Behind mouse eyes: The function and control of eye movements in mice

The mouse visual system has become the most popular model to study the cellular and circuit mechanisms of sensory processing. However, the importance of eye movements only started to be appreciated recently. Eye movements provide a basis for predictive sensing and deliver insights into various brain functions and dysfunctions. A plethora of knowledge on the central control of eye movements and their role in perception and behaviour arose from work on primates. However, an overview of various eye movements in mice and a comparison to primates is missing. Here, we review the eye movement types described to date in mice and compare them to those observed in primates. We discuss the central neuronal mechanisms for their generation and control. Furthermore, we review the mounting literature on eye movements in mice during head-fixed and freely moving behaviours. Finally, we highlight gaps in our understanding and suggest future directions for research.


A renewed interest in murine eye movements
Orienting the eyes in space is vital for sighted animals during navigation, for evading danger, or finding food (Saleem and Busse, 2023).Historically, eye movements in mice have predominantly been considered a way to stabilise or shift gaze following head movements, as observed in other afoveate visual animals (Kautzky and Busse, 2020;Land, 2019).Technical limitations to recording eye movements, and the prevailing assumption that mice preferentially rely on their whiskers, explain the limited knowledge of the dynamics and higher-order control of eye movements (Kautzky and Busse, 2020).In contrast, a significant body of research focuses on the murine visual cortex (Niell, 2011).Recently, however, camera and recording technology advances started to bridge this gap.
Here, we review recent developments in mouse eye movement research.First, we describe the types of eye movements defined in primates and relate their properties to those observed in mice.In the second section, we give a brief compendium of brain regions implicated in eye movements in mice and primates.In the third section, we focus on studies dissecting the circuits underlying oculomotor control in headfixed mice.Finally, the fourth section highlights novel findings from freely-moving rodents in naturalistic settings.

Types of eye movements and their categorisations
Five main types of eye movements are widely recognised in mammals: vestibulo-ocular reflexes (VOR), optokinetic reflexes (OKR), saccades, smooth pursuit, and vergence eye movements.A sixth type, fixational eye movements, are only occasionally included.Eye movement types can be categorised by their kinematic properties, their effect on the line of sight (gaze-stabilising versus gaze-shifting), or based on their volitional control (reflexive versus voluntary).A subset of the eye movements described in primates has been reported in mice, with differences emerging regarding their concomitance with head movements.Furthermore, mice are lateral-eyed, with larger monocular visual fields compared to primates (~180 • ) but a modest binocular overlap of approximately 40 • -50 • (Holmgren et al., 2021;Scholl et al., 2013;Sterratt et al., 2013).The binocular field of rodents extends above their head (Wallace et al., 2013), but the largest portion of it lies in front of the animal (Meister and Cox, 2013), similarly to primates, due to the pitch of the head.The average gaze position in mice during ambulation is 64 • in azimuth, referenced to the midsagittal plane, and 22 • in elevation, referenced to the Earth's horizontal plane (Oommen and Stahl, 2008).The head adopts a pitch of about − 29 • of the lambda-bregma plane, referenced to the Earth's horizontal (Oommen and Stahl, 2008).In the following section, we briefly describe the most common categorisations of eye movements in primates and relate the findings to those observed in rodents.

Gaze-stabilising and gaze-shifting eye movements
The separation between gaze-stabilising and gaze-shifting eye movements constitutes a relevant functional classification (Fig. 1).Gazestabilising eye movements compensate for head and visual world motion to stabilise the retinal image and enable clear vision (Land, 2019(Land, , 2015)).They typically manifest as a slow, centrifugal or centripetal pursuit-like phase (slow phase of nystagmus).When their orbit reaches maximal eccentricity, a fast, centripetal component (fast phase of the nystagmus, or resetting saccade) recentres the eyes.While saccades are, by definition, gaze-shifting, resetting saccades can form part of a gaze-stabilising sequence.
Gaze-shifting eye movements serve to align the eyes on a target.Foveate animals centre the foveas on the target in both the frontoparallel plane and depth.Gaze-shifting eye movements also occur in afoveate animals (Land, 2019), such as mice (Meyer et al., 2020;Parker et al., 2023).

Vestibulo-ocular reflex
The vestibulo-ocular reflex (VOR), a gaze-stabilising eye movement type, which generates conjugate eye movements in directions opposite to head movements along the rotational and translational axes.The reflex is present in both primates and mice (Fig. 2a; Straube and Büttner, 2007;van Alphen et al., 2001).The peripheral sensors driving the VOR are the vestibular organs located in the inner ear.The semicircular canals sense angular acceleration and generate the angular VOR (aVOR, often referred to simply as VOR).The otolith organs sense the gravitational force and linear acceleration and generate the translational VOR (tVOR, Harrod and Baker, 2003;Straube and Büttner, 2007).
The VOR is a plastic and adaptive reflex.In both primates and mice, the gain of VOR, defined as the multiplicative factor between the driving head velocity and the compensatory eye velocity, is under constant cerebellar control, similar to all other eye movements (Beh et al., 2017), and can also be modulated by the visual system (Carcaud et al., 2017;Faulstich et al., 2004;Schweigart et al., 1997;Stahl, 2004a); see section "Subcortical and cortical mechanisms underlying the adaptability of gaze-stabilising eye movements").The VOR is also engaged in the dark (Stahl, 2004a;Straube and Büttner, 2007), but when visual stimuli are present, the VOR and the optokinetic reflex (OKR, see next section) operate in synergy (angular visuo-vestibulo-ocular reflex, aVVOR; (Stahl, 2004a;Straube and Büttner, 2007).Furthermore, during self-movement, voluntary gaze shifts or fixating on a moving target could be compromised by the VOR.To prevent this, the efficacy of the direct VOR pathway is decreased via brainstem mechanisms in primates (VOR suppression;Cullen, 2009;Roy and Cullen, 1998).VOR suppression is likely missing in mice and other mammals which track moving objects with their head instead of their eyes (Land, 2019).In both mice and primates, the aVOR is best adapted for mid-high frequency, transient head movements (Migliaccio et al., 2011;Schweigart et al., 1997;Schweigart and Mergner, 1995;Stahl, 2004a;Straube and Büttner, 2007;van Alphen et al., 2001), although mice show lower gain for high frequency, low amplitude movements than primates (Migliaccio et al., 2011).Murine 3D aVOR is isotropic (i.e.gains are similar for the different axes of head rotations in 3D), in contrast to the primate aVOR, which shows strong gain reduction for roll movements (Migliaccio et al., 2011).The gain of the aVOR can be up-or down trained in both primates and mice (Kimpo et al., 2005).Finally, there are contradictory reports about the presence and direction of asymmetry in nasal and temporal eye movement velocity and gain during both VOR and VVOR in mice (Migliaccio et al., 2011;Voges et al., 2017).Similarly to yaw movement, there is a supero-inferior eye movement asymmetry in favour of the eye moving inferiorly during head rolls (Migliaccio et al., 2011).Studies in primates suggest that aVOR, but not aVVOR shows directional asymmetries (Khojasteh and Galiana, 2009).

Optokinetic reflex
The optokinetic reflex (OKR) is a gaze-stabilising eye movement type, displayed by both primates and rodents in response to movements of large portions of the visual field on the retina (Fig. 2b; Lappe and Hoffmann, 2000).In primates and mice, OKR gain (amplitude of eye movement / amplitude of stimulus) is typically highest for slow eye movements, complementing the operational range of the VOR (Faulstich et al., 2004;Schweigart et al., 1997;Schweigart and Mergner, 1995;van Alphen et al., 2001).The OKR can be elicited by both binocular and monocular stimuli, in the fronto-parallel plane and in the vertical plane.The fronto-parallel OKR in rodents shows high gain and minimal phase-lag when elicited by slow-moving stimuli (<10 • /s), with both features decreasing at higher stimulus velocities (Kodama and du Lac, 2016;Stahl, 2004a;van Alphen et al., 2001;Van Alphen et al., 2010).Rodent and primate OKR differ in two key aspects.First, the primate OKR is composed of an "open-loop" eye acceleration phase, termed ocular following response (OFR), during which eye movement latency is short and eye acceleration is steep.During the initial phase of the rodent OKR, eye acceleration is limited in comparison to primates (Harvey et al., 1997;Stahl, 2004a;Tabata et al., 2010;van Alphen et al., 2001).The OFR is thought to be mediated by the smooth pursuit system (Takemura et al., 2007), considered absent in afoveate animals.Second, E. Ambrad Giovannetti and E. Rancz eye velocity during the second phase of the OKR, the "closed-loop" phase, is steady or slowly decreases in rodents when OKR stimuli are presented over prolonged intervals, a phenomenon not observed in monkeys (Miles et al., 1986).This is thought to be related to the rodent's underdeveloped velocity storage system (Kodama and du Lac, 2016;Stahl, 2004a;van Alphen et al., 2001).The presence of the optokinetic afternystagmus (OKAN), an eye movement depending on the velocity-storage system in primates (Gygli et al., 2021), is not consistent throughout the mouse literature (weak in van Alphen et al., 2001, but present in França de Barros et al., 2020).Finally, rebound gaze drifts were shown to be differentially adaptable in the T-N and N-T direction, similarly to eye acceleration and deceleration (Kodama and du Lac, 2016).
In rodents, both binocular and monocular OKR stimulation have been shown to trigger asymmetrical eye movements.In the majority of the cases reported, temporo-nasal (T-N) eye movements display higher acceleration and gain than naso-temporal (N-T) ones (monocular stimulation: De'sperati et al., 1994;De'sperati et al., 1994;Douglas et al., 2005;Harvey et al., 1997;Liu et al., 2023   E. Ambrad Giovannetti and E. Rancz et al., 1997;Kodama and du Lac, 2016;Liu et al., 2023; but see Voges et al., 2017, where binocular OKR stimulation resulted in higher N-T gain than T-N, and Wakita et al., 2017, where binocular OKR resulted in equal N-T and T-N gains).This effect is pronounced for slow visual stimuli between ~10-30 • /s (Kodama and du Lac, 2016), causing the eyes to increasingly converge upon repeated stimulus presentation, a phenomenon that is only partially corrected by the resetting saccades (Harvey et al., 1997).Asymmetric responses to monocular OKR stimulation are present in infant humans and non-human primates (Atkinson, 1979), as well as in humans affected by conditions altering normal vision, like strabismus (Valmaggia et al., 2003).In adult individuals, the kinematic asymmetry disappears upon the visual cortex-dependent development of binocular OKR (Brodsky, 2019).Accordingly, lesions of the visual cortex in primates (Lynch and McLaren, 1983), but not in rats (Harvey et al., 1997), introduced an asymmetry in velocity of T-N versus N-T movements, reminiscent of what is observed in juvenile individuals.Recent reports in mice, however, showed that OKR gain and gain potentiation was impaired upon lesioning or silencing the visual cortex (Liu et al., 2016;Liu et al., 2023, see section "Subcortical and cortical mechanisms underlying the adaptability of gaze-stabilising eye movements").

