Disentangling mechanisms behind the pleiotropic effects of proximal 16p11.2 BP4-5 CNVs

Abstract

Whereas 16p11.2 BP4-5 copy-number variants (CNVs) represent one of the most pleiotropic etiologies of genomic syndromes in both clinical and population cohorts, the mechanisms leading to such pleiotropy remain understudied. Identifying 73 deletion and 89 duplication carrier individuals among unrelated White British UK Biobank participants, we performed a phenome-wide association study (PheWAS) between the region's copy number and 117 complex traits and diseases, mimicking four dosage models. Forty-six phenotypes (39%) were affected by 16p11.2 BP4-5 CNVs, with the deletion-only, mirror, U-shape, and duplication-only models being the best fit for 30, 10, 4, and 2 phenotypes, respectively, aligning with the stronger deleteriousness of the deletion. Upon individually adjusting CNV effects for either body mass index (BMI), height, or educational attainment (EA), we found that sixteen testable deletion-driven associations-primarily with cardiovascular and metabolic traits-were BMI dependent, with EA playing a more subtle role and no association depending on height. Bidirectional Mendelian randomization supported that 13 out of these 16 associations were secondary consequences of the CNV's impact on BMI. For the 23 traits that remained significantly associated upon individual adjustment for mediators, matched-control analyses found that 10 phenotypes, including musculoskeletal traits, liver enzymes, fluid intelligence, platelet count, and pneumonia and acute kidney injury risk, remained associated under strict Bonferroni correction, with 10 additional nominally significant associations. These results paint a complex picture of 16p11.2 BP4-5's pleiotropic pattern that involves direct effects on multiple physiological systems and indirect co-morbidities consequential to the CNV's impact on BMI and EA, acting through trait-specific dosage mechanisms.

Keywords: Mendelian randomization; dosage sensitivity; educational attainment; genomic disorder; mediation; multi-system disorder; obesity; phenome-wide association study; pleiotropy; structural variant.

Graphical abstract

Figure 1

Study workflow (A) Direct (horizontal) pleiotropy: The CNV causally affects multiple traits through independent mechanisms. (B) Indirect (vertical) pleiotropy: The CNV causally impacts trait₂, which in turn causally affects trait₁ and trait₃. The impact of the deletion on trait₁ and trait₃ is thus indirect and mediated by a shared mechanism, i.e., trait₂. For illustration, the 16p11.2 BP4-5 deletion is shown but the same concept applies to the duplication. (C) Overview of the study. The first two analyses aim at detecting and characterizing the pleiotropy of 16p11.2 BP4-5 CNVs through four distinct dosage models that estimate the effect of the CNV on the trait (Y) either without (analysis 1; CNV₁) or with (analysis 2; CNV₂) adjustment for one of three covariates (mediator) that could potentially mediate the CNV-phenotype association. The second part of the study aims at understanding the mechanisms through which pleiotropy arises. Bidirectional Mendelian randomization was used to investigate the causal relationship between trait-mediator pairs for which the significance of the CNV effect on the trait was affected by adjustment for the mediator (analysis 3). Support for mediation was claimed when the forward MR effect of the mediator on the trait (${\alpha }_{M\to T}$) was significant, while the reverse effect of the trait on the mediator (${\alpha }_{T\to M}$) was not significant or of smaller magnitude than the forward effect. For traits that showed a significant association with the CNV regardless of covariate adjustment, matched-control analysis, which allows simultaneous adjustment for all possible mediators, was performed (analysis 4). If the effect remains significant, the association likely reflects genuine horizontal pleiotropy. BP, break point; PheWAS, phenome-wide association study.

Figure 2

16p11.2 BP4-5 phenome-wide association study Effect sizes (beta; x axis) with 95% confidence interval (CI) of the 16p11.2 BP4-5 deletion (circle) and duplication (square) on 117 complex traits and diseases, ordered by physiological system (y axis). Effect sizes are in standard deviation units of the outcome (quantitative traits) or logarithms of the odds ratio of a logistic regression (disease traits). Deletion- and duplication-only effects that are significant under Bonferroni correction (p ≤ 0.05/117 = 4.3 × 10⁻⁴) are in blue and red, respectively, while non-significant effects are in gray. If the most significant model was the mirror or U-shape model, it is denoted with an “M” or “U,” respectively (right). BMI, body mass index; HDL, high-density lipoprotein; LDL, low-density lipoprotein; WHR, waist-to-hip ratio.

