SIV-specific cytotoxic T cells that can clear the infected cells before they go into latency, thus preventing the viral reservoir from becoming established.

It remains to be seen whether clinical studies will confirm Whitney and colleagues’ obser-vations, because substantial differences exist between SIV infection in rhesus macaques and HIV-1 infection in humans. As mentioned by the authors, the SIV dose used in their study may be much higher than the typical amount of HIV-1 involved in sexual transmission, per-haps resulting in a higher level of early viral replication. Nevertheless, the striking findings of the early seeding of the viral reservoir in mucosal and lymphoid tissues before viraemia is detected suggest that new approaches, in addition to early treatment, will be necessary to eradicate HIV-1 infection.

Kai Deng and Robert F. Siliciano are in the Department of Medicine, Johns Hopkins University School of Medicine, Baltimore, Maryland 21205, USA. R.F.S. is also at the

Howard Hughes Medical Institute, Baltimore.

e-mail: rsiliciano@jhmi.edu

1. Finzi, D. et al. Science 278, 1295–1300 (1997).2. Chun, T.-W. et al. Proc. Natl Acad. Sci. USA 94,

13193–13197 (1997).3. Finzi, D. et al. Nature Med. 5, 512–517 (1999).4. Siliciano, J. D. et al. Nature Med. 9, 727–728

(2003).5. Richman, D. D. et al. Science 323, 1304–1307 (2009).6. Whitney, J. B. et al. Nature 512, 74–77 (2014).7. Chun, T.-W. et al. Proc. Natl Acad. Sci. USA 95,

8869–8873 (1998).8. Chun, T.-W. et al. J. Infect. Dis. 195, 1762–1764

(2007).9. Hocqueloux, L. et al. J. Antimicrob. Chemother. 68,

1169–1178 (2013).10. Ananworanich, J. et al. PLoS ONE 7, e33948 (2012).11. Salgado, M. et al. Retrovirology 8, 97 (2011).12. Sáez-Cirión, A. et al. PLoS Pathogens 9, e1003211

(2013).13. Persaud, D. et al. N. Engl. J. Med. 369, 1828–1835

(2013).14. Ledford, H. Nature http://dx.doi.org/10.1038/

nature.2014.15535 (2014).15. Hill, A. L., Rosenbloom, D. I. S., Fu, F., Nowak, M. A.

& Siliciano, R. F. Proc. Natl Acad. Sci. USA http://dx.doi.org/10.1073/pnas.1406663111 (2014).

16. Hansen, S. G. et al. Nature 502, 100–104 (2013).

This article was published online on 20 July 2014.

on days 3, 7, 10 or 14 after SIV infection. Although initiating treatment on days 7, 10 and 14 significantly reduced peak plasma virus levels, treatment from day 3 completely blocked the emergence of detectable viraemia; this was also evidenced by the absence of SIV-specific antibody-based and cellular immune responses in these animals. The authors found no proviral DNA in the animals’ peripheral-blood mononuclear cells (which include CD4+ T cells) before treatment initia-tion on day 3, but pro viral DNA was already detectable in their lymph nodes and the mucosal linings of the gastrointestinal tract. This crucial finding suggests that the viral reservoir may be first seeded in the lymphoid and mucosal tissues, a result with important implications for HIV-1 eradication strategies.

Most significantly, the authors observed viral rebound in all animals after ART was stopped. This occurred even when ART that fully suppressed detectable viraemia was ini-tiated at day 3 and continued for 6 months, a treatment period that allows elimination of labile infected cells and thus reveals stable reservoirs. The observed rebound suggests that the viral reservoir is seeded surprisingly early in SIV-infected animals. However, the animals treated from day 3 showed a slightly delayed viral rebound compared with those that started ART at later times. Using a sophis-ticated model of viral dynamics, the authors show that the time to viral rebound is cor-related with total viraemia during the acute phase of infection.

These data indicate that the viral reservoir could be seeded substantially earlier than pre-viously assumed — a sobering finding that poses additional hurdles to HIV-1 eradica-tion efforts. If this evidence from SIV-infected animals reflects what happens early in HIV-1 infection in humans, it would mean that it is nearly impossible to initiate ART before viral reservoirs have seeded, because viraemia is not detectable at this point. In other words, reser-voir seeding precedes any clinical evidence of infection. However, although early treatment may not prevent reservoir seeding, it has been consistently shown to reduce the size of the latent reservoir8–10, and infected individuals who receive early treatment could have a lower barrier to cure in future eradication strategies.

Whitney and colleagues’ findings are of particular interest in light of a study last year16 reporting that a disseminated SIV infection could be cleared by vaccine-induced T-cell-based immune responses. The different out-comes of these two studies may be partly due to the fate of infected cells during acute infection. Early initiation of ART immediately stops sub-sequent new infection of susceptible cells, but does not affect the fate of cells that are already infected. A small fraction of these infected cells survive and revert back to a resting state, thereby seeding the latent reservoir (Fig. 1). By contrast, vaccinated animals have pre-existing

E V O L U T I O N

Tooth structure re-engineeredMice deficient in the EDA protein lack normal tooth features. Restoring EDA in embryonic teeth at increasing doses has now been found to recover these dental features in a stepwise pattern that mimics evolution. S A .44

Z H E -X I L U O

A fundamental connection between developmental changes and evolution has long been established1. This link

is gaining renewed emphasis2 as molecular studies shed new light on evolution by reveal-ing many genetic modifications that alter developmental processes, in turn changing an organism’s shape and structure. On page 44 of this issue, Harjunmaa et al.3 report that, by simply tinkering with the genes and signalling pathways that control the shape of develop-ing teeth, they have remade several different tooth structures in vitro. These structures draw a striking parallel with how teeth evolved from those of distant mammalian ancestors to the teeth of modern-day rodents.

Some lineages of therians (marsupial and placental mammals and their kin) that lived in the Mesozoic era, 252 million to 66 million years ago, had ‘tribosphenic’ molars4. The taller front end of the tribos-phenic lower molar, called the trigonid, had three cusps — the raised points on the crown

of the tooth — for shearing food. The lower back end, the talonid, had a basin-like surface for grinding food5,6. The trigonid and talonid of Mesozoic mammals are still recognizable in modern-day rodents, albeit in a highly modi-fied form. As rodents arose from ancestral mammals and diversified into many line-ages, cusps that were separate in the Palaeo-cene epoch 66 million to 56 million years ago7 became progressively connected by crests — which are more effective for chewing plant food — in a ‘cusp-to-crest’ dental evolution that occurred in many rodent groups8,9.

The ectodysplasin A (Eda) gene encodes a vertebrate signalling protein that is involved in the development of a wide array of struc-tures, from hair to sweat glands10. In the embryonic tooth, the EDA protein is active in enamel knots, which are signalling centres and the precursors of adult tooth structures. EDA regulates the position and size of future tooth cusps and their connecting crests8. Mice that do not express Eda lack normal cusps and crests, and instead have only basic, generalized teeth10.

3 6 | N A T U R E | V O L 5 1 2 | 7 A U G U S T 2 0 1 4

NEWS & VIEWSRESEARCH

© 2014 Macmillan Publishers Limited. All rights reserved

Harjunmaa and colleagues grew Eda-deficient teeth in vitro, and found that cusps and crests could be restored by adding EDA. They demonstrated, with the aid of computer models, that different doses of EDA alter tooth morphogenesis (the process by which structures are shaped as they develop), akin to the dental transformations that occurred as rodents evolved from their Mesozoic mamma-lian ancestors (Fig. 1). For example, the trigo-nid — the first part of the tribo sphenic molar to have evolved — is regenerated with only a small dose of EDA. However, a higher dose of EDA is required to restore the talonid, which evolved more recently5.

The cusp-to-crest morphogenesis of mouse molars is controlled by a gene network that includes genes encoding the signalling pro-teins fibroblast growth factor 3 (Fgf3; ref. 9) and sonic hedgehog (Shh)11. An increase or decrease in Fgf3 causes over- or under-development of tooth features, respectively9. Harjunmaa and co-workers found that reduc-ing the concentration of SHH in Eda-deficient teeth regenerated the ancestral features of Pal-aeocene rodents, reversing the cusp-to-crest transformation of modern-day rodents. Thus, molecular manipulations that alter tooth morphogenesis in vitro can replay evolution, either forward, to mimic the fossil record, or in reverse.

Perhaps the most exciting insight from

Harjunmaa and colleagues’ work is that ancestral structures show a more uniform response to the addition or removal of EDA or SHH than those that evolved more recently or independently in different lineages (con-vergent evolution). For example, addition of a low dose of EDA reliably restored the trigo-nid, as expected of ancestral features, which are typically evolutionarily well conserved owing to their long history. By contrast, higher doses restored the talonid in many, but not all, teeth — development of this feature was more variable in response to EDA. This is consistent with the theory6 that the talonid basin evolved convergently in different mammal lineages, but has reduced in size in some carnivoran or insectivoran mammals.

