Journal of Neurophysiology

Exploring the Superior Colliculus In Vitro

Tadashi Isa, William C. Hall


The superior colliculus plays an important role in the translation of sensory signals that encode the location of objects in space into motor signals that encode vectors of the shifts in gaze direction called saccades. Since the late 1990s, our two laboratories have been applying whole cell patch-clamp techniques to in vitro slice preparations of rodent superior colliculus to analyze the structure and function of its circuitry at the cellular level. This review describes the results of these experiments and discusses their contributions to our understanding of the mechanisms responsible for sensorimotor integration in the superior colliculus. The experiments analyze vertical interactions between its superficial visuosensory and intermediate premotor layers and propose how they might contribute to express saccades and to saccadic suppression. They also compare and contrast the circuitry within each of these layers and propose how this circuitry might contribute to the selection of the targets for saccades and to the build-up of the premotor commands that precede saccades. Experiments also explore in vitro the roles of extrinsic inputs to the superior colliculus, including cholinergic inputs from the parabigeminal and parabrachial nuclei and GABAergic inputs from the substantia nigra pars reticulata, in modulating the activity of the collicular circuitry. The results extend and clarify our understanding of the multiple roles the superior colliculus plays in sensorimotor integration.


Understanding the mechanisms responsible for the translation of sensory signals into the commands for movement is one of the fundamental goals of neurobiology. Orientation of the eyes toward external stimuli is a common choice for analyzing these mechanisms not only because it is a nearly universal behavior but also because the constant relationship between stimulus location and the direction of a gaze shift suggests that it may provide a relatively simple model for studying how sensory systems initiate and guide movement.

In the superior colliculus, the juxtaposition of cellular layers with sensory and motor functions is especially useful for studies of these mechanisms. Many types of experiments in a wide variety of species have demonstrated that visuosensory cells in the superficial layers of the superior colliculus (SL)1 encode the location of objects in a retinotopic map while premotor cells within its intermediate layers (SGI) are organized into a spatial map that encodes the vectors of the rapid shifts in gaze direction called saccades (Mohler and Wurtz 1976; Schiller and Stryker 1972; Sparks 1978; Wurtz and Goldberg 1972). Experiments have further shown that the SL visuosensory cells receive input from the retina and visual cortex, and the SGI premotor cells transmit command signals to the midbrain and pontine gaze centers that organize and initiate the saccades (Grantyn and Berthoz 1985; Guitton and Munoz 1991; Keller 1979; Ma et al. 1991; May 2006). But while these observations clearly indicate a role for the superior colliculus in the initiation of saccades, we still know very little about how its circuitry performs this function. Until recently, models of these circuits were based primarily on the morphological characteristics and arrangements of the nerve cells (Ramón y Cajal 1995) (Fig. 1A) and on extracellularly measured physiological properties, such as, the relative timing of the responses of different cell types and characteristics of their receptive and movement fields (Fig. 1B) (Mohler and Wurtz 1976). However, these models were difficult to test because the available extracellular methods did not provide sufficient detail about the nature of the synaptic interactions among the different cell types. To surmount this limitation, our two laboratories began using—for the first time in the superior colliculus—in vitro whole cell patch-clamp methods to perform detailed structural and functional analyses of the collicular circuitry in rats and mice. These rodents, like primates, generate rapid, saccade-like orienting movements of their eyes (McHaffie and Stein 1982; Sakatani and Isa 2004, 2007, 2008) and possess a laminar organization of the superior colliculus that is almost indistinguishable from the primate's (May 2006). The long-term goal of this research is to use the in vitro approach to learn how collicular circuits integrate sensory and motor signals to initiate gaze shifts. In this review, we describe some of our recent findings and discuss how they have helped resolve longstanding issues about collicular structure and function.

Fig. 1.

The great unraveled knot. A: Ramón y Cajal's drawing of the cell types and arrangements in the superior colliculus of the rabbit (Ramón y Cajal 1995). The letters indicate cell types revealed in the layers of the superior colliculus by the Golgi method. The Golgi method shows potential connections and eliminates others but doesn't establish either the connections that are formed or the nature of the synaptic interactions among the different cell types. B: models of the circuitry linking the layers of the superior colliculus proposed by Mohler and Wurtz (1976). Left: the traditional model that argues that the superficial visuosensory layers project to the intermediate premotor layers which in turn project out of the superior colliculus. Middle: pathways from the superficial and intermediate layers converge in an intercalated layer that provides the major output from the superior colliculus. Right: the layers have independent inputs and outputs. The available extracellular methods could not distinguish with certainty among these models. SL, superficial layers; SZ, stratum zonale; SG, superficial gray layer; SO, optic layer; IL, intermediate layer; IG, intermediate gray layer; IW, intermediate white layer.

