Gene-Level Pleiotropy and Therapeutic Safety

Summary¶

Analysis of 100,526 GWAS across 9,280 trait ontology terms reveals that gene-level pleiotropy exhibits a non-linear relationship with clinical drug approval.[1] While high gene-level pleiotropy flags organism-level safety liabilities and clinical trial terminations, intermediate pleiotropy (2–5 therapeutic areas) paired with protein-altering variants yields an odds ratio of 10.3 for drug approval.[1] Crucially, cell-culture essentiality screens invert this safety signal, highlighting the necessity of human genetics for predicting organism-level toxicity.[1]

Source Status¶

Preprint: This page is compiled from a bioRxiv preprint (Tsepilov et al., 2026) not yet peer reviewed. confidence: medium.

Map Scale & Ancestry Discovery Dynamics¶

Open Targets applied its Gentropy post-GWAS framework to 100,526 publicly available GWAS across 4,250 publications, evaluating over one trillion single-point statistics mapped to 9,280 trait ontology terms across 23 therapeutic areas (TAs):[1]

Fine-Mapped Yield: After eight ancestry-specific fine-mapping strategies, the analysis produced 789,453 credible sets, filtered to 520,975 qualified sets (70,618 disease, 450,357 biomarker/measurement).[1]
Gene Prioritisation Coverage: Yielded 523,409 credible-set–gene prioritisations covering 15,641 genes, 1,394 diseases, and 3,412 measurements.[1]
Genome-Wide Breadth: 77.9% of all human protein-coding genes were associated with at least one complex trait.[1]
Non-Saturating Discovery: Gene-disease associations accumulate faster than new gene discoveries, causing the average number of associated diseases per gene to rise continuously without reaching a plateau.[1]
Ancestry-Driven Discoveries: By 2024, 30% (2,469 of 8,129) of disease-associated genes were first identified in non-European populations, contributing 16,401 of 34,905 total gene-disease associations.[1]

Pleiotropy Landscape & Functional Partitioning¶

Among 8,285 disease-associated genes, 64% (5,314) are associated with more than one disease, and 57% span multiple therapeutic areas.[1] Gene set enrichment across 312 pathways partitioned pleiotropic architecture into two distinct functional regimes:[1]

High-Pleiotropy Genes (221 Pathways): Enriched in immune and inflammatory signaling, oncogenic signal transduction, and transcriptional regulation — including cytokine and interferon cascades, T-cell and B-cell receptor pathways, JAK-STAT, NF-$\kappa$B, MAPK, PI3K-AKT, WNT, TGF-$\beta$, and receptor tyrosine kinase (RTK) networks.[1]
Low-Pleiotropy Genes (91 Pathways): Concentrated in core housekeeping functions — DNA repair and replication, cell-cycle control, chromatin regulation, RNA splicing, translation, and mitochondrial oxidative phosphorylation.[1] Purifying selection restricts detectable coding variation in these core genes.[1]

gPS as an Organism-Level Safety Indicator¶

The Gene-Level Pleiotropy Score (gPS) measures the breadth of phenotypic involvement across independent locus domains.[1] Higher gPS strongly predicts membership in gene sets representing severe organismal liabilities:[1]

Gene Set Category	$\log(\text{OR})$ per Doubling of gPS	$P$-value
Cancer driver genes	0.40	$5.6 \times 10^{-22}$
Developmental disorder panel genes	0.28	$3.1 \times 10^{-19}$
Mouse knockout-lethal homologs	0.26	$3.3 \times 10^{-19}$
Clinical trials terminated for safety	0.27	$8.9 \times 10^{-10}$
LoF-constrained genes (Q4 pLI/LOEUF)	0.21	$2.1 \times 10^{-31}$
Low-constraint genes (Q1)	$-0.16$	$2.3 \times 10^{-12}$
Human knockouts (homozygous LoF)	$-0.12$	$5.9 \times 10^{-9}$

The association between gPS and clinical trial safety terminations remains significant after adjusting for directionality discordance across traits, confirming that safety liability reflects broad phenotypic involvement rather than opposing effect directions.[1]

