Whole genome sequences discriminate hereditary hemorrhagic telangiectasia phenotypes by non-HHT deleterious DNA variation

Abstract

The abnormal vascular structures of hereditary hemorrhagic telangiectasia (HHT) often cause severe anemia due to recurrent hemorrhage, but HHT causal genes do not predict the severity of hematological complications. We tested for chance inheritance and clinical associations of rare deleterious variants in which loss-of-function causes bleeding or hemolytic disorders in the general population. In double-blinded analyses, all 104 patients with HHT from a single reference center recruited to the 100 000 Genomes Project were categorized on new MALO (more/as-expected/less/opposite) sub-phenotype severity scales, and whole genome sequencing data were tested for high impact variants in 75 HHT-independent genes encoding coagulation factors, or platelet, hemoglobin, erythrocyte enzyme, and erythrocyte membrane constituents. Rare variants (all gnomAD allele frequencies <0.003) were identified in 56 (75%) of these 75 HHT-unrelated genes. Deleteriousness assignments by Combined Annotation Dependent Depletion (CADD) scores >15 were supported by gene-level mutation significance cutoff scores. CADD >15 variants were identified in 38/104 (36.5%) patients with HHT, found for 1 in 10 patients within platelet genes; 1 in 8 within coagulation genes; and 1 in 4 within erythrocyte hemolytic genes. In blinded analyses, patients with greater hemorrhagic severity that had been attributed solely to HHT vessels had more CADD-deleterious variants in platelet (Spearman ρ = 0.25; P = .008) and coagulation (Spearman ρ = 0.21; P = .024) genes. However, the HHT cohort had 60% fewer deleterious variants in platelet and coagulation genes than expected (Mann-Whitney test P = .021). In conclusion, patients with HHT commonly have rare variants in genes of relevance to their phenotype, offering new therapeutic targets and opportunities for informed, personalized medicine strategies.

© 2022 by The American Society of Hematology. Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0), permitting only noncommercial, nonderivative use with attribution. All other rights reserved.

Figures

Graphical abstract

Figure 1.

Framework for studies “toward personalized genomic interpretations” (TPGI). MALO: More severe than expected for disease; As expected for disease; Less severe than expected for disease, and Opposite phenotype. *Gene selection by any broadly relevant set of panels such as PanelApp established for 100 000 Genomes Project interpretations suggested to include additional categories less relevant to phenotypes (as in current study) to control for methodological biases. Where TPGI demonstrates evidence of relevance across a patient cohort, this supports attention to a subsequent “for personalized genomic interpretations” (FPGI) stage to be developed within mainstream medicine. WES, whole exome sequencing; WGS, whole genome sequencing.

Figure 2.

HHT genes and phenotypic associations. (A) Endothelial vasculopathy genes and variants identified in current study cohort. Note that the major HHT genes are ACVRL1 (red) and ENG (green).SMAD4 is less common, and GDF2 was more recently described as HHT causal, whereas EPHB4 and RASA1 cause separate endothelial vasculopathies (CM-AVM2 and CM-AVM1) that overlap phenotypically with HHT. (B) Schematic of the cohort of 104 patients plotted on 4 separate axes for hemorrhage (H), anemia (A), thrombosis (T), and deep-seated infection (I). Ten deep-seated infections were in association with concurrent and presumed causative pulmonary AVMs (cerebral and spinal abscesses, spinal discitis, and septic arthritis due to polymicrobial flora, particularly anaerobic and aerobic commensals of the gastrointestinal and periodontal spaces). Blue numbers indicate the number of the cohort with no events/no excess (H0, A0, T0, and I0), precipitated events (H1, A1, T1), and spontaneous events (H2, A2, T2, and I1). I1 were cerebral abscess (N = 6), spinal discitis (N = 2), spinal abscess (N = 1), septic arthritis (N = 1), recurrent bacterial endocarditis (N = 1), osteomyelitis (N = 1), and recurrent sepsis (N = 1). (C) Violin plots for ACVRL1 (left, red) and ENG (right, green) patients orientated on the 3 axes as in (B).

Figure 3.