Pathway from SL to SGI and its contribution to express saccades

Experiments have demonstrated that the retinotopic map of visual space in SGS is in register with the spatial map of saccade vectors in SGI (Schiller and Stryker 1972; Wurtz and Goldberg 1972). The registry of these sensory and motor maps might readily be explained by a simple model that postulates stimuli located within one area of the visual field activate neurons within a restricted region of the retinotopic map in SL. According to the model, this region in turn gives rise to descending columnar projections that activate the SGI premotor neurons that command gaze shifts toward the same area of the visual field. However, the initial results of both anatomical and physiological experiments seemed to support the contrary conclusion that there are no direct pathways from the visuosensory to the motor layers. For example, Sprague (1975) plotted in cats the anterograde axonal degeneration produced by small collicular lesions in an attempt to trace a pathway from SL to SGI, but, following these lesions, he found only very sparse degeneration in SGI. On the basis of results from a series of progressively deeper lesions, he argued SL is connected with SGI only indirectly via a multisynaptic pathway in which the cells at a particular depth project primarily to neighboring subjacent cells. Even the existence of a multisynaptic pathway from SL to SGI was questioned by Edwards (1980), who used autoradiographic methods in the cat to trace the connections between the layers. He could find no convincing evidence for a pathway to SGI following SL injections of tritiated amino acids and concluded that the labeled axons he could detect in SGI were only passing through this layer in route to other brain stem structures. The concurrent results of physiological experiments in monkeys seemed to support Edward's conclusion (Mays and Sparks 1980). These experiments showed that when successive saccades are made to two targets, both of which are extinguished before the onset of the first saccade, activity in SL is neither necessary nor sufficient to evoke responses in the underlying SGI cells. That is, cells at the appropriate location in the motor map of SGI discharged prior to the second saccade even in the absence of activity in the overlying SL cells, and the activity evoked in SL by the second target did not generate presaccadic activity in the subjacent SGI cells. The conclusion that pathways other than one from SL activate the premotor cells in SGI also was supported by behavioral experiments in tree shrews that showed that removing the superficial layer does not obviously impair visual orientation (Casagrande and Diamond 1974).

However, in contrast to the results of these initial experiments, more recent studies of several species, based both on extracellular injections of more sensitive axonal tracers (Behan and Appell 1992; Lee and Hall 1995; Rhoades et al. 1989) and also on intracellular injections of tracers (Hall and Lee 1997; Mooney et al. 1988a; Moschovakis et al. 1988), have provided evidence for a direct pathway from SL to SGI. These later experiments served to re-raise the question of the contributions this pathway makes to visually guided gaze shifts. Because establishing the synaptic relationships between the sensory and motor layers is fundamental for understanding the sensorimotor functions of collicular circuitry, our two laboratories addressed this question using in vitro whole cell patch-clamp recordings (Helms et al. 2004; Isa et al. 1998; Lee et al. 1997). These experiments independently provided convincing evidence for a powerful, monosynaptic, excitatory pathway from SL to SGI.

In the initial experiments, electrical stimulation in SL was used to evoke responses in SGI cells (Isa et al. 1998; Lee et al. 1997), but in later experiments, photostimulation in SL was substituted for electrical stimulation to avoid the possibility of activating fibers from unknown sources that might pass through SL in route to SGI (Callaway and Katz 1993; Dalva and Katz 1994; Katz and Dalva 1994). To photostimulate neurons in the slice, ultraviolet light from an argon ion laser was delivered through a microscope objective and focused on an area 50 μm in diameter at the focal plane to obtain localized release of glutamate from caged glutamate [gamma-(alpha-carboxy-2-nitrobenzyl) ester]. Because axons of passage lack glutamate receptors, only cell somas and their dendritic processes are activated by the uncaged glutamate. Figure 2 illustrates results from an experiment in which SL photostimulation with uncaged glutamate was combined with whole cell patch-clamp recordings in SGI (Helms et al. 2004). The results demonstrated that SL stimulation evokes individual excitatory postsynaptic currents (EPSCs) or EPSC clusters in SGI cells (Fig. 2B). The elimination of these evoked responses by adding the voltage-gated sodium channel blocker TTX into the slice bath to block action potentials confirmed that the responses were synaptically mediated EPSCs (Fig. 2C); that is, they were not evoked directly by diffusion of the uncaged glutamate to the recorded cell.

Fig. 2.

Neurons in SGS project to intermediate SG (SGI). A: photograph of the slice. The activity of a premotor cell in SGI (yellow circle) was recorded with a pipette that approached the cell from the lower left, while the sites labeled 1-10 were photostimulated. The overlay of concentric circles indicates distances. B: voltage-clamp recordings of the responses evoked by stimulation at each site. White trace at bottom indicates the onset and duration of the photostimulus. C: excitatory postsynaptic currents (EPSCs) evoked by stimulation at site 6 were completely blocked by bath application of 1 μM TTX. The responses evoked by stimulation at site 6 in B and C differ slightly because response C was evoked later in the experiment, just before the application of TTX. (Helms et al. 2004).

This pathway from SL to SGI may be responsible for the generation of the very short latency eye movements called “express” saccades, which, in primates at least, are dependent on an intact superior colliculus (Schiller et al. 1987). The reaction times for saccades have a bimodal distribution (Fischer and Boch 1983; Sommer 1994). “Regular” saccades have reaction times in the range of 150–250 ms depending on the context and on the training of the subject. In contrast, express saccades have latencies in the range of 70–120 ms. These shorter latencies approximate the sum of the latency of visually evoked responses in SGI (∼55 ms) plus the time interval between the discharge of the SGI cells and the onset of a saccade (∼20 ms) (see Edelman and Keller 1996). Thus because the timing of the collicular responses are consistent with the latency of express saccades, it is reasonable to propose that they are mediated by pathways from the retina and visual cortex that project to the SGI via a relay in SL (Isa 2002). Regular saccades, such as those studied by Mays and Sparks (1980) might depend on less direct pathways to the brain stem gaze center and on the timing of a process of disinhibition that facilitates the flow of activity to these centers (see next section).

