Input-dependent segregation of visual and somatosensory circuits in the mouse superior colliculus

Abstract

Whereas sensory perception relies on specialized sensory pathways, it is unclear whether these pathways originate as modality-specific circuits. We demonstrated that somatosensory and visual circuits are not by default segregated but require the earliest retinal activity to do so. In the embryo, somatosensory and visual circuits are intermingled in the superior colliculus, leading to cortical multimodal responses to whisker pad stimulation. At birth, these circuits segregate, and responses switch to unimodal. Blocking stage I retinal waves prolongs the multimodal configuration into postnatal life, with the superior colliculus retaining a mixed somato-visual molecular identity and defects arising in the spatial organization of the visual system. Hence, the superior colliculus mediates the timely segregation of sensory modalities in an input-dependent manner, channeling specific sensory cues to their appropriate sensory pathway.

In the mature cerebral cortex, sensory modalities are segregated into specialized areas known as primary sensory cortices. So that cortical areas ultimately respond to a specific sensory stimulus, this segregation is thought to occur during development, and it is first instructed by intrinsic factors and later by sensory experience (1–5). However, this early inception of sensory identity has not been directly demonstrated because there is no functional evidence of the sensory specificity of emerging circuits. It remains to be determined whether sensory identities arise directly as unimodal entities or whether they are initially multimodal and become specified over time. We found that the nascent somatosensory and visual pathways of mice are functionally interconnected, with both cortices responding to tactile stimulation at prenatal stages yet segregating into independent pathways at birth. This segregation happens in the superior colliculus (SC) and depends on retinal input. Blocking stage I retinal waves or removing retinal projections in embryos leads to a failure in the timely developmental segregation of visual circuits in the SC. Consequently, the activation of the visual cortex by somatosensory stimuli extends into postnatal life and is associated with long-term circuit abnormalities.

S1 and V1 emerge functionally intermingled

A functional map of the periphery is present in the mouse somatosensory cortex at perinatal stages, as shown through whisker pad stimulation in a transgenic mouse line in which glutamatergic neurons in the neocortex express GCaMP6f (hereafter CxGCaMP6f) (6). Hence, there appears to be a high level of intramodal functional organization of sensory circuits at these early stages. We used the same approach to test whether developing sensory circuits also show early specificity to stimulus modalities. We found that at embryonic stages, a somatosensory stimulus not only triggered the expected contralateral response in the barrel field of the primary somatosensory cortex (S1) but also, a bilateral response in the presumptive primary visual cortex (V1). This multimodal response was observed in 33% of the mice analyzed at embryonic day 18 (E18) but disappeared by postnatal day 0 (P0) (Fig. 1, A- to C, and movies S1 and S2), suggesting that we were revealing the end of a developmental process.

Fig. 1. Embryonic somatosensory and visual cortices originate as multimodal.

(A) (Left) Experimental design. (Right) Calcium responses in the cortex (pink) elicited by mechanical stimulation of the whisker pad in control mice at E18 and P0. (B) Proportion of the whisker pad stimulations that evoked a contralateral or ipsilateral response in S1 and V1 of control mice at E18, and in V1 from E18 to P6 (E18, n = 9 mice; P0, n = 13 mice; P3 to P6, n = 8 mice. (C) Scheme summarizing the results. (D) (Left) Experimental design. (Right) Calcium responses in the cortex (pink) elicited by mechanical stimulation of the whisker pad in embBE mice at E18 and P0. (E) Proportion of the whisker pad stimulations that evoked a contralateral or ipsilateral response in S1 and V1 of control and embBE mice at P0 (n = 19 control mice, n = 21 embBE mice). (F) Scheme summarizing the results. (G) Proportion of the whisker pad stimulations that evoked a bilateral V1 response in embBE from E18 to P6 (E18, n = 5 mice; P0, n = 21 mice; P3, n = 5 mice; P5 to P6, n = 8 mice). Scale bars, 1000 μm. Bar graphs show the means ± SEM. ns, not significant; *P < 0.05, **P < 0.01, ***P < 0.001.

We next tested whether the disappearance of this cortical multimodal response might be related to the arrival of retinal inputs to central structures, which occurs close to birth (7–9). To test this possibility, we used CxGCaMP6f mice in which their eyes were eliminated bilaterally at E14 (embBE mice) (10). Whereas whisker pad stimulation at P0 triggered only the cortical somatosensory response in control mice, in the embBE mice, this response was consistently multimodal, activating both S1 and V1 in 100% of the cases, even at E18 (Fig. 1, D to F, and movies S3 and S4). This multimodal response extended into the first postnatal days and lasted until P6 (Fig. 1G and fig. S1). Thus, it appears that somatosensory and visual circuits are not segregated by default but require the arrival of retinal input to do so.

