A collicular visual cortex: Neocortical space for an ancient midbrain visual structure

See allHide authors and affiliations

Science  04 Jan 2019:
Vol. 363, Issue 6422, pp. 64-69
DOI: 10.1126/science.aau7052

Another primary visual cortex

Most functional studies in the visual system have focused on the cortical representation of the geniculo-striate pathway that links the retina to the cortex. The parallel collicular pathway is believed to sparsely project throughout the visual cortex and have a modulatory role on cortical responses to visual stimuli. Beltramo and Scanziani found a visual cortical area that is entirely dedicated to the superior colliculus. This area can discriminate moving visual stimuli that the “classical” primary visual cortex cannot. Thus, the superior colliculus, a phylogenetically ancient structure, has its own projection in neocortex that provides this area with exquisite feature-detection abilities not found in the classical primary visual cortex.

Science, this issue p. 64


Visual responses in the cerebral cortex are believed to rely on the geniculate input to the primary visual cortex (V1). Indeed, V1 lesions substantially reduce visual responses throughout the cortex. Visual information enters the cortex also through the superior colliculus (SC), but the function of this input on visual responses in the cortex is less clear. SC lesions affect cortical visual responses less than V1 lesions, and no visual cortical area appears to entirely rely on SC inputs. We show that visual responses in a mouse lateral visual cortical area called the postrhinal cortex are independent of V1 and are abolished upon silencing of the SC. This area outperforms V1 in discriminating moving objects. We thus identify a collicular primary visual cortex that is independent of the geniculo-cortical pathway and is capable of motion discrimination.

The mammalian cerebral cortex receives sensory information from several modalities. Even within the same modality, sensory information reaches the cortex via anatomically distinct parallel pathways. The relative roles of these distinct pathways in driving cortical responses to a sensory stimulus and the extent to which their sensory representations are spatially segregated in the cortex are still matters of debate (1).

In the visual system, dorsolateral geniculate nucleus (dLGN) innervation of the primary visual cortex (V1) is considered the primary entry point of retinal input to the cortex (2). V1 lesions abolish or strongly reduce visually evoked activity in several higher cortical visual areas (36). The colliculo-cortical pathway provides visual input to the cortex from the superior colliculus (SC) via the pulvinar nucleus of the thalamus (710). Its inactivation has either no or only a slight and feature-selective effect on cortical visual responses (9, 1113). Thus, despite a clear anatomical link from SC to visual cortex, no cortical area has been identified yet whose visual responses rely entirely on visual activity originating from the SC.

Visual responses in the mouse cortical area postrhinal cortex (POR) are well documented (1416). Although generally assumed to rely on V1, their dependence on V1 has not been directly assessed. We determined the impact of V1 on POR’s visual responses in awake, head-fixed mice. We optogenetically silenced V1 while performing simultaneous electrophysiological recordings from V1 and POR, ensuring that the receptive fields of the respective recording sites overlapped (Fig. 1 and fig. S1A). Visual areas in mice are anatomically defined by their retinotopic afferent input originating from V1 (14, 16). We thus injected the anterograde viral tracer AAV1.CAG.TdTomato in the posterior portion of V1 and identified POR via transcranial epifluorescence illumination of the labeled V1 axons projecting to the visual areas surrounding V1 (Fig. 1A and fig. S2). Drifting gratings displayed on a monitor at the center of the receptive field triggered responses in 40% (81 of 209) of the units isolated from POR [average firing rate ± SEM of visually evoked responses for regular-spiking (RS), putative excitatory neurons: 1.61 ± 0.25 Hz, n = 41; for fast-spiking (FS), putative inhibitory interneurons: 3.14 ± 0.43 Hz, n = 40; 5 mice]. We silenced V1 by photoactivating cortical inhibitory interneurons expressing channelrhodopsin-2 (ChR2). This approach abolished visual responses in RS neurons across the entire cortical depth (fig. S3) and over large areas of V1 (fig. S4) (17). Despite this extensive silencing of V1, however, most of the visual response in POR persisted (Fig. 1, C and D) [21.65 ± 6.51% average decrease ± SEM in visually evoked firing rate of RS cells (P = 0.022, n = 41); 34.5 ± 5.03% of FS cells (P < 0.0001, n = 40, 5 mice; Wilcoxon signed-rank test)]. Whereas the response latencies in V1 and POR were quite similar (fig. S5), the time courses of the peristimulus time histogram (PSTH) in V1 and POR were markedly different (Fig. 1, C and D, and fig. S6).

