A Low-Correlation Resting State of the Striatum during Cortical Avalanches and Its Role in Movement Suppression

During quiet resting behavior, involuntary movements are suppressed. Such movement control is attributed to cortico-basal ganglia loops, yet population dynamics within these loops during resting and their relation to involuntary movements are not well characterized. Here, we show by recording cortical and striatal ongoing population activity in awake rats during quiet resting that intrastriatal inhibition maintains a low-correlation striatal resting state in the presence of cortical neuronal avalanches. Involuntary movements arise from disturbed striatal resting activity through two different population dynamics. Nonselectively reducing intrastriatal γ-aminobutyric acid (GABA) receptor-A inhibition synchronizes striatal dynamics, leading to involuntary movements at low rate. In contrast, reducing striatal interneuron (IN)-mediated inhibition maintains decorrelation and induces intermittent involuntary movements at high rate. This latter scenario was highly effective in modulating cortical dynamics at a subsecond timescale. To distinguish intrastriatal processing from loop dynamics, cortex-striatum-midbrain cultures, which lack feedback to cortex, were used. Cortical avalanches in vitro were accompanied by low-correlated resting activity in the striatum and nonselective reduction in striatal inhibition synchronized striatal neurons similar to in vivo. Importantly, reduction of inhibition from striatal INs maintained low correlations in the striatum while reorganizing functional connectivities among striatal neurons. Our results demonstrate the importance of two major striatal microcircuits in distinctly regulating striatal and cortical resting state dynamics. These findings suggest that specific functional connectivities of the striatum that are maintained by local inhibition are important in movement control.


Introduction
In the absence of specific sensory input or motor output, the brain nevertheless is highly active. In the cortex, such resting activity exhibits long-range spatial and temporal correlations [1][2][3], with intermittent neuronal bursts described by power laws and defined as neuronal avalanches [4]. Neuronal avalanches have been identified in spontaneous activity in vitro in isolated cortex preparations [4][5][6] as well as in vivo in rodents [7][8][9], nonhuman primates [10][11][12][13] and humans [2,14,15], suggesting that, during resting, the cortex resides close to a critical state [16,17] at which numerous aspects of information processing are optimized [18].
The scale-free nature of cortical avalanches implies maximal variability in size and synchrony of neuronal events [19,20]. When monitored in motor cortical areas, avalanches unfold without the presence of apparent movements [10,13], raising the question why even large avalanches during resting do not translate into sporadic or involuntary motor outputs. Here, we study this question in the context of forebrain loops that encompass cortex and basal ganglia and that are considered crucial for the initiation of voluntary as well as suppression of involuntary movements [21][22][23][24][25].
The main entry point from cortex to the basal ganglia is the striatum, which consists of more than 95% of γ-aminobutyric acid (GABA)-releasing spiny projection neurons (SPNs) and a small percentage of GABAergic interneurons (INs), particularly parvalbumin-positive, fast-spiking INs [26,27]. Although changes in intrastriatal inhibition have long been identified to lie at the core of many movement disorders (e.g., [23,25]), the distinct roles of SPNs and INs remain unclear. SPNs form a sparse network of inhibitory recurrent connections with each other [28][29][30], which theory and simulations suggest support competitive dynamics [31,32] that decorrelate networks [33]. In contrast, striatal fast-spiking INs provide a dense network of perisomatic inhibitory connections on SPNs, typically interpreted as cortical feedforward inhibition of SPNs [34][35][36][37]. Reducing striatal fast-spiking neuron activity induces involuntary movements in rodents [38], in line with a reduced number of those neurons in humans suffering from Tourette syndrome [39,40]. However, how inhibition in striatal microcircuits relates to cortical avalanche dynamics at rest and suppresses involuntary movements is unclear.
Here, we demonstrate in awake rats during quiet resting that cortical activity organizes as neuronal avalanches, whereas the striatum actively maintains a low-correlation state. Involuntary movements emerge from this dynamical profile through two distinct mechanisms. During nonselective reduction of inhibition in the striatum, movements emerged at low rate with little change in cortical avalanches but large increase in striatal synchrony. In contrast, when reducing inhibition from striatal INs only, movements emerged at high rate with corresponding large changes in cortical avalanches yet small change in relative striatal synchrony. In both scenarios, involuntary movements correlated with striatal and cortical bursts. To distinguish intrastriatal processing from loop dynamics, cortex-striatum-midbrain cultures, which lack feedback to cortex, were employed. Cortical avalanches in vitro were accompanied by low-correlated resting activity in the striatum and nonselective reduction in striatal inhibition synchronized striatal neurons, similar to in vivo. Importantly, reduction of inhibition from striatal INs maintained low correlations in the striatum while reorganizing functional connectivities among striatal neurons. Our findings demonstrate the importance of two major striatal microcircuits in distinctly regulating striatal and cortical resting state dynamics. We suggest that specific functional connectivities of the striatum that are maintained by local inhibition are important in movement control.
The change in pLFP rate was not paralleled by a corresponding change in cross-correlation (CC) between pLFPs, which was found to be relatively high at baseline and increased only weakly under PTX or IEM-1460 (Fig 1F, right; 4 ms bin size; rANOVA, F(2,12) = 10.13, p = 0.003; 1.16-fold increase). Because pLFPs could largely reflect synaptic input to the striatum, we additionally analyzed striatal MUA, which more directly reflects intrastriatal processing. Indeed, striatal MUA showed an increase in rate as well as an order of magnitude increase in spatial correlations for dyskinetic conditions. At baseline, spatial MUA correlations were low (r = 0.02 ± 0.01; n = 5 rats, 20 ms bin size) and increased 8-fold for IEM-1460 and even 11-fold for PTX (Fig 1G,   temporal correlation between MUA was also wider for PTX than IEM (Fig 1G, right; PTX: 60.0 ± 6.63 ms; IEM-1460: 34.3 ± 9.5 ms; half-width in the CC function). The dissociation between pLFP-and MUA-based measures is supported by the weak correlation between striatal MUA and the LFP under baseline conditions (Fig 1G, left; S4 Fig). Our findings so far suggest that striatal activity changes from a weakly correlated state during resting to a more correlated state under PTX-and IEM-1460-induced dyskinesia, with IN-mediated disinhibition causing involuntary movements at higher rate compared to nonselective striatal disinhibition.