The in vitro experiments also showed that stimulation within a column (∼500 μm diam) extending superficially from the recorded SGI cell evokes the largest and most reliable responses, but that off-axis stimuli are effective as well (Fig. 3) (Helms et al. 2004). In the experiment summarized in Fig. 3, EPSCs could be evoked by stimuli as far as 1,000 μm off-axis from the postsynaptic SGI neuron (sites 4 and 6, Fig. 3). Because photostimulation reveals only monosynaptic connections formed by the neurons that are activated by the photo-released glutamate and does not activate fibers of passage (Callaway and Katz 1993; Dalva and Katz 1994), it has sufficient spatial resolution to compare the sizes of receptive fields with the measured dimensions of dendritic and axonal terminal fields. For example, narrow field vertical cells are visuosensory SL neurons that have small dendritic field diameters in the range of 200 μm (Fig. 4A) (Langer and Lund 1974; Lee and Hall 1995). In contrast, wide field vertical cells in SL have dendrites that can span as much as 2 mm across the visuosensory layers (Fig. 4B) (Endo and Isa 2001; Isa et al. 1998; Lee and Hall 1995; Mooney et al. 1988b; Moschovakis 1988; Saito and Isa 1999). Both cell types project to SGI where the terminal fields of their axons and the dendritic fields of the SGI cells are spatially restricted; the radii of the axonal arbors rarely exceed 150 μm while the radii of the largest SGI dendritic trees are on the order of 450 μm (Hall and Lee 1997; Lee and Hall 1995). Taken together, these measurements support the conclusions that wide field vertical cells with their 2-mm-diam dendritic trees account for the broad spatial distribution of effective stimulation sites illustrated in Fig. 3C, whereas the columnar projections from the narrow field cells account for the central region that evokes the largest and most reliable responses in the underlying SGI cells (Fig. 4C). In vivo, the wide field cells may decrease the threshold for detecting moving stimuli (Mooney et al. 1988b) by temporally summing the activity generated as a stimulus sweeps across the retina, while the narrow field cells may more precisely specify their location.

Fig. 3.

Mapping synaptic inputs to a SGI cell. A: drawing of the biocytin-filled premotor cell (yellow) superimposed on a photograph of the slice. Blue circles are photostimulation sites and are to scale (∼50 μm diam). The concentric circles of 100, 300, 500, and 1,000 μm radii indicate distance. The asterisk indicates recurrent collaterals. B: black-and-white inset is a photograph of the biocytin-filled premotor cell. Axons from the cell enter the collicular commissure (single arrow) and the predorsal bundle (double arrows). C: photostimulation in SGS evokes EPSCs in this premotor cell. The corresponding numbers on the blue circles in A and C indicate the locations of the stimulation sites responsible for the voltage-clamp recordings in C. The yellow traces were recorded at the chloride reversal potential, Vh = −82 mV. D: some sites close to the cell soma evoked inhibitory PSCs (IPSCs) when the cell was clamped at a voltage depolarized to Ecl (Vh = −53 mV) for hhhred traces at 13, 17, and 18 in C and D. (Helms et al. 2004).

Fig. 4.

Dimensions of dendritic and axonal fields in SGS and SGI. A: drawings of SGS narrow (A) and wide field (B) cells intracellularly filled with biocytin to reveal the extent of their dendritic fields (Lee and Hall 1995). C: schematic drawing of cells in SGS and SGI showing the dimensions of their dendritic and axonal terminal fields. The anatomical dimensions correspond closely to the dimensions of the receptive fields mapped with photostimulation (Fig. 3).

Contributions of collicular disinhibition to the generation of saccades

Express saccades in primates are preceded by a high-frequency burst of SGI activity that occurs at about the same latency as the visuosensory activity in this layer evoked by stimulation of the retina (Edelman and Keller 1996). These saccades occur most frequently in experimental paradigms that provide the subject both a time gap between a fixation light offset and the onset of a visual target stimulus (Fischer and Boch 1983) and also a high level of certainty during the time gap about the eventual target location (Dorris et al. 1997; Fischer and Boch 1983; Paré and Munoz 1996). It has been proposed that a reduction in the level of tonic inhibition in SGI occurs during this time gap and that the combination of this disinhibiton with the activity evoked by the visual target leads to the generation of the high-frequency “bursts” of action potentials in SGI that initiate the saccades (Dorris et al. 1997; Munoz and Wurtz 1995). Orienting movements with extremely short latencies also occur in species other than primates, including cats (Baro et al. 1995; Isa and Sasaki 2002), although, presumably for technical reasons, they have not yet been demonstrated in rodents. However, very-short-latency orienting behaviors are adaptive in many situations, and it is reasonable to conjecture that they are part of a behavioral repertoire shared by a wide variety of mammals.