Saccades
Saccades are gaze-shifting eye movements with multiple definition criteria (Fig. 2c; Hessels et al., 2018).Functionally, saccades are defined as eye movements performed to rapidly reposition the line of sight between fixation periods (Hessels et al., 2018).A purely kinematic definition sees saccades as fast, conjugate, ballistic eye movements that occur in horizontal, vertical, or oblique directions.In mice, saccadic eye movements have been observed during both head-fixed and freely-moving conditions (Meyer et al., 2020;Michaiel et al., 2020;Senzai and Scanziani, 2022;Stahl, 2004a).The kinematics of saccades, termed "saccadic main sequence" (Bahill et al., 1975), are highly stereotyped, showing a regular relationship between amplitude, duration and velocity).In primates, the speed versus amplitude relationship is non-linear and shows saturation, causing the duration of saccades to scale linearly to their amplitude (Gibaldi and Sabatini, 2021).In head-fixed mice, the peak velocity of saccades scales linearly with their amplitude, showing no saturation (Sakatani and Isa, 2007).The duration of saccades, instead, saturates relatively to their amplitude.Nevertheless, in primates, saccade amplitude and peak velocity were reported to deviate from the main sequence depending on the value of the targeted stimulus (see Hayhoe and Ballard, 2005 for a review), indicating that the saccadic motor movement can be under significant top-down control and not purely ballistic.During a saccadic eye movement, a directional asymmetry occurs, with saccades in the T-N direction being consistently faster and larger in amplitude compared to the N-T direction (a phenomenon disputed in humans, see Takahashi et al., 2019 for discussion).In addition, mice showed larger post-eye movement drifts than primates (Itokazu et al., 2018;Sakatani and Isa, 2007).Lastly, saccade types can be further grouped according to their triggering stimulus and degree of volitional control.

Reflexive, externally-driven saccades
Stimulus-oriented saccades are triggered by an external stimulus.They are often centrifugal (i.e.shift the pupil away from the centre of the orbit), and do not require strong volitional control (see McDowell et al., 2008 for a review).Stimulus-oriented saccades were also described in mice (Itokazu et al., 2018;Samonds et al., 2018;Zahler et al., 2021).
Express saccades are stimulus-oriented saccades with extremely short latency, as low as 75 ms in monkeys (Fischer et al., 1984) and 100 ms in humans (McDowell et al., 2008).Stimulus-oriented saccades with 100 ms latencies have been recently reported in mice (Zahler et al., 2021).
Resetting saccades are centripetal saccades (i.e. they recentre the eye after a centrifugal movement) occurring during gaze-stabilising eye movements (OKR and VOR).They represent the quick phase of the jerk nystagmus (Stahl, 2004a), ipsiversive to the head movement during VOR and contraversive to the visual stimulus during OKR.Finally, resetting saccades in mice and humans are slower than stimulus-oriented saccades, although the directional bias in velocity persists (Garbutt et al., 2001;Sakatani and Isa, 2007).Recent reports in mice (Meyer, O'Keefe, and Poort, 2020;Michaiel, Abe, and Niell, 2020) referred to resetting saccades as "compensatory saccades", although this definition may be misleading due to the term used in a human clinical context (see below).
Compensatory saccades occur in humans as compensation for a defective VOR, prevalent in older, healthy adults (Anson et al., 2016).Compensatory saccades have not yet been described in mice.

Voluntary, internally-driven saccades
Internally-driven eye movements are of great experimental value to study top-down processes like attention and memory (Zhao et al., 2012).Even though higher cognitive processes like selective visual attention have been identified in mice (Wang and Krauzlis, 2018), the following types of saccades have not yet been reported.
Search saccades: visually guided (stimulus oriented) saccades made towards a visual target of interest.Visual search is a complex phenomenon driven by external stimuli, but not reflexive, as the saccades are under strong cognitive influence (see Hayhoe and Ballard, 2005 for a review).
Look-ahead fixations, despite their name, are fast, voluntary orienting eye movements that humans do towards objects that will be manipulated in the subsequent steps of a task or naturalistic behaviour (Pelz and Canosa, 2001).They are yet to be thoroughly described in primates, and only one report has described similar behaviour in mice (Bergmann et al., 2022).
Predictive saccades occur when the gaze is fixated on the future location of a target that is moving predictably (McDowell et al., 2008).
Antisaccades are intentional gaze shifts away from a visual stimulus (Rommelse et al., 2008).Misdirections of antisaccades are linked to dysfunctions of the frontal cortices (Rommelse et al., 2008).
Memory-guided saccades occur without a visual stimulus and are guided by the memory of a previously presented stimulus (Pierrot-Deseilligny et al., 1991).Deficits of memory-guided saccades are related to dysfunctions of the frontal cortices (Pierrot-Deseilligny et al., 1991).

Vergence
From a mechanical standpoint, vergence eye movements are defined as movements of the eyes in different directions: eyes can either converge (one or two eyes move towards each other) or diverge (one or two eyes move away from each other, Fig. 2d).In primates, though, vergence is often defined in relation to binocular vision, namely as a type of asymmetrical eye movement that serves to maintain binocular fusion in depth (Schor et al., 2002).In rodents, eye movements are largely non-conjugate, and vergence eye movements are observed as a consequence of the ocular countertilt (described in the "Eye movements in freely moving mice" section), thus serving the tVOR (Holmgren et al., 2021;Meyer et al., 2020;Michaiel et al., 2020;Wallace et al., 2013).Mice show vergence eye movements during depth OKR as well (vergence OKR), albeit with a lower gain compared to primates.Vergence OKR in mice is primarily driven by interocular velocity differences (IOVDs) and not by changes in binocular disparity cues in the visual cortex (Choi and Priebe, 2020), suggesting a purely subcortical control of vergence eye movements.A similar OKR-driven vergence was also reported in humans (Wibble and Pansell, 2020).

Smooth pursuit
Smooth pursuit is defined as the ability to voluntarily follow a E. Ambrad Giovannetti and E. Rancz moving target while holding the gaze on it (Straube and Büttner, 2007).Smooth pursuit movements are interrupted by "catch-up" saccades.Similarly to the OKR, the smooth pursuit is subdivided into an open-loop and a closed-loop phase.During open-loop, eye acceleration is exclusively dictated by the object's velocity on the retina.During closed-loop, eye motion is also taken into account, and the pursuit system tries to nullify the retinal slip of the target.In primates, smooth pursuit is best performed for horizontally moving stimuli.Smooth pursuit movements, defined as smooth, voluntary gaze-shifting eye movements, have not yet been observed in rodents.