Reduction in Striatal IN-Mediated Inhibition Induces Involuntary Movements with Little Change in Relative Striatal Synchrony
Changes in striatal MUA correlation could still reflect changes to striatal input rather than differences in local striatal processing. Specifically, the recruitment of cortico-basal ganglia loops during involuntary movements is supported by early reports on interrupting involuntary movements through cortical cooling in rodents and the emergence of synchronized cortical and striatal LFP deflections before movement onset [48].
This increase in cortical synchronization for IEM-1460 compared to PTX suggests that synchronization of striatal activity under IEM-1460 might be largely explained by changes in cortical activity. To compare the striatal change in synchrony relative to that in cortex, we normalized the average spatial correlation in the striatum by that found in cortex. Indeed, PTX-induced movements revealed a strong increase in relative striatal synchrony, whereas IEM-1460-induced movements emerged from relatively decorrelated striatal conditions ( Fig 2E). This decorrelated striatal state under normal resting conditions and during IEM-1460-induced movements was confirmed whether using cortical LFP or cortical MUA, which strongly correlated with cortical LFP during all conditions (S5 Fig). Taken together, these results establish two vastly different population scenarios for striatal induction of involuntary movements-a nonselective disinhibition, which induces movements at low rate in face of large relative striatal synchrony, and a selective reduction of IN-mediated inhibition, which induces movements at high rate with modest changes in relative striatal synchrony.

Reduction in Striatal IN-Mediated Inhibition Effectively Deviates Cortical Resting Activity from Avalanche Dynamics
We next demonstrated that the observed changes in striatal and cortical activity indeed arise from a resting state in cortex that organizes in the form of neuronal avalanches, and that, compared to PTX, IEM-1460-induced involuntary movements are more effective in introducing deviations from avalanche dynamics. Neuronal avalanches reflect spatiotemporal clusters of activity, which, besides pairwise correlations, also contain significant higher-order correlations that establish precise scale-invariant dynamics in space and time [12,13]. Cortical avalanches have been described in local populations of pyramidal neurons [7,9] and at the mesoscopic scale using nLFPs [4,8] as well as in humans using magnetoencephalography and functional magnetic resonance imaging [1,2,14,15]. In cortex, nLFPs are associated with increased firing in local synchronized neuronal populations [8,10,49]. We therefore used the nLFP (S5 Fig) to measure spatiotemporal activity clusters and quantify cortical dynamics. Fig 2F illustrates the definition of spatiotemporal avalanches using a given threshold for detection of nLFPs (black dots) and bin size, Δt, for concatenation of successive nLFPs into spatiotemporal clusters (adjacent dark gray time bins). In line with previous reports on ongoing activity in vivo [8], spatiotemporal clusters of cortical nLFPs during baseline distributed in size according to a power law with exponent α = -1.45 ± 0.08 and cut off at array size close to 32, the defining character- . Importantly, the average duration of spatiotemporal clusters was less than 10 ms under all conditions and thus approximately one order of magnitude shorter than the time between clusters, indicating that the increased probability of larger clusters under IEM-1460 did not result from coalescing clusters due to the chosen bin time, Δt.
In summary, a striatal resting state, in which IN-mediated inhibition is reduced, is highly effective in entraining cortical dynamics away from neuronal avalanches.

Low-Correlation Striatal Resting State in Response to Cortical Avalanches In Vitro
In order to dissociate changes in striatal dynamics due to intrastriatal processing versus cortico-basal ganglia-thalamic loops, we next studied striatal responses to cortical avalanches in organotypic cortex-striatum-midbrain cultures, which lack striatal feedback to cortex [51,52]. Cultures were grown on custom planar microelectrode arrays (MEAs) with two electrode fields, allowing for simultaneous recording from cortex (8×4 electrodes) and striatum (6×5 electrodes) (Fig 3A). Recordings were performed between 13 to 28 days in vitro (DIV) when the striatum was innervated by corticostriatal projection neurons [53] and densely innervated by tyrosine-hydroxylase (TH)-positive fibers (Fig 3B, left) originating from substantia nigra neurons of the midbrain culture (Fig 3B, right; 175 ± 33 TH-positive neurons, range: 37-385; n = 11 cultures; [51]). During that period, cortical and striatal population activities were This open-loop in vitro system confirmed our in vivo finding that cortical neuronal avalanches are accompanied by low-correlated periods in striatal activity. First, nLFP amplitudes, which correlate with MUA activity (Fig 3D), as well as spatial correlations between nLFP or MUA activity were smaller in the striatum compared to cortex (Fig 3F and 3G). Second, spatiotemporal nLFP clusters in cortex revealed avalanche signatures, i.e., a power law in cluster size distribution with exponent α close to −1.5 (Fig 3H, black, discrete: n = 8, power law versus exponential: LLR = 1,564-27,090, all p < 0.001 in favor of power law; α = −1.47 ± 0.02, [4]). In contrast, striatal nLFP cluster size distributions, although consistent with a power law distribution (Fig 3H, red; n = 8, LLR = 123-3,225, all p < 0.01; α = −3.04 ± 0.27), showed a more negative exponent ( Fig 3I; paired t test, t(7) = −5.9, p < 0.001); that is, the probability of large nLFP clusters was lower in striatum compared to cortex, in line with our finding of low spatial correlations in the striatum in vivo during resting activity. A similar relationship was observed when defining cluster size as the absolute sum of nLFP amplitudes (Fig 3H, continuous). The difference between cortical and striatal cluster size distributions was of dynamical nature because it was significantly reduced by bath application of PTX (4 μM; D KS between cortical and striatal cluster size distributions, n = 8, rANOVA, F(2,14) = 11.67, p = 0.001, S7 Fig).
Our open-loop in vitro model confirms our in vivo finding that resting state activity in the form of cortical avalanches is associated with a low-correlation resting state in the striatum.

