Achieving high-sensitivity for clinical applications using augmented exome sequencing

Abstract

Background: Whole exome sequencing is increasingly used for the clinical evaluation of genetic disease, yet the variation of coverage and sensitivity over medically relevant parts of the genome remains poorly understood. Several sequencing-based assays continue to provide coverage that is inadequate for clinical assessment.

Methods: Using sequence data obtained from the NA12878 reference sample and pre-defined lists of medically-relevant protein-coding and noncoding sequences, we compared the breadth and depth of coverage obtained among four commercial exome capture platforms and whole genome sequencing. In addition, we evaluated the performance of an augmented exome strategy, ACE, that extends coverage in medically relevant regions and enhances coverage in areas that are challenging to sequence. Leveraging reference call-sets, we also examined the effects of improved coverage on variant detection sensitivity.

Results: We observed coverage shortfalls with each of the conventional exome-capture and whole-genome platforms across several medically interpretable genes. These gaps included areas of the genome required for reporting recently established secondary findings (ACMG) and known disease-associated loci. The augmented exome strategy recovered many of these gaps, resulting in improved coverage in these areas. At clinically-relevant coverage levels (100 % bases covered at ≥20×), ACE improved coverage among genes in the medically interpretable genome (>90 % covered relative to 10-78 % with other platforms), the set of ACMG secondary finding genes (91 % covered relative to 4-75 % with other platforms) and a subset of variants known to be associated with human disease (99 % covered relative to 52-95 % with other platforms). Improved coverage translated into improvements in sensitivity, with ACE variant detection sensitivities (>97.5 % SNVs, >92.5 % InDels) exceeding that observed with conventional whole-exome and whole-genome platforms.

Conclusions: Clinicians should consider analytical performance when making clinical assessments, given that even a few missed variants can lead to reporting false negative results. An augmented exome strategy provides a level of coverage not achievable with other platforms, thus addressing concerns regarding the lack of sensitivity in clinically important regions. In clinical applications where comprehensive coverage of medically interpretable areas of the genome requires higher localized sequencing depth, an augmented exome approach offers both cost and performance advantages over other sequencing-based tests.

Fig. 1

A total of 5,419 genes in the MIG drawn from five data sources. The bulk (98 %) of genes came from HGMD, OMIM, and GTR with additional genes supplemented from COSMIC (67) and PharmGKB (1). Areas of vertical overlap indicate genes common across multiple sources

Fig. 2

Coverage efficiency in the medically interpretable genome (MIG). Shown is the cumulative distribution of on-target sequence coverage obtained from sequencing NA12878 across multiple platforms: Personalis Accuracy and Content Enhanced (ACE) Clinical Exome, Agilent SureSelect Clinical Research Exome (SSCR), Agilent SureSelect Human All Exon v5 plus untranslated regions (UTR) (SS), lllumina’s Nextera Exome Enrichment (NX), NimbleGen SeqCap EZ Human Exome Library v3.0 (NG), and 31× whole-genome sequencing (WGS) using an Illumina PCR-free protocol. For clinical applications, we indicate ≥20× as the minimum coverage threshold required (gray line) among all coding (left) and non-coding (right) regions. For reference, insets show an expanded distribution of sequence coverage. ACE and conventional WES data are normalized to 100× mean target coverage

Fig. 3

Relationship between the percentages of MIG exons ‘finished’ as the coverage stringency varies. The left graph shows the percentage of MIG exons (y-axis) with ≥90.0-100.0 % of bases covered at ≥20× depth (x-axis) among different platforms using data obtained on NA12878. The right graph shows the percentage of finished exons (y-axis) with 100.0 % base coverage as the local coverage depth varies ≥10-20× (x-axis). At higher coverage stringencies, ACE finishes more exons than other WGS or WES assays in regions defined as the entire exon (solid curves) or only the subset of coding-regions (circles). ACE and conventional WES data are normalized to 100× mean target coverage

Fig. 4

Relationship between GC content and the percentages of MIG exons ‘finished’ by platform. Regions with >30-80 % GC content (x-axis) represent 99 % of exons in the MIG. Finishing is determined by 100 % base coverage at ≥20×

Fig. 5

Disease-associated variants covered at ≥20× for 56 genes in the ACMG gene list. The x-axis labels indicate the total number of disease-associated SNVs (daSNVs) drawn from HGMD for each ACMG gene; and the y-axis indicates the percentage of those variants covered at ≥20×. For brevity, only the highest obtained percentage (Max over all WES) observed across all conventional WES (SS, SSCR, NX, NG) platforms is shown. Seventeen of the 56 genes failed to have some fraction of their daSNVs covered at ≥20× among any of the conventional WES platforms. On a gene basis, the platforms with the highest to lowest number of genes with constituent daSNVs adequately covered included ACE (51 genes with 100 % daSNVs covered at ≥20×), SSCR (39 genes), NX (36 genes), SS (15 genes), NG (12 genes), and WGS (2 genes). The y-axis is truncated at 95 %, with truncated points labelled accordingly

Fig. 6

Coverage gaps in Retinitis Pigmentosa and Cystic Fibrosis genes are recovered with augmented exome approaches. Chromosomal position (x-axis) is plotted against coverage depth (y-axis) averaged over multiple 1000 Genome samples, with the clinical coverage threshold (≥20×) represented by a horizontal black line. Blue areas represent mean-depth of coverage across coding and non-coding regions using the SS (light blue), and SSCR (dark blue) exomes. Areas in green represent coverage gaps ‘filled in’ by ACE. These include areas with known pathogenic variants in high GC rich areas in the RPGR gene associated with retinitis pigmentosa (a); or non-coding regions of the CFTR gene (b)