Endocranial Anatomy of the Giant Extinct Australian Mihirung Birds (Aves, Dromornithidae)

: Dromornithids are an extinct group of large ﬂightless birds from the Cenozoic of Australia. Their record extends from the Eocene to the late Pleistocene. Four genera and eight species are currently recognised, with diversity highest in the Miocene. Dromornithids were once considered ratites, but since the discovery of cranial elements, phylogenetic analyses have placed them near the base of the anseriforms or, most recently, resolved them as stem galliforms. In this study, we use morphometric methods to comprehensively describe dromornithid endocranial morphology for the ﬁrst time, comparing Ilbandornis woodburnei and three species of Dromornis to one another and to four species of extant basal galloanseres. We reveal that major endocranial reconﬁguration was associated with cranial foreshortening in a temporal series along the Dromornis lineage. Five key differences are evident between the brain morphology of Ilbandornis and Dromornis , relating to the medial wulst, the ventral eminence of the caudoventral telencephalon, and morphology of the metencephalon (cerebellum + pons). Additionally, dromornithid brains display distinctive dorsal (rostral position of the wulst), and ventral morphology (form of the maxillomandibular [V 2 +V 3 ], glossopharyngeal [IX], and vagus [X] cranial nerves), supporting hypotheses that dromornithids are more closely related to basal galliforms than anseriforms. Functional interpretations suggest that dromornithids were specialised herbivores that likely possessed well-developed stereoscopic depth perception, were diurnal and targeted a soft browse trophic niche.


Introduction
The dromornithids were large flightless birds, collectively known as 'mihirungs', whose fossils are a distinctive component of the Cenozoic avifauna of Australia, and are sometimes comparatively abundant in the Australian Neogene fossil record [1,2]. The greatest diversity of the group occurs during the Miocene [1,[3][4][5][6], but the family is known from fossils dating from the Palaeogene, with a record consisting of a mould of fossil footprints from the Eocene of Queensland [7], postcranial remains from the late Oligocene Pwerte Marnte Marnte Local Fauna (LF) in the Northern Territory [4], and trackways, probably made by dromornithids, reported from the late Oligocene of Tasmania [1,8].
The fossil record shows that the characteristic morphology of dromornithids had already evolved by the late Oligocene, and that it changed little over the next~20 million years (Ma) until the group became extinct in the late Pleistocene [6,9]. The first dromornithid named was Dromornis australis Owen, 1872, from undated deposits at Peak Downs, Queensland [1,6,[10][11][12][13]. Eight species in four genera of dromornithids are now recognised: Stirling and Zietz [14] named Genyornis newtoni Stirling and Zietz, 1896 from Lake Callabonna, South Australia, from what was originally thought to be late Pliocene to early Pleistocene [14] (p. 177), but more recently proposed to be middle to late Pleistocene [15] (p. 16) deposits. Rich [11] described D. stirtoni Rich, 1979, Ilbandornis lawsoni Rich, 1979 and I. woodburnei Rich, 1979 from the late Miocene Waite Formation, Alcoota, Northern Territory, and Barawertornis tedfordi Rich, 1979 from the late Oligocene to early Miocene Carl Creek Limestone at Riversleigh, Northern Queensland. She also named Bullockornis planei Rich, 1979 from the middle Miocene Camfield beds, Bullock Creek, Northern Territory [11] (p. 27), but the genus Bullockornis was later synonymised with Dromornis [12]. This was supported by Worthy et al. [6] upon revision of cranial material of the Bullock Creek specimens, in conjunction with the description of Dromornis murrayi Worthy et al., 2016 from Riversleigh. At the same time, Worthy et al. [6] proposed the eight dromornithid species formed two lineages, where the Dromornis lineage is monotypic throughout its range, and includes in a temporal succession D. murrayi, D. planei, D. stirtoni, and D. australis. The Ilbandornis/Barawertornis lineage comprises the more gracile taxa B. tedfordi, I. lawsoni, I. woodburnei, and G. newtoni.
Dromornithids were long considered to be ratites [11,14,[16][17][18]. All ratites exhibit reduced wing morphology and are generally large, flightless birds [15,19,20]. These features are shared with dromornithids, but Olson [21] (p. 104) succinctly opined that "large size and flightlessness do not a ratite make", and pointed out that characteristics of the dromornithid mandible, quadrate and pelvis suggested that they were likely derived from an entirely different group of birds. In more recent times, with the discovery of additional cranial elements, phylogenetic analyses by Murray and Megirian [3] concluded that dromornithids were the sister-group of the Anhimidae, and so were Anseriformes. This conclusion was reinforced by Murray and Vickers-Rich [2], who also found similarities to Anseranatidae, a closer sister-group to anatids within Anseriformes. In a phylogenetic study of the affinities of Pelagornithidae (bony-toothed birds), Mayr [22] suggested that dromornithids were likely stem Galloanseres, i.e., a sister group to Galliformes and Anseriformes. A relationship more distant from Anseriformes was also found by Worthy et al. [23], whereby with inclusion of a representative sample of extant galloanseres, representative Neoaves and palaeognaths, and key fossil taxa, Dromornis was found to have a stem-galliform relationship. Most recently, Worthy et al. [24,25] used an expanded taxon set, and resolved dromornithids as the sister group to the flightless gastornithids of Eurasia and North America, forming a galloansere clade termed Gastornithiformes Stejneger, 1885. However, the relationship of this clade to either galliforms or anseriforms within Galloanserae, was poorly resolved.
As with gastornithids [26,27], there exists convincing evidence for a herbivorous diet in dromornithids, as some specimens have been preserved with gastroliths [28] (p. 79). Individual stones, presumed to be gastroliths, are common in the fossiliferous silty unit of the Waite Formation producing the Alcoota LF wherein dromornithids are abundant ( [2] (p. 262), [29] (p. 164), [30] (Figure 8A)). Undoubted gastroliths in the form of complete or partial gizzard stone sets, are known from several specimens of the Pleistocene dromornithid Genyornis newtoni, collected on recent expeditions by THW et al.
Handley et al. [31] demonstrated significant male dominated sexual dimorphism in the largest of all dromornithids, the Miocene species Dromornis stirtoni, and revealed those birds identified as male had a mean mass of 528 kg based on tibiotarsus circumference metrics. The tibiotarsus was preferred for estimating body mass in large birds, especially in dromornithids, after statistical evaluations of several mass estimation algorithms applied across a large sample, showed that femoral metrics likely overestimated body mass for them [31,32]. These dromornithids, along with the giant aepyornithid morphotype Vorombe titan (Andrews, 1894), from the Holocene of Madagascar [33], likely comprise the largest birds to have ever evolved.
Worthy et al. [6] identified that from the Oligocene through the late Miocene, the shape of crania of dromornithids changed, with a foreshortening of the length relative to the height of the cranium (e.g., Figure 1). How the shape of the dromornithid brain changed to accommodate these temporal changes in cranial anatomy and whether there exists quantifiable differences in endocranial anatomy between the Dromornis and Ilbandornis lineages have yet to be appropriately assessed. Missing cranial areas are shown by orange stippled lines on (A) and (E). Abbreviations: ct, cavum tympanicum; ep, exoccipital prominence; mm, millimetres; orb, orbit; pb, processus basipterygoidei; po, processus postorbitalis; pp, processus paroccipitalis; pz, processus zygomaticus; rp, rostrum parasphenoidalis; rq, recessus quadratica; zfc, zona flexoria craniofacialis. Scale bars equal 40 mm.
The objectives of this study are to assess dromornithid endocast material spanning the late Oligocene to the late Miocene. This will: (1) allow a comprehensive description of morphological characteristics of the dromornithid brain and its principal innervation for the first time; (2) inform our understanding of how dromornithid brains differ morphologically from those of other galloanseres; (3) assess how Dromornis and Ilbandornis differ in endocranial anatomy; (4) assess how brain shape responded to significant changes in cranial anatomy through time; (5) identify potential functional constraints shaping the evolution of dromornithid endocranial anatomy across this period; and (6) aid in resolving the phylogenetic position of dromornithids.
Much disparate terminology has been used for the description of surface morphology of the avian cranium and brain, and in some instances with no consensus for precedence of any particular term over another (e.g., see [47,48], and references therein). We agree with Early et al. [50] who considered the term 'wulst' (wlst, Figure 2, Figure 3B-D and Figure 4C) appropriate to describe the external dorsal eminences of the internal hyperpallium [49] ( Figure 6), and lobus opticus (optic lobe; opt, Figures 2, 3B and 4B,D) to describe the external ventral eminences of the internal tectum opticus (optic tectum), or tectum mesencephali (mesencephalon). Additionally, we use cerebrum pars frontalis (rostral telencephalon; tel.r, Figure 3B,D and Figure 4C,D; see also Section 2.5), to describe the surface topology of dorsorostrolateral mesopallium and nidopallium [49] (Figure 6), rostrad of the arteria cerebralis medialis (medial cerebral artery; acm, Figure 3A-B and Figure 4B,D) dorsoventral traversal of the hemispherium telencephali (telencephalic hemisphere); cerebrum pars parietalis (caudal telencephalon; tel.c, Figure 2, Figure 3B-D and Figure 4C-D), to describe the surface topology of the mesopallium, nidopallium and arcopallium [49] (Figure 6), forming part of the caudolateral telencephalon pallial complex caudad of the traversal of the medial cerebral artery (see Section 2.5 below); ganglion trigeminale (trigeminal ganglia, tri.g, Figures 2, 3D and 4A,D) to describe ganglia of the trigeminal nerve (V) complex (see Section 3.1.3 below), inserting on the ventral surface of the optic lobe.
The crania of D. murrayi, D. stirtoni and I. woodburnei were medical X-ray CT scanned using the Siemens Somatom Force CT instrument located at the SAHMRI facility in Adelaide. Acquisition CT data were captured at a slice thickness of 0.4 mm, but with the application of an oversampling technique allowing the acquisition of twice the number of slices per detector row, an effective resolution of 240 µm was achieved for all specimens, excluding D. stirtoni (NTM P3250) which was CT scanned at a resolution of 320 µm. Acquisition data were reconstructed by M. Korlaet of Dr Jones and Partners using Siemens proprietary software.
The cranium of D. planei was scanned at the ANSTO nuclear facilities in Sydney using the DINGO neutron CT instrument, located in the OPAL reactor beam hall on thermal beam HB2. Neutron CT acquisition data were captured at low-intensity mode at 95 µm resolution and were reconstructed by Dr. J. Bevitt of ANSTO. All CT data used were isotropic.