Fixational eye movements
Fixational eye movements are complex phenomena occurring in absence of overt orienting head movements, however no studies have focused on fixational eye movements in mice thus far.Fixational eye movements comprise fast microsaccades, slow ocular drifts and higherfrequency tremors (Collewijn and Kowler, 2008;Rucci and Victor, 2015).Microsaccades are microscopic (<0.5 • ) ballistic eye movements that can only be operationally distinguished from other saccade types (Hauperich et al., 2020).Microsaccades occur frequently (1-2 Hz in humans) and can shift the visual image over several retinal receptors, on the same order of magnitude as the smallest cortical receptive fields (Rucci and Victor, 2015).Microsaccades have been shown to be under volitional control in primates (Peel et al., 2016;Willeke et al., 2019).Ocular drifts are slow eye movements (~2 • /s in humans) that compensate for microscopic head movements in freely-moving conditions in humans (Poletti et al., 2015), but recent reports have shown that they are also under cognitive control (Lin et al., 2023).Tremors are high-frequency (~90 Hz), involuntary eye movements that occur at the same time as ocular drifts, during inter-saccadic intervals.The physiological function of ocular tremors is currently unclear (Bowers et al., 2019).
At a broader scale, fixational eye movements can be gaze-stabilising, as they work to maintain the image on the retina after a saccade has occurred.From an oculomotor perspective, however, the gaze is microscopically shifted during fixation.Recent evidence suggests that both microsaccades and drifts can be modified by peripheral or foveal visual stimuli (Malevich et al., 2020) and ocular drift responses in rhesus monkeys were systematically tuned to the stimulus visual features (Khademi et al., 2024).Image stabilisation during ocular drifts and tremors is not perfect, and this residual motion on the retina was proven necessary for normal vision (Rucci and Victor, 2015).Both residual motion and microsaccades have been proposed to improve visual acuity by preventing stimulus adaptation, and to create synchronous population responses in retinal ganglion cells (Ahissar et al., 2016;Rucci and Victor, 2015).Ocular drifts in mice have been observed following OKR (rebounding gaze drift; Itokazu et al., 2018;Kodama and du Lac, 2016) and after saccade termination in head-fixed conditions (Itokazu et al., 2018;Kodama and du Lac, 2016).These drifts are possibly linked to absent vestibular input following an attempted head movement, and are likely not functionally equivalent to primate ocular drifts.Stimulus-statistic driven saccades, functionally similar to primate's microsaccades, have been reported in head-fixed mice (Samonds et al., 2018), although these too may be linked to attempted head movements.

Brain circuits underlying the generation and control of eye movements
Mechanistically, the movement of the eyes within the orbit is generated by the contraction of the extraocular muscles, innervated by the main oculomotor nuclei (oculomotor, trochlear and abducens nuclei) located in the midbrain and brainstem (see Sparks, 2002 andStraube andBüttner, 2007 for an extensive review).In primates and mice, brainstem oculomotor nuclei are under the control of tectal projections (tectoreticular pathway), in turn, orchestrated by cortical projections either directly (corticotectal) or indirectly (cortico-striatal-nigro-tectal).Historically, it was thought that each eye movement type was governed by discrete neural circuits operating in parallel, a view no longer prevailing (Büttner-Ennever and Horn, 1997;Stahl, 2004a).For a summary of subcortical and cortical areas known to be involved in oculomotor control in mice, primates, or both, refer to Table 1 and Fig. 3.

Subcortical areas
In both primates and mice, the superior colliculus (SC), the homologue of the optic tectum found in non-mammalian vertebrates (Isa et al., 2021), occupies a central role in the generation of diverse eye movements, among other orienting movements (see Cooper and McPeek, 2021 for a review).A retinotectal and a parallel corticotectal pathway, originating from the visual cortex, converge in the superficial layers of the SC (Gandhi and Katnani, 2011;Hafed et al., 2023;Isa et al., 2021).Upon multimodal stimulus integration, a motor command is generated from the intermediate and deep layers of the SC (SCm) and then relayed onto brainstem oculo-premotor nuclei, among others (Sparks, 2002).Recently, a direct tectoreticular pathway to the premotor nuclei instructing neck muscle movement (i.e.orienting head movements) has been found in mice (Zahler et al., 2023).Head-eye movement amplitude and directionality are topographically organised, showing increased amplitude along the anteroposterior axis of the SCm in both primates and mice.In primates, stimulation of the rostral portion of the SCm evokes eye-only movements and small eye-only gaze shifts, implicating this area in fixational eye movements, such as microsaccades (Basso et al., 2000).In both primates and mice, the medial and caudal portion of the SCm govern larger head-eye and orienting/avoiding body movements (Constantin et al., 2004;Freedman et al., 1996;Masullo et al., 2019;Sahibzada et al., 1986;Zahler et al., 2023).In addition to saccade control, bilateral lesions of the SC have been shown to alter vergence eye movements in humans (Ohtsuka et al., 2002).
In primates, an indirect corticostriatal pathway provides further control of saccade generation.Here, the caudate nucleus (CN), the substantia nigra pars reticulata (SNr), and the striatum (Str) form an indirect saccade generator pathway that operates via cortical disinhibition of SCm neurons (Hikosaka et al., 2000;Shires et al., 2010).While inhibitory nigrotectal projections have been functionally characterised in mice (Kaneda et al., 2008), their contribution to saccade generation remains to be demonstrated.
The cerebellum and the precerebellar nuclei are critically implicated in the modulation of saccadic and other eye movements.The oculomotor vermis (OMV) and the oculomotor region of the fastigial nucleus (FN) of the cerebellum are part of the cortico-ponto-cerebellar circuit governing conjugate horizontal eye movements, best described in primates (see Kheradmand and Zee, 2011 for a review).The OMV and FN exert a modulatory function on the accuracy of saccade amplitude and direction, and on eye acceleration during smooth pursuit.Neurons within the vestibulocerebellum, particularly the flocculus/paraflocculus and the nodulus/uvula, are involved in the control and initiation of smooth pursuit, vergence eye movements, gaze holding and VOR reflexes (Kheradmand and Zee, 2011).Lastly, the vestibulocerebellum is strongly involved in the adaptive plasticity of gaze-stabilising reflexes (Carcaud et al., 2017;Kodama and du Lac, 2016), as well as in the coordination of subtypes of eye movements (Kheradmand and Zee, 2011).
In both primates and mice, the control of gaze-stabilising eye movements is carried out by three main systems: the accessory optic system (AOS), the vestibular system, and the cerebellum.The AOS is composed of three main nuclei, located in the midbrain: the dorsal terminal nucleus (DT), the lateral terminal nucleus (LT), and the medial terminal nucleus (MT).The DT is operating in conjunction with the nucleus of the optic tract (NOT) for horizontal gaze stabilisation while the LT and MT promote vertical gaze stabilisation (reviewed in Giolli E. Ambrad Giovannetti and E. Rancz et al., 2006;Dhande et al., 2013).The nuclei of the AOS, and the vestibulo-premotor nuclei trigger eye movements in response to retinal slips or vestibular signals generated by head motion, respectively.The main afferents to the AOS are contralaterally-projecting direction-sensitive retinal ganglion cells (ON-DSRGCs; see Dhande et al., 2013;Dhande and Huberman, 2014 for a review), which optimally respond to slow-moving stimuli such as peripheral optic flow.The primary efferents of the AOS are the ipsilateral vestibular nuclei, inferior olive (IO), and pontine nuclei, sparing the oculomotor nuclei and cerebellar cortex (Giolli et al., 2006;Masseck and Hoffmann, 2009).Notably, lesions of the NOT not only compromise horizontal gaze stabilisation, but also result in impairments of smooth pursuit in primates (Yakushin et al., 2000).Finally, spinal cord afferents to the oculomotor nuclei have recently been shown to mediate locomotion-coupled gaze-stabilising eye movements in decerebrated mice, similar to other vertebrates (França De Barros et al., 2022).

