Evo-devo of mammalian molars

Blogging on Peer-Reviewed Research

I’ve written a long introduction to the work I’m about to describe, but here’s the short summary: the parts of organisms are interlinked by what has historically been called laws of correlation, which are basically sets of rules that define the relationship between different characters. An individual attribute is not independent of all others: vary one feature, and as Darwin said, “other modifications, often of the most unexpected nature, will ensue”.

Now here’s a beautiful example: the regulation of the growth of mammalian molars. Teeth have long been a useful tool in systematics—especially in mammals, they are diverse, they have important functional roles, and they preserve well. They also show distinct morphological patterns, with incisors, canines, premolars, and molars arranged along the jaw, and species-specific variations within each of those tooth types. Here, for example, is the lower jaw of a fox. Look at the different kinds of teeth, and in particular, look at the differences within just the molars.

This example — the lower teeth of a grey fox — shows the three-molar dental phenotype typical of placentals.

Note that in this animal, there are three molars (the usual number for most mammals, although there are exceptions), and that the frontmost molar, M1, is the largest, M2 is the second largest, and M3, the backmost molar, is the smallest. This won’t always be the case! Some mammals have a larger M3, and others may have three molars of roughly equal size. What rules regulate the relative size of the various molars, and are there any consistent rules that operate across different species?

To answer those questions, we need to look at how the molars develop, which is exactly what Kavanagh et al. have done.

The development of a tooth can be observed in vivo and in vitro by the formation of enamel knots. Enamel knots are small clumps of tissue that condense at the site of each developing tooth, and they are both spots where the tooth enamel will be secreted, and signaling centers: they secrete molecular signals that both recruit and repel the formation of new enamel knots, one for each cusp of the tooth.

The molar enamel knots develop sequentially, from front to back, so first M1 forms, then M2, and finally M3. This can be seen in the microscope: in the photo of embryonic mouse jaws below, on day 14 you see an oval blob (that’s M1) with a small tail to the right, or back of the jaw. By day 16, the tail of tooth tissue has developed into an independent tooth of its own, M2. M3 (not shown) will develop to the right of M2 over a week later.

a, Mouse molars develop sequentially, and the dental lamina extending posteriorly (black arrowhead) from the developing M1 gives rise to M2 at day 16. M3 forms (white arrowhead) posterior to M2 about ten days later. b, In comparison with the situation in vivo, M1 development proceeds normally in vitro and the secondary enamel knots form at day 16 (bright green). In contrast, M2 initiation is delayed in vitro. We suggest that this delay is due to a decrease in mesenchymally secreted activators (blue arrows), whereas M1 continues to inhibit M2 normally. To test this, we cut the posterior tail that forms M2 from M1. Anterior is towards the left. Scale bar, 0.5 mm.

The photos and the first set of diagrams illustrate what happens in the intact embryo. If the little scrap of jaw tissue is snipped out and put in a dish to grow in vitro, something different happens: M1 forms, but M2 is greatly delayed. This suggests that there is an activating signal from the tooth environment (the small blue arrows in the cartoons) that promotes tooth formation. In the dish, those signals are diminished, and so M2 forms much, much more slowly.

The bottom diagram illustrates another experiment. Like the in vitro intact experiment, culture the fragment of jaw in a dish … but cut it carefully to separate that tail of tooth tissue from M1. Now M2 develops on schedule! What this tells you is that M1 is secreting an inhibitor (the orange bar in the diagram) to suppress the formation of an adjacent tooth. In its normal environment, the levels of activator are high enough to overcome that suppression; in vitro, the activator is reduced and the inhibitor dominates; in vitro with the M2 primordium separated from M1, the inhibitor’s effect is reduced and M2 can form in response to the lower activator levels.