Three-Dimensional (3D) Surface Model Construction
Three-dimensional (3D) surface model construction was conducted via segmentation using Materialise Mimics v18 software, and 3D surface stereolithograph (STL) endocast models were produced from reconstructed CT data to represent the shape of the brain (Figures 3 and 4). These included the base and immediate stem of the major nerves passing from the cranium into the brain (see also Section 3.1, Section 4.4.1 and SI Section S3.1.1). Surface STL models were exported to Materialise Mimics 3-Matic v10 software for reconstruction and remeshing.

Model Reconstructions
In many fossils, structures are often lost or damaged by taphonomic processes over time, or during recovery. Where specimens are somewhat bilaterally symmetrical, as is the case of endocasts, damaged or missing structures may be digitally reconstructed based on preservation of one side, or parts of a particular endocast. The incomplete endocasts for the two fossil specimens of D. stirtoni NTM P5420 and NTM P3250 constrained their interpretation. Thus, a two-dimensional (2D) reconstruction representative of the species was derived using both endocast models (see SI Figure S8 and Section S2.1). Similarly, endocasts for specimens of D. murrayi were, respectively, damaged and incomplete, where QM F57984 preserves only the left hand side (LHS) dorsolateral endocast, and QM F57974 preserves only the ventral endocast. Consequently, a single 3D endocast model was compiled from CT data of the two specimens of D. murrayi (see SI Figure S9 and Section S2.2).

Remeshing
Remeshing of 3D STL surface models is required to optimise the quality of the triangles comprising the surface mesh, and to reduce the file size of models for landmarking (see below). Remeshing operations were carried out in Materialise 3-Matic v10 and passing of remeshed STL objects to Polygon File Format (PLY) was conducted in MeshLab v2016.12 [52].

Landmarking
To capture the shape of the endocasts, we defined seven morphological zones using semilandmark (Slm) patches, or modules (see Figure 2; SI Figures S10 and S11, Section S3.2). This approach allowed us to assess whether each morphological zone differed in the same way between taxa and along a lineage, or whether the zones differed in contrasting ways across specimens.
Additionally, the dorsal endocast of dromornithids is dominated by wulst structures (e.g., Figures 2 and 3), which are distinct with respect to those of the extant galloanseres (e.g., wlst, Figure 3 versus [vs] Figure 4, and Section 3.2.2 below). We prefer the explanation that the expansion, or hypertrophy, of the wulst in dromornithids has effectively masked the surface morphology of the rostrodorsal telencephalon (see Section 3.2.1 below). This required the segregation of telencephalic hemispheres into rostral and caudal regions, defined by the traversal of the medial cerebral artery (see acm, Figure 4B, and SI Sections S3.2.1.3-S3.2.1.5), such that telencephalic regions caudal of the medial cerebral artery might be compared across all specimens.
Digital landmarking of 3D endocast surface models was conducted in IDAV Landmark v3.6 [53] using 20 fixed (type 1) and 460 semi-(type 3) landmarks [54] for a total of 480. The set of landmarks (Lms) and semilandmarks (Slms) comprising the modules (Figure 2; SI Figures S10 and S11) were the basis for all subsequent shape assessment. Endocast landmarking protocols and descriptions of the full Modular landmark (Lm) suite are given in Supplementary Information (SI, Section S3).

Data
We used Slm modules to capture the shape of discrete regions of the brain (see Section 2.5 above). Derived from those shape data (see Section 2.6.1), we computed Measurement data (see Section 2.6.2), and acquired Surface Area data based on the 'footprint' of Slm modules as defined (see Section 2.6.3), with the exception of trigeminal ganglion data, where Measurement data were computed from trigeminal ganglia Slm modules (e.g., Figure 10C and Figure 11C), and Surface Area data were computed from the truncated faces of the maxillomandibular (V 2 +V 3 ; see Figure 2E, Figure 3B and SI Figure S5L) branch of the trigeminal (V) nerve complex (see Figures 3D and 4A,D). Collectively, those three forms of data informed the systematic assessment of morphological differences between the endocasts of individual dromornithid specimens (see Section 2.7.2, Section 3.3 and SI Section S4), and the comparison of dromornithid endocast morphology with those of the extant galloanseres (see Section 3.4 and SI Section S5).

Modular Lm Data
Three-dimensional digital shape data derived from the Modular Lm suite (Figure 2; SI Figures S3, S10 and S11) were used for all assessments of shape (see Section 2.7 below).

Measurement Data
Measurement data were calculated between Lm and Slm locations along specific transects for each specimen, using the 'interlmkdist' function in Geomorph v3.1.3 [55]; see also Section 2.7 below. Measurements for the length and width of each modular structure, capturing the directional 'curve' over a 3D surface, were calculated by adding together the distances between each Slm forming the measurement vector (see Figure 2E). Data for each paired structure (i.e., wulst, rostral and caudal telencephalon, optic lobe, and trigeminal ganglion modules) were combined, and mean Measurement values calculated (see Table 1A). Additionally, measurements describing gross endocast morphological 'vector' distances were calculated between two Lm or Slm locations (see Figure 2B,F; Table 1A). Size-standardised mean Measurement log shape ratios were calculated by the log shape ratios method [56], where species Measurement values were divided by species endocast total volume values and log 10 transformed [57] (p. 99), [58] (p. 117). The log shape ratio approach produces size-standardised shape variables from univariate data and is analogous with the Procrustes superimposition method for multivariate Lm data (i.e., GPA; see Section 2.7.1 below), where both methods correct for size while retaining shape variation [59] (p. 1389). Measurement log shape ratios are presented in text, in Table 1B, and plotted in SI Figures S12 and S13. Surface Area data for each endocast module as defined here (see SI Section S3.2) were computed directly from the surface of each endocast using MeshLab v2016.12 (see Figure 2A). Two forms of Surface Area data were acquired: (1) total endocast Surface Area; and (2) modular Surface Area in square millimetres (mm 2 ), from which mean Surface Area values for all bilateral modules (i.e., wulst, rostral and caudal telencephalon, optic lobe, and trigeminal ganglion face) were computed (see Table 1C). Size-standardised mean Surface Area log shape ratios were calculated by the log shape ratios method, where species Mean Surface Area values were divided by species endocast Surface Area values and log 10 transformed (see Section 2.6.2 above). Surface Area log shape ratios are presented text, in Table 1D, and plotted in SI Figure S14.
Additionally, due to taphonomic processes over some 14 Ma, the caudoventral endocast of Dromornis planei had suffered somewhat of a rostrocaudal ventral rotation, along with a subtle rostrally orientated 'twisting' of the caudoventral endocast with respect to dorsal endocast surfaces. This can be appreciated in the slight caudal displacement of LHS optic lobe margins in the D. planei endocast, when observed from the ventral aspect (see Figures 2F and 3D below), and in the D. planei cranium itself (see Figure 1D above). During landmarking, optic lobe and trigeminal ganglion module margins on the ventral D. planei endocast were situated with respect to existing morphological boundaries (see SI Section S3.2.1.9), without attempting to adjust margins for the subtle taphonomic distortion present in the endocast. Subsequent univariate Measurement and Surface Area data for the optic lobe and trigeminal ganglia formed the primary focus for downstream morphological assessments, as the computation of mean log shape ratio values from those paired modular data (see Section 2.6.2 above), accommodated for the subtle bilateral misalignment of ventral midbrain regions in the D. planei endocast.

Three-Dimensional Modular Shape Variation Plots
To better understand the extent of particular morphological variation between species of dromornithid, we used 3D shape variation plots to visualise the modular shape variation between individual specimens. One dromornithid species represented by black dots and blue links, is superimposed over another represented by grey dots and grey links, visualising the extent and direction of modular shape variation between the two specimens. In this manner, the morphological differences between the endocasts of each species of dromornithid were described (see Figure 5, Section 3.3 and SI Section S4).

Nervus Olfactorius
The olfactory (I) nerve transmits rostrocaudally into the bulbus olfactorius (olfactory bulb; bo, Figures 3D and 4D), through the bony foramen n. olfactorii (folf, SI Figures S4K-S4L, S5K) of the rostrodorsal cranium. The dromornithid olfactory bulb is best described by the more complete RHS lateral view of D. murrayi ( Figure 3E). The margins of the olfactory bulb are pronounced both dorsally and ventrally, but caudodorsally masked by the rostral eminence of the wulst. The caudomediolateral transmission of the olfactory bulb margins are shown by the ventral view of D. murrayi and D. planei ( Figure 3D,H, respectively) as transitioning into the rostroventral endocast without reduction in mediolateral width, as seen most notably in A. cornuta and A. semipalmata (e.g., Figure 4C,D,O,P, respectively).

Nervus Opticus
The optic (II) nerves form the chiasma opticum caudally, and pass through the os laterosphenoidale, forming the caudomedial wall of the orbit via the foramen opticum (fopt, SI Figures S4K,L and S5K), meeting rostrally at the septum interorbitale. In dromornithids, the interorbital septum rostral of the foramen opticum is robust, and the optic (II) nerves divides rostrolaterally into two well defined branches (i.e., II; Figures 3A and 4A).

Nervus Trigeminus
The trigeminal (V) nerve complex comprises three divisions. The medial or ophthalmic branch carries nervus ophthalmicus (ophthalmic [V 1 ] nerve; Figures 3B and 4B), transmitting to the trigeminal ganglia (see Section 3.2.5 below) on the ventral surfaces of the optic lobe (see Section 3.2.4 below), through the foramen n. ophthalmici (foph; SI Figures S4K,L and S5K). The foramen n. ophthalmici opens into the "lacerate (presphenoid) fossa" sensu [6] ( Figure 1D), located ventrolaterad from the foramen opticum, between the os laterosphenoidale, os basisphenoidale, os parasphenoidale and septum interorbitale (os laterosphenoidale complex) of the caudomedial wall of the orbit. In dromornithids, the foramen n. ophthalmici is paired with the foramen n. abducentis, which transmits n. abducens (abducent [VI] nerve, see below). The lateral or maxillomandibular (V 2 +V 3 ) branch of the trigeminal ganglion complex, carries n. maxillaris (maxillary [V 2 ] nerve) and n. mandibularis (mandibular [V 3 ] nerve), both of which enter the skull rostroventrolaterally at the foramen n. maxillomandibularis (fmx, SI Figures S4K,L and S5K), a single opening between the prootic and laterosphenoid bones of the skull. In dromornithids, the maxillomandibular branch is distinctive in that it is markedly elongate, compared to the extant galloanseres assessed, and transmits the maxillary (V 2 ) and mandibular (V 3 ) cranial nerves minimally 20 mm caudoventrolaterally (in Dromornis), after entering the skull at the foramen n. maxillomandibularis (see also V 2 + V 3 , Figure 3B, SI Figure S5L; and Section 3.2.5 below).