Cortical areas
In primates, a network of cortical areas within the frontal, parietal, and temporal lobes, supports the generation of eye movements.Cortical control of brainstem oculomotor nuclei is achieved via a direct corticotectal, an indirect corticotectal pathway through the striatum, a corticoreticular, and a corticocerebellar pathway (Krauzlis, 2013).Diverse areas within the prefrontal and parietal cortices are key in initiating rapid eye movements (frontal eye fields, FEF), planning and learning motor tasks (supplementary eye field, SEF; posterior parietal cortex, PPC), modulating the gain of rapid and smooth pursuit eye movements (anterior cingulate area, ACA; FEF), OKR and VOR (PPC; (Ventre, 1985a(Ventre, , 1985b;;Ventre and Faugier-Grimaud, 1986), or inhibiting saccades (dorsolateral prefrontal cortex, dlPFC).These functions have been extensively reviewed (see Pierrot-Deseilligny et al., 2004;Pouget, 2015;Rivaud et al., 1994;Lynch, 1987;MacAvoy et al., 1991).Area 8, of which the FEF is part of, contributes to vergence and fixational eye movements (Gamlin and Yoon, 2000;Gamlin, 2002; see Krauzlis et al., 2017 for a review).Within the temporal lobe, the medial temporal cortex (MTC) and medial superior temporal (MST) areas are engaged during smooth pursuit eye movements (including the OFR Kawano et al., 1994;Takemura et al., 2007).Finally, the binocular portion of the visual cortex was shown to control short-latency vergence eye movements occurring in response to anticorrelated binocular disparity cues (Masson et al., 1997).
The contribution of the cortex to the generation of eye movements in rodents is less extensively known.The supplementary motor cortex (MOs) was shown to be necessary for the generation of contralateral saccadic eye movements in mice (Itokazu et al., 2018).The primary visual cortex (VISp) and the higher visual areas were shown to promote OKR potentiation via cortico-fugal projections to the midbrain (NOT/DT), upon vestibular lesion (Liu et al., 2016) or prolonged exposure to OKR stimuli (Liu et al., 2023).In rats, a potential homologue of the FEF was found in the frontal eye field (FOF) area, governing memory-guided orienting movements (Erlich et al., 2011).Finally, in rodents, vestibular signals are processed by a widespread network of cortical areas, including primary and higher visual areas (Rancz et al., 2015;Vélez-Fort et al., 2018;Bouvier et al., 2020;Guitchounts et al., 2020;Hennestad et al., 2021;Keshavarzi et al., 2022;Parker et al., 2022b).Visuo-vestibular integration already occurs at the level of the  E. Ambrad Giovannetti and E. Rancz vestibular nuclei in the brainstem to drive gaze-stabilising reflexes.However, gain control and other modulatory influences are implemented predominantly at subsequent stages at the level of the thalamus, cortex and cerebellum (see Cullen, 2019;Cullen and Zobeiri, 2021).Parallel retino-tectal (orange arrow), retino-geniculate (green arrow) and projections from the retina to the accessory optic system (turquoise arrow; AOS, comprising the nuclei NOT/DT, MT, LT) support gazestabilising and gaze-shifting eye movements in mice.Visuo-vestibular gaze stabilisation is dependent on the AOS and its direct or indirect projections to the oculo-premotor nuclei via the vestibular premotor nuclei or the inferior olive (IO).Activity of the NOT/DT is also modulated by projections from the primary and higher visual cortex.The floccular lobe of the cerebellum receives innervation from the AOS via the IO.The flocculo-cerebellum, the semicircular canals and spinal cord modulate the plasticity of gaze-stabilising reflexes via projections to the vestibular premotor nuclei, which in turn project to the visual/vestibular thalamic nuclei and to the oculomotor nuclei.The visuo/vestibular thalamic nuclei relay vestibular premotor nuclei activity to the retrosplenial cortex (RSP), higher visual areas (HVAs) and primary visual cortex (VISp).Gaze shifts are mediated by the superior colliculus (SC), either via direct retino-tectal projections, or they undergo indirect cortical modulation via the cortico-tecal pathway.Projections from the motor regions of the SC reach the oculo-premotor nuclei (tecto-reticular pathway), which in turn project to the oculomotor nuclei.Initiation of gaze-stabilising and gaze-shifting eye movements is mediated by the contraction of the eye muscles, innervated by the third, fourth and sixth cranial nerves (CN).Indirect pathways from AOS to IO through ventral tegmental area and cortico-striatal-nigro-tectal projections are not shown.The dashed line indicates projections from VISp to the superficial layers of the SC.

Neuromodulation of eye movements
Dopaminergic, noradrenergic and cholinergic neurotransmission have been strongly implicated in visual attention-guided orienting in primates and mice (Chudasama and Robbins, 2004;Li et al., 2021;Noudoost and Moore, 2011a).While selective visual attention has been demonstrated in mice, its effect on eye movements is unclear (Wang and Krauzlis, 2018).Within the basal ganglia of non-human primates, reward signals in the caudate nucleus (CN) are used to disinhibit collicular neurons that burst before saccade generation (Hikosaka and Wurtz, 1985).Saccades in primates are indeed faster when a reward is expected, a process linked to dopaminergic signalling (Chen et al., 2014).The application of dopamine agonists in the FEF of primates was also linked to increased ipsiversive saccadic target selection (Noudoost and Moore, 2011b).Conversely, depleting dopamine in the CN disrupted the saccadic main sequence and voluntary saccades in primates (Kato et al., 1995).Accordingly, fatigue-related reduction of saccade velocity has been linked to dopamine and noradrenaline depletion (Connell et al., 2017).While monoamines modulated saccade velocity, inhibition of monoamine reuptake did not affect smooth pursuit movements (Connell et al., 2017).Lastly, acetylcholine was shown to play a role in the fixation of the eyes after a saccade in cats (Navarro-López et al., 2004), presumably by modulating the presynaptic release from PPRF neurons and local excitability of tonic firing neurons in the NPH nucleus as shown in rats (Navarro-López et al., 2004).

Eye movements in head-fixed mice
This section highlights recent reports on eye movements recorded from head-fixed rodents under stimulus-controlled conditions.We focus on recent findings about the cortical and subcortical neural correlates responsible for the plasticity of gaze-stabilising reflexes, as well as the deployment of gaze-shifting eye movements.

Subcortical and cortical mechanisms underlying the adaptability of gaze-stabilising eye movements
The kinematic properties of gaze-stabilising reflexes have been extensively characterised in rodents (Mitchiner et al., 1976;Imai et al., 2016;Kodama and du Lac, 2016;Migliaccio et al., 2011;Stahl, 2004a;Tabata et al., 2010;van Alphen et al., 2001).Pontine, cerebellar and mesencephalic mechanisms supporting the control and plasticity of eye movements received extensive attention in all vertebrates (Beh et al., 2017;Horn and Leigh, 2011;Wibble et al., 2022).On the other hand, only a few studies have addressed whether cortical circuits exert control over gaze-stabilizing reflexes.
Vestibular lesions, the experience of vestibular-visual mismatch, or continuous OKR stimulation can potentiate the gain of the OKR (Faulstich et al., 2004;Katoh et al., 2000;Liu et al., 2016;Liu et al., 2023;Wakita et al., 2017).OKR dynamics is thus not limited by the properties of directional sensitive retinal ganglion cells which drive the OKR via projections to the accessory optic system (AOS; Kodama and du Lac, 2016;Sabbah et al., 2017).Lesioning the ipsilateral flocculus, a part of the cerebellum necessary for the plasticity of gaze stabilising reflexes, was shown to affect the acceleration of OKR eye movements in the T-N direction only, suggesting that independent parallel pathways control T-N and N-T eye movements (Kodama and du Lac, 2016).Similar asymmetries in the directionality of gain potentiation have been observed for the VOR (Voges et al., 2017).VOR gain in the N-T direction correlated with increased floccular Purkinje cell activity, but only for visual stimuli contraversive to the VOR stimulus (animal and visual field moving in opposite directions).However, decreases in activity during gain-decrease learning only partially reflected the changes in VOR gain.This suggests that the locus for VOR gain-decrease, but not gain-increase learning, is not exclusively located within the cerebellar cortex (Voges et al., 2017).Indeed, reduction in VOR gain using a naturalistic visual-vestibular mismatch protocol led to strong excitability changes in the vestibular nuclei and simultaneous synaptic depression at the vestibular afferent synapses (Carcaud et al., 2017).Furthermore, lesioning the nucleus prepositus hypoglossi (NPH), the nucleus immediately upstream of the abducens nuclei and historically thought to function as a single oculomotor integrator for horizontal eye movements, showed differing effects on VOR and OKR in primates, further supporting the idea that gain stabilising eye movements are controlled by parallel systems (Kaneko, 1999).
Cortical contributions to the adaptability of the OKR have been evaluated by lesioning the cortex in various mammals.In primates (Zee et al., 1987) and cats (Tusa et al., 1989), lesions or inactivation of the visual cortex reduced OKR gain and response to higher-frequency stimuli, indicating that the cortex increases the dynamic range of the reflex in these species.In juvenile rats (Prusky et al., 2008) and monocularly deprived adult mice (Prusky et al., 2006), the visual cortex was instead shown to mediate a form of plasticity increasing the spatial frequency range of the stimuli evoking OKR responses.Furthermore, it was recently demonstrated that visual cortical projections to the nucleus of the optic tract and dorsoterminal nucleus (NOT/DT), but not other eye-movement-related structures, mediated the experience-dependent fronto-parallel OKR potentiation in mice (Liu et al., 2016).Following up on this work, it was demonstrated that direction selective pyramidal neurons in posterior HVAs have a stronger influence on OKR gain and plasticity than anterior HVAs, and that this effect is mediated by inferior olive-projecting NOT-DT neurons (Liu et al., 2023).Interestingly, the cortical contribution to OKR gain and gain potentiation occurs independently of retinal slip signals transmitted to the NOT-DT by DS-RGCs (Liu et al., 2023).For vergence OKR, silencing VISp had no significant effect, although this study did not measure reflex gain (Choi and Priebe, 2020).Regarding the cortical control of VOR, adaptation was shown to partially depend on the middle suprasylvian cortex in cats (Tusa et al., 1989;Ventre, 1985a).However, the contribution of the cortex to the gain of the VOR in mice is unknown, despite widespread activation of E. Ambrad Giovannetti and E. Rancz the rodent cortex by vestibular stimulation (Rancz et al., 2015).
In summary, the direction-dependent speed and gain asymmetry of gaze-stabilising eye movements is supported by parallel subcortical circuits governing their adaptability in both mice and primates.In addition, the visual cortex plays an essential role in OKR gain potentiation in mice.

