The proportion of ipsilateral retinal projections (IRP), the uncrossed fibers in the optic chiasm (OC), varies considerably with species. Most animal groups have laterally situated eyes and no or few IRP. However, for example cats and primates have frontal eyes and high proportions of IRP. The traditional assumption is that the latter configuration is due to the selective advantage of accurate depth perception through stereopsis (Pettigrew, 1986a; Heesy, 2009). Due to the position of the eyes, binocular vision creates two slightly different images, and this disparity provides information that the brain uses to estimate depth in the visual scene (Wheatstone, 1838; Barlow et al., 1967; Heesy, 2009). Binocular vision has been associated with ecological/behavioral factors such as predatory behavior, nocturnalism, and living in trees (Tigges, 1970; Heesy, 2009). It is functionally related to changes in the visuomotor system, high visual acuity, specializations of visual pathways in the brain, elaboration and differentiation of the visual cortex, and with orbital convergence (Heesy, 2009). Many features of primates, including specializations of the visual system, have been explained as adaptations to the occupation of an arboreal niche (Heesy, 2009). Collins (1921) proposed that binocular vision and stereopsis, i.e., reconstructing a three-dimensional world in the cortex using binocular visual information, is necessary for accurate judgment of distance in arboreal locomotion, not least during leaping, where a miscalculation may be deadly. Similarly, the fine-branch niche hypothesis of primate origins postulates that visual specializations for judgment of depth during acrobatic locomotion in a network of typically small branches was vital (Martin, 1979).

Also Crompton (1995) supported that primates' stereoscopic depth perception and orbit convergence developed for accurate judge distance, but added that it may also have been for locating food, such as fruit and insects, in an obscure and complex arboreal environment. Aforementioned hypotheses have in common the necessity for accurate judgment of distance in arboreal substrates to avoid a fatal estimation of distance. However, Heesy (2009) argued that arboreality is not exclusive to primates; some other arboreal mammals, e.g., rodents, have divergent orbits and panoramic visual fields, including flying squirrels, which may fly over 100 m between trees. Cartmill (1992) suggested that high orbit convergence and binocular visual field overlap are adaptations to the nocturnal visual predatory habits of the last common ancestor of all primates. He demonstrated that many predatory mammals relying on vision to follow and grab prey have comparatively high orbit convergence, and postulated “the visual predation hypothesis” i.e., that several characteristics of the primate visual system might have been due to alterations serving in visually based predation. Allman (1977) and Pettigrew (1978) developed the visual predation hypothesis further, suggesting that orbit convergence is associated with the development of suitable optical, and visual axes, in nocturnal taxa. The optical axis is the axis of symmetry through the cornea and lens; the visual axis means the line that fits through the point of fixation, nodal points, and area centralis (or fovea); the orbital axis is the line of symmetry of the orbit (Heesy, 2008). If visual and optic axes are more and more brought into line, the result will be that light from the object of interest can pass through the center of the lens, and close alignment of the optic and visual axes will permit nocturnal animals to view prey and other objects within binocular fields without compromising light-gathering capabilities (Heesy, 2009).

Pettigrew (1986a) expanded his avian hypothesis to include mammals. Regardless of panoramic or binocular visual fields arboreal animals are likely to use multiple cues to perceive depth during locomotion, examples are interposition (an object is partly obscuring another object), perspective, motion parallax, vergence eye movement cues, accommodation, and optic flow (Heesy, 2009). Heesy (2009) proposed that large binocular visual fields, give at least three unique potential visual advantages: enhanced light sensitivity, expanded stereoscopic depth perception, and contrast discrimination. The latter is made possible by summation of two similar images presented to each eye, which helps the brain to distinguish unwanted “noise” from useful information. Stereopsis means the computation of object solidity and depth based on binocular disparity cues (Poggio and Poggio, 1984; Cumming and DeAngelis, 2001). Stereoscopic depth perception is founded on neurons sensitive to binocular disparities. These “binocular neurons” need contribution from each eye (Poggio, 1984) and are responsible for the perception of depth, object solidity, and binocular fusion (Heesy, 2009). Stereoscopic processing in the cortex integrates/fuses percepts of retinal images from each eye into a singular mental visual image (Poggio, 1984; Poggio and Poggio, 1984; Cumming and DeAngelis, 2001). Stereoscopic depth perception precludes binocular visual field overlap since that is the portion of the visual field from which binocular parallax cues are composed; binocular parallax is generated by the slightly different views of objects that are projected onto each retina; the stereopsis hypothesis suggests that these slight discrepancies, termed binocular disparities, provide the cues from which stereoscopic depth is computed (Heesy, 2009).