Nervus Abducens
The abducent (VI) nerve inserts on the rostroventral rhombencephalon and is transmitted caudoventrally through the osseus canalis n. abducentis, after entering the skull at the foramen n. abducentis. In dromornithids, the foramen n. abducentis is paired with the foramen n. ophthalmici, forming a single bi-lobal foramen in the rostromedial os laterosphenoidale structures of the orbit (fa, SI Figures S4K,L and S5K; VI, Figure 3B and SI Figure S5L; and above).
The proximal ganglion is enclosed in the fovea ganglii vagoglossopharyngealis in the lamina parasphenoidalis of the fossa cranii caudalis, between the exoccipital and opisthotic bones. The glossopharyngeal (IX) nerve enters the cranium caudoventrolaterad from the vagus (X) nerve at the foramen n. glossopharyngeus (fg; SI Figure S4M,N), situated in the fossa parabasalis. In dromornithids, the glossopharyngeal (IX) and vagus (X) nerves separate caudoventrolaterad of the eminence of the proximal ganglion from the rhombencephalon surface ( Figure 3D). The separation of the glossopharyngeal (IX) and vagus (X) nerves in dromornithids is similar to, but occurs to some extent further distally, than the condition seen in Anhima cornuta ( Figure 4D), Gallus gallus ( Figure 4H), and Anseranas semipalmata ( Figure 4P), but distinct to that seen in Leipoa ocellata ( Figure 4L).

Rostral Telencephalon
The external morphology of the rostral telencephalon as defined here (see SI Figures S10 and S11 and Section S3.2.1.3), evident rostrodorsally of the medial cerebral artery (acm, see Figure 3A,B and Figure 4B,D) in both anseriforms, and somewhat less so in the galliforms, is notably absent in dromornithids. We prefer the explanation that the rostral telencephalon in dromornithids have been masked by hypertrophy of the wulst (see below), and that the external remnants of rostral telencephalon morphology are only visible, in rostral aspect, as twin eminences ventromediolaterad of the olfactory bulb, on either side of the rostromedial surface of the dromornithid endocast (e.g., tel.r; Figure 3B,D).

Wulst
The wulst in dromornithids are greatly hypertrophied and dominate the entire dorsal endocast morphology (see wlst, Figure 3B,D). They extend rostromediolaterally to effectively mask the olfactory bulbs (see Section 3.1.1 above), and extend rostroventrally over the most rostral eminence of the rostral telencephalon, substantially overhanging the rostroventral surface of the brain when viewed from the ventral aspect ( Figure 3D,H,L). The wulst extend rostrolaterally and mask the rostrodorsal telencephalon (see above, Figure 3), obscuring the rostrodorsal endocast morphology visible in the extant galloanseres (e.g., Figure 4). The wulst extend mediolaterally across the entire dorsal forebrain, to the lateral vallecula transition zones delimiting the boundaries between the mediolateral wulst, and the dorsolateral caudal telencephalon (see Section 3.2.3 below). The vallecula transition zones are well defined, as the wulst are strongly dorsolaterally expanded in those areas (va, Figure 3B,C). Caudodorsally, the wulst grade into the cerebrum pars occipitalis (coc, Figure 3C) in the region of the medial glandula pinealis (gp, Figure 3A,C), rostrolaterad of the dorsomedial cerebellum ( Figure 3C). Notably, dromornithid wulst structures are located somewhat rostrally on the dorsal endocast (e.g., Figure 3C,G,K), such that they do not overlap the rostromedial eminence of the cerebellum when viewed from the lateral aspect (e.g., Figure 3A,E,I); this is similar to the condition in A. cornuta ( Figure 4A,C), G. gallus ( Figure 4E,G), L. ocellata ( Figure 4I,K), and distinct to the condition seen in the anseriform A. semipalmata ( Figure 4M,O).

Caudal Telencephalon
The caudal telencephalon in dromornithids are well defined and are delimited from the wulst by the vallecula transition zone dorsolaterally (va, Figure 3B,C). The mediolateral hypertrophy of the caudal telencephalon begins approximately where the medial cerebral artery traverses the telencephalic hemisphere dorsoventrolaterally (acm, Figure 3A,B). They extend ventrolaterally, approximately level (dorsoventrally) with the fissura subhemispherica (fs, Figure 3B), and return medially, somewhat acutely, to grade into the ventrolateral optic lobe (see below). Caudally, the cerebrum pars occipitalis (coc, Figure 3C) forming part of the dorsal caudolateral caudal telencephalon, grades into the dorsorostrolateral pons and medulla oblongata structures forming the overall metencephalon complex rostromediad of the cerebellum, in the vicinity of the glandula pinealis dorsolaterally, and medially at the rostromediolateral metencephalon.

Optic Lobe
The optic lobes in dromornithids are somewhat visually inconspicuous structures in comparison to those in the extant galloanseres (opt, Figure 3B vs Figure 4B,F,J,N). They are defined by a slight lateral hypertrophy of the ventromedial endocast ventrolaterad of the fissura subhemispherica, and rostrally by transition into the caudolateral chiasma opticum and tractus opticus structures. Caudally, the optic lobes grade into the ventromediolateral metencephalon complex (see below, Figures 3 and 4).

Trigeminal Ganglia
Trigeminal ganglia receive the three divisions of the trigeminal nerve (V, see Section 3.1.3 above), and insert on the ventral surface of the optic lobe (tri.g, Figures 3D and 4A,D). The trigeminal ganglia form part of the Modular Slm suite, therefore, are described here (see SI Figures S10 and S11, Section 2.6.1 and Section S3.2.1.9). In dromornithids, the medial portion of the trigeminal ganglion carrying the ophthalmic nerve (V 1 ) separates from the lateral branch carrying the maxillary (V 2 ) and mandibular (V 3 ) nerves and exhibits a small ganglionic bridge between the two primary eminences ( Figure 3D,H,L). The characteristics of dromornithid trigeminal ganglia are distinctive in that the maxillomandibular branch transmits minimally 20 mm caudoventrolaterally (in Dromornis), after entering the skull at the foramen n. maxillomandibularis.

Cerebellum
The cerebellum, as reflected in the endocast in dromornithids, is compressed rostrocaudally and expanded mediolaterally ( Figure 3C,G,K), and more so in species of Dromornis. From the lateral aspect (see Figure 3A,E,I), the dorsal rostroventral surface forms a shelf somewhat level with the dorsal lateral semicircular duct (dsl, Figure 3B,D) of the vestibular organ (lv, Figure 3A), before turning sharply ventrally in the vicinity of the dorsolateral auricula cerebelli, to grade into the caudodorsal medulla spinalis at the osseous foramen magnum. Overall, the exposed dromornithid cerebellum and the associated ventral rhombencephalon (medulla oblongata + pons; see below), compared with the extant galloanseres, form a comparatively distinctive hind brain in these birds.

Rhombencephalon
Rhombencephalon is the collective term describing the structures of the medulla oblongata and pons, forming the caudoventrolateral areas of the hindbrain. In dromornithids, the ventral rhombencephalon surface is somewhat flat rostrocaudally and mediolaterally (i.e., not as ventrally curved as in other galloansere specimens, e.g., Figure 4), and extends further rostrocaudally than it does mediolaterally.

Key Morphological Differences between Species of Dromornis and Ilbandornis
In the following, we present a summary of results for Modular Lm data, Measurement data, and Surface Area data forms. Full descriptions of results are given in SI (see SI Sections S4 and S5).

Wulst Modules
The overall size and surface area of the wulst increased between the late Oligocene and the middle Miocene in species of Dromornis, particularly rostrodorsally, and in the ventrolateral displacement of the vallecula transition zones between the dorsolateral wulst and the dorsal caudal telencephalon (see SI Section S4.1.1). The wulst in D. stirtoni likely had a comparable dorsal profile to other species of Dromornis, in that the rostrocaudal and mediolateral profile is distinctively hypertrophied (see SI Section S4.2.1). The late Oligocene D. murrayi and middle Miocene I. woodburnei dromornithids have similar length ratios, but the mediolateral width ratio is markedly smaller in D. murrayi. Similarly, results for Surface Area data show that D. murrayi has a smaller wulst surface area ratio than I. woodburnei (see SI Section S4.4.1).
With respect to the middle Miocene dromornithids, D. planei shows greater rostrodorsal hypertrophy of the wulst, and the dorsolateral margin of the vallecula transition zones are more ventrolaterally located than in I. woodburnei. D. planei has greater overall rostrocaudal length and caudal mediolateral width ratios than I. woodburnei. These results are consistent with those for Surface Area data, which show D. planei has a larger surface area ratio than I. woodburnei (see SI Section S4.3.1). Ilbandornis woodburnei displays a deeper fissura interhemispherica transition zone, placing the medial wulst margins closer together than in both species of Dromornis assessed.

Caudal Telencephalon Modules
The overall size and surface area of the caudal telencephalon reduced between the late Oligocene and the middle Miocene in species of Dromornis (see SI Section S4.1.2), and between species of Dromornis and Ilbandornis (see SI Sections S4.3.2 and S4.4.2). However, the characteristic shape of the structures are maintained within the late Oligocene through middle Miocene species of Dromornis. The caudal telencephalon of the late Miocene D. stirtoni is mediolaterally hypertrophied and is similar to the other species of Dromornis (see SI Section S4.2.2). The main differences in the caudal telencephalon between species of Dromornis and I. woodburnei, are that caudoventral margins project further ventrally in species of Dromornis, this is reflected in caudal telencephalon dorsoventral width ratios being greater in species of Dromornis than in Ilbandornis.

Cerebellum Module
There was limited rostrodorsal hypertrophy in the cerebellum between the late Oligocene and the middle Miocene in species of Dromornis. The dorsal surface of the cerebellum in the late Miocene D. stirtoni has a similar rostrocaudal shape to the other species of Dromornis, but the dorsal cerebellum appears more ventrally orientated in D. stirtoni (see SI Section S4. . It is notable, however, that there exist rostrocaudolateral profile similarities between the cerebellum of the middle Miocene D. planei and I. woodburnei specimens, that differ from the late Oligocene D. murrayi. Where both middle Miocene species display relative hypertrophy of the rostrodorsal through dorsomedial cerebellum, affecting a rostrocaudally steeper caudal transition to the dorsal medulla spinalis, compared to the older Dromornis murrayi.

