Compartment-specific dendritic information processing in striatal cholinergic interneurons is reconfigured by peptide neuromodulation

Cholinergic interneurons are central hubs of the striatal neuronal network, controlling information processing in a behavioral-state-dependent manner. It remains unknown, however, how such state transitions influence the integrative properties of these neurons. To address this, we made simultaneous somato-dendritic recordings from identified rodent cholinergic interneurons, revealing that action potentials are initiated at dendritic sites because of a dendritic axonal origin. Functionally, this anatomical arrangement ensured that the action potential initiation threshold was lowest at axon-bearing dendritic sites, a privilege efficacy powerfully accentuated at the hyperpolarized membrane potentials achieved in cholinergic interneurons following salient behavioral stimuli. Experimental analysis revealed the voltage-dependent attenuation of the efficacy of non-axon-bearing dendritic excitatory input was mediated by the recruitment of dendritic potassium channels, a regulatory mechanism that, in turn, was controlled by the pharmacological activation of neurokinin receptors. Together, these results indicate that the neuropeptide microenvironment dynamically controls state- and compartment-dependent dendritic information processing in striatal cholinergic interneurons.


In brief
Striatal cholinergic interneurons are integral for adaptive behaviors. Williams et al. show that the operation of these neurons is context dependent. When in a state associated with salient behavioral stimuli, excitatory input has an asymmetric impact across dendritic domains. This inequality is nullified when neurons receive a modulatory signal from striatal output neurons.

INTRODUCTION
The striatum is involved in cognitive, sensory, and motor behaviors, signaling information across defined spatial domains largely inherited from topographically aligned neocortical areas. 1 Despite this apparently simple hierarchical arrangement, the striatum has a central role in adaptive behaviors, [2][3][4][5][6] and the relationship between neocortical and striatal activities is highly plastic and scalable with learning. 1,[7][8][9] The striatum is therefore considered a nexus for learning and the control of behavioral flexibility, a function suggested to be driven by the rich dopaminergic and cholinergic modulatory systems that influence striatal activities. [10][11][12][13][14] Notably, the cholinergic control of striatal function is believed to arise from a sparse, but widespread, population of intrinsic cholinergic interneurons (CINs) that express the neuropeptide receptor neurokinin-1. [14][15][16][17][18][19][20][21] Indeed, the dense intrastriatal axonal arbors of CINs are anatomically positioned to have far-reaching impacts on striatal function, targeting postsynaptic sites of striatal projection neurons, medium spiny neurons (MSNs), and interneurons, as well as glutamatergic and dopaminergic axons innervating the striatum. 18,[22][23][24][25][26][27][28][29] For example, recent work has demonstrated that CINs dynamically control the phasic release of dopamine across wide striatal territories through the nicotinic acetylcholine receptor (nAChR)-mediated driving of action potential (AP) initiation in dopaminergic axons. 26,27 Furthermore, through the activation of presynaptic muscarinic acetylcholine receptors (mAChRs), CINs control cortico-and thalamo-striatal glutamatergic excitatory synaptic drive to MSNs. 22 Functionally, CINs are dynamically engaged during behavior, 11,24,[30][31][32][33] where their specific ablation, or the blockade of AChR-mediated signaling, disrupts adaptive behaviors. 3,4 Moreover, CIN dysfunction is believed to have a significant role in a range of neuropsychiatric disorders, 14,18,34 with direct evidence also demonstrating the alteration of excitability in models of Parkinson's disease. [35][36][37] Intriguingly, CINs generate ongoing patterns of low-frequency AP firing in vivo, which are paused by salient sensory or motivational stimuli. 