Phenotype risk scores identify patients with unrecognized Mendelian disease patterns

Abstract

Genetic association studies often examine features independently, potentially missing subpopulations with multiple phenotypes that share a single cause. We describe an approach that aggregates phenotypes on the basis of patterns described by Mendelian diseases. We mapped the clinical features of 1204 Mendelian diseases into phenotypes captured from the electronic health record (EHR) and summarized this evidence as phenotype risk scores (PheRSs). In an initial validation, PheRS distinguished cases and controls of five Mendelian diseases. Applying PheRS to 21,701 genotyped individuals uncovered 18 associations between rare variants and phenotypes consistent with Mendelian diseases. In 16 patients, the rare genetic variants were associated with severe outcomes such as organ transplants. PheRS can augment rare-variant interpretation and may identify subsets of patients with distinct genetic causes for common diseases.

Fig. 1. Phenotype risk scores capture the diagnostic fingerprint of Mendelian disease in EHR data

Scores for six Mendelian diseases were calculated for clinically diagnosed cases and controls matched by age, sex, race, and record length. (A) Boxplots of PheRS for cases and controls for each disease. (B) Number of cases and statistical significance between cases and controls (Wilcoxon rank-sum test) for each disease. (C) Matrix of standardized differences in location (pseudomedian) of the PheRS between cases and controls (by row) and for each Mendelian disease definition (by column).

Fig. 2. Phenotypes and PheWAS for two variants associated with PheRS for cystic fibrosis and nephrotic syndrome

For phenotype grids (A) and (C), each row corresponds to a phenotype used in the PheRS; each column represents an individual who is heterozygous or homozygous (starred) for the variant. The bar on the left of the grid indicates the relative risk for each phenotype compared to wildtype. In grid (A), individuals clinically diagnosed with cystic fibrosis are enclosed by a blue box. PheWAS plots (B) and (D) show the PheWAS for the variant (Fisher’s exact p-value). The constituent phenotypes that define the PheRS are starred. All associations with p<0.001 are labeled. The horizontal red and blue lines show the Bonferroni correction threshold for an individual PheWAS and the nominal (uncorrected) p=0.05, respectively.

Fig. 3. Whole exome sequencing reveals second variants among individuals with high PheRS and demonstrates disease risk in heterozygotes

Each point represents an individual who is heterozygous or homozygous for the variant labeled on the left. The x-axis represents the z-score for the PheRS relative to what is expected given age and sex (using the residual from the PheRS). All individuals carry at least one copy of the variant indicated on the left; additional variants identified by whole exome sequencing or clinical chart review are labeled for each individual; homozygotes confirmed by sequencing are labeled “HOM.” Additional CFTR variants were ascertained from clinical testing in the EHR; all other individuals were sequenced for this study. Clinically diagnosed individuals are squares; all others are circles. Where additional variants were found, the association test from the discovery analysis was repeated after dropping individuals with a second variant (p-values generated using linear regression assuming dominant model adjusted for age and sex), and the p-value is recorded under the gene/variant label.

Fig. 4. PheRS enriches for variants with altered function in vitro

Representative Western blots (A) and mean phospho-ERK2 levels normalized to EPOR expression (B) in EPOstimulated HEK293T cells transiently transfected with wildtype (WT) versus variant SH2B3 constructs, EPOR, and JAK2. As expected, known variant SH2B3-R392E fails to inhibit EPOstimulated ERK phosphorylation. Similarly, SH2B3-E395K shows approximately 1.8-fold elevation of EPO-stimulated ERK activation at 10 min relative to wildtype SH2B3. RT-PCR analysis (C) and quantification (D) of WT versus variant splicing of SUOX and TG exons in HEK293T cells transiently transfected with empty minigene vector pET01, pET01 containing exons of interest flanked by 100 bp of intronic sequence, or negative control pIRES2-EGFP. Absolute change in the percent of exon-inclusion was −61% for SUOX-VAR and −39% for TGVAR. Means ± SEM; n = four (A, B) or five (C, D) independent experiments; unpaired twotailed t test; *p=0.003, **p<0.001.