Evidence suggests that the high-frequency bursts of activity in SGI are dependent on a reduction of tonic inhibition in this layer prior to regular as well as express saccades. In vivo experiments have demonstrated that substantia nigra pars reticulata (SNr) is a major source of this tonic inhibition in SGI (Chevalier et al. 1985; Hikosaka and Wurtz 1985a,b; Liu and Basso 2008), and in vitro experiments now have clarified the contribution of this inhibition to the build-up of the high-frequency bursts. In particular, these in vitro experiments show that the synaptic responses evoked in SGI neurons by SL stimulation are markedly augmented and change to high-frequency bursts when GABAergic inhibition is reduced by the addition of GABAA receptor antagonists to the slice bath (Isa et al. 1998). These experiments suggest that the reduction in the tonic inhibition from SNr “opens a gate,” thereby permitting a nonlinear signal amplification of inputs to the premotor cells that evokes the high-frequency responses.

In support of this proposal, an in vitro study by Petit et al. (1999), found that photostimulation in SGI generated strong bursts of EPSCs in neighboring premotor cells. Because the slice isolates the superior colliculus from extrinsic sources of synaptic input and because photostimulation does not activate fibers of passage, the results suggested that the nonlinear amplification is provided by excitatory connections formed by recurrent collaterals, which in both primates and rodents are known to arise from the premotor cell axons (Fig. 3) (Hall and Lee 1997; Mooney et al. 1988a; Moschovakis et al. 1988). A similar conclusion was reached in an in vitro study by Saito and Isa (2003) in which optic tract fibers were electrically stimulated during simultaneous patch-clamp recordings from a pair of SL and SGI neurons (Fig. 5). They found that the addition of the GABAA receptor blocker bicuculline amplified the evoked responses in both SL and SGI but that the amount and duration of the amplification was much greater in SGI. The same conclusion also was reached by Özen et al. 2000, who showed that single brief current pulses applied to SL generated large clusters of EPSCs in SGI cells but rarely evoked bursts of action potentials in SL cells. These in vitro results all support the idea that circuitry within the superior colliculus, and especially within SGI, plays an important role in building up the high-frequency presaccadic bursts and that this build-up is facilitated by disinhibition.

Fig. 5.

Simultaneous recordings from a pair of SGS and SGI neurons. Five traces are superimposed. A: synaptic responses in SGS (top) and SGI (bottom) neurons evoked by stimulation of the optic layer (arrow, 100 μA) in the control solution. B: evoked synaptic responses in SGS (top) and SGI (middle and bottom) after application of 10 μM bicuculline. Bottom traces: slower sweep records of the middle ones. C: drawing of SGS and SGI neurons filled with biocytin during the recordings. (Saito and Isa 2003).

A picture of how the presaccadic activity builds up over time in a large widely distributed population of SGI cells was provided in in vitro experiments by Phonghanphanee et al. (2008a), who recorded field potentials induced by electrical stimulation of a point in the SL simultaneously over time from a wide area of a slice using a 64 recording electrode array (Fig. 6A). In their control situation, they found that stimulation of the optic tract or SL evokes brief excitatory responses in SL and in the subjacent region of SGI (Fig. 6, B–D). When the stimulation was repeated with bicuculline in the slice bath, the activity spread over a wider area in both SL and SGI, but the spread was much more extensive and prolonged in SGI (Fig. 6, B, C, and E). Addition of the N-methyl-d-aspartate (NMDA) receptor blocker 2-amino-5-phosphonovaleric acid (APV) together with bicuculline to the bath (Fig. 6, B, C, and F) prevented the spread of activity, supporting the argument that NMDA receptor-mediated synaptic transmission in the intracollicular circuits contributes to the build-up of activity. A similar mechanism may be responsible in vivo for the build-up of activity in the large populations of premotor cells that discharge prior to a saccade (Sparks 1986; Sparks et al. 1976).

Fig. 6.

The spatial distribution of field responses evoked by stimulation in the SL. A: the position of microelectrodes on a slice. The boundary between SL and SGI is indicated by a white line. The electrode outlined in white was the stimulating electrode and the remaining 63 electrodes were recording electrodes. The responses from electrodes 1–4 are shown in B. B: field potentials from representative electrodes 1–4 elicited by a biphasic pulse (positive 0.1 ms and negative 0.1 ms, 180 μA) to the stimulating electrode under 4 conditions, control (Cont), 10 μM Bic, 10 μM Bic and 50 μM APV (Bic+APV), and 10 μA Bic, 50 μM APV, and 10 μM CNQX (Bic+APV+CNQX). C: the same records shown within the dotted lines in B but in a faster time sweep and focused on the events shortly after the electrical stimulation. The 64-channel field potentials recorded in the control, Bic (bicuculline), and Bic+APV (2-amino-5-phosphonovalerate) conditions, respectively. Gray curved lines indicate the boundary between SL and SGI and the ventral border of the deep layer. D–F: computed color images of field potentials reconstructed from the 63-channels plus 1 at the stimulation site extrapolated from other channels in the control (D), Bic (E), and Bic+APV (F) conditions, respectively, demonstrate the spatial distribution of responses at each time point. As indicated in the color bar, negative potentials are in red and positive ones in blue. White lines indicate the dorsal boundary of the SC, the boundary between the SL and SGI, and the ventral border of the SC. The locations of stimulating electrodes are shown by the white squares (Phongphanphanee et al. 2008a).