Fig. 1 Visual responses in POR are not driven by the geniculate-V1 pathway.

(A) Injection of the anterograde tracer AAV1.TdTomato in V1 enabled the visualization of higher visual areas. Delineated cortical areas are as shown in fig. S2. The weaker fluorescence of POR is consistent with its weaker V1 input (31). (B) Drifting gratings (diameter, 20°) were presented to awake mice conditionally expressing ChR2 in V1 inhibitory neurons. Recordings were simultaneously performed in POR and in V1. A light-emitting diode (LED)–coupled optic fiber was used to silence V1. (C) Example experiment. (Left) V1 recordings; (right) POR recordings. (Top) Heatmaps of the receptive fields of multiunit activity. Dotted circle shows position of the visual stimulus. (Bottom) Raster plots and PSTH of RS units under control conditions (black) and during V1 silencing (blue). Black horizontal bar, duration of stimulus presentation; blue horizontal bar, period of V1 silencing. (D) Summary of all experiments as described for (C) (33 non-GABAergic RS units in V1 and 41 RS units in POR from five animals). (Top) Summary average PSTHs. The PSTH values for individual units were normalized by their maximum value and then averaged together. (Bottom) Scatter plots reporting the responses of all units measured during the stimulus presentation period (0.9 s). Green data points, example units in (C). (E) Injection of the retrograde tracer CTB in POR. (F) (Left) TdTomato injection site in V1 with anterograde projection to POR (white) and CTB injection in POR (red). (Middle) CTB+ neurons in dLGN and pulvinar nuclei. (Right) Higher-magnification image of the region delineated by the square in the middle image. (G) Distribution of retrogradely labeled dLGN and pulvinar neurons along the thalamic rostro-caudal axis (ratio of CTB+ cells in pulvinar or in dLGN to the total CTB+ cells counted in the two nuclei; 45 coronal sections, three mice).

Which structure other than V1 could relay visual input to POR? The dLGN is also a source of afferent input to other visual areas (16). To determine whether POR directly receives input from the dLGN, we injected cholera toxin subunit B (CTB) in POR (Fig. 1E). The dLGN was almost devoid of retrogradely labeled neurons. The vast majority (>99%) of retrogradely labeled cell bodies in the visual thalamus were found in the pulvinar (18) [also called the latero-posterior nucleus in rodents (10)].

The pulvinar receives a massive afferent input from V1 (1, 9), and its response to visual stimuli depends on V1 (1, 19). Because silencing of V1 has a minor effect on POR activity, the pulvinar might seem a poor candidate for relaying visual activity to POR. However, the pulvinar is also a key node of the colliculo-cortical pathway because it receives direct input from SC (10, 20). If there were pulvinar neurons whose visually evoked activity was driven by SC and unaffected by V1 silencing, then such neurons could mediate responses in POR that are independent of V1. We first determined whether there are neurons in the pulvinar that are visually driven by SC (Fig. 2). Injections in V1 and SC with viral anterograde tracers revealed that axons originating from SC preferentially target the caudal pulvinar, whereas axons originating from V1 preferentially target the rostral pulvinar (Fig. 2A). To determine whether V1 and SC inputs are also separated functionally, we recorded responses from the caudal or rostral pulvinar while optogenetically silencing V1 or SC (Fig. 2, B to D). We presented dark moving dots, visual stimuli known to drive robust activity in the SC (21). Photoinhibition of SC strongly attenuated its responses to the visual stimuli, particularly in the stratum opticum (Fig. 2B) (79.34 ± 4.31% average decrease ± SEM in visually evoked firing rate; P < 0.0001, Wilcoxon signed-rank test, n = 33, 5 mice) (see also fig. S7). SC silencing similarly attenuated the visually evoked responses recorded simultaneously in the caudal pulvinar (Fig. 2B) (80.02 ± 5.96% average decrease ± SEM in visually evoked firing rate; P < 0.0001, Wilcoxon signed-rank test, n = 34, 5 mice). Silencing of V1 had little effect on visually evoked activity in the caudal pulvinar (Fig. 2C) (24.64 ± 5.33% average decrease ± SEM in visually evoked firing rate; P = 0.0004, Wilcoxon signed-rank test, n = 63, 7 mice) while strongly reducing visual responses in the rostral pulvinar (Fig. 2D) (91.43 ± 4.82% average decrease ± SEM in visually evoked firing rate; P = 0.0005, Wilcoxon signed-rank test, n = 16, 3 mice).

