Arrest is often thought of as an avoidance response, and is called freezing when the mouse is in a context with a potential threat. However, arrest may also be part of more exploratory behaviours, allowing for a pause for surveillance, or attention to the external environment (Botta et al., 2020). Direct manipulations of SC support the idea that the optic layer is important in arrest. Activation (Sans-Dublanc et al., 2021) or inhibition (Xie et al., 2021) of neurons concentrated in the optic layer induces or impairs arrest, respectively. Other manipulations of SC that evoke arrest have either included (Wei et al., 2015), or have focussed on (Zingg et al., 2017), the optic layer. Manipulation of CAMK2+ SC neurons at the border of the optic layer and intermediate layers can also elicit arrest behaviour (Wei et al., 2015; Cai et al., 2022). In common with some other neurons in the optic layer, these neurons appear to project to the lateral posterior nucleus of the thalamus (LP) and activation of their terminals in LP evokes arrest (Wei et al., 2015). For simplicity, we therefore include these neurons as part of the optic layer.

Arrest behaviour is more likely to be elicited by activation of medial SC, than lateral SC (Wei et al., 2015). This medio-lateral asymmetry is consistent with the previously hypothesised medio-lateral separation of function (Dean et al., 1989). Our hypothesis predicts that the absence of arrest behaviours during activation of lateral SC can be explained by the fact that the optic layer does not extend into the lateral most columns of SC (Figure 1; Benavidez et al., 2021).

It is well established that activation of the motor-related SC in freely moving animals produces contralateral turning movements, or biases an animal towards making them (Stubblefield et al., 2013; Masullo et al., 2019; Cregg et al., 2020; Essig et al., 2021). In head-fixed mice, activation of motor-related SC biases eye movements (Zahler et al., 2021) towards the contralateral side. Inhibiting motor-related SC does the opposite, biasing the animal towards ipsilateral movements (Stubblefield et al., 2013; Lee and Sabatini, 2021). We note that when head-fixed mice rotate a ball or wheel beneath them, most active neurons in motor-related SC prefer ipsilateral turns (Steinmetz et al., 2019) and unilateral inhibition of motor-related SC biases animals away from ipsilateral turns (Huda et al., 2020). The likely explanation is that these “ipsilateral” turns require mice to push the ball or wheel down on its ipsilateral side, a movement associated with contralateral turning during free behaviour (Huda et al., 2020).

The intermediate layers of SC appear to be particularly important in turning. Activation of PITX2+ neurons in SC, which are concentrated in the intermediate layers, evokes turning (Masullo et al., 2019). The intermediate (but not deep layers) are targeted by the substantia nigra pars reticulata (SNr; Lee J. et al., 2020), and activation of SNr terminals in SC induces turning (Villalobos and Basso, 2020). Likewise, the intermediate (but not deep layers) project to neurons in the gigantocellular nucleus that are important in turning (Cregg et al., 2020).

Lateral SC is known to be involved in turning, and there is some evidence for a role of medial SC in turning. Activation of lateral SC mostly evokes contralateral turning (Isa et al., 2020), while inhibition of lateral SC neurons can impair instinctive turning towards sounds (Vale et al., 2020) and other turning behaviours (Sooksawate et al., 2013; Isa et al., 2020). Circuits for turning may, however, include medial SC as well as lateral SC. Activation of PITX2+ neurons in medial SC induces head turns (Masullo et al., 2019). Activating progressively more medial locations of SC evoked larger pitch angles of head rotation, suggesting that medial SC is involved in making movements towards more elevated angles (Masullo et al., 2019). Pitch rotations of the head produced by medial SC are difficult to measure and have only been achieved in a limited number of experiments–more data would help strengthen the case for a role of medial SC in turning.

Superior colliculus is primarily concerned with turning towards the contralateral side, but activation of anterior-medial SC can also evoke ipsiversive head turns (Isa et al., 2020), and an ipsiversive bias has been reported for some medial SC neurons (cuneiform nucleus-projectors; Isa et al., 2020) but not others (primary auditory cortex-recipient; Zingg et al., 2017). Activation of inhibitory, GABAergic SC neurons can bias animals towards contralateral movements (Essig et al., 2021; Sans-Dublanc et al., 2021) or ipsilateral movements (Duan et al., 2021; Hao et al., 2021); the specific movements elicited by activating subpopulations of neurons in SC is therefore likely to depend on both the specific projection patterns and effects (excitatory, inhibitory) of those neurons.