11,24,[30][31][32][33] At the intracellular level, electrophysiological recordings have demonstrated that such pauses of AP firing are characterized by periods of membrane hyperpolarization, which are driven and shaped by cell-intrinsic and synaptic mechanisms. 22,24,[38][39][40][41] As tonic AP firing of CINs is broadcast widely through the striatum, 26 pauses of AP firing have been suggested to powerfully influence striatal information processing, acting to temporally frame responsiveness and synaptic plasticity. 33,38,42 Although the drivers of AP pauses in CINs have been investigated, little is known about how such pauses veto ongoing thalamo-and cortico-striatal excitatory inputs that powerfully excite CINs and drive persistent firing of MSNs. 1,22,23,31,43,44 More generally, it is unknown (legend continued on next page) ll OPEN ACCESS Article whether the dendrites of CINs are electrically excitable and what impact excitatory synaptic input received throughout the anatomically distributed dendritic trees of CINs has on AP output. 31,45 To address these questions, we examined how the dendritic excitatory input is integrated and controlled in identified rat and mouse striatal CINs using simultaneous somatic and dendritic patch-clamp recording techniques. Our results demonstrate a rich voltage-dependent and dendritic-compartmentspecific landscape of dendritic integration in CINs, which is powerfully modulated by Substance P (SP), a neuropeptide synaptically released from neighboring D1-MSNs. 21 Taken together, our findings suggest that CINs that dynamically modulate striatal information processing are, in turn, modulated by the activity of striatal output neurons.

Identification of CINs
Putative CINs were visually identified under infrared differential interference contrast (IR-DIC) microscopy by their soma size and the pattern of dendritic arborization in acute brain slices of the rat dorsal striatum. Somatic recordings demonstrated that putative CINs tonically fired APs in cell-attached and wholecell recording configurations 38 and exhibited subthreshold electrophysiological properties significantly different from physiologically identified MSNs (peak apparent input resistance: cholinergic = 49.9 ± 2.2 MU, MSN = 16.2 ± 0.8 MU, Mann-Whitney test, p < 0.0001; voltage sag: cholinergic = 14.0 ± 0.7 mV, MSN = 2.0 ± 0.1 mV, t test, p < 0.0001; current step: cholinergic = À0.82 ± 0.03 nA, n = 43, MSN = À0.93 ± 0.03 nA, n = 22; Figure 1). Anatomical reconstruction of putative CINs revealed a complex morphology characterized by large, sparsely branching dendritic trees stemming from large-volume somata and highly elaborated branched axonal arborizations (soma volume = 2,223 ± 171 mm 3 , n = 27; dendritic field area = 0.132 ± 0.01 mm 2 ; n = 15; axonal path length = 25.2 ± 8.9 mm, n = 5; Figures 1C, S1, and S2). To confirm cell-class identity, we observed that choline acetyltransferase (ChAT) immunostaining decorated the somata and primary dendrites of all anatomically reconstructed CINs ( Figures 1A and S1). In contrast, ChAT immunostaining was absent from recovered MSNs, which had spinous dendrites and morphological properties consistent with previous descriptions 46,47 (soma volume: 540 ± 23 mm 3 , n = 19; dendritic field area = 0.059 ± 0.007 mm 2 , n = 10; Figures 1C and S1). To further confirm the identification of CINs, whole-cell recordings were made from neurons of the murine dorsal striatum in brain slices prepared from mice in which channelrhodopsin-2 (ChR2) was genetically expressed exclusively in cholinergic neurons (ChAT-Cre-ChR2-EYFP mice). 48 The photoactivation of ChR2 in neurons identified as CINs by IR-DIC microscopy and electrophysiology drove the firing of a single AP in each neuron tested (480 nm, 2-4 ms full field-light pulse; n = 48 neurons; Figure S3). Analysis revealed that the shape of current step-evoked APs, the apparent input resistance, and the time-dependent anomalous rectification of genetically identified mouse CINs clustered with those of rat CINs but were divergent from those of MSNs ( Figures 1E-1I).