PTX but Not IEM-1460 Amplifies and Synchronizes Striatal Responses to Cortical Input
To further study the differential effects of PTX and IEM-1460 on striatal dynamics observed in vivo, we first confirmed that IEM-1460 selectively suppressed firing in striatal INs in our in vitro system. Whole-cell current-clamp recordings ( Fig 4A) of electrophysiologically identified INs (Fig 4B and 4C) showed that spontaneous action potential firing was significantly reduced in response to local application of 500 μM IEM-1460 ( Fig 4D, t test, t(12) = 5.9, p < 0.001). To confirm that IEM-1460 did not affect AMPA-mediated excitatory postsynaptic currents in SPNs, we recorded spontaneous up-state currents in putative SPNs in the presence of QX-314 (5 μM, intracellular) and AP5 (100 μM, bath application) to block active sodium currents and N-methyl-D-aspartate (NMDA) receptors, respectively. To minimize inhibitory postsynaptic currents, voltage-clamp recordings were performed at the estimated GABA reversal potential, V h = −59 mV. As expected, local ejection of the selective AMPA receptor antagonist DNQX significantly reduced up-state currents in all putative SPNs ( Fig 4E, Fig 5A), and background-corrected calcium transients of spontaneous activity were converted to changes in fluorescence over baseline fluorescence, ΔF/F (see Materials and Methods). Simultaneous loose-patch recordings and calcium imaging ( Fig 5B) demonstrated a linear relationship between the number of striatal spikes and corresponding peak ΔF/F amplitudes (Fig 5C), as reported previously [56,57].
Under normal conditions, spontaneous striatal population activity was characterized by irregularly occurring, near-simultaneous episodes in which most neurons participated with largely varying peak amplitudes (Fig 5D, baseline). Amplitude heterogeneity was seen both within episodes and within neurons. Within <30 s of local striatal PTX application (100 μM), peak amplitudes increased ( given that cortical disinhibition would have induced prolonged activity periods at much lower rate in this system [58]. As further control, striatal changes to intrastriatal PTX application did not depend on midbrain inputs, further supporting exclusive intrastriatal PTX action (S9 Fig). In contrast, when locally applying IEM-1460 to the striatum, average ΔF/F peak amplitudes in the local striatal population did not change (Fig 5E and 5F

Transition between Striatal States of Pairwise Correlations in Response to IEM-1460
The previous analysis provides a picture of average changes but, in general, does not capture individual alterations in ΔF/F amplitude of single neurons or pairwise correlations (i.e., CCs) between neurons [59]. That is, different constellations of amplitudes or correlations could result in the same average. Indeed, the inability of IEM-1460 to change the average CC in the striatum was contrasted by its ability to significantly change individual CCs between neurons, that is, to reorganize the functional connectivity of the striatum while maintaining a low-correlation resting state. This is illustrated in more detail in Fig 6A, Fig 6A, middle), demonstrating that individual CCs changed upon local application of IEM-1460, as can be seen in the corresponding scatterplots. Fig 6B shows density plots of CCs for all consecutive segments and cultures (PTX: n = 8; IEM-1460: n = 11). The corresponding R 2 CC values are summarized in Fig 6C, demonstrating that, similar to PTX, IEM-1460 led to a highly significant change in CCs (rANOVA; PTX: F(4,28) = 16.21, p < 0.001; IEM-1460: F(4,40) = 15.72, p < 0.001). The analysis of the change in individual ΔF/F peak amplitude averages revealed a similar picture. That is, although IEM-1460 did not lead to changes in the grand average ΔF/F (Fig 5G), it changed the ΔF/F responses in individual striatal neurons  In summary, these results strongly suggest that, under normal conditions, the low-correlation state among striatal neurons requires local GABA A -mediated inhibition and that reduction of spontaneous IN firing changes the pairwise correlation state while maintaining a low average correlation (Fig 6E).

Discussion
Movement disorders, in which the basal ganglia play a pivotal role, remain a significant public health burden [25]. Although our study confirms that suppression of involuntary movements requires intact striatal inhibition [25,38,41,48], here, we demonstrate two vastly different mechanisms of the emergence of involuntary movements when manipulating striatal inhibition. The particular dynamics that gave rise to involuntary movements involved cortico-basal ganglia loops, which, due to their recurrent feedback nature, made the analysis of the population dynamics particularly challenging from a systems' point of view. A combined in vivo and in vitro approach using microelectrode array recordings and cellular resolution population imaging enabled us to study striatal population dynamics under closed-and open-loop conditions. Through this combination of techniques, we identified a low-correlation resting state of the striatum stabilized by striatal INs from which, after disruption of IN inhibition, involuntary movements can emerge at high rate involving cortico-basal ganglia loops. These pathological dynamics might be of particular clinical relevance for humans suffering from Tourette syndrome who reveal a reduction in striatal IN number [39,40]. This scenario was distinguished from a nonselective disinhibition in the striatum, which significantly increased striatal synchrony and gave rise to dyskinetic movements at much lower rate, a condition that might be encountered in early stages of Huntington's disease when striatal neurons degenerate [22].

