Striatal local circuitry: a new framework for lateral inhibition

Heterogeneity of striatal subregions and cell-types

Functional overview

Classical models of basal ganglia function posit that activation of dSPNs disinhibits principal thalamic neurons to facilitate behavior, while activation of iSPNs ultimately inhibits thalamic neurons to suppress motor and reward-based behaviors. Dysregulation in the balance of dSPN and iSPN activity is thought to be the underlying cause of several basal ganglia movement disorders (Albin et al., 1989; DeLong, 1990). This model is supported by experiments using cell specific genetic perturbation, cell ablation, or optogenetic activation (Bateup et al., 2010; Bock et al., 2013; Durieux et al., 2009; Kravitz et al., 2012, 2010; Lobo et al., 2010; Tai et al., 2012).

Glutamate inputs drive SPN firing and, without these excitatory inputs, SPNs are quiescent and their membrane potential is very hyperpolarized. During a desirable action, dopamine is thought to modulate the neuronal excitability to acutely raise the signal-to-noise ratio through activation of G_s/olf coupled D1 receptors (D1Rs) on dSPNs and the G_i/o coupled D2 receptors (D2Rs) on iSPNs. Also, by triggering synaptic plasticity and inducing changes in gene expression, dopamine biases future behavior toward repeating desirable actions, mediating processes that underlie learning (Alexander and Crutcher, 1990; Cohen and Frank, 2009; Gerfen and Surmeier, 2011; Surmeier et al., 2014; Tritsch and Sabatini, 2012).

There is accumulating evidence for a more nuanced view of the interaction between the direct and indirect pathways in driving behavior. Recent work in the dorsal striatum using calcium imaging with fiber photometry (Cui et al., 2014; Tecuapetla et al., 2014), deep brain calcium imaging (Barbera et al., 2016), and in vivo electrophysiology in which SPN subtypes were identified with optogenetic tagging (Jin et al., 2014) or post hoc in situ hybridization (Isomura et al., 2013), have all demonstrated that both dSPNs and iSPNs are more active when an animal is performing a task or freely moving and less active when the animal is not moving or engaged in a task (though dSPNs and iSPNs have differing activity patterns). These data imply that iSPNs and dSPNs are coactive during action initiation and that the direct and indirect pathways are both necessary to facilitate actions. Further support for this idea comes from studies showing that inhibition of either iSPNs or dSPNs can disrupt action initiation or ongoing behavior (Tecuapetla et al., 2016, 2014).

Extending this idea to the ventral striatum, optogenetic inhibition of either iSPN or dSPNs in the NAc after trial start cue in an operant food task reduced motivated responding on the task (Natsubori et al., 2017), and inhibiting synaptic transmission from either iSPNs or dSPNs prevents the acute locomotor response to cocaine and methamphetamine, a behavior known to involve the NAc (Hikida et al., 2010). These studies provide evidence that both dSPNs and iSPNs are more active during actions than at rest, and activity in both SPN subtypes is needed for driving normal motivated behavior.

In addition, fiber photometry recordings to examine the activity of dSPNs and iSPNs in the NAc during an operant food responding task show that both SPN subtypes are more active after a trial start cue and reward delivery than at rest (Natsubori et al., 2017). Furthermore photometry recordings in the NAc during a risky decision task have demonstrated that iSPN activity is highest following a risky choice that led to no reward and that this activity predicts future “safe” choices (Zalocusky et al., 2016). When recording calcium transients in iSPNs during a cocaine conditioned place preference paradigm, iSPNs show marked decreases in their activity after the animal moves into the drug paired side of the chamber, similar to their response to cocaine (Calipari et al., 2016). Three main considerations with fiber photometry experiments are: a) this technique measures calcium signals, an indirect assessment of spikes, and while it is capable to detect single spikes under the right conditions, this is almost rarely verified, b) the time course of the photometry signal is largely determined by properties of the calcium indicator, not the spiking and c) the spatial information is lost when collecting photons through a fiber and as such photometry measures bulk calcium signals that are amplified when there is simultaneous activation. Thus, photometry signals in vivo are most likely a reflection of 1–3 action potentials and the signal detection is biased toward synchronized events that can summate in time. Photometry signals might reflect the activity of ensembles rather than single neurons (Alvarez, 2016). While important to consider these factors, neither of these considerations undermines the main findings that co-activation of iSPNs and dSPNs is seen during a wide range of behaviors. Altogether these studies suggest that both the direct and indirect pathways are necessary to drive behavioral output in both the dorsal and the ventral striatum.

