Excitatory input from the cortex and thalamus enters the basal ganglia (BG) through the striatum, where it is locally processed and transformed into two inhibitory, GABAergic outputs called the direct and indirect pathways7
. Each pathway arises from a distinct class of spatially intermixed medium spiny neurons (MSNs) with differing projections and molecular characteristics8
. These projection patterns suggest opponent effects on BG output: direct pathway MSNs synapse in Substantia Nigra reticulata (SNr), the BG output nucleus, whereas indirect pathway MSNs synapse in Globus Pallidus (GP), which in turn inhibits the SNr9
. Since the SNr provides GABAergic inhibition to the thalamus, which subsequently activates cortex through glutamatergic synapses, the interactions of BG, thalamus and cortex can be simplified as a closed loop, differentially controlled by the direct and indirect pathways (Supplementary Figure 1
). Anatomical evidence supports this model10,11
and the opponent roles of the two pathways on motor behavior have been recently demonstrated in adult rodents12,13
. In addition, in vivo
recordings and circuit tracing indicate that different corticostriatal inputs are processed through segregated, parallel networks6,10,11,14
. Given this organization, establishing correct wiring of the cortex-BG-thalamus circuitry poses a significant developmental challenge, requiring that functional interactions be maintained over polysynaptic networks comprised of mixed inhibitory and excitatory projections.
To investigate how striatal activity contributes to circuit development, we generated mice incapable of releasing GABA from direct or indirect pathway MSNs through conditional knockout of Slc32a1
, which encodes the vesicular GABA transporter (vgat) ()15
. GABAergic neurons lacking vgat are unable to package GABA into synaptic vesicles for release16, 17
. Pathway specificity was conferred by BAC transgenic mice that express Cre recombinase under the control of the type 1a or type 2 dopamine receptors18
. In Drd1a
Cre (D1-Cre) mice, Cre expression in GABAergic neurons is largely limited to direct pathway MSNs, although Cre is also found in non-GABAergic cortical neurons (, Supplementary Figures 2–3
). In Drd2
Cre (D2-Cre) mice, Cre expression is largely limited to indirect pathway MSNs (, Supplementary Figure 2
). Mice with Slc32a1
deletion in direct or indirect pathway MSNs showed opposing locomotor phenotypes and survived until weaning (Supplementary Figure 4
), a time of increasing reliance on reward-based complex motor actions.
Conditional knock out of Slc32a1 in direct and indirect pathway MSNs results in opposing changes to excitatory synapse number
Conditional knock out of Slc32a1 from direct or indirect pathway MSNs abolishes GABAergic output
To test the efficacy of silencing of GABA release we relied on optogenetic analysis of MSN-to-MSN collateral synapses19
. Channelrhodopsin2-mCherry fusion protein (ChR2-mCherry) was expressed in Cre-positive MSNs by injecting an adeno-associated virus (AAV) containing a double floxed inverted ChR2-mCherry transgene into the striatum at P3-5 ()20
. Whole-cell voltage-clamp recordings of ChR2-negative MSNs demonstrated light-evoked GABAergic currents in acute brain slices from P13-18 D1-Cre; Slc32a1f/+
mice (−1477 ± 352 pA, n=20) and D2-Cre; Slc32a1f/+
(−1319 ± 221 pA, n=12), which were absent in littermate Slc32a1f/f
animals (D1-Cre; Slc32a1f/f
: −2.19 ± 1.6 pA, n=7; D2-Cre; Slc32a1f/f
: −9.91 ± 2.4 pA, n=7) (, Supplementary Figure 5
We examined whether GABA release from MSNs is required for cell survival and long-range axonal projections. The proportion of Cre+
cell nuclei was 45–48% in D1-Cre and D2-Cre animals and independent of the number of conditional Slc32a1
alleles (Supplementary Figure 6
). MSNs lacking GABA release form qualitatively normal long-range axonal projections, based on examination of ChR2-mCherry fluorescence in the SNr and GP (SNr, 18–19% and GP, 15–16% fractional area coverage) (Supplementary Figure 7
). These observations confirm that MSNs lacking GABA release are maintained in normal numbers and extend grossly normal axons.
