Fossils reveal the complex evolutionary history of the mammalian regionalized spine

See allHide authors and affiliations

Science  21 Sep 2018:
Vol. 361, Issue 6408, pp. 1249-1252
DOI: 10.1126/science.aar3126

Early shifts lead to big changes

Mammals represent one of the most morphologically diverse taxonomic groups. One of the unique features underlying this diversity is variability of the spine, which facilitates everything from flexibility for speedy running and support for upright walking. Jones et al. studied a group ancestral to modern mammals—nonmammalian synapsids, or mammal-like reptiles. As forelimb function diversified, the spine developed distinct regions. These regions then differentiated further, leading to the highly varied mammalian forms we see today.

Science, this issue p. 1249


A unique characteristic of mammals is a vertebral column with anatomically distinct regions, but when and how this trait evolved remains unknown. We reconstructed vertebral regions and their morphological disparity in the extinct forerunners of mammals, the nonmammalian synapsids, to elucidate the evolution of mammalian axial differentiation. Mapping patterns of regionalization and disparity (heterogeneity) across amniotes reveals that both traits increased during synapsid evolution. However, the onset of regionalization predates increased heterogeneity. On the basis of inferred homology patterns, we propose a “pectoral-first” hypothesis for region acquisition, whereby evolutionary shifts in forelimb function in nonmammalian therapsids drove increasing vertebral modularity prior to differentiation of the vertebral column for specialized functions in mammals.

The evolution of the mammalian body plan from the ancestral amniote condition is one of the most iconic macroevolutionary transitions in the vertebrate fossil record (1, 2). A unique feature of mammals is their specialized vertebral column, which displays constrained vertebral counts but highly disparate morphologies (24). In therian mammals, the presacral vertebral column is traditionally divided into cervical, rib-bearing thoracic, and ribless lumbar regions (Fig. 1A). In contrast, the presacral vertebrae of basal amniotes are comparatively uniform and show little differentiation (Fig. 1B). The transition from an “unregionalized” to a “regionalized” presacral column is an important step in mammalian evolution and has been linked to the origin of specialized gaits and respiratory function (1, 2, 5, 6).

Fig. 1 Regionalization and heterogeneity.

(A) The therian presacral column is highly regionalized and morphologically differentiated (Mus musculus). (B) Basal synapsids display a homogeneous dorsal region with little differentiation (Ophiacodon).

Recent quantitative work has detected subtle presacral regionalization in extant snakes and limbed lizards, superficially unregionalized taxa (7). It was hypothesized that the ancestral amniote condition is “cryptic regionalization,” in which regions are present but are only subtly expressed. The global-patterning Homeobox (Hox) genes were implicated as underlying these conserved regionalization patterns. Under this model, the degree of regionalization—the number of regions present—has remained constant through mammalian evolution, whereas the amount of morphological disparity between regions (here termed heterogeneity) has increased. But this evolutionary scenario is based solely on data from extant species.

The two amniote clades—Synapsida (mammals and their relatives) and Sauropsida (reptiles, birds, and their relatives)—diverged more than 320 million years ago and have independently undergone substantial morphological transformations. Therefore, to document the evolution of the mammalian vertebral column, we must examine mammals’ extinct forerunners, the nonmammalian synapsids. Here, we examined the presacral vertebral columns of 16 exceptionally preserved nonmammalian synapsids (including “pelycosaurs,” basal therapsids, and cynodonts), one extinct amniote outgroup, and a broad range of extant salamanders, reptiles, and mammals. Using morphometric data, we quantified patterns of regionalization and heterogeneity and compared their evolution to elucidate when and how synapsid presacral differentiation occurred.

Using a likelihood-based segmented regression approach, we calculated a regionalization score for each taxon [an Akaike information criterion (AIC)–weighted average of the relative fit of one- to six-region hypotheses], producing a continuous variable that reflects the estimated number of vertebral regions (fig. S2). Similar to prior work (7), most reptiles and some extant mammals (e.g., monotremes) have regionalization scores indicating the presence of four regions (Fig. 2A), whereas therians (marsupials and placentals) most frequently display five regions. Therian regionalization scores are also more variable, probably reflecting high ecomorphological diversification of their axial system (4). Thus, data from extant amniotes alone support the null hypothesis of conserved regionalization. However, both salamanders and basal synapsids have lower regionalization scores than extant amniotes (Fig. 2A, cool colors), which demonstrates that regionalization increased independently in the sauropsid and synapsid lineages. Accordingly, we reject the hypothesis of conserved regionalization patterns in amniotes, and instead propose the hypothesis of increasing regionalization in synapsid evolution.

Fig. 2 Evolution of presacral differentiation in amniotes.

(A) Regionalization score. (B) Heterogeneity [log(mean variance)]. Warmer colors reflect more regions and greater morphological heterogeneity, respectively. Black circles, mammals; gray circles, fossil taxa; triangles, reptiles; stars, amphibians; grayed tips in (A), fossil taxa excluded because of <0.75 r2 of regionalization model. See table S5 for full taxonomic names. Ma, millions of years.