The hypoconulid is a talonid cusp in some mammals, but is enlarged and forms a sepa-rate lobe in mice. The authors found that full development of this structure requires a higher dose of EDA than does the rest of the talonid, and shows even wider variation in its response to EDA. Finally, the anteroconid in mice — another feature that arose late in rodent evolution — requires the highest EDA dose to regenerate, and shows the broadest variation when regenerated. Its position on the tooth corresponds to the ‘pseudo-talonid’ that arose in some early-divergent mammals that died out before the end of the Mesozoic. Harjunmaa and co-workers’ experiment

therefore demonstrates that modern-day mice still have the developmental potential to replicate evolutionary events that occurred long ago, in the now-extinct mammals of the Mesozoic12.

The level of EDA required to give rise to individual molar characteristics therefore seems to provide information about how robust their development is. When studying how morphogenesis drives evolution13, it will be crucial to bear in mind that the sensitivity of a particular tooth feature to EDA activity may indicate the likelihood of an evolution-ary transformation producing that feature. For example, as mentioned above, talonid-like features evolved twice — in the basal diversi-fication of modern mammals and in early-divergent groups of the Mesozoic. Variable sensitivities to gene-expression dosage and sig-nalling strength can serve as a measure of the evolutionary variability of each tooth feature, and may underpin the many convergences and reversals of tooth evolution observed in the mammalian fossil record.

Eda and Shh have varying effects on many vertebrate structures, so it can be hard to tease apart which evolutionary feature is controlled by which part of the gene network. Harjunmaa et al. have cleared this hurdle in a welcome development that brings us closer to being able to test how changes in morphogenesis affect the final shape of evolving teeth as seen in the fossil record. Genetic engineering of develop-mental processes in vitro is a fruitful way to decipher how the shape of organs or other bio-logical structures is modified by evolution. ■

Zhe-Xi Luo is in the Department of Organismal Biology and Anatomy, The University of Chicago, Chicago, Illinois 60637, USA.e-mail: zxluo@uchicago.edu

1. Gould, S. J. Ontogeny and Phylogeny (Belknap, 1985).

2. Abzhanov, A. Trends Genet. 29, 712–722 (2013).3. Harjunmaa, E. et al. Nature 512, 44–48 (2014).4. Kielan-Jaworowska, Z., Cifelli, R. L. & Luo, Z.-X.

Mammals from the Age of Dinosaurs: Origins, Evolution, and Structure (Columbia Univ. Press, 2004).

5. Crompton, A. W. Zool. J. Linn. Soc. 50 (Suppl. 1), 65–87 (1971).

6. Luo, Z.-X., Cifelli, R. L. & Kielan-Jaworowska, Z. Nature 409, 53–57 (2001).

7. Meng, J. & Wyss, A. R. J. Mamm. Evol. 8, 1–71 (2007).

8. Gomes-Rodrigues, H. et al. Nature Commun. 4, 2504 (2013).

9. Charles, C. et al. Proc. Natl Acad. Sci. USA 106, 22364–22368 (2009).

10. Mikkola, M. J. & Thesleff, I. Cytokine Growth Factor Rev. 14, 211–224 (2003).

11. Cho, S.-W. et al. Development 138, 1807–1816 (2011).

12. Luo, Z.-X., Ji, Q. & Yuan, C.-X. Nature 450, 93–97 (2007).

13. Carroll, S. B., Grenier, J. K. & Weatherbee, S. D. From DNA to Diversity: Molecular Genetics and the Evolution of Animal Design 2nd edn (Blackwell, 2005).

This article was published online on 30 July 2014.

Figure 1 | Reconstructing tooth evolution in vitro. a, As mammals evolved, their molars (one indicated in the jaw) became ever more complex, because extra tooth features evolved over time. Structures called trigonids (dark grey) evolved first, in early mammals such as the Mesozoic symmetrodonts, followed by talonids (light grey) in a group of Mesozoic therians called cladotherians. Hypoconulids (blue) evolved in Mesozoic therians, and anteroconids (yellow) in advanced rodents. The anteroconids are similar in structure to pseudo-talonids, which evolved separately (convergently) in pseudo-tribosphenic teeth in an early-divergent clade of mammals (the pseudo-talonid is also indicated in yellow). b, Deletion of the Eda gene in mouse embryos results in the loss of all of these tooth features. Harjunmaa et al.3 show in vitro that addition of the EDA protein to embryonic teeth from Eda-deficient mice in increasing doses can replay the steps of evolution. Furthermore, features that evolved longer ago respond in a less variable manner than features that evolved more recently.

Evolutiona

Developmentb

Mammalia

Rodentia

Mesozoicsymmetrodont

Pseudo-tribosphenidan

Mesozoiccladotherian

Mesozoictherian

Palaeocenerodent

Extant mouse

Talonid

Hypoconulid

Remade with least EDA, least variable

Trigonid

Remade with more EDA, variable

Convergent feature remade with most EDA,highly variable

Remade with more EDA,more variable

7 A U G U S T 2 0 1 4 | V O L 5 1 2 | N A T U R E | 3 7

NEWS & VIEWS RESEARCH

© 2014 Macmillan Publishers Limited. All rights reserved

ARTICLEdoi:10.1038/nature13613

Replaying evolutionary transitions fromthe dental fossil recordEnni Harjunmaa1, Kerstin Seidel2,3, Teemu Hakkinen1, Elodie Renvoise1, Ian J. Corfe1, Aki Kallonen4, Zhao-Qun Zhang5,Alistair R. Evans6,7, Marja L. Mikkola1, Isaac Salazar-Ciudad1,8, Ophir D. Klein2,3,9,10 & Jukka Jernvall1

The evolutionary relationships of extinct species are ascertained primarily through the analysis of morphological char-acters. Character inter-dependencies can have a substantial effect on evolutionary interpretations, but the develop-mental underpinnings of character inter-dependence remain obscure because experiments frequently do not providedetailed resolution ofmorphological characters. Herewe show experimentally and computationally how gradualmodi-fication of development differentially affects characters in themouse dentition.We found that intermediate phenotypescould beproducedbygradually adding ectodysplasinA (EDA)protein in culture to tooth explants carrying anullmutationin the tooth-patterning gene Eda. By identifying development-based character inter-dependencies, we show how topredictmorphological patterns of teeth amongmammalian species. Finally, in vivo inhibitionof sonichedgehog signallingin Eda null teeth enabled us to reproduce characters deep in the rodent ancestry. Taken together, evolutionarily infor-mative transitions can be experimentally reproduced, thereby providing development-based expectations for character-state transitions used in evolutionary studies.

In the case of extinct mammals, a large number of dental features areused as characters in phylogenetic analyses1–4, and these charactersoftenprovide the key evidence for evolutionary inferences due to the pre-ponderance of teeth in the fossil record. For reliable phylogenetic infer-ences, charactershavebeen typically considered to be independent fromeach other5–8. Although developmental factors can make charactersdependent8–11, thorough analyses of the influence of development oncharacter state changes are lacking. To approximate changes relevantto evolutionary transitions, experiments that tunemorphology gradu-ally are required. These kinds of experiments are also useful to evaluatehow, andwhether, continuous changes in underlying developmental orgenetic parameters map to continuous changes in the phenotype12–14.Here we investigated whether gradual alterations of tooth develop-

ment canproduce gradual changes in the phenotype, andwhether thesechanges reflect knownevolutionary transitions.We focused on the devel-opment of the rodent dentition, using mice carrying a spontaneouslyoccurring nullmutation in ectodysplasin (Eda) as a starting point. Thismutation was chosen because the effects of Eda on tooth morphologyare relatively subtle, causing simplification of dentalmorphologywith-out complete loss of teeth10,15, however, theEdamutation causes changesin many characters and is thus highly informative10.

Experimental tuning of morphologyWereasoned that, to approximate evolutionary transitions, fine-tuningof EDA signallingwould be required.We tracked gradual changes dur-ing development by crossingEdanullmicewithmice that express greenfluorescent protein (GFP) from the Shh locus (hereafter called ShhGFPmice16). The epifluorescence of ShhGFP mice can be used to monitortooth cuspdevelopment because Shh is initially expressed in the enamelknots,which are the epithelial signalling centres that format thepositionsof future cusps17. Later during differentiation, Shh expression spreads

throughout the inner enamel epithelium, enabling the visualization ofthe overall crown shape.First, we used EDA protein in culture at increasing concentrations

(n5 9 to 16 in each group, Supplementary Table 1, Methods) to testwhether theEdanullmorphologycouldbeengineered togradually resem-ble wild-type morphology. We cultured first lower molars starting atembryonic day 13, just before crown formation begins, and EDA pro-teinwas administered into the culturemedia at days zero and two. Thistreatment scheme restored EDA signalling during the period of firstmolar cusp patterning. At this stage, Eda is thought to regulate the sizeand signalling of enamel knots10, which in turn give rise to tooth cusps.The EDAprotein treatments restored thewild-typemouse cusp pat-

tern inculture (Fig. 1a), in agreementwithprevious experiments18,19.Wenext examined themode of cusp appearance indetail by analysingdailytime-lapse images of the cultured teeth. The results showed that increas-ing dosage of EDAcaused aheterochronic shift in cusp initiation (Fig. 1a).Specifically, some of the cusps were initiated earlier (predisplaced) asEDAconcentrationwas increased (Fig. 1a, SupplementaryTable 1). Fur-thermore, the time-lapse data showed that increasingEDAconcentrationenlarged the primary enamel knot, which in turn increased thenumberof cusps (Fig. 1b, Extended Data Fig. 1). The link between the primaryenamel knot size and cusp number (Fig. 1b) indicates that the overallsize of the tooth crown needs to reach certain thresholds to accommo-date additional cusps. From a developmental signalling point of view, aheterometric20 change in the dosageof EDAsignalling can lead to a hete-rochronic shift in timing of cusp initiation.

