Translate

Where have all the ancestors gone?

In contrast to neontologists, palaeontologists deal literally in the past. However, like their modern counterparts, most palaeontologists seem to have the opinion that although we all agree evolution is a fact, we will never have to deal with actual ancestors.

Things evolve, but there are no ancestors 

Most times when you see a phylogeny, an evolutionary history, of a long-gone group of organisms (e.g. dinosaurs), it is in the form of a stick graph – a series of dichotomies, with each taxon being on the tip of a branch. Here an example from a recent paper published in Royal Society Open Publishing by V. Fischer, R. B. J. Benson, et al. on polycotylid plesiosaurans (people like me that have no idea about extinct "thunder-lizards" know them as the four-paddled, long-necked swimming dinosaurs)

A phylogenetic tree of a subgroup of plesiosaurians (modified from Fischer et al., 2018, fig. 8a). The Bremer decay index is a support measure only seen these days in palaeontological and neo-morphological studies. >1 means that moving the branch would add >1 steps. In other words, most clades seen in this tree are just one step more optimal than their alternatives. Which is not a lot, regarding that the total tree length is 1706 steps and the pretty impressive dimensions of the compiled matrix: 270 characters scored for 118 taxa (58 MB, the file includes 50000 trees).
This tree shows three features seen in many fossil-based trees (the paper of Fischer et al. includes things not so common in palaeontological literature, so it's worth a read).
  • No branch support values known from modern-day trees such as bootstrapping or jack-knifing values are provided that could inform us about the robustness of the signal for each shown branch. Or Bayesian-inferred probabilities. The reason for this is simple: they are usually too low to show (an opinion of many authors and most reviewers, I think any value is worth showing).
  • The first member of the lineage, the oldest taxon Brancasaurus from the Berriasian (145–140 Ma), is placed as first-diverged sister to all others. Which is exactly the place where you would resolve the common ancestor in a tree. This pattern is also predominant within the clades: Older forms are resolved as early branching sisters to younger ones. Again exactly how ancestors would be resolved in trees
  • A large "basal" (i.e. root-proximal) polytomy — the tree shown is a strict consensus tree of equally optimal tree solutions, hence, the polytomy here is a "soft" one, reflecting that the data cannot decide on one alternative.

Franz Hilgendorf's famous tree depicting evolution of fossil snails (left) and its standard visualisation/reconstruction (right) as it would be presented in a phylogenetic paper since the 1980s. Note the position of ancestors and their descendants (arrows).

An implicit assumption of graphs like the one shown by Fischer et al. is that there are no ancestors in the fossil record: whatever fossil we find, it represents an extinct sister lineage of something that lived later (or still today). There is little doubt that birds are the last surviving lineage of a particular dinosaur lineage, the Coelurosauria, but the oldest birds and related feathered dinosaurs from the Mesozoic were all evolutionary dead-ends and died out. The actual ancestors of all birds, Adam and Eve Birdie, hide in obscurity, and will never be found. Why? Because the terrestrial fossil record (at least) is extremely patchy, so the probability to find an actual ancestor must be very low.


A simple evolutionary tree. Each circle represents a population at a given place in time and space. Every 10th population is represented by fossils (filled circles), the colours represent distinct (main) morphotypes. Even though most lineages (75% of the morphotypes) are represented in the fossil record, there are no ancestor-descendant relationships to worry about. Abbrev.: CA = common ancestor; PO = point of origin. [I really just counted to 10, and this worked quite well until I reached the upper part of the tree where I would have ended up with two ancestral populations, so I bent my rule a bit and took the one before or the next.]

When we look at this: It's indeed unlikely that the few fossils preserved to be found are individuals from exactly the population that evolved and diverged into new forms, species, and genera.

