Approach to Cohort-Wide Re-Analysis of Exome Data in 1000 Individuals with Neurodevelopmental Disorders

Abstract

The re-analysis of nondiagnostic exome sequencing (ES) has the potential to increase diagnostic yields in individuals with rare diseases, but its implementation in the daily routines of laboratories is limited due to restricted capacities. Here, we describe a systematic approach to re-analyse the ES data of a cohort consisting of 1040 diagnostic and nondiagnostic samples. We applied a strict filter cascade to reveal the most promising single-nucleotide variants (SNVs) of the whole cohort, which led to an average of 0.77 variants per individual that had to be manually evaluated. This variant set revealed seven novel diagnoses (0.8% of all nondiagnostic cases) and two secondary findings. Thirteen additional variants were identified by a scientific approach prior to this re-analysis and were also present in this variant set. This resulted in a total increase in the diagnostic yield of 2.3%. The filter cascade was optimised during the course of the study and finally resulted in sensitivity of 85%. After applying the filter cascade, our re-analysis took 20 h and enabled a workflow that can be used repeatedly. This work is intended to provide a practical recommendation for other laboratories wishing to introduce a resource-efficient re-analysis strategy into their clinical routine.

Keywords: exome sequencing; neurodevelopmental disorder; re-analysis.

Figure 1

Cohort structure. Cohort structure of 1040 individuals regarding sequencing approach (A); enrichment kits used for ES—Agilent SureSelect Human All Exon V6, BGI Exome capture 59 M kit or TWIST Human Core Exome Kit (B); genetic tests conducted prior to exome sequencing (C); and disorder group (D). Numbers refer to the number of individuals. NDD: neurodevelopmental disorders.

Figure 2

Filter cascade and resulting variants. The filter cascade consists of several steps that take into account the quality of the variant, the gene–disease association, and its phenotypic overlap with the symptoms of the corresponding individual. Depending on the zygosity, we filtered by the frequency of the variant in gnomAD and by other criteria, such as the variant effect, constraints, and presence in ClinVar, also considering segregation information (e.g., de novo). Numbers of manually evaluated variants are depicted for each arm of the filter cascade (bars without hatching). Initially reported (L)P variants that are covered by our filter cascade are displayed on the left in the colour of the corresponding filter criterion (bars with hatching). Notably, some variants appear in more than one filter step (e.g., a variant is both de novo and has been reported in ClinVar), and thus are counted repeatedly. In nine genes, we identified novel (L)P variants. CV: ClinVar, var.: variant, hem.: hemizygous, hom.: homozygous, LoF: loss of function, (L)P: (likely) pathogenic, MAF: minor allele frequency in gnomAD, pLI: probability of being loss-of-function-intolerant in gnomAD, Z: missense Z score in gnomAD.

Figure 3

Iteration of missense Z score and HPOsim score. Number of variants to be manually evaluated (blue) and of initially reported (L)P variants (orange) based on missense Z score thresholds (A) and HPOsim score thresholds (B) as a single filter criterion. A) Depicted lines mark the adapted cut-off of 2.2 and the previously set cut-off of 4.0. B) Depicted lines mark the applied cut-off of 0.1 and an exemplary cut-off of 0.18. Notably, not all variants have an HPOSim score.

Figure 4

Checkbox for re-analysis of large cohorts.