Computational modelling of patterningComputationalmodellinghasbeenused to simulate tooth shapedevelop-ment andevolution14,21, and thenewdevelopmental data thatweobtainedallowedus to linkmodels and experiments in unprecedenteddetail (see

1Developmental Biology Program, Institute of Biotechnology, University of Helsinki, P.O. Box 56, FIN-00014 Helsinki, Finland. 2Program in Craniofacial and Mesenchymal Biology, University of California,San Francisco, San Francisco, California 94114, USA. 3Department of Orofacial Sciences, University of California, San Francisco, San Francisco, California 94114, USA. 4Division of Materials Physics,Department of Physics, University of Helsinki, P.O. Box 64, FIN-00014Helsinki, Finland. 5Key Laboratory of Evolutionary Systematics of Vertebrates, Institute of Vertebrate Paleontology and Paleoanthropology,Chinese Academyof Sciences, Beijing 100044, China. 6School of Biological Sciences,MonashUniversity, Victoria 3800, Australia. 7Geosciences,MuseumVictoria, GPOBox 666,Melbourne, Victoria 3001,Australia. 8Genomics, Bioinformatics and Evolution Group. Department de Genetica i Microbiologia, Universitat Autonoma de Barcelona, Cerdanyola del Valles 08193, Spain. 9Department of Pediatrics,University of California, San Francisco, San Francisco, California 94114, USA. 10Institute for Human Genetics, University of California, San Francisco, San Francisco, California 94114, USA.

4 4 | N A T U R E | V O L 5 1 2 | 7 A U G U S T 2 0 1 4

Macmillan Publishers Limited. All rights reserved©2014

Methods). To model the experimental transitions, we implemented amorphodynamic model, which integrates signalling and tissue growthto simulate tooth development21, in the newToothMaker interface (Ex-tended Data Fig. 2). First we modelled a wild-type mouse tooth mor-phology corresponding to our cultured teeth (see Methods). Then, byprogressively adjusting (mutating) each parameter separately, we sim-ulated the effects of gradual changes in signalling (ExtendedData Fig. 3).The in silico simulation reproduced the fusion of cusps both on thetalonid and on the trigonid of the Eda null model, as well as rescue ofthe separate cusps observed in the in vitro experiment (Fig. 2a, ExtendedData Fig. 4). Moreover, as predicted from the experimental observa-tions, the full range of transitions from fused trigonid to separate cuspswas replicated by varying the activator parameter that induces the for-mation of enamel knots (Fig. 1a, Extended Data Fig. 4).The comparable patterns in the experiments (Fig. 1) and the model-

ling (Fig. 2) underscore the dynamic nature of tooth shape development,

and indicate that the identification of cusp homologies should rely ontopological correspondences rather thanunique, cusp specific geneexpres-sionpatterns. Furthermore, both themodel simulations and the experi-ments showed a rapid phenotypic response at low levels of activator(Fig. 2b) andEDA signalling (Fig. 2c), respectively, with smaller changesobserved at higher levels of signalling. These results imply a potentialdisjunction between rates of evolutionmeasured in the phenotype andin gene expression level, although by varyingmore than one parameterat a time, a multivariate linear relationship between expression levelsand phenotypes remains possible12,13,19.

Detailed analyses of character statesTo examine the full range of morphologies produced in the culturedexplants, we tabulated the character states comparable to the ones usedin evolutionary studies for each crown region (see Methods and Sup-plementaryTable 1). Although the tooth culturing systemdoes not pro-duce mineralized features, we were able to tabulate the presence orabsence of cusps (Fig. 3a) and relative height of the talonid (Fig. 3b)with high resolution.First, the trigonid cusps, the protoconid and themetaconid, are fre-

quently fused in Eda null teeth (40% of explants, Fig. 3c, Supplemen-tary Table 1). The separation of the protoconid and the metaconid in2.5 ngml21 EDA treated teeth reflects a transition that would predatethe evolution of the pretribosphenic mammalian pattern11,21.Next, the talonid, which in Eda null teeth is a shallow shelf lacking

well-defined cusps, was already affected in the lowest, 2.5 ngml21 EDAtreatments by an increase inheight, and in the second lowest 10 ngml21

treatments by acquisition of additional cusps (Fig. 3c, SupplementaryTable 1).These treatments, however, causedpolymorphic effects (Fig. 3c).In 47%of10 ngml21 explants, the shallow shelf gave rise to a single cusp,whereas in the remaining explants two distinct cusps formed. These twotalonid cusps correspond to the hypoconid and the entoconid cusps inwild-type teeth, andbothcusps formedconsistently startingat100ngml21

EDA (Fig. 3c).During the evolution of tribospheny in Mesozoic mammals, the

functional, three cusped talonid was added to the posterior end of thetrigonid11,22. This originated inTriassicnon-mammalian synapsidswherea single cusp, often located on the cingulid, was appended posteriorlyto the basal three-cusped trigonidmorphology22,23. Although the hypo-conid is generally agreed to have evolved before the entoconid22, our datado not allow determination of whether the single cusp in the talonid of

b

2

3

4

5

6

Cus

ps (n

) at d

ay 7

6,000 10,000 14,0002,000 18,000

Primary enamel knot size (μm2) at day 2

Day

s in

cul

ture

Eda null 10 50 WT

2

3

4

5

6

7

Mouse strain and treatment (ng ml-1)a

Figure 1 | Gradual dosage effects of EDA on Eda null mutant first lowermolars (m1). a, ShhGFP3 Eda null tooth development is rescued by EDA,with higher concentrations reproducing wild type (WT) development (Edanull:n5 15; 10 ngml21:n5 15; 50 ngml21:n5 13;WT:n5 16; all teeth listedin Supplementary Table 1: n5 113). Initiation of different parts of the toothcrown is shifted earlier with higher EDA concentrations, such as theanteroconid (black arrowheads) and the hypoconulid (white arrowheads).b, Primary enamel knot size at culture day 2 predicts the number of cusps atday 7. Eda null teeth treated with 10 ngml21 (open circles) and 50ngml21

(open diamonds) fill in the phenotypic gap between the Eda null (black circles)and wild-type (black diamonds) teeth. Anterior is towards the left in a. Scalebar, 500mm.

0.2 0.6 1.0 1.40.4 1.2 1.60.8

b

c

0.1 0.2 0.4 1.6Model parameter value (Act)

14

12

10

8

6

4

a

Sim

ulat

ion

stag

e (×

103 ) Cus

ps (n

)

Model parameter value (Act)

EDA treatment (ng ml-1)

Cus

ps (n

)

40 60 WT1,00080 1002002

3

4

5

6

4

3

2

1

5

Figure 2 | Computational modelling of gradual changes in signalling oncusp patterns. a, Computer simulations using ToothMaker (Extended DataFig. 2) of first lower molar development show appearance of cusp areas(red colour). Larger values of activator (Act) parameter increase cusps.b, Modelled wild-typemouse pattern is largely retained untilAct5 0.5. SmallerAct values quickly reduce cusps resulting in modelled teeth that resemble Edanull teeth. c, Tabulated data from culture experiments (Supplementary Table 1)show an abrupt change in cusps at low levels of EDA protein (2.5 to10 ngml21), similar to modelling data (b). Increasing EDA concentrationdoes not increase cusps beyond wild type (WT). Error bars denote s.d.Anterior is towards the left in a.

ARTICLE RESEARCH

7 A U G U S T 2 0 1 4 | V O L 5 1 2 | N A T U R E | 4 5

Macmillan Publishers Limited. All rights reserved©2014

cultured teeth is the hypoconid or the entoconid. The central locationof the single talonid cusp in our cultured teeth (Figs 1a and 3a) does,however, suggest that the effects of EDA, at least partially, mimic theearly steps of talonid evolution.Finally, the anteroconid and the hypoconulid appear already at 10–

50ngml21EDA,butunlike invivo, theirwild-type character states remainpolymorphic in cultured molars irrespective of treatments (shown byhatched colouring in Fig. 3c). Evolutionarily, these crown features appearin early rodents and are present in the basal murines2,3,24,25. The hypo-conulid, however, has been lost inmanymurine lineages, and this evo-lutionary lability appears to be reflected in the developmental data.Taken together, these data show that even though dental characters

used in evolutionary analyses may be highly pleiotropic, as shownpreviously10, transformations of character states can occur at differentthresholds of signalling (Fig. 3c). In contrast to the relatively robust trig-onid cusps, large variation in talonid structure can be reproduced bysmall changes of EDA signalling. This has major implications for theevolution of tribosphenicmammals that are diagnosed by their derivedtalonid features1,11,22, as discussed below.