And so, it is understandable, when a theoretical paper (pre-print) about the fossil record and its use for molecular dating shows the following figure.
The figure accompanying the paper of Hopkins, Bapst et al., arXiv (2018). The stars and blue lines represent fossils providing useful (scorable) data; the red lines molecular data (here: including subrecent "ancient DNA" samples).
In reference to this figure, the authors write (p.9): "Paleontological data thus complicates any analysis where a taxon needs to be assigned a single precise age, requiring some treatment to deal with persistent morphotaxa, and leading to the so-called ‘times of observation’ problem" and do not further delve what this figure shows as well:
  • Although each terminal branch (extinct and those surviving) is represented in the fossil record (and quite well), none of the internal branches is. There are no ancestral lineages in the fossil record.
  • The 'times of observation' problem relates to a reconstruction problem. If the stars and blue bars are not identical – i.e. represent phases of evolutionary stasis – we would not resolve them as part of one branch as depicted, but as extinct sister lineages.
  • Do persistent morphotaxa in the fossil record represent populations, species, or even higher biological taxa? Because if it's the latter two, we may have to deal with ancestors again (see below).
The many palaeophylogenetic studies represent the ultimate proof for NID — Non-Intelligent Design (a fun article on the topic). Everything the Lord ever shaped (or triggered to evolve) in such great quantity that we still can find it as a fossil, was a failure. Evolutionary dead-ends, with no exception. Good news for the creationists sitting in the U.S. cabinet (Pence, DeVos, Carson): All those ancient hominids, we were lucky to find, and none of you is our ancestor. And cladists: No ancestor-descendant relationships to deal with, so we are safe with just identifying sisters. And for naturalists philosophing about future life: it's those irrelevant animals and plants you are not thinking of who will radiate and evolve. For sure not the common and widespread ones.

Do ancestors get lost over time?

Of course, this is nonsense. For hominids, we can be pretty sure that some of the bones we found are from our ancestors, the hominid species/lineages which eventually evolved Homo sapiens. For the youngest part, we even have genetic proof that the formation of Homo sapiens sapiens, Modern Man, was not entirely straightforward, but involved some cross-population intercourse (an interesting question for alt-right Christians like the current vice-president of the United States: Is it sodomy to breed with a Neanderthal?). And their total numbers are considered to be very small. And we have little reason to assume that the 15–5 Ma old oak leaves you can find in hundreds or thousands in coal mines of Greece and Turkey are from extinct oaks and not from the ancestors of at least some of the oak species that live today in the Mediterranean region (and may have very similar to identical leaves).

And the same applies to any other patchy terrestrial lineage. Let's go back to our hypothetical example and add some names for populations sharing a common origin and morphotype, i.e. combine them to "species".

Ancestors and descendants. Based on lineage (common ancestry) and morphology, taxa can be defined: The four surviving ("Modern") species – P*,T*, U*, and X*; their ancestors ("Precursor") B, C, D, F, and W; and their extinct siblings E, G, H, S, and V. Bracketed names: additional "phylogenetic" species or "chronospecies" that can only be defined via time (or place).
In the example we have four surviving "modern" species: P*, T*, U*, and X*. All of them are monophyletic in a strict (Hennigian) sense, their members share an 'inclusive' common origin. If they survived until today, we can access their genetic code an establish their phylogenetic relationships: X + { P + ( T*+U*) }, in the best case perfectly mirroring the true tree. Three of those species (P*, T* and U*) are descendants of "Precursor A": P* via B (being the sole surving lineage of B) and the sister taxa T* and U* via C and D.

In my Twitter discussion on the topic, my counterpart pointed out (by the way, if you like palaeontology and are on Twitter, you should definitly follow him): "The lineage, yes, but not necessarily the direct ancestor. Ancestor is a very very precise statement: not merely an early relative." — A particular way to define them, but I think, this is splitting hairs. We don't have the data to distinguish between "mere early relatives" (i.e. aunts and uncles, great-aunts and uncles etc.) and direct ancestors (i.e. parents, grand-parents etc.), and we don't need to. Our aunts and uncles can be as similar to us than our parents.

Back in time the population that evolved into/gave birth to the next species – the direct ancestor – was part of a species: the phylogenetically first member of e.g. P* lineage was part of species B at its time. And B descended (evolved) from a member of species A.