Rhombencephalon Module
There was a dorsal displacement of the rhombencephalon surface between the late Oligocene and the middle Miocene in species of Dromornis, and the development of an overall 'flatter' mediolateral and rostrocaudal rhombencephalon profile in D. planei. As noted for the cerebellum above, the rhombencephalon structure in D. stirtoni appears to be more ventrally situated than that of D. planei. These observations suggest that the apparent ventral displacement of the dorsal cerebellum and ventral rhombencephalon surfaces in D. stirtoni, may reflect a compensatory ventral rotation of the hindbrain complex, with respect to the rostrodorsal rotation evident in the forebrain (see SI Section S4.2.4). In effect, the brain of D. stirtoni appears rotated about the median plane, whilst the position of the dorsomedial surfaces of the brain have been maintained, effectively foreshortening the overall rostrocaudal length of the D. stirtoni endocast compared with other species of Dromornis.
The overall length of the rhombencephalon between species of Dromornis and Ilbandornis are remarkably similar. However, in species of Dromornis, the rhombencephalon describes a more 'flat' rostrocaudal profile, along with greater mediolateral width ratios compared to I. woodburnei. The surface area ratio of the rhombencephalon in D. planei is markedly larger than that of I. woodburnei (see SI Section S4.3.4), and the late-Oligocene D. murrayi has a larger ratio compared with the middle Miocene I. woodburnei (see SI Section S4.4.4). These trends are similar to those noted in the comparisons of D. planei and I. woodburnei (see SI Section S4.3.4, and above), and suggest that the size of the cerebellum may have increased, particularly in the rostromedial zone, across both Dromornis and Ilbandornis lineages between the late Oligocene and the middle Miocene. However, this observation may only be tested by assessment of yet to be discovered cranial material of the Ilbandornis lineage from the late Oligocene.

Innervation
Unlike in dromornithids, the olfactory zones of extant galloanseres are wholly external to the rostral telencephalon, as revealed by their slight constriction immediately rostrad of the rostral telencephalon, particularly evident in A. semipalmata ( Figure 4M,O,P) and A. cornuta ( Figure 4A,C,D). The hypoglossal (XII) nerves have a single origin at either side of the caudoventral medulla oblongata in dromornithids, this condition is unlike the extant galloanseres, which display rostral (XIIr) and caudal (XIIc) rami ( Figure 4D,H,L,P). Results for Measurement data show the trigeminal ganglia of A. semipalmata and A. cornuta are wider than those of all other galloanseres. Surface Area data show the galliforms L. ocellata and G. gallus have the smallest surface area ratios for the face of the maxillomandibular (V 2 + V 3 ) branch of the trigeminal (V) nerve, closely followed by the dromornithids. The anseriforms A. semipalmata and A. cornuta, respectively, have the largest maxillomandibular surface area ratios for all galloanseres assessed (see SI Section S5.1).

Wulst Modules
Wulst modules in extant galloanseres are much hypotrophied in comparison with all dromornithids (e.g., Figure 3 vs Figure 4). Results for Measurement data show G. gallus has the shortest and narrowest wulst ratios, followed by L. ocellata, and the anseriforms A. cornuta and A. semipalmata, respectively. Dromornithids have substantially larger wulst surface area ratios than all galloanseres, but when endocast absolute size is accounted for, G. gallus and L. ocellata have the largest wulst surface area ratios among the extant galloanseres, and the anseriforms A. cornuta and A. semipalmata the smallest, respectively (see SI Section S5.2).

Rostral Telencephalon Modules
Evidence of the rostral telencephalon is conspicuously absent on the external morphology of dromornithid endocasts, and only present rostrally as twin eminences ventromedi-olaterad of the olfactory bulb, on either side of the rostromedial surface of the endocast (tel.r, Figure 3B,D). Consequently, detailed comparisons were not possible between the rostral telencephalon of dromornithids and those of the extant galloanseres, and we present results for the extant galloanseres only. Measurement data show G. gallus has the shortest and narrowest rostral telencephalon ratios, followed by L. ocellata, and the anseriforms A. cornuta and A. semipalmata, respectively. Similarly, results for Surface Area data show that G. gallus has the smallest surface area ratio, and A. semipalmata the largest surface area ratio of all extant galloanseres (see SI Section S5.3).

Caudal Telencephalon Modules
Results for Measurement data show that all dromornithids have relatively greater rostrocaudal length ratios compared with the extant galloanseres. However, all extant galloanseres overlap in dorsoventral width ratios. Dromornithids have the greater mediolateral endocast total width ratios of all galloanseres, and the galliforms have somewhat similar endocast total width ratios as the anseriform A. semipalmata.
Results for Surface Area data show that the dromornithids have the largest surface area ratios of all specimens, L. ocellata has the smallest surface area ratio of all specimens, and the phasianid G. gallus has the largest surface area ratio of the extant galloanseres (see SI Section S5.4).

Optic Lobe Modules
Results for Measurement data show the extant galliforms have the largest optic lobe length and width ratios, followed by the anseriforms and dromornithids. Results for Surface Area data (see SI Section S5.5) show that the largest taxa have the smallest overall optic lobe ratios, and reveal the optic lobe in I. woodburnei is larger than those of both species of Dromornis. Surface Area ratios for the anseriforms A. cornuta and A. semipalmata are somewhat similar to those of the dromornithids, and the galliforms G. gallus and L. ocellata have the largest surface area ratios of all specimens assessed.

Cerebellum Module
Results for Measurement data show the cerebellum, as exposed in the endocast in dromornithids, are notably rostrocaudally shorter and mediolaterally wider than those of all extant galloanseres, with the width ratio for A. semipalmata most approaching those of the dromornithids. Rostrad of the foramen magnum, all extant galloanseres display a gradual rostrolateral divergence of cerebellum mediolateral margins, prior to grading into the cerebrum pars occipitalis regions of the caudal telencephalon. This condition is not evident in dromornithids, which display a more abrupt mediolateral divergence of caudodorsal cerebellum margins, which are more pronounced in species of Dromornis than in I. woodburnei. Results for Surface Area data show that the dromornithids have the largest ratios of all galloanseres. Among the extant galloanseres, A. semipalmata has the smallest surface area ratio for all specimens. Leipoa ocellata overlaps with those of the dromornithids, and is most similar to that of I. woodburnei. Anhima cornuta has a larger surface area ratio than all dromornithids, and the G. gallus has the largest surface area ratio of all galloanseres (see SI Section S5.6).

Rhombencephalon Module
Results for Measurement data show that all galloanseres have somewhat similar rhombencephalon length ratios, with the exception of A. cornuta, which has the largest rostrocaudal length ratio of all galloanseres. Rhombencephalon width ratios for extant galloanseres are all larger than in dromornithids, and likely reflect the greater ventral projection, or eminence, of the rhombencephalon in the extant specimens. The overall mediolateral width ratio of the hindbrain in species of Dromornis are greater than in all galloanseres assessed, and that of I. woodburnei overlaps with those of the extant galloanseres. Results for Surface Area data show that the dromornithids I. woodburnei and D. murrayi have the smallest rhombencephalon surface area ratios, followed by A. semipalmata and G. gallus. The dromornithid D. planei has a somewhat similar surface area ratio as G. gallus, and A. cornuta and L. ocellata have the largest surface area ratios among all galloanseres (see SI Section S5.7).

Discussion
Dromornithid endocranial anatomy is described here in detail for the first time, using specimens from fossil sites in Australia spanning~20-8 Ma. In the following, (1) we summarise the morphological characteristics of the dromornithid brain, and discuss these features with respect to those of exemplars of other major galloansere clades; (2) we review potential lineage-specific morphological characteristics identified in dromornithid endocranial anatomy; (3) we comment on morphological changes observed to have occurred in the endocranial anatomy of the Dromornis lineage over time; (4) we assess potential functional implications of the dromornithid endocranial condition.

Olfactory Bulb
The olfactory bulb in dromornithids is pronounced both dorsally and ventrally in the oldest (Dromornis murrayi) species. However, in middle Miocene specimens (D. planei and Ilbandornis woodburnei), its dorsal morphology is masked by the rostral eminence of the wulst (see Section 4.1.3 below). There appears no reduction in the size of the olfactory bulb in the younger dromornithids, as from the ventral aspect, lateral margins of the organ transition into the rostroventral endocast without reduction in mediolateral width. In extant galloanseres, the olfactory bulb of Anseranas semipalmata displays hypertrophy in excess of that seen in Anhima cornuta, which is somewhat more than those of Leipoa ocellata and Gallus gallus. The olfactory zones of these galloanseres appear wholly external rostrad of the rostral telencephalon, as the olfactory bulb margins constrict somewhat prior to grading caudally into the rostral telencephalon, a condition distinct from those of dromornithids.
Taken together, the evidence shows that the wulst extends rostromediolaterally to effectively mask the olfactory bulbs in dromornithids, and so contrary to a first assessment that might consider that dromornithids had hypotrophied olfactory bulbs, we consider them to be relatively no smaller than in the other galloanseres.

Trigeminal Ganglia
Trigeminal ganglia transmit the three divisions of the trigeminal (V) nerve complex and insert on the ventral surface of the optic lobe. They are distinctive in dromornithids, in that the maxillomandibular (V 2 + V 3 ) branch passes minimally 20 mm caudoventrolaterally through the unusually thick cranium (in Dromornis), after entering it at the foramen n. maxillomandibularis. The extended transmission of the maxillomandibular branch through the cranium in dromornithids accommodates for the unusually thick honeycomblike trabecular bone separating the cortical bone defining the endocranial capsule, from the outer cortical cranial surface (e.g., SI Figure S8I). In the extant galloanseres, the transmission of these nerves from the foramen n. maxillomandibularis is markedly shorter. Notably, however, in the phasianid galliform G. gallus (see Figure 4E,F,H), the transmission of the maxillomandibular (V 2 + V 3 ) branch through the cranium, although somewhat shorter than that observed in dromornithids, is longer than observed in the other extant galloanseres. Moreover, CT data for G. gallus reveals that this is also accompanied by trabecular matrix surrounding the endocranial capsule through which the nerves transmit [Pers. Obs. Authors].
In general, the shape of the avian cranium, particularly dorsally, exhibits a close relationship to the shape of the brain within [68][69][70][71][72][73][74]. However, patterns of the brain size 'lagging behind' the body, accompanying an increase in the relative thickness of trabecular bone surrounding the endocranial capsule have been recognised in Haast's eagle Hieraaetus moorei (Haast, 1872) by Scofield and Ashwell [75], who showed that the eagle's "ten-fold" increase in body size was only accompanied by a "doubling or tripling" of endocast volume. This demonstrated lag of neuroanatomy behind rapid hypertrophic skeletal changes in a taxon, may relate to strong selection for body size, i.e., adaptation to a novel trophic niche, or artificial selection in the form of "stringent" human mediated selection for desirable phenotypes (e.g., as in the case of G. gallus see [76], and references therein). Consequently, as dromornithids became larger through the course of their evolution (e.g., [6,31]), it is likely the increase in physical size of the cranium, was accommodated for by an increase in trabecular bone enclosing the 'lagging' endocranial capsule [75] (Figure 5a).
Similarly, the G. gallus skull used for CT scanning for this project was almost certainly from a domestic chicken, and may demonstrate increasing trabecular thickness in the cranium, mediated by human selection for body size. To assess these observations more comprehensively, additional data in the form of wider sampling across galloanseres in particular, but across Neornithes in general is required, targeting taxa with demonstrated temporal increases in body size.
Results for Measurement data show the trigeminal ganglia of A. semipalmata and A. cornuta are relatively wider than those of all other galloanseres, and Surface Area data show the galliforms L. ocellata and G. gallus have the smallest surface area ratios for the face of the maxillomandibular (V 2 + V 3 ) branch of the trigeminal (V) nerve, closely followed by the dromornithids. The anseriforms A. semipalmata and A. cornuta, respectively, have the largest maxillomandibular surface area ratios for all galloanseres assessed (see Section 3.4.1). The eminence of the hypoglossal (XII) nerves which typically comprise a rostral (XIIr) and caudal (XIIc) branch in extant galloanseres, is represented by one nerve root at either side of the caudoventrolateral medulla oblongata in dromornithids, with bifurcation into rostral and caudal branches occurring within the os exoccipitale, close to the caudomediolateral surface of the cranium.

