Research Article

Lattice system of functionally distinct cell types in the neocortex

See allHide authors and affiliations

Science  03 Nov 2017:
Vol. 358, Issue 6363, pp. 610-615
DOI: 10.1126/science.aam6125

The basic modules of the neocortex

The fundamental organization of excitatory and inhibitory neurons in the neocortex is still poorly understood. Subcerebral projection neurons, a major excitatory cell type in neocortical layer 5, form small cell clusters called microcolumns. Maruoka et al. examined large regions of mouse brain layer 5 and observed that thousands of these microcolumns make up a hexagonal lattice with a regular gridlike spacing. The other major layer 5 excitatory cell class, cortical projection neurons, also form microcolumns that interdigitate with those of the subcerebral projection neurons. Microcolumns received common presynaptic inputs and showed synchronized activity in many cortical areas. These microcolumns developed from nonsister neurons coupled by cell type–specific gap junctions, suggesting that their development is lineage-independent but guided by local electrical transmission.

Science, this issue p. 610


The mammalian neocortex contains many cell types, but whether they organize into repeated structures has been unclear. We discovered that major cell types in neocortical layer 5 form a lattice structure in many brain areas. Large-scale three-dimensional imaging revealed that distinct types of excitatory and inhibitory neurons form cell type–specific radial clusters termed microcolumns. Thousands of microcolumns, in turn, are patterned into a hexagonal mosaic tessellating diverse regions of the neocortex. Microcolumn neurons demonstrate synchronized in vivo activity and visual responses with similar orientation preference and ocular dominance. In early postnatal development, microcolumns are coupled by cell type–specific gap junctions and later serve as hubs for convergent synaptic inputs. Thus, layer 5 neurons organize into a brainwide modular system, providing a template for cortical processing.

The mammalian neocortex is densely populated by diverse types of excitatory and inhibitory neurons, each with specific molecular and cellular properties, synaptic connections, and in vivo functions. Whether neocortical cell types organize into repeated structures representing common motifs for information processing has been poorly understood. Cortical columns, including orientation columns in the visual cortex and barrels in the somatosensory cortex, have patterned structures, but their cellular and circuit-level organization is largely unsolved (1). Moreover, cortical columns are restricted to specific cortical areas and therefore do not represent brainwide structural motifs.

In neocortical layer 5, subcerebral projection neurons (SCPNs) are one of the two major excitatory neuron types and have well-defined anatomical and genetic specifications. SCPNs constitute the major cortical output pathway, sending massive axonal projections to subcortical targets including the pons, spinal cord, and superior colliculus (2). Prior studies described a local arrangement of SCPNs in radial clusters with a diameter of one to two cells and tangential distances of a few cell diameters (3, 4). These clusters, here termed microcolumns, have been reported in visual and somatosensory cortical areas in mice (3) and in language areas in humans (4). To study the cellular organization of neocortical layer 5, we conducted structural and functional analyses of SCPN microcolumns and investigated whether other cell types organize into microcolumnar structures.

Spatial organization of major cell types in layer 5

We examined the three-dimensional organization of SCPN microcolumns in the mouse brain. SCPNs were retrogradely labeled by injecting fluorescent tracers into the pons (Fig. 1A and fig. S1A). After fixation and clearing, the neocortex was scanned using two-photon microscopy (Fig. 1B). Confirming previous results (3), we observed SCPN microcolumns in visual and somatosensory cortices (Fig. 1, C and D). The radial alignment of SCPNs was also conserved in the motor cortex (Fig. 1E). Statistical analyses showed a microcolumnar organization in nearly all examined neocortical regions (Fig. 1F and fig. S1, B to L). The orientation of microcolumns gradually changed along the cortex (Fig. 1F and fig. S1M) but remained approximately parallel to apical dendrites (fig. S1, N to P, and materials and methods). The radius of microcolumns was ~10 μm in visual, somatosensory, and motor cortices (Fig. 1G).

Fig. 1 Lattice organization of SCPN microcolumns.