Site of AP initiation
The properties of APs recorded from CINs were distinct from those of MSNs, exhibiting a slower and more complex rising phase and a longer time course (10%-90% rise-time: CIN = 0.28 ± 0.004 ms, MSN = 0.16 ± 0.003 ms, t test, p < 0.0001; base width: CIN = 3.10 ± 0.07 ms, MSN = 1.80 ± 0.04 ms, Mann-Whitney, p < 0.0001; Figures 1F and 1G). The complexity of the rising phase of somatically recorded APs in CINs was better resolved by calculating their first derivative, which demonstrated separation into distinct components known to represent an axon initial segment (AIS), and secondary somato-dendritic spike [49][50][51][52][53] (Figures 1H and S3). Notably, across our sample of both rat and mouse CINs, the relative amplitude of the AIS component and its coupling with the somato-dendritic spike varied between the cells (Figure 1H), producing a wide distribution of the base width of AP first derivatives, a pattern not observed for the first derivative of APs recorded from MSNs (rat CIN = 0.441 ± 0.005 ms, murine CIN = 0.414 ± 0.008 ms, MSN = 0.298 ± 0.005 ms; rat and mouse CIN vs. MSN, ANOVA, p < 0.0001 and p < 0.0001, respectively; Figure 1I).
The properties of somatically recorded APs are known to be influenced by the subcellular location of the axon, the site of AP initiation, which in some classes of mammalian neurons originates from a dendritic site, distal to the soma. [54][55][56][57][58][59] To explore whether this organization underlies the cell-to-cell variability of the coupling between the AIS and somato-dendritic components of APs in CINs, we made simultaneous somatic and dendritic recordings ( Figure 2). To our surprise, in a fraction of simultaneous recordings, APs were first detected at the dendritic recording site, when driven in response to steps of positive current delivered through the somatic or dendritic recording electrode (rat 25 of 135; mouse 4 of 18; Figures 2A-2C and S3H-S3J). By contrast, APs were first recorded at the soma in most of the simultaneous recordings (Figures 2A-2C). Pooled data of the time difference between AP onset at somatic and dendritic recording sites, therefore, revealed a predominately sloped relationship, reflecting the distance-dependent delay in the onset as APs spread from the soma to dendrites ( Figure 2D, positive times). This relationship was, however, decorated by observations where APs first occurred at the dendritic sites ( Figure 2D, negative times), together with observations where APs were detected at somatic and dendritic recording sites nearly simultaneously ( Figure 2D). We suggest these findings indicate that APs backpropagated from the soma to the dendritic recording site in most of the recordings, whereas in a fraction of dendritic recordings, APs were initiated at a site closer to the dendritic recording electrode. Consistent with this idea, we found a weak decremental relationship between AP amplitude and dendritic recording distance from the soma (slope of relationship = 12.1 mV per 100 mm; Figure 2E). However, when these data were sorted regarding AP onset, the dendritic amplitude of APs first detected at the dendritic recording sites was greater than that of APs, which were first detected at the soma, as would be predicted for APs recorded at sites close to the site of initiation (slope of relationship = À8.1 mV per 100 ms; dendrite first (>40 ms) = 81.1 ± 1.1 mV, soma first (>40 ms) = 70.5 ± 0.8 mV, t test, p < 0.0001; Figure 2F). Taken together, these data demonstrate an apparent heterogeneity in the site of AP initiation in CINs. However, as the anatomical reconstruction of rat and mouse CINs revealed somata surrounded by 4.8 ± 0.2 (n = 27), and 7.1 ± 0.4 (n = 8) primary dendrites, respectively, the frequency of detection of APs first detected at dendritic sites in untargeted dendritic recordings suggests that the axonal site of AP initiation is dendritically located in most CINs. Consistent with this interpretation, high-resolution morphological reconstruction of CINs demonstrated that the axon emerged from a dendritic site, located 39.3 ± 15 mm from the border of the soma (n = 5; Figure S2). In these reconstructions, the axon could be differentiated by its highly branched structure and diameter at the primary, secondary, and tertiary branches ( Figure S2). Furthermore, immunohistochemistry revealed the presence of Ankyrin-G immunostaining at the proximal axonal sites 57 ( Figure S2).