We then designed experiments to identify the circuits responsible for this lack of segregation and the ensuing multimodal response in embBE mice. Electrical stimulation of either S1 or V1 in the embBE mice failed to reproduce the bilateral V1 responses to whisker pad stimulation (fig. S2, A to D). Moreover, these responses were still triggered by whisker pad stimulation when the barrel field was inhibited by tetrodotoxin (TTX) (fig. S2, E and F). Hence, these experiments demonstrate that the sensory-modality interconnection in embBE mice is not driven by a cortico-cortical mechanism.

SC circuit reorganization provides sensory-modality specificity

Before reaching the cortex, sensory-modality pathways converge in nearby territories of the thalamus and also in the SC (11). Therefore, the multimodal responses observed might involve direct communication between modalities at either of these structures. The SC is a midbrain multisensory structure that receives visual input in its superficial layers (sSC) and somatosensory input in its deep layers (dSC) (11, 12). Using mice that express GCaMP6f in SC neurons (hereafter SCGCaMP6f), we found that whisker pad stimulations at P0 evoked a bilateral response in the sSC of all embBE mice, whereas no responses were evoked in the sSC of control mice (Fig. 2, A- to C, and movies S5 and S6). To confirm that the cortical multimodal response requires the SC, we inactivated the SC with TTX in embBE triple transgenic mice that expressed GCaMP6f in the cortex and SC (Cx-SCGCaMP6f) and found that the cortical response became unimodal (Fig. 2, D- to F). Similar results were found when the SC was acutely lesioned (fig. S3, A and B). Last, we looked for sensory modality changes in the circuits downstream of the SC and, using dye tracing, demonstrated that the connectivity from the SC to the thalamus was normal in embBE mice (fig. S3, C and D). Similarly, we found no differences in the specificity of the sensory modality of thalamic afferents or efferents in these embBE mice (fig. S4). Therefore, these results demonstrate that the SC mediates the multimodal cortical responses in embBE mice, and suggest that the SC is involved in the developmental segregation of sensory modalities observed in control mice.

Fig. 2. Somatosensory and visual SC circuits remain intermingled in embBE mice.

(A) (Left) Experimental design. (Right) Calcium responses (pink) in the SC elicited by mechanical stimulation of the whisker pad in control and embBE mice at P0. (B) (Left) Quantification of the data shown in (A) (n = 7 control mice, n = 7 embBE mice). (C) Scheme summarizing the data. (D) Cortical and SC calcium responses elicited by mechanical stimulation of the whisker pad in embBE mice at P0 before, during, and 1 hour after bilateral TTX injection in the SC (asterisks). (E) Quantification of the data shown in (D) (n = 6 mice). (F) Scheme summarizing the data. (G) (Left) Experimental design. (Right) Calcium responses elicited by electrical stimulation of the d SC in control and embBE slices at P0. There is spreading to the ipsilateral and contralateral superficial layers of the SC in embBE mice (arrowheads). (H) Proportion of the total area (active fraction) of the sSC and dSC layers activated by a peri-threshold stimulus in the dSC at P0 (n = 8 control mice, n = 5 embBE mice). (I) (Left) Experimental design. The lentivirus was injected into the SC at E14. (Right) Optical coronal sections from three-dimensional (3D) light-sheet images showing the EGFP lentiviral– labeled axons in the sSC of embBE mice at P0 (arrowhead). (J) Quantification of the normalized EGFP expression in contralateral sSC and dSC (n = 8 control mice, n = 4 embBE mice). Scale bars, (A) and (D) 1000 μm ; (G) and (I) 150 μm. Boxplots show the medians with the interquartile range (box) and range (whiskers). The bar graphs show the means ± SEM. ns, not significant; *P < 0.05, **P < 0.01, ***P < 0.001.

We next assessed how the SC channels somatosensory information into the presumptive V1 by testing the connectivity between the dSC (somatosensory) and sSC (visual) layers in embBE mice. At P0, whereas unilateral electrical stimulation of the dSC layers in SCGCaMP6f control slices only elicited responses in the contralateral dSC layer, the responses in the embBE mice also propagated to both the ipsilateral and contralateral sSC layers (Fig. 2, G and H, and movies S7 and S8), an effect that was not observed by stimulating the superficial layers (fig. S5). In addition, these functional results were supported by anatomical tracing. Unilateral injection of lentivirus- expressing enhanced green fluorescent protein (EGFP) into the SC produced widespread labeling of the contralateral dSC and sSC layers in embBE mice, whereas only the contralateral dSC layer was labeled in their control littermates (Fig. 2, I and J). Conversely, when the multimodal response finally switched to unimodal in the embBE mice at P6, we observed a significant reduction, as compared with P0, both in the number of axons in the sSC and in the bilateral V1 or SC activation elicited by whisker pad stimulation (fig. S6).

