Fig. 2: Phenotype-level and gene-level pleiotropy.

In the upper panel of Fig. 2, lipid metabolism influences intra-cellular lipid content in both hepatocytes and lymphocytes. However, only hepatocytes, but not lymphocytes, influence serum LDL-cholesterol. The in vivo associations between genes and serum LDL-cholesterol will capture effects across all biological contexts (e.g., cell-types) and the life-course. Therefore, GPAT estimates can indicate that an in vitro model is transportable but may not necessarily identify the true causal phenotype. We refer to this as ‘phenotype-level’ pleiotropy (a form of vertical pleiotropy) as it is characterised by alternative paths between genes and the outcome, which are through phenotypes closely related (genetically correlated) to the in vitro model phenotype. In the lower panel of Fig. 2, inter-cellular lipid content in hepatocytes and hepatocyte proliferation have distinct but shared genetic aetiology, with both phenotypes influencing serum LDL-cholesterol. The in vivo associations between genes and serum LDL-cholesterol will capture effects of both phenotypes, and so effects of hepatocyte proliferation may bias GPAT analyses centred on hepatocytic intra-cellular lipid content. However, in this context, the two phenotypes are not perfectly genetically correlated, so we may be able to statistically detect and control for bias relating to this ‘gene-level’ pleiotropy (a form of horizontal pleiotropy) using methods from the MR literature (e.g., Weighted Median).