An important question then is how to reconcile the findings from optogenetic stimulation experiments suggesting that iSPN activation inhibits movement (Kravitz et al., 2010) with those from calcium dynamics showing iSPN activation during action initiation and behavior (Barbera et al., 2016; Cui et al., 2014; Natsubori et al., 2017). The most parsimonious explanation is that during behavior, dSPNs facilitate the desired behavior, while iSPNs act simultaneously to inhibit all other competing undesirable behaviors (see more in section 3). Indeed, this concept of the basal ganglia serving as a selection mechanism between competing behavioral programs was proposed over twenty years ago by Jonathon Mink (Mink, 2003, 1996; Mink and Thach, 1993) and Okihide Hikosaka (Hikosaka, 1998, 1991), and it has been incorporated into computational models of basal ganglia function (Collins and Frank, 2014; Frank, 2005). Consistent with this, global optogenetic stimulation of iSPNs would therefore suppress all behavioral programs and inhibit movement.

If this model were correct, precise regulation of the activity of dSPNs and iSPNs that control competing behavioral programs would be required (Tecuapetla et al., 2016), making it important to determine how striatal circuitry can give rise to this coordinated activity. In order for the model to work properly, coordinated activation of dSPNs that facilitate a desired behavior and activation of iSPNs that suppress competing behaviors should be achieved. This coordination could be dictated by the inputs to the striatum, the organization of the local intrastriatal connectivity, or a combination of both. In the next sections, we will first review inputs into the striatum and then discuss the local intrastriatal circuity.

Main inputs to the striatum

The striatum receives two main types of inputs: glutamatergic and dopaminergic. Glutamatergic projections arise from most cortical regions, as well as thalamic and limbic regions. Cortical inputs descend in a topographic fashion with neighboring regions of cortex projecting to adjacent regions of the striatum (Bolam et al., 2000; Dudman and Gerfen, 2015). The DLS receives inputs from sensorimotor cortex, while the DMS is innervated by associative regions of cortex. Both DMS and DLS receive inputs from the thalamus. The NAc is targeted by associative and limbic regions of cortex, as well as amygdala, hippocampal, and thalamic nuclei (Britt et al., 2012; Bubser and Deutch, 1998; MacAskill et al., 2012). Cortical inputs target both subtypes of SPNs, with single axons making synapses onto both dSPNs and iSPNs (Doig et al., 2010). While anatomical studies have suggested that dSPNs and iSPNs receive inputs from different types of cortical neurons, with intertelencephalic (IT) neurons more likely to synapse onto dSPNs and pyramidal tract neurons (PT) more likely to synapse onto iSPNs (Lei et al., 2004), physiological studies have shown that both types of cortical neurons synapse onto both iSPNs and dSPNs (Kress et al., 2013). Retrograde viral tracing methods have suggested that there may be differences in dSPN and iSPN innervation by cortical region (Wall et al., 2013). However, a subsequent study, in which the number of presynaptic neurons labeled outside the striatum was an order of magnitude greater, found no difference in dSPN or iSPN innervation by cortical region (Guo et al., 2015). This suggests that the initial findings of segregated inputs could have suffered from sampling bias and cautions against drawing quantitative conclusions from viral methods with low infection rates.