Wulst
The morphology of the wulst is the most distinguishing feature uniting dromornithids, forming massively hypertrophied structures. The wulst extends rostromediolaterally to effectively cover the olfactory bulbs and extends rostroventrally over the most rostral eminence of the rostral telencephalon, substantially overhanging the rostroventral surfaces of the brain. The dromornithid wulst extend strongly rostrolaterally, masking the rostrodorsal telencephalon (see Section 4.1.4 below), and extends mediolaterally across the entire dorsal forebrain. The structure of the wulst in dromornithids is unlike any seen in the extant galloanseres. Results show the dromornithids all display much larger width and length, and surface area ratios, than all the extant galloanseres. When endocast absolute size is accounted for, G. gallus and L. ocellata have the largest wulst surface area ratios among the extant galloanseres, and the anseriforms A. cornuta and A. semipalmata the smallest (see Section 3.4.2).
Among taxa that were not included in these analyses, some palaeognaths display hypertrophy of the wulst.  (Owen, 1846), A. didiformis, Euryapteryx curtus gravis (Owen, 1870), and Emeus crassus (Owen, 1846) as well, all of which show wulst characteristics similar to those seen in dromornithids, wherein the vallecula, especially in the larger moa taxa, visibly extend rostrocaudally across the entire dorsolateral telencephalic hemispheres. These characteristics suggest that such characteristic hypertrophy of the wulst in large flightless birds (see also [78] (Figures 1 and 2), [79] (Figure 2a), [80] (Figures 1 and 3), [81] (Figure 1), [74] (Figures 5.3 1-6), may represent a parallel convergent modification towards enhanced visual proficiency and stereoscopic capability (see also Section 4.4.2.1 below). However, ratite taxa clearly display a lesser degree of wulst hypertrophy than that evident in dromornithids. Even the oldest dromornithid cranial fossils (e.g., D. murrayi from the~20 Ma sites of Riversleigh), display greater hypertrophy of the wulst than seen in any ratite, indicating a long dromornithid "ghost lineage" must have existed prior to any substantive fossil evidence of dromornithids in Australia (see also [6] (p. 19)). Evidence for such, comprise trackways reported from the late Oligocene of Tasmania [1], postcranial remains from the late Oligocene Pwerte Marnte Marnte LF in the Northern Territory [4], and a mould of fossil footprints from the Eocene Redbank Plains Formation of Queensland [7].

Rostral Telencephalon
Evidence of the rostral telencephalon in dromornithids is only present rostroventrally as twin eminences ventromediolaterad of the olfactory bulb, on either side of the rostromedial endocast (tel.r; Figure 3B,D). It is possible that these eminences are expanded cerebrum tuber ventrolaterale structures, as evident in A. cornuta ( Figure 4A,B,D). However, their interpretation as remnant rostral telencephalon eminences is favoured, as in all dromornithid endocasts modelled, there exist pronounced paired eminences in this rostromedial zone that are not present to the same degree in any avian endocast modelled or observed in the literature [Pers. Obs. Authors]. Further support for this, is that the positioning of these rostral eminences in dromornithids, agrees with the angle and position of rostral telencephalon eminences in specimens, when endocasts are aligned to putative "alert posture" [82][83][84][85], with reference to the horizontal positioning of the lateral semicircular canal of the vestibular organ. Additionally, the rostrocaudal transition angle of the vallecula, describing the dorsal margins of the caudal telencephalon, agree with the extension of the visible rostral eminences of the dromornithid rostral telencephalon, should the dorsolateral curve of the rostral telencephalon not be masked by hypertrophy of the rostromediolateral wulst. In support of this interpretation, the apparent rostral extension of the vallecula across the dorsolateral surface of moa endocasts figured by Ashwell and Scofield ([77] Figures 5E−5I and 6G-6I) and Corfield et al. ([49] Figure 1b D-E), and similarly in the brains of extant flightless ratites (see [78] (Figures 1 and 2), [79] (Figure 2a), [77] ( Figure 6), [49] (Figure 1B,C), [80] (Figures 1 and 3), [81] (Figure 1), [74] (Figures 5.3 1-6)), suggest that the evolution of a rostromediolaterally hypertrophied wulst in large flightless birds may effectively mask rostral telencephalon morphology. The accommodation of this apparently major change in rostrodorsal endocranial morphology in dromornithids, would have necessitated a dorsomedial displacement of the olfactory bulb, which we think is a possibility, as this condition is somewhat similar to that seen in moa (e.g., [77] (Figures 5e-l and 6g-l), [49] (Figure 1b D-E)).
Detailed comparisons of the dromornithid rostral telencephalon were not possible. However, results of comparisons among extant galloanseres show the phasianid G. gallus has the smallest rostral telencephalon, followed by the megapodiid L. ocellata, and the basal anseriform A. cornuta. The anseriform A. semipalmata has the most hypertrophied rostral telencephalon of the extant galloanseres (see Section 3.4.3).

Caudal Telencephalon
The caudal telencephalon is strongly defined in all galloanseres assessed. Results show dromornithids have relatively greater rostrocaudal length and mediolateral endocast total width ratios than the extant galloanseres. Similarly, the dromornithids have the largest caudal telencephalon surface area ratios, and the megapodiid L. ocellata has the smallest surface area ratio of all specimens. The phasianid G. gallus has the largest surface area ratio for the extant galloanseres (see Section 3.4.4).

Optic Lobe
The optic lobe in dromornithids appear somewhat indistinct, and not as well delimited as in the extant galloanseres. This is confirmed by results revealing the extant galliforms have the largest optic lobe length and width ratios, followed by the anseriforms and dromornithids. These patterns are also reflected by results for Surface Area data which show the dromornithids have the smallest optic lobe surface area ratios, followed by those of the anseriforms. The largest surface area ratios for all galloanseres assessed are shown by the galliforms G. gallus and L. ocellata. Overall, results show that when endocast size is accounted for, the largest overall taxa display the smallest overall optic lobe ratios, and reveal relatively greater hypertrophy of the optic lobe in the Ilbandornis lineage compared with species of Dromornis (see Section 3.4.5).
Additionally, during the process of segmenting the endocast models used for this project, we retained trigeminal ganglion structures (see Figure 3D,H,L) inserting on the ventral surfaces of the optic lobe, primarily to gain insight of dromornithid somatosensory capabilities (e.g., Section 4.4.1 below). This, however, resulted in the ventromedial surfaces of optic lobe Slm modules being constrained by the morphology of the trigeminal ganglia. Retrospectively, it may have been more useful to segment the trigeminal (V) complex out of the final endocast models, enabling access to the full extent of the ventral optic lobes and allowing consideration of optic lobe morphology with respect to those of the wulst and optic foramen (sensu [50]; see also [86], and (Section 4.4.2.3 below). This aspect of dromornithid cranial morphology forms part of continuing work on the brains of these giant birds.

Cerebellum
Dorsal cerebellum margins in dromornithids are characteristically rostrocaudally compressed and mediolaterally expanded. Results show that the exposed cerebellum in dromornithid endocasts are rostrocaudally shorter and mediolaterally wider than those of all extant galloanseres. Results for Surface Area data show that the galliform G. gallus has the largest cerebellum ratio, and the anseriform A. semipalmata the smallest ratio of all galloanseres assessed. Dromornis planei has the largest cerebellum surface area ratio of the dromornithids, and the ratios for A. cornuta and L. ocellata overlap with those of dromornithids. Rostrad of the foramen magnum, all extant galloanseres display a gradual rostrolateral divergence of cerebellum mediolateral margins, prior to grading into the cerebrum pars occipitalis regions of the caudal telencephalon. This condition is not evident in dromornithids, which display a much more abrupt mediolateral divergence of the caudodorsal cerebellum margins, and this is more pronounced in species of Dromornis than in Ilbandornis. Progressing rostrally, dromornithid cerebellum margins describe a somewhat parallel rostrocaudal transition (see Section 3.4.6).

Rhombencephalon
Dromornithids display relatively flat ventral rhombencephalon surfaces rostrocaudally and mediolaterally, whereas, in comparison, all extant galloanseres show much ventrally hypertrophied rhombencephalon surfaces (rho; Figure 4A,B,D). These trends are reflected by results for Measurement data showing all galloansere species, with the exception of A. cornuta, have similar rhombencephalon length ratios. Width ratios for the extant galloanseres are all larger than those of the dromornithids, and likely reflect the greater ventral projection, or eminence, of the rhombencephalon in the extant taxa. The overall width ratio of the hindbrain in species of Dromornis, are greater than in all other species, and that of I. woodburnei overlaps with those of the extant galloanseres. Results for Surface Area data show that the dromornithids I. woodburnei and D. murrayi have the smallest surface area ratios, followed by the anseranatid A. semipalmata and phasianid G. gallus. Dromornis planei has a somewhat similar surface area ratio as G. gallus, and A. cornuta and L. ocellata have the largest surface area ratios among all galloanseres assessed (see Section 3.4.7).