Differential Effects of PTX and IEM-1460 on Cortico-Basal Ganglia Dynamics and Behavior
In the current study, we used two pharmacological agents that affect striatal inhibition. The first, PTX, blocks GABA A receptors expressed in SPNs and INs. In the past, GABA A -antagonists have been extensively used to study the effect of striatal disinhibition on neuronal firing and motor behavior [41][42][43][44][45]. The second drug, IEM-1460, influences striatal inhibition indirectly by blocking AMPA-mediated inputs to striatal inhibitory INs, thus leading to the disinhibition of SPNs by reducing IN-to-SPN activity. Although PTX also affected the latter connection (i.e., IN-to-SPN synaptic transmission), we observed in our experiments distinct activity and behavioral phenotypes. In vivo, both substances increased striatal synchrony and induced involuntary movements. However, local striatal PTX increased striatal synchrony even in the absence of a closed corticostriatal loop (via globus pallidus/substantia nigra/thalamus), leading to highly synchronized striatal events in our in vivo and in vitro preparation. In contrast, local striatal IEM-1460 showed increased synchrony only in vivo, most likely due to increased synchronous input from cortex (and thalamus). That is, local striatal IEM-1460 application deviated cortical activity away from avalanches into a highly synchronized state not observed under PTX. In addition, PTX and IEM-1460 showed neuronal bursts and involuntary movements at very different frequencies. That even low doses of IEM-1460 (0.5 mM versus 5 mM) induced striatal bursts at high rate further suggests that, indeed, PTX and IEM-1460 influence cortico-basal ganglia loop activity in very different ways. Thus, although both substances induced involuntary movements, our findings suggest distinct mechanisms underlying the emergence of these movements. We also note that the relatively high frequency of involuntary movement components under IEM-1460 could suggest tremor-like spontaneous movements. However, due to their variability (S2 Movie) and intermittency (S2 Fig), these movements do not resemble a continuous tremor. These mechanisms and behavioral phenotypes need to be further explored in future studies to improve the understanding of normal and pathological conditions in the basal ganglia.

A Low-Correlation Striatal Resting State That Supports Cortical Avalanche Dynamics In Vivo
Our combined in vitro and in vivo findings identified a low-correlation resting state of the striatum that is maintained in the presence of cortical neuronal avalanches and depends on intrastriatal inhibition. Avalanche dynamics in cortex are characterized by long-range spatial and temporal correlations and are described by a power law in burst size distribution with exponent close to -1.5 [4]. The low-correlation striatal resting state dynamics qualitatively differed from cortical neuronal avalanche activity as measured by a more negative power-law exponent in vitro, indicating a spatially more confined activation of striatal neuronal populations compared to cortex. Our results further show a differential participation of two major striatal microcircuit components in maintaining and regulating ongoing striatal and cortical avalanche activity through cortico-basal ganglia-thalamic loops during resting. This finding relied on a precise quantification of the cortical resting state, that is, the measurement of cortical avalanches and the quantification of deviations from avalanche dynamics. Avalanche dynamics are robustly identified using the LFP at the mesoscopic level, although recent advances with single-cell resolution have been obtained for cortex [7]. We confirmed that cortical LFPs are related to local neuronal activity and that they organize as neuronal avalanches both in vivo and in vitro.
The low rate of neuronal population bursts and corresponding involuntary movements induced by nonspecific reduction of intrastriatal inhibition is due to a refractoriness of cortex, with an absolute refractory period >300 ms [45]. Disfacilitation of inhibitory striatal INs using IEM-1460, which left the feedback inhibition between SPNs intact, quickly abolished cortical neuronal avalanche dynamics and induced corticostriatal population bursts, often less than 300 ms apart [60]. This suggests that the striatum might require lateral inhibition between SPNs to efficiently entrain cortical activity at a subsecond time scale. This finding leads us to propose that it is not the change in average correlation or activity in striatal output but rather the specific functional connectivity of the striatum, supported by lateral SPN inhibition, that influences ongoing cortical avalanche dynamics, presumably via substantia nigra/globus pallidus and thalamus by promoting certain avalanche patterns in cortex.