Testing the experimental predictionsTo test the development-based predictions of evolutionary patterns inthe talonid, we examined the link between talonid height and cusp num-ber across mammalian species.We first tabulated relative talonid height(as percentage of the trigonid height) and cusp numbers in the entireposterior region of the tooth (the entoconid, the hypoconid and thehypoconulid; see Methods). The results show that talonid height andcuspnumber are linkeddevelopmentally (Fig. 4a, ExtendedDataTable 1),even though these traits are typically treated as independent charactersin evolutionary analyses. In the macroevolutionary context of rodents,the patterns obtained in the experiments appear to bridge the derivedstate we analysed in 35 species of extant murine rodents, already pres-ent in the extinctMiocene early murines Potwarmus andAntemus24,25,and the basal morphology found in Tribosphenomys minutus, a Paleo-cenemammal that is considered to be a basal rodentiaform2, or the imme-diate sister taxon of Glires4 (Fig. 4a, Extended Data Table 1).Because our experimental morphologies extend even beyond those

found in rodents and towards further reduced talonids (Fig. 4a), wealso measured talonid height and cusp number in 32 species of extantcarnivorans.The first lowermolar, or carnassial, of carnivorans showsargu-ably the fullest range of talonidmorphologies among extantmammals26.The correlated change in carnivoran talonid height and cusp number(Fig. 4b, Extended Data Tables 2 and 3) is reminiscent of the patternsfound in experiments on the mouse (Fig. 4a). Furthermore, this rela-tionship between talonid height and morphology was retained whenwe replaced cuspnumberwith talonid complexity using orientationpatchcount (OPC, Fig. 4c). In contrast, the trigonid morphology remainedrelatively constrained (Extended Data Table 2). OPC, which is calcu-lated as the number of discrete surfaces distinguished by differences in

orientation, has been shown to increase across the dietary spectrumfrom carnivores to omnivores to herbivores in extant mammals27,28.Therefore, even though carnivoran dental diversity is driven by ecolo-gical and functional factors, development may have an influence onwhich parts undergo adaptations more easily.Taken together, the same developmental cascade, starting from the

trigonid, may have contributed both to the initial evolution of the tal-onid at the base of mammalian evolution and to dental morphologicaldiversification during mammalian radiations. To a lesser degree, thesame pattern may hold for the anterior end of the teeth, and it is con-ceivable that an analogous developmental cascade to the one that pro-duced the talonid also produced the anterior expansion of the crownin pseudo-tribosphenicmammals29. A cascading system of activation–inhibition between teeth has been proposed to regulate molar propor-tions inmammals30, and in general,muchof the evolutionary history ofmammaliandentitionsmay consist of tinkering31with this pre-existingdevelopmental program.

Retrieving ancestral character statesDespite the overall agreement between experimental and evolutionarypatterns, there were small but important differences when comparedwith thedetails of rodent evolution.Most notably, theEdanull teethhadfused cusps, which differs from what is found in basal rodentiaforms

c

a b010 1 2 010 21 01 2 21 0 2 2 11 2 2 2 0 2 2 2 1 0 1 2

Eda null 10 50 100 500 WT

Eda null 10 50 100 500 WT

Eda null 10 50 100 500 WT

Eda null 10 50 100 500 WT

Eda null 10 50 100 500 WT

Talonid cuspsTrigonid cuspsAnteroconid cusps Hypoconulid Talonid height

Characters 210

Figure 3 | Differential sensitivities of tooth crown regions to EDA. a, Rangeof tooth morphologies and cuspal character states tabulated at the end of thecultures (day 131 7). Character state numbers above first lower molar imagescorrespond to the number of cusps present in the respective region of the crown(Supplementary Table 1). The first trait is for the anteroconid, the second forthe trigonid, the third for the talonid and the fourth for the hypoconulid.

b, Talonid height characters are tabulated as the height of the talonid (whitearrowhead) relative to the trigonid (black arrowhead). c, Differentialsensitivities of main regions of tooth crown to EDA show how different parts ofthe crown have varying trait sensitivities to EDA. For the full range of data, seeSupplementary Table 1.

0.2 0.4 0.6 0.8 1.00

Talo

nid

cusp

s (n

) 4

3

2

1

0

Talonid height

a b

Talo

nid

OP

C

c

Talonid height Talonid height

30

20

10

00.2 0.4 0.6 0.8 1.000.2 0.4 0.6 0.8 1.00

Talo

nid

cusp

s (n

) 4

3

2

1

0

Figure 4 | Testing developmental predictions on evolutionary patterns.a, Relative height of the m1 talonid (measured as percentage of the trigonidheight) in each treatment (open circles, error bars denote s.e.m.) correlates withthe number of cusps in the talonid. The experimental data bridges thecorresponding values formurine rodent species (n5 35, black circlewith s.e.m)and for a basal rodentiaform Tribosphenomys minutus (black diamond). Thereduced major axis regression slope for the experimental data includinguntreated Eda null and wild-type teeth is 3.30 and the intercept is 21.79(r25 0.891) and the corresponding values for only explants with EDA are 4.22and21.63 (r25 0.946). We note that these slopes can be consideredunderestimates due to the variably present hypoconulid cusps in culturedwild-type mouse teeth. b, c, The first lower molars of carnivoran species(n5 32) show correlated changes in the talonid height and cusp number(b) and the talonid height and talonid complexity (c)measured usingOPC. Thereduced major-axis regression slope for the graph in b is 6.53 and the interceptis21.38 (r25 0.594), and for the graph in c it is 48.24 and the intercept is210.93 (r25 0.562). For data and details see Methods and Extended DataTables 1–3.

RESEARCH ARTICLE

4 6 | N A T U R E | V O L 5 1 2 | 7 A U G U S T 2 0 1 4

Macmillan Publishers Limited. All rights reserved©2014

and close relatives2–4. This difference in morphology indicated that, inaddition to EDA signalling, other pathways needed to be adjusted inorder to produce the evolutionarily basal morphologies. To addressthis issue, we considered reducing the required cusp spacing, whichcan be experimentally adjusted beyond the normal mouse pattern bymodulating multiple signalling pathways19. We therefore next set outto engineermouse teeth that would have additional characters of basalrodents (see Methods).First, we modelled the reduction in the cusp spacing by decreasing

inhibition in the simulated Eda null teeth, which resulted in formationof multiple cusps (Extended Data Fig. 5). Next, because SHH has beenshown to inhibit cusp formation by regulating cusp spacing19,32, wecultured Eda null samples with a SHH inhibitor, thereby inhibiting theinhibitor. This treatment also circumvents the tendency of EDA to causethe formation of crests or lophs between cusps10,19,33, which are foundin evolutionarily derived rodents. The experimental results validatedthe in silicomodel simulations: inhibition of SHHsignalling in culturedEda null teeth caused the development of more distinct cusps, withouteliminating evolutionarily basal features of Eda null teeth such as theheight difference between trigonid and talonid (Fig. 5a, Extended DataFig. 6).To push the experimental system further and to retrieve features pres-

ent in the basal taxonTribosphenomys2, we inhibited SHH in developingEdanull teeth in vivo (seeMethods).Tribosphenomys cusps are colum-nar andwell separated, lacking a crest called themetalophid (the trigonidwall) that connects cusps together. The loss of themetalophid has beenlinked to the basal rodentiaforms2, but it has reappeared inmany rodentclades, including murines2,3,24,25. Our in vivo engineered tooth shapesshowed columnar and laterally separated cusps without themetalophid

(Fig. 5b), amorphology visible also at the enamel–dentin junction (Ex-tended Data Fig. 7). Although the effects of SHH inhibition were vari-able, the trough separating the protoconid andmetaconid approximatedthe pattern found in Tribosphenomys (Fig. 5c). These results indicatethat, with a relatively small number of developmental changes, mouseteeth can be engineered to express evolutionarily basal traits.

ConclusionsOur results demonstrate that many of the step-wise transitions thatare widespread in the fossil record of mammalian teeth are reproduc-ible experimentally.Whereas our results suggest that several, if not themajority, of dental traits are developmentally linked10, individual char-acters may respond to different levels of the same signal. These thresh-oldsmaywell underlie differentmorphological gradients that havebeenidentified along the tooth row34. Moreover, trait thresholds may affectevolutionary rates of individual traits differently. Suchdata in turn shouldbe useful when combinedwith analyses of character correlations35, andin weighting or ordering characters, or objectively assigning transitionweightswithincharacters coded tominimize characterdependencyeffectsin phylogenetic analyses8. Other developmental factors and signallingpathways may influence traits in other ways, but we predict that thegeneral pattern of results will hold as long as the factors affect the sig-nalling dynamics of the enamel knots. Finally, as recently proposed forthe evolution of bird and non-avian dinosaur skulls36, developmentaldata can suggest novel insights into the processes underlying hetero-chrony. A bettermechanistic basis for heterochronywill help to explainchanges in evolutionary rates, including prediction of intermediatemorphologies evenwhen they have not yet been recovered in the fossilrecord. In general, with advancing understanding of development, itwill be possible to experimentally test manymore of the known evolu-tionary transitions.

Online Content Methods, along with any additional Extended Data display itemsandSourceData, are available in theonline versionof thepaper; referencesuniqueto these sections appear only in the online paper.

Received 25 March; accepted 25 June 2014.Published online 30 July 2014.

1. Luo, Z. X., Cifelli, R. L. & Kielan-Jaworowska, Z. Dual origin of tribosphenicmammals. Nature 409, 53–57 (2001).

2. Meng, J. & Wyss, A. R. The morphology of Tribosphenomys (Rodentiaformes,Mammalia): Phylogenetic implications for basal Glires. J. Mamm. Evol. 8, 1–71(2001).

