Hardwiring the Brain: Endocannabinoids Shape Neuronal Connectivity

See allHide authors and affiliations

Science  25 May 2007:
Vol. 316, Issue 5828, pp. 1212-1216
DOI: 10.1126/science.1137406


The roles of endocannabinoid signaling during central nervous system development are unknown. We report that CB1 cannabinoid receptors (CB1Rs) are enriched in the axonal growth cones of γ-aminobutyric acid–containing (GABAergic) interneurons in the rodent cortex during late gestation. Endocannabinoids trigger CB1R internalization and elimination from filopodia and induce chemorepulsion and collapse of axonal growth cones of these GABAergic interneurons by activating RhoA. Similarly, endocannabinoids diminish the galvanotropism of Xenopus laevis spinal neurons. These findings, together with the impaired target selection of cortical GABAergic interneurons lacking CB1Rs, identify endocannabinoids as axon guidance cues and demonstrate that endocannabinoid signaling regulates synaptogenesis and target selection in vivo.

In the cerebral cortex, information processing requires the precise temporal and spatial coordination of synaptic communication among excitatory pyramidal cells, inhibitory γ-aminobutyric acid–containing (GABAergic) interneurons, and subcortical afferents (1). Cortical neurons are born in progenitor zones that are distant from their final positions, and their layer-specific patterning is achieved through extensive migration in the developing cerebrum (1, 2). En route to their destination, cortical neurons establish their synaptic connectivity patterns (3), thus providing the blueprint for their functional diversification.

In the adult central nervous system (CNS), presynaptic Gi/o protein–coupled CB1 cannabinoid receptors (CB1Rs) (4) are the targets of marijuana (Cannabis spp.)–derived psychoactive phytocannabinoids and of the endocannabinoids anandamide (AEA) and 2-arachidonoylglycerol (2-AG). Endocannabinoids released from post-synaptic neurons serve as retrograde messengers that suppress neurotransmitter release at mature cortical synapses (4). During brain development, CB1Rs are first expressed in early neural progenitors (5), with receptor levels increasing throughout neuronal specification and synaptogenesis (6). Although functionally active CB1Rs are localized to developing axonal projections (6, 7), it remains unknown whether endocannabinoids function as diffusible axon guidance factors before the growth cone differentiates into a presynaptic nerve terminal.

We defined the precise cellular distribution of CB1Rs on neuronal precursors during cortical cell migration, axonal navigation, and synaptogenesis by high-resolution laser-scanning microscopy of genetically tagged neurons, in situ hybridization, and electron microscopy (3, 8). From embryonic day 13.5 (E13.5) until birth, CB1Rs were present on pyramidal cells (Fig. 1, A to C) with their axons coursing in the intermediate zone of the developing cerebral cortex (9) (Fig. 1, D and E) and establishing the fornix pathway in the hippocampus (figs. S1 and S2). These projections primarily contained CB1Rs associated with the surface of distal axon segments, with CB1Rs being largely absent from their growth cones (Fig. 1F). In contrast, CB1Rs were expressed in GABAergic interneurons during late gestation as they were undergoing intracortical tangential or radial migration (Fig. 1G). Here, CB1Rs were preferentially found on axons and axonal growth cones (Fig. 1, H to J) coincident with establishing target-specific synaptic connectivity patterns (3, 10).

Fig. 1.