Circuits in medial SC are clearly important in triggering escape behaviours. Activation of medial, motor-related SC neurons can evoke escape or putatively escape-related backwards walking and fast forwards running (Zingg et al., 2017; Evans et al., 2018; Isa et al., 2020); inhibition of medial motor-related SC impairs the triggering of escape (Evans et al., 2018). By contrast, inhibition of lateral SC does not affect the production or speed of escape behaviour (Shang et al., 2019; Vale, 2020; Huang et al., 2021), though it can affect the direction of escape (Vale et al., 2020).

Circuits in lateral motor-related SC appear more important in triggering capture behaviours. Inhibition of lateral motor-related SC neurons (including those projecting to subthalamus- or substantia nigra pars compacta, SNc; Shang et al., 2019; Huang et al., 2021) impairs capture, that is movements towards prey (Shang et al., 2019; Huang et al., 2021; Xie et al., 2021), food or conspecifics (Huang et al., 2021).

Escape and capture are both complex behaviours, that involve a combination of actions, such as turns combined with locomotion. However, activations that induce escape and capture behaviours produce turns that appear goal directed, rather than the stereotypical ego-centric turns that comprise “turning” behaviours described above (Evans et al., 2018; Shang et al., 2019; Huang et al., 2021; Xie et al., 2021). Whether and how escape and capture behaviours recruit circuits for “turning” is not yet clear. Activation of PITX2+ neurons in intermediate layers of SC induces turning without locomotion (Masullo et al., 2019), but large-scale inhibition of PITX2+ neurons does appear to impair capture (Xie et al., 2021). Thus, while we hypothesise that turning is supported primarily by neurons in intermediate SC, and triggering of capture and escape is supported primarily by neurons in deep SC, direct manipulations of motor-related SC are yet to reveal the relative contribution of intermediate and deep parts of SC in these behaviours.

We hypothesise that SC plays a similar role in both capture and escape: the output of deep layer SC triggers goal-directed action. The specific goal of that action depends on whether that signal arises in the lateral (e.g., prey) or medial (e.g., refuge) subdivisions of the deep layer, because these subdivisions have different connections to the rest of the brain, as we review below. However, the computations performed by the lateral- and medial parts of deep layer SC are predicted to be the same in both cases.

The major sensory inputs to SC–visual, auditory and somatosensory–are organised into aligned topographic maps (Drager and Hubel, 1975; Ito et al., 2020). These maps provide a representation of the egocentric direction of an object relative to the animal’s head: the direction of an auditory or somatosensory stimulus is directly related to the head-centric direction of the object that produces them; the location of an object’s image on the retina, if eye-movements are ignored, is also a proxy for head-centric object direction. These sensory maps are aligned parallel to the surface of SC, orthogonal to the proposed laminar organisation of function. The sensory maps may therefore be important in guiding and constraining the behaviour(s) that are elicited by sensory stimuli presented at particular directions relative to the animal.

The alignment between sensory and motor maps in SC is likely to be important in turning behaviours. Visuosensory SC includes “narrow-field” cells (Gale and Murphy, 2014) that project into topographically aligned parts of motor-related SC and appear important in turning towards prey (Hoy et al., 2019). Consistently, activation of motor-related SC evokes turning towards the directions that are represented by the equivalent location in the sensory maps (Figure 2). Activation of PITX2+ neurons at specific locations in the intermediate layers evokes contralateral turns towards specific directions (Masullo et al., 2019). Activation of more posterior PITX2+ neurons evokes larger contralateral (yaw) turns, consistent with the more temporal receptive fields found in posterior SC. Activation of more medial PITX2+ neurons evokes larger pitch turns, consistent with the more elevated receptive fields found in the medial SC. Similarly, in head-fixed mice activation of more posterior SC neurons induces more temporal eye turns (Wang et al., 2015).