3. Asher, R. J. et al. Stem lagomorpha and the antiquity of Glires. Science 307,1091–1094 (2005).

4. O’Leary,M. A. et al.The placentalmammal ancestor and the post-K-Pg radiation ofplacentals. Science 339, 662–667 (2013).

5. Kluge, A. G. & Farris, J. S. Quantitative phyletics and the evolution of anurans. Syst.Zool. 18, 1–32 (1969).

6. Felsenstein, J. Maximum-likelihood and minimum-steps methods for estimatingevolutionary trees from data on discrete characters. Syst. Zool. 22, 240–249(1973).

7. Doyle, J. J. Treeswithin trees: genes and species, molecules andmorphology. Syst.Biol. 46, 537–553 (1997).

8. O’Keefe, F. R.&Wagner, P. J. Inferring and testinghypothesesof cladistic characterdependence by using character compatibility. Syst. Biol. 50, 657–675 (2001).

9. Wake, D. B. Phylogenetic implications of ontogenetic data. Geobios 22 (supp. 2),369–378 (1989).

10. Kangas, A. T., Evans, A. R., Thesleff, I. & Jernvall, J. Nonindependence ofmammalian dental characters. Nature 432, 211–214 (2004).

11. Luo, Z.-X. Transformation and diversification in early mammal evolution. Nature450, 1011–1019 (2007).

12. Polly, P. D. Developmental dynamics and G-matrices: can morphometric spacesbe used to model evolution and development? Evol. Biol. 35, 83–96 (2008).

13. Rice, S. H. Theoretical approaches to the evolution of development and geneticarchitecture. Ann. NY Acad. Sci. 1133, 67–86 (2008).

14. Salazar-Ciudad, I. & Marin-Riera, M. Adaptive dynamics under development-based genotype-phenotype maps. Nature 497, 361–364 (2013).

15. Gruneberg, H. Genes and genotypes affecting the teeth of the mouse. J. Embryol.Exp. Morphol. 14, 137–159 (1965).

16. Harfe, B. D. et al. Evidence for an expansion-based temporal Shh gradient inspecifying vertebrate digit identities. Cell 118, 517–528 (2004).

17. Jernvall, J., Keranen, S. V. E. & Thesleff, I. Evolutionarymodification of developmentinmammalian teeth: quantifying gene expression patterns and topography. Proc.Natl Acad. Sci. USA 97, 14444–14448 (2000).

Eda null Wild type

Tribosphenomys

Eda null TribosphenomysWild type

Eda null + SHH antagonist

Eda null +SHH antagonist

Days in culture2 3 4 5 6 7

b

a

c

Figure 5 | Engineeringmouse teeth to have basal character states. a,Eda nullteeth cultured with SHH antagonist (n5 9 of 11 teeth) show more andbetter separated cusps (arrowheads, compare to Fig. 1a). b, Second molarsproduced using in vivo treatment of Eda null mice with SHH antagonistparticularly show better separation of cusps compared to the Eda null teeth(n5 2 of 4 mice). Tribosphenomys minutus teeth lack crests connecting cusps(specimen V10776 shows p4–m3). c, Obliquely posterior views of molars showthe lack of the metalophid crest (arrowheads) in treated Eda null, which hasreplicated the ancestralmorphology ofT.minutus.TheTribosphenomysmolarsshown are the first (V10776, on the left) and the second molar (V10775holotype, on the right). Teeth have beenmirrored if needed to represent the leftside, anterior is towards the left in a and b and top in c. Scale bars, 500mm.

ARTICLE RESEARCH

7 A U G U S T 2 0 1 4 | V O L 5 1 2 | N A T U R E | 4 7

Macmillan Publishers Limited. All rights reserved©2014

18. Gaide, O. & Schneider, P. Permanent correction of an inherited ectodermaldysplasia with recombinant EDA. Nature Med. 9, 614–618 (2003).

19. Harjunmaa, E. et al. On the difficulty of increasing dental complexity. Nature 483,324–327 (2012).

20. Arthur, W. Evolution: a Developmental Approach (Wiley-Blackwell, 2011).21. Salazar-Ciudad, I. & Jernvall, J. A computational model of teeth and the

developmental origins of morphological variation. Nature 464, 583–586 (2010).22. Butler, P. M. Early trends in the evolution of tribosphenic molars. Biol. Rev. Camb.

Philos. Soc. 65, 529–552 (1990).23. Osborn, J. W. & Crompton, A. W. The evolution of mammalian from reptilian

dentitions. Breviora 399, 1–18 (1973).24. Wessels, W. Miocene rodent evolution and migration: Muroidea from Pakistan,

Turkey and North Africa. Geol. Ultraiectina 307, 1–290 (2009).25. Lopez Antonanzas, R. First Potwarmus from the Miocene of Saudi Arabia and the

early phylogeny ofmurines (Rodentia:Muroidea). Zool. J. Linn. Soc.156, 664–679(2009).

26. Van Valkenburgh, B. Major patterns in the history of carnivorousmammals. Annu.Rev. Earth Planet. Sci. 27, 463–493 (1999).

27. Evans, A. R., Wilson, G. P., Fortelius, M. & Jernvall, J. High-level similarity ofdentitions in carnivorans and rodents. Nature 445, 78–81 (2007).

28. Santana, S. E., Strait, S. & Dumont, E. R. The better to eat you with: functionalcorrelates of tooth structure in bats. Funct. Ecol. 25, 839–847 (2011).

29. Luo, Z. X., Ji,Q.&Yuan, C. X.Convergentdental adaptations inpseudo-tribosphenicand tribosphenic mammals. Nature 450, 93–97 (2007).

30. Kavanagh, K. D., Evans, A. R. & Jernvall, J. Predicting evolutionary patterns ofmammalian teeth from development. Nature 449, 427–432 (2007).

31. Jacob, F. Evolution and tinkering. Science 196, 1161–1166 (1977).32. Cho, S.-W. et al. Interactions between Shh, Sostdc1 and Wnt signaling and a new

feedback loop for spatial patterning of the teeth. Development 138, 1807–1816(2011).

33. Gomes Rodrigues, H. G. et al. Roles of dental development and adaptation inrodent evolution. Nat. Commun. 4, 2504 (2013).

34. Van Valen, L. An analysis of developmental fields. Dev. Biol. 23, 456–477 (1970).

35. Goswami, A. & Polly, P. D. in Carnivoran Evolution: New Views on Phylogeny, Form,and Function (eds Goswami, A. & Friscia, A.) 141–164 (Cambridge Univ. Press,2010).

36. Bhullar, B.-A. et al. Birds have paedomorphic dinosaur skulls. Nature 487,223–226 (2012).

Supplementary Information is available in the online version of the paper.

AcknowledgementsWe thank M. Fortelius, J. Eronen, I. Thesleff, P. Munne fordiscussions; S. Alto, M. Makinen, R. Santalahti, R. Savolainen, and M. Christensen fortechnical assistance; P. Schneider for the Fc-EDA-A1-protein; F. de Sauvage forHhAntag compound; Hou Yemao for tomography; and the FinnishMuseumof NaturalHistory (Helsinki, Finland), Museum of Natural History (Stockholm, Sweden), andMuseum fur Naturkunde (Berlin, Germany) for specimen loans. This work wassupportedby theAcademyof Finland to J.J.,M.M., I.S.-C., R01-DE021420 (NIH/NIDCR)and an NIH Director’s New Innovator Award DP2-OD007191 to O.D.K., an AustralianResearch Council Future Fellowship to A.R.E, and the Major Basic Research Projects(2012CB821904) of MST to Z.-Q.Z. Data are presented in the Supplementary andExtended Data Tables, available in the MorphoBrowser database (http://morphobrowser.biocenter.helsinki.fi/) and from the authors, and models can beaccessed at (http://www.biocenter.helsinki.fi/bi/evodevo/toothmaker.html).

Author Contributions E.H., J.J. and O.D.K. designed the project and wrote the initialmanuscript. E.H. and E.R. performed culturing experiments and K.S. mouseexperiments. E.H., E.R., A.K. and J.J. performed measurements and prepared images.I.J.C., A.R.E. and J.J. analysed evolutionary data. I.S.-C. constructed the computationalmodel and T.H. the ToothMaker. M.L.M., Z.-Q.Z. provided materials, observations andscientific interpretations. O.D.K. and J.J. coordinated the study. All authors discussedthe results and provided input on the manuscript.

Author Information Reprints and permissions information is available atwww.nature.com/reprints. The authors declare no competing financial interests.Readers are welcome to comment on the online version of the paper. Correspondenceand requests for materials should be addressed to J.J. (jernvall@fastmail.fm) orO.D.K. (ophir.klein@ucsf.edu).