In the dorsal striatum, both dSPN and iSPNs receive glutamatergic inputs from not only cortex, but also thalamus, and single axons of both types contact both dSPNs and iSPNs (Doig et al., 2010). However, cortical and thalamic synapses differ in their properties: cortical synapses onto SPNs have a much higher failure rate and are facilitating, while thalamic synapses have a lower rate of failure and are depressing (Ding et al., 2008). Synapses from different thalamic nuclei display a great deal of heterogeneity in their strength, short term dynamics, and SPN subtype targets (Ellender et al., 2013; Smith et al., 2009). Furthermore, through their interactions with cholinergic interneurons, thalamic neurons appear to “gate” cortical information flow, suggesting an interesting interaction between these two information streams for striatal processing (Ding et al., 2010).

The striatum is densely innervated by dopamine neurons. Dopamine neurons of the substantia nigra pars compacta target the dorsal striatum, while those of the ventral tegmental area (VTA) target the ventral striatum. Dopamine’s influence on striatal function is vast and complex due to the variety of D1-like (D1, D5) and D2-like (D2, D3, D4) receptors located on both axon terminals and postsynaptic sites in the striatum. In addition to differential effects in directly modulating D1 containing dSPN and D2 containing iSPNs, dopamine powerfully influences striatal processing through its actions on cholinergic and GABAergic interneurons as well as presynaptic terminals of inputs to the striatum. These actions of dopamine on the inputs can be acute and reversible or expressed in the form of long-term changes in synaptic strength (For reviews see (Gerfen and Surmeier, 2011; Kreitzer and Malenka, 2008). For review covering afferents to GABAergic interneurons see (Tepper et al., 2010; Tepper and Koós, 2017)).

It is now known that the midbrain provides another source of glutamate to these regions. The VTA contains a population of glutamate neurons that projects to the striatum (Hnasko et al., 2012; Yamaguchi et al., 2015, 2011, 2007) as well as a population of neurons that co-release dopamine and glutamate (Hnasko et al., 2010; Stuber et al., 2010; Tecuapetla et al., 2010). VTA glutamate neurons projecting to the NAc shell synapse onto FSIs and mediate behavioral aversion (Qi et al., 2016). In addition, it has recently been reported that nondopaminergic glutamate neurons in the VTA can also co-release GABA and that stimulation of these neurons can be transiently reinforcing (Yoo et al., 2016).

Further, dopamine neurons have been shown to co-release GABA along with dopamine in the dorsal striatum (Tritsch et al., 2012) and the NAc, though dopamine neuron IPSCs recorded in the NAc are smaller than in the dorsal striatum (Kim et al., 2015). Interestingly, dopamine neurons do not synthesize GABA through canonical GAD65 and GAD67 enzymes, but rather uptake it through the plasma membrane (Tritsch et al., 2014) and synthesize it through a non-canonical aldehyde dehydrogenase 1a1 pathway (Kim et al., 2015).

Functional GABA projections from cortex have recently been described in both NAc and dorsal striatum. The mPFC sends GABA projections to the NAc shell, and activation of these afferents produces avoidance behavior in a real time place preference paradigm (Lee et al., 2014). In the dorsal striatum cortical GABA projections from layer 5 and 6 SOM+ cells to both dSPNs and iSPNs have recently been described in both auditory and motor cortex, implying that this may be a general feature of cortex (Rock et al., 2016). These findings are in agreement with anatomical studies that identified cortical GABA projections to the striatum (Tomioka et al., 2015) and challenge the common assumption that cortical projections to the striatum are exclusively glutamatergic.

Another large source of GABA to the striatum comes from back projections from the GP. Recently identified arkypallidal, or GP-TA, neurons in the GP exclusively target the striatum with ~10,000 boutons per cell and synapse onto SPNs as well as both cholinergic and GABAergic (PV+ and NOS+) interneurons, suggesting that these cells may provide the largest extrinsic source of GABA to the striatum (Mallet et al., 2012). Functionally, in vivo recordings of these neurons during a behavior task, in which animals must abruptly abort an action, imply that arkypallidal cells may be responsible for sending a “stop” signal to the striatum to cancel evolving action plans (Mallet et al., 2016). In addition to arkypallidal cells, GP neurons that project to downstream structures send GABA projections to the striatum that target PV interneurons (Bevan et al., 1998). VTA contains nondopaminergic GABA neurons that project to the NAc (Van Bockstaele and Pickel, 1995). In addition to locally modulating the excitability of VTA dopamine neurons (Tan et al., 2012; Van Zessen et al., 2012), these neurons project to cholinergic interneurons in the NAc and can pause their firing, which plays a role in associative learning (Brown et al., 2012). Thus, in addition to GABA co-released from midbrain dopamine neurons, GABA projection neurons in midbrain also influence striatal circuits via their projections to CINs.

