ADVERTISEMENT
  About the SA Blog Network













Tetrapod Zoology

Tetrapod Zoology


Amphibians, reptiles, birds and mammals - living and extinct
Tetrapod Zoology Home

50 million years of incredible shrinking theropod dinosaurs

The views expressed are those of the author and are not necessarily those of Scientific American.


Email   PrintPrint



Theropod dinosaurs encompass a huge range of body sizes. This illustration shows a Broad-billed Hummingbird (Cynanthus latirostris) in front of a tooth of the giant allosauroid Carcharodontosaurus. Images courtesy of Terry Sohl and Christophe Hendrickx.

Some time round about 165 million years ago, the group of small, feathered dinosaurs that we call birds evolved from within the theropod radiation (theropods are the so-called ‘predatory dinosaurs’: the great group that includes animals like Tyrannosaurus and Velociraptor as well as the birds). As anyone reasonably familiar with recent palaeontological discoveries will know, we’re now aware of a large number of fossils that blur and smudge whatever distinction there might have been between ‘dinosaur’ and ‘bird’. Archaeopteryx may or may not be one of the oldest members of the bird lineage, but it’s evident that – when it was alive – it belonged to just one of several lineages of small, long-legged, fully feathered, omnivorous or predatory theropods, all of which possessed a mosaic of ‘bird-like’ and ‘unbird-like’ characters. [T-shirt design shown below available here.]

Oh yeah, there's this t-shirt design!

Indeed, if we look at theropod history across the whole of the Triassic, Jurassic and Cretaceous, we see a gradual, cumulative acquisition of bird-like features, ranging from wishbones and a pneumatised skeleton to complex feathers, a reduced, three-fingered hand, an enlarged sternum (breastbone) and tiny size.

This trend of increasing birdiness is well known and has often been commented on. Allosaurus is more bird-like than Coelophysis, Tyrannosaurus is more bird-like than Allosaurus, Ornithomimus is more bird-like than Tyrannosaurus, Velociraptor is more bird-like than Ornithomimus, and Archaeopteryx is more bird-like than Velociraptor. Incidentally, while this general trend might imply that theropods are a ‘bird factory’ (Holtz 2000) and not much else, remember that evolution is about divergence and diversification, with the umpteen lineages that split off along the way all having complex, intricate and independent histories of their own.

Even big birds look small compared to 'average' non-bird theropods. This composite shows the Mesozoic theropods Allosaurus and Carnotaurus, and a selection of Cenozoic ones. Image by Darren Naish.

The fact that the non-bird theropods closest to birds are small has long been evident, but is there really a supportable trend of size reduction in theropods along the lineage that led to birds? This idea has been both supported (Sereno 1997, Padian et al. 2001, Turner et al. 2007, Novas et al. 2012, Dececchi & Larsson 2013) and contested (Hone et al. 2005, Sookias et al. 2012), but even those studies that supported this trend looked at just a few parameters, did not combine the data on trends with that on how quickly evolution was occurring, or did not analyse data on anatomy, stratigraphic age and body size simultaneously. Frankly, compiling and running analyses that incorporate all these different lines of data is a gargantuan task. But, with the University of Adelaide’s Mike Lee at the helm, this is exactly what a group of us have just done, and our results appear today in the journal Science (Lee et al. 2014a) (see also Benton 2014). In addition to Mike and myself, the team incorporates Andrea Cau and Gareth Dyke.

By applying Bayesian inference to two different datasets of theropod anatomy (Andrea Cau’s mega-matrix project – the most character-rich dataset yet compiled for theropods – and the so-called TWiG database combined by the AMNH-based Theropod Working Group), we produced results that simultaneously account for phylogenetic position, divergence date, evolutionary rates, body size and other factors as well (Lee et al. 2014a). I’m not going to pretend that I understand all of the underlying mathematics, but if you have a basic concept of what Bayesian inference is – that it involves inferring the probability of a relationship based on the addition of further data – then you can appreciate the methodology underpinning this study. We previously applied Bayesian methods to an analysis of the bird fossil record (Lee et al. 2014b).

Diagrammatic phylogenetic tree (from Lee et al. 2014a) showing how size decreases along the theropod stem leading to the origin of birds. Decreases in size are marked in pink, increases in size in blue.

The results? As revealed by the title of the paper (and the title of this article), what has emerged is a robust and well-supported model showing a prolonged, directional trend in size reduction in the theropod lineage leading to birds: a trend that is continuous across 50 million years of theropod history, and which shows the animals at successive nodes becoming ever-smaller as we get closer to birds in the phylogeny (Lee et al. 2014a).

Theropods display a continuous, pervasive decrease in size when we look at the inferred size of ancestral species at successive nodes across the lineage leading to birds. From left to right, this illustration by Davide Bonnadonna shows the ancestral neotheropod (~220 Million years old), the ancestral tetanuran (~200 myo), the ancestral coelurosaur (~175 myo), the ancestral paravian (~165 myo), and Archaeopteryx (150 myo).

The phylogeny we recovered is an essentially ‘standard’ one for large datasets, with major clades diverging in the following order: megalosauroids, allosauroids, tyrannosauroids, ornithomimosaurs, alvarezsauroids, oviraptorosaurs, dromaeosaurids, troodontids, birds (Lee et al. 2014a). Note that, even in those groups conventionally imagined as ‘big’ (like megalosauroids and allosauroids), the sizes estimated for ancestral species are ‘only’ in the 100-200 kg ballpark (Lee et al. 2014a), meaning that classic taxa like Megalosaurus and Allosaurus should not be considered typical of the ancestral body size for these groups. Elsewhere in the phylogeny, our depiction of troodontids as closer to birds than to dromaeosaurids is now pretty familiar; it means that Deinonychosauria is not monophyletic. The recognition of a maniraptoran clade that includes oviraptorosaurs, dromaeosaurids, troodontids and birds (and not therizinosaurs and alvarezsauroids) is also looking increasingly familiar; the clade concerned was recently named Pennaraptora by Foth et al. (2014).

Theropod cladogram mapped across time, with the different colours showing the approximate size ranges present in the respective part of the tree. This is a slightly different version of the tree from the one included in the final paper (Lee et al. 2014a). Note the emergence of a large number of new lineages at round about 170 million years ago - it seems that especially rapid evolution occurred at this approximate time.

Because our results allow us to map the appearance dates of lineages onto the phylogeny, we can see that evolutionary rates across part of the lineage leading to birds occurred much faster than expected compared to the rest of the tree – up to four times faster, in fact (Lee et al. 2014a). This seemingly explains why several groups of tetanuran theropods – allosauroids, tyrannosauroids, compsognathids and others – appear near-simultaneously in the fossil record: it seems that the time intervals between their originations really were very short. Why evolution was occurring so rapidly in these animals remains, of course, an unknown.

So, birds emerged as the ‘end products’ of a sustained, long-term period of miniaturisation. It should be said that the existence of this trend will not be surprising to the vast majority of dinosaur specialists, since the idea that birds and their non-bird maniraptoran relatives are relatively small compared to other coelurosaurs, and that coelurosaurs are generally small relative to other tetanurans, has often been noted. However, confirming this size reduction as part of a grander trend that involves non-coelurosaurs and even non-tetanurans is largely novel, and our results are the most comprehensive and statistically robust produced so far.

Three theropod innovations that might have evolved as a corollary of ever-smaller size: gliding behaviour, complex, extensive feathering, and a posture where the thighs are more horizontal and the centre of gravity has shifted posteriorly. Images (top to bottom): Emily Willoughby, John Conway, and from Gatesy (1990).

What impact might this long history of miniaturisation have had on the anatomy, biology and behaviour of the theropods that were ancestral to birds? In recent years it’s been noted that a number of unusual features are associated with those smaller-bodied theropods that are antecedent to birds in the phylogenetic tree: proportionally short snouts and big brains are present in many of the most bird-like maniraptorans, as are small teeth with reduced serrations. We suggest that these features – inherited by early birds and integral to the behaviours and ecological preferences they practised at the start of their history – were most likely a direct consequence of miniaturisation (Lee et al. 2014a).

Then there’s the fact that, as we get closer to birds in the phylogenetic tree, we see an increasingly elaborate plumage, a more bird-like system of body and hindlimb orientation linked to a shift in the centre of gravity, a stiffer, slimmer tail, and a number of behaviours that involve a degree of climbing (Birn-Jeffery et al. 2012) and gliding (Dyke et al. 2013). All of these features are linked in some way to the evolution of small size, but whether they evolved because size had decreased, or whether they drove the evolution of reduced size is just about impossible to say, especially when all were occurring in concert (Lee et al. 2014a).

The great stereotype about dinosaurs concerns their giant size. But the evolution of bigness was evidently not the case across the whole of the dinosaurian tree. In fact, it was the evolution of the very opposite trend that ultimately led dinosaurs in a wholly new direction, one that allowed their survival and, ultimately, their most successful evolutionary radiation.

Andrea’s take on our project can be seen here at his Theropoda blog.

For previous Tet Zoo articles on the evolution and diversity of theropods (including birds), see…

Refs – -

Benton, M. J. 2014 How birds became birds. Science 345, 508-509.

Birn-Jeffery, A. V., Miller, C. E., Naish, D., Rayfield, E. J., Hone, D. W. E. 2012. Pedal claw curvature in birds, lizards and Mesozoic dinosaurs – complicated categories and compensating for mass-specific and phylogenetic control. PLoS ONE 7(12): e50555. doi:10.1371/journal.pone.0050555

Dececchi, T. A. & Larsson, H. C. E. 2013. Body and limb size dissociation at the origin of birds: uncoupling allometric constraints across a macroevolutionary transition. Evolution 67, 2741-2752.

Dyke, G., de Kat, R., Palmer, C., van der Kindere, J., Naish, D. & Ganapathisubramani, B. 2013. Aerodynamic performance of the feathered dinosaur Microraptor and the evolution of feathered flight. Nature Communications 4, Article number: 2489 doi:10.1038/ncomms3489

Foth, C., Tischlinger, H. & Rauhut, O. W. M. 2014. New specimen of Archaeopteryx provides insights into the evolution of pennaceous feathers. Nature 511, 79-82.

Gatesy, S. M. 1990. Caudofemoral musculature and the evolution of theropod locomotion. Paleobiology 16, 170-186.

Holtz, T. R. 2000. Theropod paleobiology, more than just bird origins. Gaia 15, 1-3.

Hone, D. W. E., Keesey, T. M., Pisani, D. & Purvis, A. 2005. Macroevolutionary trends in the Dinosauria: Cope’s rule. Journal of Evolutionary Biology 18, 587-595.

Lee, M. S. Y., Cau, A., Naish, D. & Dyke, G. J. 2014a. Sustained miniaturization and anatomical innovation in the dinosaurian ancestors of birds. Science 345, 562-565.

- ., Cau, A., Naish, D. & Dyke, G. 2014b. Morphological clocks in paleontology, and a Mid-Cretaceous origin of crown Aves. Systematic Biology 63, 442-449.

Novas, F. E., Ezcurra. M. D., Agnolín, F. L., Pol, D. & Ortíz, R. 2012. New Patagonian Cretaceous theropod sheds light about the early radiation of Coelurosauria. Revista del Museo Argentino de Ciencias Naturales, nueva serie 14, 57-81.

Padian, K., de Ricqlès, A. J. & Horner, J. R. 2001. Dinosaurian growth rates and bird origins. Nature 412, 405-408.

Sereno, P. C. 1997. The origin and evolution of dinosaurs. Annual Review of Earth and Planetary Science 25, 435-489.

Sookias, R. B., Butler, R. J. & Benson, R. B. J. 2012. Rise of dinosaurs reveals major body size transitions are driven by passive processes of trait evolution. Proceedings of the Royal Society B 279, 2180-2187.

