Endocannabinoid-dependent formation of columnar axonal projection in the mouse cerebral cortex

Significance

One of the most important morphological properties in the cerebral cortex is the columnar structure, where axonal projections occur precisely within individual vertical columns. However, how such precise axonal projections arise during development is not well known. Previous studies indicate that columnar topographical axonal projections are achieved by selective addition of axons within individual columns without identifying the molecule involved in this process. We show that the endocannabinoid is crucial for shaping columnar axonal projections of layer 4 neurons in the mouse barrel cortex by using mice with global knockout of the synthetic enzyme of the endocannabinoid 2-arachidonoylglycerol, pharmacological activation of type 1 cannabinoid receptor (CB1R), and knockout of CB1R in layer 4 neurons.

Abstract

Columnar structure is one of the most fundamental morphological features of the cerebral cortex and is thought to be the basis of information processing in higher animals. Yet, how such a topographically precise structure is formed is largely unknown. Formation of columnar projection of layer 4 (L4) axons is preceded by thalamocortical formation, in which type 1 cannabinoid receptors (CB1R) play an important role in shaping barrel-specific targeted projection by operating spike timing-dependent plasticity during development (Itami et al., J. Neurosci. 36, 7039–7054 [2016]; Kimura & Itami, J. Neurosci. 39, 3784–3791 [2019]). Right after the formation of thalamocortical projections, CB1Rs start to function at L4 axon terminals (Itami & Kimura, J. Neurosci. 32, 15000–15011 [2012]), which coincides with the timing of columnar shaping of L4 axons. Here, we show that the endocannabinoid 2-arachidonoylglycerol (2-AG) plays a crucial role in columnar shaping. We found that L4 axon projections were less organized until P12 and then became columnar after CB1Rs became functional. By contrast, the columnar organization of L4 axons was collapsed in mice genetically lacking diacylglycerol lipase α, the major enzyme for 2-AG synthesis. Intraperitoneally administered CB1R agonists shortened axon length, whereas knockout of CB1R in L4 neurons impaired columnar projection of their axons. Our results suggest that endocannabinoid signaling is crucial for shaping columnar axonal projection in the cerebral cortex.

Cerebral cortices contain arrays of cortical columns, which are the fundamental units of cortical information processing (1, 2). In the adult somatosensory cortex (S1), the excitatory feedforward relay is mediated by axons of layer 4 (L4) excitatory neurons projecting to L2/3 almost exclusively within the same column, or home barrel column, forming topographically precise columnar projections. Similar topographically precise columnar organization is seen throughout the neocortices of higher animals, and thus, columnar projection is thought to be crucial to higher-order functions in the neocortex. However, the underlying mechanism of how such precise projections arise during development has not been intensively studied. Previous studies indicated that columnar topography was achieved by selective addition of axon branches within the home column, namely, “targeted axonal growth” (3, 4) and involvement of serotonin was suggested in a previous study (5). However, its precise mechanism still remains elusive. We previously showed that spike timing-dependent plasticity (STDP) contributes to neural circuit formation and axonal retraction by strengthening and weakening synaptic connections, respectively (6, 7). Endocannabinoid-dependent long-term depression (LTD) of STDP in particular plays a crucial role in the formation of barrel-specific targeting of thalamocortical axons, because it was disrupted in mice with knockout (KO) of type 1 cannabinoid receptor (CB1R) that mediates endocannabinoid signaling in the brain (8). These results indicate that CB1R plays an important role in fine-tuning the position of axonal projections during development (6), in addition to its crucial role in functional map plasticity during critical period (9). Considering that CB1R becomes functional at the L4 terminals at around the time when L4 axons are positioned into columns (3, 10), we hypothesized that CB1R is also involved in the formation and organization of L4 axonal projections.