The DepMap / FUSIL Cell-Culture Reversal¶

In vitro cellular screens exhibit an inverted relationship with pleiotropy:[1] - High-gPS genes are depleted from DepMap essential genes and FUSIL cellular-lethal panels ($\log(\text{OR}) < 0$).[1] - Mechanistic Explanation: Cell-culture screens measure cell-autonomous survival in isolated environments, whereas gPS captures organ-system crosstalk and physiological trade-offs.[1] Consequently, targets that appear completely benign in cellular knockouts may still carry high gPS values predicting clinical safety failure in human trials.[1]

Cell Culture vs Human Genetics Safety Profiling
├── DepMap / In Vitro Knockout ──> Measures cell-autonomous viability (Depleted for high-gPS)
└── Human GWAS gPS Mapping     ──> Measures multi-organ physiological trade-offs (Enriched for trial safety failure)

Stratification of Therapeutic Target Approval¶

Evaluating 242 approved target-indication pairs supported by fine-mapped credible sets confirmed that genetic support increases drug approval likelihood (mixed-effects adjusted $\text{OR} = 3.14, P = 4.5 \times 10^{-49}$):[1]

Rare vs. Common Support: Rare-variant associations ($\text{MAF} < 0.01$) yield higher approval odds than common variants ($\text{OR} = 7.0$ vs. $3.4; P = 0.0077$).[1]
Coding Variant Impact: Protein-altering variant (PAV) support outperforms non-coding QTL support ($\text{OR} = 6.0$ vs. $3.1; P = 0.0002$).[1]
Effect Size Independence: Effect magnitude ($|\beta| > 0.5$ vs. smaller) does not alter approval rates ($\text{OR} = 4.6$ vs. $3.5; P = 0.19$). For example, both large-effect coding LoF variants and modest-effect common non-coding variants ($\text{rs11206510}, |\beta| = 0.068$) independently validate PCSK9.[1]
Non-Linear Pleiotropy Curve: Targets with intermediate pleiotropy ($\text{gPS} \le 5$) are 1.6-fold more likely to achieve approval than highly pleiotropic targets ($\text{gPS} \ge 10$; $\text{OR} = 4.8$ vs. $3.0; P = 0.008$). A logarithmic pleiotropy term significantly improves model fit ($P < 2 \times 10^{-13}$).[1]

Resolving the Target Selection Tension¶

Protein-altering variant (PAV) support strongly boosts therapeutic success ($\text{OR} = 6.0$), but PAV-supported targets also exhibit higher average pleiotropy, creating a competing safety risk.[1] Combining PAV support with intermediate pleiotropy (2–5 therapeutic areas) resolves this tension:[1]

$$\text{Approval Odds Ratio (PAV } \times \text{ Intermediate Pleiotropy)} = 10.3 \quad (\text{Relative Success } RS = 4.8)$$

This optimal profile is currently satisfied by 52 approved clinical therapies.[1] High-pleiotropy targets still outperform targets lacking any genetic support ($\text{OR} = 0.74$ relative to genetically supported targets, $P = 1.8 \times 10^{-10}$), indicating that pleiotropy should be used for risk-mitigated target prioritization rather than absolute rejection.[1]

Synthesis & Lab Interpretation: Target triage should treat gPS as a two-sided filter rather than a simple scalar score. Very low gPS often indicates under-powered GWAS or purifying selection on core housekeeping genes; very high gPS flags multi-system toxicity. The optimal target profile combines protein-altering variant support with intermediate pleiotropy (2–5 TAs).

Citations¶

[1] Tsepilov, Y. A., Suveges, D., Considine, D., Szyszkowski, S., Ge, X. J., et al. (2026). The Human Pleiotropic Map of GWAS Associations and Therapeutic Implications. bioRxiv, 2026.04.28.721048. Supports: Gentropy pipeline counts; non-saturating discovery; ancestry-specific findings; pathway enrichments; gPS safety log(OR) table; DepMap cell-culture reversal; PAV x intermediate pleiotropy OR = 10.3. Location: Main text (Figs. 1, 4, 5); Supplementary Results. Source paper: 2026.04.28.721048v1.full.md