Heterogeneity, expressed as the logarithm of the mean variance of the morphological measures for each column, also increased during synapsid evolution (Fig. 2B). Lepidosaurs and salamanders have low heterogeneity, denoting relative uniformity of the axial column; therians have much higher values, reflecting their extreme disparity; and crocodilians have intermediate levels. Most nonmammalian synapsids also have intermediate levels of heterogeneity. The outgroup Diadectes and the ophiacodontids display particularly low values, reinforcing previous assertions of homoplastic increases in mammals and archosaurs from a homogeneous ancestral condition (7). The cynodont Kayentatherium has more heterogeneous morphologies than the other fossil taxa, reflecting its position close to the mammal radiation. Given the association between heterogeneity and functional specialization of the axial skeleton in therians, the more homogeneous morphologies of most nonmammalian synapsids point toward functional conservatism.

Although regionalization and heterogeneity increased during synapsid evolution, they are not significantly related (fig. S7 and table S6, P = 0.73), meaning that simple linear change is insufficient to explain these patterns. Instead, quantitative trait modeling supports evolution toward shifting adaptive optima (multiple optimum Ornstein-Uhlenbeck models) for these data (table S7). On the basis of AIC fitting, we reconstructed two major adaptive shifts in each trait during synapsid evolution (Fig. 3 and fig. S8). The adaptive optimum for regionalization increases from around three regions in “pelycosaurs” to around four regions at the base of Therapsida, with a later shift to five regions occurring in Theria. The adaptive optimum for heterogeneity increases first at Cynodontia and subsequently within therians. Taken together, our data reveal that vertebral regionalization increased before heterogeneity increased, hence these two measures of axial differentiation evolved independently.

Fig. 3 Adaptive regime shifts in vertebral evolution.

(A) Regionalization. (B) Heterogeneity. Theta denotes adaptive optima of each regime. Pz, Paleozoic; T, Triassic; J, Jurassic; K, Cretaceous; Cz, Cenozoic.

To understand how vertebral regionalization increased in synapsids, we reconstructed region boundaries recovered in the best-fit segmented regression models (Fig. 4A). Region boundaries were then cross-referenced with developmental data, anatomical landmarks, and variation in extant species to identify homologies (Fig. 4B). In extant tetrapods, the cervicothoracic transition is correlated with Hox6 expression, rib morphology, and the position of the forelimb and brachial plexus (8). Therefore, the cervicothoracic boundary was identified by (i) the position of the posterior branch of the brachial plexus, and (ii) the location of the anterior sternal articulation or first long rib. Functional studies in Mus also show that Hox9 patterns the transition from sternal to nonarticulating ribs and that Hox10 controls the suppression of ribs altogether in the lumbar region (Fig. 4B) (9, 10). In keeping with this association, dorsal regions were defined relative to their proximity to long ribs (anterior dorsal), short ribs (posterior dorsal), or absent ribs (lumbar).

Fig. 4 Best-fit region models, region homologies, and evolutionary hypothesis.

(A) Best-fit region models for select taxa. Colors represent inferred region homologies; St. Dev., standard deviation of break locations; PS count, presacral count; R-sq, adjusted r2; % complete, total completeness; shaded region models reflect taxa with <0.75 r2 fit (excluded from evolutionary reconstructions). (B) “Pectoral-first” hypothesis for the evolution of synapsid presacral regionalization. Taxa (left to right): Ambystoma, Iguana, Edaphosaurus [redrawn from (17)], Thrinaxodon [redrawn from (2)], Mus. Width of gray bars reflects relative rib lengths and/or connection to sternum; vertical dashed lines denote cervicothoracic transition. For Mus, Hox bands correspond to vertebrae affected by functional gene manipulation (18).

Using these criteria, region homology hypotheses were constructed in key taxa for which rib or neural anatomy are known (Fig. 4B). In salamanders (and the stem amniote Diadectes; see supplementary text), three regions are recovered. The anterior break correlates with the posterior branch of the brachial plexus in Ambystoma, implying homology with the cervical region despite the lack of a true “neck” (Fig. 4B, red region). Although salamanders have poorly developed ribs, the position of the posterior break in the mid-trunk is consistent with the anterior-posterior dorsal transition in other taxa (Fig. 4, yellow and pale blue regions). This ancestral three-region pattern is retained in the most basal synapsids. In “pelycosaurs,” the first break corresponds to the inferred cervicothoracic transition based on rib length and forelimb position (e.g., v5 in Edaphosaurus, v7 in Dimetrodon), whereas the second break corresponds to the gradual transition from longer to shorter dorsal ribs, signifying cervical, anterior dorsal, and posterior dorsal homologies (Edaphosaurus; Fig. 4B).