The data from the embBE suggest that the multimodal embryonic response in control mice could also be mediated by the SC. Using the Cx-SCGCaMP6f, we found that whisker pad stimulation at E18 in control mice elicited a bilateral V1 response in 37% of the cases, which was always accompanied by concomitant bilateral sSC activation. However, activation of the sSC was not detected when the cortical multimodal response switched to unimodal at P0 (Fig. 3, A to C). We then tested whether the switch from multimodal to unimodal cortical responses involved a developmental reconfiguration of the intracollicular circuits. As such, we injected a lentivirus- expressing EGFP unilaterally into the SC at E14 and analyzed the disposition of the axons in the contralateral sSC at E18 and P0. We detected significantly more labeled axons in the sSC layer at E18 than at P0 (Fig. 3D). This unplugging of the sSC layer from the dSC layer, which was delayed in embBE mice, coincides with the peak innervation of retinocollicular axons (Fig. 3E) (7). These results suggest that the arrival of retinal axons to the SC prompts the segregation of somatosensory and visual circuits.

Fig. 3. Blocking stage I retinal waves prolongs the multimodal configuration.

(A) Whisker pad stimulation leads to the concomitant bilateral activation of both V1 and SC in control E18 mice, which switches to unimodal at P0. (B) Quantification of the data shown in (A) (n = 8 E18, n = 13 P0). (C) Scheme summarizing the data. (D) (Left) Optical coronal sections from 3D light-sheet images showing EGFP lentiviral– labeled axons in the sSC of control E18 (arrowheads) but not P0 mice. The lentivirus was injected into the SC at E14. (Right) Quantification of the normalized EGFP expression in contralateral sSC and dSC (n = 9 E18 mice, n = 8 P0 mice). (E) (Left) Coronal views of retinal axons labeled by CTB injection of the eye at E14 that reach the SC at E18 and P0. (Right) Quantification of the data shown at left (n = 9 E18 mice, n = 6 P0 mice). (F) Scheme showing early phases of retinal waves. (G) (Left) Experimental design. (Right) Cortical calcium responses at P1 elicited by mechanical stimulation of the whisker pad after bilateral injection of (left) saline, (middle) cbx, and (right) epibatidine (epb) into the eye at P0. (H) Proportion of the whisker pad stimulations that evoked a bilateral V1 response before and after the injection of saline, cbx, or epb: time (t) = 0, n = 20 mice injected with saline, n = 24 cbx, n = 17 epb; t = 1 to 2 hours, n = 4 saline, n = 8 cbx, n = 6 epb; t = 4 to 5 hours, n = 6 saline, n = 7 cbx, n = 9 epb; t = 7 to 8 hours, n = 7 saline, n = 4 cbx, n = 3 epb; t = 24 hours, n = 19 saline, n = 14 cbx, n = 13 epb; and t = 32 hours, n = 11 saline, n = 12 cbx, n = 7 epb). (I) Scheme summarizing the data. (J) (Left) Optical coronal sections from 3D light-sheet images showing lentiviral EGFP – labeled axons in the sSC of P1 mice treated at P0 with cbx (arrowheads) but not with saline. The lentivirus was injected into the SC at E14. (Right) Quantification of the normalized EGFP expression in contralateral sSC and dSC (n = 5 saline, n = 6 cbx). Scale bars, (A) and (G) 1000 μm; (D) 200 μm; (E) 250 μm; (J) 200 μm. Boxplots show the medians with the interquartile range (box) and range (whiskers). The bold lines in (H) indicate the mean, and the shading indicates the SEM. The bar graphs in (B) show the mean ± SEM. ns, not significant; *P < 0.05, **P < 0.01, ***P < 0.001.

