A 3D MRI‐based atlas of a lizard brain

Magnetic resonance imaging (MRI) is an established technique for neuroanatomical analysis, being particularly useful in the medical sciences. However, the application of MRI to evolutionary neuroscience is still in its infancy. Few magnetic resonance brain atlases exist outside the standard model organisms in neuroscience and no magnetic resonance atlas has been produced for any reptile brain. A detailed understanding of reptilian brain anatomy is necessary to elucidate the evolutionary origin of enigmatic brain structures such as the cerebral cortex. Here, we present a magnetic resonance atlas for the brain of a representative squamate reptile, the Australian tawny dragon (Agamidae: Ctenophorus decresii), which has been the subject of numerous ecological and behavioral studies. We used a high‐field 11.74T magnet, a paramagnetic contrasting‐enhancing agent and minimum‐deformation modeling of the brains of thirteen adult male individuals. From this, we created a high‐resolution three‐dimensional model of a lizard brain. The 3D‐MRI model can be freely downloaded and allows a better comprehension of brain areas, nuclei, and fiber tracts, facilitating comparison with other species and setting the basis for future comparative evolution imaging studies. The MRI model and atlas of a tawny dragon brain (Ctenophorus decresii) can be viewed online and downloaded using the Wiley Biolucida Server at wiley.biolucida.net.

Comparative studies on brain anatomy and evolution would be greatly facilitated if the animal's nervous system could be rapidly visualized in an intact head, and even in live specimens (Corfield, Wild, Cowan, Parsons, & Kubke, 2008). Magnetic resonance imaging (MRI) is a noninvasive technique that allows for such visualization. This technique is particularly useful in the case of endangered or protected species, and when working with precious museum specimens that would be destroyed in the process of extracting the brain. From a practical perspective, MRI is advantageous because it does not require any of the labor-intensive tissue processing necessary for histology.
The resulting image can be viewed in any plane, allowing for brain regions and fiber tracts to be viewed from multiple orientations throughout their rostral-caudal extent. In addition, MRI can facilitate both inter-and intraspecific comparisons as measurements can be semi-computer-automated (e.g., Lerch et al., 2008).
Here, we present a detailed description of the brain of an agamid lizard, the Australian tawny dragon (Ctenophorus decresii, Duméril & Bibron, 1837; Reptilia: Agamidae), using high-resolution MRI. In the FIGURE 1 A comparison between a coronal section from (a) an MRI image of a single brain and (b) the MRI model of thirteen brains demonstrates that the model has far superior resolution (voxel = 20 μm 3 ) compared to the image of a single brain (voxel = 50 μm 3 ) FIGURE 2 A three-dimensional rendering of an MRI of a tawny dragon (Ctenophorus decresii) head showing the position of its brain from (a) a lateral perspective and (b) a dorsal perspective. The majority of the brain (in red) is included in this model, however the olfactory tracts and bulbs (in yellow) are excluded. (c) The natural position of the brain inside the tawny dragon head is rotated 28 in the x-plane compared to the position of our model [Color figure can be viewed at wileyonlinelibrary.com] atlas we identify nuclei, fiber tracts, and other structures throughout the brain in coronal, sagittal, and horizontal orientations. Furthermore, we describe our MRI data with reference to the neuromeric/prosomeric model, in addition to the traditional columnar model, since the former is more natural as it relates to the fundamental divisions of the brain that are shared by all vertebrates (Puelles, 2009;Puelles & Rubenstein, 2015;Puelles, Harrison, Paxinos, & Watson, 2013). This atlas therefore provides a new means of understanding the structure and connectivity of the reptile brain.

| Specimen acquisition
Sixteen male tawny dragons were collected from the southern Flinders Ranges, South Australia. We euthanized each lizard with an injection of 100 mg/kg sodium pentobarbital and an equal volume of 2 mg/mL lignocaine. Each lizard was then intracardially perfused following Hoops (2015). Magnevist was added to the fixative perfusate (4% paraformaldehyde) to maximize image contrast in magnetic resonance imaging (Ullmann, Cowin, & Collin, 2010a). The brains were stored at 4 C in a solution of 0.1% Magnevist and 0.05% sodium azide in phosphate-buffered saline until imaging. The Australian National University's Animal Experimental Ethics Committee approved all research under protocol number A2011-49.