RESEARCH ARTICLE

4 8 | N A T U R E | V O L 5 1 2 | 7 A U G U S T 2 0 1 4

Macmillan Publishers Limited. All rights reserved©2014

METHODSAnimals studied.WeusedEdanullmice (that is,Tabby) fromthe strainB6CBACa-Aw-J/A-Ta (Stock JR0314, JacksonLaboratory) crossedwithShhWT/GFPCre reportermice that express green fluorescent protein (GFP) under a sonic hedgehog (Shh)-promoter16, and the ShhWT/GFPCre reportermice aswild type.Although the growthand patterning of ShhWT/GFPCre teeth have been reported to be comparable to wildtype teeth19, we tested the growth of Eda null;ShhWT/GFPCre compound mutants.TheEdanull;ShhWT/GFPCrewere 82%of the size (P5 0.05;Mann–WhitneyU-test,n5 12) and had 98% of the cusps (P. 0.5;Mann–WhitneyU-test, n5 12) of Edanull teeth after 7 days in culture. All the procedures of this study involving animalswere reviewedand approved by relevantAnimalWelfare andResearchCommittees.Organ cultures. Embryonic day 13 lower molars were dissected and cultured forseven days at 37uC with 5% CO2 using a Trowell-type organ culture as describedpreviously37. We studied lower molars because their development is best under-stood. We used Fc-EDA-A1 protein (EDA), which has been shown to rescue theEdanull phenotype ofmice and dogs10,18,38. Culturemedia were changed every twodays. The media were supplemented with the EDA protein during the first fourdays. Corresponding amounts of bovine serumalbumin 0.1% (BSA, Sigma) in PBSwere used for controls. Eda null;ShhWT/GFPCre lower molars were cultured in theEDA concentrations 2.5, 10, 25, 50, 100, 500 and 1000 ngml21 (n5 9–16 in eachgroup, littermates were used as controls). As controls, ShhWT/GFPCre molars andEda null;ShhWT/GFPCre molars were cultured. Molars were photographed dailyboth in visible light (Olympus SZX9) and in fluorescent light (Leica MZFLIII andZeiss Stereo Lumar V12). Cusp presences and initiation times were determinedfrom time-lapse images andwere based on the appearance of their enamel knots, asrevealed by the GFP reporter. Explants were excluded from the analyses if molarstilted to a point in which tabulating cusps became difficult. Molars from at leastthree different litters were used for each treatment. Shh is expressed only in theprimary enamel knot during the cap stage of development10,19, andweused the epi-fluorescence of ShhWT/GFPCre to measure enamel knot sizes.Tooth shape analyses. Because we analysed cultured teeth, which do not allowdetermination of final cuspmorphology, we evaluated cusppresence-absence statesand the height of the talonid relative to the trigonid (repeated by two observers).Nevertheless, these tabulations cover over half of the first molar characters usedpreviously in the analysis of fully mineralized Eda null mutants10. The protoconidand metaconid cusps were tabulated as fused when they were closely spaced andlacked distinct cusp tips. Themouse anteroconid typicallyhas twoobliquely placedand fused cusps (bilobed morphology). In the character tabulations the distinctlybilobed anteroconids were tabulated as having two cusps (character state 2, foundin seven cases). Tabulating the anteroconid having only one cusp did not alter thepattern of results. Due to the limitations of culture systems, the presence of hypo-conulid and bilobed anteroconids are underrepresented compared to fully miner-alized teeth in a jaw. Even if cultured longer, the number of cusps does not increasedue to the onset of cellular differentiation. Talonid height was measured as therelative height of talonid cusps relative to trigonid cusps at the intervals 0 (notobserved in the explants), 0.25, 0.33, 0.50, 0.66, 0.75 and 1.00. Talonid height wasonly measured when the teeth were tilted enough to see the crown base and cusptips (in 75% of cases, Supplementary Table 1).To map changes in tooth character states in the cultured teeth, we used the fol-

lowing five characters. Corresponding character numbers used in ref. 10 are inparenthesis (single occurrences were excluded in Fig. 3c).1 (7): Anteroconid: (0) absent; (1) single; (2) bilobed.2 (9): Trigonid (protoconid andmetaconid): (0) absent (1) fusedwith single tip;

(2) separated.3 (10): Talonid (hypoconid and entoconid): (0) absent (1) fused with single tip;

(2) separated.4 (17): Hypoconulid: (0) absent; (1) present.5 (5): Trigonid size relative to talonid size: (0) notably higher; (1) slightly higher;

(2) flat occlusal surface. Here measured talonid height values 0.25 to 0.50, 0.66 to0.75, and 1 were coded as character states 0, 1, and 2 in Fig. 3c, respectively.Comparative data and analyses.We analysedmorphologies of first lowermolarsof extant 35 murine rodent species, extinct taxon Tribosphenomys minutus, andextant 32 carnivoran species (ExtendedData Tables 1 and 2). The extant taxa rep-resent the range of diets across the phylogenywithin each group27.Morphologicalcharacters corresponding to the ones tabulated for the experimental data (cuspnumbers and talonid height) were tabulated from photographs and three dimen-sional scans (available at MorphoBrowser database at (http://morphobrowser.biocenter.helsinki.fi/))27. Alternative ways of measuring talonid height in com-parative and experimental data (for example, talonid-trigonid height difference/protoconid length) produce equivalent patterns.When lateral cusps form a singlecrest in murines, they were tabulated as having two cusps. Due to the high diver-sity of carnivorem1s, we alsomeasured complexity by first scaling teeth to a lengthof 50pixel rows. The teethwere divided into trigonid and talonid sections, and then

orientation patch count (OPC) was determined for each section as previouslydescribed27,39. After determining the surface orientation at each pixel on the digitalelevationmodel (DEM), contiguous pixels that are facing the same cardinal direc-tion (for example, north, south, southwest) were grouped into patches. The num-ber of these patches is theOPC. Reducedmajor axis regressions were performed inPAST40 with permutation tests (Extended Data Table 3). Carnivoran data wasanalysed also after calculating the mean values for species in each family.Modelling.We implemented amorphodynamicmodel on tooth development14,21

in a graphical user interface called ToothMaker. ToothMaker is written in C11and uses QT libraries. The interface consists of model controls and an interactiveOpenGLmodel viewer for visualizing the three-dimensional objects in real-time asthey are computed. Model parameter space scanning can be automated within theinterface for given parameter value ranges. ToothMaker is freely available forWin-dows, OS X and Linux at (http://www.biocenter.helsinki.fi/bi/evodevo/toothmaker.html).We usedToothMaker to obtain amorphology corresponding to a developingmousemolar. Here we used previous studies14,21 and automated parameter scanningin ToothMaker to search for parameters that reproduce the EDA effects on thecusp patterns. The model integrates experimentally inferred genetic interactionswith tissue biomechanics to simulate tooth development. The genetic parametersentail three diffusible signals; an activator inducing enamel knots, an enamel knot-secreted inhibitor of enamel knot formation, anda growth factor regulating growthof the epithelium and the mesenchyme21. Because the enamel knot differentiationis irreversible, the model represents an irreversible reaction-diffusion-like model.EDAsignalling is known to interactwithmultiple signalling pathways41, and accord-ingly our focus was to model the effects of EDA through the genetic parameters.Herewe report onemousemodel version thatwasproduced in amodestnumberofiterations (14,000), allowing efficient scanning of parameters.Other parameter com-binations producingmouse provided equivalent results. Because of the limitationsof the implementation of cellular differentiation,matrix secretion, and current limitsin computational power, the model accounts for mouse molar development up today 16–17.Suppression of SHH signalling.To prevent cusp fusion in Eda null teethwithoutthe direct stimulation of EDA signalling, we inhibited SHH signalling. PreviouslySHH inhibition has been shown to increase cusp density in culture mouse teeth,and we treated Eda null;ShhWT/GFPCre lower molars with 1mM of cyclopamine(catalogue no. C4116, Sigma, diluted in dimethylsulphoxide (DMSO)) on culturedays zero and two (similarly to EDA treatments, n5 11). Pregnant Eda hetero-zygous females (n5 4, mated with Eda null males) were treated with Hedgehogpathway inhibitor at a dose of 50 and 100mg per kg. Small molecule antagonist,acting at the level of Smoothened, was synthesized as previously described42 andadministered via oral gavage to timed pregnant females starting at day 13.25 andcontinued to 14.75 and 15.75 of embryonic development in 8 h intervals.We chosethis late treatment because it does not interfere with the development of otherorgans, although the dental effects targeted principally the secondmolars. Animalsborn to these females were sacrificed at 35 days of age when all teeth are erupted,and skulls were cleaned using a dermestid beetle colony. Only the teeth from thehigher dosage treatments were affected (n5 2 of 4 Edanull males).We used X-raytomography to reconstruct three-dimensional cusp patterns. Themouse jawswerescanned using a custom-built microCT system Nanotom 180 NF (phoenixjX-raySystems 1 Services GmbH) with a CMOS detector (Hamamatsu Photonics) and ahigh-power transmission-typeX-ray nanofocus sourcewith a tungsten anode. TheTribosphenomys minutus specimens (IVPP V10775 (holotype) and V10776) werescanned using the 225 kV microCT (developed by the Institute of High EnergyPhysics, Chinese Academy of Sciences (CAS)) at the Key Laboratory of VertebrateEvolutionandHumanOrigins,CAS. Fiji (http://fiji.sc/Fiji)VolumeViewerplugin43

was used to segment and render the teeth.Voxel resolutions used for the renderingsare 3 to 4mm for mice and 2.4 to 3.3mm for Tribosphenomys.

37. Narhi,K.&Thesleff, I. Explant cultureofembryonic craniofacial tissues: analyzingeffects of signaling molecules on gene expression. Methods Mol. Biol. 666,253–267 (2010).