Fig. 2 Anatomical and functional segregation of cortical and collicular inputs in the pulvinar.

(A) (Top) Coronal sections showing the injection sites of AAV1.CAG.TdTomato in V1 and AAV1.CAG.GFP in SC (left) and of V1 and SC projections to dLGN and pulvinar (Pulv) for three representative sections along the rostro-caudal axis. (Bottom left) Normalized fluorescent density (average ± SEM) along rostro-caudal axis of the pulvinar (green, SC input; magenta, V1 input; 54 coronal sections, three mice). Fluorescence density values were normalized to those for the coronal section displaying maximal tdTomato or GFP expression. (Bottom right) Schematic illustration of the results. (B) (Top row) The stimulus was a dark dot moving along a straight trajectory. Recordings were performed simultaneously from stratum opticum of SC (with an optrode for silencing) and caudal pulvinar. (Second row) (Left) PSTH of an SC unit. Black, control; blue, SC silencing; black horizontal bar, period of stimulus presentation; blue horizontal bar, period of SC silencing. (Right) PSTH of a simultaneously recorded unit in the caudal pulvinar. (Third row) Average PSTH for 33 (left) and 34 (right) isolated units in SC and in caudal pulvinar (five animals). (Bottom row) Scatter plots of firing rates averaged over a 450-ms interval (i.e., the response window; see materials and methods for details on the moving dot analysis) during the period of visual stimulation. Green data points, example units shown in the second row. (C) (Top row) V1 silencing and simultaneous recordings in caudal pulvinar. (Second row) PSTH of a unit in caudal pulvinar. Black, control; blue, V1 silencing. (Third row) Average PSTH for 63 isolated units (seven animals). (Bottom row) Scatter plot as in (B). (D) Similar experiment as shown in (C) but recording from rostral pulvinar (16 units; three animals).

Could the SC be disynaptically connected to POR through the pulvinar? We used anterograde trans-synaptic tracing in which transfection of virus harboring Cre recombinase in the presynaptic neuronal population leads to the conditional expression of a reporter gene in the postsynaptic neuronal population (22) (Fig. 3, A and B, and fig. S8). We injected this virus in the SC and a virus containing conditional green fluorescent protein (GFP) in the caudal pulvinar. Histological analyses revealed cell bodies expressing GFP in the caudal pulvinar and axonal arborizations densely innervating layers 4 and 1 in POR as well as other cortical visual areas, albeit with the exception of the laterointermediate area (LI), to a much lesser extent than POR (Fig. 3, A and B, and fig. S8), consistent with recent observations (20).

Fig. 3 POR is driven by the colliculo-cortical pathway.

