Modern birds possess highly encephalized brains that evolved from non-avian dinosaurs. Evolutionary shifts in developmental timing, namely juvenilization of adult phenotypes, have been proposed as a driver of head evolution along the dinosaur-bird transition, including brain morphology. Testing this hypothesis requires a sufficient developmental sampling of brain morphology in non-avian dinosaurs. In this study, we harness, for the first time, brain endocasts of a postnatal growth series of the ornithischian dinosaur Psittacosaurus and several other immature and mature non-avian dinosaurs to investigate how evolutionary changes to brain development are implicated in the origin of the avian brain. Using three-dimensional characterization of neuroanatomical shape across archosaur reptiles, we demonstrate that (i) the brain of non-avian dinosaurs underwent a distinct developmental trajectory compared to alligators and crown birds; (ii) ornithischian and non-avialan theropod dinosaurs shared a similar developmental trajectory, suggesting that their derived trajectory evolved in their common ancestor; and (iii) the evolutionary shift in developmental trajectories is partly consistent with paedomorphosis underlying overall brain shape evolution along the dinosaur-bird transition; however, the heterochronic signal is not uniform across time and neuroanatomical region suggesting a highly mosaic acquisition of the avian brain form.

Taken from manuscript's Methods section (in-text citations refer to references of manuscript):

Specimens

We amassed 85 previously published endocranial reconstructions that spans 52 species of non-avian and avian dinosaurs in addition to American alligators (Table 1). The data include interspecific sampling of 36 crown birds, a juvenile Archaeopteryx³², and 19 non-avialan dinosaurs^{1,3,17–20,33}, plus ontogenetic series of the domestic chicken (Gallus gallus; n = 14), the American alligator (Alligator mississippiensis; n = 14), and the ornithischian Psittacosaurus lujiatunensis^3,17,34 (n = 4). Among non-avialan dinosaurs, several isolated immature specimens of both ornithischian and non-avian theropod dinosaurs, including Lambeosaurus (ROM 758) and troodontids Zanabazar and an unnamed specimen IGM 100/1126^{19–21,35–38} were sampled for both ontogenetic and phylogenetic diversity. The taxonomically ambiguous tyrannosaurid specimen, CMNH 7541, has been considered a relatively immature specimen within Tyrannosauridae^39–43. The major results and conclusions of this study are not dependent on the taxonomic assignment of this specimen, but rather the ontogenetic stage of CMNH 7541. Considering that no organism attributable to a tyrannosaur similar to, and including, CMNH 7541 has preserved an external fundamental system that indicates the cessation of rapid growth and the onset of skeletal maturity^39,45, we treat CMNH 7541 as an immature individual.  The specimen and data acquisition protocols for Alligator and Gallus specimens were approved by Institutional Animal Care and Use Committee at respective institutions (Stony Brook IACUC Protocol #236370-1; Oklahoma State University Center for Health Sciences IACUC Protocol #2015-1; American Museum of Natural History 2014). Please refer to the references listed in Table 1 for additional details about those specimens. 

Imaging and Endocranial Reconstructions

Endocranial reconstructions were generated from computed tomography (CT) imaging of cranial specimens. Considering the incomplete nature of the PKUP 1053 and PKUP 1054 endocasts¹⁸, they were reconstructed again for this study. We found that the original specimens and reconstructed models were of sufficient quality for morphometric analysis. These reconstructions were paired with two other Psittacosaurus specimens, IVPP V15451 and IVPP V12617, to reconstruct an ontogenetic series for the species P. lujiatunensis. The endocast of the tyrannosaurid Alioramus altai (IGM 100/1844) was retrodeformed using a local symmetrization algorithm⁴⁶ implemented in the “Morpho” R package⁴⁷ due to its sheared appearance. Because the combined data comprise multiple sources, the scanner, scanning parameters, and segmentation protocol naturally differ across datasets (Table 1). Nevertheless, we considered variation in resulting shape data due to differences in these practices to be negligible compared to the large interspecific variation observed across archosaurs⁴⁸. However, to adopt some standardization across endocranial reconstructions, we used the “QuickSmooth” function in Geomagic Wrap v2021 (Artec 3D, Senningerberg, Luxembourg) to remove cranial nerve and vascular projections from the main body of all the endocasts at the base.

