Causality in Genetics: The Gradient of Genetic Effects and Back to Koch's Postulates of Causality

Genetics provide a robust scientific platform for establishing a relationship between a cause, for example, a genetic variant, and an effect, such as a disease phenotype. And yet, causality in genetics is a probabilistic and rarely a deterministic certainty. The causal relationship between a genetic variant and a phenotype is provisional to the conditions and the environment, such as the genetic backgrounds, in which the causal variants and the phenotype operate. The degree of probabilistic causality is in part determined by the effect sizes of the genetic variants, which typically follow a gradient ranging from minimal to large ^{1, 2}. Genetic variants (single nucleotide variants, small insertion/deletions and the structural variations) with large effect sizes are highly penetrant, exemplified by variants that are responsible for the single gene disorders with Mendelian patterns of inheritance. In such circumstances, the cause and the effect relationship is best analyzed through co-segregation and linkage analyses, whereby an LOD (Logarithm of Odds) score of 3 or greater is considered a strong evidence of a genetic linkage but not necessarily a definitive indicator of causality. On the opposite end of the spectrum are the variants that exert modest effect sizes and hence, are less penetrant. In such cases, establishing a cause and an effect relationship is more challenging, because such low penetrant variants typically do not show a clear co-segregation with the phenotype and are often found in the general population as well. Establishing the causal role of the variants is more challenging in small families, and even more so, in a single individual. Further compounding the ascertainment of causality is the influence of the genetic backgrounds (modifier genetic variants) and the environmental factors, which are expected to exert greater influence when the effect sizes of the causal genetic variants are rather small. Hence, a genetic variant with a small or moderate effect size might be penetrant in certain genetic backgrounds but not in others. Notwithstanding the effect sizes of the causal genetic variants, the modifier variants and the environmental factors contribute to penetrance of the causal variants and phenotypic variability of the disease.

In the background of the above concepts, the author suggests categorization of the genetic variants in the human genome, with regards to their pathogenic role in human diseases into five groups ²:

• 1. Disease-causing variants: This category encompasses genetic variants that benefit from the most robust evidence of causality, typically achieved through genetic linkage analysis in large families. The variants generally exhibit a high penetrance, exert large effect sizes, and are typically responsible for the single gene disorders with Mendelian patterns of inheritance. Hence, when present in an individual genome, they commonly lead to the linked disease, albeit, the severity of the phenotype is also influenced by the modifier variants, genomic factors and other determinants. Functional and mechanistic lend further support to the causal role of these variants in the pathogenesis of the linked phenotype. The disease-causing variants are rare in the population and even rarer in an individual genome. Examples of the well-established disease-causing variants are the non-synonymous or frameshift mutations in the MYH7 and MYBPC3 genes, which encode sarcomere proteins β-myosin heavy chain and myosin binding protein C3, respectively which are established causes of hereditary cardiomyopathies³. Nevertheless, not all protein altering variants in the genes known to cause single gene disorders should be considered disease-causing variants. For example, MYH7 and MYBPC3, which are among the best characterized genes for human hereditary cardiomyopathies^{3, 4}, contains a large number of non-synonymous variants that have not been linked to a clinical phenotype (http://evs.gs.washington.edu/EVS/). Not surprisingly, the population frequency of protein altering and likely pathogenic variants in the sarcomere proteins is higher than the prevalence of the hereditary cardiomyopathies in the general population^{5, 6}. Finally, it is also important to recognize the shortcomings of in silico algorithms in the accurate identification of the pathogenic variants, which is platform dependent with a modest agreement among multiple platforms ⁷. For example, only 1% of the non-synonymous variants are consistently predicted to be functional when analyzed by multiple commonly used algorithms⁷. Thus, prediction of a pathogenic role for a variant based on a single in silico platform has a high false positive rate⁷. In view of the above, categorization of a genetic variant as a disease-causing variant must be based on strong human molecular genetic data, such as linkage evidence, typically in complementation with the biological and mechanistic studies.

• 2. Likely disease-causing variants: This category of the variants are defined as the variants that show evidence of an association with the phenotype of interest along with strong mechanistic data that implicate them in the pathogenesis of the phenotype of interest. However, these variants unlike the first category do not benefit from robust human molecular genetic data, such as the linkage evidence in large families. The “likely-disease causing variants” impart the second largest effect sizes after the disease-causing variants. They often show incomplete penetrance, i.e., do not show a perfect co-segregation with the phenotype in the families. They are very rare in an individual genome and might be found in the general population, albeit with a lower population frequency than in those with the phenotype. Accordingly, the “likely disease-causing variants” are enriched in those with the disease of interest^{8, 9}. In addition to the human molecular genetic data, evidence for the causal role of these variants must be supported by the mechanistic data, such as induction of the intended phenotype in a model organism upon introduction of the variant and reversal of the phenotype upon its removal or shutting down expression of its protein. Therefore, human molecular genetic and mechanistic data are necessary to consider a variant a “likely disease-causing variant”. Despite the genetic and mechanistic evidence, the causal role of this category of variants is less certain than the “disease-causing variants”.

