Clonally related visual cortical neurons show similar stimulus feature selectivity.

A fundamental feature of the mammalian neocortex is its columnar organization. In the visual cortex, functional columns consisting of neurons with similar orientation preferences have been characterized extensively, but how these columns are constructed during development remains unclear. The radial unit hypothesis posits that the ontogenetic columns formed by clonally related neurons migrating along the same radial glial fibre during corticogenesis provide the basis for functional columns in adult neocortex. However, a direct correspondence between the ontogenetic and functional columns has not been demonstrated. Here we show that, despite the lack of a discernible orientation map in mouse visual cortex, sister neurons in the same radial clone exhibit similar orientation preferences. Using a retroviral vector encoding green fluorescent protein to label radial clones of excitatory neurons, and in vivo two-photon calcium imaging to measure neuronal response properties, we found that sister neurons preferred similar orientations whereas nearby non-sister neurons showed no such relationship. Interestingly, disruption of gap junction coupling by viral expression of a dominant-negative mutant of Cx26 (also known as Gjb2) or by daily administration of a gap junction blocker, carbenoxolone, during the first postnatal week greatly diminished the functional similarity between sister neurons, suggesting that the maturation of ontogenetic into functional columns requires intercellular communication through gap junctions. Together with the recent finding of preferential excitatory connections among sister neurons, our results support the radial unit hypothesis and unify the ontogenetic and functional columns in the visual cortex.

To identify clonally related sister cells, we used a GFP-expressing retrovirus, previously shown to label isolated ontogenetic columns of excitatory neurons[7,11,12]. The retrovirus was injected into the right ventricle in utero at embryonic day 15-17 (E15-E17, see Methods), the beginning of neurogenesis in cortical layer 2/3 (ref. 13). At postnatal day 12-17 (P12 - P17, soon after eye opening), in vivo two-photon imaging[14,15] was performed in the primary visual cortex (V1) of injected mice under anesthesia. A low density of GFP-labeled neurons was observed in layer 2/3 (1.1 ± 0.9 (s.d.) per animal, within an imaging window ~500 μm in diameter at cortical depths up to 400 μm, n = 181 neurons, 161 mice). In some cases (n = 52), we found a pair of GFP-labeled neurons aligned nearly vertically (Fig. 1a, b), with no other GFP neurons nearby, suggesting that they were clonally related sister cells. Although large tangential dispersion has been observed in some clonally related cells[16], here we focused on GFP-labeled neuronal pairs with < 120 μm horizontal separation (see Methods, Supplementary Fig. 1).

Figure 1

Two-photon imaging of clonally related sister cells and nearby layer 2/3 neurons

a, Left, fluorescence images at 190 - 260 μm from pia, showing two GFP-expressing cells that were nearly vertically aligned. Scale bar, 50 μm. Right, 3D reconstruction of the GFP pair. b, Two more examples of GFP cell pairs (arrow heads). Red, GFP; green, OGB. Left, 250 - 310 μm from pia; right, 150 - 240 μm. Scale bar, 100 μm. c, Single condition maps of fluorescence change (ΔF), computed by averaging the images during each stimulus and subtracting baseline (gray screen), for the experiment in a. Central panel, GFP (red) and OGB (green) labeling. Red circle, GFP cell. Scale bar, 50 μm. d, Orientation maps with visually responsive cells colored according to their preferred orientations, for a larger imaging area. White box, region shown in c. Scale bar, 50 μm.

To examine the functional properties of layer 2/3 neurons, we injected the calcium indicator dye Oregon Green BAPTA-1 AM (OGB-1) into a region encompassing the GFP cell pair. Orientation and direction selectivity of OGB-loaded neurons was measured with drifting grating stimuli (100% contrast, spatial frequency 0.02-0.03 cycles/°, temporal frequency, 1 - 2 Hz) presented through the contralateral eye. The mapping was made at two cortical depths to include both GFP neurons in the sister pair. We found that 875 of the 2286 OGB-loaded neurons (38%) showed significant increases in intracellular calcium in response to the grating stimuli (see Methods, Fig. 1c). Among these visually driven neurons, 75% (657/875) showed significant orientation selectivity (p < 0.05, Hotelling’s T-Square test), comparable to previous studies in rodent visual cortex[4,17]. Notably, nearby neurons often preferred different orientations with no apparent spatial organization (Fig. 1c, d), consistent with previous findings of ‘salt-and-pepper’ arrangement of orientation preference in rodent visual cortex[4,10].