In case of extinct organisms, recognition of species and other biological taxa (units) is directly linked to the morphotype as we have little other ways to define a taxon (I'm going to explore this a bit in future posts). When we put up hypothesis about evolution and phylogenetic relationships, it doesn't matter whether the fossils sample the actual population of species B that evolved into P*. Morphologically and in an evolutionary context, B is ancestral to P*, as much as A is ancestral to B. Like (some of) the first hominids are our ancestors, and the first tiny but horse-like creatures those of our modern-day horses. And wolves are ancestors of dogs. And at least some of the fossil giraffides and elephantides are the likely ancestors of the African giraffes and elephants, whereas the mammoth is an extinct sister lineage. So why should that not apply to largely extinct groups of organisms?

Of course there may be A-morphs that are not part of the ancestral sublineage that lead to B and P*, but this is something we have to live (or deal) with. C for example descended from other members of species A. This makes B and C sisters, and A their common ancestor. Hence placing A in a polytomy or as sister to B and C in a tree doesn't get it right. And calling A an ancestor of B and C is not wrong, even though some later, surviving individuals with A-morphology may be sisters of B and C.

Close-up on the evolution of the (pre-crisis) ABCD lineage. Since our example is a perfect sequence of dichotomies (no intra- or inter-species reticulation, which pretty often occurs in nature), we have a single monophyletic (A) or reciprocally monophyletic sister species (later A, B, C, then D) for each time period. But as soon as we include A, B, C, D from different time periods, we have to deal with ancestor-descendant relationship. A is the ancestor of B and C, and C the ancestor of D. 


To quote (first-time not in a response to a reviewer's comment) from Joe Felsenstein's famous 2004 book "Inferring Phylogenies", we should not only consider the possibility that some fossils are ancestral to others (and modern species/taxa), we should be encouraged to do so. Especially, when it comes to phylogenetic inference.

It's just cladistics, stupid!

But then, why do so many researchers reject the notion to find an ancestor? Solely because of the "cladistic" framework they rely on. Ancestors – or ancestral lineages and forms – make things much more difficult. No currently used inference method to infer a dated or not dated phylogenetic tree can handle ancestor-descendant relationships. They treat each taxon as a terminal and do not allow placing taxa (even much older ones) on internal branches or nodes (the original fossilised-birth-death dating being an exception of some sort, see e.g. our paper on Osmundaceae rhizomes — Warning! may include ancestors). They all assume that each fossil included in the analysis can be added as a discrete branch to the tree. And unambiguously so (something that remains a dream, even in Bayesian dating). No ancestors, no problems with the cladistic approach that only identifies sisters.

And cladistics have one great advantage: it takes no brain or prior knowledge to read a phylogenetic tree (it takes a bit more to understand, why the computer came up with that particular tree). In the plesiosaurian tree above, everyone can read who is the sister of whom. But if we move to a method that outperforms trees when it comes to ancestor-descendant relationships such as the neighbour-net, we face something like this.

A neighbour-net based on a mosasaurian matrix. See this post for equally puzzling graphs.

There is only one (never used to my knowledge for palaeontological data) method to infer something like Hilgendorf's evolutionary tree: the median network.

The median network family, a method that explicitly allows inferring ancestor-descendant relationships.
The median network shows all parsimony solutions to the data while allowing to place taxa at the nodes (the "medians"), thus, directly depict an ancestor-descendant relationship. Problem is: real-world morphological data sets are full of homoplasious characters and tree-incompatible signals. Being parsimony-based, the median networks are vulnerable to homoplasy-induced reconstruction artefacts (red branches in the graph; the commonly used parsimony-based trees are even more vulnerable). One is rarely shown is the complexity of signals in morphological matrices used to infer (parsimony) trees.

A strict consensus network showing all equally parsimonious topologies for another dinosaur data set. See this post for more details.


Consequently, we are swamped with cladograms (trees without meaningful branch lengths) and phylograms (trees with branch lengths) boosting exclusively evolutionary dead-ends, and, more recently, an increasing number of explicitly or manually dated trees (chronograms) without a single fossil – or absolute no-go: taxon – placed on an internal branch (but see the Wikipedia entry on Osmundales/Osmundaceae, guess who's behind that update).