Endocranial Morphology Distinguishing Lineages within Dromornithids
The examination of dromornithid endocasts has revealed morphological features of the dromornithid brain that may provide support for the two-lineage hypothesis (sensu [6]). We define these distinctive endocranial attributes below, with a focus on the morphological differences between Ilbandornis woodburnei and the two species of Dromornis with adequate preservation for comparison (i.e., D. planei and D. murrayi): (1) the medial boundaries of the wulst in the rostrocaudal fissura interhemispherica zone of the dorsal endocast in I. woodburnei, are notably closer together than in species of Dromornis; (2) the most ventral eminence of the I. woodburnei caudal telencephalon, in the zone of the fissura subhemispherica, are conspicuously less ventrally pronounced than in species of Dromornis; (3) the rostrodorsal cerebellum in I. woodburnei defines a more rostrally projecting mediolateral curve than in species of Dromornis, which display a flatter rostral dorsomediolateral margin; (4) the caudodorsal part of the cerebellum in I. woodburnei projects further caudally in the region of the dorsal medulla spinalis, whereas this region is caudally shorter but mediolaterally flatter in species of Dromornis; (5) the entire hindbrain (rhombencephalon, medulla oblongata and metencephalon complex [cerebellum + pons]) in I. woodburnei, is rostrocaudally longer and mediolaterally narrower than in species of Dromornis.
These morphological traits are potential lineage-specific endocranial attributes, but require confirmation through a shared presence in yet to be discovered crania of Ilbandornis lawsoni and Barawertornis tedfordi, or in presently undescribed material of Genyornis newtoni.

Temporal Changes in the Endocranial Morphology of the Dromornis Lineage
Across the~10 Ma period represented by specimens of Oligo-Miocene D. murrayi, and the middle Miocene D. planei, the orientation of the brain within the skull appears to have remained much the same, despite foreshortening of the cranium. Other than regional changes in endocast shape, for example, the rostrodorsal hypertrophy of the wulst, accompanying other trends described above, the brains of these species are generally similar, and distinctively dromornithid. The relatively major increase in the overall size of species of Dromornis, reflected by crania of D. murrayi and D. planei figured here (e.g., Figure 1A-D; SI Figures S1A-P and S4A-H), and postcranial fossils described elsewhere [1,6], appear to have not been substantial enough to affect changes in the position and orientation of the brain in D. planei relative to that of D. murrayi. However, by the late Miocene, somẽ 6 Ma after the occurrence of D. planei, the cranium of the Dromornis lineage had evolved to become even more foreshortened and dorsoventrally deeper, as manifested in Dromornis stirtoni ( Figure 1E; SI Figures S6 and S7; see also [6]). These morphological changes appear to have affected the orientation of the brain.
With regard to D. stirtoni, it is unfortunate that the state of preservation of specimens prevented the level of endocast shape assessment achieved for the other dromornithids. This was primarily due to the taphonomic characteristics of the only site preserving these giant birds (see Section 2.2 above). In turn, this limited the biological and functional inferences derived from their exceptional cranial architecture. However, we have shown from the similarity of preserved features of the D. stirtoni endocast models with those of D. planei (see SI Sections S2.1 and S4.2), that the brain of D. stirtoni does not depart greatly from the only other known dromornithid endocast morphology. However, the altered endocranial alignment to 'fit' the brain in the foreshortened cranium, resulted in rostroventral endocast surfaces in D. stirtoni being rotated rostrodorsally around the medial caudal telencephalon into a more dorsally orientated position. Additionally, it appears the brain of D. stirtoni has experienced a measure of dorsoventral compression and mediolateral expansion. These changes in the forebrain are accompanied by a more ventrally orientated hindbrain, which may reflect a compensatory ventral rotation of the hindbrain complex in the species, although the 'life position' of the midbrain in the skull of D. stirtoni has not changed appreciably from that of D. planei.
The reasons for this unusual rotation and apparent subtle compression of the D. stirtoni endocast may lie in the highly derived state of cranial morphology attained by this, the largest of the dromornithids, by the late Miocene. The cranium of D. stirtoni is unique in that the rostrocaudal cranial length is effectively about half the cranial depth, and represents the terminal state of a concerted trend in cranial foreshortening, along with an increase in bill size, of the most extreme avian cranial specialisation known [3,6]. This trend extends from the oldest known species of Dromornis, the Oligo-Miocene D. murrayi, through the middle Miocene D. planei, to the most derived late Miocene taxon D. stirtoni.

Functional Implications of Dromornithid Endocranial Morphology
Jerison [87] (p. 8) proposed the "Principle of Proper Mass" which specifies particular sensory specialisations in the vertebrate brain are correlated with concomitant hypertrophy of the neural tissue processing related information, and that the relative mass of functional neural tissue implies the relative importance of those functions in the species. Subsequent studies showed that overall brain size was in fact increased by independent hypertrophy of particular brain regions (see [88][89][90][91], and references therein), supporting Jerison's [87] observations, which became to be known as the "mosaic" model of brain evolution (sensu [90]).
That mosaic evolution characterises some, but not all, of avian brain composition has been demonstrated by several works (see [49,92], and references therein). It is acknowledged that the brain is not strictly compartmentalised into regions that process exclusive neuronal input, but rather includes levels of interconnectivity across the whole structure [92]. It is clear that particular brain nuclei share greater levels of neuronal connectivity associated with specific functions; that a hypertrophied brain region reflects a greater level of "information-processing power", and that these patterns are somewhat indicative of functional specialisation ( [49,90,[93][94][95] and references therein). Most recently, Early et al. [50] showed there exists statistically significant correlations between the external surface area of the wulst and optic lobe, and the volume of underlying brain regions.

Innervation
Characteristic mosaic correlations in the trigeminal system have previously been shown in several vertebrate taxa (see [96], and references therein). In birds, the trigeminal (V) nerve system comprises the medial portion carrying the ophthalmic (V 1 ) nerve which innervates the orbit and nasal cavity, the rostral palate and the tip of the upper bill, and forms a major sensory pathway for the skin of the head and maxillary rostrum. The maxillary (V 2 ) branch innervates the maxillary rostrum and infraorbital regions, and the mandibular (V 3 ) division innervates the entire lower bill and several mandibular and interramal regions [97][98][99][100]. The trigeminal nucleus receives exclusively proprioceptive information from the descending tract, and the principal sensory nucleus of the trigeminal system [96]. This includes not only projections from ophthalmic (V 1 ) and maxillomandibular (V 2 + V 3 ) nerves, but taste information from the tongue is conveyed, within the lingual branch of the maxillomandibular (V 2 + V 3 ) ramus, by the facial (VII) nerve to the trigeminal principal sensory nucleus, which also receives input from glossopharyngeal (IX) and hypoglossal (XII) nerves [93,98,[101][102][103][104][105]. In short, the trigeminal (V) nerve comprises the largest somatosensory cranial innervation complex, and transmits epicritic sensation from the entire facial region and mastication musculature [98,105,106]. Dubbeldam [107] proposed that differences in the trigeminal principal sensory nucleus were indicative of the functional demands of specific feeding behaviours. Gutiérrez-Ibáñez et al. [96] reported hypertrophy of the trigeminal principal sensory nucleus in species dependent on tactile input for feeding, and that bill morphology and the concentration of mechanoreceptors in the bill and tongue, strongly correlate with feeding behaviour.
The glossopharyngeal (IX) and vagus (X) nerves share the large proximal ganglion. The glossopharyngeal (IX) components of this complex comprise somatic, "special" and visceral afferent fibres. The special fibres connect with the palatine branch of the facial (VII) nerve at the cranial cervical ganglion, and are associated with sensory taste and tactile information [98,101,108,109]. The general visceral efferent fibres of the glossopharyngeal (IX) nerve supply the oesophagus and crop, exhibit size variability across taxa that show greater "distensibility" of the oesophagus [98], and are notably hypotrophied in taxa that have no crop (i.e., owls and hawks). The glossopharyngeal (IX) nerve complex bifurcates, after separation from the vagus (X) nerve at the proximal ganglion, and transmits, in two main afferent branches of the lingual and the laryngopharyngeal nerves, as the descending oesophageal nerves, innervating the tongue and the laryngeal muscles, respectively.
The vagus (X) nerve complex is the most extensive of the sensory and motor cranial nerves, wherein there are two groups of motor fibres. The first consists of "general" visceral efferent fibres which innervate the muscles and glands of the thoracoabdominal viscera, including the heart and lungs, etc., and is associated with circulation, respiration and digestion control [98]. The second consists of "special" visceral efferent fibres innervating the muscles of the pharynx and the larynx, reached via branches of the glossopharyngeal (IX) nerve ( [98]; for a contrary opinion, see Wild [103] who argued vagus (X) projections are exclusively cardiovascular and pulmonary in function).
The morphology of dromornithid ventral cranial innervation, in the form of the maxillomandibular (V 2 + V 3 ), glossopharyngeal (IX) and vagus (X) nerves, and results presented here for trigeminal ganglia, are more similar to the species of galliforms. Dromornithid trigeminal ganglia appear less hypertrophied than those of the more derived, and trophic specialist anseriforms too, this is supported by results for Measurement data. Additionally, Surface Area data describing the truncated face of the maxillomandibular (V 2 + V 3 ) branch of the trigeminal (V) nerve complex, show that the dromornithids have somewhat similar surface area ratios as the galliforms, and which collectively, are much smaller than those of the anseriforms (see Sections 3.4.1 and 4.1.2). The morphology of dromornithid ventral innervation, better represented by the endocast of Dromornis planei (see Figure 3A-D), suggests that dromornithids were likely possessed of somewhat better tactile capability than the extant galliforms (e.g., Figure 4H-L), and were likely employing greater levels of sensory input from the bill, palate, and tongue in their trophic behaviour, than the extant galliforms do.
Dromornithids, especially species of Dromornis, have extremely large, deep bills, with dorsally prominent, mediolaterally compressed culmens [1,3,6]. The herbivorous diet of dromornithids is well established ( [1], and references therein), but musculature for operation of the bill is "surprisingly limited" [6] (p. 19), and suggests that these birds were not capable of a particularly forceful bite [6], contra [3] (p. 88). For example, there is no temporal fossa on the side of the cranium for insertion of mandibular musculature, which is thus limited to the fused postorbital-zygomatic processes, and hyper-developed insertions on the orbital wall of the cranium (i.e., ma, SI Figure S4K,L). Moreover, the culmen, while large, has a lightly constructed osseous core that was only partially covered in rhamphotheca (at least it was in species of Dromornis), was highly vascularised and likely highly innervated, a combination of features conferring relatively weak biting ability [6]. The large size of the dromornithid culmen, combined with indications that they are not strengthened for food manipulation, suggest that they were primarily utilised for display, and that the distinctive morphology was likely driven by sexual selection, or by thermoregulatory requisites.
Bill architecture suggests that dromornithids were likely not consuming coarse browse requiring strong bite forces, as were some species of moa (see [19], and references therein). This contention can be tested by observations of the collections of gastroliths used to process such food. We were able to analyse total volume of gastroliths and size of stones in complete or partial gizzard stone sets from specimens of Genyornis newtoni to infer characteristics of diet [110][111][112][113], and showed (see SI Figure S15 and Table S1) that dromornithids selected gastroliths of much smaller diameter, and accumulated them in surprisingly small overall volumes compared with Dinornis moa, although the dromornithids were somewhat larger birds. Moreover, the data reveal that G. newtoni selected smaller stones than did the smaller extant emu (Dromaius novaehollandiae) which occasionally selected remarkably large gastroliths (see also [110] (Table 2), [111] (p. 26)). In fact, when overall body size was accounted for, gastrolith size ratios for the emu were larger than those of Dinornis robustus, although accumulated in smaller quantities. Notably, the stout-legged moa (Euryapteryx curtus gravis) has very similar gastrolith size and gizzard mass ratios to those of G. newtoni. Stout-legged moa are hypothesised to have exploited a diet of soft leaves and fruit, in dry scrubland and mosaic environments, in contrast to the generalist tree and shrub browsing D. robustus moa, the gizzard contents of which have been shown to comprise much coarser, low-quality fibrous leaf and twig material (e.g., [15,[114][115][116] and references therein). Complete dromornithid gizzard stone sets are rarely preserved, but a complete gizzard set is known for the Oligo-Miocene Dromornis murrayi, preserved in situ in limestone (see [28] (p. 79)), and which shows only small stones. The gizzard sets measured for the Pleistocene species Genyornis newtoni, show they selected similarly small stones. Thus, if gizzard mass and gastrolith size preferences for G. newtoni are representative of dromornithids in general, these results suggest that the fibrosity of browse the dromornithids were targeting was likely similar to that of stout-legged moa (i.e., new growth, soft leaves, and fruit), requiring less vigorous mechanical processing in the gizzard [112]. Such a diet would require specialised visual pathways for the identification and selection of specific food resources. We have shown that dromornithids may have had well-developed somatosensory and sensorimotor capabilities, and so were likely adapted for precise and selective visual browsing.