Comparing the response properties of the sister cells, however, we found that they often preferred similar orientations (Fig. 1c, d, circles). To quantify this relationship, we fitted the tuning curve of each visually driven neuron with a double Gaussian function (Fig. 2a, Supplementary Fig. 2) to identify its preferred orientation (θ). The functional similarity between each cell pair was measured by the difference between their preferred orientations (Δθ, varying between 0° and 90°). Of the 34 sister pairs in which both neurons were visually driven, 20 pairs (59%) preferred similar orientations (0° < Δθ < 30°), and only 5 pairs (15%) preferred near orthogonal orientations (60° < Δθ < 90°, Fig. 2b). The distribution of Δθ was significantly nonuniform (p = 0.0071, Kolmogorov-Smirnov test), with a strong bias towards 0. In contrast, for pairs of non-sister neurons with horizontal distance < 120 μm, the distribution of Δθ was largely flat (Fig. 2c), significantly different from the sister pairs (Fig. 2d, p = 0.018). The slight biases of the distribution for non-sister pairs towards both 0° and 90° were caused by the over-representation of cardinal orientations (horizontal and vertical) in mouse V1 soon after eye opening[17].

Figure 2

Orientation and direction preference of sister neurons

a, Orientation tuning of the GFP pair (red) and several nearby cells (indicated by numbers) in the experiment shown in Fig. 1a, c, d. Circle, mean; vertical line, ± s.e.m; curve, fitted double Gaussian function. Horizontal lines, baseline (solid) ± s.e.m. (dashed). Inset above, polar plot of orientation tuning. Scale bar, 50 μm. b, Histogram distribution of difference in preferred orientation (Δθ) between sister cells, for all visually driven (open bars) and orientation selective (fill bars) pairs. c, Histogram of Δθ between non-sister (GFP/non-GFP and non-GFP/non-GFP) pairs, for all visually driven (open) and orientation selective (filled) pairs. d, Cumulative distribution of Δθ for sister and non-sister pairs. Red, sisters (p values, difference from uniform distribution); black, non-sisters. Dashed, all visually driven pairs; solid, orientation selective pairs. Gray, diagonal line. e, f, Histograms of difference in preferred direction between sister cells (e) and non-sister cells (f).

When we restricted the analysis to cell pairs with significant orientation selectivity (n = 23), the distribution of Δθ for sister pairs showed a more striking bias towards 0 (Fig. 2b, filled bars), significantly different from both the uniform distribution (p = 0.0038) and the distribution for orientation-selective non-sister pairs (Fig. 2c, filled bars, p = 0.034). Furthermore, even with the sparse labeling, our population of sister pairs could still be contaminated by GFP neurons from separate but neighboring radial clones. Thus, the similarity between true sisters could be even stronger than that observed here.

In addition to similar orientation tuning, the sister neurons also showed a modest tendency to prefer similar directions. When their difference in preferred direction was plotted over the range 0 - 180°, we found more pairs falling between 0° and 90° (21/34 visually driven, 14/23 orientation selective pairs) than between 90° and 180° (Fig. 2e), although the difference was not statistically significant (p = 0.11 for visually driven, p = 0.26 for significantly tuned pairs, bootstrap). For the non-sister pairs, the distribution was largely symmetric (Fig. 2f).

What mechanism confers the sister neurons with similar functional properties? Previous studies in developing cortical slices have revealed spontaneous co-activation of neurons within discrete, radially oriented domains spanning multiple cortical layers, which is mediated by gap junction coupling between the neurons[18]. These domains are comparable to the radial clones in shape and size, and they could provide a blueprint for the functional columns by influencing the formation and fine tuning of chemical synapses. To test this idea, we examined the effect of disrupting gap junction coupling between cortical neurons on orientation tuning. Among all the genes encoding the gap junction protein connexin, Cx26 was found to be highly expressed in developing neocortex and strongly associated with inter-neuronal coupling[19]. We thus made in utero injection of retrovirus expressing a mutant Cx26, with a threonine in the third transmembrane helix (T135) replaced by alanine[20] (Cx26(T135A)-T2A-EGFP). Since the mutant Cx26 forms closed gap junction channels and exerts a trans-dominant-negative effect on other connexins[21], it provides a useful tool for selective disruption of gap junction communication in a small number of cortical neurons (which in this case were labeled with EGFP). When we measured orientation tuning at P12-17, we found that 35% (585/1674) of the OGB-loaded neurons were visually responsive, among which 72% (423/585) showed significant orientation selectivity (p < 0.05, Hotelling’s T-Square test), similar to the mice injected with retrovirus expressing GFP alone. Thus, expression of the mutant Cx26 in a small number of neurons caused no global disruption of V1 responses. However, in contrast to the GFP control (Fig. 2b, d), the distribution of Δθ for sister pairs showed little bias towards 0 (Fig. 3b, d), not significantly different from uniform or the distribution for non-sister pairs (Fig. 3c, p > 0.6, Kolmogorov-Smirnov test). We also found no clear tendency of the sister cells to prefer similar directions (Fig. 3e, f). Thus, in addition to their roles in prenatal neuronal proliferation and migration[22], gap junctions may also be required for coordinating postnatal functional development of sister cells, through either electrical coupling or intercellular exchanges of small molecules[23].

