databases such as ClinVar [66] or Human Gene Mutation Database [67] are helpful for classifying variants and identifying cohorts of patients with same or similar pathogenic variants. Along the line of making variant filtration more effective by growing the population database, having benign or likely benign variants catalogued in these variant classification databases are as essential as having pathogenic or likely pathogenic variants catalogued. A more rapid entry system into ClinVar is clearly needed to more promptly translate novel gene discovery into clinical impact [33].

Conclusion

In the past decade, the advent of next-generation sequencing (NGS) technology and its clinical application have made significant contribution to Medical Genetics by identifying the molecular basis of numerous known and a large number of new rare NDD. Not only has the novel gene discovery rate peaked during this period, but also the time and resources taken to molecular diagnosis for individual patients has substantially decreased with the NGS-based genomic tests becoming the first-tier test in the clinic. It is now possible to end the diagnostic odyssey earlier and provide more tailored clinical management. However, it is quite sobering that as we presented here, there are still about 50% of patients with highly suspected genetic diseases, referred for exome sequencing that remain without a molecular diagnosis, and the next decade will focus on improved methods to detect currently cryptic pathogenic variants. The workflow in place as described in figure 1 has been established quite stably in the current medical system and we expect that it will continue to stay the same for a while. However, we foresee and propose that in a near future, as sequencing price continues to decrease, WGS combined with RNAseq will replace CMA and CES and become the first-tier test performed in all NDD patients as soon as they start showing symptoms.

Precision health or precision medicine has been a buzzword in the past few years with the core of it being characterizing the genetic or molecular basis of a health condition to personalize healthcare. It is important to recognize that the rare disease community has been practicing precision medicine for a long time by tailoring clinical management based on the molecular basis of a disease and leading the effort in developing tools and standards for identifying and interpreting genetic information. It should also be noted that the vast majority of variants clearly causing a human phenotype are rare and cause a Mendelian phenotype despite massive efforts to reveal common variant contributions to common phenotypes. Thus, the rare disease community will continue to serve as the model for impactful translational genomic medicine.

Highlights.

Rate of diagnostic yield of Clinical Exome Sequencing is not increasing

Rate of new genetic disease discovery is slowing

Novel methods are needed to improve diagnostic and discovery rates