Pharmacological Approach Using IEM-1460 to Reduce Striatal IN Firing
The striatum receives excitatory and inhibitory input from various sources that are part of intricate feedback loops, such as cortex [61,62], thalamus [63], globus pallidus [64], and substantia nigra [65]. Using an open-loop in vitro system that exhibits the same resting state dynamics as in vivo, i.e., cortical neuronal avalanches, allowed us to isolate those aspects of striatal dynamics and corresponding microcircuits that underlie the observed dynamical changes in vivo. We confirmed in vitro that nonselective reduction of intrastriatal inhibition using PTX synchronizes striatal action potential firing, in line with the striatal changes at the multi-unit and LFP levels in vivo. Importantly, we could demonstrate in vitro that avalancheinduced striatal firing remains decorrelated when reducing IN firing with IEM-1460, in line with a low increase in relative striatal synchrony observed in vivo under those conditions ( Fig  2E). Taking advantage of monitoring local clusters of striatal neurons at high spatial resolution using intracellular calcium imaging, our in vitro approach allowed us to dissect the apparent discrepancy between the strong effect of IN manipulation in the absence of major changes in striatal synchrony. We could demonstrate in vitro that IEM-1460 strongly affected which neurons were (co-) active with little or no influence on the average CC between striatal neurons and average single neuron response amplitudes, respectively. We believe that this finding introduces the specific functional connectivity maintained dynamically by local striatal inhibition as a major factor in the regulation of activity in cortico-basal ganglia loops.
We note that changes in single neuron activity under IEM-1460, quantified by the coefficient of determination, were comparable to PTX (Fig 6C and 6D). In addition, local IEM-1460 injections in vivo induced strong changes in striatal neuron activity at high as well as ten times lower concentrations, with corresponding strong changes in cortical avalanche dynamics. This suggests that even small reductions in IN inhibition induce strong dynamical changes in cortico-basal ganglia loops.
In addition to the reduction of AMPA-mediated currents for receptors lacking the glutamate receptor 2 (GluR2) subunit, IEM-1460 has been shown to reduce NMDA receptor currents at high concentrations, thereby possibly influencing SPN input. However, the effectiveness of blocking NMDA currents is two orders of magnitude smaller compared to the effect on AMPA currents [66]. Importantly, such NMDA receptor blockade would have been expected to decrease SPN neuron firing. However, our in vitro results showed no change in average response amplitude in SPNs, and a subset of SPNs even increased their responses to IEM-1460 application (see also [38]). IEM-1460 has been shown to target AMPA receptors in cholinergic INs, and cholinergic neurons via inhibitory neuropeptide Y-positive neurogliaform neurons can influence SPN firing [67][68][69]. However, the absence of correlations between cholinergic INs and SPNs in nonhuman primate recordings [70] suggests that this pathway might not dominate striatal resting activity in vivo or in general during cortical neuronal avalanches. Accordingly, even after blockade of cholinergic transmission, selective reduction of IN firing using IEM-1460 can induce hyperkinesia [38].
Our pharmacological approach in vitro allowed for the manipulation of intrastriatal circuits in the absence of cortico-basal ganglia feedback loops and inhibitory inputs originating from globus pallidus or midbrain. However, our approach is not able to exclude other possible sources that might contribute to the observed decorrelation effect, such as inputs from diverse striatal IN classes or from a newly described corticostriatal inhibitory pathway [71]. While cell-specific manipulations can be achieved in striatal SPNs and INs using optogenetic techniques in transgenic mice [67,72], changing SPN firing to precisely test whether inhibition between SPN maintains an intrastriatal low-correlation resting state currently faces two caveats. First, manipulation of SPN firing does not allow differentiating between intrastriatal (i.e., SPN-to-SPN) and loop (i.e., striatopallidal/-nigral) connectivity. Second, SPN firing was used as a readout of the network state to calculate response amplitudes and CCs. Dissecting the functional role of feedback inhibition between SPNs would require an opto-or pharmacogenetic approach that directly manipulates SPN-to-SPN synapses, for which techniques are currently being developed [73,74].
In our in vivo experiments, MUA did not allow us to differentiate striatal cell types involved, and even single-unit analysis in the striatum is limited in mapping waveforms to identifiable cell types [75,76]. Although our multi-unit and LFP analyses identified the differential effect of PTX and IEM-1460 on cortico-basal ganglia loops, we were unable to demonstrate the corresponding specific changes in striatal IN firing. Naïvely, one might assume that the net effect of IEM-1460 in vivo reduces IN firing; however, as shown by the dramatic changes in loop activity, loop reverberation does not permit interpretations of changes in IN activity based on direct drug action alone. Therefore, we extended our in vivo approach to in vitro, in which we recorded up to a hundred striatal neurons at single-

Potential Cellular Mechanisms That Maintain a Low-Correlation Striatal Resting State and Prevent the Emergence of Involuntary Movements
The average low correlation in the striatum was maintained when manipulating INs, supporting the view that lateral (i.e., feedback) inhibition between SPNs [28][29][30] might be responsible for the low-correlated striatal resting state, in line with prediction from theory and simulations [31,32]. That feedback inhibition can affect the population of striatal neurons was indeed shown in acute slices through antidromic electrical activation [77] (with potential contributions from pallidal-striatal projections [64]). The increase in striatal synchrony upon nonselective reduction of intrastriatal inhibition using PTX is also in line with another study [78], in which acute slices were activated by electrical cortical stimulation or NMDA receptor activation.
Computational studies suggest that networks of inhibitory neurons with realistic connectivity regimes for lateral inhibition reduce the level of activity, increase the contrast of responses, i.e., decorrelate striatal input [79,80], and can cause transitioning between striatal cell assemblies [32,81]. Although lateral inhibition is not the only mechanism by which neuronal network activity can be decorrelated [82], it can greatly enhance pattern decorrelation, as recently shown in a computational network model of neurons with threshold nonlinearities [33]. Thus, threshold nonlinearities [83] and corticostriatal connectivity [62] are likely to contribute to the observed decorrelation in striatal activities.
The above models have in common that they require the collective inhibitory influence of cell groups. We propose that the large number of SPNs and wide distribution of measured synaptic strengths [35,36,84] provide the basis for lateral inhibition to affect striatal output and, consequently, future cortical activity [85]. Striatal inhibitory INs, on the other hand, might influence the functional connectivity of SPNs, thus promoting changes between different states of low correlation in the striatum that might encode specific motor programs. This idea is supported by a recent study [86], which found that striatal fast-spiking INs increase their firing particularly during the choice execution period in a choice task.
In the context of our study, we propose that a change in IN activity promotes switching between low-correlation states in the striatum, which entrain cortico-basal ganglia loops supported by lateral inhibition between striatal projection neurons. In summary, our results uncover different dynamical influences of two major intrinsic striatal microcircuits in regulating cortico-basal ganglia resting activity important for the suppression of involuntary movements in normal behavior.

Materials and Methods
All animal procedures were in accordance with National Institutes of Health guidelines. Animal procedures (protocol numbers LSN-01 and LSN-12) were approved by the National Institute of Mental Health Animal Care and Use Committee.