(A) Anterograde trans-synaptic tracing. AAV1.hSyn.Cre was injected in SC and AAV8.CAG.Flex.GFP was injected in caudal pulvinar. AAV1.CAG.TdTomato injected in V1 was used to identify higher visual areas. Delineated cortical areas are as shown in fig. S8. (B) (Left) Double labeling of axons targeting POR. Red, V1 afferents (for POR localization); green, caudal pulvinar afferents. (Inset) GFP expression in the caudal pulvinar. (Middle) Magnification of the region marked by the rectangle in the left panel (GFP channel) and summary distribution of GFP fluorescence (caudal pulvinar afferents) along POR cortical depth (five animals). (Right) Fluorescence density distribution of trans-synaptically labeled caudal pulvinar axons across cortical areas. Green bars, normalized averages and SEM (five animals) across 10 distinct cortical areas. For each animal, the fluorescence density of each area was normalized to that for POR. Each symbol corresponds to a different animal. Black triangle, experiment on the left. (C) SC silencing and POR recording. The visual stimulus was a dark dot moving at a speed of 30° per second along the two cardinal and two oblique axes for 0.67 or 3 s (i.e., the time required to cover, at 30° per second, trajectories of 20° or 90° of visual space, respectively). (D) (Left) Raster plot and PSTH of a POR unit under control conditions (black) and during SC silencing (blue). (Middle) Aligned average PSTH for 96 isolated units (five animals). The average PSTH was generated by aligning the response of each isolated POR unit to the center of the “SC response interval” for the same stimulus (time 0 s) (see materials and methods for a description of the moving dot analysis; see also fig. S9). (Right) Scatter plot of firing rates of POR units, averaged over a 450-ms window (i.e., the response window; see materials and methods description of the moving dot analysis) within the SC response interval. Green data point, example on the left. (E and F) Results of experiments similar to those shown in (C) and (D) but with a circular drifting grating patch as the stimulus (diameter: 20°; 44 isolated units; two animals). The average PSTH was aligned to the visual stimulus onset. Scatter plot, neuronal responses measured during stimulation period (0.9 s), as in Fig. 1D. (G) SC silencing and POR recording. The visual stimulus was a dark dot moving at 30° per second along a straight trajectory of 20° of visual space (stimulus duration, 0.67s). The superficial layers of SC were silenced via local application of TTX. (H) (Left) Raster plot and PSTH of an isolated POR unit under control conditions (black) and ~20 min after the beginning of TTX infusion in SC (red). The gray portion of the raster plot indicates the initial ~20 min of TTX infusion. (Middle) Average PSTH, aligned to the visual stimulus onset, for 44 isolated units (three animals). (Right) Scatter plot as in (D).

To determine whether visually evoked activity in POR depends on SC, we performed simultaneous recordings from POR and SC while optogenetically silencing SC (Fig. 3, C to F). We presented dark moving dots and ensured that the receptive fields at the recording sites in POR and in the superficial layers of SC overlapped (see materials and methods and figs. S1B and S9). SC silencing almost completely abolished the response of POR to the dots (Fig. 3D) (93.77 ± 1.95% average decrease ± SEM in visually evoked firing rate; P < 0.0001, Wilcoxon signed-rank test, n = 96, 5 mice), across cortical depths in a homogeneous manner (fig. S10). The reduction of POR responses to dots was particularly pronounced for the portion of the POR receptive field that overlapped with the receptive field silenced in SC. That was evident for dot trajectories covering a large fraction of visual space, where SC silencing created a “scotoma” in the response of POR neurons (fig. S9). To directly compare the effect of SC silencing with that of V1 silencing, we also presented drifting gratings (Fig. 3, E and F) identical to those used to measure V1 silencing (Fig. 1, C and D). Optogenetic silencing of SC suppressed the response of POR to drifting gratings (Fig. 3F), highlighting the much stronger impact of SC than V1 on visually evoked activity in POR (79.83 ± 4.46% average decrease ± SEM in visually evoked firing rate; P < 0.0001, Wilcoxon signed-rank test, n = 44, 2 mice).