Our data point to the convergent addition of a fourth region in distinct locations in sauropsids and synapsids. In sauropsids, a fourth region is detected anterior to the brachial plexus, suggesting a novel cranial region within the neck (Iguana; Fig. 4B, purple region). Sauropsids exhibit more variation in cervical count than do synapsids (11), providing a potential connection between neck plasticity and cervical modularity in this lineage. Conversely, in basal therapsids and cynodonts, a fourth region is detected posterior to the cervicothoracic transition (Thrinaxodon; Fig. 4B, orange region). In Thrinaxodon, the first break corresponds to the cervicothoracic transition and first full-length rib (v7–8), the second break lies in the middle of the long rib series (v12–13), and the anterior-posterior dorsal boundary falls at the transition from long to short ribs (v19–20). These regions conform to the ancestral cervical region (red), a novel pectoral region (orange), and the ancestral anterior dorsal (yellow) and posterior dorsal (pale blue) regions. Therian mammals display an additional break within the posterior dorsal region that differentiates the ribless lumbar region (Mus; Fig. 4B, blue region).

Considering the pattern of region acquisition, we propose a “pectoral-first” hypothesis for the evolution of mammalian presacral regionalization (Fig. 4). Under this hypothesis, “pelycosaurs” retained the three-region ancestral amniote condition. In basal therapsids, addition of a fourth “pectoral module” accompanied the reorganization of the pectoral girdle and forelimb. Unlike “pelycosaurs,” therapsids are characterized by reduction of the pectoral girdle dermal bones and increased shoulder mobility (1, 12). Medial extrinsic shoulder muscles (e.g., levator scapulae, serratus ventralis) originating on the scapula are thought to have expanded their axial insertions during synapsid evolution (12). As these vital body-support muscles attach directly onto the underlying vertebrae and ribs, shifts in pectoral morphology and function likely drove divergent neck- and shoulder-selective regimes in the axial skeleton, providing impetus for increased regionalization (1, 12, 13). Further, the vertebrae, medial extrinsic shoulder muscles, and dorsal border of the scapula all develop directly from somitic mesoderm (primaxial), signifying strong developmental ties between these structures (14).

It has been proposed that the muscular diaphragm evolved from an unmuscularized septum or “proto-diaphragm” via co-option of shoulder muscle precursor cells that were later canalized into a distinct cell population by repatterning of the posterior neck (15). Reorganization of the anterior column and pectoral girdle in therapsids may have facilitated this transition by increasing cervicothoracic modularity and remodeling shoulder musculature. Subsequent fixation of the cervical count at seven in nonmammalian cynodonts is hypothesized to represent the appearance of the mammalian-style muscular diaphragm (6). Thus, anterior regionalization that had initially been associated with shoulder evolution in early therapsids was likely later exapted in cynodonts in response to selection for increased ventilatory efficiency (5, 15).

A “lumbar module” evolved later in therian mammals. Evolution of the lumbar region in mammals is associated with Hox10, which functions to repress rib formation and patterns lumbar identity in Mus (10) (Fig. 4B). Convergent loss or gain of lumbar ribs in multiple fossil theriiform clades suggests high plasticity of this character early in therian evolution (16). Within therians, lumbar count and morphology vary, and this is reflected by translocation of the (morphometrically defined) region boundary in our sample. Because the lumbar region plays a critical role in mammalian locomotion, it is predicted that region variability is related to ecological specialization caused by clade-specific functional overprinting.

Regional differentiation is “the major structural difference between reptilian and mammalian vertebral columns” (13), yet its evolution has never been quantitatively examined. Our findings show that regionalization and heterogeneity—the two aspects of vertebral differentiation—evolved independently. Forelimb reorganization in therapsids drove initial increases in regionalization as a result of developmental and functional connections between the pectoral girdle and underlying vertebrae. High heterogeneity and presumed functional diversity did not appear until crown mammals. The combination of a regionalized axial skeleton with heterogeneous vertebral morphologies ultimately enabled mammals to become specialized for a remarkable diversity of ecologies and behaviors.

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S8

Tables S1 to S7

References (1949)

References and Notes

Acknowledgments: We thank R. Asher, B. Brainerd, D. Brinkman, T. Capellini, C. Capobianco, J. Chupasko, J. Cundiff, K. Jakata, T. Kemp, C. Mehling, A. Millhouse, M. Omura, A. Resetar, J. Rosado, B. Rubidge, R. Smith, K. Smithson, C. Tabin, R. Tykoski, I. Werneburg, and B. Zipfel. Funding: Supported by NSF grants EAR-1524523 (S.E.P.) and EAR-1524938 (K.D.A.) and by an AAA Postdoctoral Fellowship (K.E.J.). Author contributions: Study design, K.E.J., K.D.A., and S.E.P.; methods, K.E.J., P.D.P., and S.E.P.; data collection, K.E.J., K.D.A., S.E.P., J.J.H., V.F., J.K.L., and S.T.; data analysis, K.E.J.; manuscript preparation, K.E.J., K.D.A., and S.E.P. Competing interests: The authors have no competing interests. Data and materials availability: Data are available in table S5 and Dryad (doi:10.5061/dryad.jm820mg). Code is available via github (
View Abstract

Stay Connected to Science

Navigate This Article