A, anterior; P, posterior; M, medial; L, lateral. (A) SCPNs were labeled with retrograde tracers. L5, layer 5; Sup, superior colliculus. (B) Labeled brain at age 5 weeks. Gamma correction was applied uniformly. (C to E) Sagittal [(C) and (D)] and coronal (E) optical sections of (B). (F) Detection of microcolumnar alignment. (Top left) The density of SCPNs (circles) in the two cylindrical volumes relative to other SCPNs was determined for various orientations. (Right) Each panel shows the result for cells around the corresponding square in the bottom left panel. Minimum-maximum ranges are given in the materials and methods. (Bottom left) Estimated microcolumn orientations [data from the brain in (B)]. (G) Structure of microcolumns. Each cortical area was divided into multiple subregions of about equal size. In each subregion, the SCPN density was measured at different tangential distances from other SCPNs (top left) and normalized to the average density in the subregion. Green line and shading, mean and SEM, respectively, across subregions (two mice at 5 weeks of age). The troughs at 16 to 24 μm were significant (P < 0.005 for the three areas, two-tailed sign test against 1). (H) Tangential section image of the brain in (B). Retrogradely labeled SCPNs are shown in black. Thickness, 200 μm. (I) Enlarged image. Red dots are the estimated centers of microcolumns. Gamma correction was applied uniformly in (H) and (I). (J) Autocorrelogram of the SCPN density distribution in (H) (1539 cells). (K) Autocorrelograms for the visual (2676 cells) and somatosensory (3982 cells) areas of the brain in (B). (L) (Left) Two-dimensional power spectrum of the density distribution of 15,765 SCPNs in the brain in (B). Colored circles represent the centers of the peak frequency components used to reconstruct the waves in (H) and (I). (Right) Power spectrum for 17,307 SCPNs in another mouse. Colored circles are at the same positions as those in the left panel.

Analyses of the organization of microcolumns have been performed previously in brain slices (3, 4), but their two-dimensional lateral arrangement in the cortex has not been determined. A two-dimensional Fourier analysis of the SCPN distribution revealed a periodicity of 30 to 45 μm (P < 0.001 in three of three mice; fig. S1, Q to S). We further analyzed the periodicity with a correction for microcolumn tilt (fig. S1T). Tangential section images (Fig. 1, H and I) and the autocorrelogram of the SCPN distribution (Fig. 1J) suggested an approximately hexagonal pattern, which was observed in multiple cortical areas (Fig. 1K). We found a sixfold symmetry (P < 0.01; fig. S1, U to X) but no other rotational symmetries. Consistently, the two-dimensional power spectrum had six peaks ranging from 24 to 30 cycles/mm, two of which were located on the anterior-posterior axis and the other four of which were at lateral positions (Fig. 1L; computed for areas containing ≥2,000 microcolumns). Individual microcolumns were positioned near the intersections of the three waves reconstructed from the six peaks of the power spectrum (Fig. 1, H and I).

Layer 5 contains another major type of excitatory neuron that innervates the cerebral cortex: cortical projection neurons (CPNs) (2). We labeled CPNs in Tlx3 (T-cell leukemia homeobox 3)–cre/Ai9 mice (5) (green in Fig. 2A and fig. S2, A to F) and, in parallel, visualized SCPN microcolumns by retrograde labeling (magenta in Fig. 2, A and B, left). In layer 5b—the lower part of layer 5, where SCPNs are present—the density of CPNs radially aligned to SCPNs was lower than the average density of CPNs (P < 0.01; Fig. 2B, middle), indicating that CPNs were excluded from SCPN microcolumns. Moreover, CPNs were radially aligned to each other in an orientation parallel to SCPN microcolumns (P < 0.01; Fig. 2B, right). Thus, CPNs are organized into cell type–specific microcolumns that interdigitate with SCPN microcolumns. CPNs in layer 5a also adopted a microcolumnar arrangement (fig. S2G).

Fig. 2 Cell type–specific microcolumnar organization of CPNs and inhibitory neurons.

Analyses of layer 5b cells in mice at age 5 to 12 weeks. Photographs show the somatosensory area. White dashed lines mark the border between layers 5a and 5b. (A and B) Analysis of SCPNs and CPNs. Colored lines in (B) show cell density for 31,304 SCPNs and 12,560 CPNs in two mice (determined similarly to the results shown in Fig. 1G). (Left) The mean and SEM (shading) calculated among multiple subregions in the cortex. The trough at 15 to 25 μm was significant (P < 3.1 × 10−5, two-tailed sign test against 1). (Middle and right) Gray, 100-surrogate data generated by random positioning of CPNs. Dashed lines, highest and lowest 2.5% of surrogates. Black line, median of surrogates. The trough in the right panel at 15 to 20 μm was significant (P < 0.01). (C to F) Analysis of inhibitory neurons, shown similarly to the middle panel in (B) (data from three mice): (C) 5727 SCPNs and 1990 PV+ cells, (D) 6179 SCPNs and 1793 SOM+ cells, (E) 7601 CPNs and 2830 PV+ cells, and (F) 8772 CPNs and 3249 SOM+ cells. In (E) and (F), the orientation of microcolumns was estimated from that of the apical dendrites.