38. Mauldin, E. A., Gaide, O., Schneider, P. & Casal, M. L. Neonatal treatment withrecombinant ectodysplasin prevents respiratory disease in dogs with X-linkedectodermal dysplasia. Am. J. Med. Genet. 149A, 2045–2049 (2009).

39. Wilson, G. P. et al. Adaptive radiation of multituberculate mammals before theextinction of dinosaurs. Nature 483, 457–460 (2012).

40. Hammer, Ø., Harper, D. A. T. & Ryan, P. D. PAST: paleontological statisticssoftware package for education and data analysis. Pal. Electron. 4, http://palaeo-electronica.org/2001_1/past/issue1_01.htm (2001).

41. Fliniaux, I., Mikkola, M. L., Lefebvre, S. & Thesleff, I. Identification of dkk4 as atarget of Eda-A1/Edar pathway reveals an unexpected role of ectodysplasin asinhibitor ofWnt signalling in ectodermal placodes.Dev. Biol.320,60–71 (2008).

42. Yauch, R. L. et al. A paracrine requirement for hedgehog signalling in cancer.Nature 455, 406–410 (2008).

43. Schindelin, J. et al. Fiji: an open-source platform for biological-image analysis.Nature Methods 9, 676–682 (2012).

ARTICLE RESEARCH

Macmillan Publishers Limited. All rights reserved©2014

Cus

ps (n

)

Primary enamel knot size (Sqrt of area)40 60 80 100 120 140

1

3

5

7

2

4

6

Extended Data Figure 1 | Primary enamel knot size predicts cusp number.Size of the primary enamel knot (day 2) and cusp number (day 7) for Eda null,Eda null1 10ngml21 EDA, Eda null1 50 ngml21 EDA, and wild-type teeth.Reduced major axis regression for square root of mm2 are 0.0533 x20.791,r25 0.613,P, 0.0001, n5 46. Enamel knot size does not increase substantiallywith higher EDA concentrations.

RESEARCH ARTICLE

Macmillan Publishers Limited. All rights reserved©2014

Extended Data Figure 2 | The ToothMaker modelling interface andmorphodynamic model of tooth development. The model parameters can bechanged manually or scanned automatically (Options-menu). For descriptionsof parameters, see Methods. The figures illustrate the growth factor

concentration (secreted from the enamel knots) showing the future cuspareas. Initial activator concentration (Ina) is not used in the model. Themodel can be downloaded at (http://www.biocenter.helsinki.fi/bi/evodevo/toothmaker.html).

b

fD+sD f+ –rgM–rgE

–ceS–teS–iD+hnI

–tnI–aD–tcAWT

Cus

ps (

n)

Change (%) from the wild type parameter values

a Act– Da– Int– Inh+ Di–

Set– Sec– Ds+ Dff+ Egr– Mgr–

-80 -60 -40 -20 0 -80 -60 -40 -20 0

6

4

2

6

4

2

Extended Data Figure 3 | Scanning parameter space to produce gradualchanges. a, Parameters producing variation in cusp number were scanned at 10percent intervals up to 90 percent change from the wild-type (WT) mouse.Growth factor domains, produced by enamel knots, were used to tabulate cuspsnumbers (threshold 5 0.04). b, Simulated teeth show the minimum number ofcusps that can be produced by changing each parameter. Only parameter Actproduced single-cusped morphology. Plus and minus signs after eachparameter denote to the direction of parameter change that produced a

decrease in cusp number. Act 5 activator auto-activation, Da 5 activatordiffusion, Int 5 inhibitor production threshold, Inh 5 Activator inhibitionby inhibitor, Di 5 inhibitor diffusion, Set 5 growth factor productionthreshold, Sec 5 growth factor secretion rate, Ds 5 growth factordiffusion, Dff 5 differentiation rate, Egr 5 epithelial proliferation rate,Mgr 5 mesenchymal proliferation rate. All simulations were run for a fixednumber of iterations (14,000).

0

1

2

3

4

5

6

0 1,000 2,000 3,000

Cus

ps (n

)

Signaling domain size (arbitary units)

0.1 0.2 0.3 0.4

0.5 0.6 0.7 0.8

0.9 1.0 1.1 1.2

1.3 1.4 1.5 1.6

a b

Extended Data Figure 4 | Simulating EDA effects. a, Simulated shapesproduced by changing the activator parameter (Act) from 0.1 to 1.6 at 0.1interval. b, The size of inhibitor domain (at arbitrary threshold 0.85) at iteration

1,000 and corresponding cusp number at iteration 14,000 approximates therelationship between primary enamel knot size and cusp number in real teeth.

ARTICLE RESEARCH

Macmillan Publishers Limited. All rights reserved©2014

Extended Data Figure 6 | Rescuing cusps in Eda null teeth by inhibitingSHH. Eda null teeth cultured with SHH antagonist show variablemorphologies with tightly packed cusps (arrowheads). In addition, in

roughly half of the cases (n5 4 of 11 teeth) portions of the first molar appear toform from the fusion with the developing second molar (two bottom rows).Scale bar, 500mm.

Inh = 800, Di = 0.2

Inh = 20, Di = 0.2 Inh = 10, Di = 0.2

Inh = 800, Di = 0.01 Inh = 800, Di = 0.0001

Simulated Eda null tooth

Extended Data Figure 5 | Simulating reduction of inhibition in Eda nullteeth. Reducing activator inhibition by inhibitor (Inh) or diffusion of inhibitor(Di) results in formation of multiple cusps in simulated Eda null molar. Theeffects are variable depending on the parameter values, and the lability of thesystem appears to be corroborated in the in vitro experiments.

Eda null

Tribosphenomys

Wild type

Eda null + SHH antagonist

Extended Data Figure 7 | In vivo inhibition of SHH in Eda null embryoscauses the formation of separate cusps without crests. Obliquely anteriorviews and tomography sections (along the plane of the dotted line) of secondmolars show the lack of a crest (metalophid, arrowheads) in treated Eda nulland Tribosphenomys minutus (V10775). Enamel in sections shown in bluecolour except in Tribosphenomys fossil which did not allow segmentation ofenamel due to high degree of mineralization. Scale bar, 500mm.

Extended Data Table 1 | Comparison of talonid height and cuspnumber in rodents

Anterior cusps

1.542.002.852.782.923.003.003.133.20

3

54454445444543465444444344435464434

Experiment or species

Eda null (0 ng/ml EDA)Eda null (2.5 ng/ml EDA) Eda null (10 ng/ml EDA)Eda null (25 ng/ml EDA)Eda null (50 ng/ml EDA)Eda null (100 ng/ml EDA)Eda null (500 ng/ml EDA)Eda null (1000 ng/ml EDA)Wild type (0 ng/ml EDA)

Tribosphenomys minutus

Aethomys hindei Anisomys imitatorApodemus agrariusArvicanthis niloticusBandicota indicaBerylmys bowersiBunomys coelestisChiropodomys sp.Crateromys schadenbergiDasymys sp.Dephomys sp.Grammomys dolichurusHybomys univittatusHydromys chrysogasterHylomyscus stellaHyomys goliathLemniscomys striatusLeopoldamys sabanusLeptomys elegansMalacomys sp.Mallomys rothschildiMastomys natalensisMus musculusNesokia indicaNiviventer rapitNotomys mitchelliiOenomys hypoxanthusPelomys campanaePogonomys sp. Praomys jacksoniRhabdomys pumilioStochomys longicaudatusSundamys muelleriUromys caudimaculatusZelotomys sp.

Posterior cusps

1.001.001.461.782.002.292.402.632.50

2

43443334343443434333333233334433433

Talonid height

0.430.620.750.780.880.931.000.960.97

0.75

1.001.001.001.001.001.001.001.001.001.001.001.001.001.001.001.001.001.001.001.001.001.001.001.001.001.001.001.001.001.001.001.001.001.001.00

Cusp numbers tabulated for the entire anterior (anteroconid 1 trigonid) and posterior (talonid1

hypoconulid) parts of the m1 and corresponding heights of the talonid relative to the trigonid. Forexperimental data, averages listed for explants with data on both cusps and talonid heights(Supplementary Table 1). For rodent taxa and specimens, see Methods.

ARTICLE RESEARCH

Macmillan Publishers Limited. All rights reserved©2014

Extended Data Table 2 | Comparison of talonid height and cusp number in carnivorans

Cusp numbers and OPCs tabulated for the entire anterior (anteroconid1 trigonid) and posterior (talonid1 hypoconulid) parts of them1 and corresponding heights of the talonid relative to the trigonid. For taxaand specimens, see Methods.