To explore how a dendritic origin of the axon in CINs influences the properties of somatically recorded APs, we calculated the second derivative of AP waveforms to accentuate the separation of AIS and somato-dendritic components 52 ( Figure 2G). In recordings where APs were first detected dendritically, which we refer to as recordings made from axon-bearing (AB) dendrites, the AIS component of the second derivative of APs was greatest in amplitude at the dendritic recording site ( Figures 2G, S3H, and S3I). By contrast, in recordings where APs were first detected at the soma and subsequently spread to the dendritic recording site, which we refer to as non-axon-bearing (nAB) dendritic recordings, the AIS component was greatest in amplitude at the soma ( Figures 2G, S3H, and S3I). Analysis revealed that the ratio of the amplitude of the AIS component recorded from dendritic and somatic sites showed a characteristic relationship, with the AIS component being largest in amplitude at AB-dendritic sites and dramatically attenuating as it spread to the soma (Figure 2H). This relationship suggests that the coupling between the AIS and somato-dendritic components of APs at the soma should exhibit a voltage-dependent relationship, in a manner analogous to the antidromic invasion of APs. 49 To test this idea, we made somatic recordings from mouse ChAT-Cre-ChR2-EYFP CINs and evoked single AP firing by the photoactivation of ChR2 (2-4 ms, 480 nm light pulse) as the membrane potential was progressively hyperpolarized by the injection of tonic holding current ( Figures S3E-S3G). Analysis revealed the initiation threshold for ChR2-evoked APs was highly voltage dependent, becoming progressively more negative as the somatic membrane was hyperpolarized ( Figures S3E-S3G). This change in the AP threshold was accompanied by an increased separation between the AIS and somato-dendritic spike components, until, at negative potentials ($À90 mV), the AIS component could be evoked in isolation ( Figures S3E-S3G). Notably, however, the initiation threshold for APs evoked by somatic current steps did not exhibit the same relationship in these recordings, remaining relatively stationary as the membrane was hyperpolarized (Figure S3F). Together, these data show that ChR2-evoked depolarization acts to drive axonal AP initiation, suggesting that such light-evoked APs should display contrasting properties at (B and C) Cumulative probability distributions of the time difference of AP onset between somatic and dendritic recording sites for rat (black symbols) and mouse (magenta symbols) CINs. Note the dendritic onset of APs (negative times) evoked by somatic (B) and dendritic (C) positive current steps in a fraction of rat and murine CINs. (D) Relationship between the time difference of AP onset and dendritic recording distance from the soma. Note the sloped relationship, decorated by observations where APs were first detected dendritically (negative times). (E) Decremental distance-dependent relationship between AP amplitude and dendritic recording distance. The average somatic amplitude of APs (±SD) is shown. The line represents a linear fit to the data. (F) Relationship between the amplitude of dendritically recorded APs and the time difference of AP onset between somatic and dendritic recording sites. Negative times represent APs that were first recorded at dendritic sites. The data have been fit by linear regression (line). (G) Calculation of the second derivative of AP waveforms reveals that the axon initial segment (AIS) spike is largest in amplitude at an axon-bearing dendritic recording site (blue, upper traces) and attenuates as it spreads to the soma (black, upper traces), preceding the somato-dendritic component. Inversely, in a recording made from a non-axon-bearing dendrite the AIS spike is largest in amplitude at the soma (lower traces). Same neurons as illustrated in (A). (H) Pooled data illustrating the relationship between the ratio of the amplitude of the AIS recorded from dendritic and somatic sites, and the time difference in AP onset. The data have been fit by a single exponential function (line). nAB-and AB-dendritic sites. To directly test this idea, we made simultaneous somatic and dendritic recordings from murine CINs and drove AP firing by the photoactivation of ChR2 or the injection of steps of somatic and dendritic current. When recordings were made from AB-dendritic sites, APs were first detected dendritically, and membrane hyperpolarization revealed in isolation an AIS component ( Figures S3H-S3J). Notably, the amplitude of the ChR2-evoked AIS component was similar at depolarized and hyperpolarized potentials when recorded from the ABdendrite, but was attenuated at the soma; consequently, at the hyperpolarized potential the AIS component failed to trigger an somato-dendritic spike ( Figures S3H-S3J). By contrast, in dendritic recordings from murine CINs where APs were first detected somatically, the dendritic amplitude of the AIS component was attenuated relative to the soma at both depolarized and hyperpolarized potentials ( Figures S3H-S3J). Taken together, therefore, these functional data reveal that cell-to-cell variability in the placement of the dendritic origin of the axon determines the properties of somatically recorded APs in rat and mouse striatal CINs.