More recent in vitro experiments by Kaneda et al. (2008b) also suggest that, in addition to the burst generation mechanism in the SGI, a NMDA receptor-dependent amplification mechanism is implemented in the deeper portion of the SL. These experiments showed that GABAB receptor-mediated inhibition in SL influences the duration of bursts in SGI by controlling the amplitude of visual responses in SL. In their experiments, block of GABAB receptors by the application of the antagonist CGP52432 prolonged the duration of the EPSC bursts in SL. These results suggest that, in the absence of CGP52432, GABA activates GABAB receptors in SL to reduce burst duration.

Role of pathways from SGI to SL in saccadic suppression

As the eyes move during a shift in the direction of gaze, the optical projection of the visual field sweeps across the retina at the same speed in the opposite direction, but the resulting retinal stimulation does not produce either the perception of visual motion during the eye movement or the triggering of subsequent, unwanted shifts in gaze (Volkmann 1968; Volkman 1962). The mechanisms responsible for this “saccadic suppression” have been a topic of study since the time of Sherrington (1918), particularly because it is an example of the general problem of how an organism distinguishes between self- and environmentally induced stimulation. Previous experiments designed to identify the mechanisms responsible for saccadic suppression in primates suggested that neither retinal insensitivity to the rapid saccade induced movement of the visual field nor inhibitory proprioceptive feedback from the extra-ocular muscles or tendons plays an essential role. For example, extracellular recordings in SL demonstrate that cells can respond vigorously to visual stimuli moving at the velocity of saccades (Goldberg and Wurtz 1972). Moreover, the observations that the suppression of activity in SL can even begin before the onset of the saccade and can occur during saccades in the dark indicate that saccadic suppression is not dependent on reductions in the level of retinal activity during the saccade (Goldberg and Wurtz 1972). Similarly, the proposal that the suppression is mediated by inhibitory input from receptors in extraocular muscles or tendons that detect the eye movements not only is contradicted by the experiments demonstrating that the suppression begins before the onset of the saccade but also by studies that have shown that it occurs when the eyes are paralyzed by retrobulbar injections of lidocaine HCl (Richmond and Wurtz 1980; Volkman et al. 1968).

An alternative hypothesis proposes that premotor neurons provide inhibitory corollary discharges that function to suppress the visual activity evoked during the eye movement. Our in vitro experiments provided both anatomical and physiological support for this hypothesis and suggested further that the circuitry resides within the superior colliculus (Lee et al. 2007). Specifically, the results of two types of experiments provide evidence that a GABAA receptor-mediated inhibitory pathway to SL arises in SGI. First, in GAD 67-GFP knock-in mice, in which GABAergic neurons express green fluorescent protein (GFP) (Tamamaki et al. 2003), the GFP-labeled SGI neurons were injected with the axonal tracer biocytin. The results revealed a population of GABAergic neurons in SGI that gives rise to a prominent projection to SL (Fig. 7A). Second, whole cell patch-clamp recordings from both rats and mice detected large inhibitory currents in SL cells evoked by photostimulation in SGI (Fig. 7B). This inhibitory pathway is monosynaptic and is blocked by GABAzine, confirming that the inhibition is generated by GABAA receptor-mediated inhibitory PSCs (IPSCs). While the projections to SL from individual GABAergic SGI cells have a crude columnar organization, the large population of premotor cells that are active prior to a saccade (Sparks et al. 1976) would activate large numbers of the GABAergic cells and therefore be expected to globally suppress activity throughout the visual field.

Fig. 7.

Mechanism for saccade suppression. A: drawing of a GAD 67-GFP-positive GABAergic neuron that has its soma and dendrites in SGI (black) and projects to SGS (red). B, bottom right: outward inhibitory postsynaptic currents (IPSCs) evoked in a SGS cell (black dot) by photostimulation at sites 1-5 (top). The cell was clamped at −30 mV to increase the amplitude of the IPSCs. Bottom left: camera lucida drawing of the narrow field SGS cell. C: circuit model proposing that recurrent collaterals from SGI premotor cells contact SGI GABAergic cells that project to projection cells in SGS (Lee et al. 2007).

These results provided the rationale for a model of a circuit mechanism responsible for saccadic suppression that is located in the superior colliculus (Fig. 7C). The model proposes that the recurrent axon collaterals from premotor cells activate neighboring GABAergic cells that in turn inhibit the visuosensory cells in SL. Direct evidence for premotor input to the GABAergic cells was provided by more recent experiments that showed that antidromic stimulation of premotor cell axons that have left the colliculus in the predorsal bundle evokes EPSCs in SGI GABAergic cells. These EPSCs can be blocked by the glutamate receptor blocker 6-cyano-7-nitroquinoxalene-2,3-dione (CNQX), showing that they are synaptically mediated by the SGI premotor cells (Phongphanphanee et al. 2009). IPSCs evoked in SL by the same antidromic stimulation were blocked by the GABAA receptor blocker GABAzine, confirming that the premotor cells in SGI excite GABAergic cells that inhibit SL cells. According to the model, the attenuation of activity in the SL neurons reduces the excitatory input they provide to the premotor cells in SGI during the saccade and thus decreases the likelihood that eye-movement-induced retinal activity will trigger additional unwanted saccades. The input to the GABAergic cells from the premotor cells might also help ensure that the duration of the saccadic suppression approximates the duration of the visual stimulation that occurs during the saccade. SL also projects to the lateral geniculate and lateral posterior nuclei in the dorsal thalamus, both of which relay visuosensory information from the retina and/or superior colliculus to visual areas of the cortex (May 2006). The model proposes that inhibition of the colliculothalamic pathway contributes to the suppressed perception of the visual field movement that occurs during saccades (Judge et al. 1980; Volkmann 1962) by reducing the level of activity in these thalamocortical relays to visual cortex (Noda 1975).