In Vivo Recordings
Male Sprague-Dawley rats (5-8 wk old) were used for behavioral assessment and/or chronic recording of LFPs and MUA in the cortex or striatum. To study the influence of the striatal inhibitory mechanisms, two different substances were microinjected into the dorsal striatum (AP: 0.9-1.5 mm, ML: 2.2-2.6 mm, 4.2-5.5 mm from cortical surface) through a chronic cannula (26 gauge, 1-2 mm projection; Plastics One, Roanoke, VA, United States): (1) PTX (Sigma-Aldrich), a GABA A -receptor antagonist, and (2) IEM-1460 (Tocris Bioscience), an antagonist of GluR2-lacking AMPA receptors selectively expressed in striatal INs [46,47,87]. Implantation of the cannula guide and the recording array was done under isoflurane anesthesia (1.5%-4%, 100% oxygen) and presence of the analgesic ketoprofen (5 mg/kg, subcutaneous). All cannula guides, cannulas, and recording arrays were sterilized. During the implantation surgery, care was taken to avoid blood vessels. To prevent unnecessary brain injury, the dura was carefully ruptured and the cannula guide and/or recording array was slowly lowered at a rate of~150-200 μm per min. Ketoprofen was given for up to 2 d post surgery, and animals were allowed to recover for 2-5 d before recordings. Spontaneous activity for LFP and MUA analyses (see below) was recorded up to 2 wk post surgery from superficial layers of the primary motor cortex and/or the somatosensory forelimb region (AP: 0.5-2.2 mm, ML: 3.2-3.5 mm, and 0.2-1.1 mm from cortical surface) using 8×4-MEAs (8 shanks with 4 electrodes each, plus additional reference electrode implanted along the anterior-posterior axis; 28-32 working electrodes; 200 μm inter-electrode spacing; 23 μm electrode diameter; Neuronexus, Ann Arbor, MI, US), or the dorsolateral striatum (AP: 0.7-2.1 mm, ML: 3.2-3.5 mm, 3.2-3.5 mm from cortical surface) using 16-channel MWAs (8×2 electrodes plus additional reference wire implanted along the anterior-posterior axis; 14-16 working electrodes; 150 μm inter-wire distance; 0.6-0.9 MO impedance; Microprobes, Gaithersburg, MD, US). The ground wire was connected to a scull screw located~1 mm posterior to lambda. Data were recorded for at least 30-60 min at 30 kHz using a Cerebus data acquisition system (Blackrock Microsystems). After baseline recordings, 0.8-1.5 μl sterile drug solution (PTX, 1 mM; IEM-1460, 5 mM) was injected at a rate of 0.3 μl/min for 3-5 min. The internal cannula was left in place for 1-2 min post injection, after which recordings were performed. Animals were allowed to recover for 1 d before the next recording session. In n = 3 rats, we tested a ten times lower IEM-1460 concentration (0.5 mM) and found that two out of three rats showed involuntary movements that were of similar nature as under higher IEM-1460 concentration, that is, intermittent movements at high rate in the contralateral front paw.
Animal behavior was video-recorded with a Logitech c920 camera (10-30 frames per s, fps) for behavior-only recordings, or simultaneously with LFP and MUA using a triggered CMOS camera (40 fps; Thorlabs). Involuntary movements were analyzed using custom scripts in Matlab (Mathworks, MA, US). We defined a "movement" signal as 1 minus the frame-toframe correlation for a region of interest (i.e., contralateral front paw or neck), and involuntary movements were extracted by applying a threshold of 2-3 SDs (Fig 1C and 1D). Only periods during which animals were resting (i.e., no locomotion, cage exploration, or grooming) were included in the video analysis. After recordings were finished, brains were dissected and the locations of cannula and electrode placements were confirmed in a subset of animals.
In total, data from 17 rats were analyzed in this study. All but one rat were chronically implanted with a cannula guide for local drug infusion in the striatum. A subset of rats was implanted with an MWA in the striatum (n = 8 rats) or an MEA in the superficial layers of cortex (n = 5 rats). A list of all rats and the recordings and observations is given in S1 Table. Organotypic Culture Preparation Coronal slices from rat cortex (350 μm thick, postnatal days 0-2; Sprague Dawley), striatum (500 μm thick), and midbrain (substantia nigra pars compacta; 500 μm thick) were cut on a vibratome (VT1000 S, Leica, Wetzlar, Germany) in ice-cold, sterile Gey's balanced salt solution (0.4% D-glucose) and cultured on poly-D-lysine coated and plasma-/thrombin-treated carriers to allow proper tissue adhesion [52]. After tissue adhesion to the carrier, standard culture medium was added (600 μl of 50% basal medium, 25% HBSS, 25% horse serum, 0.5% glucose, and 0.5% of 200 mM L-glutamine; Sigma-Aldrich) and changed every 3-4 DIV. At 1, 8, and 20 DIV, 10 μl mitosis inhibitor (0.3 mM uridine, 0.3 mM ARA-C cytosine-β-D-arabinofuranoside, and 0.3 mM 5-fluoro-2 0 -deoxyuridine) was added for 24 h to prevent excess glia cell formation. Cultures were incubated at 35.5 ± 0.5˚C. Carriers were either coverslips for calcium imaging and patch recording experiments or 60-channel, planar MEAs for the recording of LFPs and MUA (see below). Cultures on coverslips were incubated in a roller tube incubator at 0.6 rotations/min, and MEA cultures were incubated on a rocking storage tray at ±75˚, 0.25 cycles/min (±25˚, 0.6 cycles/min during the recording sessions except for developmental data, Fig 3C).

In Vitro Multielectrode Array Recordings
Planar titanium nitride MEAs with 60 channels (59 recording electrodes plus one reference electrode; 200 μm inter-electrode distance, 30 μm electrode diameter) were obtained from Multichannel Systems (Reutlingen, Germany). For the developmental recordings, a standard 8×8 layout was used. For all other MEA recordings, a custom layout with two sub-arrays for cortex (8×4, 31 electrodes) and striatum (6×5, 28 electrodes) was used. Both sub-arrays were separated by 1,200 μm (Fig 3A). Data were recorded at 25 kHz for MUA or 1 kHz for LFP using an MEA1060 amplifier and the MC Rack software (Multichannel Systems). Spontaneous activity for the developmental data (20 min) and the experiments with PTX bath application (4 μM, 60 min) was recorded in culture medium under sterile conditions. Washout recordings were done 24-48 h after the culture medium was replaced with conditioned medium collected 3-4 d before the experiment. All recordings were performed at 35 ± 0.5˚C after 2 wk in vitro if not stated otherwise.