Another novel inhibitory projection from the bed nucleus of the stria terminalis to the striatum has also been described, terminating on patch and exo-patch SPNs in the dorsal striatum (Smith et al., 2016). Using tract tracing, another study identified an extrinsic source of acetylcholine arising from brainstem nuclei, the pedunculopontine nucleus and the lateral dorsal tegmental nucleus (Dautan et al., 2014). The function of these newly identified projections still needs to be determined. Other minor afferents to the striatum release neuromodulators, such as serotonin from the dorsal raphe (Miguelez et al., 2014) as well as histamine and oxytocin from hypothalamus (Bolam and Ellender, 2016; Dölen et al., 2013).

Axo-axonal connections

In axo-axonal synapses, neurotransmitter release from one synaptic bouton activates receptors located presynaptically on another bouton or axon fiber and act directly to suppress, facilitate, or even trigger neurotransmitter release from the “postsynaptic” axon terminal, sometimes without affecting the firing of action potentials from the postsynaptic neuron (Figure 2). While anatomical evidence for axo-axonal synapses in the striatum is still lagging behind, functional data indicates that presynaptic heteroreceptors can be activated quickly and reliable upon neurotransmitter release from other terminals (Cachope et al., 2012; Kosillo et al., 2016; Threlfell et al., 2012; Shin et al., in revisions). A recent example of an axo-axonal connection that was functionally identified in the striatum is between CINs and dopamine (DA) fibers. Synchronized release of ACh from cholinergic interneurons can activate presynaptic nicotinic acetylcholine receptors (nAChRs) located on DA axons and evoke DA release in both ventral and dorsal striatum (Cachope et al., 2012; Threlfell et al., 2012). Via presynaptic muscarinic M5 receptors on DA axon fibers, ACh can further enhance DA transmission in the striatum (Shin et al., 2015). Through activation of presynaptic muscarinic M2/M4 receptors on glutamatergic terminals from cortex and thalamus, ACh can also suppress glutamate transmission onto SPNs (Ding et al., 2010). GABA transmission from FSIs onto SPNs is also inhibited by ACh through presynaptic muscarinic receptors (Koós and Tepper, 2002). In addition, the existence of nAChRs on glutamatergic terminals has been suggested (Kaiser and Wonnacott, 2000). Further confirmation that CIN stimulation and/or ACh release from these interneurons can produce these modulatory effects on GABA and glutamate transmission will be important for the understanding the spatial and temporal limitation of ACh transmission

Other neurotransmitters act presynaptically in the striatum as well: through activation of presynaptic D2 receptors, DA inhibits its own release from DA terminals through autoreceptors, GABA transmission from iSPNs (Dobbs et al., 2016a; Tecuapetla et al., 2009), and acetylcholine release from CINs. It is proposed to inhibit glutamate transmission from cortical afferents (Bamford et al., 2004), though because cortical D2 receptor expression is sparse, postsynaptic effects of dopamine on glutamate transmission likely play a large role in observed modulation (Higley and Sabatini, 2010). Further, single NPY-NGF interneuron firing can inhibit cortical glutamate release through presynaptic GABA_B receptors, suggesting an axo-axonal synapse (Logie et al., 2013).

