(click for larger image)

Reconstruction of O. reburrus by M. Collins. The precise arrangement of the anteriormost region remains somewhat conjectural.

Halkieriids are Cambrian animals that looked like slugs in scale mail; often when they died their scales, called sclerites, dissociated and scattered, and their sclerites represent a significant component of the small shelly fauna of the early Cambrian. They typically had their front and back ends capped with shells that resembled those we see in bivalve brachiopods. Wiwaxiids were also sluglike, but sported very prominent, long sclerites, and lacked the anterior and posterior shells; their exact position in the evolutionary tree has bounced about quite a bit, but some argument has made that they belong in the annelid ancestry, and that their sclerites are homologous to the bristly setae of worms. One simplistic picture of their relationship to modern forms was that the halkieriids expanded their shells and shed their scales to become molluscs, while the wiwaxiids minimized their armor to emphasize flexibility and became more wormlike. (Note that that is a very crude summary; relationships of these Cambrian groups to modern clades are extremely contentious. There’s a more accurate description of the relationships below.)

Now a new fossil has been found, Orthozanclus reburrus that unites the two into a larger clade, the halwaxiids. Like the halkieriids, it has an anterior shell (but not a posterior one), and like the wiwaxiids, it has long spiky sclerites. In some ways, this simplifies the relationships; it unites some problematic organisms into a single branch on the tree. The question now becomes where that branch is located—whether the halwaxiids belong in a separate phylum that split off from the lophophorate family tree after the molluscs, or whether the halwaxiids are a sister group to the molluscs.

Here are some photos of the actual fossils.

(click for larger image)

O. reburrus from the Middle Cambrian Burgess Shale. (A to D) Dorsal view. (A) Holotype, ROM 57197; (B) ROM 57837; (C) ROM 57835; (D) ROM 57839. White arrow in (C) indicates a bended spine. (E to F) Ventral view. (E) ROM 57836; (F) ROM 57838. Only the doublure of the shell is visible in (E). Images were obtained by light microphotography on uncoated material. Scale bars, 1 mm. Cu, cultrate; Gu, gut; Sc, dorsal sclerites; Sh, shell; Sp, dorso-lateral spines.

This work doesn’t yet resolve exactly where to place the halwaxiids on the family tree—there are a couple of possibilities ilusstrated below. (Note also the mention of Odontogriphus, which I wrote about before; it’s like an early Cambrian slug without the armor).

(click for larger image)

An outline of lophotrochozoan phylogeny showing the two most plausible positions of the halwaxiid O. reburrus, depending on the assumed polarity of sclerite acquisition and biomineralization in the associated taxa Kimberella (Kim.), halkieriids (Hal.), Odontogriphus (Odo.), siphogonuchitids (Sip.), and Wiwaxia (Wiw.). Dashed lines indicate alternative interpretations of the phylogeny. The first hypothesis (hypothesis 1) accepts Odontogriphus (and probably Kimberella) as stem-group mollusks, with the halwaxiids as a sister group of mollusks. In this latter clade, chitinous sclerites are first acquired (in Wiwaxia), followed by their biomineralization in the siphogonuchitids. Members of this latter group, however, are stratigraphically older and appear to have a simpler scleritome (10). Halkieriids would then reacquire a more complex scleritome [similar to that of Wiwaxia] and shells. In the sister group represented by Orthrozanclus, the sclerites demineralize, and the posterior shell is probably lost (or highly reduced). The second hypothesis (hypothesis 2) treats the halwaxiids as monophyletic, with the further implication that Odontogriphus (and probably Kimberella) are stem-group lophotrochozoans. In hypothesis 2, the earliest halwaxiids are the siphogonuchitids with a mineralized scleritome of two types of sclerite and with a shell composed of fused sclerites. Shells are then acquired, along with a third type of sclerite, in the halkieriids. Demineralization of sclerites occurs in Orthrozanclus, and (finally) complete shell loss occurs in Wiwaxia. A cladistic analysis gives some support for hypothesis 1, but the best tree is not robust.

While the relationships are still a bit murky, one important thing is that the murkiness unites three phyla, the molluscs, annelids, and brachiopods, and the difficulty in placing their ancestors into discrete lineages implies a tighter kinship between these groups in the Cambrian than you’d guess from the taxonomy—oh, what we could learn from a few Cambrian DNA samples!

Conway Morris S, Caron J-B (2007) Halwaxiids and the Early Evolution of the Lophotrochozoans. Science 315(5816):1255-1258.


  1. #1 David Marjanovi?
    March 2, 2007

    I’ve just read the article and the supplementary information. It contains the worst cladistic analysis of the decade. Twenty taxa and twenty-three characters… ARGH! Are we back to the 1970s!?! As you can guess, 23 characters are not even sufficient to establish the monophyly of the ingroup. (That’s right — the outgroup is part of the big bush.) And then they blissfully proceed to bootstrap the results. I wonder if a bootstrap test on such a teeny-tiny matrix is capable of saying anything of statistical significance.

    Plus, they don’t seem to have understood that PAUP* cannot deal with inapplicable characters. If you code something as “-”, PAUP* will interpret it as a gap, even if the rest of the matrix is not DNA or protein. We can only hope they had the settings on interpreting gaps as missing data (which is fortunately the default) rather than as a 5th base/21st amino acid.

    I like their hypotheses, but they should have tested them.

  2. #2 David Marjanovi?
    March 2, 2007

    No matter how many taxa and characters one take into account, phylogenetic trees based on cladistic methods are highly unreliable as they have mislead zoologists for the last decades. Molecular phylogenetics have refuted almost all of the “clades” based on cladistics.