Figure 3

Mediation of 16p11.2 BP4-5 pleiotropy through anthropometric traits and educational attainment (A) Effects (beta) of 16p11.2 BP4-5 CNVs on traits with adjustment for potential mediators (y axis)—i.e., body mass index (BMI; top row), educational attainment (EA; middle row), or height (bottom row)—against those without adjustment (x axis), stratified according to the best (i.e., most significant) association model (shape; columns). Only associations that were significant prior to or become significant after adjustment are plotted. Traits are colored according to physiological systems. Size reflects whether the effect is Bonferroni significant (p ≤ 0.05/117 = 4.3 × 10⁻⁴) after adjusting for the potential mediator. Traits losing significance upon adjustment are labeled (see B for BMI-adjustment of deletion-only associations). Gray dashed diagonal represents the identity line. (B) Enlargement of the area delimited by a red dashed rectangle in (A), showing the effect of BMI adjustment for deletion-driven associations, using the same legend as in (A). (C) Schematic representation of the links between 16p11.2 BP4-5 CNVs, potential mediators, and assessed traits. Covariate-adjusted phenome-wide association studies (PheWASs) identified CNV-trait associations that are dependent on either of two tested factors (i.e., BMI or EA; thick gray arrow) in (A). This can happen through mediation, collider bias, or confounding. Bidirectional Mendelian randomization (MR) assesses the genetically determined causal effect of the putative mediator on the trait (forward effect, red arrow) and of the trait on the mediator (reverse effect; dark blue arrow), with arrows proportional to causal effect sizes. When the forward effect is larger than the reverse one, mediation is a likely scenario; when the reverse effect is larger, the putative mediator likely acts as a collider; absence of causal effects likely indicates presence of an unobserved confounder, U. Depending on the scenario, adjustment for the mediator in the regression analysis might or might not be appropriate. (D) Bidirectional forward (red) and reverse (dark blue) MR effects with 95% confidence interval (CI) of potential mediators (left y axis) on traits (right y axis) for all mediator-trait pairs that either gained or lost significance upon adjustment for the mediator. Phenotypes labeled with “^∗” exhibit a Pearson correlation between 0.4 and 0.7 with the mediator. Non-significant effects (p > 0.05/54 = 9.3 × 10⁻⁴) are semi-transparent. ALT, alanine aminotransferase; apnea, sleep apnea; CKD, chronic kidney disease; conduction, cardiac conduction disorders; CRP, C-reactive protein; HDL, high-density lipoprotein; HTN, essential hypertension; IGF1, insulin-like growth factor 1; lipid, lipidemias & lipoprotein disorders; OA, arthrosis; SHBG, sex hormone binding globulin; T1D, type 1 diabetes; TG, triglycerides; urate, serum urate; valves, cardiac valve disorders; WHRadjBMI, waist-to-hip ratio (WHR) adjusted for BMI.

Figure 4

16p11.2 BP4-5 CNV carrier individuals matched-control analyses Comparison between deletion (DEL) and duplication (DUP) carrier individuals (darker shade) and their respective matched control subjects (DEL or DUP “ctrl”; lighter shade) for 23 traits that remained Bonferroni-significant (p ≤ 0.05/117 = 4.3 × 10⁻⁴) after individually adjusting for body mass index (BMI), educational attainment (EA), or height in Figure 3. Phenotypes labeled with “^∗” exhibit a Pearson correlation between 0.4 and 0.7 with either BMI or EA. For quantitative traits, data are represented as boxplots without outliers: boxes show the first (Q1), second (median, thick line), and third (Q3) quartiles; lower and upper whiskers show the most extreme value within Q1 minus and Q3 plus 1.5 × the interquartile range, respectively. Data points for CNV carrier individuals are shown as gray dots. Sample size is indicated as n. p values of two-sided t tests comparing CNV carrier individuals to matched control subjects are reported. For diseases, prevalence in percentage with standard errors is depicted. Number of cases over total sample size is indicated. p values of two-sided Fisher tests comparing CNV carrier individuals to matched control subjects are reported. “ns” indicates p > 0.05. Traits are colored according to physiological systems. ALP, alkaline phosphatase; AST, aspartate aminotransferase; BMD, bone mineral density; COPD, chronic obstructive pulmonary disease; depression, recurrent depressive disorder; FVC, forced vital capacity; GGT, gamma-glutamyltransferase; HbA1c, glycated hemoglobin; WHR, waist-to-hip ratio.