The temporal and spatial coincidence of CB1R localization with endocannabinoid availability during corticogenesis. (A) Schemes of the telencephalon at the embryonic days indicated. Colored symbols refer to particular structures in adjoining photomicrographs. At E14.5 to 16.5 (fig. S2B), CB1R mRNA is preferentially expressed in pyramidal cells of the hippocampus (B) and cerebral cortex (C), with CB1R immunoreactivity localized to developing long-range axons, coexpressing growth-associated protein 43 (GAP43), in the intermediate zone (9) (D and E). (F) CB1R+ processes, axons emitted by pyramidal neurons, in the fimbria. A three-dimensional reconstruction of a process is depicted in a semitransparent manner. A dotted line encircles the individual profile (1) shown to the right. (2) Adjacent CB1R+ processes (dotted line) with 3,3′-diaminobenzidine (DAB)–Ni reaction end products (black) precipitating on the inner plasmalemmal surface (arrows) after the use of an antibody recognizing the C terminus of CB1R. (G) Hippocampal interneurons (arrows) express CB1R mRNA at E18.5. (H) At birth, CB1Rs are spatially associated with GABAergic axons (arrows) navigating locally in the hippocampus. (I) Reciprocally perpendicular projections of a single CB1R+ growth cone from the hilus of newborn rat hippocampus. Numbers indicate the positions of planar images. Arrowheads indicate the truncated axon. DAB precipitation fills the cytoplasm, which also contains numerous vesicles (arrows). (J) CB1Rs concentrate in growth cone particles (GCPs) relative to total cortical lysates, as shown by Western analysis. (K) DAGLα predominates in the neocortex at birth and (L) is expressed by pyramidal cells. Arrows point to gold particles indicating the precise subcellular localization of DAGLα. (M) Similarly, DAGLβ is expressed by pyramidal cells in the neonatal cortex. (N) A putative GABAergic presynaptic bouton on a DAGLβ+ dendrite is marked by arrows. (O) A CB1R+ GABAergic axon (arrows) is targeted toward a DAGLβ interneuron (*) in the hippocampus at E18.5. (P and Q) NAPE-PLD is first expressed at E18.5 and is preferentially targeted to dendritic spines (arrows) in neocortical pyramids. (R) Some vertically migrating GABAergic interneurons possess NAPE-PLD expression. Arrows denote NAPE-PLD in the leading process. Abbreviations are defined in SOM text. Scale bars, 6 μm in (H), (K), (M) to (O), (Q), and (R); 50 μmin (E); 100 μm in (B) to (D), (G), and (P); and 500 nm in (F), (I), and (L).

CB1R expression is spatially and temporally coordinated with endocannabinoid synthesis during brain development: sn-1-diacylglycerol lipases α and β (DAGLα/β), which synthesize 2-AG, predominated in elongating long-range telencephalic axons at around E14.5 (fig. S2) (11). By E18.5, however, DAGLα/β expression was localized to postsynaptic dendrites of glutamatergic pyramidal cells (Fig. 1, K to N), so that focal 2-AG release could control axonal growth cone steering and the formation of CB1R+ terminals (Fig. 1O). Postsynaptic DAGLα expression exceeded that of DAGLβ in the perinatal neocortex, in agreement with previous observations establishing DAGLα as the major postnatal DAGL isoform (11). Unexpectedly, the expression of N-acyl-phosphatidyl-ethanolamine-selective phospholipase D (NAPE-PLD), an enzyme participating in AEA synthesis (4), was delayed until E18.5, at which age NAPE-PLD was strongly associated with cortical pyramidal cells (Fig. 1, P and Q). GABAergic interneurons did not possess these endocannabinoid synthetic enzymes until their engagement in radial intracortical migration, coincident with the onset of postsynaptic target selection (Fig. 1R) (3).

To identify the functions of endocannabinoid signaling in axonal growth and guidance, we established GABAergic interneuron cultures with high CB1R expression (12). Early growth cones of quiescent axons contained CB1Rs localized to their leading filopodial tips (Fig. 2A). Morphological growth cone differentiation was associated with a gradual enrichment in CB1Rs that were invariably present in motile filopodial tips at the leading edge of the growth cones (Fig. 2, B and C, and movie S1), which are critical for sensing guidance cues (13). ApplicationofAEA for 15 to 30 min induced CB1R translocation from filopodial tips to the central growth cone domain of GABAergic interneurons (Fig. 2, D and D′) and triggered receptor internalization and retrograde transport in hippocampal neurons and PC12 pheochromocytoma cells recombinantly expressing CB1Rs (fig. S3). Agonist-induced activation and trafficking of CB1Rs translated into downstream signaling in cultured interneurons, because AEA (100 nM), like brain-derived neurotrophic factor (BDNF) (14), induced significant phosphorylation of Erk1 and Erk2 (Erk1/2) (15) in the central growth cone domain within 10 min (Fig. 2, E and E′). Studies in the growth cone particulate fraction isolated from embryonic rat cortices corroborated our in vitro findings by showing significant Erk2 phosphorylation that peaked 5 to 10 min after AEA (2 μM) application (Fig. 2F).