We also investigated the arrangement of the two most prevalent inhibitory neuron types in layer 5—parvalbumin-expressing (PV+) and somatostatin-expressing (SOM+) cells (6)—using fluorescent immunostaining in three-dimensional samples (7) (Fig. 2, C to F). PV+ and SOM+ cells aligned radially with SCPNs (P < 0.01; Fig. 2, C and D) but not with CPNs (Fig. 2, E and F, and fig. S2H), indicating a selective alignment of inhibitory neurons to excitatory neuron microcolumns.

In vivo neuronal activity of microcolumns

In vivo microcolumn activity has been only indirectly inferred from studies of immediate early gene expression in fixed slices (3). We therefore investigated microcolumn activity in vivo using an adeno-associated viral (AAV) vector that expresses the Ca2+ indicator G-CaMP6 (8) almost exclusively in SCPNs (Fig. 3, A and B, and fig. S3), most likely owing to tropism. Ca2+ signals were obtained from awake mice using two-photon volume imaging (120 to 240 μm thick, 1.6 to 2.4 volumes/s). The orientation of SCPN microcolumns was approximated from the axes of the apical dendrites. Data from four representative SCPNs in the binocular visual cortex, recorded without visual stimulation, are shown in Figure 3, C to G. Cells 1 to 3 (tangential distance of <15 μm) showed synchronous Ca2+ signals, whereas cell 4 (tangential distance to the other three cells of >25 μm) exhibited no synchronization with other cells (Fig. 3E). In accord, the temporal correlation of Ca2+ traces was higher among cells 1 to 3 than between cell 4 and other cells (Fig. 3G). We calculated the average correlation as a function of the tangential distance (Fig. 3H, left). The actual correlation values at tangential distances of <15 μm were greater than those for random surrogates (Fig. 3H), whereas the actual values at tangential distances of >20 μm were almost at the level of those for the surrogates (Fig. 3H), indicating synchronized activity within radially aligned SCPNs (Fig. 3, H and I, and fig. S4, A to D). We also confirmed significant synchronization within individual microcolumns (materials and methods). The observed correlation was not caused by light contamination (Fig. 3, E and F; fig. S4, E to G; and materials and methods). Similar results were obtained in the primary somatosensory and motor cortices (Fig. 3, J and K; fig. S4, A to G; and materials and methods).

Fig. 3 Synchronized activity in SCPN microcolumns.

Data from mice at age 11 weeks. (A) Diagram of imaging. Data from the binocular visual cortex are shown in (B) to (I). (B) Expression of G‐CaMP6 in SCPNs. (C) Example cells (1 to 4). (D) Temporally averaged images of the cells in (C), top view. Green dotted lines, cell contours. (E) Ca2+ traces of the cells in (C). Magenta lines and arrows, synchronized peaks of cells 1 to 3. Cyan lines, large peaks in only one of cells 1 to 3, indicating that light contamination was undetectable. ΔF/F, relative change in fluorescence. (F) Data at “F” in (E). Left, Ca2+ traces. Right, fluorescence images (top view) at the time frames indicated by dotted lines in the left panel. Green dotted lines, cell contours. Peaks in Ca2+ traces were accompanied by a signal increase across the entire cell body, indicating that they represent neuronal activity but not light contamination. (G) Correlation coefficients of Ca2+ traces for cells 1 to 4. (H and I) Dependence of the average correlation on the distance (1209 cells in three mice). (H) Dependence on tangential distance. Green, actual data. Gray, data from 1000 random surrogates. Black dotted and solid lines, top 5% and median of random surrogates, respectively. (I) Dependence on radial distance. Pairs with a tangential distance of <10 μm were analyzed. (J) Somatosensory cortex (1529 cells, three mice). (K) Motor cortex (1302 cells, three mice).