Turner, A. H., Pol, D., Clarke, J. A., Erickson, G. M. & Norell, M. A. 2007. A basal dromaeosaurid and size evolution preceding avian flight. Science 317, 1378-1381.

Darren Naish About the Author: Darren Naish is a science writer, technical editor and palaeozoologist (affiliated with the University of Southampton, UK). He mostly works on Cretaceous dinosaurs and pterosaurs but has an avid interest in all things tetrapod. His publications can be downloaded at darrennaish.wordpress.com. He has been blogging at Tetrapod Zoology since 2006. Check out the Tet Zoo podcast at tetzoo.com! Follow on Twitter @TetZoo.

The views expressed are those of the author and are not necessarily those of Scientific American.





Rights & Permissions

Comments 87 Comments

Add Comment
  1. 1. Heteromeles 2:25 pm 07/31/2014

    Great news, Darren?

    I wonder if this pattern is because smaller body size enables two things: shorter generation times, and more individuals in the same space. This is the elephant vs. mouse issue. Are there more mouse species because there’s a trend towards rodent miniaturization while elephants show the reverse? Or is it because you can put more mice than elephants in a square kilometer, they breed much faster, and the combination of numbers and fast generation time means that more mutations will crop up, along with smaller habitat requirements that allow tiny specialists to coexist than large specialists?

    This model would predict that, when correcting for phylogenetics, there will generally be more species on the small side of the clade. We can all think of some ceratopsians which would be great counter-examples, but it’s probably testable.

    Link to this
  2. 2. Trisdino 2:40 pm 07/31/2014

    This is, if not interesting, then at least rather nice news. I, like so many other paleo enthusiasts, was obviously already aware of this trend, but being able to see it more or less conclusively proven is very helpful when dealing with skeptics, and just in education in general.

    On a slightly different note, were there any other, non-maniraptoran groups with a tendency towards size reduction on a major scale? I can think of several small dinosaurs, but they represented offshoots of a larger group, mainly containing animals that averaged at a far larger size.

    Link to this
  3. 3. Halbred 3:11 pm 07/31/2014

    I had the same thought. Offhand, I can’t think of any. Seems like every major group of ornithischian shows an increase in body size (or remains the same size, like heterodontosaurs). Sauropods obviously just kept getting bigger and bigger, although dicraeosaurs do show a tendancy toward smaller sizes and stranger proportions (Brachytrachelopan says hello).

    Link to this
  4. 4. SeanMcCabe 4:50 pm 07/31/2014

    It’s hard to tell which branches on the tree are supposed to be which. I can make out Ceratosaurs (I think.), and that’s about it.

    Also, #2, are you Trisdino from the Speculative Evolution Forums?

    Anyway, why do think the ancestral condition was always becoming smaller yet the offshoots got so big on most occasions? It seems to reverse when they split from the main branch, except in Birds.

    Link to this
  5. 5. DavidMarjanovic 5:08 pm 07/31/2014

    It’s hard to tell which branches on the tree are supposed to be which.

    All is revealed in the supplementary information.

    when they split from the main branch

    That’s where the “tree” metaphor breaks down. A phylogenetic tree does not have a main branch.

    Link to this
  6. 6. Yodelling Cyclist 5:31 pm 07/31/2014

    Oh, Heteromeles beat me too it. just what thinking, and well phrased to boot.

    I was also just going observe (and I can’t be the first) how awesome pneumaticised bone is. High strength, low mass allows huge sizes without excessive collection of skeletal elements (e.g. Calcium) in rapid times: hence sauropods and tyrannosaurs, also permits high strength to weight ratios and so relatively easy access to flapping flight: hence birds and pterosaurs. It reminds me of my materials science lectures: modern (post 1800) high strength metals permit new structures.

    Link to this
  7. 7. Trisdino 5:33 pm 07/31/2014

    To Sean, I doubt there are any other people with my rather odd username out there, so yes, that would probably be me, though at this point, I will not be spending anymore time there.

    Anyway, on the subject of lineages heading towards major decreases in size, again, I just can not think of any. There are some with minor tendencies towards smaller size, but they appear to either be, or could be explained as, temporary stages in evolution in response to present conditions, instead of any larger scale evolutionary theme. So really, can anybody else think of a lineage which was actively decreasing in size on a major scale?

    Link to this
  8. 8. Yodelling Cyclist 5:33 pm 07/31/2014

    Edit: “Just what I was thinking”.

    Plus birds are always highly speciose. Could the feathers act as easily modified species identifiers permitting rapid divergence?

    Link to this
  9. 9. Tayo Bethel 5:48 pm 07/31/2014

    Perhaps competition for food was one of the many factors that drove theropod miniaturization, as well as omnivory and arboreal behavior. Interguild competition is well established as a driving force in evolution. Some species become smaller, others larger, and behavior changes in response. The Tyranosauridae is an example of a family that might best illustrate this point. Starting out as small predators, they evolve toward ever increasing body size, eventually becoming the apex predators wherever they occurred. One can well imagine a highly competitive predatory guild where some of the members became smaller, broadened their dietary preferences and even in some cases exited the guild altogether by evolving dietary preferences of their own. Arboreality would be selected for as a very handy escape behavior. :) This is already overlong … I’ll stop here.

    Link to this
  10. 10. SeanMcCabe 6:28 pm 07/31/2014

    Just making sure. Thought to be fair, you were right on in that last argument. Conservation is important.

    I generally define “main branch” as more speciose of the two lines. So, for example, Tyrannosauroidea would be a side branch to (most) other tyrannoraptorans. But you get my point anyway. The ancestral line from relation to birds is gradually getting smaller, and the side branches from the bird-line perspective generally get larger. Why is this?

    Link to this
  11. 11. Heteromeles 7:20 pm 07/31/2014

    So far as animal clades showing decreases over time, I’d look at insects like beetles as another example. Scorpions might work too, if you stretch the time scale back far enough.

    I’d also point out that plants kind of show this pattern: AFAIK, annuals generally evolve from perennial ancestors, for example, and new flowers are often smaller than old ones.

    Link to this
  12. 12. Jerzy v. 3.0. 9:44 pm 07/31/2014

    Congrats for the article!

    I guess tetrapods as a whole are shrinking. Ichtyostega et al were huge, 2 m animals. And insectivorous lineages are usually becoming on average smaller and more speciose.

    BTW, here comes my favorite wacky theory that the first birds went through the stage where juveniles used powered flight but adults were flightless. One reason may be that juveniles were arboreal to escape predators and avoid competition with adults. There is a number of birds where juveniles fly better than adults (streamer ducks, snowcocks, megapodes, giant coot etc).

    Link to this
  13. 13. Cahokia 11:04 pm 07/31/2014

    Theropod miniaturization seems reminiscent of the trend towards smaller body sizes in advanced non-mammalian synapsids.

    Link to this
  14. 14. SeanMcCabe 11:05 pm 07/31/2014

    I’ve heard it suggested before that dromaeosaur juveniles would’ve been able to fly, but adults couldn’t. Not from a 100% credible source, but not an overly bad one either.

    Link to this
  15. 15. DavidMarjanovic 7:24 am 08/1/2014

    So really, can anybody else think of a lineage which was actively decreasing in size on a major scale?

    This hasn’t been tested much, due in part to the long-held assumption (“Cope’s rule”, actually by Depéret) that everything always gets bigger by default.

    That said, it’s not like theropods as a whole got smaller and smaller. That’s only if you pick one particular branch. Pick another, and you’ll find a size decrease followed by a size increase, or even more complex patterns.

    I generally define “main branch” as more speciose of the two lines.

    The problem with this is that you have to wait a long time till one of them becomes more speciose; they both start out as a single (Hennigian) species. And they can still switch places several times after that.

    Link to this
  16. 16. Tayo Bethel 9:02 am 08/1/2014

    @powered flight in dromaeosaurids:

    It would be interesting to test whether dromaeosaurids of any kind were capable of powered flight.Do any dromaeosaurids display any of the characteristics associated with powered flight (enlarged, keeled sternum, fused wishbone, etc.? Perhaps juvenile dromaeosaurids of large species were more arboreal than adults simply because they were smaller. Many dromaeosaurids retain large, curved, laterally compressed claws similar to those of felids, and even large felids can climb,if not as well as smaller species. Of course, then we come to the question of why juveniles would have been more arboreal at all, which is a topic which has been argued since the discovery of dromaeosaurids. Dromaeosauridsociality and whatnot. And the paper which made comparisons between dromaeosauridsand Komodo dragons. I personally don’t take the speculations in this paper too seriously. Dromaeosaurids belong to the Archosauria,the living examples of which at least display well-developed parental behavior and social systems. And even some lizards display parental behavior and complex family life.

    …This was way off topic. Sorry:) I do tend to ramble.

    Link to this
  17. 17. Heteromeles 10:02 am 08/1/2014

    Another possibility: energy-hungry organs. This is the idea that brains take a lot of energy to run. If you assume, following Carl Sagan (that, cough, noted paleontologist), that relative brain size is increasing over time, and that nervous tissue is energetically more expensive than muscles and bones, then the logical tradeoff is that body size will decrease as relative brain size increases.

    Obviously, there are niches in which large body size is a benefit (rumination comes to mind) so this isn’t a universal trend. Still, intelligence and culture only work with sufficient brains (sorry Matt, I don’t think a civilization of sauropods would work). As adaptable predators become more common, it might drive more animals to become smarter in something like a red queen race.

    This is just a thought. I’m also cogitating about how one would go about disproving these ideas, and I don’t think that’s impossible either.

    Link to this
  18. 18. JoseD 11:06 am 08/1/2014

    SeanMcCabe: “I’ve heard it suggested before that dromaeosaur juveniles would’ve been able to fly, but adults couldn’t. Not from a 100% credible source, but not an overly bad one either.”

    I’ve only heard that suggestion from 1 source & it’s pretty bad (See “Bad”: http://jd-man.deviantart.com/journal/SD-Good-semi-good-and-bad-dino-sources-1-351589315 ).

    Link to this
  19. 19. SeanMcCabe 11:08 am 08/1/2014

    I was referring to this, and the three posts linked in the first paragraph. Might be outdated or just flat out wrong, but interesting none the less:
    http://gwawinapterus.wordpress.com/2011/10/12/velociraptors-on-air-analysis-of-deinonychosaur-clades/

    Link to this
  20. 20. Tayo Bethel 11:29 am 08/1/2014

    SeanMcKay:
    Probably flat-out wrong. Sure, juvenile theropods sometimes filled different ecological niches and were differently proportioned than adults, but powered flight requires radical changes in the body to be achieved–changes that wouldn’t simply fade away with maturity. So arboreal juvenile dromaeosaurids? Possible. Flighted juvenile dromaeosaurids?Very, very unlikely.

    Link to this
  21. 21. Tayo Bethel 11:36 am 08/1/2014

    Heteromeles:
    Has it been remarked on, or is this just my intuition that at least among terrestrial tetrapods, the most intelligent (at least as we social mammals recognize it) forms are all generalized, social omnivores? Delphinids are an obvious exception to this rule, being aquatic and piscivorous, but otherwise display the same traits. Have delphinids become smaller and perhaps more intelligent overtime?

    Link to this
  22. 22. SeanMcCabe 12:00 pm 08/1/2014

    Does bring up a few interesting points thought. AFAIK he still stands by what is said in those posts.

    Link to this
  23. 23. jtdwyer 1:10 pm 08/1/2014

    Offhand, might there have been some correspondence between atmospheric oxygen levels and size?

    Link to this
  24. 24. JoseD 1:52 pm 08/1/2014

    SeanMcCabe: “I was referring to this, and the three posts linked in the first paragraph. Might be outdated or just flat out wrong, but interesting none the less:”

    Definitely the same source & definitely bad for reasons discussed in my previous comment’s link.