Fig. 2.

Agonist stimulation induces CB1R removal from filopodia and Erk1/2 phosphorylation in the central growth cone domain. (A to C) CB1Rs are present in axonal growth cones of GABAergic interneurons in vitro. CB1Rs are transported from the initial neurite segment through the neurite stem (arrow-heads) to the growth cone, where they concentrate in filopodial tips (arrows) (movie S1). The structural classification of growth cones was based on whether microtubules in the central domain were bundled, spread, or looped (29). (D and D′) AEA (100 nM) induces CB1R removal from filopodia. *P < 0.05, n = 13 to 15 filopodia per group from two independent experiments. Numbers indicate the periods of AEA exposure, in minutes. (E) CB1R activation induces Erk1/2 phosphorylation (pERK) in the central growth cone domain. (E′) Quantitative analysis of Erk1/2 phosphorylation 10 min after stimulation. oP <0.01, *P <0.05 (n = 10 growth cones per group). BDNF (100 ng/ml) was used as a positive control (14). (F) AEA induces Erk1/2 phosphorylation in growth cones isolated from embryonic rat cortices. β-III tubulin served as a loading control. Error bars represent SEM. Scale bars, 3 μm in (A) to (C), 2.5 μmin(E), and 1 μmin(D).

Direct involvement of chemotropic (endo) cannabinoid signaling in growth cone guidance was tested by assaying the growth cone turning responses of GABAergic interneurons induced by the synthetic CB1R agonist WIN55,212-2 (1618). A microgradient of BDNF, an established chemoattractive factor for GABAergic interneurons (12), induced attractive turning (16), whereas WIN55,212-2 at a concentration of 20 μM in the micropipette and ∼200 nM at the growth cone (16) elicited growth cone repulsion (Fig. 3, A and B, and table S1). In n = 13 out of 20 motile growth cones (65%), WIN55,212-2 induced growth cone collapse and neurite retraction within the 60 min of recording (Fig. 3, A and B, and table S1). The lack of directional growth cone turning or neurite retraction in a control [dimethyl sulfoxide (DMSO)] gradient excluded recording artefacts produced by positive ejection pressure or the vehicle solution itself. Coapplication of the CB1R antagonist AM251 (1 μM simultaneously in the pipette and bath solution) converted WIN55,212-2-induced chemorepulsion to attractive growth cone turning without significantly affecting the rate of neurite extension (Fig. 3B and table S1). Together with the lack of detectable CB2 cannabinoid receptor expression in cultured GABAergic interneurons (12), these findings indicate that WIN55,212-2-induced chemorepulsion is mediated by CB1Rs.

Fig. 3.

CB1R activation induces ROCK-dependent growth cone repulsion. (A) Neurites of cultured rodent GABAergic interneurons before and after 60 min of drug application. The microgradient direction is indicated by arrows. Arrowheads identify the neurites studied. Scatter plots show individual growth cone turning responses. Negative values represent neurite retraction. (B) Histograms of averaged growth cone turning responses in control and after drug treatments. Bath concentrations of the CB1R antagonist AM251 and the ROCK-selective inhibitor Y-27632 were 1 μM and 50 μM, respectively (18). (C) Growth cones of X. laevis spinal neurons contain CB1Rs. (D) Composite drawings of individual neurons at 3 hours were made by superimposing cell bodies at the colored dot, and the path of each neurite was traced. AEA inhibits growth cone turning toward the cathode in the DC-EF. (E) Time course of EF-induced growth cone turning. WIN55,212-2 (50 nM, solid blue circles) and AEA (100 nM, solid red circles) significantly inhibit cathodal growth cone steering over 5 hours in a continuous DC-EF, whereas growth cones in corresponding vehicle control cultures (open blue and red circles) turn increasingly toward the cathode. (F) The angle and frequency of cathodal growth cone turning (>10°) are each inhibited by both WIN55,212-2 and AEA but not by WIN55,212-3 after 3 hours. Negative angles indicate steering toward the cathode. Data represent means ± SEM. Sample sizes are indicated in parentheses. The statistical analysis is referred to in table S1. Scale bars, 4 μmin(A), 10 μmin(C), and 100 μmin(D).

