Mammal evolution analyses using molecules

Now that
the large reptile tree (LRT) has grown to encompass a large gamut of mammals based on shared morphological traits, it’s time to compare it with prior studies based on molecules. Some scientists say that molecular studies that do not include fossil taxa should take precedence over morphological studies that do include fossils. Some studies combine extant DNA and extinct morphological data. In any case, it is important that all pertinent taxa are included and that unrelated taxa are excluded — and that suprageneric taxa are avoided. And finally, stand back and check your work to make sure it makes sense (more on that below).

The base of the Placentalia in the LRT
begins with small, omnivorous. plesiomorphic Monodelphis. This taxon gives rise to a number of small fur balls, all similar in size and shape, but differing subtly and nesting at the bases of more diverse and derived clades. In succession the following clades split off: Carnivora (includes moles), Glires, arboreal mammals, tenrecs/odontocetes, edentates and finally the large herbivores splitting mesonychids, desmostylians and mysticetes from elephants, sirenians and ungulates. This study provides a gradual accumulation of traits from small plesiomorphic generalists to large derived specialists and includes extinct taxa. Importantly, the basalmost taxon is very much like a basal marsupial — as it should be!

By comparison
Meredith 2011
 – begins with Afrotheria (elephants/ sirenians/ elephant shrews/ tenrecs/ golden moles) + edentates, arboreals (sans bats)/ Glires, and finally moles/shrews/hedgehogs + pangolins/carnivores + bats + artiodactyls (including hippos + whales).  This study does not provide a gradual accumulation of traits from small plesiomorphic generalists to large highly derived specialists and does not include extinct taxa. The basalmost taxa are not close to any marsupials in appearance.

Margulies et al. 2007 – essentially repeat this topology. This study has the same problem.

Tree of Life project 1995 – begins with edentates + pangolins, then Glires + arboreals + insectivores + (carnivores + creodonts) + artiodactyls and whales +  aardvarks, + perissodactyls + hyracoids + tethytheres (elephants, embrithopods, desmostylians and sirenians).. This study has the same problem.

Song et al. 2015 – begins with edentates + elephants/ tenrecs, insectivores + bats + ungulates + carnivores + other ungulates + whales, Glires, tree shrews, primates. This study has the same problem.

In a condescending tone
Asher, Bennett and Lehmann 2009 added their research to the topic of mammal phylogeny. Note how often these authors use the word ‘believe’ with regard to the best efforts of prior scientists, none of whom put faith ahead of evidence.

“In the not so distant past, there was a lot of uncertainty regarding how clades of living mammals were interrelated. Many mammalian systematists believed that sengis (Macroscelididae or ‘elephant shrews’) were closely related to rabbits and rodents, that pangolins (Pholidota) were ‘edentates’ along with anteaters, or that tenrecs (Tenrecidae) and golden moles (Chrysochloridae) were ‘insectivorans’ along with shrews and hedgehogs. Some believed that hyraxes (Procaviidae) were part of the Perissodactyla, and others thought that bats were so close to primates that the non-echolocating ones actually were primates, or at least close enough to make Chiroptera paraphyletic. In contrast, the consensus today on each of these issues is not only quite different, but also resolved with a substantial level of confidence. Questions regarding character evolution among living mammals now have the decisive advantage of a relatively well-resolved tree.”

Asher, Bennett and Lehmann 2009 – begin with a basal split between Atlantogenata (edentates + elephants + elephant shrews) and Boreoeutheria (primates/ rodents + insectivores + carnivores + bats + ungulates (including whales). This study has the same problem(s). And I, for one, have no ‘substantial level of confidence’ in its results. A ‘relatively well-resolved tree’ that does not provide a series of taxa with gradually accumulating derived traits is no match for a completely resolved tree topology that does provide that gradual accumulation. Let’s keep our thinking caps on. 

Does anyone else see
that in each of these studies, bats and ungulates nest as closely related? That the highly specialized edentates and elephants nest basal to the little furry opossum-like omnivores? The LRT does not have these problems. And yes, I’m picking the low-hanging fruit, but those kinds of problems are your clue that it is best to ditch DNA for major clade interrelationships (but keep DNA for congeneric and criminal studies) and stick to morphology when you create your own tree topology). That way you can visually check your results! Stand back from your cladogram before you publish it and see if all nodes and branches form a continuous and logical sequence with only gradual changes apparent between sister taxa. And that basal taxa look like outgroup taxa. That’s why I show my work.