    SeanMcCabe: “Does bring up a few interesting points thought. AFAIK he still stands by what is said in those posts.”

    Interesting, yes, but as Tayo Bethel said, just plain wrong on many levels.

    Link to this
  25. 25. Gigantala 2:11 pm 08/1/2014

    Yes, I disagree with my own early assessments on the matter.

    Link to this
  26. 26. SeanMcCabe 3:49 pm 08/1/2014

    As for the reasons given that it’s a bad source, 1 is situational. Saying that someone is not at all trustworthy because they don’t have a degree, is TBH, stupid, 2 has nothing to do with his credibility at palaeontology, and 3 is a good point, thought not having sources should raise flags rather then prove it’s no good. Not saying it’s a good source, but that it’s being said to be bad for the wrong reasons. He’s just flat out incorrect, if he’s wrong, which he probably is.

    Link to this
  27. 27. naishd 4:27 pm 08/1/2014

    Sean McCabe and others: for the benefit of those who haven’t been following the details of your discussion (cough)… what are you talking about? Are you discussing claims discussed on another blog?

    Link to this
  28. 28. Gigantala 4:34 pm 08/1/2014

    Eh, my sources have been generally fail proof. The problem is that I don’t bother to quote.

    Link to this
  29. 29. Gigantala 4:36 pm 08/1/2014

    @naishd: See comment #19

    Link to this
  30. 30. naishd 4:39 pm 08/1/2014

    Ok, you’re talking about stuff at the gwawinapterus blog? Ok, gotcha, thanks. I’m just wary that it sounds – in some of the comments above – as if you’re commenting on stuff here at Tet Zoo!

    Link to this
  31. 31. SeanMcCabe 4:55 pm 08/1/2014

    Well, either way, there are far certainly worse blogs when it comes to palaeontology. And not many better when it comes to Speculative Evolution. His article on gliding not having to do with the evolution of flight has some merit to it as well.

    Link to this
  32. 32. JoseD 6:25 pm 08/1/2014

    I don’t want to get too far off topic, so I’m just going to say the following & leave it at that.

    SeanMcCabe: “Saying that someone is not at all trustworthy because they don’t have a degree, is TBH, stupid,”

    Which is why I never said that. I specifically said, “They’re non-experts who act like they’re experts” (which is bad for reasons discussed in this link: http://reptilis.net/2008/09/14/jfc-lockjaw/ ).

    “2 has nothing to do with his credibility at paleontology,”

    Actually, based on what I’ve read, it’s very relevant ( http://jerseyboyshuntdinosaurs.blogspot.com/2014/07/idols-and-idolatry.html ).

    “and 3 is a good point, thought not having sources should raise flags rather then prove it’s no good.”

    It proves a source is bad if that source outright refuses to source his work & pretends to be right regardless of the evidence (which was often the case at least b-4 he was perma-banned from DA/JPLegacy/etc).

    “Well, either way, there are far certainly worse blogs when it comes to paleontology.”

    While technically not as bad as “Dr. Pterosaur” & the like (I.e. He’s not denying fundamental concepts like the dino-bird relationship), he’s bad in a similar way b/c he does similar things (E.g. Pretending to be right regardless of the evidence). That’s my point.

    Link to this
  33. 33. assumingdinosaur 7:35 pm 08/1/2014

    What does this mean for the assortment of purported giant basal coelurosaurs like Lourinhanosaurus and Chilantaisaurus?

    re:bad sources: Dr. Pterosaur/Socrates is so horrifically, fundamentally wrong that the only appropriate response is to totally disregard him.

    Link to this
  34. 34. Allen Hazen 9:10 pm 08/1/2014

    Tayo Bethel (#21)–
    There are enough species of delphinium, with enough disparity in body size (a large Orca has what ? ???30 times??? the mass of a Common dolphin (guessing both masses)) that this might make an interesting study… IF we had any idea how the different species compare in intelligence. Most of the hee-haw and hype about dolphin intelligence in the 1960s and 1970s came from studies of (count them!) ONE species,”the” (the species has since been subdivided) Bottlenose Dolphin. Chosen, I suspect, because it often frequents near-shore environments and so is easier to capture (and just possibly adapts more readily to life in confined quarters): I think we have basically no useful information about how intelligence might vary across the Delphinidae.

    Link to this
  35. 35. Allen Hazen 9:12 pm 08/1/2014

    (And whatever I typed instead of “delphinidae” in the first line of the preceding, I’m pretty sure it wasn’t “delphinium”.)

    Link to this
  36. 36. Therizinosaurus 12:59 am 08/2/2014

    Looking at figure S8 based on Andrea’s matrix, doesn’t it seem that the apparent size reduction within Coelurosauria until you get to Pygostylia is due to using immature specimens? The first decrease is at the base of Coelurosauria, which everyone agrees with.

    But the second is between Zuolong and Tugulusaurus, and who knows how old the Tugulusaurus holotype was? We could ignore Tugulusaurus and still have small Coelurus in the next node up, but the latter has unfused sacral centra and open dorsal and proximal caudal neurocentral sutures (Carpenter et al., 2005). Thus is was young, so can’t be used as evidence of basal tyrannosauroids being so small.

    The third red arrowed size decrease is between Sinocalliopteryx and compsognathids (Compsognathus, Huaxiagnathus, Sinosauropteryx). Yet the latter are not adults, as e.g. the largest Compsognathus has some completely open cervical neurocentral sutures and unfused sacral centra (Peyer, 2006), Huaxiagnathus has completely separated proximal caudal neurocentral sutures (Hwang et al., 2004), and Sinosauropteryx’s largest specimen has unfused sacral ribs and neurocentral sutures on its proximal caudal centra (Currie and Chen, 2001).

    The next arrow leads to Pennaraptora, but this is based on the small size of Protarchaeopteryx and Caudipteryx. All published specimens of the latter have open neurocentral sutures, and unfused sacral centra and ribs, and both genera have rounded sternal plates with no processes unlike other oviraptorosaurs, paravians or Pelecanimimus. Once these are removed, it’s small Avimimus vs. large caenagnathoids.

    The next arrow is the greatest change, but is based purely on the young Epidexipteryx holotype.

    The final arrow of decreased size before short-tailed birds is due to Aurornis and Anchiornis. At this point with the above examples eliminated, any decrease in size in these two nodes is not a trend, since the next five nodes are all larger in size until you get to small pygostylians.

    In the TWG analysis (Analysis 2, fig. S9), the two size decreases between basal coelurosaurs and birds are based on 1. Caudipteryx again; 2. small Rahonavis being weirdly the most basal deinonychosaur. Note the latter is a result not found in any other analysis I know of, no doubt due to this being from Choiniere’s Haplocheirus matrix with so many uncoded characters that it featured in my and David Marjanovic’s manuscript critisizing this flaw (note also the paraphyletic carnosaurs, unique dromaeosaurid phylogeny, etc.). Without young Caudipteryx or flukily basal Rahonavis, the trend disappears.

    Would have been cool to see less complete size outliers with estimated femoral lengths (e.g. Deinocheirus, Sinotyrannus, Balaur). Also I think taxonomic inclusion is highly important here. Using ornithomimosaurs as an example, you have mid-sized Harpymimus and smaller but seemingly younger Shenzhousaurus. But besides huge Deinocheirus, we also have large Beishanlong and tiny Hexing. So what would the basal size actually be?

    Sorry for the largely negative response. I have no issue with the evolutionary rate aspect, which is Andrea’s matrix is certainly the best for.

    Link to this
  37. 37. naishd 6:55 am 08/2/2014

    Hi Mickey — you raise lots of interesting points that require response. I don’t time to do that now but I do want to make two quick points. One is that an analysis like this obviously involves the ‘averaging out’ of a large quantity of size data: we didn’t include all taxa, of course, but we included enough such that any generalisation in part of the tree can be seen to be based on a reasonable set of taxa. In the case of Epidexipteryx – to take one example – even if it were removed, it’s surrounded in the phylogeny by small-bodied taxa, meaning that the reported size reduction compared to earlier maniraptorans is still there. As for the inclusion of taxa where the specimens don’t seem to have reached full skeletal maturity – those compsognathids, Caudipteryx and so on – I would agree that they haven’t fully fused up their neurocentral sutures etc., and also that we included them as if they’re representative of the adult condition. However, for several reasons I think we can accept their sizes as being representative of adult or near-adult conditions, in which case their use in our analysis (and those of others) is acceptable.

    Finally (for now) note that – in the dataruns where there are a few abrupt changes (this relates to your comment about the TWiG dataset), we were careful not to claim that a significant continuous trend was present: rather, those reconstructions (which used parsimony methods alone, not Bayesian ones) were a ‘reality check’. If the trees and ancestral sizes from the Bayesian and parsimony reconstructions were very different, we’d have cause to worry – but they were broadly similar.

    Link to this
  38. 38. naishd 7:03 am 08/2/2014

    One more thing in connection with the “but those might be juveniles” comment, remember that it’s misleading to imply that a specimen isn’t representative of (approximate) adult size just because it has unfused neurocentral sutures in parts of its column. Many adult archosaurs – sexually mature, at ‘normal’ adult size etc – still have unfused sutures in parts of their column; it doesn’t mean that they can be dismissed as juveniles.

    Link to this
  39. 39. DavidMarjanovic 8:22 am 08/2/2014

    Do any dromaeosaurids display any of the characteristics associated with powered flight (enlarged, keeled sternum, fused wishbone, etc.?

    Forget about the wishbone. :-) All theropods have a wishbone, and that’s been known since the turn of the millennium. Coelophysis has it.

    Dromaeosaurids do have rather large sterna, like oviraptorosaurs.

    If you assume, following Carl Sagan (that, cough, noted paleontologist), that relative brain size is increasing over time

    This is Marsh’s so-called law, which hasn’t been investigated in many decades. It’s been forgotten rather than actually disproved, AFAIK.

    Choiniere’s Haplocheirus matrix with so many uncoded characters that it featured in my and David Marjanovic’s manuscript critisizing this flaw

    No need for the past tense. I should have time to work on that long-delayed manuscript in just a few weeks; the work on it overlaps with the work on the next big manuscript anyway. :-)

    In case anyone’s wondering, the manuscript is about the widespread problems of 1) people leaving out known data in their data matrices, keeping it as “unknown”, no doubt because they think it wouldn’t make a difference, so coding it would be a waste of time – but it does make a difference to the resulting tree, as we show for several cases; 2) reviewers not reviewing supplementary data.

    remember that it’s misleading to imply that a specimen isn’t representative of (approximate) adult size just because it has unfused neurocentral sutures in parts of its column. Many adult archosaurs – sexually mature, at ‘normal’ adult size etc – still have unfused sutures in parts of their column; it doesn’t mean that they can be dismissed as juveniles.

    Well, for a study like this, it’s obviously best to compare specimens at comparable ontogenetic stages. The easiest one to recognize is skeletal maturity, after which not much growth happens anymore. That criterion would leave you with a rather small dataset, perhaps too small for your tastes. Another candidate would be sexual maturity, after which quite some growth can still happen (though it not necessarily does); how easy is that to recognize in fossils without looking at the bones from the inside?

    However, for several reasons I think we can accept their sizes as being representative of adult or near-adult conditions

    Are the reasons explained in the paper?