Perinatal retinal waves drive the SC circuit reorganization

Developing retinal axons convey a stereotyped pattern of spontaneous activity known as retinal waves. Stage I retinal waves, which are mediated by gap junctions, last approximately from E17 until P1 in mice (Fig. 3F) (13), and therefore, we tested whether these waves might play a role in directing intracollicular sensory-modality segregation. We blocked stage I retinal waves by injecting the gap junction blocker carbenoxolone (cbx) (13, 14) into both eyes of CxGCaMP6f mice at P0. To confirm that cbx blocks gap junction-–mediated retinal waves in vivo, we recorded spontaneous activity in V1, the frequency of which was reduced significantly 1 hour after the administration of cbx at P0 (fig. S7, A and B). This reduction was not due to the degeneration of retinofugal axons because these axons reached both the thalamus and the SC (fig. S7C), and visual spontaneous activity was recovered at P6 (fig. S7, A and B). Accordingly, acute administration of epibatidine, a high-affinity cholinergic agonist that inhibits postnatal stage II retinal waves (15–17), significantly reduced the frequency of spontaneous activity in V1 at P6, whereas cbx had no such effect at this stage (fig. S7, D and E). Altogether, these experiments show that most of V1 activity at the perinatal stage in vivo is driven by gap junction–mediated retinal waves.

We next recorded the cortical activity after whisker pad stimulation in CxGCaMP6f mice treated with cbx at P0 (Fig. 3G). Seven to 8 hours after cbx administration, stimulation of the whisker pad triggered in all mice a multimodal response that was elicited in 44% of the trials, and that rose to 64% after a second injection of cbx on the following day (Fig. 3, G to I, and movies S9 and S10). By contrast, this multimodal response was not triggered by the injection of saline or epibatidine (Fig. 3, G to I). Cbx administration to Cx-SCGCaMP6f mice at P0 showed concomitant bilateral responses in the V1 and SC at P1 (fig. S8, A to C, and movie S11), which were abolished by TTX acute injection into the SC (fig. S8, D and E). As in the embBE, lentiviral tracing in the SC of cbx-treated mice showed the abnormal invasion of axons from the dSC layer into the sSC layer at P1 (Fig. 3J). Last, we checked whether the responses to whisker pad stimulation eventually switch to unimodal in the cbx-treated mice. When we performed the whisker pad stimulation in the CxGCaMP6f and triple Cx-SCGCaMP6f mice, we found that the multimodal responses both in the V1 and SC were almost absent by P6 (fig. S8, F to I). These data show that the timely segregation of somatosensory and visual pathways require a perinatal reorganization of intracollicular circuits that depends on gap junction-–mediated retinal waves.

Failure to timely segregate drives long-lasting circuit alterations

Although reversed by the end of the first postnatal week, the transient expansion of the multimodal phase may have long-term effects on the specification of sensory circuits. For example, perinatal blockage of retinal activity might cause enduring changes in the transcriptional program of SC layers. To assess this, we performed bulk RNA-sequencing (RNA- seq) of the sSC and dSC layers at P6 in mice treated with saline or cbx at P0 to P1 (Fig. 4A). A principal components analysis (PCA) revealed that sSC and dSC cells clustered according to their anatomic origins (fig. S9A). Moreover, a differential expression analysis (DEA) in saline conditions revealed 1544 differentially expressed genes (DEGs) enriched in the sSC as opposed to 1,549 in the dSC layers (Fig. 4A and table S1). Among the DEGs enriched in each population, we found genes previously identified as layer-specific markers, including those encoding transcription factors such as Rorb and Barhl1 in the sSC layer or Pou4f1 and Pou4f2 in the dSC layer (18, 19). To determine whether cbx may influence SC layer–specific genes, we compared the RNA-seq data between mice that received saline or cbx and found that cbx significantly modified 24% (370 genes) of the sSC-specific genes and 21.1% (326 genes) of the dSC-specific genes (Fig. 4B, fig. S9B, and table S2). Among these, cbx induced the expression of a large proportion of the dSC-specific genes (285 out of 326 genes) in the sSC layer, and that of sSC-specific genes (168 out of 370 genes) in the dSC layer. Likewise, it provoked the down-regulation of layer-specific gene expression to levels resembling those in the other layer (Fig. 4C and fig. S9, C and D). Together, these results demonstrate that the somatosensory and visual molecular identity of SC layers was altered by the blockade of stage I retinal waves (Fig. 4D).

Fig. 4. Transcriptomic SC alterations following blockade of perinatal retinal waves.

(A) (Left) Scheme of the RNA-seq experiments of sSC and dSC tissue from P6 mice treated with saline and cbx at P0-P1. (Right) Heatmap of normalized, regularized logarithm (Rlog) z-score of expression, and unbiased clustering of significant DEGs between the sSC and dSC in saline- treated mice. Each row represents a gene, and the columns are biological replicates. The color code indicates the normalized expression for up-regulated genes in yellow versus down-regulated genes in purple. Highlighted genes are those previously identified as layer- specific markers. (B) Volcano plot showing the significance and P value distribution of DEGs. The light brown and dark brown dots indicate sSC and dSC DEGs, respectively, and the green dots and the percentages indicate layer-specific DEGs shifted by cbx, with the top 10 protein coding genes listed in every region. (C) Heatmaps of the normalized regularized logarithm (Rlog) z-score of expression and unbiased clustering of layer-specific DEGs whose expression was modified by cbx in the sSC and dSC regions. (D) Scheme summarizing the data.

