This work provides a much-needed systematic evaluation of the relationships between familial (FH) and polygenic (PRS) risks and their contributions to disease susceptibility in a panel of common diseases (1). The results have clear implications for risk-stratification, but as discussed in subsequent letters to the editor (2, 3) the results and their underlying data may also contribute to an understanding of causation in these diseases. Mars et al (1) do implicitly draw a conclusion regarding causation in describing in the Abstract the striking finding illustrated in their Fig 7: “…low PRS compensated completely for the risk implied by positive family history” (my emphasis). The potential mechanisms accounting for this compensation are not explicitly discussed in Mars et al (1) or in the related correspondence (2, 3). I suggest that a simple mechanism arises naturally in a model of familial inheritance based on segregation of very rare causal variants with large effects. This mechanism may be testable in existing FinnGen data.

Taking the disease I’m most familiar with as an example (T2DM), and ignoring for the moment any non-genetic FH effects and other confounders, FH+ then represents 50% transmission of causative variants while hiPRS and lowPRS are potential markers of 100% and 0% transmission respectively. Rather than compensating for FH effects, lowPRS is then a marker of absence of transmission of causative variants. The cumulative incidence at 80 years (Ic) in the T2DM data from FH+ in Fig 7 illustrates this interpretation: taking lowPRS (Ic = 0.29) and hiPRS (Ic = 0.69) to represent 0% and 100% transmission of causative variants, a predicted Ic for 50% transmission can be calculated as the mean (= 0.49) and compared to the FH+ estimate (= 0.47) giving a predicted/estimated ratio (P/E) = 1.04. The model provides a plausible explanation for the T2DM data. The same calculation applied to all of the diseases represented in Fig 7 gives a mean P/E = 1.06 ± 0.07 (SD), supporting a general applicability of the model in these data.

The above calculations are consequences of the also striking linear relationships between FH prevalence and PRS across most of the panel of diseases (Fig S2). In the case of T2DM a linear regression of FH prevalence by decile from Fig S2 on PRS expressed as Z-scores corresponding to decile means in a normal distribution gives R = 0.98. A model in which elements of PRS are linked to very rare, perhaps private causative variants running in families provides a simple explanation for these data. Such causative variants would not be statistically identifiable in GWAS-like approaches.

This model in which PRS effects are mediated through linkage to large effect variants predicts that elements of PRS would track these large effects in families and under favourable circumstances PRS or some elements of them could show evidence of segregation. In testing for segregation it would be necessary to consider the contributions of other genetic and non-genetic contributors to disease expression. In the case of T2DM excess adiposity, which is almost a necessary condition for T2DM expression, and is under interacting genetic and environmental influences would need to be incorporated into the model.

We have obtained evidence supporting this type of model (segregation) in the genetics of body composition using distribution analysis of a latent adiposity phenotype vs FH in a modest sample (4) and of BMI vs FH in a very large sample (NHANES (5)). In both studies we used FH of diabetes as a surrogate for FH of excess adiposity justified by the very close clinical association between the two conditions. I suggest that a two-hit model involving independent familial transmission of excess adiposity and T2DM susceptibilities could form the basis of a segregation analysis in the FinnGen data. The plausibility of this model can be illustrated using the Ic data in Fig 7. The model predicts that 25% of FH+ individuals would inherit both susceptibilities and with full expression would therefore account for 25% of the excess incidence giving a predicted Ic = 0.47, equal to the (digitised) measured outcome. The same model applied to the FH+ hiPRS group assuming 100% transmission of T2DM and 50% of excess adiposity susceptibilities predicts Ic = 0.64, slightly lower than measured (0.69) but within the displayed confidence interval.

The extent to which this model if supported could lead to the identification of causative variants is unclear to me. I assume that it would involve analyses of data which included all available related individuals and hence could track extended family structures. Perhaps the extent to which adjustment of genomic analyses for principal component scores could obscure relevant signals of relatedness should be considered.

1. N. Mars et al., Am J Hum Genet 109, 2152 (2022).
2. S. Li, J. L. Hopper, Am J Hum Genet 110, 1221 (2023).
3. N. Mars et al., Am J Hum Genet 110, 1224 (2023).
4. A. B. Jenkins et al., PLOS ONE 8, e70435 (2013).
5. A. B. Jenkins, M. Batterham, L. V. Campbell, http://dx.doi.org/10.1101/749606 (2019).