    Link to this
  40. 40. AndreaCau 9:18 am 08/2/2014

    Mickey,
    Note that S8 is the parsimony analysis, where size is considered as a discrete character and thus eventual variation among a single state is not detected. In fact, the rough size states in that parsimony analysis cannot detect any size trend among non-coelurosaurians, which instead results using continuous values. Our result is based mainly on the Bayesian analysis integrating morphology and time, and using the femur data as a continuous value, not from a small series of discrete states in the parsimony test.
    I think that the term “immature specimen” is in pary not appropriate here. How could we compare the different ontogenetic stages of most of the theropod specimens known? Everybody usually uses terms as “juvenile”, “subadult” and “adult” and more or less implicitly assumes that those terms refer to (1) universally fixed discrete states that (2) could be compared among different individuals of different taxa in the same way as we compare – for example – the number of premaxillary teeth and conclude that Allosaurus is intermediate between Torvosaurus and Pelecanimimus and is at the same “stage” as Neovenator. Probably, both assumptions are false. Is the holotype of Tyrannosaurus the best representation of adult Tyrannosaurus? Or is it Sue? Both and none of them, since the “adult size of a taxon” is probably impossible to define in an accurate way if the taxon is based on just a single specimen (or even half a dozen of specimens).
    You mentioned open neurocentral suture as evidence of immature stage, but we also know that neurocentral fusion is not indicative of arrested growth, and that sexual maturity (a biologically more important feature than skeletal fusion) is not linked to neurocentral suture. Thus, we did not exclude specimens based on unfused neurocentral suture alone. We excluded values when several lines of evidence indicated that the largest known femur for that taxon was not indicative of a post-hatchling stage of development. In an ideal world, we would have used just taxa with a robust set of istological samples from a large known population… something actually (and probably, never) available for fossil theropods.
    Furthermore, given the mostly random process of fossilisation, there is no reason to assume that “juvenile” taxa are sampled only among Coelurosauria or that our analysis is biased by such sampling. For example, the Allosaurus femur we used is probably not from the largest known Allosaurus individual, but is the largest described in literature. Is that particular Allosaurus at the SAME ontogenetic stage as the Tyrannosaurus femur used? or was it at the same ontogenetic stage as that of Coelurus type? Probably, that question is just meaningless. Since a large set of random and indeterminable factors determines the fossil record, it’s impossible to extract “exact” values from the specimens that represent “THE adult size of that taxon” (whatever it means).
    Thus, I find no problems in using taxa described in literature as “subadult and/or adult”, as they provide a more or less random sample of size ranges among theropods. Also, note that we used the Log10 of femur length, not their linear size, to avoid any allometric effect that could biase our analyses. If specimens of various ontogenetic stages are randomly sampled among the whole ingroup (as I think), or mostly immature individuals are used among Coelurosauria (as you argue), why a consistent miniaturisation trend (that is, a phylogenetic signal in body size variation) emerges in the avian stem alone and not in other parts of Coelurosauria?
    We knew that the exact age of Tugulusaurus is unknown, in fact, as for other uncertain ages, we followed a strategy using the mean value of the age range for a taxon. I acknowledge that this is a provisional solution, but it’s similar to other age calibrations present in literature: future implementations of the analysis may incorporate age ranges among the data, testing the effect of using alternative age values in topology calculation.

    I agree that a larger taxon sample would be better (and a way for including those taxa missing femur data)… and I’m working on that direction ;-)

    Thanks for considering my dataset at least the best available for evolutionary rate analysis.

    Link to this
  41. 41. Jenny Islander 12:25 pm 08/2/2014

    Is it plausible that there could also be a connection to changing flora? Like this:

    New species mix with, perhaps, more usable nutrients per unit of plant mass;

    Less survival value in being big enough to haul around a big gut;

    Tendency for smaller herbivores to be more common relative to big ones than previously;

    Less survival value in being specialized for bringing down big prey species, more survival value in ability to live on a range of prey sizes (think tiger vs. leopard);

    Tendency for smaller carnivores to be more common relative to big ones than previously.

    Total amateur, grain of salt, etc.

    Link to this
  42. 42. Tayo Bethel 1:50 pm 08/2/2014

    @David:

    I don’t doubt that dromaeosaurids had a wishbone,since the fact that all theropods had one is pretty well known by now. Living theropods with reduced flight capabilities are known to have incompletely fused clavicles, though this probably doesn’t apply to dromaeosaurids, whose forelimbs were far from degenerate.

    BTW Comments on delphinids: Delphinium is a flowering plant :) and should probably never be mentioned here again unless it’s the favorite food of an amazing tetrapod that considers it a rare delicacy :) You are right about the number of species of delphinids for which intelligence has been testedl I think a numberof studies have also been done on orcas, though. THey seem every bit as intelligent as their bottlenosed cousins. What would be an equally interesting, if logistically challenging study would be to conduct tests for intelligence across the whole of Cetacea.

    Link to this
  43. 43. Heteromeles 3:42 pm 08/2/2014

    @David 39: If you assume, following Carl Sagan (that, cough, noted paleontologist), that relative brain size is increasing over time

    This is Marsh’s so-called law, which hasn’t been investigated in many decades. It’s been forgotten rather than actually disproved, AFAIK.

    Thanks David. That helps. Interestingly, Kruzweil’s Singularity (the idea that we’re inevitably charging forward into an exponential advance in intelligence ending in superhuman AIs) seems to have something like Marsh’s Law embedded in it, rightly or wrongly. Therefore, I’d suggest that Marsh’s Law would make an extremely good punching bag, erm, topic for phylogenetic studies. Is there good evidence that encephalization quotients increase in lineages over time? Or is this yet another example of the selective quoting of evidence that Kurzweil and others have been criticized for? The nice thing is that if you do the research, you get a paper out of it, whether you prove it or disprove it, because either answer makes a good story.

    Link to this
  44. 44. LeeB 1 6:29 pm 08/2/2014

    “Intelligence” in Steno bredanensis ( the rough toothed dolphin) has also been remarked on; at least two specimens have been trained to get a reward only when they do a new trick that they invent themselves.
    The first one that was trained thus came up with so many new tricks that they trained a second one just to see if the first one was some kind of dolphin genius; apparently the second one was similarly innovative.

    I don’t know the literature reference for this; I saw this on a television program on whales and dolphins years ago but the Steno concerned were being kept in an aquarium in Hawaii.

    LeeB.

    Link to this
  45. 45. LeeB 1 7:47 pm 08/2/2014

    Hopefully this works; there is a paper on this here: http://escholarship.org/uc/item/9cs2q3nr

    LeeB.

    Link to this
  46. 46. ectodysplasin 7:51 pm 08/2/2014

    @assumingdinosaur

    re:bad sources: Dr. Pterosaur/Socrates is so horrifically, fundamentally wrong that the only appropriate response is to totally disregard him.

    Socrates is a well-known and documented creationist troll. He’s also something of a homophobe, an Islamophobe, and an antisemite. So it’s not just that he’s wrong.

    Link to this
  47. 47. SeanMcCabe 9:26 pm 08/2/2014

    Mortimor, Naish, Cau: Wouldn’t it have a very large effect on the analysis to not include extremes like Deinocherius and Hexing? Not that I can at all tell which taxa are included, so sorry if the extremes are coded from here.

    Jenny:If I understand what you’re saying here, it’s an interesting point. But, going by this, shouldn’t the Sauropods and Ornithichians, as well as other theropod lineages, also be getting smaller? Tyrannosaurs, Ceratopsids, Hadrosaurs, Sauropods, Ankylosaurs, etc all got significantly larger as the Bird-line was getting smaller. But indeed some environmental factor could indeed relate to this.

    It could be a genetic change. Or perhaps the bird line ancestral condition was mammal/small prey hunting, and all the other, gigantic lines adapted from this.

    And I can relate to the Amateur, grain of salt situation….

    Link to this
  48. 48. CS Shelton 2:15 am 08/3/2014

    Forgive my foolery on anything I mess up. BFA, y’all.

    The Big DM -
    “The problem with this is that you have to wait a long time till one of them becomes more speciose; they both start out as a single (Hennigian) species. And they can still switch places several times after that.”

    This makes me think of artiodactyls and perissodactyls. Both had turns where they were the majority of hoofed animals. Did they alternate at any point, or was it perissodactyls up to a point and then successive radiations of artiodactyls after that?

    Heteromeles –
    “I’m also cogitating about how one would go about disproving these ideas, and I don’t think that’s impossible either.”

    That’s the good scientists in action. Think of a cool idea and immediately think of how it could be wrong. Rather the opposite of certain fellows that broke all comment records forever recently…

    JT @23 -
    “Offhand, might there have been some correspondence between atmospheric oxygen levels and size?”

    I’ve heard that too and wonder if any of the smarties here can drop some science on us about that. Another possibility is that mammals just aren’t built to achieve larger dino sizes and the fact they beat birds* to dominating the large-bodied niches prevents giants like that from happening again.

    Then again, the largest land mammals before we jacked up the program were pretty dang hefty. Could they have gotten even bigger without super-predatory brain-monkeys in the way?

    *This is assuming birds would be the primary competitors. The “Age of Turtles” should have given us sauropod sized beasties, yo. How cool would that have been?

    Link to this
  49. 49. Therizinosaurus 2:36 am 08/3/2014

    Darren wrote- ” In the case of Epidexipteryx – to take one example – even if it were removed, it’s surrounded in the phylogeny by small-bodied taxa, meaning that the reported size reduction compared to earlier maniraptorans is still there.”

    Ah, not so. The surrounding taxa (jeholornithids and Xiaotingia/Archaeopteryx in Bayesian; deinonychosaurs and oviraptorosaurs in MP) are larger. So in fig. S8 for instance, the main spine of the tree wouldn’t turn from green to red until Aurornis’ level if Epidexipteryx were removed.

    “As for the inclusion of taxa where the specimens don’t seem to have reached full skeletal maturity – those compsognathids, Caudipteryx and so on – … However, for several reasons I think we can accept their sizes as being representative of adult or near-adult conditions, in which case their use in our analysis (and those of others) is acceptable.”

    I’d be interested in those reasons. To continue the Caudipteryx example, you treated its sister group Microvenator as juvenile, but Caudipteryx shares all of the same open sutures, is only ~17% larger, and neither has been ontogenetically asssessed via histology. Yet one is considered representative of at least near-adult size and the other is not.

    “Many adult archosaurs – sexually mature, at ‘normal’ adult size etc – still have unfused sutures in parts of their column; it doesn’t mean that they can be dismissed as juveniles.”

    Yet specimens with fused sutures are more likely to be larger and so presumably older (Irmis, 2007). So it seems like a good rule to follow, even if there will be exceptions. This is especially true for coelurosaurs where no grade/clade seems to be paedomorphic for suture fusion (e.g. Ornitholestes’ holotype shows compsognathid-grade taxa could have neurocentral and sacral fusion).

    Link to this
  50. 50. irenedelse 2:53 am 08/3/2014

    Well, I’m late to the party (hooray for holidays), but first off, congratulations on your high-profile etc. paper! \o/
    Does the trend in the Theropod bird lineage need explaining or is it just what you see when a lineage “finds” a new niche to exploit? Here, we have Theropods getting into niches available only with smaller size…

    Link to this
  51. 51. Therizinosaurus 4:02 am 08/3/2014

    Before I address Andrea’s comment, I think I’ve found a fatal flaw in the basic structure of the analysis. I’m no mathematician or programmer, but when you have a group at the tip of the tree that’s tiny (ornithothoracines), won’t every node leading to that be on average smaller than the more basal node when analyzed this way? At least until you get far enough away from that tiny group (in Bayesian Dataset 2 that seems to happen at Tetanurae).