Role of horizontal connections within SGI in saccade selection

Experiments in primates have provided evidence that low-frequency activity generated by the premotor cells in SGI prior to the presaccadic high-frequency bursts plays a role in selecting the direction of an ensuing saccade by influencing which premotor neurons generate the high-frequency bursts (Glimcher and Sparks 1992). Some models of the mechanism responsible for this process of saccade selection hypothesize that local excitation and long-range inhibition mediate a “winner-take-all” competition among ensembles of SGI cells during which the activity levels of premotor cells in one location of the motor map is built up while those in other locations are suppressed (Arai et al. 1994; Das et al. 1996; Kim and Basso 2008; Van Opstal and Van Gisbergen 1989). One model for this competition mechanism proposes that while the amplification of SGI premotor responses by local recurrent excitation facilitates the transition from low-frequency activity to high-frequency presaccadic bursts at one location, the suppression of command signals for incompatible saccades in other areas of the motor map is mediated by wide-ranging inhibitory connections of local interneurons that also are activated by the recurrent excitation. Lee and Hall (2006) combined photostimulation and whole cell patch-clamp recording to test this model in vitro. Uncaging glutamate (spot diameter, 10–50 μm) in SGI evoked clusters of brief synaptic currents caused by activation of neurons presynaptic to the recorded cell (Fig. 8, A and B). For a series of photostimulation sites extending horizontally from the recorded cells along both the rostrocaudal and mediolateral axes of the SGI, the evoked responses included both excitatory and inhibitory synaptic currents. Robust EPSCs were evoked consistently within 300 μm of the soma. Only small and rare EPSCs could be evoked from sites between 300 and 500 μm, and none were detected from still more distant horizontal sites. These measurements are consistent with the dimensions of the dendritic fields and the local axonal terminal fields of SGI cells in both rodents and primates (Fig. 4) (Hall and Lee 1997; Helms et al. 2004; Moschovakis et al. 1988) and further support the proposal that the recurrent axonal collaterals that arise from cells in this layer contact the dendrites of neighboring cells to amplify the responses and to impose the dimensions of the surround excitatory field. Evoked IPSCs were maximal within 100–500 μm of the recorded cell and decreased rapidly to baseline within 1,000 μm (Fig. 8C). Although the baseline in Fig. 8C indicates that spontaneous IPSCs were occasionally recorded during stimulation beyond 500 μm, evoked responses were rare. Similar estimates of the extent of the inhibitory surround were obtained anatomically by injecting the tracer biocytin into GFP-positive neurons in SGI of the mouse; most of the labeled GABAergic axons branched and terminated in dense arbors within 500 μm of the cell soma. Some of these neurons also possessed axon branches that extended horizontally as far as 1,000 μm from the cell, but, as predicted by the photostimulation experiments, the number of these branches and their terminals was small (Sooksawate et al. 2005).

Fig. 8.

Horizontal inhibition in SGI. A: photograph of a slice showing the location of a patch-clamped SGI cell (●) and photostimulation sites 1-7. B: strong IPSCs were evoked in the cell by stimulation at sites 1-3 but the responses decreased sharply beyond 300 μm (Vh = −20 mV). C: this summary of the results from multiple cells shows that the number of IPSCs decreases sharply over the 1st 300 μm from the cell and reaches baseline at around 500 μm (Lee and Hall 2006).

Because photostimulation was used, the measurements made by Lee and Hall (2006) estimate the dimensions of the excitatory and inhibitory surrounds mediated by the monosynaptic connections of SGI cells, but they do not address whether long range inhibitory influences could be mediated by multisynaptic pathways. To address this question, a recent in vitro study from Isa's laboratory measured the extent of multisynaptic excitation and inhibition within both SL and SGI (Phongphanphanee et al. 2008b). In this study, horizontal slices were made that preserved the architecture of the local circuits in either SL or SGI. Whole cell recordings were made of responses that were evoked by electrical stimulation at various sites in the slices. Multisynaptically mediated EPSCs and IPSCs were isolated from each other by clamping the membrane voltage of the recorded cells at the reversal potentials of IPSCs or EPSCs. The results showed that multisynaptic excitation and inhibition could be evoked by repetitive stimulus pulses applied several hundred micrometers away from the SGI cells but that the extent of the inhibition was equal to or smaller than that of the excitation and, as a net effect, excitation was the dominating influence. The spatial extent of excitation in these in vitro experiments might be explained by in vivo experiments in the cat reported by McIlwain (1982), who found that electrical stimulation within SGI could evoke action potentials in cells within 1.5 mm of the stimulation site. This extent is consistent with the idea that wide-ranging inhibition in this layer might be mediated by long range excitatory inputs to inhibitory cells with more local projections, but McIlwain (1982) also pointed out that the spread of excitation could be mediated by multisynaptic pathways that link widespread areas of SGI or by activation of axons that originated outside of the superior colliculus. In contrast, “Mexican hat” local excitation and long-range inhibition was observed in the SL slices (Phongphanphanee et al. 2008b). In conclusion, these results indicate that an organization of local circuitry consistent with a winner-take-all mechanism for saccade selection is more evident in SL than in SGI. In SGI, the spatial properties of the inhibition suggest that it contributes to local spatiotemporal response properties of the cells. Thus models that argue that wide-ranging competitive interactions in SGI could mediate a winner-take-all mechanism (Trappenberg et al. 2001) need to be expanded to include the contributions of circuitry within SL.

