Next-generation phenotyping integrated in a national framework for patients with ultrarare disorders improves genetic diagnostics and yields new molecular findings

Abstract

Individuals with ultrarare disorders pose a structural challenge for healthcare systems since expert clinical knowledge is required to establish diagnoses. In TRANSLATE NAMSE, a 3-year prospective study, we evaluated a novel diagnostic concept based on multidisciplinary expertise in Germany. Here we present the systematic investigation of the phenotypic and molecular genetic data of 1,577 patients who had undergone exome sequencing and were partially analyzed with next-generation phenotyping approaches. Molecular genetic diagnoses were established in 32% of the patients totaling 370 distinct molecular genetic causes, most with prevalence below 1:50,000. During the diagnostic process, 34 novel and 23 candidate genotype-phenotype associations were identified, mainly in individuals with neurodevelopmental disorders. Sequencing data of the subcohort that consented to computer-assisted analysis of their facial images with GestaltMatcher could be prioritized more efficiently compared with approaches based solely on clinical features and molecular scores. Our study demonstrates the synergy of using next-generation sequencing and phenotyping for diagnosing ultrarare diseases in routine healthcare and discovering novel etiologies by multidisciplinary teams.

Fig. 1. Workflow in the TRANSLATE NAMSE project and phenotypes in which exome sequencing was performed.

a, Patients with a suspected rare disease were referred to a MDT and deeply phenotyped using HPO terminology. If a genetic etiology was considered likely, exome sequencing was performed. The MDT then evaluated the molecular findings and could order additional analyses for variants of uncertain significance or variants in potentially novel disease candidate genes (created with BioRender.com). b, Exome sequencing was performed predominantly in children. The main indications for exome sequencing in children were neurodevelopmental disorders. In adults, the main indications were neurological/neuromuscular disorders. In both children and adults, the least common disease categories were ‘cardiovascular’, ‘endocrine, metabolic, mitochondrial, nutritional’ (emmn) and ‘hematopoiesis/immune system’ (his). c, Phenotypic similarities between patients, as encoded according to their HPO terms, were visualized with UMAP. As reference, all OMIM diseases were included using their HPO annotations (gray background dots). For each patient, color coding indicates allocation to disease groups, in accordance with the leading clinical feature. An overlap is evident for patients in the neurodevelopmental and neuromuscular groups (aquamarine and blue clusters), which indicates high phenotypic similarity. This precludes the unequivocal assignment of these patients to a diagnostic group. The triangles indicate patients who contributed to the identification of a novel, high-evidence gene–phenotype association.

Fig. 2. Diagnostic yield of exome sequencing depends on age and disease group.

a,b, The diagnostic yield differed according to age group (adult/child) (a) and disease category (b). For all disease categories, with the exception of cardiovascular, the diagnostic yield was increased by novel DGGs and high-evidence candidate genes (dark-colored tip of the bar). The absolute number of solved cases in which a variant was found in an established disease gene is given at the bottom of each bar, and the number of solved cases attributable to a novel DGG or high-evidence candidate gene is given at the top of each bar. The entire TRANSLATE NAMSE exome sequencing cohort was considered for a and b (n = 1,577). Diagnostic yield between disease categories were compared using two-sided Fisher’s exact test. P values were adjusted by Bonferroni correction. ***P < 0.001; exact corrected P values: neurodevelopmental (ndd) versus neurologic neuromuscular P = 5.4 × 10⁻⁵, ndd versus organ abnormality P = 5.2 × 10⁻⁵, ndd versus emmn P = 5.9 × 10⁻⁴, ndd versus his P = 1.1 × 10⁻¹¹. emnn, endocrine, metabolic, mitochondrial, nutritional; his, hematopoiesis/immune system.