    I can’t read the size values for Dataset 1 due to the low resolution in the pdf, but using figure S5 for the Baysian results of Dataset 2… compsognathids have a basal size of 2.213 then the next node up is alvarezsaurs with a LARGER basal size of 2.306 then next is the almost identical Enigmosauria/Oviraptoriformes with 2.3. So really the trend is compsognathids being smaller than the 2.316 Ornitholestes outgroup, then alvarezsaurs being larger, then enigmosaurs being about the same. Yet the size values on the spine of the tree are 2.371 at the split of Ornitholestes and more derived taxa, to 2.325 when compsognathids split, to 2.299 when alvarezsaurs split, to 2.237 when enigmosaurs split. This supposedly indicates a significant size change downward, but as I explained above that’s not what each individual clade shows. So I wonder if the tiny birds make all of the birdward clades smaller ancestrally in this program?

    In a similar way, in that figure ornithomimosaurs are ancestrally a huge 2.45 despite their two most basal taxa being 2.3222 Coelurus and 2.281 Shenzhousaurus. But ornithomimids themselves are ~2.5-2.8, so these high values drag up the supposed ancestral sizes of ornithomimosaurs to be larger than any of the basal members.

    Or at basal Tetanurae where Piatnitzkysaurus is smaller than the four nodes birdward of it, but your results still have the Piatnitzkysaurus to orionidan move as being a decrease in size. Similarly, megalosauroids are ancestrally smaller than the three nodes birdward of them, but the tree’s labels would indicate a size decrease happened leading from megalosauroids to those larger taxa.

    Link to this
  52. 52. Therizinosaurus 5:15 am 08/3/2014

    Andrea wrote- “Our result is based mainly on the Bayesian analysis integrating morphology and time, and using the femur data as a continuous value…”

    That Bayesian tree is figure S1, which Science screwed up by giving it horrible resolution. Another thing I notice is that taxa which weren’t used due to lacking femora or being young are colored the same as the assumed ancestral state. This is quite misleading, since e.g. Balaur is huge for a tailed bird but is colored like a bird half its size, Patagopteryx is big for an ornithuromorph but colored as if small, ditto for patagonykines. Figure 1 has the same issue. It doesn’t affect the analysis of course, but does give the illusion of a less noisy signal.

    But as for the tree itself, most of the points I used for the MP tree work here as well (e.g. Caudipteryx and Protarchaeopteryx). There are some differences, but the same issues apply. For instance, troodontids plus birdier things are smaller than dromaeosaurids due to basal troodontid Jinfengopteryx being small, but who knows what age it is? Perhaps more importantly, any differences between the Bayesian and MP trees represent at best relationships that aren’t established. So only relationships shared by both trees should be used to defend your hypothesis. Considering the apparent flaw of the analysis itself I describe above, any nitpicks of the tree now seem less useful though.

    You spend a lot of time on the difficulties and complexities of determining ontogenetic stages of theropods, but you did that yourself in this paper. Why did you not use Tanycolagreus but did use Coelurus? Or why exclude Garudimimus but not Shenzhousaurus? So even you admit there are characters which indicate specimens are probably younger or older, which can be used to put individuals of any group on a continuum of young to old compared to each other. We don’t need to know exactly which stage Liaoxiornis and Stygivenator are compared to each other to know both are too young to use as adult sizes.

    “If specimens of various ontogenetic stages are randomly sampled among the whole ingroup (as I think), or mostly immature individuals are used among Coelurosauria (as you argue), why a consistent miniaturisation trend (that is, a phylogenetic signal in body size variation) emerges in the avian stem alone and not in other parts of Coelurosauria?”

    I never argued the juveniles used were mostly coelurosaurs. I agree there is a shrinking near the base of Coelurosauria, and note that e.g. Eustreptospondylus and Sinraptor dongi are immature but treated by you as adults. But based on the apparent flaw I commented on above, I’d say the stronger trend emerges on the avian stem due to the small size of ornithothoracines getting irrationally dragged stemward but with decreasing influence as you get closer to the stem.

    “We knew that the exact age of Tugulusaurus is unknown, in fact, as for other uncertain ages”

    I meant ontogenetic age, not temporal/stratigraphic age. I have no problem with your method to estimate stratigraphic age.

    Link to this
  53. 53. Therizinosaurus 5:26 am 08/3/2014

    Finally, and most easily, Tayo Bithel wrote- “Do any dromaeosaurids display any of the characteristics associated with powered flight (enlarged, keeled sternum, fused wishbone, etc.?”

    Enlarged sterna are known from Pelecanimimus to birds, including Bambiraptor, Sinornithosaurus, Microraptor, Tianyuraptor, Changyuraptor, Saurornitholestes or Dromaeosaurus, Velociraptor, Tsaagan and probably Deinonychus based on all the ossified sternal ribs. A keel is absent from all known dromaeosaurid sterna. A fused furcula is present in most theropods including Bambiraptor, Sinornithosaurus, Microraptor, Tianyuraptor, Changyuraptor and Velociraptor.

    Link to this
  54. 54. DavidMarjanovic 6:07 am 08/3/2014

    sexual maturity (a biologically more important feature than skeletal fusion)

    Whether it’s biologically important is not important :-) for your study; what’s important is that the specimens are all at comparable stages, whichever those actually are and whatever that actually means.

    However, log-transforming the data must of course have helped: it decreases the differences between large specimens.

    I’d suggest that Marsh’s Law would make an extremely good punching bag, erm, topic for phylogenetic studies.

    Hee. :-)

    Is there good evidence that encephalization quotients increase in lineages over time?

    Impressionistically, Eocene mammals apparently have consistently smaller brains than their extant relatives; that’s where Marsh got the idea. But that doesn’t necessarily mean there’s a driven trend going on; and that has not, AFAIK, been tested.

    Link to this
  55. 55. naishd 7:32 am 08/3/2014

    Thanks for continuing comments. Some brief responses to Mickey (I’ll try and add more detail later on)..

    (1) Yes, the small size of avialans will cause nearby nodes to be reconstructed as small. But EVERY clade and taxon in the analysis exerts a similar influence – so the end results are driven by the entirety of the data. As explained/pointed out above, the size of reconstructed concestors is estimated based on all the taxa that surround it in the cladogram – the sizes of the ancestral neotheropod, ancestral tetanuran, and ancestral coelurosaur, for example, really are getting smaller whether birds are present in the cladogram or not.
    (2) The phylogenetic positions of taxa were as recovered via our running of Andrea’s dataset: we didn’t move any taxa in anticipation of obtaining a trend. Nor did we choose taxa to score for femur length based on any anticipated results. The results fell out once everything was analysed – there was no inherent bias to get a big picture trend here.
    (3) I think we can all agree that some specimens are less ideal than would be perfect in view of ontogenetic stage and so on. But should we have excluded all the specimens concerned? As David said above, that might have left us with far fewer specimens/taxa that we might like – as usual, we can only work with what we have.

    Link to this
  56. 56. SeanMcCabe 7:41 am 08/3/2014

    Do the amateur comments not matter at all, or what?

    Link to this
  57. 57. naishd 7:42 am 08/3/2014

    Sean — I have no idea what you’re getting at, please explain.

    Link to this
  58. 58. SeanMcCabe 8:08 am 08/3/2014

    Nevermind. I was in a bad mood. See 41, 47

    Link to this
  59. 59. naishd 8:20 am 08/3/2014

    Please don’t take it personally — it’s hard to keep up with things when comment threads become long; I overlooked the comments concerned only because they’re (now) far upthread. I’ll respond to your comments when time allows, and thanks for leaving them in the first place.

    Link to this
  60. 60. AndreaCau 9:57 am 08/3/2014

    @Mickey:”I think I’ve found a fatal flaw in the basic structure of the analysis. I’m no mathematician or programmer, but when you have a group at the tip of the tree that’s tiny (ornithothoracines), won’t every node leading to that be on average smaller than the more basal node when analyzed this way?”

    Mickey, thanks for helping us in finding that “fatal flaw” we (and our reviewers) were unable to found… Our size estimation was not just averaging values at nodes. [Furthermore, what is relevant in our study is not the (already recognised) presence of a size reduction toward ornithothoracines (as they are the smallest theropods), but the time and phylogenetic inclusiveness of that trend.]
    Size evolution was estimated using a Brownian motion model, and all femur data was log transformed to avoid an asymmetry between size increase vs reduction. Furthermore, size estimation was not just “an average” between the values of the terminal taxa/nodes included, but was estimated taking into account lineage duration (that is, distance of terminal taxa lingeages from their origin node).
    From the suppl. material:
    “Lineage durations (branch lengths) are integral to Brownian motion models, since
    large changes are less likely on short branches. Thus, the reconstructed ancestral value for a
    node will be most influenced by fossil taxa separated from that node by short branch lengths.
    For instance, the node representing the ancestral tyrannosauroid is reconstructed as small
    (FL10=2.49, ~54kg; see Fig. 1), consistent with previous proposals (43-45). Even though two
    included taxa (Yutyrannus, Tyrannosaurus) are huge, the small Guanlong (FL10=2.54, ~81kg)
    is closest to the ancestral tyrannosauroid node in terms of branch lengths, and exerts the
    strongest influence on the reconstructed state.”.

    The above extract is the reason I think the inclusion of additional taxa would not change the main result: for example, among the non-included taxa, biggest allosauroids are all Cretaceous, thus not closer to the allosauroid basal node than other missing taxa like Shidaisaurus, “Szechuanosaurus zigongensis” and so on. Biggest megalosauroids are Late Jurassic and Cretaceous, with smaller form of Middle Jurassic age. Deinocheirus is latest Cretaceous (and based on the new data, I suggest is much more derived than previously suggested based on holotype alone). Therizinosaurus is latest Cretaceous. Biggest (Mesozoic) avialans are Late Cretaceous. Biggest abelisauroids are Late Cretaceous. All basal members are usually smaller and older than their derived relatives: this empirical evidence suggests that our result is a real phenomenon and not just a spurious effect of some sort of “fatal flaw”.
    Hope soon to perform such larger analysis: the data set is almost ready, but it’s a sort of analysis that requires a lot of computation time, not only for the main analysis but also for the check analyses and eventual simulation tests.

    In the supplementary material, freely available from Science page of our study, we described in detail the theory and methodology of our analyses: please, read carefully that and eventually the mentioned literature, before commenting in such a naive way (sorry, Mickey, but it was not dissimilar from some BAND-ists comments on cladism done without understanding in detail the basis for phylogenetic analysis).

    I repeat: we performed simulation tests showing that there is NOT a size trend among the whole clade and, at the sime time, that the size reduction trend along the avian stem is NOT due to chance (including, some artefact in data collection). In particular, note that we used two indipendently developed data set, differing in taxon and character sampling, topology details, timing of cladogenesis and estimated size values at nodes: nevertheless, both analyses indicate a size trend along the avian stem and absence of a size trend along Theropoda as a whole.
    Also, the general trend is consistent with what we know of theropod lineages, with biggest members of each lineage usually more derived and later than basalmost, oldest members. The analysis of size variation (along clades and along time) supports a general trend along the avian stem.

    Since all the data used and the methods followed are freely available online (from Science page of our study, and from Dryad), I encourage you to re-analyse the data, and eventually perform alternative test, for example adding/removing taxa. Both BEAST and Bayestraits are freely available online. All is thus open and available for everyone interested to check: I’ll be very happy to receive comments and corrections on the actual analyses, and any suggestion useful for the improvement of this relatively young methodology.
    Everyone seems happy with Open Source and data sharing: so, use the shared data and provide valid counter-arguments instead of just comments that way in a blog.
    I don’t want to result arrogant! but we spent a lot of time in data collection, data elaboration and re-analysis of the data during the peer-review, after receiving a long list of detailed revisions. In every step of our study, we supported our arguments on a solid theoretical ground: such effort is much more complex and hard than just writing a naive comment on a blog. The analysis is probably preliminary, and additional works is not only necessary but welcome: but it cannot be considered just as the result of a “fatal flaw”.
    Unfortunately, for most readers, this asymmetry could be not that evident…

    Link to this
  61. 61. DavidCerny 12:14 pm 08/3/2014

    Congratulations on the great paper! It’s nice to see Bayesian methods finding their way into dinosaur paleontology. (I know that this paper and the morphoclock analysis weren’t the first to apply them to a Mesozoic dinosaur dataset, but they were probably the first to make full use of their potential.) I have a few questions about details of the analysis that don’t seem to be mentioned in the supplementary materials:

    What tree prior did you use? Yule, birth-death, or something else? One reason I ask is that several other Bayesian tip-dating analyses have used the Yule process (Pyron 2011; Wood et al. 2013) and have been criticized for that choice (Ronquist et al. 2012; Heath & Moore 2014), since the Yule process assumes an extinction rate of zero, which is unrealistic for an analysis that includes fossils.