| Magnetic resonance imaging
Whole-brain images of 13 tawny dragon brains (e.g., Figure 1a) were acquired using a Bruker Avance 11.74 Tesla wide-bore spectrometer (Ettlingen, Germany) with a micro-2.5 imaging probe capable of generating magnetic gradients of 1.50 T/m. Brains were immersed in Fomblin (perfluoropolyether, Grade Y06/6, JAVAC, Sydney, Australia) and placed in a 10 mm diameter Wilmad tube using a custom-built plastic holder (Hyare et al., 2008). Parameters used in the scans were optimized for gray-white matter contrast in the presence of Magnevist.
For comparison, two brains were embedded in agarose and vibratome-sectioned at 70 μm. Brain sections were stained for 5 min using the DNA-binding stain SYBR-green (Life Technologies Australia, Melbourne, Australia), rinsed in phosphate-buffered saline, mounted in Fluoro-Gel (ProSciTech, Brisbane, Australia), and imaged using Olympus fluorescence light microscopes.

| Model generation and analysis
To ensure consistent measures of brain morphometry all images were first manually masked such that consistent coverage of brain structures and nerve endings was achieved. In the tawny dragon the olfactory bulbs are small and separated from the brain on long stalks ( Figure 2) and we were unable to stabilize their location in the Wilmad tube. Therefore, the olfactory bulbs were included in the masked regions. The manually masked areas were then set to the background value such that they were not included in subsequent calculations. Thirteen brain image datasets of 50 μm 3 resolution were first reoriented to standard rostro-caudal orientation. All images were then corrected for B0 intensity inhomogeneity using the N3 algorithm (Sled, Zijdenbos, & Evans, 1998). An image with a good signal to noise ratio and no obvious artifacts was then manually selected from the group to create an initial model by blurring. All images were then recursively matched to this evolving model of average structure to create a minimum deformation average with a resulting resolution of 20 μm 3 (Figure 1b). The details of the model creation process can be found in Janke and Ullmann (2015). The fitting stages in this case started at a resolution of 1.28 mm and finished with a resolution of 80 μm 3 . The model finished with a resolution of 20 μm 3 .
To compare the natural orientation of the tawny dragon brain to the orientation of our model, a representative scan was acquired of a brain within a fully intact tawny dragon head. The brain was automatically segmented from this scan using a combination of registration to FIGURE 3 Our coordinate system for the lizard brain, demonstrated using a three-dimensional view of the lizard brain model with coordinate axes indicated. The bar in the lower right corner = 1 mm  (Paxinos & Franklin, 2013;     the constructed model, using the MINC toolkit (Vincent et al., 2016), and manual corrections. The linear rotational component of the automatic registration was used to measure the angle of alignment of the brain in the skull with respect to our model.

| RESULTS
The tawny dragon brain model described here can be viewed online and downloaded in NIFTI format from wiley.biolucida.net. The model represents the spatial positioning and intensity of each neural structure based on the nonlinear averaging of thirteen tawny dragon brains.
Using the intrinsic three-axis nature of MRI-atlases (Ullmann, Cowin, Kurniawan, & Collin, 2010b), we established a coordinate system with x-coordinates running medio-laterally, y-coordinates running rostrocaudally, and z-coordinates running ventro-dorsally, as per convention ( Figure 3). The midline of the brain, which divides the two hemispheres, has been designated as the plane x = 0. The center of the epiphysis (defined as the y plane in which the diameter of the epiphysis reaches its maximum) has been designated as the point (x,y) = (0,0), following studies which use the parietal eye as the point (x,y) = (0,0) (Greenberg, 1982 could be easily implemented in evolutionary neuroscience to, for example, digitally "extract" lizard brains from the surrounding tissue in an MRI. We registered our model to a representative MRI scan of an intact tawny dragon head, scanned ex-vivo, to demonstrate the position of the brain (Figure 2a,b). During the registration process, we observed that our atlas is not in the natural orientation of the tawny dragon brain; it is rotated by 28 in the x-plane (Figure 2c). The tawny dragon head MRI is also available for viewing and download from wiley.biolucida.net.
From our atlas, we were able to identify over 200 structures including areas, nuclei, fibre tracts and ventricles (Table 1). Whenever possible, the terminology of ten   We have identified the major anatomical divisions of our atlas according to the columnar (Table 2) and neuromeric (Table 3) models. The boundaries between neuromeres are often seen as transverse, dark strips separating grisea. They sometimes run paral-  Table 1 fourth ventricles appear black as they have filled with Fomblin, the oil used to immerse the brain during imaging (e.g., Figure 16). Because of this variation, we have outlined ventricles with white dashed lines.
The laminar morphology of some brain regions is readily distinguishable in our atlas, particularly in the cerebral cortex, optic tectum and cer- The reptilian optic tectum has a marked laminar organization consisting of cell layers separated by fiber layers; in some reptiles a total  Table 1 FIGURE 39 A coronal histological section (left panel) of a nuclear-stained posterior telencephalic section of a tawny dragon (Ctenophorus decresii) brain demonstrates similar anatomical features to an equivalent section through the MRI model (right panel). The plane according to our coordinate system is indicated in the upper right corner. The scale bar = 1 mm. A list of abbreviations is found in Table 1 of fourteen layers have been described (Ramón, 1891). These have been grouped in six main layers or strata , which are readily distinguishable in our model (Figure 47b). The optical layer is only slightly darker than the adjacent superficial grey and fibrous The meninges can be seen as thin, light structures around the edge of the brain, particularly in images of the brain stem (e.g., Figure 17).
Droplets of the aqueous storage solution can get trapped around the brain when transferring them to Fomblin for imaging. These appear as bright areas in some images, for example the spaces between the optic tectum and the epiphysis ( Figure 12) and between the optic tectum and the cerebellum ( Figure 14).  Table 1 FIGURE 41 A coronal histological section (left panel) of a nuclear-stained mid-diencephalic section of a tawny dragon (Ctenophorus decresii) brain demonstrates similar anatomical features to an equivalent section through the MRI model (right panel). The plane according to our coordinate system is indicated in the upper right corner. The scale bar = 1 mm. A list of abbreviations is found in Table 1 4 | DISCUSSION

