A high-throughput sequencing test for diagnosing inherited bleeding, thrombotic, and platelet disorders

Abstract

Inherited bleeding, thrombotic, and platelet disorders (BPDs) are diseases that affect ∼300 individuals per million births. With the exception of hemophilia and von Willebrand disease patients, a molecular analysis for patients with a BPD is often unavailable. Many specialized tests are usually required to reach a putative diagnosis and they are typically performed in a step-wise manner to control costs. This approach causes delays and a conclusive molecular diagnosis is often never reached, which can compromise treatment and impede rapid identification of affected relatives. To address this unmet diagnostic need, we designed a high-throughput sequencing platform targeting 63 genes relevant for BPDs. The platform can call single nucleotide variants, short insertions/deletions, and large copy number variants (though not inversions) which are subjected to automated filtering for diagnostic prioritization, resulting in an average of 5.34 candidate variants per individual. We sequenced 159 and 137 samples, respectively, from cases with and without previously known causal variants. Among the latter group, 61 cases had clinical and laboratory phenotypes indicative of a particular molecular etiology, whereas the remainder had an a priori highly uncertain etiology. All previously detected variants were recapitulated and, when the etiology was suspected but unknown or uncertain, a molecular diagnosis was reached in 56 of 61 and only 8 of 76 cases, respectively. The latter category highlights the need for further research into novel causes of BPDs. The ThromboGenomics platform thus provides an affordable DNA-based test to diagnose patients suspected of having a known inherited BPD.

Figure 1

Breakdown of the 300 samples sequenced with the ThromboGenomics platform. The width of each box is proportional to the number of individuals it represents. The 4 main categories are shown as labels in italics. The shaded area in each box shows the proportion of samples in which pathogenic or likely pathogenic variants were identified with the ThromboGenomics platform. Note that the mother of a hemophilia A patient from the “suspected” group appears in shading in the box representing the “unaffected” group.

Figure 2

Technical evaluation of the ThromboGenomics platform. (A) Histogram of mean autosomal target coverage for 300 samples. (B) Mean fraction of exonic (solid black) bases and HGMD variants (dashed red) covered at least at 0X, 1X, …, 50X, showing that 99% of the targeted exonic regions are covered by at least 50 sequencing reads. (C) Coverage profile for the ITGA2B gene.

Figure 3

Sample identity assurance. (A) The het/hom ratio vs the aut/X ratio is used to infer the gender of each individual. One sample from a male individual with an abnormally high aut/X ratio was substantially more degraded than all others. (B) A scatterplot of the first two principal components derived from the 1000 Genomes genotypes, with individuals colored by major population, and projected ThromboGenomics individuals shown as black circles if they have fewer than 7 candidate variants and triangles if they have at least 7 candidate variants. For clarity, admixed American HapMap individuals are not shown. There is a lower density of ThromboGenomics individuals with African or East Asian ancestry but they all have at least 7 variants, whereas ∼80% of ThromboGenomics individuals with European ancestry have fewer than 7 variants.

Figure 4

Candidate variants per sample. (A-C) Bar plots of the number of candidate SNVs, indels, and CNVs per individual. (D) Scatterplot of the Bayes factor vs the observed over expected reads ratio for each CNV called by ExomeDepth and the thresholds distinguishing different levels of changes in zygosity. Note that the number of called CNVs is slightly biased upwards relative to the number of true CNVs because a single underlying CNV can sometimes be coded as multiple adjacent calls by the ExomeDepth algorithm. The number of CNVs surviving filtering is slightly elevated relative to the number of surviving indels, because we include CNV calls with a Bayes factor down to 4.5 for maximum sensitivity and because CNVs do not undergo any external cohort-based frequency filtering. Het, heterozygous.

Figure 5

Case study. (A) HPO encoded phenotype of a case in the “suspected” category, visualized as a graph using the hpoPlot package. Note that “abnormality of leukocytes” is also an “abnormality of the immune system” (not shown). (B) The ratio between observed and expected read depth over the PLAU gene for the case (red) and superimposed over the 95% confidence interval (gray shaded area). In the lower panel, the central position of each exon of the PLAU gene is shown as a vertical bar and the gene coordinates are provided on the horizontal axis. The data indicate that the case carries an additional copy of the PLAU gene (Bayes factor = 145), which is compatible with a diagnosis of suspected Québec platelet syndrome.

Figure 6

HPO-based prioritization. (A) HPO profile of a case with BSS encoded as a graph. Note atypical presence of hearing impairment, which is likely unrelated to the BSS. The plot beneath the graph shows the similarities between the patient profile and each gene in which the case has a candidate variant. The profile of GP1BB is the most similar out of the 4 genes with candidate variants. (B) For each of the 109 HPO-coded cases for which a causative variant was assigned by the MDT, the similarity is shown between the case profile and the profiles of the genes in which the case has a candidate variant. The similarity to the gene containing the variant(s) determined to be pathogenic or likely pathogenic in each case (red circles) and the similarity to other genes containing VUS (gray dashes) are shown. Case index 1 corresponds to the BSS case shown in (A).

Figure 7

HGMD variants and corresponding MAFs in ExAC. (A) Truncated log-scale barplot showing the number of HGMD variants by HGMD phenotype (edited for improved legibility). (B) Log-scale barplot showing the number of HGMD variants binned by MAF in ExAC. (C) Histogram of the 140 variants in BPD genes with a MAF in ExAC exceeding 1/1000 (ie, belonging to the blue bins in panel [B]) broken down by HGMD phenotype (edited for improved legibility). The individual Phred-scaled MAFs of the variants (ie, such that 30 corresponds to 1/1000 and 20 to 1/100) are superimposed on the histogram and colored by whether they are classified as disease causing (DM and DM? categories). ADAMTS13, a disintegrin and metalloproteinase with a thrombospondin type 1 motif, member 13; IgE, immunoglobulin E; IVF, in vitro fertilization.