Results

We first examined whether columnar targeting was impaired in animals in which CB1R was unable to be activated. For this purpose, we used mice that lack the enzyme diacylglycerol lipase α (DGLα) that synthesizes the endogenous cannabinoid 2-arachidonoylglycerol (2-AG) (11). In mice that genetically lack this gene (DGLα^−/−), we found numerous L4 axons extending beyond the home columnar borders (Fig. 1A). We calculated the fraction of axon length outside the home column throughout the layers (L1-L6) and strictly within L1-L3 (see Materials and Methods), which demonstrated that more axons were found outside the home column (about 20% differences) in DGLα^−/− in both L1-L6 (83.6 ± 1.8% vs. 69.2 ± 1.8%, P < 0.0001, Wilcoxon test) and L1-L3 (73.7 ± 2.7% vs. 59.6 ± 2.7%, P = 0.0003), although there were no significant differences in the total axon length (P = 0.159) (Fig. 1B). Previously, we showed that cannabinoid signaling is necessary to confine the thalamocortical axons below the L4-L2/3 border (6). We examined whether this phenotype could be reproduced in DGLα^−/− mice and found that a larger fraction of thalamocortical axons were observed above the L4-L2/3 border in DGLα^−/− mice (P < 0.0001) and outside of the barrels (P < 0.0002, SI Appendix, Figs. S1 A and B and S2), although there were no significant differences in total axon length (P = 0.30) or branch number (P = 0.66, SI Appendix, Fig. S1B). In the previous study, we demonstrated that thalamocortical synapses exhibit STDP with only LTD (all-LTD STDP) with L2/3 cells from the second postnatal week, which underlies the retraction of thalamocortical axons from L2/3 (6). To confirm that timing-dependent LTD at the thalamocortical synapses is impaired in DGLα^−/−, we tested negative timing stimulations (post-before-pre, −15 ms) to thalamus and L2/3 pyramidal neurons and found that timing-dependent LTD could not be induced in DGLα^−/− mice (SI Appendix, Fig. S3 A–D). This result suggests that 2-AG signaling is essential for timing-dependent LTD at the thalamocortical synapses, and this form of LTD might underlie the construction of the columnar projection of L4 axons. One study indicated that L4 axons in immature animals spread across columns more than in mature animals, but gradually become columnar during development to achieve the mature columnar pattern (3). At L4-L2/3 synapses, Hebbian STDP with long-term potentiation (LTP) and LTD, by pre-before-post and post-before-pre timing stimulation, respectively, begins to be induced after P13-P14 (7, 10). We assumed that LTD of Hebbian STDP may play a role in columnar shaping of L4 axons. To address this, we first confirmed that at L4-L2/3 vertical synapses, stimulation with negative (post-before-pre) timing caused LTD in a DGLα-dependent manner. We found that in DGLα^−/−, negative timing stimulation did not induce LTD, unlike in DGLα^+/+ (P < 0.0001, Fig. 2 A–E).

Fig. 1.

Disrupted columnar projection of L4 spiny stellate axons in matured (P19-P23) DGLα^−/− mice. (A) (Top) Confocal image of L4 spiny stellate axons (Left) and its tracing showing layers and barrel columns (Right). Axons are in white (Left) or black (Right), and dendrites are in red. CP, caudate putamen; Hp, hippocampus; WM, white matter. Arrowheads indicate barrel borders. (Middle and Bottom) Example traces of L4 axons (black) and dendrites (red). (B) Quantification of total axon length (Left, 11,234.6 ± 393.0 and 11,218.8 ± 561.2 µm for wild type [^+/+] and knockout [^−/−], respectively; P = 0.159) and topographic distribution with regard to home column (Middle, 83.6 ± 1.8% [n = 31] and 69.2 ± 1.8% [n = 31] for all layers [L1-L6] for ^+/+ and ^−/−, respectively, P < 0.0001; and Right, 73.7 ± 2.7% and 59.6 ± 2.7%, for ^+/+ and ^−/−, respectively, P = 0.0003 for L1-L3). Colored symbols in the graph correspond to individual cells with the same symbols in traces of sample cells in A. NS, not significant. (Scale bars in morphological picture and illustrations, 200 µm.)

Fig. 2.

Impairment of spike timing-dependent LTD in DGLα^−/− mice. (A) Schematic illustration of the experimental procedure. pyra., pyramidal cell. (B and C) Examples of successful induction of LTD-STDP in ^+/+ (B) and failed induction in ^−/− (C). (Right) Sample recordings at baseline (gray) and 20 min. after paring (black for ^+/+ or red for ^−/−). (D) Population data for LTD-STDP. (E) Quantification of LTD-STDP. 68.5 ± 3.7%, 99.7 ± 3.0%, and 100.8 ± 3.7% of control for ^+/+ (n = 18), ^+/− (n = 28), and ^−/− (n = 18). ****P < 0.0001. (F) Experimental design of testing spike timing-dependent plasticity at L4 and the adjacent L2/3 connection. (G) An example of LTD-STDP induced by negative timing stimulation. WT, wild type. (H and I) Population data of LTD-STDP induced by negative timing (60.8 ± 4.9% of control, P < 0.0001, n = 17). (Scale bars in illustrations, 200 µm.)

Next, we tested whether synapses from L4 axons to L2/3 neuron in the adjacent column could exhibit timing-dependent LTD. For this experiment, we used rather young mice, aged between P13-P21, because CB1R starts to function from around P13 (10). We found that negative timing stimulation caused LTD (Fig. 2 F–I) quite reliably at these synapses (Fig. 2 H and I, P < 0.0001, n = 17). As previously demonstrated (12, 13), spiking of L2/3 neurons is preceded by that of L4 neurons due to feedforward GABAergic inhibition by parvalbumin interneuron (PV-GABA) activity in L4 of the same column (14) when driven by thalamic or sensory inputs. This implies that L4 neurons in a given column and L2/3 neurons in the adjacent columns rarely spike sequentially; rather, adjacent L2/3 should be dominantly controlled by L4 cells directly underneath within the same column. Under STDP with LTD, only the strictly sequential pre-before-post spiking within 10∼25 ms leads to strengthening, but uncorrelated spiking leads to weakening (6, 10, 15). These results suggest that L4 axon terminals invading neighboring columns are likely to have difficulty in establishing synaptic connections with the L2/3 cells of those columns.