In contrast, silencing GABA release had profound consequences for the number of excitatory synapses received, which are typically associated with dendritic spines21,22
. In the rodent striatum, inhibitory synapse density is relatively constant throughout postnatal development, while excitatory synapse density rises dramatically between P10 and P2123
. Therefore, we restricted analyses to pairs of littermate control and Slc32a1
null mice at P14-15. We identified MSNs of each pathway using ChR2-mCherry expression delivered by either Cre-On or Cre-Off AAVs (Supplementary Figure 8a
. Even within this narrow developmental window, we detected an increase in miniature excitatory postsynaptic current (mEPSC) frequencies, but not amplitudes, among same genotype littermates (frequency, P14: 0.14 ± 0.06 Hz, P15: 0.34 ± 0.1 Hz; amplitude, P14: 14.4 ± 1.0 pA, P15: 13.9 ± 0.4 pA) (Supplementary Figure 8b
). In mice lacking GABA release in direct pathway MSNs, mEPSC frequency was reduced in both direct and indirect pathway MSNs compared to heterozygous sibling controls (direct pathway MSNs, D1-Cre;Slc32a1f/+
: 0.22 ± 0.04 Hz, n=18, D1-Cre Slc32a1f/f
: 0.08 ± 0.02 Hz, n=16; indirect pathway MSNs, D1-Cre Slc32a1f/+
: 0.32 ± 0.07 Hz, n=14, D1-Cre;Slc32a1f/f
: 0.1 ± 0.03 Hz, n=15) (). Conversely, mEPSC frequency was increased in MSNs of mice with output-silenced indirect pathway (direct pathway MSNs, D2-Cre;Slc32a1f/+
: 0.11 ± 0.04 Hz, n=9, D2-Cre;Slc32a1f/f
: 0.28 ± 0.07 Hz, n=6; indirect pathway MSNs, D2-Cre;Slc32a1f/+
: 0.15 Hz ± 0.06, n=5, D2-Cre;Slc32a1f/f
: 0.37 Hz ± 0.04, n=14) (). mEPSC amplitudes were largely unaffected by either manipulation (Supplementary Figure 8c-d
Changes in mEPSC rates were paralleled by alterations in dendritic spine density, indicating concurrent changes in structural and functional correlates of excitatory synapse number. Dendritic spine density of all MSNs was decreased by silencing the direct pathway and increased by silencing the indirect pathway (D1-Cre, direct pathway, 0.84 ± 0.03 spines/µm vs. 0.45 ± 0.05, n=6 and 7; D1-Cre, indirect pathway, 0.68 ± 0.05 spines/µm vs. 0.41 ± 0.02, n=5/group; D2-Cre, indirect pathway, 0.56 ± 0.03 spines/µm vs. 1.02 ± 0.05, n=5 and 6; D2-Cre, direct pathway, 0.41 ± 0.02 spines/µm vs. 0.63 ± 0.08 n=6/group) (). Together these data show that the degree of excitatory innervation of MSNs is determined by striatal output, such that excitatory hypo- or hyper-innervation is triggered by silencing direct or indirect pathway, respectively.
Complementary, pathway-dependent effects on excitatory input could be due to cell-intrinsic differences in direct and indirect pathway MSNs. Alternatively, common activity-dependent wiring rules could be differentially activated by each perturbation of BG output. For example, a net increase in SNr inhibitory output, caused by silencing direct pathway MSNs, could diminish thalamic and cortical activity, decreasing glutamate release and glutamatergic synapse formation in the striatum. In contrast, silencing the indirect pathway could have the converse set of effects (Supplementary Figure 1b-d
). The observation that direct or indirect pathway MSNs show similar perturbations despite selective deletion of Slc32a1
only in MSNs of the Cre-expressing pathway () provides support for the circuit-level model.