When it comes to whales
Geisler et al. 2011 – nested fossil and extant odontocetes and mysticetes arising from Zygorhiza. and Georgiacetus, two archaeocetes. The toothed taxa, Janjucetus, Mammalodon and Aetiocetus were nested as basal mysticetes. Sus (pig), Bos (cattle) and hippopotamidae (hippos) were outgroup taxa. This study appears to be accurate when it comes to extant whales. But this team assumed whales were monophyletic and thus haven them a common ancestor with fins and flukes. By contrast the LRT found toothed whales arising from toothed tenrecs and baleen whales arising from desmostylians, all of which have a long diastema (toothless region of the jawline) and dorsal nares.

Asher RJ, Bennett N and Lehmann T 2009. The new framework for understanding placental mammal evolution. BioEssays 31:853–864.
Geisler JH, McGowen MR, Yang G and Gatesy J 2011. A supermatrix analysis of genomic, morphological, and paleontological data from crown Cetacea. BMC Evolutionary Biology 11:112.
Margulies EH et al. 2007. Analyses of deep mammalian sequence alignments and constraint predictions for 1% of the human genome.
Meredith RW et al. 2011. Impacts of the Cretaceous terrestrial revolution and KPg Extinction on Mammal Diversification. Scence  334(6055):521-524.
Song S, Liu L, Edwards SV and Wu S 2015. Resolving conflict in eutherian mammal phylogeny using phylogenomics and multi species coalescent model. PNAS 109(37)14942-14947.

21 thoughts on “Mammal evolution analyses using molecules

  1. There are excellent textbooks available that cover the subject of molecular phylogenetics. I would recommend consulting one before you address molecular phylogenetic analyses again.

    • Unfortunately Mr Peters is completely uninterested in methodology and the theory behind the methods he uses and discusses. As a quantitative worker, I have many times tried to point out the methodological issues in his analysis, where his matrix violates basic cladistics principles, where the things he says about phylogenetics are either misunderstandings, misinterpretations, or just plain wrong (and demonstrably so). He simply doesn’t care. The only answer you’ll get is that he has more fossils in his matrix than others, ergo his analysis is better.

  2. Not so, Neil. Remember, all of my taxa look like their sisters. And given a large number of nesting opportunities, they choose where they want to nest with minimal restriction caused by taxon exclusion. Moreover, I have shown time and again that DNA studies do not deliver similar looking sisters across large phylogenetic differences. If any current sisters should nest elsewhere, please tell me which ones and where should they go. It’s time to reexamine the DNA paradigm. It cannot be confirmed by morphology when the two clash.

    • None of those points are justification for neglecting study of phylogenetic methodology. Until you actually learn something about the methods you’re using, no one will take you seriously

      • Quite right. I find it odd that David doesn’t question his results more every time a result appears that is contrary to the consensus. And that the basic reply of every specialist is that he reconstructed the specimen wrong, be it pterosaurs, theropods, ceratopsids, whales, amphibians, etc.. Everyone in multiple evolutionary fields is wrong except for him. Here’s a question for David- if we got every phylogenetics worker in the world to sign a petition saying your analysis is fatally flawed and your results almost entirely unhelpful, would you agree?

  3. In ‘the states’ we call this a ‘snipe hunt.’ Here’s the definition: A snipe hunt or fool’s errand is a type of practical joke that involves experienced people making fun of credulous newcomers by giving them an impossible or imaginary task.

    Please, be specific. No nebulous or fantasy threats or promises. Put your cards on the table. Which listed taxa do not belong together? And where do they belong instead? And when you say wrong, please be specific? Not just how wrong, but exactly where wrong. Send images if necessary. In other words: act like a scientist. If I can do it, you can do it. And don’t forget, several of my earlier nestings are being confirmed by other workers rediscovering something that’s been on the ‘net for five years. It’s just a matter of time before others join them. Finally, recognize that paradigms are damn hard to change. Since I have no paradigms and let the data take me to results, I don’t have the same problems that you do. More later. Thank you for your continuing interest and readership.

    • And in England, we call this “wilful ignorance”. How can you complain we’re setting you an impossible task? Surely asking that you research the basic concepts surrounding a method you are using and another you are criticising is the barest minimum you should be doing. I’ve had considerably less time on this earth than you, and I’ve managed to find the time to learn this stuff. It’s really is not impossible. It doesn’t require that much specialist knowledge, and there’s plenty of readily availble sources of info. Snipe are real birds and it is perfectly possible to hunt them. Just requires a bit of effort.