At L4-L2/3 synapses, CB1R starts to function only from the third postnatal week; before P12, negative timing stimulation fails to induce LTD at this synapse (10). One previous study demonstrated that, at P8-P11, the morphology of L4 axons is immature and not columnar, and adult columnar topography is established during P14-P26 (3). These observations support the idea that endocannabinoid signaling through CB1R is involved in the shaping of the columnar arrangement of L4 axons. Thus, we tested whether CB1R is involved with the fine-tuning of the columnar projection of L4 axons by examining the developmental changes in projection of L4 axons in DGLα^−/− mice from P7 to P23. In accordance with the previous study (3), at P7-P8, L4 axons mainly innervated infragranular layers, with only a few axons in L1-L3 (Fig. 3A and SI Appendix, Fig. S4). Later, the axons gradually turned upward, and more and more axons were found in L1-L3. However, topography varied from cell to cell and was mostly noncolumnar until around P12. Thereafter, projections gradually became more organized, and around P20, the adult columnar organization is established. In DGLα^−/−, although developmental changes were similar up to P12, thereafter they were substantially different from that of DGLα^+/+; there were many cells with noncolumnar projections remaining. For quantification, we measured total axon length, fraction of axon length within the home column across all layers (L1-L6), and that in L1-L3. There were no significant differences in these parameters before P12 between DGLα^+/+ (n = 39) and DGLα^−/− (n = 14) mice (total axon length, P = 0.5653 [Wilcoxon test], fraction of axon length in the home column for L1-L6 and L1-L3, P = 0.3358 and P = 0.3855 [Wilcoxon test], respectively, Fig. 3B, Left). However, during P13-P18, although there was no significant difference in total axon length (P = 0.1871), the fraction of axon length in the home columns was significantly less in DGLα^−/− (n = 19) than in DGLα^+/+ (n = 41) mice, both in L1-L6 (P = 0.0008) and L1-L3 (P = 0.0019) (Fig. 3B, Right). When the developmental time courses of these three parameters were compared, no significant differences were noted in the total axon length between DGLα^+/+ and DGLα^−/− mice (Kolmogorov–Smirnov [K–S] test, P = 0.5603, Fig. 3C, Left), but the fraction of the axon length in the home column was significantly different for L1-L6 (K–S test, P = 0.0002) and for L1-L3 (K–S test, P = 0.0123) between DGLα^+/+ (n = 111) and DGLα^−/− (n = 64) mice (Fig. 3C and SI Appendix, Fig. S5). These results suggest that the shaping of the columnar projection of L4 cells is mediated by endocannabinoid signaling that becomes functional after P13 (10), with possible involvement of CB1R-dependent synaptic weakening between L4 axons and L2/3 cells (10).

Fig. 3.

Development of L4 axon projection in DGLα^+/+ and DGLα^−/− mice. (A) Representative examples of the developmental changes in L4 axon projection in DGLα ^+/+ and DGLα^−/−. (B) Comparison of total axon length (5,311.4 ± 333.7 μm vs. 5,448.4 ± 557.0 μm for DGLα^+/+ [n = 39] and DGLα^−/− [n = 14], respectively, P = 0.5653), percent of axons within home column in L1-L6 (70.2 ± 2.3% vs. 74.5 ± 3.8% for DGLα^+/+ and DGLα^−/−, respectively, P = 0.3358), and that in L1-L3 (53.8 ± 4.7% vs. 61.5 ± 7.8% for DGLα^+/+ and DGLα^−/−, respectively, P = 0.3855) during P7-P12 and P13-P18 (total axon length, 10,509.6 ± 351.9 µm vs. 11,344.4 ± 517.0 µm for DGLα^+/+ [n = 41] and DGLα^−/− [n =19], respectively, P = 0.1871; percent of axons in L1-L6, 78.1% ± 2.1% vs. 65.2% ± 3.0% for DGLα^+/+ and DGLα^−/−, respectively, P = 0.0008 and L1-3, 69.2% ± 2.6% vs. 54.1% ± 3.8% for DGLα ^+/+ and DGLα ^−/−, respectively, P = 0.0019). NS, not significant. (C) Developmental time course of the three parameters. Each point represents a single cell. Shaded areas indicate 90% confidence interval. (Scale bars, 200 µm.)