RESEARCH ARTICLE

Macmillan Publishers Limited. All rights reserved©2014

Extended Data Table 3 | Reduced major axis regression analyses between talonid height and talonid cusp numbers and complexity

Data

ExperimentsExperiments (explants with EDA) Line from Tribosphenomys to MusCarnivoran species (cusps) Carnivoran families (cusps) Carnivoran species (OPC) Carnivoran families (OPC)

Slope (95% CI)

3.30 (1.76–4.01) 4.22 (3.08–5.07) 46.53 (4.60–7.73) 6.63 (4.68–7.84) 48.24 (33.37–56.50) 50.83 (26.74–62.61)

Intercept (95% CI)

-0.79 (-1.39–0.56) -1.63 (-2.38– -0.67) -1-1.38 (-2.02– -0.28) -1.28 (-2.09– -0.07) -10.93 (-15.64– -2.04) -12.14 (-20.62–1.64)

r

0.8910.946

0.5940.8740.5620.812

p

0.00030.0005

0.00010.00020.00010.0011

2

Talonid tabulations include all the posterior cusps. Explants with EDA include only Eda null teeth culturedwith EDAprotein (2.5ngml21 to 1,000ngml21). Values for carnivoran families are based on themeans ofspecies for each family (n59). For details, see text and Methods.

W W W. N A T U R E . C O M / N A T U R E | 1

SUPPLEMENTARY INFORMATIONdoi:10.1038/nature13613

Supplementary Table 1 | Day of culture for the appearance of cusps and correspondingcharacter codings. Asterisk indicates explants in which final cusps were tabulated at day 6,otherwise cusps were tabulated at day 7. 1 = EDA ng/ml, 2 = Anteroconid day, 3 = Protoconidday, 4 = Metaconid day, 5 = Talonid day, 6 = Entoconid day, 7 = Hypoconid day, 8 =Hypoconulid day, 9 = Comments, 10 = Cuspal characters, 11 = Cusp n, 12 = Talonid/Trigonidheight character state, 13 = Measured Talonid/Trigonid height.

ID 1 2 3 4 5 6 7 8 9 10 11 12 13

EDAIBAL1 0 2 Fused 4 0110 2 0 0.5EDAIBAL2 0 2 3 6 0210 3 0 0.33EDAIBAL3 0 2 4 6 0210 3EDAIBAL4 0 2 Fused 6 0110 2 1 0.66EDAIBAL5 0 2 3 5 0210 3 0 0.33EDAIBAL6 0 2 4 7 0210 3EDAIBBL4 0 2 Fused 5 0110 2 0 0.5EDAIBBR2 0 2 Fused 5 0110 2 0 0.33EDAIBDL5 0 2 3 6 0210 3 0 0.25EDAIBEL5 0 2 3 5 0210 3 0 0.5EDAIBER1 0 2 Fused 6 0110 2 0 0.33EDAIBIR2* 0 2 3 5 0210 3 1 0.66EDAIBJL2* 0 2 Fused 5 0110 2 0 0.5EDAIBJL5 0 2 3 5 0210 3 0 0.33EDAIBLR5 0 2 3 5 0210 3 0 0.332.5IBAL1 2.5 1 3 4 0210 3 0 0.52.5IBDR1 2.5 4 1 3 4 1210 4 1 0.752.5IBBL1 2.5 1 Fused 5 0110 2 1 0.752.5IBCL1 2.5 1 3 6 0210 3 0 0.332.5IBDR2 2.5 1 3 4 0210 32.5IBBL2 2.5 1 3 4 0210 3 2 12.5IBER2 2.5 1 3 4 0210 3 1 0.662.5IBCL2 2.5 1 3 4 0210 3 0 0.52.5IBFR2 2.5 1 3 5 0210 3 0 0.510IBAL1 10 2 3 5 0210 3 0 0.510IBAL3 10 5 1 3 4 1210 4 0 0.510IBAR2 10 4 1 3 4 4 1220 510IBBL4 10 4 1 3 4 4 1220 5 1 0.7510IBBR2 10 4 1 3 4 5 1220 510IBBR3 10 5 1 3 4 1210 4 1 0.7510IBBR4 10 4 1 3 4 5 1220 5 1 0.7510IBCL2 10 4 1 3 4 4 1220 5 1 0.7510IBCL3 10 5 1 3 4 1210 4 1 0.7510IBCL4 10 5 1 3 4 5 1220 5 1 0.7510IBCR2 10 4 1 3 4 4 1220 5 2 110IBCR4 10 5 1 3 4 5 1220 5 1 0.7510IBDL4 10 5 1 3 5 1210 4 1 0.7510IBDR3 10 1 3 4 0210 3 1 0.7510IBDR4 10 4 1 3 4 1210 4 2 125IBBL1 25 4 1 3 4 4 1220 5 1 0.7525IBCL1 25 4 1 3 4 4 1220 5 1 0.7525IBAL2 25 4 1 3 4 4 1220 525IBFR2 25 4 1 3 4 6 1220 5 2 125IBER2 25 4 1 3 4 1210 4 1 0.7525IBCL2 25 4 1 3 4 4 7 1221 6 2 125IBDL3 25 4 1 3 4 6 1220 5 2 125IBAR3 25 4 1 3 4 5 1220 5 1 0.7525IBAR4 25 1 3 5 0210 3 0 0.525IBAR5 25 1 3 4 0210 3 0 0.550IBAL3* 50 4 1 3 4 4 1220 550IBBL1 50 4 1 3 4 1210 4 2 150IBBL2 50 1 3 5 0210 3 1 0.6650IBBL3* 50 4 1 3 4 4 6 1221 6 2 150IBCL1 50 6 2 3 5 1210 4 1 0.6650IBCR1 50 4 1 3 4 1210 4 1 0.7550IBCR3* 50 4 1 3 4 4 6 1221 6 2 150IBDL1 50 6 2 3 5 1210 4 1 0.7550IBDR2 50 4 1 3 5 1210 4 1 0.7550IBDR3* 50 4 1 3 4 4 6 1221 6 2 150IBER2 50 4 1 3 4 4 6 1221 6 2 150IBFR2 50 4 1 3 4 4 7 1221 6 2 150IBGR2 50 4 1 3 4 4 6 1221 6 2 1100IBAR1 100 4 1 3 4 4 1220 5 2 1100IBBL1 100 4 1 3 4 4 1220 5100IBCL1 100 4 1 3 4 4 1220 5 1 0.75100IBCL2* 100 4 1 3 4 4 1220 5100IBDR1 100 4 1 3 4 4 7 1221 6 2 1100IBEL1 100 4 1 3 4 4 1220 5100IBFL1 100 5 1 3 5 5 1220 5 1 0.75100IBFR2 100 5 1 3 4 1210 4100IBGL1 100 4 1 3 4 4 1220 5 2 1100IBGR2* 100 4 1 3 4 4 6 1221 6 2 1100IBHR2* 100 4 1 3 4 4 1220 5 2 1500IBER1 500 4 1 3 4 4 1220 5 2 1

SUPPLEMENTARY INFORMATION

2 | W W W. N A T U R E . C O M / N A T U R E

RESEARCH

500IBER2* 500 4 1 3 4 4 bilobed anteroc 2220 6500IBFR1 500 4 1 3 4 4 1220 5500IBGR1 500 4 1 3 4 4 1220 5500IBHR1 500 4 1 3 4 4 7 1221 6 2 1500IBIL2* 500 4 1 3 5 4 1220 5 2 1500IBIR1 500 4 1 3 4 4 1220 5500IBJR1 500 4 1 3 4 4 6 bilobed anteroc. 2221 7500IBKR1 500 4 1 3 5 5 1220 5 2 1500IBLR1 500 4 1 3 4 4 6 1221 6500IBMR1 500 4 1 3 4 4 bilobed anteroc. 2220 6500IBNR1 500 4 1 3 4 4 6 1221 6 2 11000IBAL1 1000 4 1 3 4 4 6 bilobed anteroc. 2221 7 2 11000IBAR1 1000 4 1 3 truncated m1 1200 31000IBBL1 1000 4 1 3 4 4 1220 51000IBBR1 1000 4 1 3 4 4 7 1221 6 2 11000IBCL1 1000 4 1 3 4 4 1220 5 2 11000IBDL1 1000 4 1 3 5 5 6 1221 61000IBEL1 1000 4 1 3 5 5 7 1221 6 1 0.661000IBER1 1000 4 1 3 5 5 1220 5 2 11000IBFL1 1000 5 1 4 truncated m1 1200 31000IBFR1 1000 4 1 3 5 5 1220 5 2 11000IBGR1 1000 4 1 3 4 4 6 1221 6 2 11000IBHR1 1000 4 1 3 4 4 6 1221 6 2 1WTIBAL1 0 4 2 3 4 4 7 1221 6 2 1WTIBAL2 0 5 2 3 4 4 1220 5 2 1WTIBAL3 0 4 1 3 4 4 6 1221 6WTIBBL1 0 5 2 3 4 4 7 1221 6 1 0.66WTIBBL2 0 5 2 3 4 4 1220 5WTIBBL3 0 4 1 3 4 4 1220 5WTIBCL2 0 5 2 3 4 5 1220 5 2 1WTIBCL3 0 5 1 3 4 4 bilobed anteroc. 2220 6WTIBDL1 0 4 2 3 4 4 6 1221 6 2 1WTIBDL2 0 4 1 3 4 4 bilobed anteroc. 2220 6 2 1WTIBDL3 0 4 1 3 4 4 1220 5WTIBEL3 0 4 1 3 4 4 1220 5 2 1WTIBFL1 0 4 2 3 4 4 bilobed anteroc. 2220 6 2 1WTIBFL3 0 5 1 3 5 5 1220 5WTIBGL1 0 4 1 3 4 4 6 1221 6 2 1WTIBGR1 0 4 1 3 4 4 6 1221 6 2 1