The relationship between topographic sensory maps and other actions (escape, capture, arrest) is less clear. Activation experiments show that motor-related medial SC is important in escape behaviour, and visuosensory medial SC represents the overhead visual field. If this topographic alignment were important for escape behaviour, then escape behaviours should be more easily elicited by stimuli in the upper visual field (cf., Dean et al., 1989), which would include aerial, and tall ground-based predators. Indeed, an expanding black disc on a screen (“looming stimulus”) presented to the upper visual field usually elicits a rapid escape to refuge when one is present (Yilmaz and Meister, 2013). Whether looming stimuli from other visual directions can induce escape responses is less clear, but limited work suggests that a looming stimulus in front (Zhou et al., 2019) or below (Yilmaz and Meister, 2013; Zhou et al., 2019) a mouse does not elicit the same rapid escape. Capture can be directed towards stimuli in the lower visual field (Hoy et al., 2016; Vale et al., 2020), and perhaps specific parts of the lower visual field (Hoy et al., 2016, 2019; Michaiel et al., 2020; Holmgren et al., 2021; Johnson et al., 2021). Whether capture behaviours can be evoked by a stimulus presented to the upper visual field remains to be seen. Behavioural work suggests that arrest can be induced by a visual stimulus presented to either the upper or lower visual field (De Franceschi et al., 2016; Procacci et al., 2020).

Connections of visuosensory SC, and the optic layer in particular, are consistent with a role in arrest. The thalamic area LP and pontine area laterodorsal tegmental nucleus are innervated by the optic layer (Xie et al., 2021), and are involved in arrest (Yang et al., 2016; Shang et al., 2018). Midbrain area parabigeminal nucleus (PBG) has reciprocal connections with visuosensory SC (including the optic layer; Zhang et al., 2019; Tokuoka et al., 2020), and may be involved in the production of arrest following escape (Shang et al., 2018). Primary visual cortex innervates the visuosensory SC, including the optic layer, and activation of these terminals in SC (Liang et al., 2015) or of SC neurons post-synaptic to them (Zingg et al., 2017) induces arrest, while inhibition of primary visual cortex impairs arrest to light flashes (Liang et al., 2015). Interestingly, ventral stream higher visual areas primarily project to optic layer (Wang and Burkhalter, 2013) but whether they also have a role in arrest is not yet known. By contrast, while inhibition of primary auditory cortex also impairs sound-induced arrest (Li et al., 2021), primary auditory cortex’s projection to SC does not innervate the optic layer (Benavidez et al., 2021), and inhibition of SC does not impair these sound-induced arrest behaviours. The stimulus selectivity of visuosensory SC is generally broad (e.g., Gale and Murphy, 2014; De Franceschi and Solomon, 2018), and the sensory signals in these layers are therefore likely able to guide many or even most behaviours. Particular pathways through visuosensory SC may nevertheless be more important for some behaviours than others (e.g., Reinhard et al., 2019); whether projections from visuosensory to motor-related SC are particularly important for arrest, remains to be determined.

Central amygdala (CeA) is an end target of many of the pathways that project from the optic layer of SC. LP is indirectly connected to CeA through the basolateral amygdala (Fadok et al., 2018), and PBG projects directly to CeA (Shang et al., 2015). Activation of medial SC populations concentrated in (NTSR1+) (Gale and Murphy, 2014), or including (CAMK2+) (Wei et al., 2015), the optic layer activates CeA and induces arrest (Sans-Dublanc et al., 2021), and manipulating CeA alters freezing responses to visual looming stimuli (Zelikowsky et al., 2018). CeA may promote arrest or freezing via several potential pathways, perhaps even through its projection to periaqueductal grey (PAG) (Tovote et al., 2015; Vaaga et al., 2020; Yu et al., 2021).

Intermediate layers of SC are likely to be particularly important in turning. CHX10+ gigantocellular nucleus neurons, which are a potential route through which lateral SC promotes contralateral turning, receive input from intermediate layers (but not deep layers) of SC (Cregg et al., 2020). Intermediate layers (but not deep layers) are innervated by the substantia nigra pars reticulata (Lee J. et al., 2020), and manipulation of that input induces turning (Villalobos and Basso, 2020). Dorsal stream higher visual areas, which may play a role in representing turn directions (Itokazu et al., 2018; Odoemene et al., 2018), specifically target intermediate layers (Wang and Burkhalter, 2013).