The alterations to the modality-specific identity of sSC and dSC may somehow affect the stereotypic organization of incoming inputs to these layers. Within the visual system, retinocollicular axons are organized in eye-specific segregated clusters in the sSC (7, 20), and therefore, we tested whether eye-specific segregation is disrupted in cbx-treated mice. Cholera toxin subunit B (CTB) injection into the eyes of perinatally cbx-treated mice (fig. S10A) showed a compromised eye-specific segregation in the sSC at P15. Whereas in the saline condition, contralateral and ipsilateral axons segregated into dorsal and ventral sSC compartments, respectively, the ipsilateral axons in cbx-treated mice invaded the dorsal compartment and mingled with the contralateral axons (Fig. 5, A and B, and movies S12 and S13). In addition, small dye crystals placed in V1 at the azimuth and elevation axis at this stage also showed alterations in the fine-scale organization of the geniculo-cortical pathway (Fig. 5, C and D, and fig. S10, B to D). Together, these results demonstrate that early perturbations in retinal activity leads to long-lasting circuit reconfiguration at central stations within the visual pathway.

Fig. 5. Long-term alterations in eye-specific segregation and retinotopy after blockade of perinatal retinal waves.

(A) (Left) Experimental design. (Right) Coronal view of the SC showing axons from the ipsilateral (CTB-555) and contralateral (CTB-647) eye at P15 in saline- and cbx-treated mice. (B) Quantification of the data in (A) (contralateral, n = 10 saline, n = 6 cbx.; ipsilateral, n = 7 saline, n = 5 cbx). (C) (Left) Experimental design. (Right) Coronal view of the dLG showing back label from DiI and DiD crystals in V1 at P15 in saline- and cbx-treated mice. (D) Quantification of the data shown in (C) (azimuth, n = 7 saline, n = 5 cbx; elevation, n = 6 saline, n = 5 cbx). Scale bars, 100 μm. The bold lines in (B) indicate the mean, and the shading indicates the SEM. Bar graphs show the mean ± SEM. ns, not significant; *P < 0.05, **P < 0.01.

Discussion

Our findings demonstrate that visual and somatosensory pathways emerge as multimodal circuits, and that during perinatal life, they segregate in a manner that is orchestrated in the SC. The multimodal-to-unimodal transition requires a reconfiguration of intracollicular circuits at birth, in which dSC layers disconnect from the sSC layers. This reconfiguration drives the specificity of sensory circuits, so that whisker pad stimulation exclusively triggers S1 responses in the cortex after birth. The switch from multimodal to unimodal requires early retinal activity because perturbation of stage I retinal waves prevents the reconfiguration of the circuits in the SC and leads to the extension of the multimodal phase into postnatal life.

Therefore, these data broaden our understanding of retinal wave function by revealing their instructive role in the acquisition of sensory modality specificity, which is beyond their classic role in the postnatal refinement of visual circuits (21–25).

The segregation of visual from somatosensory systems must occur in a limited developmental window because any delay in this segregation will cause long-lasting changes in the intra-modal organization of visual circuits. The mechanisms that mediate the final closing of the multimodal phase in embBE and cbx-treated mice remain unexplored, although they may include the influence of passive whisking during the first days of life (26) or the progressive assembly of inhibitory cells into the networks of the SC, as described for cortical networks (27). A longitudinal analysis of both the molecular identity of collicular neurons, and the maturation of their connectivity in physiological and in manipulated scenarios, will help identify the cell types and potential factors involved in this circuit segregation and plasticity.

Last, our findings reveal that the SC, a highly conserved structure in vertebrates (28, 29), participates in the construction of brain regions that appeared more recently in phylogenetic terms such as the neocortex, a role that goes far beyond the SC’s well- established sensorimotor and multimodal integrative functions (30). From the phylogenetic perspective, it has been stated that the visual cortex has inherited an increasing number of functions from the SC related to the processing of visual features (31–33). Our results spotlight the ontogenetic perspective, in which the developing SC exerts a master control on cortical specification and configuration of visual circuits. Thus, we believe that a deeper understanding of the functional development of phylogenetically ancient structures is crucial to understand how the neocortex is formed and its functional areas are specified.

Data and materials availability

All the data are available in the main text or the supplementary materials. The RNA-seq data generated in this study has been deposited in the NCBI Gene Expression Omnibus repository under accession GSE198112 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE198112).