Modulation of collicular circuitry by extrinsic inputs


In some instances, the in vitro slice preparation permits analyses of the contributions of afferent pathways that arise outside of the superior colliculus. For example, in a recent in vitro study from Isa's laboratory (Kaneda et al. 2008a), the pathway from SNr, which is located in the base of the midbrain, to the superior colliculus, which is located at the top, was included within an individual slice (Fig. 9, A–C). When the SNr was stimulated while recordings were made from SGI cells, IPSCs were recorded not only in the non-GABAergic premotor cells but also in GFP-labeled GABAergic interneurons (Fig. 9, D–F). These results indicated that the SNr inhibits GABAergic interneurons in the SGI as well as excitatory projection neurons, which is consistent with a recent report that electrical stimulation of the SNr in primates sometimes shortens the latencies of contraversive saccades (Basso and Liu 2007). Therefore it has been suggested that the SNr input plays a role in determining the spatiotemporal pattern of SGI local circuit activity in addition to directly modulating the level of excitation of the premotor neurons (Hikosaka and Wurtz 1985a,b).

Fig. 9.

Characteristics of nigral inhibition in SGI GABAergic neurons. A: schematic of configuration for stimulating and recording electrodes. Whole cell recordings were made from GABAergic cells in the SGI. Five cathodal concentric bipolar electrodes with a tip distance of 300 μm were placed on the SNr. B: evoked IPSCs from stimulation site 1 in A were recorded at different holding potentials. C: the IPSCs evoked by stimulation in SNr were reversibly blocked by application of the GABAA receptor antagonist gabazine (10 μM) (Kaneda et al. 2008b).


The parabrachial nucleus gives rise to a very prominent cholinergic input to SGI in diverse species (Beninato and Spencer 1986; Hall et al. 1989; Woolf and Butcher 1986). This same region of the pons has widespread projections to other brain regions, including the thalamus (Fitzpatrick et al. 1989), and is believed to play an important role in regulating levels of alertness and attention (Billet et al. 1999; Lu et al. 1993; McCormick 1989; Steriade and Llinàs 1988). In vitro studies by Isa and his group (Li et al. 2004; Sooksawate and Isa 2006; Sooksawate et al. 2008) demonstrated the cholinergic input to SGI suppresses GABAergic synaptic transmission at presynaptic terminals and generates inward nicotinic and muscarinic currents that depolarize the premotor neurons. Both of these mechanisms would enhance the excitability of SGI neurons. In support of these results, administration of nicotine depolarized the SGI neurons (Fig. 10A), and when electrical stimulation of the optic tract (OT) was combined with nicotine application in slice preparations, subthreshold excitatory synaptic responses in SGI neurons were converted to bursts of action potentials. These results suggested that depolarization of SGI neurons caused by nicotinic receptor activation in these cells can gate the signal transmission from the SL to the SGI (Fig. 10, B—D, from Isa and Sparks 2006). Behavioral support for this facilitation was provided by studies in awake primates, which showed that injections of nicotine into the superior colliculus dramatically reduce saccade reaction times from regular saccade latencies between 150 and 200 ms to express saccade latencies in the range of 100 ms (Fig. 11, A and B) (Aizawa et al. 1999; Watanabe et al. 2005). Thus methods as disparate as in vitro physiology in rodents and behavioral studies in awake primates suggest that the parabrachial region may influence saccade reaction times as a function of level of alertness or attention toward a particular region of the visual field (Isa 2002; Isa and Sparks 2006).

Fig. 10.

Cholinergic influences in the SGI. A: depolarization induced in a SGI neuron by application of 30 μM nicotine. B: experimental design of the experiment. Whole cell recordings were made from a SGI neuron and the effect of electrical stimulation to the optic tract (OT) was measured. C: effect of nicotine application (30 μM) on excitatory postsynaptic potentials (EPSPs) induced by stimulation of SGS. 1: control responses. 2: responses during nicotine application. 3: responses after wash out of nicotine. D: time course of the effect of nicotine application (Isa and Sparks 2006).

Fig. 11.

Effects of nicotine injection (10 mM, 2 μl) into the monkey superior colliculus on saccade reaction time (SRT) during the GAP (170 ms) visually guided saccade paradigm. A horizontal bar indicates the period of nicotine injection (10 min). A: SRTs are plotted against time. ○, trials before injection; *, trials during and after nicotine injection. B: distribution of saccadic reaction times before (lower) and during and after the injection of nicotin (Aizawa et al. 1999).