Figure 3

Effect of expressing dominant-negative mutant of Cx26 in sister neurons

a, Two example experiments showing tuning curves of sister pairs (each pair in a box) in mice injected with retrovirus containing Cx26(T135A)-T2A-EGFP. b, Histogram of difference in preferred orientation (Δθ) between sister neurons for all visually driven (open) and orientation selective (filled) pairs in mice expressing mutant Cx26. c, Histogram of Δθ between non-sisters in these mice. d, Cumulative distribution of Δθ for sister pairs expressing mutant Cx26 (cyan; p values, comparison with uniform distribution), those expressing GFP only (red, same as Fig. 2d), and for non-sister pairs in mice injected with retrovirus containing Cx26(T135A)-T2A-EGFP (black). Dashed, all visually driven pairs; solid, orientation-selective pairs. e, f, Histograms of difference in preferred direction between sisters (e) and non-sisters (f).

In addition to the selective disruption of gap junction coupling in a small number of neurons with retrovirus, we also tested the effect of systemic application of a gap junction blocker CBX in the mice injected with retrovirus expressing GFP alone. Since gap junction coupling between cortical neurons declines rapidly in the second postnatal week[24], we injected CBX (IP, 10-20 mg/kg) daily for the first postnatal week to disrupt the inter-neuronal communication during early postnatal development but not at the time of imaging. We found that CBX injection similarly disrupted the functional similarity between sister neurons (Supplementary Fig. 3). However, CBX injection also caused an overall reduction in the percentage of visually driven neurons (225/1121, 20%). This suggests that gap junctions may play an important role in early postnatal development of normal visual responses, although with systemic application of CBX it is difficult to rule out its potential non-specific effects on cortical development.

Intracortical excitatory connections are highly non-random[25], organizing the neurons into fine-scale subnetworks[26]. A recent study showed that sister neurons in the same radial clone are much more connected to each other than to nearby non-sisters[11], suggesting the radial clones as a basis for the subnetwork organization. The high connectivity between sister cells should also contribute to their functional similarity observed in our study. However, the inputs from the sister cells alone are likely to be insufficient to determine the stimulus selectivity, since each neuron receives inputs from ~1,000 other neurons, while each radial clone only consists of tens of neurons[6]. Other factors, such as common inputs to the sister cells, may also play important roles. In mouse V1, layer 2/3 neurons with similar orientation tuning are shown to be preferentially interconnected[10]. A significant fraction of these neurons may be sister cells, exhibiting similar orientation tuning (Fig. 2) and preferential connectivity[11].

Although the columnar structure is long thought to be a fundamental organizational principle of the neocortex, the existence of a basic processing unit has remained controversial[8]. While the anatomical minicolumns observed in adult cortex[27] are believed to arise from ontogenetic columns, the relationship between the functional columns and mini/ontogenetic columns remained speculative[1]. Our results demonstrate a direct correspondence between them in V1, at least in superficial layers where neurons are most orientation selective[28]. Contrary to the notion of random organization, our study shows that orientation tuning is organized in columns even in rodent visual cortex. The fine spatial scale of ontogenetic columns may also explain the extraordinary precision of the orientation map in cat visual cortex[4,29]. The inter-species difference in the macroscopic cortical organization may be due to differences in the horizontal connections between ontogenetic columns[1], which can lead to either a smoothly varying map or apparent salt-and-pepper organization[30]. Thus, our results support the view that the ontogenetic columns, rather than the macroscopic functional columns, constitute the basic units of cortical processing.

METHODS SUMMARY

Retrovirus was injected into the right ventricle of each mouse embryo in utero at E15-17. At P12-17, the injected mice were anaesthetized with urethane (1 g/kg) and chlorprothixene (5 mg/kg), sometimes supplemented with isoflurane (0.5 - 1% in O2). A 1.5 mm-diameter craniotomy was made above V1 for two-photon imaging. After the GFP-expressing cells were identified, nearby layer 2/3 neurons were labelled with Oregon Green 488 BAPTA-1 AM (OGB-1) via bolus loading. For measuring orientation tuning, 8-12 repeats of drifting sinusoidal gratings were presented in 8 directions spanning 0° - 360° in a pseudo-random sequence. Neuronal responses were measured at the two depths where the GFP-expressing neurons were found. To quantify orientation and direction preference, we fitted each measured tuning curve by a double Gaussian function. To block gap junctions, 10-20 mg/kg CBX was injected daily (IP) over the first postnatal week.