Voltage-dependent impact of dendritic excitatory input
We next explored the electrical architecture of CINs. Simultaneous somato-dendritic recordings demonstrated that somatic voltage responses, evoked by long (1 s) steps of the negative current (À100 pA), efficiently spread to the dendritic sites ( Figure 2I), and were only fractionally attenuated, at steady state, when recorded up to 160 mm from the soma (holding potential = À61.5 ± 0.2 mV, peak apparent input resistance = 131.3 ± 3.4 MU, voltage sag = 5.6 ± 0.2 mV; soma-to-dendrite voltage-transfer-slope = À0.03 per 100 mm, n = 101; Figure 2J). To explore whether this was also the case for transient excitatory postsynaptic potential (EPSP)-shaped voltage responses, we injected current waveforms at somatic and dendritic recording sites to generate families of simulated EPSPs (sEPSPs; current kinetics: t rise = 0.5 ms, t decay = 10 ms; Figures 3 and S4). When delivered from membrane potentials near rest, the amplitude of the simulated synaptic current required to generate AP firing was found to be only weakly related to the dendritic site of generation, and, on average, marginally greater at dendritic recording sites compared with the soma, indicating the near uniform impact of dendritic excitatory input at the site of AP generation (dendritic V m = À61.6 ± 0.4 mV; average recording distance = 75 ± 2 mm; rheobase: soma = 0.70 ± 0.04 nA, dendrite = 0.80 ± 0.07 nA, Wilcoxon test, p = 0.0058, n = 66; ratio = 1.10 ± 0.03; slope = 0.15 per 100 mm). Notably, when dendritic recordings were separated into those made from AB-and nAB-dendrites, the amplitude of the simulated excitatory synaptic current required to evoke AP firing from nAB-dendritic sites was 1.16 ± 0.03 times greater than that required at the soma (soma = 0.68 ± 0.04 nA; dendrite = 0.80 ± 0.06 nA, Wilcoxon test, p < 0.0001; n = 52; Figures 3A and 3B). By contrast, recordings from AB-dendrites revealed that the current threshold for AP generation was less than that at the soma, suggesting that input to this dendritic region is privileged for the control of AP output (soma = 0.67 ± 0.06 nA, dendrite = 0.52 ± 0.04 nA, ratio = 0.80 ± 0.04, t test, p = 0.003; n = 13; Figures S4A-S4D).