    In your Syst Biol paper, you wrote that BEAST – unlike MrBayes – wasn’t able to correct discrete character data for ascertainment bias (i.e., sampling only variable or parsimony-informative characters). Was that still true of the version used for the present analysis? According to the supplementary materials, Andrea Cau’s Megamatrice actually does include autapomorphies and constant characters, but previous papers have suggested that it is pretty much impossible to include constant characters in a way that satisfies the assumptions of the model of evolution (you can never know if you sampled the right number of them), and correcting for the bias is the only viable solution (Lewis 2001).

    Finally, the maximum clade credibility tree is an unusual way of summarizing the posterior distribution; most papers report the 50% majority-rule consensus from the MCMC run. Was there any specific reason for this choice? Did you try to construct the 50% consensus as well? If so, were there any interesting differences between the MCC tree and the consensus tree? (I know I could find out myself using the XML file deposited in Dryad, but I’m not really familiar with BEAST. Sorry if these questions come across as superfluous.)

    Refs:

    Heath TA, Moore BH 2014 Bayesian inference of species divergence times. 277–318 in Chen M-H, Kuo L, Lewis PO, eds. Bayesian Phylogenetics: Methods, Algorithms, and Applications. Chapman & Hall/CRC

    Lewis PO 2001 A likelihood approach to estimating phylogeny from discrete morphological character data. Syst Biol 50(6): 913–25

    Pyron RA 2011 Divergence time estimation using fossils as terminal taxa and the origins of Lissamphibia. Syst Biol 60(4): 466–81

    Ronquist F, Klopfstein S, Vilhelmsen L, Schulmeister S, Murray DL, Rasnitsyn AP 2012 A total-evidence approach to dating with fossils, applied to the early radiation of the Hymenoptera. Syst Biol 61(6): 973–99

    Wood HM, Matzke NJ, Gillespie RG, Griswold CE 2013 Treating fossils as terminal taxa in divergence time estimation reveals ancient vicariance patterns in the palpimanoid spiders. Syst Biol 62(2): 264–84

    Link to this
  62. 62. Andreas Johansson 1:46 pm 08/3/2014

    David Cerny wrote:
    Also, the general trend is consistent with what we know of theropod lineages, with biggest members of each lineage usually more derived and later than basalmost, oldest members.

    That sounds rather Cope’s Rule-y.

    Is it just that big forms tend to be late? That wouldn’t really require explanation in radiating groups – if the number of species is increasing, a randomly distributed character would be more likely to be found among more numerous later forms than among less diverse early ones. Or is there directional trends in avg size?

    Link to this
  63. 63. MesozoicMammal 4:01 pm 08/3/2014

    Excellent paper, congrats to those involved (Darren, Andrea, anyone elso in the comments!).

    Just as a quick note to something that came up in the comments (Heteromeles and others) there have been papers looking at encephalization quotients in various mammal groups including hominins, a recent high profile one is Rowe et al 2011. The discuss 2 pulses of EQ increase in the lineage leading from Cynodontia to crown Mammalia. It would be super interesting to run the kind of analysis Darren reports here on the Mesozoic mammal data set, as the 2 EQ pulses of Rowe et al are based really only on 2 individual taxa and their places in a highly reduced phylogeny. Alternative phylogenetic positions and more taxa would potentially drastically change this!

    Rowe, T. B., Macrini, T. E., & Luo, Z. X. (2011). Fossil evidence on origin of the mammalian brain. Science, 332(6032), 955-957.

    Link to this
  64. 64. Yodelling Cyclist 6:56 pm 08/3/2014

    Cope’s rule makes some sense – if you are to develop large body size you need to harvest resources efficiently, which implies that you are well/highly adapted for a given ecosystem – specialised to a degree. If that ecosystem is perturbed, you die off and oops, your lineage grew through time (as you specialised and grew through intraspecific competition if nothing else) and then you went extinct.

    Not an idea that works in all cases, alas.

    Link to this
  65. 65. Therizinosaurus 4:06 am 08/4/2014

    Andrea- The way I explained myself does sound arrogant and naive, as if I saw something that you and your coauthors did not. A better way to say it is that your methodology doesn’t seem to be biologically realistic to me. As you said, “the reconstructed ancestral value for a
    node will be most influenced by fossil taxa separated from that node by short branch lengths.” The problem is that it is still seemingly _somewhat_ influenced by more distantly related taxa. So if the ‘derived’ group is large, their large size will influence the calculated size of ancestral taxa several nodes away, even if known ‘basal’ taxa in that area are all small. Thus my ornithomimosaur example.

    This same thing seems to be happening with your main bird line conclusion, except in this case, the ‘derived group’ (birds) are small. Thus each successive sister group to birds is reconstructed as smaller than it would otherwise seem to be because the smallness of birds is influencing ancestors. But this makes no sense biologically because the future has no influence on the past. You state the procedure used was developed to “infer the geographic spread of viruses (in two dimensions)” and Spencer and Wilburg (2013) note the Markov model you use for your Bayesian analyses is based on a model of molecular evolution and that “little has been done, however, to
    empirically test its efficacy” for morphological analyses. Given this, it doesn’t seem crazy to question if it gives realistic results here, unlike BANDits that question MP morphological analyses which were developed for that purpose and have had numerous tests over the decades.

    With all of this in mind, how would the data be different if shrinkage didn’t start until e.g. troodontids+birds? Because right now all of the calculated basal sizes for Tyrannosauroidea, Compsognathidae, Ornithomimosauria, Alvarezsauria+Therizinosauria, Oviraptorosauria and Dromaeosauridae seem about the same color of green, though I can’t read the actual values in figure S1 (care to send me a high res copy? :) ). So currently your paper’s thesis seems to be “our model indicates if ~160 MY old basal avialans were small, every theropod node leading to them was shrinking regardless of the size of preserved basal taxa in those lineages.” It’s an interesting idea, but given that the model’s effectiveness hasn’t been previously established I wonder how much any conclusion based on it means.

    To test if it’s just these avialans that are skewing the data, what happens if you delete all troodontids and taxa closer to birds from Dataset 1? Do you still get shrinking along the coelurosaur lineage then?

    Reference- Spencer and Wilberg, 2013. Efficacy or convenience? Model-based approaches to phylogeny estimation using morphological data. Cladistics. 29(6), 663-671.

    Link to this
  66. 66. AndreaCau 5:12 am 08/4/2014

    David Cerny,

    -What tree prior did you use?

    We used the simple sostitution model, with empirical base frequencies and the gamma distribution for site heterogeneity.

    -In your Syst Biol paper, you wrote that BEAST – unlike MrBayes – wasn’t able to correct discrete character data for ascertainment bias (i.e., sampling only variable or parsimony-informative characters). Was that still true of the version used for the present analysis? According to the supplementary materials, Andrea Cau’s Megamatrice actually does include autapomorphies and constant characters, but previous papers have suggested that it is pretty much impossible to include constant characters in a way that satisfies the assumptions of the model of evolution (you can never know if you sampled the right number of them), and correcting for the bias is the only viable solution (Lewis 2001).

    I assume that any morphological analysis of fossils intrinsecally undersamples the morphological disparity. This is true even for parsimony-based analyses. Therefore, what matters is not to “completely satisfy” that assumption, but to approaches it as much as possible, reducing the bias present in traditional (parsimony) datasets that a priori exclude autapomorphies. (My dataset was exapted for this as I’ve tried to include as much characters as possible). Note that even the analyses based on molecular data undersample the “total set” of relevant features, as usually focus on particular loci or genome parts.

    -Was there any specific reason for this choice [the use of the MCC tree]?

    It’s BEAST default.

    -Did you try to construct the 50% consensus as well?

    No (explained below).

    -If so, were there any interesting differences between the MCC tree and the consensus tree?

    Yes, the MCC is the tree with the highest product of probability for each clade. The consensus tree, well, is just a consensus among the sampled trees (as in parsimony analyses), and . There is not a “best” tree representation, each following particular assumptions and thus focusing on aspects not addressed by other representations. I’m currently learning new ways to draw the results of these analyses beyond the mere “cladogram”…

    Link to this
  67. 67. naishd 5:37 am 08/4/2014

    With reference to comment # 65 – I don’t have time, right now, to respond in detail to any of these criticisms, sorry (today is deadline for a major editing job). Mickey – seriously, I don’t think what you say reflects our results nor shows appreciation of how the study was done. It stated clearly in the paper that the estimated sizes of concestors at all nodes on the bird line are smaller than those at preceding nodes – this result is not (of course) a reflection of animals anticipating the future, but is a genuine pattern that we recovered and would occur regardless of what’s going on in Maniraptora.

    And of course the Markov model has been criticised. Bayesian methods in general are disliked by many people who work on phylogenetics. Doesn’t mean we shouldn’t try to use them.

    I think you’re looking at this paper too ‘superficially’ – you keep pointing to specific little issues whereas the whole project is about a big-picture trend which is robustly supported by covarying variables.

    Link to this
  68. 68. DavidCerny 7:20 am 08/4/2014

    @ Mickey Mortimer #65:

    Spencer and Wilburg (2013) note the Markov model you use for your Bayesian analyses is based on a model of molecular evolution and that “little has been done, however, to empirically test its efficacy” for morphological analyses.

    Well, Spencer and Wilberg published their paper in a journal that has been increasingly dedicated to “proving” that parsimony is superior to all other methods, despite mounting evidence to the contrary. I wouldn’t expect it to provide a fair and balanced critique of Bayesian methods.

    The specific argument you cite isn’t very sound either. Yes, Lewis derived his Mk model as a generalization of a molecular substitution model (JC69). However, what remained of the molecular model are the following assumptions: (1) characters have several states, (2) the states all have the same equilibrium frequency, (3) the transitions between all states are equally probable*. It may be argued that these assumptions are highly unrealistic (then again, a model isn’t meant to be realistic), but they are cleary no less applicable to morphology than to molecules.

    *This can be relaxed, of course, by ordering the states.

    @ Andrea Cau #66:

    Many thanks for the answers. I may have phrased some of my questions inaccurately, though:

    -What tree prior did you use?

    We used the simple sostitution model, with empirical base frequencies and the gamma distribution for site heterogeneity.

    That’s not what I meant. Your analysis estimated the joint posterior density of a large number of parameters (topology, node ages, branch rates, ancestral sizes, parameters of the substitution model, and some others). In Bayesian inference, a prior has to be specified for each of those parameters. The prior distribution of topology and node ages is usually called the “tree prior” (though there are many other names: time-tree prior, tree process prior, branching process prior etc.). Some commonly used tree priors are the Yule and birth-death processes. While the supplementary materials provide information about some of the priors you used (e.g., the prior on rates), they don’t mention what your tree prior was.

    I assume that any morphological analysis of fossils intrinsecally undersamples the morphological disparity. This is true even for parsimony-based analyses.

    Indeed – it’s probably especially true of parsimony analyses.

    Therefore, what matters is not to “completely satisfy” that assumption, but to approaches it as much as possible, reducing the bias present in traditional (parsimony) datasets that a priori exclude autapomorphies.