Visual Pathways
There are three principal visual pathways in birds: (1) the thalamofugal pathway transmits visual signals from the retina via the optic lobe, to the principal optic nucleus of the dorsal thalmus, and thence to the wulst; (2) the tectofugal pathway transmits via the optic lobe to the nucleus rotundas of the thalmus, and proceeds to the entopallium of the caudal telencephalon; and (3) the third visual pathway transmits via the optic lobe through retinal recipient nuclei in the accessory optic system and pretectum, and projects to several regions of the brain, including the cerebellum (see [49,[117][118][119][120], and references therein).

Wulst
The wulst are composed of two main regions, the larger "visual" region, located dorsally and extending caudodorsally, which receives retinal projections, and a smaller rostral somatosensory region, receiving "substantial" somatosensory and kinesthetic input [94,106,[121][122][123]. The thalamofugal pathway, incorporating the wulst, has been shown to be primarily involved in binocular vision capability, and global stereopsis ( [94,120,[124][125][126] and references therein). Iwaniuk et al. [94] showed that the size of the wulst was significantly correlated with more frontally orientated orbits and broader binocular fields [120,127], and argued changes in the relative size of the wulst were indicative of increases in somatosensory and motor processing capabilities [48,124,128,129]. Additionally, wulst structures are hypertrophied in species that forage using tactile information from the bill [120,124,130,131].
Dromornithids may have had a well-developed thalamofugal pathway as they display particularly hypertrophied wulst structures. Potential indications of the kinds of adaptive selection driving dromornithid wulst morphology, may lie in the extraordinarily similar morphology of avian predators such as Barn owls (Tyto alba), and the Australian Tawny frogmouth (Podargus strigoides), which have a similarly hypertrophied wulst structures too (e.g., [124] (Plate 26), [120] (Figure 2), [79] (Figure 2c), [49] (Figure 1bK), [132] ( Figure 3A,B), [74] (Figures 5.3 34 and 35)). Barn owls are nocturnal, and possess exceptional low-light visual proficiency and binocularity, or stereopsis [133][134][135][136][137]. Similar specialisations typical of low-light and stereoptic visual proficiency have been recognised in the Tawny frogmouth, which have highly developed visual systems [120,124], and are thought to possess stereoscopic vision [125]. Stereoptic proficiency has been shown to facilitate accuracy in nocturnal prey capture in caprimulgid taxa [125], and spatial, or "topographical cues" associated with feeding activities in Columba and Gallus [126], and corvids [131,138]. Furthermore, it has been advanced that taxa which use tactile information for foraging (see also Section 4.4.2.2 below) show somewhat hypertrophied rostral wulst structures [120,124], likely as the mandibular (V 3 ) division of the trigeminal (V) cranial nerve, innervating the lower bill (see Section 4.4.1 above), terminates in the rostrodorsal mesopallial regions of the brain [99,130,139,140]. As such, the development of stereopsis in birds has been linked with the presence of a well-developed wulst, evidence of which is proposed as compelling indications for the presence of stereoscopic vision specialties in fossil material [125] (p. 220).
Therefore, future analyses of dromornithid orbit size, optic foramen and skull orientation may provide additional insights into the hypertrophy noted in wulst structures for these birds. Cranial anatomy unambiguously shows that dromornithids had large, forward facing eyes (see Figure 1; SI Figures S1, S4 and S6; see also [6] (Table 1)). However, inference as to whether dromornithid retinal topography was structurally adapted (i.e., corneal diameter, retinal cell density/type, etc.), for sensitivity to low light conditions, may only be made by interpretation of orbit shape and size from skeletal remains.
In a large study assessing the relationship between corneal diameter and axial length of the avian eye, Hall and Ross [141] showed that species adapted to light-limited (scotopic or crepuscular) habitats, have larger corneal diameters and axial lengths, relative to those active during well-lit (photopic or diurnal) conditions. Hall [142] showed that there exists a close relationship between corneal diameter and axial length of the eye, and that metrics describing the osseous structures of the orbit were "well associated" with photic activity in birds. In support of these observations, several additional studies have shown that in nocturnal birds, eye shape increases relative to skull length, they display larger orbit diameters relative to depth, and orbits are more frontally orientated (e.g., [94,143,144] and references therein, but see also [131]). Animals that exploit low-light environments have evolved in one of two ways: (1) by enlargement and orientation of the visual system (i.e., increasing orbit size and binocular overlap); or (2) they develop enhanced sensitivity of somatosensory and tactile systems (e.g., [144], and references therein). Among ratites, kiwi are the only confirmed nocturnal taxon, and have evolved reduced eye size and distinct endocranial morphology associated with the somatosensory and tactile systems strategy (i.e., (2) see [79,145]). All other ratites are diurnal, as were the extinct NZ moa [77], and inspection of the shape of their brains (e.g., [75] (Figure 2), [79] (Figure 2a,b), [77] (Figure 6g-l), [145] (Figure 1b A-E), [80] (Figures 1 and 3), [81] (Figure 1), [74] (Figures 5.3 1-6)), show rostromediolaterally hypertrophied wulst structures in these large flightless birds, which may represent evolution for enhanced visual proficiency (i.e., strategy (1) above). However, no ratites display the massively hypertrophied state of the wulst seen in dromornithid dorsal endocasts.
Martin [131] advanced that binocular vision in birds may be used for the inspection of food items held in the bill, for bill control during the process of foraging and food provision to chicks, and not primarily for stereopsis. He also argued that binocular vision in the control of locomotion is a secondary function, as spatial information may be provided by each eye independently. Considering dromornithid cranial morphology displays large, forward facing orbits, and their dorsal endocranial morphology is dominated by the wulst, these birds probably adopted the strategy of enlargement and orientation of the visual system to develop good stereoscopic vision and accurate depth perception. This would preadapt them to being specialised browsers capable of selecting individual fruit and leaves from within complex browse. Additionally, such combined features of cranial and endocranial morphology also raises the possibility that dromornithids were adaptively selected for low-light visual proficiency along the nocturnal-diurnal gradient (scotopic sensu [142,146]). However, we prefer the more 'parsimonious' explanation that features of dromornithid cranial and endocranial anatomy are more likely associated with foraging and locomotion within complex diurnal environments.

Cerebrum
The rostral and caudal telencephalon-the nido-and mesopallial structures of the cerebrum-are recognised to form a complex with "integrative" functions (see [93], and references therein). Thus, we describe functional interpretations for the cerebrum as a whole.
The internal structure of the dorsal and rostrocaudolateral cerebrum incorporates four main subdivisions: the hyperpallium, incorporating the wulst, the mesopallium, incorporating the rostro-and caudodorsal telencephalon, the nidopallium, incorporating the rostro-and caudolateral telencephalon, and the arcopallium incorporating the caudoventral telencephalon. The rostromediolateral telencephalon incorporates the medial striatopallidal complex (striatum + pallidum), overlain rostrocaudally and dorsolaterally by the nidopallium, which is similarly overlain rostrocaudally and dorsolaterally by the mesopallium (see [48] (Figure 1C), [49] (Figure 6)). Dubbeldam and Visser [147] showed that the caudolateral nidopallium receives arcopallial afferents sourced through the medial nidopallium, and contains a complex pattern of terminal fields. They identified a strong connection between the striatopallidal complex and the mediolateral nidopallium, arguing that in the mallard (Anas platyrhynchos) there exists two major telencephalic circuits: one relaying through the pallidum, and the other through the striatum (e.g., [48] (Figure 3)), and that these afferent circuits play a major role in feeding behaviour. The telencephalon has been associated with a wide range of behaviours including feeding, taste, tactile sense, taste discrimination, vocalisation, and with high levels of cognition and complex tasks ( [49], and references therein). Furthermore, stereotyped species-specific behaviour [93,148], pecking accuracy [149], and the processing of visual information such as brightness, colour, and pattern discrimination [118], have been attributed to processes within the caudolateral telencephalon. Pettigrew and Frost [130] showed that the maxillary (V 2 ) division of the trigeminal (V) nerve, which innervates the upper bill (see Section 4.4.1 above), transmits to extensive terminal fields in the region of the rostrodorsal mesopallium (e.g., [150] (Figure 3b), see also [139]). Similarly, Dubbeldam et al. [99] showed that ascending maxillary (V 2 ) and mandibular (V 3 ) trigeminal (V) projections transmitted rostrodorsally, via the nucleus basalis, to mesopallial terminal fields [140]. These sensorimotor and somatosensory projections were related to the "detection" of food particles, particularly in low-visibility feeding in anseriforms [151][152][153][154], and food grasping in columbiforms [140,155] and passeriforms [156]. Located in the caudal telencephalon, the medial spiriform nucleus receives projections from both somatosensory and visual parts of the wulst (see [121]; and Section 4.4.2.1 above), along with arcopallial projections associated with "motor planning information" from the caudoventral telencephalon, and projects to the cerebellum (see Section 4.4.2.4 below), forming the "telencephalic-cerebellar loop" (sensu [150] (p. 9)).
A notable caveat when interpreting differences in caudal telencephalon morphology, is that changes may arise from differential hypertrophy of any of several internal structures, each of which vary in function (e.g., [157], and above). Results show that the caudal telencephalon in dromornithids are strongly hypertrophied, and along with the apparently hypotrophied rostral telencephalon, they are morphologically similar to galliforms in the condition of the cerebrum (see Sections 4.1.4 and 4.1.5). Extant galliforms are generally visual foragers and exploit less somatosensory information associated with trigeminal (V) projections than tactile specialists like anseriforms [152][153][154]. Results show surface area ratios for the face of the maxillomandibular (V 2 + V 3 ) branch of the trigeminal (V) nerve complex in dromornithids are more similar to those of the galliforms, than they are to the anseriforms, which are distinctly larger (see Section 3.4.1). These morphological trends suggest that the tactile capability of dromornithids was likely reasonably well developed, at least comparable with the extant galliforms, but they were certainly not tactile sensory specialists.