Calcium Imaging
Calcium imaging was performed on coverslip cultures loaded with 50 μM OGB (Life Technologies, NY, US) dissolved in 10 μl pluronic F-127 (20% in DMSO; Life Technologies, NY, US) and 790 μl freshly prepared artificial cerebrospinal fluid (ACSF) [88]. ACSF was bubbled with 95% O 2 and 5% CO 2 and contained (in mM): 124 NaCl, 3.5 KCl, 10 D-(+)-glucose, 26.2 NaHCO 3 , 0.3 NaH 2 PO 4 , 1.2 CaCl 2 , and 1 MgSO 4 . Cultures were incubated for 60-90 min in a roller tube incubator and perfused with ACSF (flow rate~100 ml/h) for 20-30 min before imaging. After baseline recordings (5 min), PTX (100 μM) or IEM-1460 (500 μM) was ejected locally in the striatum at a rate of 12 μl/min for 5 min using glass pipettes with a tip diameter of~80-100 μm, while intracellular calcium was simultaneously imaged. Washout conditions were recorded 10-20 min after the drug application ended. Drug spillover to the cortex was prevented by using a two-compartment chamber in which a glass coverslip separated the bath between the cortical and striatal tissue. The glass coverslip was positioned~300 μm above the tissue and sealed with agar pieces around the recording chamber. ACSF and drug flow was directed away from cortex. This approach was highly efficient in avoiding any drug spillover to cortex, as shown in S8 Fig. All recordings were performed between 13-28 DIV.
Regions of interest (ROIs) were manually selected by identifying typical cell bodies ( Fig  5A), and background subtraction was performed by automatically subtracting the fluorescence signal from a dark background region within the area of two cell body diameters. All fluorescence values are expressed as relative change in fluorescence from baseline, denoted by ΔF/F, and measured as percentage. Formally, ΔF/F is defined as the percentage change in fluorescence over baseline, that is, ΔF/F = 100 (F ROI −F 0 )/F 0 , where F ROI and F 0 denote the background-corrected fluorescence intensities in the ROI and the baseline, respectively. For the spike-triggered detection of fluorescence changes (Fig 5B and 5C), the baseline was calculated as the average fluorescence 50 ms before the spike. For all other analyses, the baseline was calculated from a 30-s sliding window as the average of the 50% smallest values (i.e., excluding transients that correspond to neuronal activity). To allow for a more robust detection of calcium transients, successive increases in fluorescence (ΔF/F > 0) were summated, and the threshold detection was performed on this summated signal. The percentage of spuriously detected ΔF/F-peaks was lower than 0.5% (n = 8 neurons).
For the estimation of the rate of up-state events in striatal neurons, we used the summed widefield (bulk) fluorescence signal within the field of imaging. Because up-states among striatal neurons are correlated [51,89] and driven by cortical input, this approach gave a good approximation of the input rate arriving from cortex.

Patch-Clamp Recordings
Patch pipettes were pulled from borosilicate glass using a P-97 micropipette puller (Sutter Instrument, CA, US), and had a resistance of 5-10 MO. For all recordings, pipette resistance and capacitance were compensated for. Loose-patch and cell-attached recordings were performed in voltage-clamp mode using patch pipettes that were filled with regular ACSF. Whole-cell patch-clamp recordings were done in current-clamp mode with an intracellular solution containing (in mM) 132 K-gluconate, 6 KCl, 8 NaCl, 10 HEPES, 2 Mg-ATP, and 0.39 Na-GTP, or voltage-clamp mode with an intracellular solution containing (in mM) 132 CsMeSO 3 , 1 CsCl, 10 HEPES, 2 Mg-ATP, 0.39 Na-GTP, and 5 QX-314. The intracellular solution was kept on ice during the experiment. For the local application of PTX (100 μM) or IEM-1460 (500 μM), a second patch pipette was placed in close vicinity (<60 μm; Fig 4A) of the patched soma and drugs were ejected at 50-55 mmHg. As expected, all recorded neurons (5/5) responded to ejection of DNQX (50 μM, Fig 4E). Significance was calculated by comparing 100 s baseline plus 100 s washout with 80 s of drug data using Student's t test.

Immunohistochemistry and Confocal Imaging
A subset of cultures was used for post-hoc immunostaining of TH. Cultures were rinsed in phosphate buffered saline (PBS), fixed in 4% paraformaldehyde for 40-60 min, and incubated for 2 h at room temperature in blocking solution (10% normal goat serum and 0.5% Triton X-100 in PBS). For all subsequent steps, a carrier solution consisting of 1% normal goat serum and 0.3% Triton X-100 in PBS was used. Cultures were incubated for~12 h at 4˚C in a TH-antibody solution (1:1000, antimouse, Immunostar, WI, US), washed three times for 10 min each, incubated 1-2 h at room temperature in secondary antibody solution (1:1000, Alexa 555 anti-mouse, Invitrogen, NY, US), and washed again three times for 10 min each at room temperature. Before the confocal imaging, cultures were rinsed in PBS and mounted on coverslips using a fluorescence-preserving mounting medium (Vector Laboratories, CA, US).
Confocal images were obtained with a Zeiss LSM 510 using a 63× oil immersion objective (numerical aperture 1.4, 0.6 μm optical thickness). For the cell counting of TH-positive neurons in the substantia nigra pars compacta, images were obtained with a high-speed scanning confocal microscope (Leica TCS SP5 II, 10× objective) with tile scan function. Cell counts were obtained from maximum z-stack projections (42 μm thick, 6 μm optical thickness). For TH-positive neurons that were organized in dense clusters, only well-distinguishable somata or somata with distinct (dark) nucleus were counted (Fig 3B, right panel), thus likely underestimating the actual number of dopaminergic neurons.