This can be tested with more quantitative detail. These mice were marked with a green fluorescent protein attached to the Sonic hedgehog (Shh) promoter — any cell that turns on the Shh gene glows green. The enamel knots express Shh, so you can observe the pattern of knot formation day by day. The photos below show two explanted embryonic jaws; the top row is intact, and the second row has been cut to isolate M1 and its inhibitory effect. The growth of both M2 and M3 were accelerated when M1 was cut away, as shown in the graphs that illustrate the frequency of appearance of the second and third molars in cut and uncut explants.

a, The epifluorescence of cultured ShhGFP teeth allows daily monitoring of the enamel knots to test whether cutting the posterior tail (dashed line) accelerates molar initiation (white arrowheads). b, c, Cumulative percentage curves show that, in comparison with the intact explants (solid lines), the cut explants (dashed lines) at day 14 (blue) and day 13 (red) have an accelerated initiation of M2 (b) and M3 (c). In vivo M2 and M3 initiation times are marked with dotted vertical lines. Mann?Whitney U-tests on M2 and M3 age differences between intact and cut explants after 12 days of culture are all P ≤ 0.001. n = 28 and n = 25 for day 14 intact and cut explants, respectively, and n = 15 and n = 10 for day 13 intact and cut explants, respectively. Scale bar, 0.5 mm.

What this suggests is a simple model for how molar growth is regulated. There is an activator molecule which promotes growth of the enamel knots, and which is secreted from the surrounding tissue. There is also an inhibitor molecule that is secreted by the enamel knots and suppresses the formation of adjacent knots. By adjusting the relative potency of these two molecules, the organism can achieve a range of relative tooth sizes. If the activator effectiveness is increased while the inhibitor is held constant, we’d expect the teeth to get larger and in particular, for M2 and M3 to become relatively larger. If the activator is held constant and the inhibitor is strengthened, M1 will stay the same size, but M2 and M3 will become increasingly smaller in proportion. It’s like having two dials or verniers, one regulating overall toothsize and another regulating M2 and M3′s size relative to M1. You’d be able to generate a wide range of tooth morphologies with just two regulators.

This can be assayed quantitatively. In (a) below, the effect of removing inhibition (by cutting the tissue to isolate M1) is measured: reduced inhibition leads to more equal sized posterior molars.

In (b), the investigators modulated the degree of inhibition by controlling the timing of the explants and cutting; the graph plots the ratio of M3/M1 size against the ratio of M2/M1 size. Tooth morphologies where M1 is much larger than the other two would be plotted on the bottom left of the chart, and as the M2 and M3 molars get closer in size to M1, the points would rise … and this would correspond to decreasing the inhibitor. The interesting thing about the developmental data plotted there is that the data all falls on a straight line. The formula for tooth size of all three molars seems to be simply described by the relative effective concentrations of two factors, a, the activator, and i, the inhibitor. The relative size of each molar can be predicted by one simple formula:

1 + [(a-i)/i](x -1)

where a and i are the relative concentrations of the activator and inhibitor, and x is the position in the tooth row. This formula predicts that M2 will always have an area equal to one third the total molar area.

a, Removal of inhibition results in earlier initiation and more equal-sized posterior molars. b, Changes in inhibition provide a trajectory through the morphospace in which more equal-sized molars are found with low inhibition (day 13 explants; error bars denote s.e.m). In contrast, increasing inhibition (day 14 explants) leads to smaller posterior molars and eventually the lack of M3. c, The molar proportions of 29 species of murine rodents (black circles; Mus musculus is marked with an open circle) fall close to the experimental data (crosses and dashed line). We note the lack of M3 when M2 is about half the size of M1, in both the experimental and the macroevolutionary data. For the experimental data, the slope drawn through the means of day 14 and day 13 molar sizes is 1.848 and the intercept is -0.833. When the 12 cut explants without M3s (all except one were day 14) are plotted separately (b), the resulting reduced major axis regression slope is 1.519 and the intercept is -0.673. When M1 sizes just before they reach their asymptotic sizes are used, approximating the growth stage of measured M2s and M3s, the reduced major-axis regression slope is 2.024 and the intercept is -0.997. For the macroevolutionary data (c), the reduced major-axis regression slope is 2.150 and the intercept is -1.219 (r2 = 0.740).

What this suggests is a kind of morphological see-saw, pivoting on M2. As the regulators make M1 larger and larger, they also make M3 smaller and smaller. Conversely, as M3 gets larger, M1 gets smaller.

Now look at (c) above. This isn’t developmental data: it’s macroevolutionary data. The authors looked at tooth sizes in 29 different mouse species, plotted M3/M1 against M2/M1, and presto — they see the same simple relationship that was observed in the developmental experiments. These observations suggest that all mice have the same two knobs controlling a and i, the same underlying developmental mechanisms, but the morphological variation is induced by turning the two knobs to different values. We can see how evolution has tweaked and fine-tuned and diddled with these two parameters to produced the different arrangements of teeth in different species of mice.