Although the superficial layers of the mouse visual cortex show weak clustering of neurons with similar response properties (9, 10), cellular clustering for visual functions in layer 5 is poorly understood. We analyzed the orientation preference and ocular dominance of SCPN microcolumns in the binocular visual cortex. We presented drifting gratings with six different orientations (Fig. 4A) and determined the preferred orientation for each SCPN that exhibited orientation-selective responses. The difference in the preferred orientation of SCPN pairs with a tangential distance of >20 μm was similar to that of randomly selected pairs (~45°; Fig. 4C, left). In contrast, SCPN pairs with a tangential distance of <10 μm and pairs within individual microcolumns had a significantly smaller difference (~32°; Fig. 4C, left; fig. S4H; and materials and methods). The similarity was observed even when the radial distance was as large as ~80 μm (Fig. 4C, right, and fig. S4H). We also determined the ocular dominance index (ODI) by stimulating both eyes alternately (Fig. 4B). The ODI was similar for SCPN pairs with a tangential distance of <10 μm and within individual microcolumns (Fig. 4D and materials and methods).

Fig. 4 Orientation preference and ocular dominance of SCPN microcolumns.

Data from the binocular visual area at age 10 to 11 weeks. (A and B) Example of radially aligned SCPNs. (A) (Left) G-CaMP6 labeling. (Right) Individual (light blue) and average (dark blue) responses to grating stimuli delivered to the contralateral eye. (B) Responses to contralateral and ipsilateral stimulation, averaged across orientations. (C) Difference in the preferred orientation of SCPN pairs (1621 orientation-selective SCPNs in six mice). (Left) Median values among cell pairs plotted against the tangential distance. Red, actual data. Gray, data from 1000 random surrogates. Black solid and dotted lines, median and bottom 5% of random surrogates, respectively. P = 0.002 for the tangential distance of <10 μm. (Right) Dependence on the radial distance. Pairs with a tangential distance of <10 μm were analyzed. The radial bin width was 50 μm. (D) Mean difference in the ocular dominance index (ODI), shown similarly to (C), where the blue line is actual data (2136 visually responsive SCPNs in seven mice). P = 0.007 for the tangential distance of <10 μm in the left panel.

Chemical and electrical synaptic connections of microcolumns

We investigated the synaptic circuits that could coordinate microcolumnar neuronal activity. Whole-cell patch clamp recordings were obtained from two to four enhanced green fluorescent protein (EGFP)–labeled SCPNs in acute slices prepared from the visual cortex of Crym-egfp mice (3) at ~4 postnatal weeks (Fig. 5A), when in vivo synchronized activity was already present (fig. S5A). We first examined mutual connections and failed to detect preferential connections between radially aligned SCPNs (fig. S5B), consistent with previous findings (11, 12). We next examined common synaptic inputs to a pair of neurons, which generate synchronized excitatory postsynaptic currents (EPSCs) with a typical time difference of <1 ms (1315) (Fig. 5, B and C, and fig. S5, C and D). When all EPSCs were included for the analysis of synchronized EPSCs, we found no preference for radially aligned SCPNs (fig. S5E). We subsequently examined large EPSCs, which are particularly important for mammalian brain function (1618). When we analyzed the largest 7.4% of EPSCs induced by presynaptic spikes (corresponding to the largest 2% of all recorded EPSCs, including those not induced by presynaptic spikes; Fig. 5D; fig. S5, F and G; and materials and methods), the probability that radially aligned SCPNs (tangential distance of <7.5 μm) had synchronized synaptic activity was higher than that expected for uniform random connections (P = 0.0036, one-tailed binomial test; Fig. 5E), and more than seven times that for pairs with only slightly longer tangential distances (7.5 to 22.5 μm; P = 0.0012, one-tailed Fisher’s exact test; Fig. 5E). The results were robust against changes in the parameters used for the analyses (fig. S5H). In contrast, we found no preference for tangentially aligned SCPNs (P = 0.63, one-tailed binomial test, and P = 0.51, one-tailed Fisher’s exact test; Fig. 5F). These results suggest that SCPNs in individual microcolumns preferentially receive strong synaptic inputs from common presynaptic neurons (materials and methods).

Fig. 5 Convergent strong inputs to SCPN microcolumns.