      As for giving specific instances where you’re wrong, I’ve done that. I’ve more than once pointed out specific issues with your matrix. I’ve pointed out the redundant characters, single characters dealing with multiple aspects of morphology, key characters missing. I’ve pointed to references illustrating how your constant claim of a fully resolved tree is in not a good argument that you’re right, and illustrated using both simple examples and readily available references how your claim that you don’t need to add more characters is false and based on misquoting a single reference. You just refuse to listen, and more than that parade your ignorance by making arguments like clades being defined by plesiomorphies. I gave you some specific characters to code, a mere five, and your response boiled down to “meh, effort.”

      No matter how reasonable your results look, if your methodology is flawed you can’t trust them. There may be instances where your tree is right (and there’s a couple of instances where I think you may be on to something) but if you are, it’s in spite of your analysis, not because of it.

  4. Neil, you don’t need to know about internal combustion to drive a car. And when I ask for specifics, I’m asking for, as you know, specific taxa that should not nest together (repeated ad infinitum). Please resend the five characters directly or through this site. I do not recall that offer. The claim that I don’t need to add characters is not so much based on a reference, but on experience. The current list is sufficient to lump and separate a list nearing 900 taxa, way more than you expected, perhaps. The fully resolved tree, is not the only justification. That all sisters greatly resemble one another does not seem to make an impression on you. That all subsets of the tree, increasing the character/taxon ratio, are fully resolved and make sense also does not seem to persuade you. See the LRT as a series of smaller trees. That may help you accept what is happening here. I’m sure the same issues were argued when quantum physics was introduced. Hard to understand, swallow, come to grips with… but it works. Maybe, in the end, that’s all that matters.

    • You don’t need to know how an internal combustion engine works, but you do need to know the rules of the road. And when you disobey the rules of the road, you risk causing an accident. I’m not asking you to learn the precise coding of PAUP or whatever program you use, I’m asking you to learn the basic methodological concepts.

      Your argument that your method works is basically “I did it, I like the answer, ergo it works.” No, that all sisters greatly resemble one another does not make to an impression on me; vague resemblance is extremely subjective, and is not expected anyway when you have an incomplete and patchy fossil record heavily influenced by largerstatten. No, that all subsets of the tree, increasing the character/taxon ratio, are fully resolved and make sense also does not persuade me. I pointed out the flaw with this reasoning last time you made the point. Yes you’re increasing the character-taxon ratio, but only by adding in characters that are autapomorphic (and therefore parsimony-uninformative) in the subset of the tree. The large reptile tree is not a series of smaller trees, it is a single tree produced by a single matrix.

      Finally, when answering my criticisms by saying “when I asked for specifics, I meant give specific taxon combinations”. it comes across as a little pathetic. “People are welcome to criticise my work, as long as its an aspect of my work I want them to criticise.”

      The characters I suggested were: craniomandibular joint positioned posterior to the occiput or level; postparietals fused or paired; posterior shaft of the humerus convex around the exit of the entepicondylar foramen; Squamosal contribution to the post-temporal fenestra or not; tabular – split-like, expanded laterally or longer than wide; parsphenoid “wings” present or absent. I have also suggested changing the composition of the temporal fenestra characters so the fenestrae are defined by the bones bordering them. I have made other suggestions on this point, but we’ll leave it at this for now.

      • Also, Neil, when I add these five traits to the current 229 and score them for all 850+ taxa, what is your hypothesis on how they will change the tree topology from wrong to right?

  5. Your comments on my arguments are shading toward condescending.
    1. “I like it” has never entered the conversation
    2. “vague resemblance” – no we’re talking about everything in the bones from tooth shapes to relative toe phalanx proportions
    3. “heavily influenced by lagerstatten” – no. right now the mammals, for instance, are mostly extant and the rest are rare one-offs. Same with the rest of the taxa.
    4. character/taxon ratio in subsets – some parsimony uninformative, but nevertheless, the ratio rises for parsimony informative traits when this happens. Try it before you insult it.
    5. “a little pathetic” – no, if the method is flawed, how can the results be correct? Galileo should have dropped balls in a vacuum. Maybe the method is good enough, which is all it need be. You can describe an elephant with 20 questions. You don’t need 40. or 400.

    I appreciate your efforts to criticize, but doggone it, please come up with something I can use.

    to your characters:

    postparietals fused or paired; – or tiny? or absent?

    posterior shaft of the humerus convex around the exit of the entepicondylar foramen; – or not? or foramen absent? what is the opposite of convex in this case? squared off? flat? anything but?