Our previous study demonstrated that CB1R activation at thalamocortical axon terminals causes both functional weakening (i.e., LTD) and morphological retraction (6). Thus, we asked whether activation of CB1R at L4 terminals also triggers similar morphological remodeling. To this end, we administered CB1R agonist, Δ⁹-tetrahydrocannabinol (THC), or WIN 55,212–2 (WIN) to animals intraperitoneally during either P1-P12 or P13-P18 and examined L4 axon morphology at P12-P14 and P18-P23, respectively. We found that THC administration (3 mg/kg body weight) before P12 did not influence the total axon length (P = 0.635) nor the columnar projection in L1-L6 (P = 0.528) or in L1-L3 (P = 0.931) (Fig. 4 A, C, and D). In contrast, THC or WIN administration during P13-P18 significantly reduced the total axon length (P = 0.0067 [vehicle-THC], P = 0.0002 [vehicle-WIN]), while the CB1R agonist administration induced no significant changes in the fraction of axon length in the home column in L1-L6 (P = 0.353 [vehicle-WIN] and P = 0.664 [vehicle-THC]) or in L1-L3 (P = 0.308 [vehicle-WIN] and P = 0.576 [vehicle-THC]) (Fig. 4 B, E, and F), which is in clear contrast to the results of DGLα^−/− mice at the corresponding ages (SI Appendix, Fig. S6). This result strongly indicates that cannabinoid agonists applied intraperitoneally during P13-P18 causes retraction of L4 axons without any topographical specificity.

Fig. 4.

Shortened L4 axons without changing their topography by CB1R agonists administered after P13. (A and B) Examples of neurobiotin-stained L4 axon images obtained by confocal microphotography after agonists or vehicle administration. (C and E) Schematic illustration showing experimental design and time schedule for administration of drugs and neurobiotin injection. (D and F) Quantification of L4 axon projection after CB1R agonist administration at P1-P12 (D, Left, total axon length, 7,221.7 ± 268.1 µm vs. 6,836.1 ± 599.4 µm for vehicle (n = 8) and THC (n = 13), respectively, P = 0.635; middle, percent of axons in L1-L6, 83.3 ± 3.5% vs. 79.1 ± 4.6% for vehicle and THC, respectively, P = 0.528; Right, percent of axons in L1-L3, 67.5 ± 5.6% vs. 66.7 ± 5.9% for vehicle and THC, respectively, P = 0.931) and at P13-P18 (F, Left, total axon length, 9,457.5 ± 518.8 µm, 7,749.3 ± 406.3 µm, 6,973.4 ± 290.2 µm for vehicle [n = 11], THC [n = 20], WIN [n = 17], P = 0.0067 [vehicle-THC], P = 0.0002 [vehicle-WIN]; Middle, percent of axons in L1-L6, 83.6 ± 3.7%, 81.6 ± 3.0%, 88.1 ± 2.6% for vehicle, THC, WIN, respectively, P = 0.353 [vehicle-WIN] and 0.664 [vehicle-THC]; Right, percent of axons in L1-L3, 77.3 ± 3.8%, 74.2 ± 3.5%, 83.2 ± 3.9% for vehicle, THC, WIN, respectively, P = 0.308 [vehicle-WIN] and P = 0.576 [vehicle-THC]). (Scale bars, 200 µm.)

Finally, we further tested the idea that columnar projection of an L4 spiny stellate cell is shaped by CB1R signaling of its own cell by examining whether columnar shaping of L4 axon projection is disturbed when CB1R is deleted in L4 neurons. For this purpose, we deleted CB1R from the subset of L4 neurons in a Cre-dependent manner by introducing Cre and enhanced green fluorescent protein (EGFP) using in utero electroporation in CB1R-floxed (CB1R^fl/fl) mice (Fig. 5 A–D). We confirmed that our in utero electroporation did not affect the thalamic inputs to S1 cortex (SI Appendix, Fig. S7). Then, at the age of P18-P23, axon morphology was examined in green fluorescent protein (GFP)-positive and -negative L4 spiny stellate cells in the same slices. We found that although GFP-negative cells showed normal columnar projection, those lacking CB1R exhibited noncolumnar projections (Fig. 5A). Similar to the results from DGLα^−/− mice, no significant differences were noted in the total axon length (P = 0.1221) but the fraction of axons in the home column in L1-L6 and in L1-L3 were smaller in CB1R-KO cells than in control cells (P < 0.0001, both, Fig. 5C and SI Appendix, Fig. S8). To further characterize the expansion of lateral distribution of L4 axons, we measured the fraction of axon length within 20 µm vertical stripes aligned from the soma (see Materials and Methods), either throughout all layers (L1-L6) or only in the L1-L3 area. The result indicates that axons of CB1R-KO cells spread more widely both in L1-L3 and in L1-L6 than control cells (P = 0.0054 and 0.0069 for L1-L3 and L1-L6, respectively, K–S test, Fig. 5E). In line with these findings, our data suggest that L4 neurons of DGLα^−/− mice spread their axons more widely than those of DGLα^+/+ mice (SI Appendix, Fig. S9). We found that a significant reduction of axonal lateral distribution between P7-P12 and P13-P18 in DGLα^+/+ mice, although there were no significances between P13-P18 and P19-P23. In contrast, no significant reduction of axonal lateral distribution was observed in DGLα^−/− animals either between P7-P12 and P13-P18 or between P13-P18 and P19-P23 (SI Appendix, Fig. S9). In contrast to the prominent effect on axonal projection, we found that CB1R is not involved in the asymmetrical distribution of dendrites (16–20) (SI Appendix, Fig. S10). These results indicate that CB1R is indispensable for the columnar shaping of axonal projection of L4 neurons.