Next, we analyzed whether AEA and WIN55,212-2 modify directional growth cone steering of Xenopus laevis spinal neurons, which also express CB1Rs (Fig. 3C). Growth cones steadily turned toward the cathode of a direct current (DC) electric field (EF) of 150 mV/mm (Fig. 3, D and E), a gradient that mimics the ∼400 mV/mm DC-EF present naturally in the developing Xenopus neural tube (19) [supporting online material (SOM) text and fig. S4]. Bath application of either AEA or WIN55,212-2, but not its inactive stereoisomer WIN55,212-3, diminished both the mean angle turned toward the cathode and the frequency of cathodal turning in a dose-dependent fashion (Fig. 3, E and F, and table S1). Reminiscent of the ability of WIN55,212-2 to induce chemorepulsion (Fig. 3A), AEA or WIN55,212-2, but not WIN55,212-3, increased the frequency of EF-induced cathodal repulsion (anodal attraction) relative to vehicle controls (Fig. 3F).

Dynamic polarization of the growth cone cytoskeleton underlies chemotropism and galvanotropism: Extending microfilaments predominate on the side of the growth cone nearest to a chemoattractant source, whereas localized collapse of the actin cytoskeleton occurs near a repellent stimulus (18, 19). Cytoskeletal integrity in axonal growth cones is controlled by members of the Rho family of small guanosine triphosphatases (GTPases) (19, 20), which act as molecular switches that transduce extracellular stimuli to the actin cytoskeleton (18). Accordingly, WIN55,212-2 (2 μM) treatment selectively increased the GTP-bound active state of RhoA but not cumulative RhoA, -B, and -C activity (20) in primary cortical cultures after 5 min of stimulation (Fig. 4A). The involvement of CB1Rs in this process was confirmed by the lack of RhoA activation in the presence of AM251 (2 μM; Fig. 4B). AEA (2 μM) induced RhoA activation in a manner similar to that of WIN55,212-2 (Fig. 4B). Spatially restricted activation of RhoA in the collapsing growth cone is associated with filopodial retraction and growth cone repulsion in response to chemical and electrical extracellular cues (18, 19) through activation of the serine-threonine kinase Rho kinase (ROCK) and subsequent phosphorylation of myosin light chains (19, 21) (fig. S5). Pretreating interneuron cultures with the ROCK-selective inhibitor Y-27632 (50 μM) (18) abolished WIN55,212-2-induced neurite retraction and converted CB1R-mediated neurite repulsion into chemoattraction without significantly affecting the rate of neurite extension (Fig. 3, A and B, and table S1). The agonist-induced coupling of CB1Rs to RhoA is consistent with the concept that Rho family GTPases control growth cone integrity by stimulating contraction of the actin cytoskeleton on the side facing the repellent gradient, thus leading to growth cone collapse (19, 22). The conversion of growth cone steering decisions upon ROCK inhibition suggests that CB1Rs also activate the Cdc42 or Rac pathways, primary transducers of BDNF-induced chemoattraction (18, 20), whose spatial and temporal antagonism with RhoA may be sufficient to switch (endo)cannabinoid-induced chemorepulsion to attractive turning (18).

Fig. 4.