The cholinergic input to the superior colliculus from the parabigeminal nucleus (Pbg) terminates primarily in SL (Baizer et al. 1991; Graybiel 1978; Jiang et al. 1996; Sherk 1979). In vivo experiments in the rat by Binns and Salt (1997, 2000) showed that the nicotinic cholinergic receptor agonist lobeline depresses visually evoked activity in SL. This effect was unaffected by the GABAA antagonist bicuculline but was blocked by the addition of the GABAB antagonist CGP 35348. Because nicotinic receptors are found on retinal terminals in SL (King and Schmidt 1991; Prusky and Cynader 1988), Binns and Salt (1997, 2000) proposed that the cholinergic input to this layer from the parabigeminal nucleus facilitates the release of glutamate from retinal terminals. They argued further that glutamate released from these retinal terminals excites GABAergic neurons that form GABAB receptor synapses with SL cells. The GABAB receptor-mediated inhibition in turn reduces the magnitude of the visually evoked responses. Similar conclusions were reached in in vitro patch-clamp experiments by recording the responses of SL projection cells to optic tract stimulation before and during the puff-application of acetylcholine (ACh) (Lee et al. 2001). These experiments showed that ACh reduced the amplitude of EPSCs evoked in neurons that were identified as projection cells by prelabeling them with a retrograde axonal tracer placed in the visual relays of the dorsal thalamus, the lateral geniculate, and lateral posterior nuclei. As in the in vivo experiments by Binns and Salt, the reduction of the evoked EPSCs was blocked by the addition to the slice bath of the GABAB receptor blocker CGP 52432 but not by the addition of the GABAA blocker bicuculline (Fig. 12A). More recently, Endo et al. (2005) also studied the role of the cholinergic input to SL, by in vitro measurement of the effects of a brief air pressure application of ACh on SL cells. They found that ACh elicited nicotinic inward currents both in GABAergic cells labeled by GFP, and in unlabeled, presumably non-GABAergic cells (Fig. 12B). They also confirmed the evidence from the previous two studies demonstrating that ACh acts presynaptically on retinal terminals and produces an inhibitory influence on SL projection cells. However, in contrast to the previous studies, they also were able to demonstrate a GABAA receptor-mediated component to this presynaptically induced ACh inhibition (Fig. 11B). One possible reason for this apparent discrepancy may be that, in the earlier experiments, stimulating the retinal input activated multiple sources of GABAA input to SL and that the summed effects of blocking these different sources may have masked the specific effects of bicuculline on the ACh modulation of visual responses.

Fig. 12.

A: the contribution of GABAB receptors to acetlycholine (ACh)- induced reductions in the EPSCs evoked by stimulation of the optic tract. The control trace was obtained before the puff-application of ACh, the ACh-labeled trace was obtain after a 20-s application of ACh and the recovery trace 2 min after ACh application. A1: the application of ACh reduces the amplitude of the evoked EPSC. A2: the addition of the GABAB receptor blocker CGP 52432 blocks the effect of ACh, indicating that the ACh-induced reduction in the evoked response is mediated by GABAB receptors (Lee et al. 2001). B: the outward current responses to puff-applied ACh in the superficial layer neurons are dependent on nAChRs. The outward current response was preceded by an inward current response in a GFP-negative neuron. Bicuculline (Bic, 10 μM) inhibited both responses (Endo et al. 2005).

This cholinergic pathway from the parabigeminal nucleus to SL apparently has a homologue in non-mammals called nucleus isthmi (Wang et al. 2000). Nucleus isthmi has been implicated in a winner-take-all mechanism that contributes to target selection in the optic tectum (Wang et al. 2000). While like nucleus isthmi and the optic tectum, the parabigeminal nucleus has reciprocal, retinotopically organized connections with the superior colliculus (Jiang et al. 1996; Sherk 1979), additional experiments are needed to determine whether they share similar functions and mechanisms.

Concluding remarks

In 1937, the great Spanish neurobiologist Ramón y Cajal argued that “The great enigma in the organization of the brain was the way in which the nervous ramifications ended and in which neurons were mutually connected.” The circuits formed by the neurons still comprise a “great unraveled knot” and, like Gordian's knot, it is difficult to unravel. Although in vitro experiments cannot cut the knot with a single slice, they do provide an unparalleled opportunity to tease circuits out from the knot with a level of detail not easily attained with traditional anatomical and physiological methods. While limitations to the in vitro approach—including the separation of neurons within the slice from their extrinsic sources of input, the immense diversity in neuronal types and their synaptic connections, and the caution that must be exercised when generalizing results among different species—are daunting, we are convinced that it has the power to significantly deepen our understanding of sensorimotor mechanisms at the cellular level.


The research was supported by National Eye Institute Grant EY-08233 to W. C. Hall and Human Frontier Science Program, Ministry of Education, Culture, Sports, Science and Technology of Japan Grants 13854029, 18019007, and 18200027 to T. Isa.


We thank P. Lee for help with the illustrations and comments on the text.


  • 1 In the text, the label SL includes both the optic layer, stratum opticum (SO), and the superficial gray layer, stratum griseum superficiale (SGS). Although SO is classified as a fiber layer, it contains populations of cells that share many properties with cells in SGS. In the figures, where useful, the borders between these layers are indicated.


View Abstract