Putative Loss-of-Function Variants More Common Than Thought

Sequencing the genomes of healthy people uncovered the presence of many more rare diseases than expected. More than three percent of the U.S. population may have a genetic condition compared to previous estimates of less than 0.02 percent, according to a study published June 4 in the American Journal of Human Genetics. The authors say this study shows that genome sequencing information can dramatically improve prediction of disease. Furthermore, they say the study demonstrates the feasibility of "iterative phenotyping."

"Today, we tend to deliver medical care based on the expected response of the average patient," said National Human Genome Research Institute (NHGRI) Director Eric Green, M.D., Ph.D., in a statement. "Eventually, we want to deliver medical care based on individual genomic differences that enable more precise ways to prevent and treat disease. These findings move us closer to that reality."

It is recognized that genome analysis reveals much higher levels of putative lossof- function (pLOF) than true LOF estimates (800 versus 100 variants per person). pLOFs can include nonsense, frameshift, and splice site alterations. Understanding the clinical implications of pLOF is complicated in real-life sequencing scenarios, as bioinformatics solutions are not yet capable of delivering accurate predictions of pathogenicity. Efforts are further complicated by limitations in discovery of novel phenotypic associations due to the use of individuals selected for predefined phenotypes in most genotype-phenotype correlation studies.

"With the increasing use of next-generation sequencing technologies for predictive medicine, it is critical to be able to predict the consequences of pLOF, especially in individuals without preexisting clinical diagnoses," write the authors.

In the present study, the researchers sequenced the exomes of 951 participants of the ClinSeq cohort (aged 45 to 65 years). Consequences of pLOF variants were characterized using iterative phenotyping in which participants were invited to a follow-up clinic visit if a family history was insufficient to confirm or rule out a diagnosis. Sequencing data was filtered for pLOF variants in genes likely to cause a phenotype in heterozygotes (with an autosomal-dominant inheritance pattern).

After filtering for quality of the sequencing data, researchers identified an average of 100,664 variants per individual, with an average of 484 pLOF variants per individual. Further filtering for pLOF variants that were considered highly likely to cause a phenotype in the heterozygous state yielded 82 variants in 103 participants. Of the 79 individuals available for follow-up assessments, the overall yield of positive phenotypes was 43 percent (n=34). For the 34 participants with findings or family histories that could be attributed to the variant, 28 variants were detected in 18 genes. An additional two participants had indeterminate findings (2 variants in 2 genes), while 43 had no personal findings and a negative family history for the trait (34 variants in 28 genes).

"Although our positive predictive rate was 43 percent, it should be emphasized that these data altered the risk of a rare, autosomal-dominant disorder in these 79 participants from baseline (1/500 to 1/500,000) to approximately half," conclude the authors, led by Jennifer Johnston, Ph.D., from NHGRI. "Although it is true that not all of the phenotypes detected here are medically actionable, this study serves as a proof-of-principle that there might indeed be predictive value in healthy genomes and exomes, once our mutation prediction algorithms improve and broaden to encompass all genes and many mutations."

Of the 42 individuals who underwent follow-up phenotyping, 21 individuals (variants in 18 different genes) were negative on examination. Another 20 individuals with harmful mutations and a detectable phenotype upon examination didn't know they had a genetic condition. These conditions ranged from mild to potentially serious. Some of the findings included:

• Mild features such as short digits with a HOXD13 variant; deafness with a KCNQ4 variant; dystonia with a SGCE variant; and blistering of the feet with a KRT16 variant.
• Three variants resulted in biochemical phenotypes of protein S, factor XI, or TNFRSFB, but LOF variants in these genes are often non-penetrant for severe disease features.
• Some undiagnosed phenotypes with a more significant risk of morbidity and/or mortality to participants, but typically with later onset, included lipodystrophy with a PPARG variant; decreased lung function with a SFTPC variant; and left ventricular non-compaction with a MYH7 variant.
• Nineteen participants had variants in genes previously identified cancer susceptibility variants (BRCA1/2, FLCN, MSH6, PMS2, RAD51D, SDHC, XRCC2), but only six of these individuals had a clear family history of the associated cancers.

"For less-severe phenotypic features that might be present but underdiagnosed in the general population, such as hearing loss, confirming the causation of the variant is difficult and over-interpretation is possible," write the authors. "For individuals without a clear family history of disease, variant identification and interpretation is more difficult because it is possible that the identified variant is not causative, and non-penetrance must always be considered."

Fifteen identified variants were in genes on the American College of Medical Genetics and Genomics' list of genes to be considered for return of incidental findings. pLOF variants in these genes were identified in a total of 23 individuals, 16 of which were evaluated for phenotypic features. Seven of the 16 were positive for associated phenotypic findings and/or positive family history, while nine individuals did not have associated phenotypes or a positive family history. The researchers say determining whether these individuals are nonpenetrant versus the variant being non-causative is "crucial" in the decision to return these variants.

Takeaway: While this study adds to the hope that genomic information can be used predictively, much more needs to be understood about the nature of penetrance and causative nature of genetic variants.