Hematological genes and variants. (A) Genome positions of the variant-containing genes in the study cohort on genome ideogram indicating positions of the 56 study genes with variants (red) and the two major HHT genes ACVRL1 and ENG. Variants present in patients with an identified variant in HHT genes are indicated for ACVRL1 (purple lines) and ENG (blue lines). To preserve anonymity, data for the single SMAD4, GDF2, and EPHB4 families are not illustrated. (B) Chromosomal distributions of variant-containing genes per 100 Mb of gDNA (blue circles/lines), number of variants per Mb (red shaded background), and variants with CADD score >15 (red circles/lines) per Mb of gDNA by chromosome. (C) Population-level burden of genetic damage in the 6 HHT panel genes (blue) and 75 study hematological genes (red), as detailed in supplemental Table 1. The study genes were subcategorized by variant presence (Variant genes; N = 56) and absence (No-variant genes; N = 19) in the study cohort, and P values were calculated by Dunn’s post Kruskal Wallis: (Ci) Gene damage index (GDI) phred scores, a genome-wide, gene-level metric of the mutational damage that has accumulated in the general population that performs well at removing exome variants in genes irrelevant to disease. (Cii) Residual variation intolerance scores (RVIS) that perform better for detection of genes in which newly identified variants are more likely cause a recognized disease.

Figure 4.

Burden of deleterious variants by categories of genes and phenotypic severity profiles. (A) The CADD score distribution of variants identified in the study cohort (red) compared with gene-level MSC scores, indicating the lower limit of the 90% and 95% confidence intervals for deleterious CADD scores in individual genes (gray violin plots). Vertical dotted lines represent overall CADD scores 5 and 15 and the violin plot median and interquartile ranges. Note that CADD scores rank the deleteriousness for all 9 billion single nucleotide variants, and millions of small indels and splice site variants, based on machine learning trained on diverse genomic features derived from surrounding sequence context, gene model annotations, evolutionary constraint, epigenetic measurements, and functional predictions. (B) The 5 process-level categories, respective category contributions in the study cohort, are shown for the total number of genes, and all variants,, with each box representing the category’s percentage for the row, as indicated by the heat scale. (C) The burden of deleterious (CADD >15) variants in the 5 categories of genes for each predefined phenotypic subcohort. Filled pie-chart regions quantify the number of CADD >15 variants per patient in the category (gray if pooling across all gene categories; red if gene category specific; fully filled if 1 variant per patient). The 10 variants with CADD scores of 1 to 5 (likely benign) and 36 with CADD scores between 5 and 15 (of uncertain significance) are effectively included in the remaining white pie chart regions to enhance clarity. For each set of 6 pie charts, gene category placements are consistent. The upper row indicates total number of CADD >15 variants across all categories, and the remaining upper row pie charts focus on hemorrhage (upper middle for coagulation genes, upper right for platelet genes). The lower row presents the red cell gene categories (lower left for hemoglobin genes, lower middle for erythrocyte [red cell] membrane genes, and lower right for erythrocyte enzyme genes). The set of 6 gray-lined empty pie charts indicate there were no patients in the predefined A1 category. (D) Heat maps for number of variants per patient in each phenotypic subcohort, in which the precipitated and spontaneous H1/H2, T1/T2, and A1/A2 subcohorts have been pooled. Trends highlighted in the text are denoted by a white cross or black star. *P < .0286; **P < .01.

Figure 5.

HHT patient subcategories and variation. Quantitative phenotypic measurements categorized by presence, absence, and CADD score of variants. Data points represent all datasets captured in the 104 patients across their clinical assessments, plotted according to the presence or absence of variants in the genes for (A) hemoglobin production, (B) red bood cell membrane production, (C) red blood cell enzymes, (D) coagulation and (E) platelets as listed in supplemental Table 1. Two indices are provided for each category of genes, but there were also no differences observed between categories for other indices examined (data not shown).

Figure 6.

Variant burden in HHT cohort compared with general population. (A) Number of variants in the 75 study genes by category, within the gnomAD 3.1.2 dataset (blue) and current cohort (red) per genome. (B) gnomAD 3.1.2 dataset loss of function (pLOF) variants compared with CADD >15 genes in study cohort per genome, with mean and standard error illustrated. (C) Fold-enriched total number of deleterious variants per category, in all 75 genes, for HHT CADD >15 variants compared with pLOF variants in gnomAD 3.1.2 dataset. (D) Comparison of fold-enriched deleterious variants limiting to HHT CADD >15 variants and gnomAD pLOF variants in the same genes. P value calculated by Mann-Whitney test.