Fig. 3. Mode of inheritance and disease burden are dependent on autozygosity.

a, Pie chart showing the distribution of modes of inheritance (MOI) for all diagnoses (n = 510). Most disease-causing variants occurred de novo and on an autosome. At least 75% of all autosomal recessive diagnoses could have been identified by expanded carrier screening (slice). b, Box plots of autozygosity for each MOI (n = 375). Individuals are indicated by gray dots. Autozygosity was substantially increased in individuals with autosomal recessive disorders due to homozygous variants. In the box plots, the center lines indicate the median values, and the bottom and top edges of the boxes are the first (25%) and the third (75%) quartiles. The whiskers extend to the minimal and maximal data points with a maximum distance of 1.5 interquartile ranges from the edges of the box. c, Bar graphs illustrating MOI in individuals with low (<2%, n = 313) and high (>2%, n = 62) autozygosity. On the right, the autosomal dominant de novo rate has been used for normalization. Individuals with high autozygosity had a higher relative burden of recessive diseases, mainly due to the presence of homozygous pathogenic variants. The box plots present the median as the center line, the upper and lower quartiles as box limits, and 1.5× the interquartile range as the whisker length (in the style of Tukey). AD, autosomal dominant inheritance, variant inherited or of unknown origin; AD (de novo), autosomal dominant inheritance with de novo variant; AR (comp het), autosomal recessive inheritance with compound heterozygous variants; AR (hom), autosomal recessive inheritance with homozygous variant; mt, mitochondrial inheritance; XL, X-linked inheritance.

Fig. 4. Most variants identified in TRANSLATE NAMSE exome sequencing cohort cause ultrarare disorders that were first associated with a gene in the last decade.

a, Comparison of the number of (likely) pathogenic variants per gene in TRANSLATE NAMSE relative to the frequency of submission of (likely) pathogenic variants to ClinVar. Genes are ordered from left to right according to a decreasing frequency of ClinVar submissions. The black line corresponds to the complementary cumulative distribution (1 − CDF; cumulative distribution function) of ClinVar submissions. Diagnostic variants in TRANSLATE NAMSE (counts displayed on the right axis) were plotted as dots above their respective gene and in the color corresponding to the year in which the gene was first described as being associated with the respective disease. b, Variant counts in TRANSLATE NAMSE in genes with high (first quartile, Q1) to low (Q4) counts of submissions per gene in ClinVar. The genes in Q1–Q4 each cover approximately 1/4 of the submissions of likely or confirmed pathogenic variants to ClinVar, as shown on the x axis in a. Variants in the same gene are grouped in horizontal blocks. c, Bar graph showing the number of variants relative to the time interval in which the gene was first described as being associated with the respective disease. Note that 59 genes listed in the recommendations for reporting of secondary findings (version 2) of the ACMG were excluded from the analyses to counteract potential biases in ClinVar due to submissions of secondary findings. TNAMSE, TRANSLATE NAMSE; vars, variants.

Fig. 5. Machine learning identifies features relevant to the diagnostic yield and can support variant prioritization.

a, The coefficient paths of regression analysis using the LASSO are shown. Only features that are included in the final model and are present in at least 5% of the cases that were used for training are depicted. The more to the left [lower ln(λ)] a coefficient path starts to deviate from the x axis, the more informative the corresponding feature is in terms of predicting the diagnostic yield. Features with positive coefficients increase the diagnostic yield. In contrast, features with negative coefficients render a monogenic cause less likely. For example, dysfunction of higher cognitive abilities and ataxia are associated with a higher diagnostic yield (clinical features are colored according to their higher-order HPO groups; for details, see Supplementary Note). An algorithm to predict the diagnostic yield (YieldPred) was developed on the basis of these data and can be found online (https://translate-namse.de). b, The performances of variant prioritization approaches were compared. All disease-associated genes were ranked using the respective variant prioritization method. Subsequently, the proportion of cases detected with the correct disease-associated gene (sensitivity) was shown as a function of the number of disease-associated genes considered, beginning at the top score. The following four approaches for variant prioritization were tested in solved cases from the PEDIA cohort (n = 94): (1) only a molecular pathogenicity score (CADD) with top-10 accuracy of 48%; (2) feature-based score (CADA) in addition to CADD with top-10 accuracy of 68%; and (3 and 4) a gestalt score from facial image analysis (GestaltMatcher) alone or in addition to both CADD and CADA referred to as PEDIA score with top-10 accuracy of 82%. Note that the bold lines indicate the observed top-k accuracy and bootstrapped 95% CIs are indicated by the lighter shading around the lines. MRI, magnetic resonance imaging; abn., abnormality; con., congenital; dysf., dysfunction; psych., psychiatric; sym., symptoms; sec, secondary.