(A) A visual cortex slice of a P24 Crym-egfp mouse. SCPNs and recorded cells were labeled by EGFP expression and biocytin injection, respectively. White arrows indicate recorded neurons. (B) Current traces (top) and EPSC onsets (bottom) of a SCPN pair. Asterisk, coincident EPSCs. (C) Cross-correlogram of EPSC rates of the pair in (B). Bin size, 2 ms. (Inset) Cross-correlogram shown on a longer time scale. (D) Tangential and radial distance between simultaneously recorded SCPN pairs. Red and gray dots indicate SCPN pairs with and without synchronized EPSCs, respectively (n = 183 pairs; P21 to P28). (E) Black, probability that a SCPN pair had synchronized EPSCs. Error bars, 95% confidence intervals estimated using the binomial distribution. Blue, average probability. Fractions indicate the number of pairs with synchronized EPSCs out of the total number of recorded pairs. Black numbers are for pairs in the first (<7.5 μm) and second (7.5 to 22.5 μm) bins. Blue numbers are for all pairs. (F) The same analysis as in (E) but for the radial distance. **P < 0.01; n.s., not significant (P ≥ 0.05); one-tailed Fisher’s exact tests.

Microcolumns are present at postnatal day 6 (P6) (3), when chemical synapses are still infrequent (19, 20), suggesting the possibility that microcolumn neurons have cellular interactions other than chemical synaptic connections during cortical development. We therefore investigated gap junction–mediated electrical coupling, which is implicated in the development of neuronal circuits (21), in acute visual cortex slices (Fig. 6, A and B, and fig. S6, A to D). At P6 to P7, about half of neighboring SCPN-SCPN and CPN-CPN pairs exhibited electrical coupling (37 of 75 SCPN-SCPN pairs, 49%; 44 of 80 CPN-CPN pairs, 55%; Fig. 6C), whereas only 12% of neighboring SCPN-CPN pairs were coupled (7 of 58 pairs; Fig. 6C). The coupling probabilities of SCPN-SCPN and CPN-CPN pairs were significantly higher than that of SCPN-CPN pairs (P < 10−5 for both, two-tailed Fisher’s exact tests). Further, the coupling coefficients of SCPN-SCPN and CPN-CPN pairs were significantly greater than those of SCPN-CPN pairs (Fig. 6C). Electrical coupling became undetectable by the end of the second postnatal week (Fig. 6C).

Fig. 6 Cell type–specific microcolumnar electrical coupling during development.

(A and B) Recordings in visual cortex slices of Crym-egfp mice at P6 to P7. Numbers indicate individual neurons. Green, EGFP-expressing SCPNs. Magenta, retrogradely labeled CPNs. Light blue and white, recorded neurons labeled by biocytin injection. (A) A slice at P6. (B) (Left) Recorded neurons. (Right) Black, hyperpolarizing pulses injected into a neuron. Dark blue, the average membrane potential response of the other neuron. (C) Coupling coefficients of neighboring pairs. Center-to-center distances, <25 μm (P6 to P7) and <30 μm (P10 to P11 and P14 to P15). Numbers of recorded pairs are shown on the horizontal axis. The vertical axis is truncated. (D) Distribution of coupled (filled circles) and noncoupled (open circles) SCPN-SCPN pairs (n = 123 pairs). (E) Dependence of coupling coefficient on the distance. Triangles, average coupling coefficient of pairs with a tangential distance of <20 μm, plotted against the radial distance. Circles, average coupling coefficient of pairs with a radial distance of <20 μm, plotted against the tangential distance. Error bars, SEM. In the statistical tests between orientations, pairs grouped to both orientations were excluded. (F) Comparisons between pairs with a different tangential distance. Fractions indicate the numbers of coupled pairs out of all tested pairs. Error bars, 95% confidence intervals for the binomial distribution. Right panel in (F), two-tailed Fisher’s exact test; other panels, two-tailed Mann-Whitney-Wilcoxon tests. ***P < 0.001; **P < 0.01; *P < 0.05; n.s., not significant (P ≥ 0.05).

In the radial orientation, the coupling coefficient of SCPN-SCPN pairs was largely independent of distance up to 50 μm (Fig. 6, D and E). In contrast, in the tangential orientation, the coupling coefficient rapidly decreased and approached zero at a distance of >30 μm (Fig. 6, D and E). Consequently, radially aligned pairs had larger coupling coefficients than tangentially aligned pairs at the same distance (Fig. 6E). Moreover, pairs with a tangential distance of <15 μm had greater coupling coefficients and coupling probabilities than those with larger tangential distances (Fig. 6F). A similar radial bias was found for CPN-CPN coupling (fig. S6, E and F). These results suggest that gap junctions preferentially couple neurons within individual microcolumns (materials and methods).

