Field of Science

Where Do You Put Your Camels?

A dromedary Camelus dromedarius dares you to say what you make of it. Copyright John O'Neill.

In any discussion of the conflicts that may exist between morphological and molecular data in phylogenetic analysis, hippos and whales are bound to come up sooner or later. The claim in the late 1990s that these are each other's closest living relatives (and hence, that whales are nested fairly deeply within the artiodactyls, or even-toed hoofed mammals) was greeted with amazement, incredulity and more than a little skepticism. The story even caught the interest of the general public through news stories like this one, meaning that many people who are otherwise unfamiliar with the trivia of mammalian phylogeny may have picked up this detail. Since then, the whale-hippo relationship has been tested, re-tested and examined again, using every data source available. But the insertion of the whales was not the only way that molecular data mixed up the artiodactyl family tree. There was also the question of where one put the camels.

Based on anatomical data, it had previously been generally agreed that camelids (including camels and llamas) were most closely related to the ruminants, the group including such artiodactyls such as cattle, deer or giraffes. Both camelids and ruminants regurgitate cud pellets from the stomach back to the mouth in order to break their food more efficiently*, and both camelids and ruminants have a stomach divided into chambers with food only travelling to the rear section of the stomach after it has been re-chewed. They do differ in that whereas ruminants have the stomach divided into four distinct chambers, camelids only possess three; the rear two chambers (the abomasum and omasum) are not clearly differentiated in camelids. They were also united by features of the dentition, such as the presence of distinctly crescent-shaped cusps on the rear teeth. This latter feature lead the camelid+ruminant grouping to commonly be referred to as the Selenodontia (the 'moon-teeth').

*Yes, giraffes do chew cud. Yes, the cud does travel all the way between the stomach and the mouth each time.

However, the advent of molecular analyses cast doubt on this long-accepted arrangement. Instead of supporting the expected Selenodontia clade, molecular analyses placed camelids as the sister group to all other artiodactyls, with the ruminants instead being sister to the whales+hippos clade (with a pigs+peccaries clade the next clade out). This implied that the shared features of camelids and ruminants had arisen convergently (or else all other artiodactyls had reverted to a state considered more plesiomorphic for the group as a whole). In support for such a proposition, one might point to the ecological similarities in play. Camelids and ruminants are more specialist browsers than the non-selenodont artiodactyls, which are commonly more omnivorous (pigs and peccaries) or even carnivorous (whales).

Alternative phylogenies of artiodactyls, based on morphological (left) and molecular (right) data, from Spaulding et al (2009).

However, even if one is willing to credit that the 'selenodont' characters may have been the result of similar dietary pressures, one must also consider the issue that there are a number of fossil artiodactyl groups with selenodont or quasi-selenodont features. Examples of these include the Protoceratidae, a North American group that has commonly made some sort of appearance in popular books on fossil animals due to the weird home arrangements of some species, and the semi-bipedal Anoplotherium. In an influential morphological study of artiodactyl relationships, Gentry & Hooker (1988) referred to the possibility that some of these groups might be members of the selenodont stem, sitting outside the exclusive camelid+ruminant clade. Obviously, if selenodonts were not monophyletic, fossil 'selenodonts' might be aligned to either camelids or ruminants, but they couldn't be connected to both. Most studies that posited selenodont polyphyly, however, looked at living taxa only and did not consider extinct groups.

The most detailed study that I've found so far that considers the relationship between data from fossil taxa and from molecular sources in artiodactyl phylogeny is that published by Spaulding et al. (2009). This combined analysis of both morphological and molecular data produced results that were largely concordant with the latter, generally supporting placement of camelids as the sister group to all other artiodactyls (it's worth noting, mind you, that the size of the molecular data set used was considerably larger than that for the morphology, and an analysis of their morphological data only resulted in selenodont monophyly). The various 'proto-selenodonts' were scattered to the stems of various Recent clades. Protoceratids, for instance, were associated with ruminants rather than with camelids*. There are still a number of groups that remain yet to be analysed, but they've made a start.

*Another result of their analysis that is not directly relevant to the selenodont question, but cannot go unremarked upon, is that their tree indicates that Andrewsarchus, a lead contender for the title of largest terrestrial mammalian carnivore ever, might some sort of giant entelodont. I don't know how much I should read into this—not all of Andrewsarchus' potential relatives were included in Spaulding et al.'s analysis—but that's the sort of result that one just wants to be true.


Gentry, A. W., & J. J. Hooker. 1988. The phylogeny of the Artiodactyla. In: Benton, M. J. (ed.) The Phylogeny and Classification of the Tetrapods vol 2. Mammals pp. 235–272. Clarendon Press: Oxford.

Spaulding, M., M. A. O’Leary & J. Gatesy. 2009. Relationships of Cetacea (Artiodactyla) among mammals: increased taxon sampling alters interpretations of key fossils and character evolution. PLoS ONE 4(9): e7062. doi:10.1371/journal.pone.0007062


  1. Vote of thanks from the mammal enthusiasts for this post!
    Worth noting that Darren Naish had several posts, back in 2009, on Andrewsarchus and what the state of play about its relationships was at that time (typing "Andrewsarchus" into Firefox's search engine generated 4 references).
    Protoceratids are, I think, only one of the extinct Artidactyl groups whose affinities, as between Camels and Ruminants, have been... problematic.

  2. Is there any principled way to weight molecular versus morphological characters? Saying that one change in morphological state is counts as as much change as a SNP sounds the height of arbitrarinesss, but what wouldn't?

  3. Most weighted phylogenetic studies these days use some variant on implied weighting, which calculates the weighting based on the distribution of character states within the data set itself (so, for instance, characters that show a fair degree of correlation in their state distribution to others may be weighted higher than characters that seem to be distributed fairly randomly). This avoids any a priori biases about what one might expect to be important, but does depend on your data-set not being somehow inherently biased itself. For instance, to use the post as an example, I've wondered if 'chambered stomach' and 'crescentic molars' might end up weighted highly because they correlate with each other, even if they only correlate because they're induced by the same selective pressures.

    There's also the question of just how much of a weighting gradient one should introduce, and as far as I know there's no really objective way as yet of doing that. A common work-around is to run the analysis at a number of different weighting settings and see how that changes the results.


Markup Key:
- <b>bold</b> = bold
- <i>italic</i> = italic
- <a href="">FoS</a> = FoS