Skip to content

Rare-Variant Genetic Architecture of Depression

Summary

Whole-exome sequencing (WES) of 320,356 UK Biobank participants demonstrates that rare protein-truncating variants (PTVs) in constrained genes ($pLI \ge 0.9$) significantly increase depression susceptibility across multiple phenotype definitions.[1] Rare PTV burden ($OR = 1.16$) and common-variant polygenic risk scores ($OR = 1.34$) act independently and additively on liability without multiplicative interaction.[1] Crucially, pairwise rare-variant genetic correlations across phenotypic definitions display structural discordance ($ARI = -0.167$) relative to common-variant correlation spaces, reflecting distinct evolutionary pressures on rare versus common risk alleles.[1]

Phenotypic Heterogeneity & Seven Case Definitions

Because psychiatric traits exhibit marked clinical heterogeneity depending on ascertainment, Tian et al. evaluated exome-wide burden across seven case definitions in UK Biobank:[1]

  1. EHR (Electronic Health Record): Primary care and hospital episode records ($N_{\text{case}} = 28,623$).[1]
  2. GPpsy: GP consultations for depression/anxiety ($N_{\text{case}} = 104,188$).[1]
  3. Psypsy: Psychiatrist consultations ($N_{\text{case}} = 30,230$).[1]
  4. SelfRepDep: Self-reported depression at interview ($N_{\text{case}} = 37,294$).[1]
  5. DepAll: Broad combined definition ($N_{\text{case}} = 119,747$).[1]
  6. lifetimeMDD: DSM-5 lifetime major depression via CIDI-SF questionnaire ($N_{\text{case}} = 32,875$).[1]
  7. MDDRecurr: Recurrent MDD via CIDI-SF questionnaire ($N_{\text{case}} = 19,812$).[1]
UK Biobank Depression Phenotype Ascertainment & Architecture
├── Broad / Clinical Health Records (EHR, GPpsy, Psypsy, DepAll)
│   └── Enriched for rare PTV burden (EHR OR = 1.17, p = 3.57e-18)
└── Structured Questionnaire Cohorts (lifetimeMDD, MDDRecurr)
    └── Highly enriched for common polygenic background; discordant in rare-variant space

Exome-Wide Rare Variant Burden

Rare variants ($\text{MAF} < 1.0 \times 10^{-5}$) were classified by functional severity into protein-truncating variants (PTV: stop-gained, frameshift, essential splice-site), missense variants partitioned by MPC score (MPC $> 2$ for damaging, $1 < \text{MPC} \le 2$, $0 < \text{MPC} \le 1$), and synonymous controls:[1]

  • Concentration in Intolerant Genes ($pLI \ge 0.9$): Exome-wide PTV burden in loss-of-function-intolerant genes was strongly associated with depression risk, reaching peak significance in EHR-defined cases ($OR = 1.17, 95\%\text{ CI } 1.13–1.21, P = 3.57 \times 10^{-18}$).[1] Damaging missense burden ($\text{MPC} > 2$) was also significant ($OR = 1.08, 95\%\text{ CI } 1.05–1.23, P = 8.52 \times 10^{-6}$).[1]
  • Absence of Signal in Tolerant Genes ($pLI < 0.9$) & Synonymous Variants: No significant association was observed for PTV or missense burden in $pLI < 0.9$ genes or for synonymous variants ($P > 0.05$), confirming that associations represent genuine functional constraint rather than population stratification or inflation.[1]

Additive Polygenic and Rare-Variant Model & Variance Decomposition

Joint logistic regression models combining a common-variant PRS (derived via PRS-CS from a meta-analyzed GWAS of $N = 157,304$ cases) and rare variant carrier status demonstrated strict additivity:[1]

Liability Variance Decomposition (EHR Depression)

$$\text{Liability Variance Explained } (h^2_{\text{liability}}) = \underbrace{\text{PRS}{\text{common}}}} \;+\; \underbrace{\text{PTV{\text{rare}}}} \;+\; \underbrace{\text{Missense{\text{damaging}}}$$

  • Common-Variant PRS: $OR = 1.34$ per SD increase ($95\%\text{ CI } 1.31–1.37, P = 4.77 \times 10^{-184}$), explaining 2.51% of liability variance.[1]
  • Rare PTV Burden: $OR = 1.16$ per allele ($95\%\text{ CI } 1.12–1.21, P = 4.26 \times 10^{-17}$), explaining 0.22% of liability variance.[1]
  • Rare Damaging Missense Burden: $OR = 1.07$ per allele ($95\%\text{ CI } 1.04–1.11, P = 6.26 \times 10^{-5}$), explaining 0.06% of liability variance.[1]
  • Zero Epistatic Interaction: Testing interaction terms ($\text{PRS} \times \text{Rare Carrier}$) yielded no significant effects across all seven definitions ($P > 0.05$), proving that rare damaging variants confer susceptibility independently of common polygenic background.[1]