Fig. 5.

Knockout of CB1R in a subset of L4 spiny stellate cells resulted in cell-specific noncolumnar projection. (A) (Top) Lower magnification of DAPI, neurobiotin staining, GFP fluorescence, and merge of these images. Square regions in merge, a and b are enlarged below (Middle and Bottom, Left) together their tracings on the Right side for two example cells of GFP- (a, CB1R^+/+) and GFP+ (b, CB1R^−/−) with neurobiotin (red). (Middle and Bottom, Right) Examples of Neurolucida images of L4 spiny stellate cells showing axon morphology of control cells (Middle) and that of neighboring CB1R-KO cells (Bottom). CP, caudate putamen; WM, white matter; Hp, hippocampus. (Scale bars, 200 µm.) (B) Schematic illustration showing experimental design and time schedule for in utero electroporation and neurobiotin injection. (C) Quantification of total axon length (Left, 9,246.3± 456.5 µm vs. 8,256.1 ± 438.6 µm for control [n = 39] and KO [n = 36], respectively, P = 0.1221, t test), fraction of axons in home column in L1-L6 (Middle, 77.7 ± 2.4% vs. 64.1 ± 2.3% for control and KO, respectively, P < 0.0001, t test), and that in L1-L3 (Right, 70.8 ± 2.7% vs. 50.8 ± 2.6%, for control and KO, respectively, P < 0.0001, t test). (D) Photomicrographs of immunohistochemical staining using CB1R antibody, showing that CB1R-KO neurites (GFP+) were not stained by CB1R antibody (white). Lower magnification image covering all layers (Left) and higher magnification images (different magnifications) of the squared region (Right). (Scale bars, 100 μm [Left], 50 μm [Right Top], and 10 μm [Right Bottom]). (E) Fractions of axon length within vertical tiles of 20 μm width for L1-L3 (Left) and for L1-L6 (Right) from the initial point of axons (soma) were compared between control (black) and CB1R-KO cells (orange). Mean and SEM (shadow) values are plotted against the distance from the cell body.

Discussion

Previous studies indicate that “targeted growth” or ”precise addition” of neurites within the home column plays the main role in topographic sharpening of L4 axons, but the underlying mechanisms are not known (3, 4). Our findings clearly show that endocannabinoid signaling is crucial for columnar sharpening of L4 axons, because in both DGLα^−/− mice and CB1R-KO L4 cells, L4 axons remained expanded over the column and thus immature.

In other systems, mistargeted neurons or axon branches are removed by apoptosis (21) or eliminated by pruning without cell death (22), both of which can cause reduction of the absolute length of axons. Because the absolute length of mistargeted L4 axons did not decrease with development, apoptosis or pruning was thought unlikely to play a role in the previous study (3). Notably, our results demonstrated that the total axon length of either DGLα-KO or CB1R-KO L4 cells did not differ significantly from that of wild-type mice, although CB1R agonists did stimulate axon terminal retraction or remodeling when applied intraperitoneally (Fig. 4), which is consistent with our previous study (6). CB1R activation has been shown to induce retraction of neurites and axonal growth cones in hippocampal cultures or in cells in the living brain due to the contraction of the actomyosin cytoskeleton (23) or other mechanisms (24, 25). On the other hand, CB1R activation has been reported to cause neurite outgrowth (26, 27) or function as an axon guidance cue as well (28). These results seem to indicate that activation of CB1R causes distinct effects, presumably depending on experimental conditions, such as the nature of target cells, molecules involved in the observed phenomenon, or doses of agonist/antagonist. The diverse effects of CB1R activation may underlie not only the axonal retraction at the column border but also the neurite outgrowth within columns, resulting in little differences in total axon length. An additional possibility might be that the loss of CB1R-dependent LTD shifted the balance toward even more LTP in the home column, stabilizing more axons there in the KOs than in wild type, masking a decrease in axon length.