Axo-dendritic and axo-somatic connections

Collateral transmission between SPNs

In addition to their long range projection targets, SPNs extend dense axon collaterals that stay within the striatum or nucleus accumbens and target other SPNs (Chang and Kitai, 1985; Pennartz et al., 1991; Somogyi et al., 1981; Wilson and Groves, 1980). These GABAergic synapses are mostly formed on distal dendrites of SPNs (Wilson and Groves, 1980) and are relatively weak compared to FSI inputs that are made at the soma (Czubayko and Plenz, 2002; Koós et al., 2004; Tunstall et al., 2002). The connectivity is sparse in the sense that only ~30% of paired recordings between SPNs produce a synaptic response, and these connections are all unidirectional.

The network of collaterals axons can provide the circuit with a sufficient degree of lateral inhibition to shape the output of the striatum. Indeed, despite the early work showing weak synaptic connections between discrete pairs of SPNs, synchronized stimulation of axon collateral from multiple SPN can provide a strong source of inhibition to other striatal SPNs (Dobbs et al., 2016). The properties and regulation of this network of lateral inhibition are discussed in more detail below. In addition, we present some models for the organization of the lateral inhibition and its implications for circuit level function.

Anatomical and functional evidence

Early anatomical studies identifying SPNs in the dorsal striatum as the projection neurons of this region made note of their dense axons collaterals within the striatum (Grofová, 1975; Kemp and Powell, 1971). Indeed, some of the earliest studies identifying SPNs, including those by Ramon y Cajal (1911), identified them as interneurons due to their intrastriatal axon branches and the limitations of the Golgi technique (Tepper et al., 2010). By combining antidromic stimulation recordings with cell filling and anatomical reconstruction, it became more clear that SPNs have both projections out of the striatum and an extensive collateral plexus within the striatum (Preston et al., 1980). The use of light and electron microscopy in the early 1980s allowed both Wilson and Groves and Somogyi and colleagues to show that SPN axons, in addition to projecting out of the striatum, target other SPNs within the striatum (Somogyi et al., 1981; Wilson and Groves, 1980). As the idea of separate direct/indirect pathway SPN subpopulations began to emerge, Kawaguchi and colleagues demonstrated that SPNs form intrastriatal axon collaterals regardless of the downstream projection target (i.e. whether that cell was an iSPN or dSPN) (Kawaguchi et al., 1990). Electron microscopy was combined with immunocytochemistry to show that iSPNs and dSPNs were synaptically interconnected and synapses were formed between SPNs of different subclasses (Yung et al., 1996). Anatomical studies in the NAc confirmed that similar to the dorsal striatum, NAc SPNs have dense axon collaterals that remain within the accumbens (Chang and Kitai, 1985; Pennartz et al., 1991) in addition to the long range axon projections to the midbrain and pallidum.

Despite the wealth of anatomical evidence, the resulting functional effect of this high density of local synapses between SPNs has been harder to grasp. Electrophysiological studies using antidromic stimulation hinted at early evidence of functional GABAergic synapses from the axon collaterals in both dorsal striatum (Park et al., 1980) and nucleus accumbens (Chang and Kitai, 1985). Despite this encouraging functional evidence, one of the first studies directly addressing the existence of functional synapses using sharp intracellular electrode recordings in pairs of SPNs in the dorsal striatum of rats found no evidence for functional synaptic connections (Jaeger et al., 1994). Another study using paired whole cell recordings between fast-spiking PV interneurons and SPNs found that PV cells form strong GABAergic connections to SPNs (Koós and Tepper, 1999). Together, these findings suggested that PV interneurons provide a larger source of GABAergic inhibition to SPNs than SPNs collaterals do as previous hypotheses based on the anatomical data suggested (Tepper et al., 2008).