• The three TRIM63 variants (p.A48V and p.I130M and p.Q247*) recently identified in small families and index cases with HCM are considered as “likely disease-causing variants”⁹. The p.Q247* variant (rs148395034), which is a premature stop-codon mutation, was identified in two small families with HCM⁹. Hence, evidence of genetic linkage could not be established because of the small size of the families. The p.Q247* is a loss-of-function variant and has a population frequency of 0.001 in the Caucasians and <0.0001 in the African Americans (http://evs.gs.washington.edu/EVS/ and http://browser.1000genomes.org). In accord with the low population frequency of this variant in the general population, Ploski et al. have identified the p.Q247* variant in a 22-year old Polish professional soccer player who underwent genetic screening by whole exome sequencing (WES) because of the prolonged QTc interval of 470 msec and an episode of 8-beat non-sustained ventricular tachycardia at a heart rate of 150 bpm ¹⁰. Neither the probands nor his 47-year old mother had evidence of HCM¹⁰. This finding, while not unanticipated based on the known population frequency of this variant, raises the question of the causality of this variant in HCM.

• One potential explanation for the absence of HCM in the carriers of the p.Q247* variant is incomplete and age-dependent penetrance of this variants, as also noted by Ploski et al. The variant was originally identified in the older individuals with HCM⁹. In addition, phenotypic expression of the p.Q247* variant might be influenced by a number of other factors including the genetic background of the individuals. Despite detection of this variant in the general population, several lines of evidence support its pathogenic role in HCM, as described by Chen et al⁹. Rare TRIM63 variants were enriched in the HCM population, and the p.Q247* variant exhibited a total loss of function (E3 ubiquitin ligase activity). When introduced into mice using an inducible system, it resulted in cardiac hypertrophy with preserved systolic function, a phenotype resembling HCM in humans⁹. Moreover, shutting down expression of the mutant protein led to reversal of the phenotype. The TRIM63 variants p.A48V and p.I130M (rs140523053 and rs377334933, respectively), also implicated in HCM, are also rare, and each has with a population frequency of <0.001. They are also functional variants, and considered pathogenic in the cell and animal models⁹. Finally, one has to consider the pre-test likelihood of the disease in the clinical interpretation of the genetic findings. Identification of the rare variants in a disease-population with a higher pre-test likelihood has greater clinical implications than an incidental finding of the variant in the general population (a lower pre-test likelihood of the disease). Nevertheless, in spite of the genetic and mechanistic evidence and enrichment of the rare variants in the TRIM63 gene in the HCM population, TRIM63 variants are categorized as “likely-disease causing variants”⁹.

• 3. Disease-associated variants: Disease-associated variants are defined as variants that are associated with the disease but do not benefit from genetic or mechanistic data to implicate them in the pathogenesis of the disease. They might simply be in linkage disequilibrium with the actual causal variants. These variants are typically common variants and exert small effect sizes. The vast majority of the variants associated with the phenotype through genome-wide association studies (GWAS) of the complex traits are placed in this category (www.genome.gov/gwastudies). The causal role of the “disease-associated variants”, identified through GWAS or otherwise, in the pathogenesis of the phenotype of interest is typically unsettled, even though some might be functional or reside within genes involved in the pathogenesis of the phenotype. Many are expected to be simply DNA markers in linkage disequilibrium with the actual pathogenic variants. To assess the causality of these variants one has to consider the strength of the evidence of the observed association with the phenotype, replication of the observed association in independent study populations, biological plausibility, and gene-dose effects. Finally, experimentation is necessary to discern the potential causal role of this category of the variants in the pathogenesis of the associated phenotype.

