New research shows that biases in genetic studies of Alzheimer’s disease can distort risk predictions and derail medical progress, but a new approach offers hope of uncovering the truth.
Study: Pervasive biases in proxy genome-wide association studies based on parental history of Alzheimer’s disease. Image credits: TanyaJoy / Shutterstock.com
Genome-wide association studies (GWASs) are commonly used to study the genetic basis of various diseases, including Alzheimer’s disease (AD). To increase their power to capture associations, it has been proposed to include parental or family history of AD to improve sample size, a process sometimes called GWAS-by-proxy (GWAX).
Despite this progress, a recent study in the journal shows Nature genetics found significant and widespread causes of errors in GWAX that require urgent correction.
The importance of GWAX biases
The extensive use of GWAX in AD research is attributed to its ability to collect data from middle-aged populations to estimate the incidence of AD and identify risk factors for late sequelae. In fact, GWAX facilitated the discovery of 75 genetic loci for AD risk, expanding knowledge of the disease at the molecular level.
Previous studies have shown that GWAX increases the power of GWAS due to the comparable effect size estimates for the top single nucleotide polymorphisms (SNPs) and high genetic associations with GWAS. However, the GWAX methodology is often associated with measurement errors.
Unless clearly defined, children of parents with dementia may not distinguish between AD and non-AD in their reports, despite showing different genetic backgrounds. Furthermore, combining GWAS and GWAX in the analysis without compensating for heterogeneity may reduce heritability estimates.
Persistent and common prejudices
AD GWAX has been associated with persistent selection biases, such as survival and participation biases, in several studies. Although these biases are often observed in epidemiological and clinical studies, they are rarely discussed in genetic research.
Selection bias refers to any bias that arises during the process of selecting research participants, leading to systematic differences between the selected individuals and the reference population..”
Selection bias is common in large-scale biobanks, such as the United Kingdom Biobank. GWASs based on the UK Biobank and follow-up studies, including Mendelian randomization (MR) and genetic correlation studies, are thus prone to confusion.
Indeed, the current study confirmed the presence of non-random selection and reporting by participants, demonstrating genetic associations between survey participation and parental health history awareness.
GWAX causes conflicting genetic associations
An analysis of genome-wide data showed only seven traits associated with both GWAS and GWAX. In comparison, GWAS-based studies reported a protective effect of education, while GWAX-based analyzes showed the opposite.
Ten studies observed similar patterns for genetic associations in AD education. Case-control studies and family history-based proxy studies showed negative and positive risk associations, respectively, while meta-analyses of both GWAS and GWAX showed intermediate results.
Education is a social factor that promotes longevity, improves parent-child relationships and increases general health awareness. However, these factors can also influence AD risk; therefore, education and cognition contribute to several biases observed in AD genetic studies.
GWAX distorts epidemiology and ical genetic studies
In the current study, researchers conducted two epidemiological applications that examined the genetic associations between cognition and AD. Initially, the researchers predicted cognition in later life using a polygenic risk score (PRS) approach for AD based on GWAS, GWAX, and a mixed meta-analysis, all of which generated negative associations.
The second application involved the use of an MR approach to estimate the magnitude and direction of the effect of education on AD.
GSUB for AD GWAX and possible biases
The authors used a novel GWAS-by-subtraction (GSUB) approach to investigate potential mechanisms for the bias in AD GWAX, which incorporates improved implementation strategies. GSUB also measures spurious AD GWAX effects, which are genetic signals that arise from confounding factors that influence family history rather than AD itself.
Using GSUB, the non-AD component was found to be negatively associated with multiple health outcomes, suggesting survival bias, as people reporting parental AD had long-lived parents. Protective alleles associated with conditions that promote longevity would thus misleadingly increase the risk of AD.
Furthermore, children aware of parental AD may be closer to their parents, indicating the confounding effect of socioeconomic stratum and family structure. GSUB also identified that the UK Biobank survey may not have distinguished between parental AD and non-AD dementia.
Conclusions
Our findings have important implications because they reveal an urgent, pervasive, yet under-researched problem hidden in plain sight.”
The researchers of the current study report the differences between GWASs and GWAXs, which have the potential to significantly mislead efforts focused on diagnosis, treatment, and drug development processes. While GSUB is a promising approach, it requires validation and adjustment for biases present in case-control GWASs.
Journal references:
- Wu, Y., Sun, Z., Zheng, Q., et al. (2024). Pervasive biases in proxy genome-wide association studies based on parental history of Alzheimer’s disease. Nature genetics. doi:10.1038/s41588-024-01963-9.
- Pederson, E. M., Wimberley, T., and Vilhjalmmson, B. J. (2024). A cautionary tale for Alzheimer’s disease GWAS by proxy. Nature Genetics. doi:10.1038/s41588-024-02023-y.