In the developing neocortex, clonally related excitatory neurons are preferentially coupled by gap junctions, but the coupling becomes infrequent by the end of the first postnatal week (~2% at P6) (21). In contrast, radially aligned SCPNs are mostly nonsisters (3) and frequently coupled at P6 to P7 (Fig. 6, D to F), suggesting that the coupling observed in this study occurred between nonsister pairs. In accord, half of neighboring nonsister SCPN pairs had electrical coupling at P6 to P7 (18 of 38 pairs; fig. S6, G to J).


We discovered that wide areas of neocortical layer 5 are organized into a cellular lattice system composed of cell type–specific microcolumns (fig. S7). The functional modularity suggests that single microcolumns perform elementary circuit functions that collectively constitute large-scale parallel processing. The descriptions of microcolumns in multiple cortical areas in mice [(3) and this study] and humans (4) suggest that the lattice system is a neuronal architecture common to cortical functions as diverse as sensory, motor, and language processing. The coordinated in vivo activity of SCPN microcolumns and their convergent inputs indicate that they constitute a brainwide system of modular, repeated synaptic circuits and discrete cortical output channels.

Several mammalian species, but not rodents, possess ocular dominance columns and orientation columns. The typical width of an ocular dominance column is ~500 μm, and the preferred orientation of orientation columns changes gradually across the cortical surface. In mice, neighboring microcolumns (distance of ~40 μm) had no apparent similarity in visual responses. We hypothesize that in species that have orientation columns and ocular dominance columns, neighboring microcolumns may be progressively arranged to have similar functions, thereby contributing to the anatomy of known cortical columns. Orientation columns have a width roughly similar to microcolumn spacing (1) and exhibit a hexagonal arrangement (22); therefore, they may be constructed on the basis of the lattice system.

Previous studies demonstrated that clonally related neurons show electrical coupling in the early neonatal stage (21) and later exhibit similar orientation preference (23, 24). In contrast, our findings show that microcolumns are composed of clonally unrelated cells (3) that have specific electrical coupling during P6 to P7, when cortical synapses are being generated (19, 20). The transient coupling may synchronize neuronal activity and promote the development of microcolumn-specific circuits through Hebbian-like mechanisms. Gap junctions may also facilitate microcolumnar clustering through cell adhesion properties (25). Microcolumns might be structurally linked to “neuronal domains”—radial neuronal clusters observed in neonatal cortex that are suggested to have gap junction coupling (2628). Because neuronal domains span multiple cortical layers, microcolumns might also be present in other cortical layers.

Supplementary Materials

Materials and Methods

Figs. S1 to S7

References (2939)

Data S1

References and Notes

  1. Acknowledgments: The AAV vector was a gift from A. Yamanaka (Nagoya University; CREST, Japan Science and Technology Agency) and is available from Nagoya University under a material transfer agreement. We thank C. Yokoyama for thoughtful discussions and manuscript editing, A. Miyawaki and H. Hama for valuable support with the anatomical experiments, S. Kondo and K. Ohki for useful advice on in vivo microscopy, S. Tonegawa and H. Okamoto for helpful discussions on manuscript preparation, and H. Kazama and E. I. Moser for critical reading. We also thank K. Kiso, N. Matsumoto, E. Ohshima, and M. Kishino for technical assistance and the RIKEN Brain Science Insitute–Olympus Collaboration Center for providing imaging equipment. This work was supported by research funds from RIKEN to T.H. and Grants-in-Aid for Scientific Research from MEXT (the Ministry of Education, Culture, Sports, Science and Technology of Japan) to T.H. (Innovative Areas “Mesoscopic Neurocircuitry,” 22115004), N.N. (16K14565), S.T. (24700344), and S.S. (25890023). Neuron coordinate data are available in the supplementary materials. H.M., N.N., S.T., S.S., and T.Y performed the Ca2+ imaging. N.N. performed the gap junction experiments. S.T. performed the synaptic connectivity experiments. S.S. performed the anatomical experiments. T.Y. performed the visual response analyses. H.M., N.N., S.T., S.S., T.Y., and T.H. analyzed the data and wrote the paper. T.H. conducted the research.
View Abstract

Stay Connected to Science

Navigate This Article