    I agree that the inclusion of autapomorphies is an important step toward fulfilling the requirement to sample all characters. I should have emphasized that I’m primarily interested in how you dealt with constant characters – whether by not using them and correcting for the resulting bias, or by including as many of them as possible.

    Note that even the analyses based on molecular data undersample the “total set” of relevant features, as usually focus on particular loci or genome parts.

    That’s true, but if you sequence a particular 500-bp locus and find out that 350 sites are variable within your taxon sample and 150 are constant, then that’s it. What I (and Lewis) meant is that this is, in principle, impossible when compiling a morphological data matrix. When you score 350 variable characters and decide to include some constant ones, you can’t know when you should stop: after you have 50 of them? 150? 500?

    However, perhaps this is more of a hypothetical problem and the Mk model might be quite robust to violating its assumption that the “proper” number of constant characters are included.

    Link to this
  69. 69. John Harshman 9:50 am 08/4/2014

    Thus each successive sister group to birds is reconstructed as smaller than it would otherwise seem to be because the smallness of birds is influencing ancestors. But this makes no sense biologically because the future has no influence on the past.

    This would seem to be a much broader critique than you want to make it. Everything you say would apply as much to parsimony as to any explicitly model-based approach. More broadly, it would apply to phylogenetic analysis of extant species. And even more broadly, to all historical science, in which we try to estimate past events by observing data collected recently. I doubt you would want to apply it that broadly, but how can you restrict your argument just to what you want to criticize?

    But fortunately, I think the argument is invalid. The future doesn’t determine the past, but we can use the future to estimate the past; in fact that’s pretty much all we ever do. That’s what Bayesian analysis does, and that’s what parsimony does.

    Link to this
  70. 70. DavidCerny 10:37 am 08/4/2014

    … okay, I may not be familiar with BEAST, but I can open an XML file. Lee et al. used the birth-death process as their tree prior, unlike Pyron or Wood et al.

    Link to this
  71. 71. MMartyniuk 7:49 pm 08/4/2014

    I think one very good point raised here that sounds like a potential follow up study, is that when were talking about evolutionary trends, the maximum size, average size, etc. of an ancestral taxon probably dens matter much. What matters is *size at sexual maturity*, especially in animals that have been shown to reach maturity relatively early in ontogeny. If a Deinocheirus or Tyrannosaurus could breed at half or less maximum size, wouldn’t entering the femur length of a large adult actually cloud any trend that may be present? This could actually push the overall trend for size reduction further down the tree. But then I guess you’ve got two questions: when and how did size at maturity begin to decrease, and when and how did maximum post-maturity size start to decrease.

    Link to this
  72. 72. AndreaCau 4:27 am 08/5/2014

    @David,
    I forgot to refer you to the actual file freely available on our Dryad page.

    @Mickey,

    your critique to the “future influencing the past” is rethorical, and a bit hypocritical since the same argument works for the phylogenetic analyses you usually perform. We reconstruct and ancestral condition based on the taxa our analysis indicate descended from that ancestor. In this respect, a parsimony optimisation works the same way: inferring the past ancestral node based on the future terminal OTUs branching from that node. Our approach adds an additional source of information: the time. I think that in an historical science as phylogenetic evolutionism in part is (and must be), the time-distance from an ancestor is as relevant as the mere character optimisation.
    Stupid over-simplicistic example: hope everyobody agrees that the last common ancestor of Archaeopteryx and Passer (that we place at 163 Ma) was morphologically more similar to Archaeopteryx than Passer, not because Archaeopteryx is “morphologically primitive” but bacause Passer is chronologically much more distant than Archaeopteryx from the 163 Ma ancestor (13 Mys vs 163 Mys), thus, a rough optimisation of its features would consider the “Archaeopteryx body plan” more than “the Passer body plan”. This is not a return to a gradistic approach (we continue to operate in a hennighian system where only synapomorphies define clade: but evolution is more than just synapomorphies, as plesiomorphies and autapomorphies are relevant in recosntructiong the whole picture). It’s just a probabilistic assumption, that the more time the more morphological change (at least as null hypothesis that evolution happens). For example our approach explains why the aberrant Balaur is so aberrant: our analysis indicates that it belongs to a very long lineage that accumulated a lot of autapomorhies along its history: it’s position is therefore among those most biased by Long Branch Attraction (and its uncertain placement among Paraves confirms this).

    Link to this
  73. 73. AndreaCau 4:46 am 08/5/2014

    @David: When you score 350 variable characters and decide to include some constant ones, you can’t know when you should stop: after you have 50 of them? 150? 500?

    I follow a different approach. The difference between costant character, autapomorphy and synapomorphy is a posteriori result of taxon sampling, cannot be define a priori. I score all characters I consider as phylogenetically relevant for a taxon sample larger than the one will be the aim of my analysis: the exclusion of those taxa not members of the proper ingroups determine what character is costant and what is a potential synapomorphy. Note that many local autapomorphies are homoplastic features that are synapomorphies of some clades and convergently result as present along one or more terminal branches.
    In my theropod dataset I followed that procedure. The number of costant/autapomorphic characters is thus not assumed by me, but is just a conseguence of the particular taxon sample. I think this is a realistic approach, since we cannot exclude that what we actually consider as an autapomorphy will result a synapomorphy once new taxa are added, and what is considered as a costant character will result a plesiomorphy once new taxa are added. In short, taxon sample matters.

    Link to this
  74. 74. Augray 10:37 pm 08/5/2014

    Congratulations to Darren (and Andrea) on your publication!

    This study would seem to be another blow to BCF (which Darren outlines at http://scienceblogs.com/tetrapodzoology/2009/06/08/birds-come-first-hypothesis/ and
    http://scienceblogs.com/tetrapodzoology/2009/06/10/birds-come-first-no-they-dont/ ). I admittedly don’t entirely understand the methodology of your new paper, but I would assume that if the bird stem lineage were a group of small arboreal archosaurs, then the pattern wouldn’t show a steady decrease, but an “oscillation” around a small average size.

    And this got me thinking… just to play Devil’s Advocate, what if the trend in your study is the result of taphonomic bias in the fossil record? After all, until about twenty years ago, there were only three bird species known from the Mesozoic, and this was blamed on the small size and fragility of bird skeletons. Is it possible that a similar situation would explain the absence of small theropods, arboreal or not?

    I suspect that the above scenario is ruled out just by the statistical improbability that it would result in the pattern seen in your paper, but I’m admittedly uncertain about that.

    Finally, if anyone is unaware of Andrea’s blog at http://theropoda.blogspot.ca/, I highly recommend it. Although it is in Italian, Google Translate does a half decent job… most of the time.

    Link to this
  75. 75. Therizinosaurus 3:31 am 08/6/2014

    Darren Niash wrote- “It stated clearly in the paper that the estimated sizes of concestors at all nodes on the bird line are smaller than those at preceding nodes – this result is not (of course) a reflection of animals anticipating the future, but is a genuine pattern that we recovered and would occur regardless of what’s going on in Maniraptora.”

    If that’s so, do you still get this result of shrinking coelurosaurs if troodontids and taxa closer to birds than them are deleted from Dataset 1?

    David Cerny wrote- “Spencer and Wilberg published their paper in a journal that has been increasingly dedicated to “proving” that parsimony is superior to all other methods, despite mounting evidence to the contrary. I wouldn’t expect it to provide a fair and balanced critique of Bayesian methods.”

    I certainly don’t follow the phylogenetic analysis literature very closely, but do you then have examples of where a Bayesian approach to morphology has been tested and shown to be accurate? If there were studies showing Bayesian analyses of morphology were closer to molecular results than MP morphological analyses were, for example, that would be great evidence. But just claiming there’s a bias is a poor argument, even if there is a bias, which could be true for all I know.

    John Harshman wrote- “This would seem to be a much broader critique than you want to make it. Everything you say would apply as much to parsimony as to any explicitly model-based approach.”

    Or would it? In an MP analysis, the state of a ‘derived’ group doesn’t affect the reconstructed ancestral states of ‘primitive’ groups if those primitive groups have known morphologies. You can chop off birds in an MP analysis and get the same reconstructed ancestral states for other coelurosaurs, but that doesn’t seem to be the case for this one. It’s as if we have Jurassic birds with 8 sacrals, and their outgroups are Cretaceous deinonychosaurs with 5 sacrals, then Cretaceous oviraptorosaurs with 5 sacrals, then Cretaceous therizinosaurs with 5 sacrals, then Cretaceous ornithomimosaurs with 5 sacrals, etc.. MP would say 8 sacrals are a bird character and everything previous had 5, but this analysis would seem to say the basal paravian had 8 sacrals, and the basal pennaraptoran had 7, as did the basal maniraptoran, and the basal maniraptoriform had 6. It drags the future state into past nodes further than I think makes sense biologically.

    Now maybe this is correct and mass reversals leading to known Cretaceous taxa having few sacrals or larger size is how evolution worked (BCF will need reevaluation then), but man is it contradictory to how we’ve treated morphological cladistics until now. Just imagine the diagnostic character lists from such assumptions- “Maniraptora is diagnosed as having character X, though the first four clades that branch from it don’t show that character, but the earliest member of the fifth clade has it and lived earlier, so we propose the basal maniraptoran had it too, and it was seemingly lost in later members of the first four clades.” This is how the size character seems to work, so that instead a discovery based on current methods, it’s a hypothesis based on a methodology palaeontologists don’t use for any other character’s history. Again maybe it’s right, but I think many of us have issues with age and distance having such precedence over preserved morphology.

    Link to this
  76. 76. naishd 4:25 am 08/6/2014

    “If that’s so, do you still get this result of shrinking coelurosaurs if troodontids and taxa closer to birds than them are deleted from Dataset 1?”

    Yes. Did you read the paper? The reconstructed ancestors at successive nodes are each smaller than the one at the preceding node.

    Link to this
  77. 77. Therizinosaurus 9:16 am 08/6/2014

    “Yes. Did you read the paper? The reconstructed ancestors at successive nodes are each smaller than the one at the preceding node.”

    By “reconstructed ancestors” do you mean the hypothetical concestors along the bird line, as labeled in figure 1? Because it’s those values I’m saying are skewed in Bayesian analyses by known early basal birdy taxa (Aurornis, Anchiornis, etc.) being small. Note in the MP analyses, the size stays the same from basal coelurosaurs until birds (minus the affects of juvenile Epidexipteryx in Dataset 1 and weirdly basal Rahonavis in Dataset 2).

    If you mean the estimated basal size of each clade that branches off the bird line, this isn’t true in Dataset 2 at least, where compsognathids are reconstructed as smaller ancestrally than alvarezsaurs or enigmosaurs (fig. S5). I can’t read figure S1 for those values but they seem to be the same shade of green.

    Link to this
  78. 78. naishd 10:16 am 08/6/2014

    Hi Mickey. Yes, I mean the concestors along the bird line. The implication that the skewing of inferred sizes of root-ward taxa in the tree by crown-ward taxa is problematic or faulty is erroneous, because a divergence at a node obviously includes individuals/populations/taxa that are evolving in the crown-ward direction; ergo, the small size of crown-ward taxa should indeed influence the reconstructed size/anatomy of the concestor at the node. I can only say again that this analysis works by analysing proxies for size (log transformed femur length) at the same time as stratigraphic age and phylogeny. So far as we can tell from every test (this is a Science paper: it went through a crapload of checking, re-checking and criticism), the trend we report – sustained miniaturisation across tetanuran nodes – is a real signal, even when there are doubts about specific taxa and specimens. And, yes, Science screwed us with the low res of some of the figures. We didn’t know that would happen (if you need hi-res ones, contact the corresponding author).