• 4. Functional variants with unknown clinical consequences: This category of the variants are defined as variants that exert biological effects, such as influencing the mRNAs and proteins levels of their respective genes but have not been associated with the clinical phenotypes. Each human genome contains approximately 13,500 non-synonymous and a large number of regulatory variants of which several thousands are predicted to be functional^{11, 12}. The findings of the ENCODE (Encyclopedia Of DNA Elements) project point to the presence of very large number of regulatory element in introns and intergenic regions and hence, putative functionality of the variants residing in these regions of the genome¹². A considerable number of the variants located in the regulatory regions, through cis or trans mechanisms affect mRNA levels of their corresponding genes¹³. It is also estimated that each human genome to contain approximately 100 loss-of-function variants including about 20 homozygous loss-of-function variants that totally inactivate the genes in which they reside¹⁴. However, the vast majority of the putative or known functional variants is not associated with a specific clinical phenotype and remains clinically orphans. The prevalence of these variants in an individual genome and their population frequencies are expected to be high but their effect sizes to be small.

• 5. Variants with unknown biological or clinical significance: This category comprises the yet-to-be characterized variants, which include the vast majority of the approximately 4 million sequence variants in each human genome. They are not known to have biological functions and have not been associated with a phenotype. Evidently, characterization of these variants might lead to their re-classifications.

Genetic diversity of the humans, which is in part due to rapid expansion of the human population during the last 400 generations since the agricultural revolution, further complicates identification of the causal variants¹⁵. Given the error rate of DNA replication and editing machinery set at 1×10⁻⁸ per nucleotide, each meiosis (genome duplication) introduces ~ 30 de novo variants^{11, 16, 17}. Accordingly, the explosive population growth during the last 10,000 years or so has led to introduction of a very large number of new and hence, by definition, rare variants into the population genome pool, which have not been subjected to selective purification adequately. Therefore, it is not surprising that the vast majority of the genetic variants in the human population are rare and typically population-specific^{7, 18}. And yet, the rare variants comprise ~ 95% of the putatively functional variants⁷. Consequently, the private nature of each genome poses considering challenges in identification of the causal variant for a specific phenotype in a given individual. Likewise, a private or rare variant might co-segregate with the phenotype in a family by chance alone, albeit the chance of a random co-segregation inversely relates to the size of the family, as each informative meiosis reduces the chance of a random co-segregation by ~ 50%. Thus, it not surprisingly likely pathogenic variants in genes previously associated with hereditary cardiomyopathies or cardiac arrhythmias are occasionally identified in the exomes of the apparently normal individuals^{10, 19, 20}. This observation also extends to other Mendelian diseases, such as the maturity-onset diabetes mellitus of the young (MODY), whereby the pathogenic variants, defined as rare, conserved and protein damaging variants, are also identified in the healthy individuals in the general population²¹. The discoveries point to the challenges encountered in the accurate identification of the causal variants, simply based on population frequencies, or segregation with the phenotype in small families, evolutionary conservation, the effect on the protein structure or biological functions. The recent findings also highlight potential contributions of the compound “causal” genetic variants to the pathogenesis of the phenotype, even in single-gene disorders^22–26. Collectively, these discoveries occasionally challenge the conventional cause and effect relationship and point to the complexity of genetic and non-genetic etiological determinants of the clinical phenotype^2,21.

Given the plethora of the rare variants in each genome, how to determine the causal role of such variants in human diseases? The first and an important step in establishing genetic causality is to provide robust genetic evidence, typically through linkage analysis in large families or through showing a statistically significant enrichment of the rare variants in the candidate gene of interest in individuals with the phenotype (a gene-centric approach). In the absence of the genetic evidence, in vitro functional studies, while supportive, alone are inadequate to discern the functional variants from the disease-causing or likely disease-causing variants. The model organisms are valuable in supporting the causal role of the candidate variants, if introduction of the candidate variants into the model organism results in a phenotype that resembles the intended human phenotype. However, the approach is not without its shortcomings and alone is an inconclusive evidence of causality²⁷. Finally, resolution or reversal of the phenotype upon turning off the candidate variant protein in an animal model offers additional line of evidence in support of the causality. One might argue that the above conditions are somewhat analogous to the Koch's postulates of causality, which Walter Koch's envisioned to establish the etiology of infection diseases more than a century ago, as they apply to genetic disorders²⁸. The analogous elements of the Koch postulates in genetic causality might be considered as follows:

1. The causal variants must be found and enriched in the families or cases with the phenotype

2. The candidate causal variants must be functional and pathogenic (novel or rare, conserved, and protein altering)

3. Introduction of the variants into an experimental model should cause a phenotype that resembles the phenotype in the humans

4. Removal (deletion or silencing) of the candidate causal variants should reverse the phenotype.

The TRIM63 variants, described by Chen and colleagues, met these criteria and hence, are considered “likely disease-causing variants” for HCM, as were presented in the original article⁹.