Subcortical and cortical mechanisms underlying the generation of saccades in mice
Head-fixed mice perform horizontal eye movements kinematically similar to primate saccades.Recent studies have probed the subcortical and cortical circuits required to generate horizontal saccades and proposed functional roles for them.In addition, measurements of attempted head movements during stimulus-oriented and spontaneous saccades have revealed that eye movements made by head-fixed mice are likely to be always accompanied by head-movements.
Early studies have demonstrated that head-fixed mice make spontaneous, rapid conjugate eye movements in the horizontal plane (see section "Saccades"; Sakatani andIsa, 2007, 2004).Moreover, visual stimuli (i.e.natural scenes) were sufficient to evoke horizontal saccades whose size was related to the image statistics (Samonds et al., 2018).Saccade size could be predicted by the minimum distance needed to decrease the correlation in neuronal population responses.In support of this hypothesis, mice made larger saccades when visual acuity was further reduced by monocular deprivation (Samonds et al., 2018).These saccades could functionally correspond to primate microsaccades, as saccade sizes were in the same range as the receptive fields of cortical neurons in both animals.Furthermore, mice were also shown to innately perform stimulus-oriented, conjugate horizontal saccades towards tactile (air-puff to the ear or whiskers) and, to some extent, auditory stimuli (Zahler et al., 2021).Surprisingly, visual stimuli did not evoke innate saccades (Zahler et al., 2021), but mice could be trained to perform saccades targeted to an LED light (Itokazu et al., 2018;Sato et al., 2019).Nonetheless, due to the strong correspondence between head movements and eye movements in mice, the interpretation of the findings in (Itokazu et al., 2018;Samonds et al., 2018;Sato et al., 2019) is likely confounded by the lack of measured attempted head-movements.
The strong link between saccadic eye movements and orienting head movements is undisputed in all mammals (Land, 2019(Land, , 2015)).Indeed, a distinguishing factor of spontaneous versus stimulus-oriented saccades

Box 1
Methods for eye tracking in freely-moving mice.
Recordings of eye movements in head-fixed mice have been carried out for decades using magnetic search coils and video-oculography, however the search coil technique has been shown to alter eye movements in mice (Stahl et al., 2000).Videography has its own limitations as well.First, using the centre of the pupil as a proxy to the line of sight relies on the assumption that the two coincide.However, in foveated animals, such as primates, the optical and visual axis of the eye differs due to the off-centre position of the fovea on the retina (Artal, 2014), introducing a known error.Second, isomorphic pupil dilation is assumed, such as the pupil can be fitted well with an ellipse to determine its centre, but shifts in the pupil centre were described under some pathological conditions in humans.Third, an insufficient sampling rate may lead to significant errors in the estimation of saccade parameters, although there are existing algorithms that can recover peak velocities, but not acceleration and deceleration (Wierts et al., 2008).Despite these limitations, videography with high-speed cameras is still the best method in rodents, which are afoveate and no anisomorphic pupil diameter changes have been observed so far.The 3D angular position of the line of sight can then be calculated using either model-based geometric transformations (Sakatani and Isa, 2004), or by using the corneal reflection of an external reference light.The latter method also circumvents the potential difficulty in estimating the pupil's edges when the pupil is dilated beyond the visible surface of the eye (as proposed by Sakatani and Isa, 2007).However, it requires additional hardware and careful calibration, thus is limited to head-fixed recordings (Imai et al., 2016).
The first method developed to record high-resolution eye movements in freely-moving mice used a magnetic approach (Payne and Raymond, 2017).This technique relies on placing a miniaturised magnet onto the surface of the eyes, whose rotation is then measured with a sensor fixed to the skull.Video-oculography, instead, consists of recording pupil movement with head-mounted, high-speed cameras, first employed in rats (Wallace et al., 2013) and subsequently in mice.Meyer, Poort et al. (Meyer et al., 2018) were the first to combine an inertial measurement unit (IMU) with head-mounted CMOS image sensors with a frame rate of around 60 Hz, arguably suboptimal to measure saccade kinetics.The eyes are illuminated by IR LEDs and the cameras record images reflected by infrared "hot" mirrors (reflecting infrared while passing visible light) positioned in front of each eye.Variations of this setup, employing cameras or LEDs instead of IMUs to track head position, have been used by independent labs (Holmgren et al., 2021;Michaiel et al., 2020).The pupil and head position measurements have been combined with neuronal activity recordings using tetrodes (Meyer et al., 2018) or linear silicon probes (Parker et al., 2023(Parker et al., , 2022b)).In parallel to video-oculography, hardware and software solutions were developed to see the world through the eye of the mouse.One strategy has been to direct an additional camera ("world camera") towards the outside world (Parker et al., 2022b) and transform head-centric coordinates into retino-centric coordinates using a shifter network (Parker et al., 2022b;Yates et al., 2023).To approximate cyclotorsion, a TiO2 spot can be applied on the cornea and tracked in each image frame (Holmgren et al., 2021).Alternatively, cyclotorsion can be approximated with models based on variations in the head pitch (Parker et al., 2023(Parker et al., , 2022b)).Without a world camera, a digitised version of the behavioural arena and room has been rendered onto each eye during offline analyses (Holmgren et al., 2021).
Several coordinate frames with their references need to be defined to capture orienting behaviours and their relationship to sensory organs and neuronal processing (Meyer et al., 2020;Stabio et al., 2018;Wallace et al., 2013).The neural computations underlying the generation of these maps and the transformations between them are the very basis of perceiving and interacting with the external world.First, the pitch and roll orientation of the head should be referenced to the Earth's horizontal and the lambda-bregma and interocular lines, respectively.This necessitates careful stereotaxic implantation of IMUs or post hoc calibrations.The yaw orientation of the head can only be referenced internally (e.g. to the line connecting the first two vertebrae) as mice do not have a robust magnetic compass.Relative yaw values likely suffice for all experiments unless explicitly studying vestibulo-collic reflexes.Second, the position of the eye referenced to the head can be determined using camera images.The horizontal eye axis connecting the eye corners can be used to determine the 2D pupil position from camera images.A geometric model of the eye can then be employed to transform this into 3D coordinates (Sakatani andIsa, 2007, 2004).Finally, torsional eye movements should be measured to determine the precise orientation of the retina in the eye socket, which is necessary to map gaze onto retinotopically organised brain areas.The definition of these coordinate frames and the transformations between them is not trivial.A consensual common framework and detailed methodological reporting are needed to make future studies comparable.
E. Ambrad Giovannetti and E. Rancz is the relative timing between attempted head and eye movements.Attempted head movements occurred immediately after saccades evoked by various sensory stimuli (Zahler et al., 2021) or by stimulating the superior colliculus optogenetically or electrically (Zahler et al., 2023(Zahler et al., , 2021)).However, in the case of spontaneous saccades, head-movement attempts always started before the saccade (Meyer et al., 2020;Zahler et al., 2021), resembling the saccade-and-fixate behaviour described in freely-moving animals (see section "Eye movements in freely-moving mice").Recently, a third type of orienting eye movement was described in head-fixed mice (Bergmann et al., 2022), albeit attempted head movements were not measured.When mice navigated in a floating plus-maze while being head-fixed, saccades occurred together with asymmetrical whisking towards the direction of the future turn, regardless of body turns, thus in a "look ahead" fashion.
The subcortical circuits governing the sensorimotor transformations underlying orienting head-eye movements are conserved between mice and primates.In both species, electrical microstimulation along the anteroposterior axis of the SCm evokes both saccades and attempted head movements with progressively larger amplitude, revealing a comparable topology of saccade generator circuits in the rodent and primate SCm (Isa et al., 2021;Wang et al., 2015;Zahler et al., 2023).In mice, electrical stimulation along the anteroposterior axis of the SCm specified saccade endpoints in eye-in-head coordinates, as well as stereotyped head displacement attempts of increasing amplitude (Zahler et al., 2023).Similarly, when saccades were evoked by an air-puff stimulus, direct optogenetic silencing or activation of SCm neurons was sufficient to bias saccades endpoints and the directionality of attempted head movement (Zahler et al., 2021).These results show that eye-and head movements are inseparable in mice.Accordingly, a single population of SCm neurons was shown to innervate distinct brainstem nuclei, which separately control head displacement and saccade endpoints (Zahler et al., 2023).In contrast, neural activity and stimulation of the primate rostral SCm is associated with small eye-only movements, possibly corresponding to gaze shifts towards near-foveal targets and fixational eye movements (Gandhi and Katnani, 2011;Klier et al., 2003).The lack of eye-only movements in mice may be a consequence of the large retinal and cortical receptive fields, making small eye movements functionally futile.
The cortical contribution to saccadic behaviours in mice received increased attention lately.(Itokazu et al., 2018) trained mice to perform stimulus-oriented saccades towards a LED target.These saccades were faster than those typically observed in primates for saccades of the same amplitude, reaching angular speeds of about 300 • /sec.Directional asymmetries in saccadic velocities were also observed, with T-N saccades being faster and larger than N-T saccades, a finding disputed in humans (Takahashi et al., 2019).Electrical stimulation of MOs, but not VISp or higher visual areas (HVAs), elicited saccades of variable amplitude, while acute silencing of the supplementary motor area (MOs) impaired the acquired saccadic behaviour (Itokazu et al., 2018).However, upon prolonged silencing of the ipsilateral MOs, contralateral cortical plasticity changes restored the acquired saccadic eye movement behaviour (Sato et al., 2019).MOs may control saccade generation through projections to the SC or the contralateral striatum (Itokazu et al., 2018;Oh et al., 2014).In addition, sparse but direct projections from the frontal cortex to brainstem reticular orienting circuits have also been reported (Usseglio et al., 2020).Additionally, continuous saccade adaptation, a term used to describe the plasticity keeping the motor and sensory systems in registration, has been shown to involve both the oculomotor vermis, a part of the cerebellum, and cortical networks in humans (Guillaume et al., 2018).
In summary, head-fixed mice show diverse saccadic behaviour occurring spontaneously or in response to various stimuli.The superior colliculus and the frontal cortex can generate mouse saccades like in primates.The topological saccade map in the murine SCm encodes stereotyped saccade endpoints and fixed amplitude head displacement attempts.