    Apologies to Sean: haven’t forgotten your questions…

    Link to this
  79. 79. DavidCerny 11:06 am 08/6/2014

    @ Andrea Cau #73:

    Wow, that approach actually seems to be a good approximation of what is proposed here and considered to be near impossible!

    @ Mickey Mortimer #75:

    I certainly don’t follow the phylogenetic analysis literature very closely, but do you then have examples of where a Bayesian approach to morphology has been tested and shown to be accurate? If there were studies showing Bayesian analyses of morphology were closer to molecular results than MP morphological analyses were, for example, that would be great evidence.

    It’s strange that you want a case of Bayesian inference outperforming parsimony just to show it gives accurate results. In most cases, Bayesian and parsimony morphological trees are far more similar to each other than either of them is to a molecular tree (see, for example, Glenner et al. 2004; Nylander et al. 2004: Figure 2; Wiens et al. 2010). Often, the only difference is that the Bayesian tree is more resolved than the parsimony one, and that’s still somewhat misleading, since collapsing the nodes with low posterior probabilities would decrease the resolution of the Bayesian tree as well.

    It seems to me that the fact that Bayesian inference with the Mk model almost always produces the same tree as parsimony means it passes the test you (and Spencer & Wilberg) are calling for.

    But just claiming there’s a bias is a poor argument

    True, but that definitely wasn’t my whole argument. By listing the assumptions of the Mk model, I was trying to show that Spencer and Wilberg’s (2013: 664) claim that “[i]t is an attempt at equating phenotypic change to an explicit model of genotypic change without accounting for unknown variables” was incorrect. True, the model might be oversimplified, but it’s not as if it tried to squeeze morphology into an alien framework whose basic assumptions aren’t applicable to it. (See above – surely in such a case we would expect the resulting topologies to be grossly wrong, which they are not.)

    It’s as if we have Jurassic birds with 8 sacrals, and their outgroups are Cretaceous deinonychosaurs with 5 sacrals, then Cretaceous oviraptorosaurs with 5 sacrals, then Cretaceous therizinosaurs with 5 sacrals, then Cretaceous ornithomimosaurs with 5 sacrals, etc.. MP would say 8 sacrals are a bird character and everything previous had 5, but this analysis would seem to say the basal paravian had 8 sacrals, and the basal pennaraptoran had 7, as did the basal maniraptoran, and the basal maniraptoriform had 6.

    No, that’s not how it works. Your example doesn’t even apply to what Lee et al. did. You seem to be describing ancestral state reconstruction on a fixed topology, but the analysis we’re talking about estimated ancestral states and topology simultaneously – along with a large number of other parameters, each of which exercised an influence on the others. The reconstruction of ancestral sizes by Lee et al. doesn’t depend on any specific topology (or node ages, or branch rates, etc.) being true.

    It’s not trivial to predict what states Bayesian inference would reconstruct in the case you describe. Even when using an empirical Bayes approach, which works on a fixed topology, you can’t be sure what the actual analysis would show based on the information you have. A lot of other thing would have to be specified as well (branch lengths, for example). There is absolutely no reason to expect that you would see the number of vertebrae gradually increasing as in your example. When topology and other nuisance parameters are not fixed to specific values but integrated out, as in Lee et al., guessing the result in this way is outright impossible, unless you can do high-dimensional integration in your head.

    Refs:

    Glenner H, Hansen A, Sørensen M, Ronquist F, Huelsenbeck JP, Willerslev E 2004 Bayesian inference of the metazoan phylogeny: a combined molecular and morphological approach. Curr Biol 14: 1644–9

    Nylander JAA, Ronquist F, Huelsenbeck JP, Nieves-Aldrey JL 2004 Bayesian phylogenetic analysis of combined data. Syst Biol 53(1): 47–67

    Wiens JJ, Kuczynski CA, Townsend T, Reeder TW, Mulcahy DG, Sites JW Jr 2010 Combining phylogenomics and fossils in higher level squamate reptile phylogeny: molecular data change the placement of fossil taxa. Syst Biol 59(6): 674–88

    Link to this
  80. 80. AndreaCau 4:32 am 08/7/2014

    @Mickey wrote: It’s as if we have Jurassic birds with 8 sacrals, and their outgroups are Cretaceous deinonychosaurs with 5 sacrals, then Cretaceous oviraptorosaurs with 5 sacrals, then Cretaceous therizinosaurs with 5 sacrals, then Cretaceous ornithomimosaurs with 5 sacrals, etc.. MP would say 8 sacrals are a bird character and everything previous had 5, but this analysis would seem to say the basal paravian had 8 sacrals, and the basal pennaraptoran had 7, as did the basal maniraptoran, and the basal maniraptoriform had 6.

    I’ve tested Mickey’s hypothesis (that is a character optimisation using Bayesian inference), using Bayesian character reconstruction on our branch calibrated topology and will show the result on my blog. In short, 5 sacrals is the reconstructed state from all the neotheropod nodes until the Balaur+more derived birds node. Increase in sacral number is placed in some subclades, e.g., the neoceratosaurs, derived ornithomimosaurs, eudromaeosaurids, derived unenlagiines, and derived birds.

    Sorry Mickey if I result arrogant, but, again, it is evident from your comment above that you don’t understand how our analysis worked. In particular, the number of sacrals is a discrete character (as all the 1549 excluding the log femur character), and all discrete ones were not analysed the same way the femur continuous characters was, only the latter was analysed using the Brownian motion model, that – I repeat – is not just averaging values on a fixed topology.

    All the references to the theoretical basis of our study are mentioned and explained in the supplementary material of our paper, all freely available: as I wrote before, please, read it carefully, and read the relevant literature (most available online).
    Sorry for the tone if it seems rude.

    Link to this
  81. 81. DavidMarjanovic 5:04 pm 08/9/2014

    I think we can all agree that some specimens are less ideal than would be perfect in view of ontogenetic stage and so on. But should we have excluded all the specimens concerned? As David said above, that might have left us with far fewer specimens/taxa that we might like – as usual, we can only work with what we have.

    Well, sometimes you can’t work with what you have, and then you can’t work at all. I’m not saying you should necessarily have ditched all the taxa known only from skeletally immature individuals; and I’m not saying either that if you had done that, you’d have been left with too few to work with; I’m only saying that the principle applies that sometimes the world really does suck, sometimes an analysis really is impossible to do, and in that case one must pack up and go home instead of using inadequate data and getting “garbage in, garbage out”, a law against which no method is immune. Again: I’m not saying the input is garbage; I’m saying it needs to be demonstrated that it isn’t garbage.

    Not long ago, I reviewed a manuscript where the authors, among many other things, estimated changes in diversification rates over time by dating nodes on a tree in which all terminal taxa were extant. That is plainly impossible to do. Several published papers have done such analyses, but the results are completely worthless: there’s no extinction in such trees. If you don’t know the past well enough, you can miss entire radiations. Sometimes, the best you can do with the data you have is not good enough.

    And this got me thinking… just to play Devil’s Advocate, what if the trend in your study is the result of taphonomic bias in the fossil record? After all, until about twenty years ago, there were only three bird species known from the Mesozoic, and this was blamed on the small size and fragility of bird skeletons. Is it possible that a similar situation would explain the absence of small theropods, arboreal or not?

    The small size and fragility of “bird” skeletons (for a rather large value of “bird”) still has to be blamed for the absence of “birds” from most continental Mesozoic sites. Only in sites with extremely, unusually good preservation do “birds” turn up.

    Link to this
  82. 82. DavidCerny 11:46 am 08/10/2014

    Several published papers have done such analyses, but the results are completely worthless: there’s no extinction in such trees.

    Working on extant taxa alone does not amount to assuming that extinction doesn’t occur. In fact, the most commonly used diversification model in molecular phylogenetics is called the birth-death process, where “death” – of course – refers to extinction. The idea is that while a speciation rate of 0.95 and an extinction rate of 0.45 may produce the same net diversification rate as a speciation rate of 0.5 and an extinction rate of 0, they will lead to different distributions of speciation ages on the resulting tree. There are papers showing that the power of molecular data to estimate the extinction rate reliably is quite low, particularly when certain common but unrealistic constraints are relaxed (e.g., Rabosky 2010), which supports your objection, but that still doesn’t mean that analyses based on extant taxa ignore extinction out of hand.

    Ref:

    Rabosky DL 2010 Extinction rates should not be estimated from molecular phylogenies. Evolution 64(6): 1816–24

    To return to the topic of the article, Mike Lee has uploaded a hi-res version of the supplementary materials to academia.edu:

    http://www.academia.edu/7832790/Lee_M.S.Y._Cau_A._Naish_D._Dyke_G.J._2014._Sustained_miniaturization_and_anatomical_innovation_in_the_dinosaurian_ancestors_of_birds._Science_345_562-566_NOTE_SM_below_has_hi-res_figures_which_were_pixellated_on_the_Science_SM_pdf_

    Link to this
  83. 83. DavidMarjanovic 12:57 pm 08/13/2014

    Point taken. Still, if cladogenesis and extinction are too unevenly distributed – radiations and mass extinctions, say, instead of just background extinction –, the age distribution of the nodes will likely be misleading.

    Link to this
  84. 84. greg_t_laden 12:06 am 08/19/2014

    I would love to hear Darren’s take on the Oxygen question. This time period does correspond to an increase in oxygen. Also, during previous periods, flying insect size seems to track atmospheric O2 levels, until the birds show up, then the insects never get big again.

    (see: Matthew E. Clapham and Jered A. Karr. Environmental and biotic controls on the evolutionary history of insect body size. PNAS 2012 109 (27) 10927-10930; published ahead of print June 4, 2012, doi:10.1073/pnas.1204026109)

    Link to this
  85. 85. naishd 1:38 pm 08/19/2014

    Hi Greg. I’m pretty ambivalent on the idea that changing O2 curves might have influenced dinosaur evolution (or, indeed, insect evolution): the correlations are not as tight as some authors have implied, there are always other factors involved (insect body size takes a dive at the time that terrestrial vertebrates diversify, for example), and Mesozoic O2 curves show that the time of greatest relevance as goes dinosaur evolution (the Late Triassic and much of the Jurassic) had O2 levels similar to those of today, or lower.

    Indeed, I don’t think there’s good evidence for an atmospheric O2 percentage higher than c. 21% during the time when theropods underwent their major diversification and shrinkage. Several studies have looked at the large-scale patterns of dinosaur diversification and found that the general evolutionary patterns are more likely driven by innovations in biology and anatomy, not by climatological, atmospheric or geological events (Sander et al. 2011, Sookias et al). 2012).

    Ref – -

    Sander, P. M., Christian, A., Clauss, M., Fechner, R., Gee, C. T., Griebeler, E. M., Gunga, H.-C., Hummel, J., Mallison, H., Perry, S., Preuschoft, H. Rauhut, O., Remes, K., Tütken, T. Wings, O. & Witzel, U. 2011. Biology of the sauropod dinosaurs: the evolution of gigantism. Biological Reviews 86, 117-155.

    Sookias, R. B., Benson, R. B. J. & Butler, R. J. 2012. Biology, not environment, drives major patterns in maximum tetrapod body size through time. Biology Letters 8, 674-677.

    Link to this
  86. 86. greg_t_laden 1:57 pm 08/21/2014

    Darren,

    Thanks for that clarifying comment.

    Link to this
  87. 87. greg_t_laden 10:08 pm 08/26/2014

    It suddenly occurs to me that the ensmallening trend might be a statistical artifact.

    Link to this

Add a Comment
You must sign in or register as a ScientificAmerican.com member to submit a comment.

More from Scientific American

Scientific American Holiday Sale

Give a Gift &
Get a Gift - Free!

Give a 1 year subscription as low as $14.99

Subscribe Now! >

X

Email this Article

X