To test for cell-autonomous, Slc32a1
-dependent regulation of glutamatergic innervation, we deleted Slc32a1
in a small and sparse subpopulation of striatal neurons by injecting small quantities of AAV encoding Cre-mCherry into D2-GFP;Slc32a1f/f
mice at P0-1 (Supplementary Figure 9a-b
). Analysis of mEPSCs and dendritic spines in neighboring Cre+
MSNs (identified by nuclear mCherry fluorescence) belonging to the direct and indirect pathway (identified by GFP expression) revealed no changes in mEPSC frequency or dendritic spine density (Supplementary Figure 9c-e
). Thus, Slc32a1
-dependent, cell-intrinsic mechanisms are unlikely to contribute to differential effects on glutamatergic innervation seen with silenced direct or indirect pathways.
We examined the consequences of manipulating striatal activity specifically during the period of striatal excitatory synaptogenesis23
using the RASSL (G-protein-coupled r
olely by a s
in direct or indirect pathway MSNs (). hM4D is activated by a blood-brain-barrier permeable molecule clozapine-n-oxide (cno), allowing non-invasive in vivo
manipulation of neural activity. hM4D couples to the Gαi/o
pathway, which reduces action potential firing in many cell types (Supplementary Figure 10
) by activating K+
. AAV carrying the double floxed inverted hM4D-mCherry transgene26
was injected in D1- or D2-Cre mice at P0-1 and, starting at P8, cno (1 mg/kg) or saline were administered subcutaneously 2×/day to littermate pups. Consistent with our hypothesis, animals with large bilateral injections of AAV to dampen activity of many direct or indirect pathway MSNs, respectively, down- or up-regulated mEPSC frequency and dendritic spine density, relative to saline-injected littermates (direct pathway MSNs: frequency, 0.8 ± 0.1 Hz vs. 0.36± 0.07 Hz, n=18 and 25; spine density 0.96 ± 0.04 spines/µm vs. 0.49 ± 0.05 spines/µm, n=5/group; indirect pathway MSNs: frequency, 0.04 ± 0.01 Hz vs. 0.27 ± 0.09 Hz, n=7 and 8; spine density, 0.54 ± 0.05 spines/µm vs. 0.98 ± 0.06 spines/µm, n=5/group) (, Supplementary Figure 12a
). Similar results were observed with the Adora2a-Cre
mouse, based on the adenosine 2A receptor promoter, another Cre driver line for indirect pathway MSNs (Supplementary Figure 11
In vivo, developmentally-restricted postnatal manipulation of activity in direct and indirect pathway MSNs results in opposing changes to excitatory synapse number
In contrast, sparse, unilateral hM4D delivery aimed to minimally impact circuit dynamics, did not alter mEPSC frequency and dendritic spine density in hM4D-expressing neurons relative to uninfected neighboring MSNs of the same pathway (, Supplementary Figure 12b
). These data also indicate that hM4D-mediated perturbation of G-protein-coupled intracellular signaling pathways is not responsible for the changes in excitatory synapses.
Three lines of evidence – analyses of non-manipulated MSNs in pathway specific conditional Slc32a1 knockouts, of sparse Slc32a1 null MSNs, and of the effects of widespread vs. sparse hM4d manipulation of neural activity – support the hypothesis that circuit-level patterns of activity, rather than cell-intrinsic mechanisms, determine the degree of glutamatergic innervation of striatal MSNs. Inherent to this proposed mechanism is the assumption that the BG output modulates corticostriatal or thalamostriatal glutamatergic activity, which, in turn, regulates excitatory synaptogenesis in the striatum. To probe this mechanism, we directly tested (1) the ability of glutamate exposure to drive spinogenesis in the striatum and (2) the effect of manipulating corticostriatal activity on striatal spinogenesis and synaptogenesis.