Optic Lobe
The optic lobe forms a crucial component of the visual pathway system. Hellmann et al. [158] (p. 395) characterised the optic lobe as a "relay station" for transmitting ascending visual output to the forebrain (see Section 4.4.2.1 above), projecting descending output to the premotor regions of the hindbrain (see Section 4.4.2.4 below), and comprise multiple "retinotopically organised, and functionally specific" cell types. So called "optic flow" (sensu [159]), are retinal stimuli generated by self-motion through stationary environments (see [150,160] and references therein). Optic flow stimuli are analysed by recipient nuclei in the accessory optic system and pretectum, wherein the lentiformis mesencephali, or pretectal nucleus, responds to "moving large-field" visual stimuli and generates optokinetic response for the control of posture, eye movement stabilisation, and compensatory movement during locomotion ( [117,150,[160][161][162][163][164] and references therein), facilitated by processes within the cerebellum [87,165].
Optic lobe structures in dromornithids are somewhat hypotrophied in comparison to those of the extant galloanseres. Iwaniuk et al. [166] showed that the optic lobe of the nocturnal Night parrot (Pezoporus occidentalis) was relatively reduced compared with those of diurnal parrots, but orbit size did not differ between them. As noted above (see Section 4.4.2.1), eye shape, corneal diameter and retinal topography are better predictors low-light sensitivity than eye size alone (e.g., [86,141,142]). What is more, Kay and Kirk [167] (p. 238) showed variation in 'retinal summation', or the ratio of photoreceptor cells in the eye which interact with a single ganglion cell in the optic (II) nerve, is indicative of the level of visual sensitivity, acuity, and photic activity. This implies that large eye size, accompanied by small optic nerve size, is indicative of nocturnal habitus.
These forms of data are often not accessible from fossil material, and as we have shown, there is much variation in overall optic lobe size across known diurnal species of galloanseres assessed here. This implies that without corroborating evidence of eye structure and optic nerve absolute size, the size of the optic lobe alone is likely not a reliable proxy for inferring photic activity for fossil taxa.
Given that optic lobe structures are primarily relay stations, their absolute size is not definitively correlated with either nocturnal or diurnal behaviour (see also [168]) although this can resolved to an extent by comparisons of orbit and foramen opticum measurements. That dromornithids have large, forward facing eyes (see Section 4.4.2.1), robust optic nerves (see Section 3.1.2), and enlarged morphology of trigeminal (V) innervation (see Section 4.4.1), further supports our argument that dromornithids were likely diurnal taxa with a strong reliance on stereopsis and trigeminal (V) somatosensory input. These results also suggest that caution is required when deriving functional or behavioural inference from comparisons of optic lobe absolute size, without accounting for the collective morphology of somatosensory "circuits" (e.g., [150] (p. 9)), such as those of the thalamofugal, tectofugal and third visual pathways (see Section 4.4.2).

Hindbrain
The cerebellum has long been associated with motor integration and posture control in birds [87]. Visual signals are projected through the third visual pathway, via the retinal-recipient nuclei of the optic lobe, to the cerebellum [87,117,165,[169][170][171], where they facilitate obstacle avoidance responses. Additionally, Pakan and Wylie [165] suggested that folia VI-VIII of the cerebellum may be involved in "steering" functions, and Iwaniuk et al. [172] showed that VI and VII folia are hypertrophied in birds they classified as "strong fliers", and showed some evidence to support correlation of hypertrophy of the cerebellum rostral lobe with "strong hindlimbs" in birds. Additionally, due to the dura mater, or dural envelope, which encloses the brain in life, a surface endocast derived from the internal osseous endocranial capsule, may not entirely reflect actual brain morphology, although this is much less of a problem in birds than in other archosaurids (e.g., [71,173] (p. 263), [74] (p. 61)). This is apparent in the lack of folia definition in dromornithid cerebellum morphology presented here, but which was almost certainly present in life. Additionally, it is recognised that an endocast can only capture surface structures of the brain, and that where a brain region is partially occluded by another structure, the endocast will underestimate total size of that region. This is the case for the cerebellum where it is occluded rostrally by the cerebrum. Thus, analyses of the cerebellum only relate to exposed parts of this region, but we consider these were analogous between compared taxa. Therefore in our analyses, the overall shape of the exposed dromornithid cerebellum is captured, and the broad metrics we use to describe the cerebellum illustrate well the differences in dorsal hindbrain shape across specimens.
Results for the dromornithid hindbrain, show the shapes described by the dromornithid metencephalon (cerebellum and rhombencephalon), are certainly distinct from those of the extant galloanseres. Measurement data show that the dromornithid cerebellum and rhombencephalon differ from the extant galloanseres in both length and width ratios. However, results for Surface Area data show the dromornithids overlap with the extant galloanseres in metencephalon surface area ratios. So the dromornithid hindbrain is not particularly dissimilar in overall size to those of the extant taxa, and apparent visual differences between caudal endocasts across specimens may lie in the particularly distinct hypertrophy of the dromornithid forebrain, several ratios of which certainly do differ from those of the extant galloanseres (e.g., Section 4.1.3 above). Additionally, the trend for hypertrophy of the rostrodorsal cerebellum noted in both middle Miocene dromornithid taxa, compared to that of the late Oligocene species (e.g., Section 3.3.3), may represent selection for enhanced neural capability with respect to hindlimb sensorimotor processes, facilitated within the cerebellum rostral lobe [172], as overall body size in dromornithids increased from the late Oligocene through the middle Miocene. Such morphological trends suggest that dromornithids likely maintained the capacity for capable locomotion through complex environments associated with third visual pathway processes in the hindbrain, at least comparable with the extant galliforms assessed here.

Conclusions
In this study, we comprehensively described dromornithid endocranial morphology for the first time, comparing endocasts for three species of Dromornis and Ilbandornis woodburnei to those of four species of extant basal galloanseres spanning the galliformanseriform dichotomy. We compared endocasts of the Oligo-Miocene Dromornis murrayi, the middle Miocene D. planei, and the late Miocene D. stirtoni, to describe endocranial changes associated with neurocranial foreshortening in a temporal series along the Dromornis lineage. The endocasts of species of Dromornis were compared to that of the middle Miocene Ilbandornis woodburnei, revealing minimally five morphological differences between Ilbandornis and Dromornis (e.g., relating to the medial boundaries of the wulst in the rostrocaudal fissura interhemispherica zone; the caudoventral regions of the caudal telencephalon; the morphology of the rostrodorsal and caudodorsal cerebellum; and the morphology of the ventral cerebellum + pons hindbrain complex), collectively supporting the two-lineage hypothesis. Overall, dromornithid brains display morphology supporting hypotheses that they are more closely related to basal galliforms than anseriforms, in that the rostral positioning of the wulst on the dorsal endocast; the form of the maxillomandibular (V 2 + V 3 ), glossopharyngeal (IX) and vagus (X) nerves, and the morphology of trigeminal ganglia, are more similar to the species of galliforms assessed. Functional interpretations suggest that dromornithids were specialised herbivores that likely possessed well-developed stereoscopic depth perception and targeted a soft browse trophic niche.
The retention of a relatively large cerebellum and associated hindbrain morphology in dromornithids, such as those shown by the volant galloanseres assessed, raises the question as to why dromornithids maintained the capacity for capable movement through complex environments (see above). That these functional attributes were selected for during early dromornithid evolution is a possibility.
During the transition from the early Cenozoic through the Eocene, Australia was blanketed by predominantly warm to cool-temperate rainforest, which began to open into sclerophyllous vegetation on higher ground only during the Oligocene (see [174]). These forests became progressively drier during the transition from the Oligocene through the late Miocene, eventuating in the establishment of sclerophyllous fire-sensitive woodlands as the dominant continental floras during the Miocene ( [1], see also [175]). Dromornithid evolution has a long history, and the key characteristics of the dromornithid brain were likely assembled when they first evolved into large flightless birds sometime during the Palaeogene. During those times, dromornithids would have occupied highly complex forested environments, where the capacity for acute stereoscopic vision, the taxon's trophic preferences established, and the capability for navigating complex environments were maintained from their flighted ancestors. Dromornithids likely co-opted these traits in adapting to and exploiting the steadily drying Australian environment through their known temporal range, traits that persisted in the last dromornithid taxon of the late Pleistocene.
During the middle Miocene through to the late Pleistocene, dromornithids were likely increasingly restricted to riparian zones, or sclerophyllous woodland remnants proximal to water courses or temporary water bodies. Dromornithid fossils from these times are almost exclusively recovered from deposits derived from ancient water courses or shallow lakes. For example, their bones are relatively abundant in the fluviatile deposits of Bullock Creek LF, and Alcoota LF in the Northern Territory (see Section 1); and are common in several lacustrine sites around Lake Callabonna in the Eyre basin of Southern Australia, where dromornithid bones are found within the clays of ancient lake margins wherein birds were mired, evidently while in search of water during dry periods (e.g., [176] (p. 208), [15] (p. 3)). Consequently, from the late Miocene in particular, dromornithids may have become increasingly 'tethered' to constricting sclerophyllous woodland remnants in a drying environment, circumstances which, along with temporal overlap with newly arrived humans during the late Pleistocene [177], potentially contributed to the eventual extinction of the group.
The cranial and endocranial morphology of dromornithids is unlike any attained in the evolution of birds, and represents distinct morphological adaptations to progressively changing Australian Cenozoic environments. The specialist browsing diet of these giant birds resulted in the unique juxtaposition of large body size, deep, narrow, and elongate bills, with frontally directed eyes, affording stereoscopic vision on crania bearing minimal musculature, and like much of the idiosyncratic Australian fauna of the past, dromornithids represent combinations of unique adaptations now lost.

Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable. Data Availability Statement: All primary data generated in this study are available in SI and Table 1 above. Image-stack CT data are available on request, see ReadMe.txt file (SI).