Diversification and extinction dynamics buried in the sea floor (6 Comments)

nature11815-f1.2

Atlantic Ocean basin lithosphere ages and locations of 73 sampled drilling sites from Peters et al. On average, nearly 500m of younger sediments overlie oceanic crust at each site (black dots). Reprinted by permission from Macmillan Publishers Ltd: Nature, (Peters et al. 2013) © 2013.

Unraveling the relative roles of biotic interactions and abiotic fluctuations on diversification and extinction is a major tenant of evolutionary biology. Exploring the impact of species interactions and climate variability on local population extinction and colonization is a relatively easy task at ecological time scales. But when it comes to linking these issues to species extinction and diversification in geological deep time the ability to tease apart the living from the non-living may become a formidable task. The recent paper by Shanan E. Peters and colleagues is one of such rare occasions in which these issues are tackled successfully.

This study capitalizes on the emerging field of macrostatigraphy, which aims at connecting plate tectonics and protracted physicochemical changes in the ocean-atmosphere system to spatiotemporal shifts in deep-sea sedimentation. The oceanic lithosphere is covered by a sediment layer having a few hundred meters on average. Oceanic floor is constantly being created along mid-oceanic rifts, so that the sediment layer is thicker the farther away from the rift. Armed with this knowledge, Peters et al. gathered basin-scale data on the lithology, thickness and age of sediments packages recovered from 73 drilling offshore sampling sites spread through the Atlantic Ocean (see figure). At the same time, they assembled global stratigraphic ranges of 671 species of planktonic Foraminifera inhabiting the world oceans from the Jurassic period to the present. Foraminifera are amoeboid protists extremely widespread, abundant and diverse, and synthetize a shell made of calcium carbonate which fossilize readily.

After correlating both time series, a fascinating pattern emerged: species diversity and extinction rates seem to be controlled by tectonically and climatically forced changes in ocean chemistry and dynamics throughout the last 160 m.y. Strikingly however, the rates of species origination seem to be basically decoupled from any measure of macrostatigraphic quantity, suggesting indeed that biological factors (i.e., species interactions) might be controlling diversification. Interestingly, these patterns break down transiently during certain periods, such as during the aftereffects of the Chicxulub asteroid impact at the end of the Cretaceous. The bottom line? In geological time extinction appears largely not selective, but diversification seems the opposite.

Some may say that the results in Peters et al. are not that compelling: after all, they resonate with the famous view from paleontology on planetary changes and mass extinctions: ¿Is it bad genes, or bad luck? (Ref. 1). On the other hand, diversification as driven by natural selection is a rather simple demographic process: if you have a trait correlated with some measure of breeding success or survival, and link it with a transgenerational response to that correlation through genetic inheritance, you end up with adaptive evolution. It logically follows that if the phenotypic correlation is high and heritability is large, natural selection becomes strong and operates quite fast. But, do we really see a consistent phenotypic change in natural populations across tens to hundreds of generations? Well, the answer is not really that much (Ref. 2). How to interpret this?

Natural selection is a process with a lot of subtleties, and the answer is not simple. But I’ve always thought that, to a large extent, this apparent paradox boiled down to an insidious effect of the metaphor of “survival of the fittest”. For me, it is difficult to regard long-term evolution by natural selection as a process by which certain traits are optimized under a set of environmental conditions, in the sense that trait values outside the optima are deleted across a run of selective events. Natural selection is not a law, is a probability distribution: at any given moment you can bear a trait in the optima, and still fail to survive or reproduce due to contingencies, so, as an “unlucky” one, you will fade away with the “unfits”; and the remaining phenotypes will persist, irrespective of their properties. This is a more nuanced view of evolution, a sort of “non-survival of the non-fit” (Ref. 3), and is a view increasingly supported by paleobiological studies such as the one by Peters et al.