    Squamosal contribution to the post-temporal fenestra or not; – or post temporal fenestra absent? Unfortunately occipital and palatal traits are not often presented in figures and photos.

    tabular – split-like, [splint-like?] expanded laterally or longer than wide [splint-like is longer than wide by definition]; – or absent? or fused?

    par sphenoid [parasphenoid?] “wings” present or absent. – what about just barely visible?

    I have also suggested changing the composition of the temporal fenestra characters so the fenestrae are defined by the bones bordering them. – so, in your view, a UTF or LTF is not what is seems, but should be defined another way?

    I appreciate your efforts, Neil, but traits need to be listed to consider all available possibilities, that’s why the score “or not” works so well. And to my question. How are these going to change the tree topology? What’s your hypothesis? I need to know before I start so we can confirm or refute your hypothesis, which is good Science. If you’re just asking me to add ingredients to the recipe to see what happens, that’s really the definition of a snipe hunt.

    • Neil’s posts are spot-on. To his characters, you illustrate here one of the several reasons your characters are so flawed.

      The postparietal character should ONLY be coded to distinguish fused vs. paired elements. Size or absence is a different measure, so would be a different character.

      Similarly, the absence/presence of the foramen is a different characters than the shape of the edge around it. No foramen? Code it inapplicable. And yeah, straight to concave is the opposite of convex.

      Ditto here- post-temporal fenestra presence/absence is a separate character.

      And ditto again- tabular absence or fusion (which you far too often assume for reduced elements in each taxon without convincing argument) are separate characters.

      Parasphenoid wings “just barely visible” are still present. You could make an ordered character with that being the intermediate state, if you ordered any of your characters. Which again, is a problem since you don’t.

      And no, the state “or not” is _terrible_ for analyses, because it artificially groups anything not like the specified states as more similar to each other. Just think if you had a character “color- red (0); blue (1); or not (2).” Then you’d get clades supported by not being red or blue. But why should green and purple be counted as being more similar to each other? Think of each state as if it’s part of a diagnosis or a genetic mutation.

      Because you form characters so poorly, in part due to these reasons, the data you’re feeding PAUP isn’t in the format that its algorithms are meant to make trees from. I told you this years ago and you haven’t changed it. Even if your anatomical interpretations were completely accurate, your results would be poor because you’re feeding PAUP so many inaccurate assumptions. Doesn’t it mean anything to you that no one involved in phylogenetic analysis has told you your Reptile Tree analysis is well designed?

    • As Mickey says, when you’re dealing with presence or absence of a bone, and the morphology of the bone, these need to be separate characters. This is just an example of where a little study into the basic principles of phylogenetics would help you. Wilful ignorance…

      “UTF or LTF is not what is seems, but should be defined another way?”
      Yes. You keep saying that the temporal fenestra of what most people consider synapsids is present in other clades, so it is not a good character. But in these other clades it is bounded by different bones, and so it is more likely it is convergent

      1. Yes it has. Your only justification for why your method works is it produces results that *you* think seem reasonable

      2. The resemblances are vague as they are based only on the characters you deem worthy

      3. The palaeozoic and Triassic tetrapod record is incredibly patchy and hideously afflicted with largerstatten. I’ve spent a goodly portion of my career demonstrating this.

      4. Basically your argument is “yes I have very few characters relative to my taxon list, but when I only count a small portion of the taxon list, the ratio rises.” The phrase “well duh” springs to mind. How is that in any way relevant to how accurate your tree is? The only reason the ratio rises is because you are including a load of characters that are relevant to a different subest of the tree and are autapomorphic for the subset of interest.

      5. And how do you know your method is good enough when you take no time to research the methods? Your only argument that the method is good enough is “questions I ask are sufficient to lump and separate all the taxa and every node”. That is not a justification. Saying “I get a result” is not an argument that the result is correct and therefore not a justification that the method is sufficient.

      Basically I can think of only three reasons why you would not want to study the methodology.
      1) seems like too much hard work
      2) you don’t think it important to know about the methods you use
      3) you’re scared to find out how flawed your analysis is

      None of those reasons are grounds for good science. None are deserving of respect. All show you to be either willfully ignorant or intellectually dishonest. That is why I sound condescending.