No ancestors, no worries! Evolution, who cares? Just stay on the safe side. But without hypotheses about ancestor-descendant relationships, we are kept clueless where all those beautiful forms come from and what was their origin, what triggered their change? Evolution is change over time, and this requires to fill the links with live. Early evolutionary biologists who had to use their brains and not computers to infer phylogenetic relationships, were not afraid of ancestors at all.

The current header of The Genealogical World of Phylogenetic Networks showing the doubly heretic (from a cladistic viewpoint) evolutionary graph published by Franz Hilgendorf over a century ago (see also this post by David Morrison). Not only that some fossils are placed as ancestors of other, it also includes a secondary reticulation, hence shows an evolutionary network rather than an evolutionary tree.


Face the enemy, and stop hiding behind semantics

An ancestor or ancestral form, separating both is impossible in practise until we invent time-machine to genetically sample what we dig out as fossils, will inevitably inflict topological ambiguity, because the tree needs to place that ancestor either as sister to all, or to part of its offspring (in the best-case scenario). Both of which is wrong, because as an ancestor, it should be placed on the root branch of the clade including all its offspring. And although there are no ancestors in the fossil record, there is a hell of topological ambiguity in inferences based on data matrices that include fossil taxa, especially when from different time periods (see links provided below).

Just take our example to keep thing simple. Let's assume we have a fine fossil record, pre-crisis every third time period is covered, post-crises every second. Geography-/population-wise we lack only every second region. The fossil record would look like this, and just based on shared morphological character suites (traits), we would identify seven extinct species (A to S, Z) in addition to the four surviving species P* to X*.

A well sampled fossil record, on which we can base on phylogenetic analyses.

Now let's assume further that the morphology perfectly reflects the evolution of our group (which, in the real world, is hardly the case, but that is a different story). We can then score a (perfect) matrix of morphological traits and infer a tree (in the perfect case, it doesn't matter which optimality criterion/method is used, parsimony, maximum likelihood, Bayesian inference or distance-based will all find the same). Considering the time-span covered by each (morpho)species, we would get the following sort-of-dated tree (chronogram) for our fossil sample.

The standard ("cladistic") solution: a time-aware tree for our example. Note the aspect-wise similarity to the real-world plesiosauran tree and Hopkins et al.'s figure.

Because our matrix is perfect, our tree has no misleading branch. But all ancestral species are either not resolved or as short branched sisters of their descendants, the evolutionary pathways A → B → P*, another part of A → C which then diverges into S + T* and U*, are not visible. The tree does not provide any idea about how A, clade BP*, and clade CST*U* are related to each other, instead we face a so-called "polytomy". The second large "basal" polytomy is inevitable because the earliest species are the result of a fast ancient radiation followed by complete lineage sorting (meaning that each emerging lineage turned out to be morphologically coherent and distinct to all others, but we have no possible data to decipher the sequence of the speciation process). And there's little difference between ancestor-descendant relationships and actual sister relationships, the only clue may be the length of the terminal branches. What we would like to know, regarding evolution (being an evolutionary biologist, and not a creationists or lamarckist), is hardly depicted.

The evolutionary tree that we would like to show, well, should show, because it reflects all aspects of the true tree captured in the fossil data and provides an evolutionary hypothesis, could look like this.

An evolutionary tree (a 'coral-type' tree metaphor), providing a solution to the same problem while showing ancestor-descendant relationships across time.

Because B fossils overlap with P* fossils evolved from B, we can infer that latest B are sisters of P*, and earlier B include their ancestors. But this requires the a priori realisation that B is the ancestor of P*. And provided a perfect matrix (and inference methods that can deal with or visualise ambiguous signals such as support consensus networks and neighbour-net splits graphs or can place taxa at internal nodes such as the median networks), we would conclude that our Z fossils are just the left-over of the ancestral species, which included the common ancestor of our entire group. [Unfortunately such an evolutionary tree would conflict with the current naming rules, which are either based on Hennig's original concept (only monophyletic taxa should be named) or cladistics (often called "phylogenetic" species concept). Following the rules, Species A, B, and C would need to remain nameless. However, no one is really following the rules.]