 

  1. Raup, D. M. & Gould, S. J. 1992. Extinction: Bad Genes or Bad Luck? W. W. Norton & Co., NY, USA.
  2. Siepielski, A. M., DiBattista, J. D. & Carlson, S. M. 2009. It’s about time: the temporal dynamics of phenotypic selection in the wild. Ecology Letters, 12, 1261-1276.
  3. Den Boer, P. J. 1999. Natural selection, or the non-survival of the non-fit. Acta Biotheoretica, 47, 83-97.

February 13, 2013

Comments Leave a Comment

Please review our commenting policy.

  • One challenge with these sorts of data is that correlations among time series often are a very misleading guide to the underlying processes that generated the series. For instance, in population ecology, a strong correlation between year-to-year fluctuations in abundance and year-to-year fluctuations in some abiotic variable was once thought to indicate lack of density dependence. But in fact it’s just the opposite: such correlations indicate *strong* density dependence! I know nothing of paleoecology, and I’m curious whether this point has ever been discussed in the paleo literature. Since it seems like it might well be relevant.

    For discussion see http://dynamicecology.wordpress.com/2012/12/27/how-not-to-test-causality-observationally/. Not my clearest post, but there’s at least a link to a good recent paper on this issue in a population ecology context (Ziebarth et al. 2010 Ecology Letters)

    • Yes, I know Ziebarth et al.’s paper, and it’s great. Those classic points on correlation vs. causation in time series were dealt with in detail by many others before; I recall, for example, the excellent book “Regulation and stabilization paradigms in population ecology” by Pieter den Boer (Yes, the one in Ref. 3 above!) and Joannes Reddingius, written in 1996. Few people appear to have read this book anyway. It’s a pity, because it’s great, whether you agree with their arguments or not.

      With regard to Peters et al.’s correlation analysis, I think there is a subtle difference with respect to this problem: in the same time series, there is a clear connection between extinction rates and macrostatigraphic quantities, but not between diversification rates and the later. For me, this qualitative difference between the “ups” and “downs” within a single time series and putative causal factors is additional evidence that there is indeed a causal connection among them. I’m not sure if your example of correlations with climate in a density-dependent world is intended only as an illustration of the problem; but I think I can make a connection between Peters et al.’s results and population ecology: phase-dependencies in the autoregressive structure (the classic lynx and voles).

      But if your point is the classic “we cannot infer causation from correlation”, all I can say is that I obviously agree. This is not a limitation of the statistical method, but a limitation imposed by our ignorance. No SEM will ever solve this…

      • I’m curious: do you think most paleoecologists are aware of the points Ziebarth et al. make (points that are also made by others, BTW)? Correlating time series of diversity, diversification rates, and extinction rates with time series of abiotic variables, and then drawing conclusions from that about the underlying processes that generated the time series, is something paleoecologists do a lot, isn’t it? Is it common for paleoecologists doing that sort of work to make the fallacious Andrewartha and Birth argument: that because we see a strong correlation between, say, diversity and some abiotic variable, that diversity isn’t “regulated” and is just driven by exogenous variation? I have no idea because I don’t know the paleo literature at all. But if it is common for paleoecologists to misinterpret these sorts of correlations as evidence for lack of density dependence, then it seems like there might be a paper worth writing here. A sort of Ziebarth et al. for paleoecology. What do you think? Or is Ziebarth et al.’s point actually already widely recognized in paleoecology?

        p.s. Reddingius and den Boer are a bit like the late Robert Peters in my experience. Interesting and provocative to read, especially for students, but also idiosyncratic and needing to be taken with a grain of salt (perhaps a large grain…)

      • I’m not a paleoecologist, so I don’t know the state of the matter in this sub-discipline. But it should be kept in mind that the particular arguments from Ziebarth et al. arise from their analysis of the standard ARMA model; this is the most well-known model in statistical time series analyses (that’s why they use it), and, as every model, it makes some assumptions. For example, the ARMA framework supposedly serves as a useful approximation to non-linear dynamics. But I think this is a risky assumption. Moreover, the statistical theory of the ARMA framework holds well for stationary time series, but not that well for non-stationary ones.