Eye movements in freely-moving mice
Measuring eye movements in freely-moving, lightweight animals with lateral eyes, like rodents, is not a trivial task.The challenge has been recently overcome using miniaturised magnetic (Payne and Raymond, 2017) or camera-based eye-tracking devices in rats (Wallace et al., 2013) and mice (Meyer et al., 2018).Combined with recordings of head movements (see Box 1), these studies were the first to describe head-eye coupling dynamics in mice.This is important to understand how the integration of movement-and locomotion-related information with visual signals in the cortex (summarised in Box 2) creates perception and behaviour (reviewed recently by Saleem and Busse, 2023).

VOR-linked eye movements
Compared to head-fixed conditions, freely-moving mice show an increase in eye movements and their variability and divergence, partially due to the vestibular reflexes compensating for head motion (Payne and Raymond, 2017).In rats, eye movements were found to be predominantly non-conjugate during open-field exploration (Wallace et al., 2013).Eyes moved in opposite directions when the head rolled or in response to changes in head pitch (upward and outward for negative pitch and vice versa), a phenomenon termed ocular countertilt.Similarly, in freely-moving mice, a variety of VOR-linked, non-conjugate, gaze-stabilizing eye movements, including torsional eye movements, have been observed (Holmgren et al., 2021;Meyer et al., 2020).These changes in eye position, and thus inferred gaze direction during head tilt, could serve to stabilise the visual field along the horizontal plane so that the celestial and the ground sampling portions of the retina are kept in place (Saleem, 2020).However, not all eye movements of freely-moving mice proved to be gaze-stabilising.(Meyer et al., 2020(Meyer et al., , 2018) ) observed that a fraction of eye movement variability, particularly in the horizontal direction, was unexplained by pitch/roll head movements.These conjugate eye movements were generated during head yaw (Meyer et al., 2020;Payne and Raymond, 2017) and were characterised by a triphasic sequence: an initial gaze-stabling slow movement (eyes and head counter-rotate), followed by a fast-resetting saccade (eyes and head rotate in the same direction), and a final slow compensatory eye movement that recentres the eyes (Fig. 4).In addition, saccades showed a faster speed in the T-N direction, similar to head-fixed animals, leading to changes in binocular field width during the head turn (Meyer et al., 2020;Payne and Raymond, 2017).This triphasic sequence occurs as a function of the angular VOR and is the well-described "saccade and fixate" eye movement, which is highly conserved across all vertebrates from lamprey to man (Land, 2019).In humans, saccade-and-fixate behaviour happens during "aimless looking around" when the gaze is principally carried by the head, so the sum of eye movements is gaze stabilising (Land, 2019).Note that during this so-called saccade and fixate behaviour the gaze is shifted by the head displacement, and the eye movement serves the sole purpose of stabilising the gaze, so "fixate, saccade, fixate" or simply aVOR would be a more precise naming.
There is a limited amount of data regarding the neuronal mechanisms underlying combined head-eye movements in mice, contrary to primates (for reviews see Leigh and Zee, 2015;Pelisson and Guillaume, 2009).In freely-moving mice, optogenetic stimulation of SCm evoked a movement sequence where a saccade is accompanied by a slightly delayed head turn and is followed by a recentring gaze stabilising eye movement linked to VOR (Zahler et al., 2023).This is similar to the movement sequence evoked by electrical stimulation of the SC in head-free primates (Constantin et al., 2004;Freedman et al., 1996), suggesting conserved underlying neuronal circuits.In contrast to the saccade and fixate behaviour, the SCm-evoked head-eye movement lacks the initial head-and consequent gaze stabilising eye-movement (phase two in Fig. 4).This suggests that direct stimulation of the SCm triggers head-eye movements akin to stimulus-evoked saccades in head-fixed mice (Zahler et al., 2021), while the saccade and fixate sequence is probably initiated outside of the SC.
While both the murine and primate SCm govern orienting head-eye movements, crucial functional differences exist between the two species.In primates, population activity in the SCm encodes a topographic gaze shift map (Sparks, 1999;Klier, 2003).Accordingly, head and eye movements evoked at the same SCm location vary inversely depending on the initial eye position (Freedman et al., 1996) to ensure invariant gaze displacement.In mice, however, Zahler et al. (2023) showed that only eye movements (i.e.saccade endpoints), but not head movements vary with initial eye position upon SCm stimulation (compare Fig. 11 in Freedman et al., 1996with Fig. 1 in Zahler et al., 2023).Furthermore, the topographic map along the anterior-posterior axis showed increasing head displacement attempts and more eccentric saccade end-points, meaning that saccade end-points pre-compensated for the magnitude of the VOR resulting from the larger head movements.Therefore, in mice but not in primates, the final gaze position upon SCm stimulation will depend on the initial eye position and is independent of the site of stimulation, which instead determines head displacement.Functionally, this arrangement in mice ensures the pupil is recentred after the head has completed its turn (Zahler et al., 2023), which is not the case in primates (Freedman, 2008).
In summary, head-eye coupling dynamics in mice consist of both non-conjugate and conjugate eye movements, which occur to compensate for changes in head tilt and yaw, respectively.Regarding spontaneous gaze shifts, saccades follow the (attempted) head movement in both head-fixed (Meyer et al., 2020;Zahler et al., 2021) and head-free conditions (Meyer et al., 2020).However, when saccades are evoked by sensory or direct SCm stimulation, saccades always precede head movements in both head-free (Zahler et al., 2023) and head-fixed animals (Zahler et al., 2023(Zahler et al., , 2021)).While in primates SCm topographically encodes gaze displacement, in mice, it drives head displacement and the corresponding eye movement to recentre the pupil.

Prey capture
Mice rely on their vision to accomplish several behaviours (Saleem and Busse, 2023).For example, mice use visual cues to catch crickets, a predatory behaviour they perform innately (Hoy et al., 2016).Different groups have implemented variations of the cricket hunting task to study the head-eye coupling dynamics and the underlying visual processing mechanisms performed by mice during visually guided behaviours.
Head-eye coupling dynamics observed during free exploration were shown to be preserved during visually guided tasks.The "saccade and fixate" pattern is present, as eyes do not change their position systematically when tracking a visual stimulus but instead move to compensate for the orienting head movements (Meyer et al., 2020).Head pitch and torsional gaze-stabilising eye movements did not significantly differ between free exploration and during prey capture behaviour (Holmgren et al., 2021).However, while these head-eye coupling dynamics were unaltered, head-eye movements became highly structured during prey capture tasks.Michaiel et al. (2020) have shown that (i) mice stabilise the eyes at a neutral vergence during approach by maintaining the head in a neutral pitch position; (ii) the prey is kept within the binocular visual field; (iii) mice employ a saccade and fixate strategy consisting of head turns and accompanying eye movements that progressively decrease in amplitude as a function of their distance to the target.Separate experiments by Holmgren et al. (2021) showed that mice perform orienting head-eye movements to maintain the prey in the part of the visual field with the least optic flow, termed focus of expansion (FOE; Gibson, 1950).It is within the FOE where the binocular overlap is maintained with the highest probability (Holmgren et al., 2021), strengthening the hypothesis that mice use binocular cues to track prey.Consistently, it was shown that monocular enucleation affected all phases of the prey hunt and that mice maintain a strongly negative head pitch when approaching crickets to keep the prey in sight above their nose (Johnson et al., 2021).The discrepancy in head pitch observed by Johnson et al. (2021) and Holmgren et al. (2021); Michaiel et al., (2020); Oommen and Stahl (2008) is unlikely to reflect a biological difference, as the head pitch was measured relative to different references.Finally, the last phase of the prey approach was shown to consist of a sharp elevation of head pitch, which stereotypically preceded a bite-and-grab event sequence (Johnson et al., 2021).How the eyes move during this sharp head movement and whether any additional mechanisms to prevent the ocular countertilt reflex are in place remain unknown.