    Oh man. Molecular phylogenetics is cladistics too. (Except for neighbor-joining. That is phenetics — it’s not phylogenetics in the first place.)

    Most ironically, the lophotrochozoan taxon was only proposed based on molecular data (Hox genes, 18S RNA,…) and now paleontologists (unfortunately have to) use this very, very unreliable technique to fit their fossils into the recent molecular phylogenies.

    Excuse me? Because DNA fossilizes so badly, they can’t use it. Conway Morris got his idea independently.

    Not that molecular phylogenetics can resolve every detail of the animal tree, but here are some major concepts on which entire evolutionary theories have been based upon and that have been proven wrong:

    I agree with all your examples. None of them has ever been found by a serious (read: many taxa, loads of characters) morphological cladistic analysis. In fact, morphological support exists for each of the molecular-based hypotheses you imply here; it just has never been adequately tested. Here in the Mesozoic dinosaur world we’ve passed the 300-character mark for matrices with less than 50 taxa, and the limits (currently 75 taxa and 438 characters) are pushed every few years.

    and taking into account that maximal parsimony can never account correctly for the loss of characters

    What do you mean, the fact that it doesn’t try to account for long-branch attraction? That’s an entirely valid point. So, however, is the opposite: maximum likelihood assumes that each position evolves with one of few velocities; parsimony makes no such assumption, thus in effect it assumes that each position evolves at its own speed – and behold, there are plenty of realistic situations where parsimony performs better than likelihood (and Bayesian methods).

    B. Kolaczkowski & J. W. Thornton. 2004. Performance of maximum parsimony and likelihood phylogenetics when evolution is heterogeneous. Nature 431:980-984.

    Incidentally, Bayesian methods can be used with morphological data, and have been several times recently.

  3. #3 David Marjanovi?
    March 2, 2007

    You sound like you’re making valid criticisms, but it’s way over my head–can I get you to explain your terminology? I’m comfortable enough dogpaddling around in the same pool as biologists, but you’re way over in the deep end…

    Let’s see if I can manage a very short explanation: To reconstruct a phylogenetic tree, you look at the organisms – but you only look at shared derived character states. We have five fingers per hand, and so does every self-respecting opossum, while horses have a single finger per hand; but in spite of this we are more closely related to the horses than to the opossums – having five fingers per hand is the normal state, it doesn’t tell us anything. Doing this systematically, and applying the principle of parsimony* in some way (either directly, as here, or indirectly to account for mutation rates, as usually done in the molecular world), is called cladistics; it produces falsifiable hypotheses, so its adoption (from the late 1960s onwards, never mind its invention in 1950 in East Germany) has changed phylogenetics from an art to a science.

    * The principle that, of all hypotheses that explain the data equally well, those that require the smallest number of ad hoc assumptions must be preferred. In this case it means that those hypotheses that require the smallest number of convergence events and reversals win.

    Now about the numbers. If you have too little information, you won’t get a fully resolved tree. Trivially, if you use fewer characters than taxa ( = species or larger groups [or smaller ones...]) in your analysis, you can’t get a fully resolved tree. In theory, having as many characters as taxa should suffice, but this is only true if neither convergence nor reversals ever occur, and that’s a quite unrealistic assumption. Conway Morris & Caron have 20 taxa and 23 characters; it’s no wonder they get a broad bush rather than a tree.

    The ingroup is the group in the internal relationships of which you’re interested. That the ingroup is monophyletic with respect to the outgroup – shares a common ancestor that was not an ancestor of the outgroup – is assumed and not tested. The outgroup is there to tell us that having 5 fingers per hand is in fact normal.

    There are theoretical studies on whether it’s good to keep adding more characters and more taxa to phylogenetic analyses. Answer: Adding characters stops adding new information at a certain ratio of characters per taxon that seems never to have been reached. Adding taxa is always good. This is because the signal adds up while the noise – random convergence & reversals – cancels itself out. I’m used to see large matrices being used to tackle comparatively small problems (2 years ago a study was published that used 347 characters just for one group of sauropod dinosaurs) and still getting less-than-ideally resolved trees, so if this tiny matrix were the best that can be done in tackling as large and interesting a problem as, well, basic animal phylogeny, I’d have to get quite desperate.

    Any more questions? :-]

  4. #4 David Marjanovi?
    March 2, 2007

    OK, so that wasn’t “very short”… Try this tutorial instead. Highly recommended!

  5. #5 Torbjörn Larsson
    March 3, 2007

    maximum parsimony

    Ah, what the correct keyword can teach you! Thank you, David. I thought bayesian methods (BM) in cladistics were used for parsimony analyzes. They can do that in physics, or at least it provides a measure for doing that. (Finding the most parsimonious model in regards free parameters.)

    But Wikipedia explains they do not do that here, it is the much more common likelihood analysis BM are used for.

    And I enjoyed seeing parsimony analysis as looking for the minimum of reversals. A while ago I read a wordy philosophical tract because it showed just that parsimony leads to fewest reversals in knowledge. And here biologists knew that all along! That removed one of the two facts philosophers have uniquely taught me. Philosophy 1 – Science 10^6, or thereabouts. :-)

  6. #6 David Marjanovi?
    March 3, 2007

    Where was the article published, please?

    This week’s Science.

    And I enjoyed seeing parsimony analysis as looking for the minimum of reversals.

    No, the minimum of homoplasy = convergence + reversals (reversals being convergence on an ancestor).