The physiological importance of CB1R-mediated growth cone repulsion. (A) Time course of WIN55,212-2 (2 μM)–induced Rho GTPase activation (n = 3 cortical cultures per data point from a representative experiment; mean ± SEM). (B) Both AEA and WIN55,212-2 activate RhoA through CB1Rs, because this response is inhibited by pretreatment with AM251, a CB1R antagonist. Sample sizes are given in parentheses (four independent experiments; mean ± SEM). **P = 0.008, *P = 0.023, oP = 0.014. (C) In layer (L) 2/3 of the mouse somatosensory cortex, VGLUT3 labels inhibitory terminals of GABAergic basket cells (25) coexpressing CB1Rs and VGAT (arrows). Asterisk denotes the pyramidal cell soma. (D) In CB1Rf/f;Dlx5/6-Cre mice (23), the lack of CB1Rs in GABAergic interneurons is accompanied by the redistribution of VGLUT3-containing inhibitory afferents in L2/3. (E and E′) Cre recombinase–mediated CB1R knockout leads to impaired postsynaptic target selection in CB1Rf/f;Dlx5/6-Cre mice, as indicated by the altered distribution and density of VGLUT3+/VGAT+-labeled perisomatic terminals on L2/3 pyramidal cells. Data are means ± SEM, **P = 0.024, *P = 0.030 (n = 5 mice per genotype, ≥32 cells per animal). Scale bars, 16 μmin(C) and35 μmin(D).

We addressed the in vivo significance of the above findings in adult mice (4.5 to 6 months of age) lacking endocannabinoid-mediated retrograde signaling at cortical inhibitory synapses (23) because of conditional CB1R deletion in GABAergic neurons by Cre-mediated recombination redirected by intergenic regulatory sequences of the genes Dlx5 and Dlx6 (CB1Rf/f; Dlx5/6-Cre mice) (24). We identified perisomatic GABAergic terminals that would otherwise have expressed CB1Rs in the neocortex and hippocampus of CB1Rf/f; Dlx5/6-Cre mice by their coexpression of the vesicular GABA (VGAT) and vesicular glutamate 3 (VGLUT3) transporters (25, 26) (Fig. 4C and fig. S6). Analysis of the distribution of VGAT+/VGLUT3+ boutons in layer 2/3 of the neocortex revealed a significant increase in the probability of pyramidal cells receiving VGAT+/VGLUT3+ inputs (Fig. 4, D to E′), indicating impaired postsynaptic target selection of cortical interneurons lacking CB1R-mediated endocannabinoid signals. These changes occurred in the absence of altered interneuron migration or neurochemical specification (fig. S7).

Our findings outline an essential developmental role for endocannabinoid signaling in growth cone steering decisions, identifying endocannabinoids as a class of signaling molecules that regulate axon guidance. This function is evident in diverse CB1R-expressing neuron populations as demonstrated in vitro by growth cone turning assays on CB1R-expressing GABAergic interneurons in rodents and Xenopus spinal cord neurons. Our evidence that CB1R-mediated endocannabinoid signals underpin growth cone steering by chemical and electrical extracellular directional cues, both of which are present in the embryonic CNS, suggests that endocannabinoid signals play an unexpectedly fundamental role in axonal pathfinding and neuritogenesis. The coincidence of intrinsic endocannabinoid synthesis in elongating long-range and GABAergic axons demonstrates that autocrine endocannabinoid signaling (11) contributes to the initial growth of axons, whereas target-derived endocannabinoid signals control axonal navigation and positioning. Our conclusion is also supported by neuroanatomical findings showing endocannabinoid synthetic enzymes in cortical neurons coincident with the enrichment of CB1Rs in neuronal growth cones during the critical period of postsynaptic target selection. Genetic evidence in CB1Rf/f;Dlx5/6-Cre mice (24) reveals that interneuron-specific deletion of CB1R-mediated endocannabinoid signaling is sufficient to relieve the restriction of axonal elongation and branching (12), synaptogenesis (27), and target innervation mediated by endocannabinoid signals in neurochemically identified cortical interneurons. The present study also expands our understanding of how prenatal exposure to the CB1R agonist Δ9-tetrahydrocannabinol (THC), the major psychoactive constituent in cannabis, could affect CNS development and induce cognitive and behavioral deficits enduring into adolescence of THC-exposed offspring (28). Thus, our data, together with THC-induced impairment of synapse formation (27), imply that maternal marijuana consumption may affect neurodevelopment through sustained CB1R activation and disrupt the proper positioning, postsynaptic target selectivity, and functional differentiation of developing axons.

Supporting Online Material

Materials and Methods

SOM Text

Figs. S1 to S7

Table S1


Movie S1

References and Notes

View Abstract

Stay Connected to Science

Navigate This Article