The requirement of CB1R signaling indicates that neuronal activity is involved in the shaping of columnar projection. However, Bender et al. demonstrated that sensory input is not required to achieve columnar projection from their finding that whisker deprivation did not affect the columnar shaping (3). We also confirmed their conclusion by examining the relationship between whisker deprivation and its effect on axonal arborization (SI Appendix, Fig. S11). How could these two apparently contradictory findings be reconciled? One possibility might be that spontaneous activity is sufficient for shaping the columnar projection. We showed that L4 terminals invading L2/3 of the adjacent columns could exhibit LTD-STDP. This type of STDP, even if it is accompanied with LTP component resulting from pre-before-post firing order, leads to LTD when the pre- and postsynaptic cells are activated randomly, as shown in previous studies (6, 10, 15). This is because the timing delay for causing potentiation (up to 25 ms) is much shorter than that for causing weakening (up to 200 ms). Thus, random firings of pre- and postsynaptic cells produce weakening stochastically (6, 10, 15). The role of spontaneous activity in topological target segregation is known in retinogeniculate projection during development (29). There are also other studies showing that patterned visual experience is not necessary for the development of either ocular dominance or orientation maps (30, 31), indicating that the instructive nature of neural activity can take the form of patterns of spontaneous activity during development (32).

Materials and Methods

Experiments were approved by the animal care and use committee of Saitama Medical University, and conducted according to the institutional guidelines of the care and handlings for the experimental animals of the Saitama Medical University, Indiana University, Niigata University, The University of Tokyo and Osaka University Graduate School of Medicine, and the Japan Neuroscience Society.

Animals.

DGLα-KO (DGLα^−/−) in a C57BL/6 background were generated as described previously (11). CB1R floxed mice (CB1R^fl/fl) in a C57BL/6 background were generated as described previously (33). Experiments shown in Figs. 2 F–I and 4 were performed in C57BL/6 mice. All experiments including electrophysiology and morphological analysis were performed without informing the experimenter of the genotype.

Slice Preparation.

Mice were deeply anesthetized with isoflurane (Pfizer) and decapitated. The brain was rapidly transferred to ice-cold slicing artificial cerebrospinal fluid (ACSF) consisting of (mM): 124 NaCl, 3 KCl, 1.2 NaH₂PO₄, 1.3 MgSO₄, 2 CaCl₂, 10 glucose, 26 NaHCO₃, bubbled with 95% O₂–5% CO₂, and balanced at pH 7.4 (295–305 mOsm). Thalamocortical slices (350–400 µm) were prepared from C57BL/6 mice and mutants at P7–P24 using a rotor slicer or vibratome, as described previously (34, 35). Slices were immediately transferred to a holding chamber where they remained submerged in oxygenated ACSF for at least 1 h.

Electrophysiology.

Slices were transferred to a recording chamber that was placed on the stage (RC-40LP, Warner Instrument Corp.) of an upright fixed-stage microscope (BX51WI, Olympus or Scientifica Patch Pro, Scientifica). Whole-cell patch pipettes (5–7 MΩ) were used to record the membrane voltage from visually identified cells in L4 or L2/3 under differential interference optics. The temperature was maintained at 27–30 °C using TC-344 Temperature Controller (Warner). Micropipettes were pulled from borosilicate, thick-walled glass capillary tubing (Sutter Instruments). The pipette solution contained (mM): 130 potassium methane sulfonate, 10 KCl, 10 Hepes, 0.5 potassium ethylene glycol tetraacetic acid, 5 magnesium adenosine triphosphate, 1 sodium guanosine-5′-triphosphate, 10 sodium phosphocreatinine, pH 7.3 (295 mOsm). Responses were recorded using a MultiClamp 700A or 700B amplifier (Molecular Devices) in current-clamp mode. Signals were low-pass filtered at 3–5 kHz, digitally sampled at 10–20 kHz, and monitored with the pCLAMP software (Molecular Devices). Upon inserting the electrode in the bath, the stray pipette capacitance was compensated, as was the bridge balance, through a built-in circuit of the amplifier. In recordings from L4 barrel neurons, we focused on excitatory spiny stellate cells. For this purpose, recordings were selectively obtained from nonpyramidal neurons. To distinguish either excitatory or inhibitory cells, an increasing current of square pulses, typically of 100–500 ms duration, were injected to test the firing patterns, as described previously (10, 14). Exclusion of pyramidal neurons after reconstruction of cell morphology following neurobiotin injection further confirmed the cell type. In experiments for testing STDP (Fig. 2 A–I and SI Appendix, Fig. S3 A–D) at L4-L2/3 or thalamus-L2/3 synapses, recordings were obtained from visually identified L2/3 pyramidal cells. Tests of the firing patterns in response to square current injections were also performed for confirmation of the cell type. For afferent stimulation, a concentric bipolar stimulating electrode (Frederick Haer & Co) was placed in L4 or in the ventrobasal area (VB) of the thalamus, through which electrical stimuli consisting of square pulses for 100 μs, up to 0.5 mA, were applied every 7,500 ms (0.133 Hz), unless otherwise specified. Upon establishing whole-cell recording of synaptic responses in response to afferent stimulation, their monosynaptic nature was examined by short (typically < 4 ms from VB, depending on age, see Kimura et al. (14) for details) constant latency, following 1 Hz afferent stimulations without failure or latency jittering. To avoid responses resulting from antidromic activation of corticothalamic axons, only excitatory postsynaptic potentials (EPSPs) with paired pulse depression that displayed no supernormality were adopted (14, 36).