Forget about cladistics, and focus on putting up hypothesis about evolution

So rather than come up with highly particular definitions of what can be called an ancestor to ensure that no fossil can be one, so one can keep on inferring more or less supported "sister relationships", we should explore the methods and data we have at hand by all means to identify potential ancestor-descendant relationships. And depict them. Make trees evolutionary graphs again! Let's start using our necessarily imperfect inferences to put forward truly evolutionary hypotheses for our groups, and not just computer-generated, pretty trivial stick graphs.

Because Pandora's Box known as Mother Nature playing evolution has more in stock. Would we recognise an ancestor as such, even if we find it? When you see how plastic morphologies are in modern species and (widespread) genera, there is a chance that some ancestors may differ from all their offspring, their descendants ... which brings us back to the patchy terrestrial fossil record: how can we be sure that all traits we observe on a fossil (which are quite a lot in case of dinosaurs and other vertrebrates) are representative for their population, species, and genus?


Some recent posts on signals in matrices including extinct organisms
 More philosophical stuff
Practical tips
Many more interesting posts that go beyong trees (or under their skin) can be found on the Genealogical World of Phlyogenetic Networks.
 

5 comments:

  1. It's sad that no one has commented on this page since it was set up. Someone could have mentioned the work that Richard Zander was doing in producing trees that show direct ancestry. One can read about it here:

    https://en.wikipedia.org/wiki/Evolutionary_taxonomy

    A note on terminology: cladists would have you think that "phylogenetic trees," by definition, means trees with all organisms at branch tips. Zander accordingly calls his trees "evolutionary trees." When I mentioned this to a zealous cladist, he called this move "unfair!" and thus displayed his use of double standards.

    ReplyDelete
    Replies
    1. I enjoyed his papers, too, having crossed them when I was just a fledgling. His paper on branch support is one of “two papers you may want to read before inferring trees from morphological (or other) data”. If more people would have read his papers, more reviewers acknowledged the issues in them, a lot of pointless and inherently instable re-classifications could have been avoided.

      We also used "evolutionary tree" often in contrast of "phylogenetic tree" because reviewers made us believe and misconceived that the latter is a propriatory trade mark of cladistics. And we never just wanted to infer phyl(ogen)y, we wanted to reconstruct the evolution of our study objects.

      But of course, since phylogenetics are (substantially older) than cladistics, such a distinction may not be really necessary. We just need to correct any cladist telling us otherwise, with the very concrete example at hand: the paper just reviewed by the cladist peer. (It was tough at the start, after a decade of battling the Mighty Beasts in the Forest of Reviews, we faced little to no resistance.)

      David Morrison has pondered such terminology issues in various posts on the Genealogical World of Phylogenetic Networks.

      A nice early collection of very evolutionary trees are in his 2012 post Evolutionary trees: old wine in new bottles?

      While originally (before Hennig, Farris, etc.) phylogenetic and evolutionary graphs (trees or networks) where and still usually are synonymous, it could make sense to discern between them because phylogenetic methods can be handy to classify, categorise, sort or just visulise non-biological data including data sets that are not the product of evolutionary processes, i.e. not depicting ancestor-descendant relationships (such as U.S. breast sizes, world happiness, Jovian and Saturnian moons)

      If I use a phylogenetic method like Maximum likelihood or Neighbour-Net, I infer a phylogenetic tree or (meta-)phylogenetic network.
      But only if this phylogentic graph depicts an evolution of some sort (biological or a-biological, e.g. the evolution of Linux systems), it is an evolutionary tree or network.
      Personally, to depict the inferred evolutionary pathways, I do prefer other metaphors: cactusses and corals. But since they are interpretations of algorithm-based inferences, not computer-generated graphs, reviewers usually refuse them as "Results", one has to move them to "Discussion" or the supplementary information.

      Delete
    2. PS Regarding the lack of comments. This is not surprising. First, my Res.I.P blog is not really trending: right now, this particular post has accumulated 666 "views" (quite fittingly...)

      A rule-of-thumb is that 25%-50% of Blogger-views are bots. Plus, there are probably multiple views by the same recipient (only Google will know).

      Let's say this post has been viewed by a 100– 200 different humans.

      My posts are generally long, too long for most blogpost consumers: the number of actual readers will hence be even lower. Maybe ~50.

