The Making of Calibration Sausage Exemplified by Recalibrating the Transcriptomic Timetree of Jawed Vertebrates

Abstract

Molecular divergence dating has the potential to overcome the incompleteness of the fossil record in inferring when cladogenetic events (splits, divergences) happened, but needs to be calibrated by the fossil record. Ideally but unrealistically, this would require practitioners to be specialists in molecular evolution, in the phylogeny and the fossil record of all sampled taxa, and in the chronostratigraphy of the sites the fossils were found in. Paleontologists have therefore tried to help by publishing compendia of recommended calibrations, and molecular biologists unfamiliar with the fossil record have made heavy use of such works (in addition to using scattered primary sources and copying from each other). Using a recent example of a large node-dated timetree inferred from molecular data, I reevaluate all 30 calibrations in detail, present the current state of knowledge on them with its various uncertainties, rerun the dating analysis, and conclude that calibration dates cannot be taken from published compendia or other secondary or tertiary sources without risking strong distortions to the results, because all such sources become outdated faster than they are published: 50 of the (primary) sources I cite to constrain calibrations were published in 2019, half of the total of 280 after mid-2016, and 90% after mid-2005. It follows that the present work cannot serve as such a compendium either; in the slightly longer term, it can only highlight known and overlooked problems. Future authors will need to solve each of these problems anew through a thorough search of the primary paleobiological and chronostratigraphic literature on each calibration date every time they infer a new timetree, and that literature is not optimized for that task, but largely has other objectives.

Twenty-five well-justified fossil calibrations for primate divergences

“…The putative leptictidan Gypsonictops is known from the Late Cretaceous (Campanian-Maastrichtian; Kielan-Jaworowska et al, 2004) but has also been reported from much older, Turonian deposits (Cohen and Cifelli, 2015), although this material remains to be formally published; if Gypsonictops is indeed a leptictidan, and if leptictidans are crown clade placentals, then this would push the origin of Placentalia considerably earlier than the K-Pg boundary. However, this remains uncertain (see Springer et al, 2019;Marjanović, 2021), and the recent study by Velazco et al (2022) found leptictidans (including Gypsonictops) to fall outside Placentalia. The preponderance of current fossil evidence appears to support an origin of Placentalia and of the major placental superorders closer to the K-Pg boundary (see also Budd and Mann, 2022); we explicitly take this into account by proposing that this calibration be modelled as an offset exponential prior distribution.…”

“…In such analyses, fossil taxa for which molecular data are unavailable cannot be included directly as terminals/tips, but instead can be used to calibrate the age of one or more nodes in the phylogeny ("nodedating"; Nguyen and Ho, 2020). Because fossils are assigned a priori to particular nodes under this approach, it is crucial that each calibrating fossil is assigned appropriately, based on the most accurate and up-to-date information (Parham et al, 2012;Marjanović, 2021).…”

“…However, as discussed at length by Marjanović (2021), identifying appropriate fossil calibrations is not trivial, and even comparatively recent lists of calibrations (e.g., Benton et al, 2015) now appear to be at least partly out of date. Such issues affect molecular clock analyses of primates.…”

“…Although fossil calibrations have been used to specify normal distributions on the ages of nodes in some studies (e.g., Perelman et al, 2011), it seems more appropriate to view the minimum age of a calibrating fossil as providing a minimum bound on the age of that node (Benton and Donoghue, 2007;Benton et al, 2009;Ho and Phillips, 2009;Parham et al, 2012). Wherever possible, it is appropriate to also specify a maximum bound, otherwise there is no direct constraint on the maximum age of a calibrated node (Phillips, 2016;Marjanović, 2021). If maximum and minimum bounds are specified, then a prior probability distribution between these bounds (and, if these bounds are "soft", outside them as well) also needs to be specified (Ho and Phillips, 2009).…”

“…Analytical methods for inferring maximum bounds and prior probability distributions on calibrations have been proposed (Marshall, 2008;Wilkinson et al, 2011;Nowak et al, 2013;Matschiner et al, 2017;Matschiner, 2019;Claramunt, 2022), but these often require estimates of parameters such as diversification and/or sampling rates that are not always easy to obtain, even for primates (but see Silvestro et al, 2014;Herrera, 2017;. For this reason, maximum bounds are typically based on somewhat subjective interpretations of available phylogenetic and fossil evidence (as in Benton et al, 2015;Roos et al, 2019;Marjanović, 2021; and most calibrations used by dos Reis et al, 2018). However, dos Reis et al (2018) based their prior distributions for two nodes -namely the ages of crown Primates and crown Anthropoidea -on the analytical estimates of Wilkinson et al (2011); nevertheless, we note that the maximum bounds of both of these calibrations (88.6 Ma for crown Primates, 62.1 Ma for crown Anthropoidea) seem implausibly old based on the fossil record (see our proposed calibrations for both of these nodes below).…”