To test whether glutamate release in the striatum is sufficient to induce spinogenesis, we focally stimulated aspiny segments of P8-11 MSN dendrites with 2-photon laser uncaging of glutamate27
. Sparse viral delivery of Cre was used to activate TdTomato reporter fluorescence in a small subset of MSNs in D2-GFP;Ai14-lsl-TdTomato
mice. 2-photon laser-scanning microscopy was used to monitor dendritic structure during glutamate stimulation (). Stimulation power was calibrated to evoke moderate ~7 pA currents from existing spines (Supplementary Figure 13
). Using protocols that robustly induce synaptogenesis in layer 2/3 pyramidal neurons of developing cortex27
, we observed growth of a new dendritic spine at the glutamate uncaging site in ~50% of trials. The probability of spine growth was similar in direct and indirect pathway MSNs (), was unaffected by blockade of GABAA
receptors (52% and 50% success percentage for GFP+ and GFP− MSNs in control conditions, and 56% and 51%, respectively, in GABAA
antagonist), and matched prior results in cortical pyramidal cells27
. Thus, glutamatergic stimulation is sufficient to drive spinogenesis in the developing striatum.
Corticostriatal activity drives synaptogenesis in MSNs
To test whether the postnatal activity of corticostriatal projections drives synaptogenesis, we used the Rbp4-Cre
mouse line to express hM4D in corticostriatal neurons emanating from deep cortical layers (, left
mice were injected with the double floxed inverted AAV carrying hM4D-mCherry at P0-1 and, starting at P8, cno (1 mg/kg) or saline were administered subcutaneously 2×/day to littermate pups. For MSNs of both pathways assayed at P14-15, mEPSC frequency and spine density were decreased after cno treatment, with no change in mEPSC amplitude (direct pathway MSNs: frequency, 0.66 ± 0.2 Hz vs. 0.25 ± 0.05 Hz, n=12 and 9; spine density 0.60 ± 0.05 spines/µm vs. 0.38 ± 0.03 spines/µm, n=7 and 6; indirect pathway MSNs: frequency, 0.65 ± 0.14 Hz vs. 0.25 ± 0.04 Hz, n=16 and 16; spine density, 0.58 ± 0.03 spines/µm vs. 0.31 ± 0.02 spines/µm, n=7 and 6) (). Since inhibition of Rbp4-Cre+
corticostriatal projections caused no change in locomotion at P14-15 (Supplementary Figure 14
), activity-dependent control of striatal synaptogenesis is not secondary to behavioral changes.
To determine whether activity-dependent changes in striatal synaptogenesis persist into early adulthood, we inhibited the activity of corticostriatally projecting Rbp4-Cre neurons during P8-P15 and examined excitatory innervation in indirect pathway MSNs at P25-28 (). Both mEPSC frequency and spine density were reduced in cno-treated mice, compared to saline injected controls (indirect pathway MSNs: frequency, 0.73 ± 0.12 Hz vs. 0.29 ± 0.07 Hz, n=28 and 26; spine density, 0.83 ± 0.02 spines/µm vs. 0.64 ± 0.02 spines/µm, n=10/group) (). Thus, perturbations of cortex-BG-thalamus circuit activity in early life can have lasting effects into adulthood.
Our data reveal that the balance of activity in direct/indirect pathways dictates postnatal excitatory innervation of the striatum. Since manipulating striatal output alters the structure of its input, activity must act recurrently, through a polysynaptic, multi-stage circuit. The simplest explanation for these findings is that glutamate release from cortical and thalamic axons in the striatum promotes the formation or stabilization of glutamatergic synapses onto MSNs. Perturbations that result in relative activation of the direct pathway compared to the indirect (and hence activate cortex and thalamus), will drive glutamatergic synapse formation in the striatum. Collateral connections among MSNs20
refine BG activity and may contribute to these effects. Such activity-dependent processes may be essential for refining locomotion and reward-based behavior.
These mechanisms are fundamentally distinct from those believed to underlie topographic organization of sensory cortices during postnatal development. In these cortical areas, waves of neural activity pass information about sensory maps from one stage of processing to the next28,29
and these mappings are translated into synaptic connectivity through spike timing-dependent plasticity rules30
. In contrast, striatal inputs from widespread cortical areas show only modest topographic organization. Instead of sensory maps, synaptic networks throughout the cortex-BG-thalamus loop are thought to reflect parallel, segregated loops of mixed inhibitory and excitatory projections. Here we show that activity propagates through cortex-BG-thalamus circuits to specify synaptic networks based on the balanced output of direct and indirect pathways in a manner that can, through positive feedback, select for recurrent closed loops.