Discordance Between Rare and Common Genetic Correlation Spaces

Using Burden Heritability Regression (BHR), pairwise rare-variant genetic correlation matrices ($r_g$) were estimated across the seven depression definitions and compared to common-variant $r_g$ matrices.[1]

To quantify structural alignment between rare and common clustering, the Adjusted Rand Index ($ARI$) was computed:[1]

  • PTV vs. Missense Rare-Variant Clustering: $ARI = 1.00$ ($95\%\text{ CI } 0.52–1.47$), demonstrating perfect concordance between PTV and damaging missense rare-variant clustering.[1]
  • Rare PTV vs. Common-Variant Clustering: $ARI = -0.167$ ($95\%\text{ CI } -0.68 \text{ to } 0.35$), indicating complete structural discordance between rare and common genetic architecture spaces.[1]

In rare-variant space, EHR cases cluster tightly with GPpsy, Psypsy, SelfRepDep, and DepAll, whereas in common-variant space, EHR depression is an outlier least correlated with questionnaire definitions.[1] This structural divergence indicates that health-record ascertainment prioritizes acute neurobiological disruption driven by rare PTVs, whereas questionnaire definitions (lifetimeMDD) reflect common polygenic background.[1]

Biological Convergence & Candidate Genes

  • Cross-Disorder Neurodevelopmental Overlap: Rare PTV burden in depression was significantly enriched in gene sets implicated in exome studies of neurodevelopmental conditions (developmental delay $P = 1.2 \times 10^{-4}$, autism spectrum disorder $P = 3.4 \times 10^{-3}$, schizophrenia $P = 2.1 \times 10^{-3}$).[1]
  • Brain Tissue Specificity: Rare PTV burden was concentrated in brain-expressed genes ($OR = 1.26$ vs. $1.03$ for non-brain genes).[1]
  • Gene-Based Candidate Discoveries:
  • SLC2A1 (GLUT1): Associated with EHR depression via damaging missense burden ($OR = 6.01, 95\%\text{ CI } 3.12–11.58, P = 3.0 \times 10^{-7}$).[1] SLC2A1 encodes glucose transporter 1; impaired glucose transport across the blood-brain barrier impairs neuronal metabolic flux.[1]
  • NOG (Noggin): Associated with Psypsy depression via damaging missense burden ($OR = 8.43, 95\%\text{ CI } 3.84–18.50, P = 2.0 \times 10^{-6}$).[1] NOG inhibits bone morphogenetic protein (BMP) signaling, governing adult hippocampal neurogenesis.[1]

Synthesis: This study provides empirical support for the concepts established in Distinct Genetic Architecture in Trait Tails and RICE. Common PRS and rare coding burdens contribute additively to liability, but their underlying correlation spaces diverge ($ARI = -0.167$), proving that exome sequencing captures neurobiological etiology distinct from common-variant GWAS.

See Also

Citations

[1] Tian, R., Ge, T., Kweon, H., Rocha, D.B., Lam, M., Liu, J.Z., Singh, K., Levey, D.F., Gelernter, J., Stein, M.B., Tsai, E.A., Huang, H., Chabris, C.F., Lencz, T., Runz, H., & Chen, C.-Y. (2024). Whole-exome sequencing in UK Biobank reveals rare genetic architecture for depression. Nature Communications, 15, 1755. Supports: 7 case definitions; exome PTV burden OR = 1.17; additive liability model (2.51% vs 0.22%); BHR genetic correlation ARI = -0.167; candidate genes SLC2A1 and NOG. Location: Main text (Figs. 1–3); Supplementary Information. Source paper: s41467-024-45774-2.md

[2] Weiner, D. J., Nadig, A., Jagadeesh, K. A., Dey, K. K., Neale, B. M., Robinson, E. B., Karczewski, K. J., & O'Connor, L. J. (2023). Polygenic architecture of rare coding variation across 394,783 exomes. Nature, 614, 492–499. Supports: Burden Heritability Regression (BHR) formulation and rare coding heritability estimation principles. Location: Methods; Results. Source paper: s41586-022-05684-z.md