Testing for genetic assimilation with phylogenetic comparative analysis: Conceptual, methodological, and statistical considerations

Abstract

Genetic assimilation is a process that leads to reduced phenotypic plasticity during adaptation to novel conditions, a potentially important phenomenon under global environmental change. Null expectations when testing for genetic assimilation, however, are not always clear. For instance, the statistical artifact of regression to the mean could bias us toward detecting genetic assimilation when it has not occurred. Likewise, the specific mechanism underlying plasticity expression may affect null expectations under neutral evolution. We used macroevolutionary numerical simulations to examine both of these important issues and their interaction, varying whether plasticity evolves, the evolutionary mechanism, trait measurement error, and experimental design. We also modified an existing reaction norm correction method to account for phylogenetic nonindependence. We found (1) regression to the mean is pervasive and can generate spurious support for genetic assimilation; (2) experimental design and post hoc correction can minimize this spurious effect; and (3) neutral evolution can produce patterns consistent with genetic assimilation without constraint or selection, depending on the mechanism of plasticity expression. Additionally, we reanalyzed published macroevolutionary data supporting genetic assimilation, and found that support was reduced after proper correction. Considerable caution is thus required whenever investigating genetic assimilation and reaction norm evolution at macroevolutionary scales.

Trade-offs between baseline thermal tolerance and thermal tolerance plasticity are much less common than it appears

“…At the core, inflated support for the TOH, or any other hypothesis, because of regression to the mean is a problem of identifying the appropriate null expectation (Hertz, Huey, & Stevenson, 1993). A lack of association between baseline thermal tolerance and thermal tolerance plasticity is not always the correct null hypothesis based on the methodological approaches taken (Gunderson & Revell, 2022). By recognizing such situations and accounting for them, we can more quickly develop a clear picture of the forces that influence the evolution of physiological plasticity, and in what situations different mechanisms are important.…”

“…How general is this outcome? Methods I reanalyzed data from published studies that tested and found support for the TOH, focusing on work conducted at the intraspecific level (see Gunderson & Revell (2022) for analysis of the same issues at a macroevolutionary scale). I started with data from the studies summarized in Table 1 of van Heerwaarden & Kellermann (2020), and added as many subsequently-published data sets as I could identify.…”

“…The "Metric" definitions are as follows: CTmax = critical thermal maximum; LD50 = temperature of 50% mortality; CCRT = chill coma recovery time; CCO = chill coma onset. in Kelly & Price (2005) and Gunderson & Revell (2022). All analyses were conducted in R (R Development Core Team, 2021).…”

“…Deery et al (2021) found initial support for the TOH at the individual level among Anolis lizards with respect to heat hardening, but subsequently conducted a data resampling analysis (Ghalambor et al, 2015; Jackson & Somers, 1991) and concluded that the relationships were spurious. Finally, Gunderson & Revell (2022) used simulations to show how pervasive regression to the mean is when testing the TOH at macroevolutionary scales. They implemented a post-hoc method for adjusting plasticity data to account for regression to the mean (Kelly & Price, 2005), and found that doing so reduced support for the TOH in a previously-published dataset (Armstrong, Tanner, & Stillman, 2019; Gunderson & Revell, 2022).…”

“…Finally, Gunderson & Revell (2022) used simulations to show how pervasive regression to the mean is when testing the TOH at macroevolutionary scales. They implemented a post-hoc method for adjusting plasticity data to account for regression to the mean (Kelly & Price, 2005), and found that doing so reduced support for the TOH in a previously-published dataset (Armstrong, Tanner, & Stillman, 2019; Gunderson & Revell, 2022). In sum, studies that attempt to account for regression to the mean in tests of the TOH have generally found that support for the hypothesis declines.…”