LFP and MUA Analyses
For in vitro recordings, MUA was detected by band-pass filtering at 300-4000 Hz and subsequent thresholding (-5 SD of each trace). LFPs were band-pass filtered at 1-100 Hz for the developmental data, which contained dominant frequency components above 50 Hz, and 1-50 Hz for all other analyses. nLFP deflections were detected by finding the minimum value of the LFP signal that crossed a given threshold z (measured in SDs). Previous studies showed that cortical nLFPs are associated with increased activity and synchrony in local firing [8,10,49]. That cortical nLFP are correlated with cortical multi-unit firing was confirmed in this study (S5 Fig). We furthermore found this relationship in the striatum in vitro (Fig 3D) despite small amplitudes of striatal nLFPs. The SD was determined for each channel individually and estimated from 2-3 s of baseline activity (z = -4.5). The threshold value z was varied to confirm the robustness of the reported power-law exponents (see also [10,50]). For the power spectral analysis of the developmental data, ±500 ms around nLFP threshold crossings were analyzed. The power spectrum was calculated by using the fast Fourier transform with a Hann window function. Averages for individual cultures were calculated across all channels and subsequently normalized (integral over the entire frequency range normalized to unity) before calculating the average over all cultures. All in vitro data were analyzed using the phaseneutral filter implementation filtfilt in Matlab and the Neuroshare library (http://neuroshare. sourceforge.net) for data import.
For in vivo recordings, presumable multi-unit spikes were extracted from the high-pass filtered signal (>250 Hz) by applying a threshold at -6 times the root mean square of the signal using the Cerebus Central software (Blackrock Microsystems). Because movements could cause artifacts in the high-pass filtered signal, thresholded waveforms were subsequently offline-sorted using the Offline Spike Sorter (Plexon Inc., Dallas, TX, US). Only electrodes were used for analysis for which MUA could be isolated from movement artifacts based on the typical biphasic waveform of multi-unit spikes (Fig 1G, inset). Calculation of LFPs was performed as described for in vitro using the entire signal for each electrode for the estimation of SD (z = -2.5). For the avalanche analysis, z was varied to confirm the robustness of the estimated power-law exponents (see above). In the striatum, MUA was associated with pLFP deflections (S4 Fig; Fig 1G, left). We therefore extracted pLFP deflections (z = 2.5-3) from in vivo striatal recordings.

CC Analyses
CCs were calculated from binned time series (rasters) of p/nLFPs, multi-unit spikes, or from continuous ΔF/F traces. Values for p/nLFP and MUA rasters were discrete and corresponded to the number of p/nLFP or spike events per bin, respectively. The raw CC between two time series, x t and y t , was defined as denotes the expected value operator, τ the time lag, and μ and σ denote mean and SD, respectively. CC for n/pLFP or MUA rasters were shuffle-corrected by subtracting CC shuffle (average of ten repetitions) from CC. The calculation of CC for calcium imaging data was performed on the ΔF/F traces. The average CC was reported as the average value across all electrode or neuronal pairs for time lag τ = 0 if not stated otherwise. In Fig 6, individual CCs were analyzed.

Detection of Spatiotemporal nLFP Clusters and Avalanche Analysis
Rasters of nLFP events that crossed a predefined threshold, z, were created by binning the nLFP times with bin size Δt = 2-4 ms [4,11,50]. Previous studies showed that cortical nLFPs can be used as a readout of cortical synchronized population activity (see also S5 Fig) [8,10,49] to measure the propagation of spatiotemporal activity clusters. Due to the predominantly local propagation of activity [11], compact 8×4 MEAs (Fig 2A in vivo, Fig 3A in vitro) were used as described above. From the recorded nLFP rasters, spatiotemporal clusters were extracted by finding cascades of nLFP events that were separated by at least one bin width ( Fig 2F). The size of a cluster was defined as the number of nLFPs within the cluster ("discrete," Figs 2G and 3H, left). Alternatively, cluster sizes can be defined as the sum of absolute nLFP amplitudes ("continuous," measured in μV; Fig 3H, right), resulting in a continuous distribution [4]. Neuronal avalanches are defined by a distribution of cluster sizes that follows a power law with exponent -1.5 [4] up to the number of electrodes in the recording array. Importantly, the power law is invariant to the number of electrodes used in the recording array up to the so-called "cut-off," which is given by the number of electrodes in the recording array. This property allows for a robust estimation of the power law exponent [11,50], as described below.

Statistical Analyses
Power-law exponents were estimated using a maximum-likelihood approach [50,90]: ln p a ðs i Þ Low-Correlation Striatal Resting State Dynamics during Cortical Avalanches denotes the log-likelihood of observing the vector of given cluster sizes s = (s 1 ,. . .,s n ) assuming a power law with exponent α, that is, In cortical networks, the cut-off is typically at the system size, N, which is given by the number of electrodes in the cortical array (see [4,11,50]). Thus, cortical event size distributions were fitted on the range from one to the number of electrodes in the cortical array. Correspondingly, exponents for striatal distributions are reported for a model that ranged from one to the number of electrodes in the striatal array.
For the comparison of power law versus exponential distribution (the expected distribution for independent neuronal activity), we used the LLR test [50,90]: where l(α|s) denotes the log-likelihood for a power law with exponent α, and l(λ|s) the log-likelihood for an exponential distribution with parameter λ For the comparison of distributions to a power law with exponent -1.5, or across different experimental conditions, we used the KS statistic [50] where P data denotes the cumulative distribution of the data and P compare the cumulative distribution of the reference power-law model [i.e., P compare ðxÞ ¼ X x s¼1 p a ðsÞ] or data from a different experimental condition.
For paired comparisons of two or more means, we used the paired Student's t test and repeated-measures ANOVA with Bonferroni correction, respectively. Values are expressed as mean±standard error of the mean if not stated otherwise.  (DIV 19) showing the cortex (ctx) and striatum (str), the coverslip (cv, approximately 300 μm above tissue) for bath compartmentalization, and the pipette (p) for dye application. The gaps between coverslip and chamber were sealed with agar pieces (not visible in picture). The white dashed line shows the approximate border between cortex and striatum. The white and black squares show approximate locations of the imaging regions for cortex and striatum, respectively. ACSF flow was from cortex to striatum. SR101 ejection as indicated in