In 2002, the first evidence of monosynaptic GABAergic connections between SPNs was demonstrated by Tunstall and colleagues, followed shortly by Czubayko and Plenz (Czubayko and Plenz, 2002; Tunstall et al., 2002). Tunstall and colleagues used sharp intracellular electrode recordings in adult rat striatal slices and found that 9/45 pairs of SPN formed unidirectional synapses that were blocked by GABA_A antagonists. In addition to the sparse connectivity (20% of pairs), the synaptic responses were of small size (IPSP amplitudes of a few hundred microvolts) and showed high rate of failure. As a consequence, averaging hundreds of events was required in order to resolve IPSPs, possibly explaining the discrepancy with the previous study of SPN pairs. Similarly, using young rats and organotypic cultures, Czubayko and Plenz found further evidence for the existence of functional synapses between SPNs. They found synaptic connections between 5/38 (13%) of pairs of SPNs with high failure rates (~0.5) in rat slices, in agreement with the connectivity rates reported by Tunstall et al. IPSPs reported in Czubayko and Plenz were an order of magnitude larger than those reported in Tunstall et al. (~2 mV vs. ~250 µV) likely due to difference the recording configurations (sharp electrode vs whole-cell). Two years later, Taverna and colleagues showed that SPNs in the NAc also form functional collateral synapses with each other (Taverna et al., 2004). Like in the dorsal striatum, these connections are relatively sparse (13/38 pairs, 34%), mediated by GABA_A receptors, and mainly unidirectional in nature.

How can this hypothesized connectivity pattern aid in the processing of inputs?

The impact that this proposed pattern of axon collateral transmission has on the activity of the SPN clusters will have to be determined and tested in future studies. To illustrate the possible ways in which the connectivity patterns in this conceptual model can aid on the processing and filtering of the excitatory inputs to the striatum, we applied a standard conductance-based formalism for the individual neurons and synapses that reflects the main network properties rather than reproducing the underlying biology. In this simple model all cells are identical and receive identical inputs. We consider two patterns of synaptic connectivity and determine how the different network scenarios affect the firing of action potentials of the SPN clusters in the two representative functional units.

In a model without lateral inhibition, each SPN cluster fires action potentials at relatively high frequency in response to excitatory inputs, and the clusters fire in synchrony because the inputs are identical and the cells are homogeneous (Figure 5A). Addition of lateral inhibition with an all-to-all connectivity pattern and equal strength among connections (symmetric connectivity patterns) results in synchronous firing at lower frequency (Figure 5B). Weakening transmission from dSPNs to introduce the experimentally reported asymmetry in the connectivity strength among SPN subclasses, results in non-trivial firing patterns where iSPN clusters become more dominant (Figure 5C). iSPN clusters fire in anti-phase bursts and dSPN clusters fire a single spike at every cycle. While it is interesting to see that functional units are taking turns, the low activity of SPNs clusters will not be conducive to the generation of motor behaviors. Going back to the pattern of Figure 5B, removal of the crossed connections between the functional leaves the originally proposed organization of the lateral inhibition (structured, Figure 5D) which produces synchrony in all clusters, similar to the firing pattern observed in Figure 5B (with slightly lower frequency). This illustrates the different balances that can be operating in the network to produce the same type of patterns. However, when these structured connections are no longer symmetrical between SPN subclasses, the firing patterns can change drastically. As shown in the example of Figure 5E, the asymmetric structured pattern can result in synchronized activation of dSPN and iSPN clusters belonging to different competing functional units where iSPN and dSPN clusters fire in phase, while the other clusters are silent. In other words, some clusters are shut down while others are activated in phase. Note that the excitatory inputs to the silent SPN clusters are still present and have not changed and as such this shows that, at least in this conceptual model, the lateral inhibition can indeed filter excitatory inputs to SPN to shut down the firing of a selected subset of iSPN and dSPN clusters. This example illustrates how the lateral inhibition organized in the proposed “structured” fashion and with the asymmetry in the connections can generate the predicted pattern of co-activation of iSPNs and dSPNs that is observed in vivo and that is conducive to activation of behavior A and inhibition of behavior B. In this way, lateral inhibition can help to striatum in its important task of selecting motor actions.