Parallax is a function of the interocular distance, which limits the functional utility of stereopsis. The maximum range in amphibians and birds is less than 1 m, and mammals also have a short range over which stereopsis functions (Collett, 1977). In humans vergent eye movements occur most frequently in the space corresponding to the arm's length (Viguier et al., 2001). Since the earliest true primates were likely to be very small, with small interocular distance, their stereoscopic range was probably much shorter than their maximal leaping distance, which contradicts that primate binocularity evolved to judge distance for arboreal leaping (Heesy, 2009).

Archeological evidence have suggest that grasping and visual adaptations evolved asynchronously in primates, i.e., adaptations for manual and pedal grasping evolved long before primates' typical visual characteristics (Silcox et al., 2007; Kirk et al., 2008). Silcox et al. (2007) proposed that a shift to increased herbivory diet characterized the origin of primates, which would contradict the visual predation hypothesis. Moreover, modern phylogenies have united tarsiers with anthropoid primates (monkeys, apes, and humans) (Jameson et al., 2011). Studies in tarsiers (Tan and Li, 1999; Melin et al., 2013) have suggested that many hallmarks of the anthropoid visual system had already evolved in the last common ancestor of tarsiers and anthropoid primates, which also challenges the visual predation hypothesis.

Hypotheses about the evolution of bifocal vision/stereopsis are largely restricted to mammals and/or birds. Ward et al. (1995) stated that the variation in IRP among non-mammalian vertebrates is mysterious and lacks a relationship with nocturnalism or predatory lifestyle. A recent eye-forelimb (EF) hypothesis proposed that receiving visual feedback of limb movements in the limb-steering hemisphere was fundamental in the evolution of vertebrate visual systems (Larsson, 2011) and the primary basis for the substantial variation in proportion of crossed and uncrossed retinal ganglion cells (RGC). The aim of this review was to examine OC anatomy and retinal specializations in a variety of vertebrates with respect to the EF hypothesis and to explore associations with motor behavior.

The OC is the structure of the ventral diencephalon midline where RGC axons either cross the midline or return to the ipsilateral hemisphere (Figure 1). Thus, fibers destined for the right or left hemispheres are separated here. The arrangement of the OC is not consistent among species, and shows notable differences in gross architecture (Figure 2) (Jeffery and Erskine, 2005). Developmental mechanisms shaping this region may vary among closely related animals (Jeffery and Erskine, 2005; Herrera and Mason, 2007).

The EF hypothesis postulates that increased IRP is useful in animals with frontal eyes that regularly use their forelimbs frontally, while reduced IRP is functional in animals using the forelimbs laterally (Larsson, 2011). Limb movement is a multimodal event engaging motoneurons, proprioception, and, possibly, touch, and vision. In essentially all vertebrates each cerebral hemisphere receives little visual, tactile, and proprioceptive information from the ipsilateral forelimb, while multisensory information from the contralateral limb is abundant (Figure 3). Since the involved motor nuclei, somatosensory areas, and vision neurons are situated in the same hemisphere, the neuronal pathways involved are relatively short. However, that would not have been the case without evolutionary modifications of crossed/uncrossed projections in the OC. Reduction in the number of interhemispheric connections reduces brain size (Ringo et al., 1994) and speeds processing (Poffenberger, 1912; Berlucchi et al., 1971). This neural architecture creates hemispheres that are autonomous with respect to motor control and perception of limb movements, and multimodal information about a moving forelimb will converge in the same cerebral hemisphere. The inner ear is also primarily connected to the contralateral hemisphere and hence the hemisphere also receives rapid auditory feedback about the limb that it controls. The OC structure may be functional in the storage and recollection of time-based sequences that is necessary for making predictions, recognizing time-based patterns, and generating behaviors. The neocortex can be viewed as a memory system that builds a model of the world for inference, prediction, and behavior (Hawkins et al., 2009). These authors suggested a hierarchically organized memory system in which time-based sequences of multimodal character are primarily stored locally. Hemispheric autonomy with respect to forelimb coordination is likely to facilitate the storing and recalling of multimodal information in forelimb maneuvers, thereby improving motor learning.

Thus, visual feedback from limbs to the limb-steering hemisphere is likely to be highly adaptive. Crossing, or non-crossing, in the OC determines which hemisphere receives visual feedback in forelimb maneuvers, and elaborate mechanisms to regulate IRP proportions exist in the form of multiple axon guidance genes (Herrera and Mason, 2007).

