A Practical Guide to Estimating the Heritability of Pathogen Traits

Abstract

Pathogen traits, such as the virulence of an infection, can vary significantly between patients. A major challenge is to measure the extent to which genetic differences between infecting strains explain the observed variation of the trait. This is quantified by the trait’s broad-sense heritability, H2. A recent discrepancy between estimates of the heritability of HIV-virulence has opened a debate on the estimators’ accuracy. Here, we show that the discrepancy originates from model limitations and important lifecycle differences between sexually reproducing organisms and transmittable pathogens. In particular, current quantitative genetics methods, such as donor–recipient regression of surveyed serodiscordant couples and the phylogenetic mixed model (PMM), are prone to underestimate H2, because they neglect or do not fit to the loss of resemblance between transmission partners caused by within-host evolution. In a phylogenetic analysis of 8,483 HIV patients from the United Kingdom, we show that the phenotypic correlation between transmission partners decays with the amount of within-host evolution of the virus. We reproduce this pattern in toy-model simulations and show that a phylogenetic Ornstein–Uhlenbeck model (POUMM) outperforms the PMM in capturing this correlation pattern and in quantifying H2. In particular, we show that POUMM outperforms PMM even in simulations without selection—as it captures the mentioned correlation pattern—which has not been appreciated until now. By cross-validating the POUMM estimates with ANOVA on closest phylogenetic pairs, we obtain H2 ≈ 0.2, meaning ∼20% of the variation in HIV-virulence is explained by the virus genome both for European and African data.

Parallel Likelihood Calculation for Phylogenetic Comparative Models: the SPLITT C++ Library

“…The POUMM R-package implements this approach by embedding the SPLIT-based likelihood calculation in a Metropolis sampler with coerced acceptance rate available from the R-package (Scheidegger 2017) (Section 3.1, Supplementary Information). We tested this approach during a POUMM analysis of a transmission tree from the HIV epidemic in UK (N=8483) reported elsewhere (Mitov & Stadler 2018). This showed faster MCMC convergence (Fig.…”

“…To illustrate the use and to test the SPLITT library, we developed two variants of the so called phylogenetic mixed model (PMM) – the original PMM assuming a Brownian motion process (Lynch 1991; Housworth et al . 2004), and its recent extension to an Ornstein-Uhlenbeck process, the POUMM, which we and other authors have used previously to analyze the evolution of set-point viral load in HIV patients (Mitov & Stadler (2018) and references therein).…”

“…(2014); Bertels et al . (2017); Mitov & Stadler (2018). While big data should provide the needed statistical power to fit a complex model, the time needed to perform a full scale Bayesian fit often limits the choice to a faster but less informative ML-inference, or a Bayesian inference of a simplified model.…”

Reliable Phylogenetic Regressions for Multivariate Comparative Data: Illustration with the MANOVA and Application to the Effect of Diet on Mandible Morphology in Phyllostomid Bats

“…For example, we estimated phylogenetic signal using Pagel's ߣ model, which can be viewed as a phylogenetic mixed model (PMM) combining a Brownian motion process with independently and normally distributed errors (Housworth et al 2004;Clavel et al 2019). Even more flexible PMMs could be envisaged, for example based on the Ornstein-Uhlenbeck process (Mitov and Stadler 2018); they are already available in the penalizedlikelihood framework (Clavel et al 2019). Future improvements should also consider models where each trait may have its own level of phylogenetic signal -as was done in lower dimensional settings (e.g., Freckleton 2012; Ho and Ané 2014) -rather than assuming a common (or average) signal across trait as we did here.…”

Heritability of the HIV-1 reservoir size and decay under long-term suppressive ART

“…Heritability can be estimated using resemblance estimators, which measure the relative traitsimilarity within transmission clusters. Established methods include parent-offspring (PO) regression [24][25][26][27][28] and analysis of variance with mixed-effect models 29,30 . Additionally, phylogenetic comparative methods can also be used to estimate heritability by measuring the association between observed trait values from individuals and their transmission tree inferred from pathogen sequences 24,25 .…”