Brief Report: Evaluating the Diagnostic Yield of Commercial Gene Panels in Autism

Genetic diagnosis in autism is limited by the ability to robustly determine the relevance of putatively pathogenic genetic variation. Genomic research in autism is progressing quickly, enabled by advancements in next-generation sequencing technologies and the subsequent establishment of large-scale sequencing cohorts and pedigree-based sequencing cohorts (Glahn et al., 2019; Ní Ghrálaigh et al., 2020). To date, many genes have been identified as having some link to autism (Satterstrom et al., 2020). The Simons Foundation Autism Research Initiative (SFARI) Gene database (Abrahams et al., 2013), collates more than 990 genes for which there is evidence of association with autism, however the clinical utility of this database is limited by the absence of systematic curation of gene-condition relationships (Schaaf et al., 2020). Despite this progress in autism genomic research, major challenges remain in the development of targeted gene panels with substantial clinical utility in autism.

At case-level, gene discovery is complicated by the nature of autism as a complex condition with a large degree of phenotypic heterogeneity. A candidate pathogenic variant may be evaluated, in the majority of autism cases, as being contributory to the genetic risk rather than being wholly causative of an individual’s condition. At cohort-level, studies discovering “autism genes” are compounded by an apparent lack of specificity to autism. For example, in individuals affected by both autism and intellectual disability, genes identified show relevance to both autism and other neurodevelopmental disorders (Myers, Challman, Bernier, et al., 2020). For these reasons, the development of effective gene panels to aid autism diagnosis is extremely complicated.

Despite these limitations, commercial gene panels are available and marketed for use in autism diagnosis. Hoang et al. (2018) evaluated many of these gene panels, clearly demonstrating their heterogeneity (Hoang et al., 2018). Their survey shows large variability in the number of genes being tested by panels, lack of consensus in the genes selected for inclusion, as well as variability in the reporting of laboratory qualification and reporting protocols.

Here we estimate the clinical utility of commercial gene panels marketed for use in autism. Diagnostic yield, which is the proportion of cases interrogated for which a genetic cause can be determined, is a strong measure of the clinical utility of a sequencing technology. Feliciano et al. (2019) estimate the diagnostic yield of whole exome sequencing to be 10.4% in the initial 457 families enrolled in the Simons Powering Autism Research (SPARK) cohort (Feliciano et al., 2019). A ‘likely pathogenic’ variant was identified in a further 3.4% of families studied. This estimate comes from the identification of a variant that fulfils either the ‘likely pathogenic’ or ‘pathogenic’ criteria, according to American College of Medical Genetics and Genomics (ACMG) standards (Richards et al., 2015).

Gene panels relevant to autism are presented in Table 1. To determine the clinical utility of each autism gene panel, variants meeting ‘likely pathogenic’ or ‘pathogenic’ criteria in the SPARK cohort can be limited to those within the gene set of each panel, respectively. In doing so, we ask how many of the pathogenic variants identified by Feliciano et al. would have been identified in the SPARK cohort with application of an autism gene panel, instead of application of whole exome sequencing. The diagnostic yield of each gene panel, estimated with respect to Feliciano et al. analyses, is presented in Table 1. The diagnostic yields range from 0.22% to 10.02%, with most gene panels achieving a diagnostic yield below 3%.

Gene discovery in autism is ongoing. Most genes included in the commercial gene panels are autism relevant. This is illustrated by the inclusion of a large proportion of the panel-specific genes in the SFARI Gene database (Abrahams et al., 2013)(Table 1). Gene selection for inclusion in autism panels is variable in relevance, ranging in overlap with SFARI Gene from 15.15% to 100%. Diagnostic yield of the gene panels and size of the panel were found to be positively correlated, (r = 0.82, p = 3.033e-05), indicating an increased number of genes per gene panel enables detection of a clinically relevant variant in a greater number of individuals. No significant correlation between percentage overlap with SFARI Gene and diagnostic yield was detected.

Considering the low diagnostic yield of the gene panels that were investigated, we can infer that, while the gene selection for inclusion in autism gene panels is evidence-based, these gene lists are not extensive enough to justify use in autism diagnosis, a complex trait for which hundreds of genes have been associated. Critically, if the application of a targeted gene panel to an affected individual’s genome returns negative for pathogenic variation, one cannot conclude that a causative variant is not present. Rather, it is more likely that genetic causes have been missed due to the limited application of the gene panel.

The GeneDx “Autism/ID Xpanded Panel” represents the autism gene panel with the highest number of individuals for which a genetic diagnosis would have been obtained with its application (10.02%). This diagnostic yield is comparable to that of whole exome sequencing, 10.4% (Feliciano et al., 2019) and that of chromosomal microarray sequencing with a median diagnostic yield of 8.1% (Savatt & Myers, 2021). However, important to note is that this gene panel targets many more genes (n = 2,641) than some of the smaller gene panels, for example GENETAQ “Autism” panel (n = 27), with a diagnostic yield of just 1.53%. The positive correlation of diagnostic yield associated with inclusion of a larger number of genes, reflects well the complex genetic architecture of autism and the number of loci expected to be associated. This raises the question whether autism is an appropriate candidate for the development of commercial gene panels, that are reliant and limited due to the size of the gene panel, the cost and current knowledge of the genetic basis of autism, and questions whether developments should focus on application of sequencing technologies with a broader coverage, such as whole genome sequencing. Expanding beyond targeted autism genes, whole genome sequencing presents the opportunity to explore more of the human genome and, ultimately, to further increase the diagnostic yield in autism (Yuen et al., 2015). Progress in non-coding variant annotation and interpretation, accompanied by a decrease in sequencing costs, may further popularize the clinical use of whole genome sequencing. Currently, whole exome sequencing is proposed as the first-tier diagnostic test for neurodevelopmental disorders (Srivastava et al., 2019). The diagnostic yield in autism using clinical exome sequencing has been estimated at 6.1% in autism (20% overall yield in neurodevelopmental disorders) (Martinez-Granero et al., 2021). Genotyping chips have limited clinical utility for rare genetic variation of SNVs and should not be used to guide health decisions without validation (Mn et al., 2021). Autism genetic testing as minimal as cytogenetic microarray and Fragile X testing alone may be all that is feasible in a clinical setting, which is currently the situation in Ireland.

Provided the relevant expertise and infrastructure for variant interpretation are available and cost effective, gene panels have potential for clinical utility. However, current evidence does not support their applicability in autism (Buxbaum et al., 2020; Myers, Challman, Martin, et al. 2020). Achieving the ultimate goal of a comprehensive autism gene panel will require uniform robust phenotyping to account for the heterogeneity in autism presentation. Application of a formal evidence-based gene curation framework, such as that proposed by Schaaf et al. (Schaaf et al., 2020), would account for the degree of certainty in autism diagnoses in studies reporting association and account for co-morbid diagnoses, providing consistency throughout gene discovery. To conclude, evaluation of the diagnostic yield of commercial gene panels marketed for autism determines that they are currently of very limited clinical utility.