Closing the genotype-phenotype loop for precision medicine

A core paradigm in precision genomic medicine is the ability to identify the underlying molecular mechanism of a disease in the individual patient and to match the specific therapeutic interventions deployed to this mechanism. In neoplastic disorders, where much of the fundamental disease biology can itself be efficiently studied ex vivo, genomic sequencing has enabled physicians to combine the identification of specific driver mutations, arising de novo in a tissue and inciting tumor formation^1,2. Empiric testing of specific targeted therapies and even parallel co-clinical modeling of the interaction between tumor, host and therapy have accelerated therapeutic innovation³ to significantly improve outcomes in several major neoplasms.⁴

While this paradigm has succeeded in oncology, for germline disorders that contribute to prevalent chronic cardiovascular disease (CVD), attribution of molecular mechanisms at the level of the individual patient has proven considerably more difficult. While genetic and genomic studies have allowed discovery of numerous disease genes for cardiomyopathies⁵, arrhythmias⁶, aortopathies⁷, and other disorders, recent studies demonstrate more complex relationships between genotype and phenotype in these syndromes. Genomic sequencing of large cohorts has uncovered many of the same variants previously annotated as pathogenic mutations^8,9, leading to variant reclassification and the conclusion that some reported genes reported to cause inherited heart disease are likely spurious. Conversely, a few variants that appear causal in a familial context appear to have no related phenotype in population studies¹⁰. These uncertainties emphasize that genomic sequencing deployed as a discovery or diagnostic tool in individual patients often fall short of the statistical rigor needed to establish pathogenicity with the certainty. Co-segregation of damaging variants and a trait in large numbers of individuals within families and multiple independent occurrences among individuals with related phenotypes increases that certainty. Prioritizing individual variants based only on tissue expression and the involvement of similar genes in a disease are not in themselves inappropriate, but their track record as predictors of pathogenesis indicates the need to develop more rigorous approaches to attributing causality and understanding disease mechanism at the level of the individual patient.

Among issues that confound the attribution of causality, we recognize several of particular importance. These include the limited specificity of most clinical phenotypes, focused analyses of coding sequences that exclude non-canonical splice signals, regulatory elements, and structural variants, and a profound paucity of data on potential conditioning environmental variables. To address these needs the NHLBI created the Trans-Omics for Precision Medicine (TOPMed) Program¹¹, a repository of genome sequences, RNA expression, DNA methylation, and metabolite profiles from large CVD cohorts with extensive phenotyping and outcomes data.

Closing the genotype-phenotype loop

In this issue of Circulation, McNally and colleagues present state of the art strategies to assess the causality of newly identified variants in potential CVD genes¹². The team identified a premature stop codon (R255X) in the myosin binding protein H-like (MYBPHL) gene among three relatives with dilated cardiomyopathy (DCM), prominent AV conduction disease, and atrial arrhythmias.

While this premature stop codon occurs in a gene that is structurally related to myosin binding protein C (MYBPC3), a definitive cardiomyopathy gene, there are several features of the R255X variant in MYBPHL that illuminate the complexity of establishing pathogenicity and genotype-phenotype correlations. The R255X variant occurs at significant frequency in sequencing databases from generally healthy individuals, suggesting that it is not under heavy negative genetic selection, as occurs when variants cause early onset morbid conditions. Recognizing that similar variants in MYBPC3, produce a range of phenotypes (hypertrophic or dilated cardiomyopathies or predisposition to adverse ventricular remodeling), the authors analyzed additional DCM patients but identified only one with the MYBPHL R255X variant. Next, they turned to experimental approaches, documenting high atrial but low ventricular expression of MYBPHL with an unusual punctate subcellular localization, suggesting a novel role for this ‘myofilament’ protein. Harnessing patient-derived iPS to generate cardiomyocytes, they established that the R255X allele fails to make stable MYBPHL protein. Finally, they extensively characterized mice in which the MYBPHL gene was deleted and demonstrated contractile defects in both heterozygotes and homozygote mutant mice. While MYBPHL null mice had atrial arrhythmias that are similar to those observed in the R255X patients, there were no baseline AV conduction abnormalities and only with isoprenaline infusion was there ventricular conduction delay. Together these data buttress MYBPHL as a candidate DCM/arrhythmia gene, but as is clearly (and laudably) stated by this thoughtful research team, there remains the real possibility for a chance association of the MYBPHL R255X variant and DCM in this particular family.

What would we need to know to be able to definitively attribute causality to a variant in the context of the non-specific phenotypes currently employed in clinical and preclinical medicine? Despite extensive experimental investigation and while appreciating that some of the discordances reflect species or allelic differences, the totality of data still limits the certainty with which it is possible to attribute causality. We speculate that more human data could provide additional and potentially powerful evidence. It would be vital to know how many deleterious coding variants in other genes were identified in the affected individuals in this family. It would also be of interest to know if there were non-coding abnormalities in genes already implicated in DCM with conduction disease. A specific molecular and functional defect shared between patients and experimental models would also aid in understanding the likelihood that this gene defect is truly the cause of the morbid disease in this family. Ultimately, measuring identical and specific phenotypes in humans and model systems to complement the tight genetic orthologies that we already can identify will accelerate a probabilistic understanding of causality in individual patients. Such a probabilistic understanding of mechanism will herald the arrival of precision diagnosis and the potential this brings to improve disease management.