| MRI as a method for studying comparative neuroanatomy
To create an atlas with the best possible resolution, we have used a non-linear image averaging strategy to create an 'idealized' model of a tawny dragon brain (Janke & Ullmann, 2015). Unlike histology, in MRI brain size impacts the level of discernable detail. For example, an MRI atlas of a monkey brain is able to delineate 720 structures in an image with a 0.5 mm 3 voxel size (Maldjian, Daunais, Friedman, & Whitlow, 2014), whereas an MRI atlas of a cichlid brain is able to delineate only 54 structures in an image with a 50 μm 3 voxel size (Simões, Teles, Oliveira, Van der Linden, & Verhoye, 2012).
Though the absolute voxel size in the cichlid atlas is much smaller than the voxel size in the monkey atlas, voxel size relative to brain size is much smaller in the monkey atlas. This provides a two-fold benefit to the monkey atlas: the larger absolute voxel size provides greater signal intensity, while smaller relative voxel size provides greater spatial resolution. Together, these factors allow for much more precise structural delineation in larger brains. This is an important consideration for comparative neuroscience, where comparisons are often made  Table 1 FIGURE 43 A coronal histological section (left panel) of a nuclear-stained anterior rhombencephalic section of a tawny dragon (Ctenophorus decresii) brain demonstrates similar anatomical features to an equivalent section through the MRI model (right panel). The plane according to our coordinate system is indicated in the upper right corner. The scale bar = 1 mm. A list of abbreviations is found in Table 1 between brains that differ in size by orders of magnitude. Using multiple MRIs to create a non-linear average brain model can help offset these issues in species with small brains.