Gap-jumping and pole-descent-cliff tasks
Depth perception in mammals can be achieved using both monocular and binocular cues.Recent studies summarised below have shown that mice normally rely on binocular cues to estimate distances during visually guided tasks, but can resort to monocular cues.
Two independent groups have recently employed naturalistic, nonhead-restrained tasks to study the link between head-eye coupling, depth perception, and distance estimation in mice (Boone et al., 2021;Parker et al., 2022a), employing variants of the previously developed visual cliff test (Fox, 1965).In the gap-jumping task (Parker et al., 2022a), mice were shown to switch to an equally successful monocular strategy if one eye was occluded.This strategy involved a longer decision time associated with a lower head pitch and increased vertical head movements, suggesting temporal integration of motion parallax cues to estimate depth.Monocular mice only used the monocular part of VISp to successfully execute the task, as binocular VISp suppression had no effect, while monocular VISp suppression decreased the success rate.When binocular VISp was acutely and intermittently suppressed in binocular mice, some of the strategy changes were evident, but the success rate decreased.Furthermore, mice did not make systematic changes in eye vergence angles that could not be explained by head movements, demonstrating that mice do not move their eyes to increase binocular overlap preceding the jump.These data show that while mice normally use binocular cues to estimate distance, they are able to switch to a monocular strategy when one eye is occluded.Boone et al. (2021) combined the visual cliff task, where animals prefer visually closer surfaces to more deep ones, with a pole descent in order to force them to sample mostly the binocular part of the visual field.Contrary to Parker et al. (2022a), in this task, mice with one eye sutured did not take longer to decide and failed to distinguish between different depths.Previous studies had identified disparity-tuned neurons with diverse tuning properties, in both VISp and HVAs (La Chioma et al., 2019;Samonds et al., 2019;Scholl et al., 2013).The neuronal disparity tuning in binocular VISp (Samonds et al., 2019) could explain the accuracy of depth estimation in this task, providing indirect evidence for using absolute disparity selectivity in VISp to estimate depth.However, depth perception can be supported by non-stereoscopic binocular cues (Chopin et al., 2019).Causal manipulations must be performed to establish the function of disparity-tuned neurons during tasks requiring distance estimation.Lastly, why chronically impaired binocular vision significantly altered the success of mice in the pole descent cliff task but not in the gap-jumping task needs further clarification.

Retinal and cortical specialisations in support of visually guided behaviours
The relevance of reorienting the binocular field during behaviour has been corroborated by newly described retinal and cortical specialisations.
During the approach phase of the cricket hunting task, the prey is sampled by the temporal (i.e.lateral) portion of the retina (Holmgren et al., 2021).Ipsilaterally projecting retinal ganglion cells (ipsi-RGCs) in the ventro-temporal portion of the retina sample the environment with enhanced spatial resolution (Bleckert et al., 2014;Huberman et al., 2008) and are thought to support stereopsis in mammals (Wilks et al., 2013).Furthermore, a subset of ipsi-RGCs was shown to be strongly excited by prey-mimetic stimuli (Johnson et al., 2021).Indeed, the specific ablation of ipsi-RGCs, a mere 2 % of retinal ganglion cells, resulted in markedly poorer prey capture behaviour (Johnson et al., 2021).These findings show that specialised RGCs in the temporal retina support prey capture and suggest mice use stereoscopic cues during hunting.
Downstream of the retinal specialisations, van Beest et al. ( 2021) discovered an intriguing region in the visual cortex of mice, spanning portions of the primary and higher visual areas, which shows enhanced spatial representation.This area, termed the focea, contains an overrepresentation of binocular regions of space (i.e. more neurons sampling the binocular field of view), with neurons having smaller receptive fields in some HVAs.Mice made compensatory head-eye movements that align the focea to the frontal-facing direction during open-field or object exploration, aligning the focea at the FOE, similar to what was observed by Holmgren et al. (2021).It is tempting to compare the murine focea to the cortical region innervated by the fovea in other animals, as these show anatomical and functional resemblance.
Altogether, these studies revealed that head-eye movements in mice are finely tuned to achieve optimal sampling of the visual world during visually guided behaviours, a phenomenon known as active sampling.Retinal and cortical specialisations in mice can be linked to functional binocular vision, like in other mammals.Accordingly, both binocular and stereoscopic cues can be used for prey capture and depth estimation, albeit monocular cues can also be used in some cases.

Concluding remarks
Over the last decades, the appeal of the murine visual system has dramatically increased among system neuroscientists.Large-scale recordings and manipulation of neural activity have recently been combined with high-resolution video-oculography and head tracking in The pervasiveness of task-irrelevant and task-relevant movement signals in cortical sensory areas is a well-described phenomenon in the mouse (see Parker et al., 2020 for a review).Proprioceptive (i.e.head, eye, body position or movement) and locomotor signals in primary sensory areas have been proposed to represent a corollary discharge signal.This can be used to i) inhibit the sensory reafference generated by the self-motion (Miura and Scanziani, 2022;Schneider et al., 2018), ii) generate a prediction of the sensory input generated by self-movements (Leinweber et al., 2017), or iii) modulate the gain of visual responses via neuromodulatory signals (hypothesised by Leinweber et al., 2017).In the context of visual processing, head-eye movements and locomotion constantly undermine the perceptual stability of the world and are differentially encoded by neurons in the murine visual cortex (Vélez-Fort et al., 2018;Bouvier et al., 2020;Guitchounts et al., 2020;Hennestad et al., 2021;Abdolrahmani et al., 2021;Keshavarzi et al., 2022;Parker et al., 2022b).Specifically, visual and vestibular inputs contribute to angular head-velocity tuning in the cortex, with HVAs being more strongly tuned to a combination of both stimuli than to each stimulus alone (Hennestad et al., 2021).Neurons in VISp and the HVAs, VISpm, VISam, Visa (Abdolrahmani et al., 2021) and VISrl, VISa, VISal (Itokazu et al., 2018) were also shown to increase firing prior to stimulus-elicited saccades, with premotor activity being as much as 4-fold larger than responses to visual stimuli (Abdolrahmani et al., 2021).On the other hand, new lines of evidence on visual sampling in mice and macaque monkeys suggest that head-eye movements trigger a stereotyped sequence of neuronal responses across VISp only when the movements are linked to a shift in gaze (Parker et al., 2023;Talluri et al., 2023).Thus, the motor/vestibular component of gaze-stabilising eye movements was not represented in the primary visual cortex, but whether it is in HVAs remains unknown.Locomotor signals were also shown to modulate neurons in the visual cortex.Species-specific differences emerge when comparing locomotion-induced activity in the visual cortex.Specifically, while neuronal activity in VISp positively correlated with running in mice (Niell and Stryker, 2010), firing in the primary visual cortex was suppressed during locomotion in marmosets (Liska et al., 2023).These diverging results may be partially explained by anatomical differences between primates and mice, whose visual cortex is more strongly innervated by premotor areas (Hovde et al., 2022;Leinweber et al., 2017;Markov et al., 2014) and where neuromodulatory systems show different anatomical motifs (Disney and Robert, 2019).Lastly, it was recently shown in mice that a non-visual directional signal relayed by the lateral posterior nucleus of the thalamus (the homologue of the primate pulvinar) is required to achieve saccadic suppression in VISp (Miura and Scanziani, 2022), hence to instruct the neurons in VISp that the motion perceived on the retina is self-generated.This serves to minimise perturbations of visual flow caused by the saccade.Mechanisms of saccadic suppression may involve the relay of a corollary discharge signal generated in SCm upon the initiation of movement via the pulvinar, as shown for primates, or rely on proprioceptive signals produced from the extraocular eye muscles (see Wurtz et al., 2011 for a review).E. Ambrad Giovannetti and E. Rancz mice.This, together with the strive to probe neural circuits in naturalistic tasks, has ushered in a series of studies endowed with unprecedented translational and ethological relevance.These studies revealed that, in unrestrained conditions, mice reorient their heads to optimally sample visual information both during open-field exploration and visually guided tasks.Eye movements are tightly coupled to the VOR, which suggests that virtually all eye movements of freely-moving mice are functionally gaze-stabilizing.Furthermore, mice were shown to rely on binocular vision during naturalistic tasks and marked binocular disparity tuning is present throughout their visual cortex, despite their eye movements hindering stable stereopsis.Lastly, there is evidence of cortical influence on gaze stabilising reflexes.Future studies should address the extent of cortical influence on the murine saccade and fixate behaviours, in naturalistic, ethologically relevant settings.
Importantly, measurements of attempted head movements during stimulus-oriented and spontaneous saccades allowed for the reconciliation of divergent behavioural findings in head-fixed versus freelymoving mice, and to compare these with primate behaviour.The lower extent and variability of eye movements observed in head-fixed versus freely-moving mice is indeed reflecting the lack of vestibular input resulting from head-fixed preparations (Meyer et al., 2020).In freely-moving mice, all gaze shifts were part of the saccade-and-fixate behaviour, during which the head moves first and the following gaze shift is driven by a resetting saccade.In line with this finding, all spontaneous eye movements of head-fixed mice were preceded by attempted head movements.Notably, stimulus-evoked saccades (triggered by external stimuli or SCm stimulation) result in an eye-first, head-second type of movement, in both head-fixed rodents and primates (Table 2).It is not yet known whether these gaze shifts occur in freely moving mice, e.g. during stimulus-or memory-guided behaviours, and whether they are under cortical control.Lastly, these findings indicated that mouse eye movements are always linked to head movements, a phenomenon not observed in primates.A potential explanation, requiring experimental validation, was posited by (Land, 2019), who suggested that primates evolved to perform "eye-only" gaze shifts due to the inertia and energetic cost of moving their head, higher in comparison to other animals.
In conclusion, mouse eye movement dynamics is a powerful yet poorly exploited tool to investigate the input-output transformation exerted by mammalian subcortical and cortical circuits.The brain likely uses oculomotor signals to infer the sensory consequences of voluntary and involuntary eye movements (Brooks and Cullen, 2019).Accordingly, oculomotor behaviour (Lakshminarasimhan et al., 2020) and dysfunction (Morita et al., 2020) reflect the predictions (and prediction errors) driving active sampling (Keller and Mrsic-Flogel, 2018;Perrinet et al., 2014).Causal relationships between cortical control of eye movements within the predictive processing framework can be established in the future by perturbing the murine visual system in ethologically relevant settings.Furthermore, it will be essential to establish consensual descriptors of head, eye and gaze positions in a common coordinates framework with universally agreed references and transformations.Finally, the century-long advances in peering through eye movements to glimpse aspects of human decision-making (Spering, 2022) could be a foundation for mouse research, promising a mechanistic understanding of the underlying neuronal processes.