Areas involved in turning appear to preferentially connect to lateral SC. Many of the connections of lateral SC (Benavidez et al., 2021) are known to have a role in producing movements towards particular egocentric directions, either of the body (including gigantocellular nucleus; Cregg et al., 2020) or its parts (motor cortex; fastigial nucleus of the cerebellum; motor thalamus; Guo et al., 2014, 2017; Hira et al., 2015; Morandell and Huber, 2017; Gao et al., 2018; Heindorf et al., 2018; Mayrhofer et al., 2019). Lateral SC also provides output to the medullary reticular nucleus (Esposito et al., 2014) that may be important in movements of the forelimb contralateral to SC (Ruder et al., 2021), and to the parvicellular reticular nucleus that may also be important in forelimb (Ruder et al., 2021) and tongue-jaw movements (Han et al., 2017).

While there is good evidence that lateral SC is involved in turning movements, PITX2+ SC neurons are found in medial- as well as lateral SC (Masullo et al., 2019), and these neurons are known to be involved in turning. The connection pattern of PITX2+ neurons is similar to that found for non-specific tracing from lateral SC (Masullo et al., 2019; Xie et al., 2021). This suggests that while areas involved in turning have stronger connections with lateral SC, they are also connected to medial SC.

Superior colliculus also sends projections to basal ganglia nuclei, including ventral tegmental area (VTA; e.g., Zhang et al., 2019; Zhou et al., 2019) and the subthalamic nucleus. Whether these SC projections help generate specific turning actions, or more complex behaviours, is not yet clear (Hughes et al., 2019; Zhou et al., 2019; Barbano et al., 2020).

Deep layers of SC are connected to areas that are thought to be important in triggering more complex movements. Deep lateral SC is connected to areas (Benavidez et al., 2021) involved in triggering capture [including zona incerta, medial part; substantia innominata; ventromedial hypothalamus, ventrolateral part (Lee et al., 2014; Zhao et al., 2019; Zhu et al., 2021)]; and activation of SC terminals in the zona incerta (Shang et al., 2019; Xie et al., 2021) or substantia nigra pars compacta (SNc; Huang et al., 2021), facilitates capture. Lateral SC also projects to the lateral hypothalamic area (Venner et al., 2019; Benavidez et al., 2021), also potentially involved in capture (Li Y. et al., 2018). Lateral SC receives input from the dentate and interposed nuclei of the cerebellum, areas which might be involved in triggering goal-directed movements of the forelimb (Dacre et al., 2021) and other body parts.

Deep medial SC is connected to areas involved in evoking fast locomotion (including cuneiform nucleus and magnocellular reticular nucleus; Capelli et al., 2017; Caggiano et al., 2018) and triggering escape (including dorsal periaqueductal grey, dPAG; lateral parabrachial nucleus; lateral septum; ventromedial hypothalamus, dorsomedial/central part, VMHdm/c; Han et al., 2015; Kunwar et al., 2015; Deng et al., 2016; Tovote et al., 2016; Evans et al., 2018; Azevedo et al., 2020; Sun et al., 2020). Activation of terminals of SC neurons in the PAG evokes mild running in head-fixed mice (Wang et al., 2019a) and “wild running or backward fleeing behaviours” in freely moving mice (Wei et al., 2015). Some medial SC neurons project to both cuneiform nucleus and dPAG (Isa et al., 2020).

Whether triggering of behaviour is the preserve of deep layers, or also involves intermediate layers is not yet clear. Capture-associated connections (zona incerta, medial part and ventromedial hypothalamus, ventrolateral part) also contact intermediate layers (Benavidez et al., 2021). In the case of escape, DRD2+ and PITX2+ neurons both have patchy labelling in intermediate layers of SC, but have different projections, and activation of the former can trigger escape, suggesting that some intermediate layer neurons are involved in triggering escape (Masullo et al., 2019; Montardy et al., 2021; Xie et al., 2021). However, deep layers of SC alone are innervated by VMHdm/c (Benavidez et al., 2021), and dPAG also gets more input from the deep layers of SC than intermediate layers (Evans et al., 2018). VMHdm/c and dPAG have roles in triggering escape behaviour, so their connection to the deep rather than intermediate SC would suggest the deep layers are more important in triggering these behaviours.