| The columnar and neuromeric models of brain organization
The study of brain structure requires a model of brain organization that sets easily recognized landmarks that help identify neural structures along pre-established axes (Puelles, 2009). In these models, the relative topological positions of the brain divisions should be invariant, independent of differences in size and shape arising through development or evolution (Nieuwenhuys & Puelles, 2015;Nieuwenhuys, ten Donkelaar, & Nicholson, 1998). Two models are currently used to interpret brain morphology, the columnar and neuromeric/prosomeric models.
The columnar model of neural divisions has been the predominant model of the second half of the twentieth century. It was based on the discovery of distinct functional columns in the spinal cord, the alar plate or dorsal horn and the basal plate or ventral horn. The model was then applied to the brain (Herrick, 1910;reviewed by Puelles, 2009). Thus, the diencephalon was described as containing several dorsoventral columns, including epithalamus, dorsal thalamus, ventral thalamus and hypothalamus. This description is still used by many neuroscientists and is found in the majority of textbooks. However, the columnar model is increasingly being recognized as unnatural  Table 1 FIGURE 45 A coronal histological section (left panel) of a nuclear-stained posterior rhombencephalic section of a tawny dragon (Ctenophorus decresii) brain demonstrates similar anatomical features to an equivalent section through the MRI model (right panel). The plane according to our coordinate system is indicated in the upper right corner. The scale bar = 1 mm. A list of abbreviations is found in Table 1 because it does not consider the curvature of the longitudinal brain axis and the true morphogenetic divisions specified during development (reviewed by Puelles, 2009).
The neuromeric model (called the prosomeric model when discussing the forebrain) was employed by neuroembryologists during late nineteenth and early twentieth centuries, and was based on the periodic transversal bulges (called neuromeres) in the neural tube wall during embryonic development (Kupffer, 1906;Orr, 1896;Puelles, 2009;Puelles et al., 2013). This model has recently experienced a resurgence due to its suitability for explaining the expression patterns of developmental regulatory genes and their mutant phenotypes, the results of experimental studies such as transplants and fate mappings, and the trajectories of major fiber tracts (Díaz & Glover, 2002;Martínez, Marín, Nieto, & Puelles, 1995;Puelles, 2009;Puelles et al., 2013;Puelles & Rubenstein, 1993Shimamura, Hartigan, Martínez, Puelles, & Rubenstein, 1995). The model is already applied in widely used brain atlases, such as the last edition of the rat brain atlas , the Allen Developing Mouse Brain Atlas (http://developingmouse.brain-map.org/), and the chicken brain atlas (Puelles, Martínez-de-la-Torre, Paxinos, Watson, & Martinez, 2007). It is starting to be incorporated into MRI atlases (Watson et al., 2017).
The neuromeric model is powerful for comparative purposes since the same developmental units are found in all vertebrates (Medina, 2006;Puelles et al., 2007;Puelles & Medina, 2002). For these reasons, in this study we used the neuromeric model as our preferred paradigm to interpret MRI data, with the hope that this will be more useful for future functional and evolutionary studies using our atlas. The boundaries between neuromeres were identified as dark transversal strips (i.e., thin, cell poor areas) between grisea (which appear lighter). Fiber tracts, easily followed in our 3D atlas, are also useful for understanding the neuromeric organization of the tawny dragon brain, as their main trajectories are often either longitudinal (i.e., parallel to the alar-basal boundary) or transverse (i.e., parallel to the divisions between neuromeres). Though this model is based on the natural divisions of the brain and therefore is more desirable than the columnar model, the columnar model remains dominant in everyday use. Therefore, we provide Table 2, Table 3, and Figure 46 comparing the major brain divisions and subdivisions according to each model. The major differences occur in the forebrain, due to the different interpretation of the longitudinal (rostrocaudal) axis and, consequently, opposite view of the transverse (dorsoventral) divisions.