This is exceedingly cool. We’re looking directly at the laws of correlation in the development and evolution of this one feature.

It gets even cooler, though. P. David Polly sees this simple relationship reported in mice, and charges into the collections at the Indiana University Zooarchaeology Laboratory, and starts measuring relative tooth sizes in species other than mice. The relationship holds! (Mostly, that is.)

Data from 35 species, compiled from measurements of specimens in the Indiana University Zooarchaeology Laboratory, show that Kavanagh and colleagues’ model can largely account for patterns of molar-size proportions in mammals. The white region is consistent with their model; the broken line is the relationship they predicted. The bat-eared fox and opossum fit the model despite having four molars. Three species of bear do not fit the model (in that early arrest of M3 development is required in addition to a change in a/i); nor does the horse (though there is no obvious explanation for this). The raccoons have a proportionally larger M2 than expected for species without an M3.

That’s the same M3/M1 vs. M2/M1 plot, but now done for carnivores, ungulates, primates, rodents and rabbits, marsupials and bats. The relationship predicted by Kavanagh et al. is the dotted line. The white area is the zone where the general relationship of M1>M2>M3 or M1<M2<M3 holds true; the middle point is where M1=M2=M3. Look at that — almost everything falls into the white area where the general morphology can be explained by a simple two factor model.

The exceptions are particularly interesting. Bears have something else going on: they have a larger middle molar than either M1 or M3, which doesn’t fit the model; the postulated explanation is another factor that induces a developmental arrest of M3. Horses are also odd in having a smaller M2 than either M1 or M3, a circumstance that is not yet explained.

This is a wonderfully satisfying model. It explains most of the data, and suggests relatively simple generalizable regulatory mechanisms, and at the same time, it opens interesting new questions. What is the molecular identity of a and i? The authors speculate a bit about known signaling molecules and inhibitors expressed in the developing teeth: BMPs, Activin A, and ectodysplasin. There are also those titillating exceptions that could probe the rule. How did such different mammals as bears and horses modify the general rule, and is there a specific functional advantage to their differences?

Most satisfying of all, though, is seeing an aspect of morphology that shows a law-like behavior, following simple predictively useful rules across such a wide range of animal species. Pattern becomes a consequence of clean mathematical rules of form, an idea that a Cuvier could appreciate, and at the same time, we can see exactly where genetic variation and selection can step in to generate and stabilize particular patterns, with the regulation of just a few developmentally significant processes.

Kavanagh KD, Evans AR, Jernvall J (2007) Predicting evolutionary patterns of mammalian teeth from development. Nature 449:427-432.

Polly PD (2007) Development with a bite. Nature 449:413-415.


  1. #1 David Marjanovi?, OM
    October 7, 2007

    Homo sapiens (specific epithets are never capitalized) fits nicely: M1>M2>M3. IIRC chimpanzees have the opposite, which also fits.

    Is it true that the bat-eared fox (Otocyon megalotis) has four molars? That would be unique among placentals (among eutherians even), wouldn’t it?

  2. #2 David Marjanovi?, OM
    October 7, 2007

    I had understood that systematists preferred to use anatomical features without major functional importance (such as the palate) because they are less subject to misleading convergent evolution. What is the situation actually?

    The palate has no functional importance?!?

    Of course it has plenty. Just about anything has. That kills half of the argument.

    Then there’s the fact that real spandrels (braincase characters are traditionally thought to be such, though I’m not aware of an actual test of that idea) can evolve in any direction they bloody well want, including convergence. That kills the other half of the argument.

    The fashionable term these days is “total-evidence approach”, and for a good reason: it is not likely that, say, the teeth and the braincase will both converge between the same two animals; therefore, if both sets of characters indicate the same relationships, those relationships are more likely to be real. That’s why phylogenetic analyses use at least 300 characters these days, often 350 or more, and why most use at least 3 times as many characters as species or whatever, both tendencies rising. When possible, people also like to add molecular data to the mix.

  3. #3 David Marjanovi?, OM
    October 8, 2007

    By far not all mammals have four bumps on their molars. The normal (original) condition is actually to have three in a straight row, and that goes back much farther than mammals. What helps is the presence of molars that can be distinguished from premolars.

    And no, the Cambrian was no different. Look at the time involved, and look at the relationships of the animals involved.