      • Neil, your approach is so degrading and inappropriate. Like any human, I tend to disregard those who practice blackwashing. A little sugar on the pill will help everything else you say go down better. For instance, you could remind yourself that several of the relationships I have recovered using my ‘flawed’ methods were later ‘discovered’ by PhDs using perfected methods. Most of the rest of the issues have not been tested yet.

        Of your 1,2,3 thoughts, #1 seems right, but it also would be too much hard work for any other worker you would like to impose this on. #2 Important? that’s always relative: i.e.. how important? I’m pushing forward on several nodes now, with little time, at present, to look back. When things settle down, retrospection will be more appropriate. #3 Scared about flaws? I’ve often said, and you seem to have forgotten, I’m eager to uncover any and all flaws in scoring. However, not in the character list at present (see #2) which, despite the pinging and smoke keeps getting me to new destinations. As you know, I uncover scoring flaws on a regular basis. The problem, from your perspective is, repair of mistakes usually cements recovered relationships. Yes, that can happen too.

        By the way, are you conceding that the topology of the tree is correct? I don’t hear any corrections with reasons coming from you on that matter. If so, I appreciate your efforts to nip and tuck the rough edges to withstand any and all criticism.

      • Moreover, Neil and Mickey and MOST importantly, I don’t expect ANYONE to use my character list. So why spend the time to fix it? No PhD in his/her right mind will ever use it. And we all know that.

        On the other hand, I DO hope and encourage others to use various subsets of the taxon list that the LRT recovers. It’s just a list of genera and specimens. No controversy there. So far, several PhDs have done so and that’s great. Others may follow.

        This is my pursuit of happiness. After making every discovery, and there have been so many, seeing others confirm them is where my future satisfaction will come from. Hope that settles things between us. I know this frustrates you.

        The taxa are flawless. The characters and scoring will always be flawed to some degree.

    • Oh and I made the mistake of telling you how I thought these characters would affect the results before. You used it as an excuse to just code a single relevant taxon. I want you to code them for every taxon and see how it affects the result. No preconceptions, no just trying to prove me wrong. Code them without bias

  6. Interesting outlook… and yet, somehow, the questions I ask are sufficient to lump and separate all the taxa and every node. In looking at other characters wrought by other workers I thought of them as insufficient or inefficient for not covering all available possibilities, which “or not” takes care of by definition. I’ll put some thought into this.

    • But our computer programs are not looking for an “or not”!!!! They don’t know what the character states mean. They don’t know that the first two character states deal with the morphology of the bone and the third deals with its absence. The computer is working with numbers, and has a set of assumptions about those numbers. By having a character where some states describe the morphology of the bone and some deal with its presence or absence, you are violating those assumptions. And if you studied the methodology you would know this, and you would know how badly it can affect the results.

      Read this

      Click to access 57422f3308aea45ee84a3805.pdf

  7. So

    “1 seems right, but it also would be too much hard work for any other worker you would like to impose this on. ”

    Given most workers doing these analyses learn these concepts while they are undergrads, I would say you have a very low bar for what constitutes “too much work.” I hold every published/attempting to be published phylogeny to these standards. The fact that you are self taught means nothing to me. It does not preclude you from making new discoveries, but nor does it excuse you from having to learn everything we had t learn. Yes, educating yourself in these matters takes a lot of time. It takes a lot of effort. It requires trawling the literature (and without tutors to aid you in your search; yet another cross you bear as a self educated palaeo). But that’s just the way it is. Becoming an expert takes time and effort, and you can’t take short cuts.

    “Important? That’s always relative. But how important?”

    I gave you a reference demonstrating in great detail precisely how important good character construction is. I assume you just ignored it, so spoiler alert, the answer is “very important”. Your excuse that your analysis is “good enough” simply does not fly, and all the reasons are in that paper, along with specific examples showing how it can make a difference.

  8. I did not ignore it. Take a look at today’s blogpost. I cover the Simoes et al. 2016 who reexamine, as you know, Gauthier et al.. 2012 and Conrad 2008, who having the best of everything, failed to come to a consensus — and likewise Simoes et al. were not able to bring the two studies into harmony. They report, “This indicates that the problems of character constructions as discussed here are inherent to most of our current notions of character construction methods for squamate
    cladistic analyses.”

    They also report, “Morphological data can help in resolving deep nodes or branching patterns in fast evolving areas of a tree (Beutel et al., 2011) and solve relationships that cannot be considered by genomic data.”

  9. David Peters, Why are you bantering with these people? The LRT is history-making and will guide research for generations to come, long after these snipes are forgotten.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.