Limitations of direct-to-consumer genetic testing for age-related macular degeneration

Advances in genetics have improved our understanding of the pathophysiology of age-related macular degeneration (AMD) and therapeutic targets along the complement cascade.^1–3 Simultaneously, public interest in at-home genetic testing is surging.^4,5 Several so-called direct-to-consumer genetic testing (DTCGT) products for estimating disease risk are commercially available. The utilization of DTCGT has increased as it provides relatively low cost information on genealogy, pharmacogenetics, cancer predisposition and disease risk estimation which can be used to guide lifestyle modifications. ⁶ For this reason, ophthalmologists, and in particular retina specialists, should be familiar with their potential benefits and limitations.

While the methods of estimating disease risk used by DTCGT companies differ, there are several similarities. Here we describe the polygenic scoring method used by 23andme (23andme, Inc., San Francisco, CA)- a biotechnology and personal genomics company that offers DTCGT for health-related conditions and ancestry testing (further details are provided on their website). ¹ Briefly, a machine learning model is built using phenotype and genotype information. The phenotype information is obtained via a self-reported consumer survey, validated with published literature on the same age, sex, and ancestry population, and checked with generalized linear models. The genotype information is obtained from consumer saliva DNA and is assayed using microarray technology. A dataset consisting of all 23andme customers who consented to participate in research is created. Exclusion criteria are applied, such as systematically and sequentially eliminating related individuals (defined as identity-by-descent of more than 700 centimorgans) keeping the individual belonging to the less prevalent phenotypic class. Next, phenotypic training validation and testing cohorts are defined in the European ancestry dataset (most genetics research participants), and hyperparameter and optimization methods are used for the non-European dataset. Then genome-wide association study is performed and used as a quality control metric. The model is trained, and a polygenic score (PGS) – an estimated value of genetic liability to having a phenotype built using regression methods. at individual-level data. Results of the PGS are reported to the consumer as binary (increased vs. not increased risk) and as an estimated likelihood of developing a condition by some target age. ⁷ While the methods appear scientific, we highlight two primary limitations to treating physicians. First, DTCGT offers consumers an incomplete assessment of risk. Second, the utility of the results – even if accurate – is unclear.

The genetics of AMD are complex. Single nucleotide polymorphism (SNP), a type of genotyping test used by many DTCGT companies, only considers a small subset of the genes since the genes must be pre-specified. For instance, just two mutations in the AMD-associated genes CFH and ARMS2/HTRA1 ⁸ which account for only around 50% of AMD heritability, ⁹ are included in the 23andMe genotyping. Orozco et al. identified 15 potential risk loci for AMD and 15 probable causal genes. ¹⁰ Also, SNP testing does not evaluate upstream regulators and modulators of AMD-associated genes or epigenetic factors, which may play a role in AMD and may therefore modulate genetic risk. ¹¹ While using whole genome sequencing (WGS) tests offered by companies like Sequencing (Sequencing, Pasadena, CA) and Nebula Genomics (Nebula Genomics, California St., CA) avoids the limitation of genotyping, it does not include phenotypic and clinical factors in risk estimation and provides many genetic variations of unknown significance, making data interpretation difficult.

The common AMD genes incorporated in genotyping based DTCGTs analysis bias the test in favour of people of European ancestry and limit generalizability of result^7,12 Association between genetic variants and AMD varies by ethnicity. Maugeri et al. ¹³ found that the Y402H (rs1061170) SNP in the CFH gene's association with AMD is weaker in Asian cohorts than in European cohorts. Kuo et al. showed that the prevalence of the C risk allele of the Y402H SNP in Japanese people is disproportionately lower than in Caucasians (7% vs 34%). ¹⁴ Frequency of the Y402H SNP has also been shown in other studies to vary by ethnicity, ranging from 35-40% in European and African populations, to less than 10% in East Asian populations. ¹⁵ Despite similar prevalence of this higher risk allele in European and African populations, rates of late AMD are lower in African people, demonstrating varying genotype-phenotype relationship across ethnicity. ¹⁵

Genetics cannot fully predict AMD risk. Smoking has been associated with a twofold increased risk of AMD. ¹⁶ High lutein and zeaxanthin diets have been associated with a reduced risk of advanced AMD. ¹⁷ Diet-gene interaction has been shown to regulate AMD risk ¹⁸ : higher omega-3 long-chain polyunsaturated fatty acid (LC-PUFA) intake is associated with a decreased risk for late AMD in individuals with low-risk CFH genotypes, an effect not seen in those lacking those low-risk variants. ¹⁸

The clinical utility of an “increased risk” result is unclear. Genetics alone is unable to address patients main concern- the risk of vision loss. Vision loss in AMD occurs in the setting of disease progression which DTCGT does not assess. The test result therefore provides little useful information for patients, and may instead lead to undue anxiety. ¹⁹

The DTCGT companies’ present privacy models are insufficient. Although most companies adhere to privacy control measures and offer varying levels of privacy protection, government access, and hacker attacks are still possible. A strict privacy model that incorporates user anonymity, decentralized data access control, access request audibility, and secure data analysis would be required. ²⁰

Results from DTCGT are challenging to interpret even by retina specialists; AMD pathogenesis is multifactorial, DTCGT risk calculation is insufficient, and preventative measures based on genetics are lacking. For these reasons, estimating the risk of AMD is not useful in guiding clinical decision-making. ²¹

Potential improvements for DTCGT in predicting AMD risk include increasing the number of genotypes under investigation and incorporating other genetic and epigenetic factors to provide a more comprehensive assessment of AMD risk. Addressing the current bias towards people of European ancestry, future research should focus on identifying and incorporating genetic variants associated with AMD in diverse populations, enhancing the accuracy and applicability of DTCGTs across different ethnic backgrounds. Integrating environmental factors and lifestyle data into DTCGTs would enable a more holistic approach to AMD risk estimation, while incorporating disease progression data and clinical parameters would improve their clinical utility and help guide clinical decision-making. These advancements, combined with the implementation of more robust privacy models and data security measures, could significantly enhance the accuracy and usefulness of DTCGTs for AMD risk prediction.

In conclusion, despite the increasing popularity of DTCGTs for risk assessment of many heritable diseases, including AMD, these tests have significant limitations. Ophthalmologists should be prepared to address these limitations, as we will all undoubtedly encounter patients seeking clarity about their DTCGT reports. Finally, as recommended by the American Academy of Ophthalmology Task Force on Genetic Testing: “Avoid routine genetic testing for genetically complex disorders like AMD and late-onset primary open-angle glaucoma until specific treatment or surveillance strategies have been shown in 1 or more published prospective clinical trials to be of benefit to individuals with specific disease-associated genotypes. In the meantime, confine the genotyping of such patients to research studies”. ²²