Retinal ganglion cells departure from the retina and propagate in a certain direction by distinguishing guidance cues that are expressed in optic pathway structures, counting the optic disc, optic stalk, OC, and optic tract (Petros et al., 2008). Such guidance cues comprise cell surface ligands e.g., semaphorins in the optic stalk and ephrin B2 in the OC, and a variety of soluble elements for example netrin-1 in the optic disc and Slit1 in areas neighboring the OC (Erskine and Herrera, 2007). In most animals few or no RGC are linked to targets in the ipsilateral side of the brain, while the majority of RGC axons, after crossing the midline at the OC, will be linked to targets in the contralateral hemisphere, located at the ventral-medial hypothalamic (vHT) area (Kim et al., 2014). Subsets of vHT cells express molecules that determine the direction, which RGC axons take at the OC (Kim et al., 2014). For example it has been shown that vHT radial glial cells express ephrinB2, which binds to ephrinB1 receptors expressed in ventral-temporal RGC axons and drive back axons toward the ipsilateral optic tract (Nakagawa et al., 2000; Williams et al., 2003). Other factors e.g., a neuronal cell adhesion molecule expressed in the vHT have been suggested to support the growth of RGC axons across the vHT midline (Williams et al., 2006; Erskine et al., 2011; Kuwajima et al., 2012). Thus, molecular cues guide retinal axons to approximate regions along the rostrocaudal axis of the target. Activity-dependent mechanisms exist that refine this crude map (Hiramoto and Cline, 2014). Hiramoto and Cline showed that retinal mapping in Xenopus laevis might be influenced by the animal's motion direction. The axonal guidance system that determines the OC architecture is an expanding area of research, however many vital mechanisms remains to be discovered (Kim et al., 2014). How axon guidance genes influence, alternatively might be influenced by evolutionary change of species' motor behavior and ecology have been little explored.

The EF hypothesis includes that the selective value of fast and accurate visual steering of limbs, particularly the forelimbs, may result in evolutionary change that turn axonal guidance genes on or off. Thus, the idea presented here is that the OC is an evolutionarily active structure, key to preserving short neural pathways and hemispheric autonomy when eyes or limbs undergo crucial modification to adapt to new ecological niches.

The placement of the eyes, retinal specializations, and the OC architecture seem to be associated with motor behavior in a variety of vertebrates. The necessity of supervising and efficiently processing movements may explain the low proportion of IRP in most fishes, reptiles, and birds. The EF hypothesis accommodates why IRP emanate exclusively from the temporal retina in animals with limbs. Abundant IRPs in primates and most predatory animals is also in accordance with the hypothesis.

There is lack of data on to what extent, and how skillfully, canidae use visually guided paw movements in prey capture, and the relatively high proportion of IRP (20–22%) (Hogan et al., 1999; Lee et al., 1999) in dogs may be at odds with the EF hypothesis. However, wolves are recognized to use their feet and claws for catching/manipulating prey and digging (Moskowitz, 2013).

Visual control of the feet and bill seems to have influenced the evolution of retinal specialization in cockatoos (Coimbra et al., 2014) and possibly in the kakapo and the New Caledonian crow, reinforcing the concept of the primary role of binocular vision in birds as visual control of body appendages (Martin, 2014). The EF hypothesis would predict binocular vision and IRP in species with forelimbs that habitually operate in front of the animal. A predatory lifestyle in terrestrial mammals such as cats or bears typically involves the forelimbs, which are usually frontally positioned and used to hold and manipulate prey, while aquatic predators such as dolphins primarily use the mouth. Thus, the EF hypothesis may explain both why terrestrial mammalian predators possess high proportions of IRP and predatory aquatic animals such as Trionyx and bottlenose dolphins have no IRP. Trionyx and cetaceans are likely to only take advantage of visual feedback to the limb-steering hemisphere, and the EF hypothesis suggests that evolutionary change in position and function of the forelimb resulted in a gradual and complete loss of IRP. That implies that the multiple axon guidance genes, which regulate IRP proportions (Herrera and Mason, 2007) are largely influenced by natural selection. The relatively high proportion of IRP in vertebrates with secondary loss of the limbs demonstrates that such evolutionary change may go in either direction, depending on motor behavior/anatomy in the ecological niche of species.

The primate visual system is highly suited to supervise tasks within the hand's working space (Hadjidimitrakis et al., 2011; Larsson, 2013); indications on that are the switch of ocular dominance during reaching maneuvers (Khan and Crawford, 2001), superior spatial discrimination (Richter et al., 2007) and shorter reaction time in the inferior visual field (Danckert and Goodale, 2001) where the hand commonly operates. Moreover, the fundamentals of stereopsis, vergent eye movements, largely take place within arm's length (Viguier et al., 2001). 3D-shape selective neurons that are co-localized with neurons engaged in grasping (Theys et al., 2015) also supports the EF hypothesis. The EF hypothesis does not contradict the premise that abundant IRP in the OC combined with frontal vision were important in the evolution of stereopsis. However, this review suggests that stereopsis was a byproduct of the evolution of accurate eye limb coordination. The predator hypothesis (Ward et al., 1995), as well as the nocturnal restriction hypothesis (Vega-Zuniga et al., 2013), applies mainly to mammals, whereas the EF hypothesis appears applicable to vertebrates in general. The EF hypothesis might be evaluated through comparative analyses of mammalian and non-mammalian associations among IRP, eye convergence, and visual guidance of the limbs. Such studies are needed to understand the evolution of OC anatomy, retinal specialization, and their association with motor behavior and ecology in vertebrates.