Glossary
• Conjugate and non-conjugate eye movements are movements of both eyes occurring in the same direction (i.e.N-T for the right eye and T-N for the left, or vice versa) or movements of both eyes in opposite directions, respectively.• Convergent and divergent eye movements are non-conjugate eye movements occurring in the T-N or N-T direction for both eyes, respectively.• Ocular countertilt is a slow eye movement compensating for head pitch and roll rotations.• Nystagmus is an involuntary eye movement characterised by cycles of smooth pursuit eye movements in one direction followed by a saccade in the other direction.It can be either physiological or pathological.
• Saccade and fixate is an eye movement sequence present in all mammals.It starts with a head movement accompanied by slow VOR-driven eye movements, followed by a resetting saccade, and is finished by VOR until the head stops moving.• Binocular vision is the ability to process monocular images simultaneously.• Binocular fusion is the second grade of binocular vision; both monocular images, which are different due to the position of the eyes in the head, are merged into a single one.Binocular fusion requires both eyes to focus on the same object.• Stereopsis is the highest grade of binocular vision, supporting the ability to see in three dimensions.Stereopsis relies on binocular disparity tuning, which occurs after binocular fusion.Stereopsis facilitates depth perception and distance estimation thanks to changes in disparity tuning with self-motion.• Disparity-tuning refers to the response of neurons in VISp and HVAs to differences in monocular images (binocular disparity).Disparity tuning drives cortex-dependent vergence eye movements in primates but not in mice.Crossed disparities occur when the object is beyond the focus plane; uncrossed disparities occur when the object is before the focus plane.Disparity-tuned neurons (also known as stereosensitive neurons) are described in terms of their sensitivity to left and right eye stimuli (receptive fields) or to the sum or subtraction of the images presented to both eyes.Disparity tuning does not imply stereopsis.
• Interocular velocity differences are binocular cues not requiring binocular fusion to elicit eye movements.IOVDs are the differences in object motion in both eyes (i.e.monocular cues deriving from both eyes).IOVDs were shown to drive cortex-independent vergence eye movements in head-fixed mice.• Parallax is the apparent displacement of an object from two different lines of sight.Motion parallax refers to the phenomenon that objects closer to the observer move faster than those further away and can function as a monocular depth perception cue.

1)
In what way does the visual cortex participate in the generation of gaze-shifting and compensatory eye movements across different behaviours in mice?Is the integration of bottom-up information with top-down signals in HVAs needed to produce a seamless sequence of eye movements?Do e.g.corticotectal VOR-suppressing mechanisms exist in mice? 2) Are there any eye-first gaze shifts in freely moving mice?While the neuronal architecture is present, studies have so far failed to identify any.Future development of freely-moving behavioural tasks to evoke gaze shifts externally (by sensory stimuli) or internally (e.g. in a memory-guided fashion) could provide the answer.3) How do retinal, collicular and cortical specialisations interact in escape and orienting behaviours?Are head-eye movements always organised in the same fashion?4) Is there any stereopsis in mice?Which binocular cues (e.g.interocular velocity difference or disparity) are used for depth estimation, and can they also serve other functions (such as contrast sensitivity or brightness perception)?5) Finally, comparable reporting of eye and head movements across foraging, hunting, social interaction and escape behaviours will lead the way to establish the anatomical and computation substrates of ethological behaviours in mice.

Fig. 1 .
Fig. 1.Eye movement types categorised by their origin (reflexive or voluntary) and function (gaze-shifting or stabilising).Asterisks mark eye movements observed in primates but not yet in rodents.

Fig. 2 .
Fig. 2. Murine eye movements.a) VOR in head-fixed or body-restrained mice evoked by a vestibular stimulus in the dark.The visual fields of the eyes are shaded in yellow and blue, and the dotted line indicates the visual axis.a1, eye position at rest.a2, counterclockwise (CCW) and a3 clockwise (CW) movement of the head causes conjugate contraversive eye movements.a4, representative traces of optic flow (none), eye and head (red) movements over time.b) OKR in head-fixed mice evoked by a rotating drum of black and white gratings.b1, eye position at rest.b2, CW drum rotation results in CW eye movement; b3: CCW drum and eye rotation.b4, representative traces of the optic flow (drum rotation, red) and movements of the eyes and the head (none).c) Saccades in head-fixed mice.c1, a representative CCW saccade; red and blue lines indicate end eye position.c2, average amplitude and speed of saccades in head-fixed mice, modified from (Meyer et al., 2020).The grey shaded area indicates the saccade.d) Vergence eye movements.Convergence: non-conjugate movement in the T-N direction.Divergence: non-conjugate eye movement in the N-T direction.Note that the T-N movement is faster and larger in amplitude than the corresponding N-T movement, leading to the widening of the binocular field during VOR, OKR and saccades.

Fig. 3 .
Fig. 3. Peripheral afferents and brain regions involved in visuo-vestibular signal processing and transformation into an eye movement command in rodents.Parallel retino-tectal (orange arrow), retino-geniculate (green arrow)and projections from the retina to the accessory optic system (turquoise arrow; AOS, comprising the nuclei NOT/DT, MT, LT) support gaze-stabilising and gaze-shifting eye movements in mice.Visuo-vestibular gaze stabilisation is dependent on the AOS and its direct or indirect projections to the oculopremotor nuclei via the vestibular premotor nuclei or the inferior olive (IO).Activity of the NOT/DT is also modulated by projections from the primary and higher visual cortex.The floccular lobe of the cerebellum receives innervation from the AOS via the IO.The flocculo-cerebellum, the semicircular canals and spinal cord modulate the plasticity of gaze-stabilising reflexes via projections to the vestibular premotor nuclei, which in turn project to the visual/vestibular thalamic nuclei and to the oculomotor nuclei.The visuo/vestibular thalamic nuclei relay vestibular premotor nuclei activity to the retrosplenial cortex (RSP), higher visual areas (HVAs) and primary visual cortex (VISp).Gazestabilising eye movements are also coupled to locomotion via spinal-extra ocular afferents to the oculomotor nuclei.Gaze shifts are mediated by the superior colliculus (SC), either via direct retino-tectal projections, or they undergo indirect cortical modulation via the cortico-tecal pathway.Projections from the motor regions of the SC reach the oculo-premotor nuclei (tecto-reticular pathway), which in turn project to the oculomotor nuclei.A direct contribution of the supplementary motor cortex (MOs) to eye movements has also been described in mice.Initiation of gaze-stabilising and gaze-shifting eye movements is mediated by the contraction of the eye muscles, innervated by the third, fourth and sixth cranial nerves (CN).Indirect pathways from AOS to IO through ventral tegmental area and cortico-striatal-nigro-tectal projections are not shown.The dashed line indicates projections from VISp to the superficial layers of the SC.
signals in the visual cortex.

Fig. 4 .
Fig. 4. Saccade and fixate eye movement sequence in freely-moving mice.a) Representative trajectories of eye-in-head (magenta), gaze (purple), and head position (orange) over time during the saccade and fixate sequence.The shaded area indicates the gaze-shifting saccade; the various phases of the movement are indicated with numbers.b) Phases of the saccade and fixate behaviour.The orange arrow indicates the position and movement of the head; the dotted line indicates the gaze position for each eye; the purple arrow shows the centre of the binocular field.1, eye, gaze, and head position at rest.2, slow, gaze-stabilising phase: the head turns in the CW direction and the eyes move in the opposite direction.3, fast, gaze-shifting phase: eyes saccade towards the direction of the head.4, slow, gazestabilising phase: the head keeps turning and eyes move in the opposite direction to recentre the gaze.

Table 1
List of brain regions known to contribute to the generation of eye movements in primates and/or mice.Only the most relevant subcortical nuclei are listed.The colour code indicates the species: primate (blue), mouse (magenta), or both (orange).Asterisks mark eye movements only described in primates.

Table 2
The sequence of gaze shifts under different conditions.EF -eye first; HF -head first.