Microfossils have a ubiquitous and well-studied fossil record with temporally and spatially fluctuating diversity, but how this arises and how major events affect speciation and extinction is uncertain. We present the first application of PyRate to a micropaleontological global occurrence data set, reconstructing diversification rates within a Bayesian framework from the Mesozoic to the Recent in four microfossil groups: planktic foraminiferans, calcareous nannofossils, radiolarians and diatoms. Calcareous and siliceous groups demonstrate opposed, but inconsistent, responses in diversification. Siliceous groups increased origination from ~104 Ma, maintaining high rates into the Cenozoic. Calcareous microfossils diversification rates significantly decline across the Cretaceous–Paleogene boundary, while rates in siliceous microfossil groups remain stable until the Paleocene–Eocene transition. Diversification rates in the Cenozoic are largely stable in calcareous groups, whereas the Paleogene is a turbulent time for diatoms. Diversification fluctuations are driven by climate change and fluctuations in sea surface temperatures, promoting selectivity in both microfossil composition and foraminiferal size. Extinctions appear induced by changes in anoxia, acidification, and stratification, while speciation tends to be associated with upwelling, productivity, and ocean circulation. These results show promise for further quantitative analyses in micropaleontological diversity studies and effects of major transitions in the fossil record. Despite extensive occurrence data, regional diversification events were not recovered, neither were some global events. These unexpected results show the need to consider multiple spatiotemporal levels of diversity and diversification analyses, and implies occurrence data sets of different clades may be more appropriate to testing some hypotheses than others.

Microfossil data used in this study was collected from NSB (Neptune Sandbox Berlin). NSB holds information of over one million species occurrences of several microfossil groups pulled from scientific drilling programs. Four fossil groups were downloaded: planktic foraminifera, calcareous nannofossils, diatoms, and radiolarians into R using the package ‘NSB Companion’ v.2.1. The data was then processed to correct any synonymous species, removing non-valid taxa, and a dating uncertainty of ±0.25 my was applied to occurrence sample ages. The data was then manipulated into a suitable format table of taxon-occurrence range entries so it could be processed by PyRate.

PyRate implements three different preservation models (the default non-homogeneous Poisson process, homogeneous, and time-dependent Poisson processes) that vary the allowed heterogeneity of the preservation rates through time. All three of these preservation models were run on all four fossils groups in separate analyses, while compositional effects on preservation rates were assessed by grouping calcareous microfossils (foraminiferans and nannofossils) and siliceous microfossils (radiolarians and diatoms) into two further analyses. PyRate analyses for each taxonomic group were run as array jobs of the ten replicates generated on the Blue Pebble cluster at the University of Bristol (Advanced Computing Research Centre) using the recommended reversible jump Markov Chain Monte Carlo algorithm, generating 10⁷ iterations, and sampling every 5,000 iterations. Convergence was assessed in Tracer version 1.7, using the effective sample size (ESS) to assure the posterior distribution is suitably represented. Although not all model runs completed the iterations within the 72-hour walltime, all model runs converged (ESS > 200) providing representative values for preservation rates and species longevity of each species.

Model selection for each microfossil group used a maximum-likelihood framework to allocate the best fitting preservation. The fit of each model was calculated in the Akaike Information Criterion (AIC), providing a statistical method of assessing the fit of qualitatively different preservation models. The sensitivity of the analyses to occurrence dating precision was investigated by running four additional PyRate analyses with decreasing occurrence ranges (±0.20, ±0.15, ±0.10, ±0.05 my) using default preservation model settings (NHPP). 

Neptune Sandbox Berlin (NSB) is an open source database of microfossil occurrence data including planktic formainifera, calcareous nannofossils, radiolarians, and diatoms. 

Thousands of marine plankton microfossil species are catalogued from hundreds of deep-sea ocean drilling sections. The data includes age models for all sections and the geochronologic data used to create these age models.

To access the database you will need to email Dr Johan Renaudie, MfN Berlin - johan.renaudie@mfn-berlin.de or Dr David Lazarus, MfN Berlin - david.lazarus@mfn-berlin.de to gain a username and password. See Renaudie et al. (2020) for more information about the creation, maintenance, and features of the database.

PyRate is readily available to download from Daniele Silvestro's GitHub: https://github.com/dsilvestro/PyRate

Using PyRate we implemented the reversible jump Markov Chain Monte Carlo algorithm, generating 10⁷ iterations, and sampling every 5000 iterations.

Convergence was assessed in Tracer v1.7 (Rambaut et al. 2018) using the effective sample size (ESS) to ensurethat the posterior distribution was suitably represented.

Tracer is freely available from: http://tree.bio.ed.ac.uk/software/tracer/

All analyses were completed within Python and R which are both free to access and download. All files uploaded to this dataset use Python, R, Microsoft Word, Microsoft Excel, any image viewer compatible with .tiff files, and any .txt document viewer.