Induction Protocol for STDP.

Single postsynaptic action potentials were evoked by somatic current injection using the smallest current possible (typically < 1.5 nA for 5–10 ms). The stimulus intensity to presynaptic cells, either in L4 or in the thalamus, was also adjusted to evoke EPSPs with a single component, without notches in the rising or decaying phases. The pairing interval was defined from the onset of EPSPs to the peak of each action potential. A total of 90 pairings was applied to presynaptic and postsynaptic cells with fixed delays. Stimulation patterns were constructed using a custom program written with LabVIEW (National Instruments) running on a Windows computer and delivered through an interface (USBX-I16P, Technowave) connected to the stimulator. For quantification of LTD, we measured the amplitude of EPSPs at 20 min after the end of induction protocol.

Staining Thalamocortical Afferents with DiI Crystals.

For morphological characterization of the thalamocortical afferents, living thalamocortical slices (300–400 μm) were prepared from P7 DGLα^+/+ and DGLα^−/− mice in exactly the same manner as the slices for electrophysiology. Slices were kept in a holding chamber in oxygenated ACSF. One or two tiny crystals of DiI (Molecular Probes) were inserted into the VB region of the thalamus with a 30-gauge needle under a dissection microscope. The slices containing DiI crystals were fixed by immersion into 4% paraformaldehyde (PFA) containing 4% sucrose in phosphate-buffered saline (PBS) at 4 °C for 1–2 d (overnight). Slices were moved to PBS with 0.01% NaN₃ and kept in the dark for 4–8 wk at room temperature for dye diffusion. To identify the L4 barrel pattern and boundary, DAPI (Invitrogen) staining (1:10,000) was also performed for 1 h at room temperature.

Stained axons were carefully checked, and only those running deep from and in parallel with the section surface were selected to avoid incomplete representation of axonal arbors by the slicing process. The arborization patterns of identifiable single DiI-labeled axons were acquired in stacks of 0.5–28-μm-interval images using confocal laser scanning microscope (FV1200, Olympus) 4×, 10×, and 20× objectives. Single thalamocortical axons were identified and reconstructed above the layer L5/L6 boundary, and all the arbors found within layers L1–L5 were traced using Neuron J or Neurolucida 360 (MBF Bioscience) software.

Neurobiotin Injection.

For morphological examination of L4 axons, 0.5∼1% of neurobiotin (Vector Laboratories) was included in the pipette solution. After whole-cell recordings, the slices containing neurobiotin-loaded cells were fixed by immersion in 4% PFA in 0.2 M phosphate buffer (PB) for 1 h at room temperature, followed by incubation in PB and 5% Triton-X100 (Sigma-Aldrich) for 2 d. Then, slices were incubated with streptavidin-fluorescein (Alexa Fluor 488 conjugate, Thermo Fisher Scientific, or DyLight 549, Vector Laboratories). After three rinses in PBS, the slices were embedded with ProLong Diamond Antifade Mountant (Invitrogen, Thermo Fisher Scientific), coverslipped, and sealed with Paper Bond (Kokuyo). DAPI staining was also performed for identification of the barrel boundary as described above.

Drugs Injection.

WIN 55, 212–2 mesylate (WIN) was obtained from Tocris Bioscience. A stock solution of WIN was made at a concentration of 5 mg/mL with ethanol, which was diluted (2%) by a solvent solution (5% Kolliphor EL, Sigma-Aldrich, in 0.9% NaCl) to a final concentration of 5 mg/kg. Δ⁹-THC was obtained from the National Institute on Drug Abuse Drug Supply Program and provided by Ken Mackie. A stock solution of Δ⁹ -THC was made at the concentration of 100 mg/mL with ethanol, which was diluted (0.3%) by 1.7% ethanol and 98% of solvent solution as above to a final concentration of 3 mg/kg. Vehicle solution was composed of 2% ethanol and 98% solvent solution. Injection solutions were made immediately before use, vortexed quickly, then administered. All stock solutions were kept at −80 °C.

Reconstruction and Quantification of Axonal Morphology.