      Given the provenance of my post's views, I suppose most of the &plusm; regular human viewers are people familiar with my work, inevitably sharing my philosophies (stemming from dealing with real-world data not coming from a philosophical viewpoint). They have no reason to comment; they simply enjoy reading and maybe pick up the one or other idea.

      That leaves maybe a dozen or two new ones accidently crossing this when browsing for a definition. Most will be freshlings (an establish researchers knows the truth, doesn't have to look anything up, right?) They may find it very interesting but wouldn't feel confident to comment. At best they discuss it, try to apply it in their own research. At worst, they discuss with their cladist supervisor and are informed how wrong they are.

      Now, there are a lot of established researchers strongly opposing the content of this (or any other my posts, and our peer-reviewed papers). They wouldn't have any reason to comment either, because any faith thrives on the simple ignorance of facts contradicting one's believes. Let's say such a person would read beyond the first line through the entire text: such a person's could only be profoundly negative.

      So, it's actually a good sign to have no comments. It shows, in the optimal case, that the faithful, if any of them has read this, can't think of an argument against it :)

      Delete
  2. Glad to hear from you, Das Grimm. I've posted to forums of your caliber from time to time, but one or two years "too late" and almost never got a response.

    Let me tell you a little bit more about the cladophile [1] whom I told you about. His name is John Harshman, and we have regularly crossed swords in the Usenet newsgroups talk.origins and sci.bio.paleontology since December 2010. We tolerate each other because we are the two best paleontology and evolution and systematics "experts" [2] in these newsgroups and we convey enough scientific information on our posts to keep each other reasonably content. I have learned a great deal in this way.

    [1] My term for those who will not countenance paraphyletic taxa nor any offshoot like trees showing direct ancestry, such as Richard Zander's, in peer reviewed papers in leading journals. Harshman is such a fanatical cladophile, he has complained about this term even though it does not have any negative connotations, probably because he knows how much I dislike that attitude.

    [2] Right now, there is an ongoing discussion about the proper use of the terms "paleontologist," "amateur paleontologist," and "real paleontologist." My stance is that neither of us qualifies as any of the above: to be amateur paleontologist, I would have to participate in at least one fossil hunt in each of ten years, and so far I'm only up to three. To be a "real paleontologist" I would also have to contribute significantly to a research article in a respected peer-reviewed journal. And to be a "paleontologist" without qualification I would have to be paid for this kind of work on a regular basis.

    This discussion is important because there is no such thing as a licensed paleontologist. This is in contrast to the existence of licensed geologists.

    ReplyDelete
    Replies
    1. Regarding the rejection of paraphyletic taxa, it's important to realise that this the official scientific standard, also among neontologists. I very early stopped discussing this with those you called “cladophiles”. It's really just the people who work at the coalface of evolution (within generic lineages, above and below species, where all the fun stuff happens) and at the intersection of neo- and palaeontology who beg to differ—we like to give fossils Linnéan binominals, and this is simply impossible in the presence of potential ancestors/ ancestral forms without either collapsing well-circumscribed, coherent holophyletic modern-day taxa or introducing paraphyletic genera (see my posts on “What is an angiosperm?” for a prominant case). Professional (i.e. being paid to do science), ex (i.e. not being paid anymore) or amateur (i.e. never been paid to do research): anyone really dealing with the data we handled, simply cannot adhere to the theory of cladistic systematics. Which, by the way, much differs from the practise: we have numerous “non-monophyletic” accepted genera and modern-day species; and for a vast number of fossil genera/species, it has never been tested, or cannot be e.g. for any fossil-species (and extinct genus) based on a single skeleton.

      From a purely professional point of view, speaking a formely tax-paid scientist dealing with evolution and biodiversity, it's utterly instable and pointless to define all biological taxa, including genus-species pairs – the biological units we ultimately name (and count for biodiversity) – merely on their position in an inferred phylogenetic tree.

      But terminology is always something a scientific community should be able to discuss and adapt, where needed, and even on a case-by-case basis. There is no ultimate model or truth. Mother Nature is simply too complex, no matter how fancy the (thought) model, it will always beg to differ in not few cases.

      Delete

Enter your comment ...