Because SC sends sparse inhibitory GABAergic projections to the dLGN (21), our optogenetic activation of GABAergic projection neurons in SC might suppress dLGN neurons. The dLGN projects to other visual areas in addition to V1 (16) (although not POR) (Fig. 1E). If visual responses in POR depended on a geniculate input relayed via other visual areas, rather than on a collicular input, optogenetic activation of GABAergic SC neurons could reduce putative geniculate-mediated visual responses in POR. We therefore blocked neuronal activity in SC with tetrodotoxin (TTX) injected in the stratum opticum of SC (Fig. 3, G and H, and fig. S11). Again, we ensured that the receptive field of the TTX injection site matched the receptive field of the recording site in POR. TTX application abolished visually evoked responses in POR to both moving dots (Fig. 3H) (94.46 ± 3.18% average decrease ± SEM in visually evoked firing rate; P < 0.0001, n = 44, 3 mice) and drifting gratings (fig. S11) (92.46 ± 3.79% average decrease in visually evoked firing rate; P < 0.0001, Wilcoxon signed-rank test, n = 39, 3 mice). Baseline activity in POR was unaffected, indicating that the suppression of visual responses was not due to a direct action of TTX on POR.

Compared with V1, does POR capture distinct properties of the visual world? The ability to respond to small objects moving in the field of view is a characteristic property of the SC (21). We compared the response of V1 and POR to small moving objects. The response of a neuron to a dot moving on a monitor may report different aspects of the stimulus. In one case, the neuronal response may simply report local changes in luminance within the neuron’s receptive field that occur as the dot moves along its trajectory. If so, the exact sequence of the changes in luminance along the trajectory may not be relevant for eliciting a response. In the other case, the neuron may selectively respond to the motion of the dot, i.e., to changes in luminance occurring sequentially at adjacent spatial positions. We therefore presented two different stimuli: a “moving dot” and a “random dot” (Fig. 4A). The moving dot stimulus consisted of a small dot shifting position incrementally along a linear trajectory through the receptive field of the neuron (the same type of stimulus as used in experiments described above). For the random dot stimulus, the dot positions along the same trajectory were randomized (Fig. 4A). Except for the temporal order of the dot positions, the two stimuli were identical. V1 neurons responded almost equally well to moving and to random dots (Fig. 4B) [25.19% (67 of 266) of the units isolated from V1 responded to moving and/or random dots; average firing rate ± SEM for moving dot: 1.62 ± 0.19 Hz; for random dot: 1.13 ± 0.16 Hz; n = 67; 5 mice]. For all responsive V1 neurons, the moving dot produced only a 1.4-fold-higher average firing response than the random dot. The response of POR neurons to the same stimuli, however, showed a marked preference for the moving dot over the random dot (Fig. 4C). For all responsive POR neurons, the moving dot produced a 9.2-fold-higher average firing response than the random dot [33.73% (56 of 166) of the units isolated from POR responded to either stimulus; average firing rate ± SEM for moving dot: 2.77 ± 0.5 Hz; for random dot: 0.3 ± 0.05 Hz; P < 0.0001, Wilcoxon signed-rank test, n = 56; 5 mice]. This was not because POR neurons responded much more to moving dots than did V1 neurons, but mainly because POR neurons responded much less to random dots than did V1 neurons. Receiver operator characteristics (ROC) analysis showed that an independent observer is better at discriminating moving from random dots based on the response of POR than the response of V1 neurons (Fig. 4D). Finally, we used moving dots to compare the sizes of receptive fields of POR and V1 neurons (fig. S12). Neurons in POR had significantly larger receptive fields than those in V1 (medians ± interquartile ranges of receptive field area; POR: 735.5 ± 631 square degrees, n = 62, 5 mice; V1: 157 ± 71.25 square degrees, n = 31, 8 mice; P < 0.0001, Wilcoxon rank sum test).

Fig. 4 POR neurons discriminate moving dots from random dots.