        I still don’t get why you keep talking of density dependence in this context (better richness-dependence?). Perhaps you are using it as a synonym of negative feedback? In the sense of “time-series with a stationary distribution (that is, regulated)”? If this is the case, that’s an arbitrary assumption you are making: note that, in the specific case of the paper by Peters et al. on Foraminifera, and others alike, you are looking at a trajectory of 160 millions of years of planetary species richness that look everything but stationary (see their Fig. 3); I can’t even start to imagine what sort of “carrying capacity” we must assume around which species richness fluctuate for millions of years! You are dealing with continents drifting for thousands of miles, whole ecosystems being born and destroyed, new species arising and disappearing, an asteroid messing up things, and, after all, you find a qualitatively differing effect of abiotic factors on extinctions vs. diversification. Here, time scales definitely matter.

        The kinematics we see might look the same, but the dynamics you think of probably is not. I understand your point, and I think is a good one; but I also think it is not something that can “shake the foundations of a discipline”. If your argument is purely a statistical one, that’s OK, because it is (or should be) a well-known one for everyone interested in time series analysis. But the results from Ziebarth et al. are neither original nor universal, because their modeling framework is neither arbitrary nor unavoidable.

        So dive into the literature, and write that paper in case you don’t find what you are looking for! I would surely read it!

  • Sorry, should’ve been clearer. Yes, by “density-dependence” I mean (in this context) “richness-dependence”. Is the per-species rate at which species richness of some clade changes over time dependent on its current or past richness? Glad you got the gist despite my not being very clear about my thinking.

    Absolutely, stationarity and the lack thereof is a big issue here. Indeed, quite possibly so big in many paleoecological cases (especially those spanning the longest time periods) that it just swamps all other interpretive issues.

    And absolutely, the issues Ziebarth et al. raise are by no means original to them, they certainly should be familiar to anyone trained in time series analysis. It’s just a nice teaching paper aimed at people who aren’t aware of these issues but should be.

    No, I don’t think that the issues raised by Ziebarth et al. would somehow destroy the foundations of the entire field of paleoecology. It’s commenters on my post who suggested that I was out to destroy entire fields, not me! Even if the methodological and interpretive issues discussed in Ziebarth et al. actually are news to most paleoecologists (and I have no idea if they are or not), I’m sure it’s not the case that paleoecology will simply have to start from scratch once these methodological and interpretive issues are pointed out! No established field of science has such flimsy foundations as to collapse entirely because of one methodological or interpretive mistake! As illustrated by the fact that lots of population ecologists used to make serious mistakes testing for density dependence. But yet the field of population ecology survived those mistakes, continued to collect useful data and obtain reliable results on many questions, and eventually rectified those mistakes.

    Apologies for assuming that you’re a paleo person who could give me a quick overview of whether paleo types are aware of the issues discussed in Ziebarth et al. Yes, of course I could dig into the paleo literature. But when one wants a quick overview of the literature in an unfamiliar area, in order to decide if it’s *worth* digging into the literature, the most efficient approach is to ask a knowledgeable colleague. ;-)

    • The point on “shaking the foundations of a discipline” is of course mine! With this I meant that the paleo people are indeed worried by serious problems with the precision and interpretation of theirs time series: for instance, macrostatigraphy is a rather novel area still trying to resolve some methodological issues with sampling error and interpretations, such as the presence of hiatuses in the series or the inherent trend of sedimentation and fossilization to co-occur beyond external causes. In my view, paleoscientist indeed take rather seriously statistical analysis and issues with their data; Peters et al. 2013, for instance, devote a lot of space in their paper discussing purely methodological problems and their impacts on interpretations. And, with respect to statistical procedures, if you explore the free software PAST (PAlaeontological STatistics*), developed by paleontologists, you will get a glimpse on what this people are able to use to crunch their data!

      Whether these methods are always fully justified within their field and given the usual limitations of the paleo data is a different and quite interesting issue.

      * It is amazing the amount of statistical methods available in just 4 Mb.!

Leave a Reply