| Comparison with other squamates
Although all lizards share a basic pattern of brain organization, there are divergences in morphology that are related to the widespread morphological, ecological, and behavioral differences between species. For instance, the optic tectum is larger in diurnal lizards than in nocturnal ones, and the size of the cerebellum is related to the type of locomotion, being smaller and simpler in limbless than quadrupedal liz- The divisions of the lizard brain according to different models of brain organization. We have delineated the major neural subdivisions according to both the neuromeric and columnar models of brain organization in (a) selected coronal sections and (b) a sagittal section to demonstrate how these different models partition the brain. White text labels the major rostro-caudal neural subdivisions, which are delineated by broken white lines. Yellow text labels the major dorso-ventral neural subdivisions, which are delineated by broken yellow lines. Black text labels commonly used minor regional designations within the major subdivisions. The plane of each section according to our coordinate system is indicated in the lower left corner. hp = hypothalamic prosomere, m = mesomere, p = prosomere, r = rhombomere [Color figure can be viewed at wileyonlinelibrary.com] quadrupedal diurnal lizard (Gibbons, 1979;Osborne, Umbers, & Keogh, 2013;Osborne, Umbers, Backwell, & Keogh, 2012).
In the tawny dragon, we have identified the four classical cortical areas of the lizard brain: the medial, dorsomedial, dorsal, and lateral cortices (Striedter, 1997). In the dorsomedial cortex there is a cell plate visible in the inner plexiform layer (the CPDMCx in Figure 10), close to the ventricle and associated with a small but distinct ventricular ridge and a thickening of the overlying dorsomedial cortex. A similar organization is also observed in Agama agama ( Figure 1B of Wouterlood, 1981). In other lizards, this inner cell plate is not as evident, although some cell clusters can be observed in a similar position (Martinez-Guijarro, Desfilis, & Lopez-Garcia, 1990;Medina et al., 1992;Smeets et al., 1986). In gekkonids, lacertids, and iguanids, the cell clusters are more numerous in the inner plexiform layer of the dorsal cortex instead of the dorsomedial cortex. Some of these form a plate referred as the cell plate of Unger (Lacerta: Medina et al., 1992; Gecko: Smeets et al., 1986) or the supraventricular layer (Iguana: Northcutt, 1967). In the green anole a cell plate is visible in the medial and dorsomedial cortices; Greenberg (1982) labelled it the dorsomedial interposition.
The identification of the dorsal pallium in reptiles has been controversial, for example see Butler (2011) versus Puelles (2006. Based on genoarchitecture during development, Desfilis et al. (2018) proposed that it is located in a very rostral and medial position, resembling that of the avian dorsal pallium, or Wulst. This area shows a different cytoarchitecture compared to medial, dorsomedial and dorsal cortices, which appear more caudally (Desfilis et al., 2018  pallium (Desfilis et al., 2018). These two sectors of the dorsal ventricular ridge are evident at the caudal telencephalic levels of our atlas (e.g., Figure 10), since they appear as two light areas separated by a dark (i.e., cell poor) strip of tissue. In Iguana these two areas are also separated by a cell-poor lamina (Northcutt, 1967). In birds, the corresponding regions are the nidopallium and arcopallium, which are again separated by a cell-poor lamina (Desfilis et al., 2018).
In our model, the most prominent component of the ventrocaudal pallium is the nucleus sphericus, a structure that exhibits substantial variation in size and complexity between species. This nucleus is involved in vomerolfaction and receives massive afferents from the accessory olfactory bulb (Lanuza & Halpern, 1997;Lohman & Smeets, 1993;Martínez-García, Olucha, Teruel, Lorente, & Schwerdtfeger, 1991). The degree of development of this nucleus likely relates to its chemosensory function. In species that use the vomeronasal system extensively, such as snakes and lizards with forked tongues, the spherical nucleus occupies a large proportion of the dorsal ventricular ridge (Cooper, 1995;Halpern, 1980;Schwenk, 1993). In species with a reduced vomeronasal organ, this nucleus may be practically nonexistent, as is the case in Anolis (Greenberg, 1982). In the tawny dragon, the spherical nucleus appears to be of intermediate size, similar to Iguana (Northcutt, 1967 Vercken, Massot, Sinervo, & Clobert, 2007;Zamundio & Sinervo, 2003). One common finding is variation in reproductive strategy between color morphs Osborne, 2005bOsborne, , 2005aOsborne et al., 2012;Teasdale et al., 2013;Wellenreuther, Svensson, & Hansson, 2014;Yewers et al., 2015;Zamundio & Sinervo, 2003). However, little attention has been paid to neural differences between morphs, despite their obvious potential role in driving behavioral variation, including reproductive strategies (but see LaDage et al., 2009LaDage et al., ,2013. Another Ctenophorus species, the painted dragon (C. pictus), is also color polymorphic (Healey, 2008;Olsson, Schwartz, Uller, & Healey, 2009;Tobler, Healey, & Olsson, 2011), but most remaining Ctenophorus species are monomorphic, making this genus an ideal system for studying the evolution of color polymorphism and its relationship to behavioral and neural variation. Further work using color polymorphic lizard species holds great potential in elucidating the neural underpinnings of different reproductive strategies.

| CONCLUSIONS
This is the first time, to our knowledge, that an MRI atlas of a lizard brain has been produced. MRI is an innovative technique used frequently in the medical sciences. Here, we have added the first reptile organisms. The resolution obtained in this atlas is significantly higher than that of other atlases for animals with similarly-sized brains. We hope this atlas provides inspiration to further the study of the reptile brain, the correlation between brain structure and function, and the study of brain evolution, particularly using comparative methods. Only by advancing research in all these fields can we understand the general principles of vertebrate brain organization and identify selective pressures and mechanisms behind variation in the functional organization of the brain. We aspire to develop a range of MRI atlases representing, as much as possible, the diversity of vertebrates. Our goal is to make these universally available through a virtual museum, similar to those provided by brain collections in traditional brick-and-mortar museums (Iwaniuk, 2010(Iwaniuk, ,2011, and more recently by the on-line brain collections such as the Comparative Mammalian Brain Collection (http:// neurosciencelibrary.org) and BrainMaps.org (http://brainmaps.org/).

| DATA ACCESSIBILITY
The MRI model of a tawny dragon brain (Ctenophorus decresii), the dragon brain atlas, and the MRI of a tawny dragon head are freely available for download from the Wiley Biolucida Server at wiley.biolucida.net.    Table 1 FIGURE 31