(A) (Left) The stimulus was a dark dot moving for 1 s at a speed of 30° per second along a straight trajectory. (Middle) Random dot: The stimulus was made of the same frames as the moving dot but was presented in random order. Each iteration showed a different random sequence. (Right) Recordings were performed in V1 or POR. (B) Raster plot and PSTH of a V1 unit in response to a moving dot (left) or a random dot (middle), and the scatter plot (right) of average firing rates of V1 units during the period of stimulus presentation (1 s) for moving versus random dots (67 units, five animals). Green data point, example unit shown on the left. (C) As in (B) but for POR (56 units, five animals). Note that the shown POR unit did not respond to random dots. (D) (Left) ROC curve for the V1 isolated unit shown in (B). (Inset) Distribution of spikes for moving dots (black bars) and random dots (red bars). (Middle) ROC curve for the POR isolated unit shown in (C). (Right) Comparison of area under ROC curve (average ± SEM) for all responsive POR and V1 units. Black filled data points denote P < 0.001 for difference in spike distributions between responses to moving and random dots (Wilcoxon rank sum test). Gray bars show areas under ROC: for V1, 0.55 ± 0.007, 67 units, five mice; for POR: 0.69 ± 0.015, 56 units, five mice (P < 0.0001, Wilcoxon rank sum test). Green data points, example units shown in (B) and (C).

These results show that a cortical area, POR, is driven by visual information conveyed via the colliculo-cortical pathway rather than by the geniculate-V1 pathway. The weak impact of V1 silencing on POR is consistent with the weak projection of the former onto the latter area (14). The superior ability of POR to discriminate moving objects compared with that of V1 is in agreement with previous descriptions of enhanced motion selectivity in lateral cortical visual areas (23, 24). However, though this selectivity to moving objects was believed to result from the hierarchical cortical processing of geniculate inputs entering V1 (23, 24), our data indicate that this property of POR depends on collicular input and could be directly inherited from SC. Similarly, the expanded representation of the upper visual hemifield reported in POR (24, 25) could also be directly inherited from the equally biased representation in SC (26). Because the colliculo-cortical pathway is believed to be phylogenetically older than the geniculate-V1 pathway (27), it is tempting to regard POR as an ancestral primary visual area.

Previous reports of SC lesions in cats (12) and rodents (9) described only a slight reduction, if any, of visually evoked activity in various visual cortical areas. Furthermore, in rodents, this reduction depends on the features of the stimulus (9). However, none of these recordings was performed in POR. In primates, SC silencing and lesions also have little or no impact on visual response of several visual cortical areas unless V1 has been ablated first (11). Although primate vision may rely less on SC than in other mammals, again, these recordings were not performed in the parahippocampal cortical areas TF and TH, the primate areas most similar to POR (28). On the other hand, the driving rather than modulatory impact of SC on the thalamus reported here is not without precedent. In primates, the SC drives the mediodorsal nucleus of the thalamus (29) to relay corollary oculomotor activity to the frontal eye fields. Primates can also partially recover from the inability to detect visual stimuli caused by V1 lesions, a phenomenon called blindsight and that involves SC (30). Whether some aspect of blindsight depends on the impact of SC on visual cortex remains to be established.

These results show a spatial and functional segregation of the sensory representation of the two main visual pathways to visual cortex, the geniculate-V1 and colliculo-cortical pathways, and define a specialized cortical area whose responses to visual stimuli are driven by the colliculo-cortical pathway.

Supplementary Materials

Materials and Methods

Figs. S1 to S12

References (3141)

References and Notes

Acknowledgments: We thank all the members of the Scanziani lab for discussions about the project and comments on the manuscript; H. Karten for advice during the course of the study; L. Frank, R. Nicoll, and E. Feinberg for critical reading of the manuscript; and M. Mukundan, J. Evora, N. Kim, and Y. Li for technical support. Funding: This project was supported by the European Molecular Biology Organization Long Term Fellowship, the Human Frontier Science Program Long Term Fellowship, and the Howard Hughes Medical Institute. Author contributions: R.B. and M.S. designed the study. R.B. conducted all experiments and analyses. R.B. and M.S. wrote the paper. Competing interests: The authors declare no competing interests. Data and materials availability: All data and analyses necessary to understand and assess the conclusions of the manuscript are presented in the main text and in the supplementary materials.

Stay Connected to Science

Navigate This Article