Morphometric Data

We adopted the same high-density 3D landmark scheme from previous studies to characterize the overall shape of endocasts and their major regions, including the cerebrum, optic lobe, cerebellum, and brainstem^3,33,34 (Fig. 1c; Table S1). The ‘patch’ tool in Landmark Editor v3.6⁴⁹ was used to place discrete landmarks that are anatomically defined, curve semi-landmarks that delimit regions, and surface semi-landmarks that capture the surface morphology within regions on both left and right sides of the endocast. We did not collect coordinate data from the olfactory tract and bulbs due to their incomplete preservation in the fossils, as well as their diminutive size in most crown birds. We acknowledge that endocasts are generally accurate representations of actual brain morphology among archosaurs^34,50, but their correspondence is relatively low in the hindbrain due to greater presence of tissues surrounding the brain. This correspondence is lower in crocodilians²⁷ and presumably in non-avialan dinosaurs, especially the presence of dorsal eminence in some endocasts^19,20,51. We have noted in the manuscript where the lower brain-endocast correspondence in the hindbrain, particularly the ‘cerebellum’ region, may influence the interpretation of results.

The coordinate data were subjected to generalized Procrustes alignment^52,53 minimizing total bending energy, allowing semi-landmarks to slide on the mesh surface^54,55 using the ‘slider3d’ and ‘gpagen’ functions in the “Morpho”⁵⁶ and “geomorph”⁵⁷ R packages. After alignment, (semi-)landmarks on the right side were removed to remove redundant shape information while minimizing artifacts caused by alignment of one-sided coordinate data of bilaterally symmetric structures^58,59. The resulting shape data to be analyzed are comprised of 14 discrete landmarks, 49 curve semi-landmarks, and 56 surface semi-landmarks. For region-specific analyses, the size and shape of each region was extracted by calculating centroid size and performing localized alignment on a subset of landmarks that characterize each region.

Time-Calibrated Phylogeny

To perform phylogenetic comparative analysis, we constructed a time-calibrated phylogeny that includes all the sampled species. Because the crown group sampling is equivalent to a previous study³, fossil taxa were grafted to the existing tree file (additional details on the base tree construction³). These additional taxa included Corythosaurus, Hypacrosaurus, Psittacosaurus, Gorgosaurus, and Tyrannosaurus, which were incorporated into the existing phylogeny based on the mid-point of range of first occurrence age recorded on the Paleobiology Database (paleobiodb.org). When the maximum age of a species was identical to that of its clade, the age of the internal node was set to equally bisect the parent and descendent branches (‘Equal Branching Method’)⁶⁰.

Analysis

All statistical analyses for this project were performed in R v4.2.1⁶¹. Patterns of endocast shape variation were visualized as morphospaces for the cerebrum, cerebellum, optic lobe, and brainstem regions of the endocasts and were constructed using scores from PCAs made from shape data. The functions physignal, procD.lm, and procD.pgls of the geomorph package were used to assess the level of phylogenetic signal, allometry, and evolutionary allometry, with 1000 pseudoreplications. The multivariate statistical tests outlined above were chosen due to their robusticity against type I errors and loss of power that often accompanies differences in specimen and landmark sampling^62–64. Allometric trends were visualized by plotting the PC1 of residuals from the overall shape-to-size relationship against scores along this allometric relationship⁶⁵. We used Morpho package’s “CAC” function to extract CAC scores and residuals from shape data. Statistical differences between endocranial shapes and allometric trajectories between clades were tested with the procD.lm function, where differences in clades were tested on data accounting for log-transformed centroid size. In addition, we used the “trajectory.analysis” function in the ‘RRPP” R package⁶⁶ to test for differences between vectors of evolutionary and developmental shape change in total shape space.