The Changing Landscape of Genetic Risk: Age-Related Effects of Polygenic Scores for Suicidal Behaviors

Technical Summary: The Changing Landscape of Genetic Risk: Age-Related Effects of Polygenic Scores for Suicidal Behaviors

Problem Statement
Suicidal thoughts and behaviors (STBs) represent a critical public health challenge with significant mortality and morbidity. While twin and family studies have established moderate heritability for STBs (estimates ranging from 0.17 to 0.55), and recent genome-wide association studies (GWAS) have identified specific genetic loci and polygenic risk scores (PRSs), a gap remains in understanding how these genetic risks manifest across the lifespan. Most existing genomic research treats genetic risk as a static factor, despite evidence that STB prevalence and manifestation vary significantly from adolescence through adulthood. This study addresses the question of whether the predictive power of current PRSs for suicide ideation (SI) and suicide attempt (SA) varies by developmental stage, specifically examining if genetic influences are stronger during adolescence or later in young adulthood and mid-life.

Methodology
The study utilized data from the National Longitudinal Study of Adolescent to Adult Health (Add Health), a nationally representative cohort followed over approximately 22 years across five waves (Wave 1: 1994–1995, ages 12–18; Wave 5: 2016–2018, ages 33–43).

• Genotyping and PRS Calculation: Genotyping was performed on saliva samples collected at Wave 4 using Illumina Human Omni1-Quad and Omni-2.5 Quad BeadChips. After quality control, the analytic sample included 9,974 participants of European-like ancestry. Polygenic Risk Scores for Suicide Ideation (SI-PRS) and Suicide Attempt (SA-PRS) were calculated using summary statistics from the most recent GWAS (9, 8) via PRS-CS, which utilizes a continuous shrinkage prior to weight SNPs across the entire p-value spectrum. PRSs were residualized for the first 10 ancestry principal components.
• Phenotypes: Suicide ideation and attempts were assessed at all five waves. Ideation was defined as seriously thinking about suicide in the past 12 months. Attempt was defined as any attempt in the past 12 months. Notably, at Waves 1–3, attempt assessment was contingent on endorsing ideation, whereas at Waves 4–5, attempts were assessed independently.
• Statistical Analysis:
  1. Logistic Regression: Cross-sectional logistic regressions were performed at each wave to assess the association between SI-PRS/SA-PRS and the binary outcomes of ideation and attempt, adjusting for sex, socioeconomic status (SES), and ancestry. P-values were adjusted for multiple comparisons using the Benjamini-Hochberg False Discovery Rate (FDR) method.
  2. Structural Equation Modeling (SEM): An exploratory bivariate autoregressive SEM was fitted using OpenMx in R. This model utilized Full-Information Maximum Likelihood (FIML) to handle missing data. It modeled the longitudinal autoregressive structure of the data, the contingency of attempts on ideation at Waves 1–3, and the residual covariances at Waves 4–5. The model simultaneously estimated the effects of SI-PRS and SA-PRS on all ten phenotypic measures (ideation and attempt across five waves) while adjusting for sex and SES as definition variables.

Key Results
The study found that the association between polygenic risk and STBs was not uniform across development but was concentrated in later waves corresponding to young adulthood and early mid-life.

• Prevalence: Suicide ideation was most prevalent in adolescence (Waves 1 and 2: ~13–15%) and decreased in adulthood (Waves 3–5: ~7–8%). Suicide attempts were ~4% in adolescence (contingent on ideation) and dropped to ~1–2% in adulthood (assessed independently).
• Logistic Regression Findings: After FDR correction, significant associations were observed only in later waves:
  • Wave 4 (ages 24–32): SI-PRS was significantly associated with suicidal ideation (OR = 1.18, FDR q = 0.008).
  • Wave 5 (ages 33–43): SA-PRS was significantly associated with suicidal ideation (OR = 1.18, FDR q = 0.032).
  • No significant associations were found between PRSs and STBs at Waves 1–3 (adolescence and transition to young adulthood).
• SEM Findings: The autoregressive model confirmed the logistic regression results. After adjusting for covariates and model structure, three PRS effects remained statistically significant (FDR corrected):
  1. SI-PRS associated with Wave 4 ideation (β = 0.08).
  2. SA-PRS associated with Wave 5 ideation (β = 0.10).
  3. SA-PRS associated with Wave 5 attempts (β = 0.17).
  • The model-implied correlations for these pathways were modest (ranging from 0.08 to 0.15).
• Covariates: Female sex and low SES were consistently associated with higher mean liabilities for suicidal phenotypes across all waves.

Key Contributions

1. Developmental Specificity of Genetic Risk: The study provides empirical evidence that current PRSs for STBs, which are derived primarily from adult GWAS samples, have stronger effect sizes during young adulthood and early mid-life (ages 24–43) compared to adolescence (ages 12–24).
2. Methodological Integration: The paper combines cross-sectional logistic regression with a complex, longitudinal autoregressive SEM to disentangle developmental trajectories from static genetic effects, accounting for the specific assessment protocols (contingent vs. independent) used across the Add Health waves.
3. Assessment of Ideation-Action Contingency: The data revealed that at Waves 4 and 5, suicide attempts occurred in individuals who did not report recent ideation, challenging the assumption that attempts are strictly contingent on recent ideation and suggesting that assessments should capture attempts independently of ideation endorsement.

Significance and Claims
The authors claim that their findings highlight a critical limitation in the current application of PRSs: they are less effective at identifying genetic risk during adolescence. The study suggests that the "changing landscape" of genetic risk implies that genetic influences on STBs may be more pronounced or detectable in young adulthood and early mid-life.

The paper posits that these results underscore the necessity for expanded GWAS efforts that include age-stratified samples, particularly adolescents, to capture developmental-specific genetic risks. The authors state that such advancements are required to improve the predictive power of PRSs across the lifespan and to inform targeted, age-specific prevention strategies. They explicitly note that the current PRSs, based on adult-heavy discovery samples, may miss adolescent-specific genetic vulnerabilities. The study concludes that understanding these developmental windows is essential for better leveraging genetic information for risk assessment and intervention, though they caution that formal tests of age moderation or mechanistic explanations were not conducted in this specific analysis.