The axonal arborization of neurons, which were examined to determine whether staining was consistently intense along the axon length without obvious truncation of processes in the slice surface (xy plane), were selected as sample candidate. Of these, only those that showed consistent staining in depth direction (z axis) were selected (SI Appendix, Fig. S12), and confocal images of these neurons were acquired in stacks of 0.5–27-µm intervals under microscopy (4×, 10×, 20× objectives; FV1200, Olympus or 5×, 10×, 20×; LSM 710, Zeiss or 10×, 25× objectives; TCS SP8, Leica) and reconstructed using Neurolucida 360 software. Barrel outlines were identified by DAPI staining as described above. Barrel column boundaries were defined by parallel lines bisecting the septa on either side of the barrel containing the neurobiotin-filled cells or the home barrel column. Axon length and spatial distribution covering all layers (L1-L6) or L1-L3 in the entire slice and within the home column were calculated using the custom-made analysis programs based on MATLAB (Mathworks). Then, the percent of axon segments located in the home column divided by whole area was calculated both in L1-L6 and L1-L3.

To determine the tangential distribution of axons, both L1-L6 and L1-L3 regions were divided into 20 µm radially oriented columns centered with the initial point of the axon. Axonal length was calculated within each 20 µm column. The resulting distributions were normalized by the total axon length in L1-L6 or L1-L3. The mean slope of the distribution for both directions (medial and lateral) was calculated by averaging across both flanks of the distributions. Differences in the shape of the spatial distributions of the axons were tested using K–S statistics.

In Utero Electroporation.

The pCAG-Cre vector (Addgene plasmid # 13775; a gift from Connie Cepko) and AAV pCAG-FLEX-EGFP-WPRE (Addgene plasmid # 51502; a gift from Hongkui Zeng) were introduced into L4 neurons of the somatosensory cortex in CB1R floxed mice (CB1R^fl/fl) using in utero electroporation as previously described (37). Briefly, pregnant mice at embryonic day (E)13.5 were deeply anesthetized. Electric pulses (40 V for 50 ms, five times at 950 ms intervals) were delivered via forceps-shaped electrodes (CUY650P2 or CUY650P3, Unique Medical Imada) connected to an electroporator (CUY21, Nepa Gene). Plasmids were dissolved in water (1 µg/µL).

Immunohistochemistry following Single-Cell CB1R-KO Experiments.

Slice preparation, electrophysiology with neurobiotin injection, and subsequent fixation and permeabilization were performed in the same manner as described above. Additional immunohistochemical staining was performed for confirming the lack of CB1R, identification of barrel boundary, and GFP expression. Nonspecific binding was blocked with 10% donkey serum. The following antibodies were applied overnight at 4 °C: CB1R (rabbit, 1:200, Frontier Institute), GFP (rat, 1:1,000, Nacalai Tesque), NeuN (mouse, 1:1,000, MilliporeSigma), and VGluT2 (guinea pig, 1:200, MilliporeSigma). For identification of the barrel boundary, immunohistochemical staining of VGluT2, NeuN, or DAPI (1:10000, MilliporeSigma) was used. The following fluorophore-conjugated secondary antibodies were used: Alexa Fluor 405 (rabbit, Jackson, 1:1,000), Alexa Fluor 488 (rat, Jackson, 1:1,000), or Alexa Fluor 647 (mouse, Jackson; 1:1,000); the immunolabeled sections were washed and then examined under a confocal laser scanning microscope (FV1200, Olympus).

Whisker Deprivation and Cytochrome-Oxidase Staining.

Beginning at P12, selected rows C and D of whiskers were plucked from the right side of the face under isoflurane anesthesia. Deprivation was maintained by plucking these whiskers every other day until electrophysiological recordings using a neurobiotin-containing micropipette were made on P18-P23. Following the electrophysiology, slices were fixed and axonal morphology was examined as described above. Additional fresh sections were stained for cytochrome-oxidase activity to confirm correspondence between individual whiskers and recorded barrels in the slices. For this purpose, sections were incubated in a solution containing 0.01% cytochrome C (Sigma-Aldrich) and 0.05% 3,3′-diaminobenzidine (Sigma-Aldrich) in PB (pH 7.4) for 2 h at 37 °C. For identification of the barrel boundary, immunohistochemical staining of VGluT2 (guinea pig, 1:200, MilliporeSigma) and DAPI (1:10000, MilliporeSigma) were used. For the secondary antibodies, Alexa Fluor 488 (mouse, Jackson, 1:1,000) and Alexa Fluor 647 (guinea pig, Jackson; 1:1,000) were used. Immunolabeled sections were washed and then examined under a confocal laser scanning microscope (LSM710, Zeiss).

Statistical Analysis.

All data were acquired blindly and tested for normal distribution using a Shapiro-Wilk test. Significance was determined using parametric Student’s t test for normal distribution unless noted otherwise, or nonparametric Wilcoxon rank sum test was used for others. All values are expressed as mean ± SEM.

Data, Materials, and Software Availability

All study data are included in the article and/or supporting information.