This model is meant to serve as a proof of concept that the patterning of the lateral inhibition between SPNs would be able to transform synchronous inputs to all cells into selective activation by a subset of SPNs, while silencing others. It is a simplified model with a small number of cells created for illustrative purposes. Thus, it does not capture all known aspects of SPN biology, most importantly, as detailed earlier, the fact that synapses from other SPNs are formed mainly onto the distal dendrites. Dendritic modeling work suggests that inhibition in distal dendrites can exert powerful shunting of excitatory inputs synapsing on dendrites, as glutamate inputs to SPNs tend to do (Gidon and Segev, 2012). However, it is unlikely that a single SPN in vivo is capable of suppressing spikes in other SPNs. In order for the computation illustrated with the model to occur in vivo, there would likely need to be some degree of synchronized activation of axon collaterals, which we have recently demonstrated can shunt action potential firing (Dobbs et al., 2016a). Thus, synchronized firing of ensembles would likely be needed to affect firing. How reasonable is that assumption based on the data? As discussed previously (see “Evidence of striatal functional units” Section 4), in vivo electrophysiology and calcium imaging data indicates evidence for synchronized firing of SPNs during specific aspects of behavior. Though cells in close proximity might not always fire in synchrony (Stern et al., 1998), given the broad reach of SPN axons and dendrites, these units need not consist of cell bodies in close proximity. Indeed, the model predicts that SPNs that are co-active are not connected by collateral synapses and thus co-active SPNs could be spaced further beyond the region of overlap between their axonal arborization and the dendritic field. Considering the sparse, disparate pattern of cells that express immediately early genes during behaviors like context induced cocaine locomotor sensitization and the wide spatial distribution of correlated, task responsive neurons (Bakhurin et al., 2016), it seems likely that the cells in these units would, in fact, not be in close proximity (but note (Barbera et al., 2016).

Another possibility that has been suggested (Koós et al., 2004; Plenz, 2003), is that local collaterals of SPNs are important for controlling transition to depolarized, plateau “upstates”, which prime SPNs for firing, rather than affecting the spiking itself. As SPN collaterals synapse mainly on distal dendrites (Oorschot et al., 2013) , and plateau upstate potentials are generated by regenerative events in the distal dendrites (Plotkin et al., 2011), GABA released from SPN could provide sufficient shunting to prevent this transition. Indeed, in vitro studies show that GABA antagonists prolong SPN upstate duration, suggesting that GABA inputs plays a role in upstate timing (Vergara et al., 2003). By controlling the upstate transitions that precede spiking and bursting, as well as directly shunting excitatory inputs, inhibition in the dendrites would still can provide a powerful means of effecting cell output (Gidon and Segev, 2012; Lovett-Barron et al., 2012). Thus, the local SPN synapses between and within units could be controlling the excitability of neighboring cells in the way proposed by the model, without necessarily directly suppressing spiking.

Open Questions

This proposed model is intended as a preliminary proposition to stimulate discussion and research regarding how local circuitry within the striatum might be organized given known functional and behavioral data. As a result, there are many open questions regarding this framework. How many cells do these units consist of? How much overlap is there between units? How might these units arise? Are they fixed developmentally, completely plastic based on learning, or somewhere in between? Given the known role of the striatum in certain forms of learning and the numerous examples of plasticity in the region, it is likely that any organization in the region is malleable. Dopamine release during behavior can modulate the strength of the lateral inhibition to alter the connectivity pattern in ways that it favors the activation of selected FUs and the inhibition of others to ultimately raise the signal-to-noise ratio and reinforce desirable behaviors. To what degree is this the case, and are certain units more flexible and plastic than others? Furthermore, how would the various interneurons and their numerous known synapses detailed earlier fit within this functional unit framework?

Concluding remarks

Given the growing diversity of striatal inputs, cell-types, and connections described in recent years, it is important to develop a framework for how the circuitry within the striatum is wired to drive processing of information and affect output. Here we present a hypothesis on how one aspect of the striatal circuitry, the lateral inhibition between SPNs, could be organized given available data. It is our hope that this will drive the future work needed to both determine the validity of this model and to understand how other elements of striatal circuitry could fit within this framework.