As CINs exhibit a bimodal membrane potential distribution in vivo, characterized by periods of depolarization and AP firing followed by long hyperpolarized pauses, a pattern of activity thought to be essential for the gating of striatal information transfer and plasticity, 38 we next investigated the efficacy of dendritic excitatory input when the membrane potential of CINs was hyperpolarized (dendritic V m = À74.6 ± 0.57 mV, n = 66; Figure 3). Strikingly, in the nAB-dendritic tree, membrane hyperpolarization remodeled the impact that dendritic excitatory input had on AP output ( Figure 3A). At hyperpolarized potentials, the simulated excitatory synaptic input required to drive AP firing from nAB-dendritic sites was dramatically enhanced and was significantly greater than the threshold somatic current (dendrite = 2.92 ± 0.17 nA, soma = 1.70 ± 0.08 nA, Wilcoxon test, p < 0.0001, n = 57; ratio = 1.73 ± 0.06; Figures 3A and 3B). Furthermore, membrane hyperpolarization remodeled the pattern of AP backpropagation into nAB-dendrites, replacing the close similarity between the amplitude of backpropagating APs (BPAPs) driven by somatic and nAB-dendritic sEPSPs at depolarized potentials (soma sEPSP = 63.5 ± 1.0 mV, dendritic sEPSP = 60.9 ± 1.1 mV, Mann-Whitney, p = 0.147; n = 51; Figures 3A and 3C), with a dramatic amplitude disparity at hyperpolarized potentials (soma sEPSP = 51.9 ± 1.5 mV, dendritic sEPSP = 36.5 ± 1.7 mV, Mann-Whitney, p < 0.0001; n = 52; Figures 3A and 3C). Analysis of the diminutive BPAPs evoked by dendritic sEPSPs in nAB-dendrites at hyperpolarized potentials revealed that the time course was also accelerated when compared with that evoked from depolarized potentials (dendritic sEPSP-evoked BPAP width: depolarized = 1.48 ± 0.07 ms, hyperpolarized = 0.83 ± 0.02 ms, Mann-Whitney, p < 0.0001; n = 52; Figure 3D). In contrast, in AB-dendritic recordings, this voltage-dependent transformation did not occur, and the threshold simulated excitatory synaptic current required to drive AP firing remained lower than that required at the soma (soma = 1.64 ± 0.16 nA, dendrite = 1.24 ± 0.09 nA, Wilcoxon test, p = 0.0003, ratio = 0.79 ± 0.05; n = 15; Figure S4). Taken together, these data reveal that the impact of dendritic excitatory input on AP output of CINs is regulated in a dendritic compartment-specific manner, where the efficacy of nAB-dendritic excitatory input is heavily attenuated at hyperpolarized potentials, whereas excitatory input to AB-dendritic sites is further privileged for the generation of AP output.

Dendritic compartment-specific control of neuronal outputs
To understand this compartment-specific control of the efficacy of dendritic excitatory inputs, we next explored the factors that shape the amplitude and time course of sEPSPs generated at the nAB-dendritic sites from hyperpolarized potentials ( Figures 3A and 4). At the site of generation in the nAB-dendritic tree, the injection of families of simulated excitatory synaptic inputs revealed that the resultant sEPSPs underwent a characteristic transformation as the amplitude of the driving excitatory current was increased, displaying a highly rectified relationship, until at intensities just below the threshold for the generation of AP firing, a transient spike-like potential was generated, which crowned the peak of sEPSPs (current amplitude = 2.81 ± 0.16 nA; n = 59; Figures 3A and 4A). Pooled data revealed the amplitude of this spike-like component increased as dendritic recordings were made from increasingly remote nAB-dendritic sites (slope = 26.8 mV per 100 mm; Figure 4B). The spike-like component was recruited in a voltage-dependent manner, as small amplitude sEPSPs were characterized by simple electrotonic charging and peaked at times following the excitatory current, whereas larger amplitude sEPSPs peaked at times corresponding to the underlying excitatory current because of the genera-tion of transient spike-like potentials ( Figures 4A, 4C, and 4D). Simultaneous somatic recording demonstrated that nAB-dendritic spike-like potentials powerfully attenuated as they spread to the soma, where these potentials were barely detectable in many cases (attenuation = 12.9-± 1.4-fold, n = 85 events; Figures 4A, 4D, and 4E). Consistent with a regenerative  (legend continued on next page) ll OPEN ACCESS Article activation mechanism, the nAB-dendrite spike-like component was attenuated by the bath application of the sodium channel blocker tetrodotoxin (TTX, 1 mM; n = 9; Figures 4F and 4G). However, in the presence of TTX, nAB-dendritic sEPSPs remained highly rectified, and large-amplitude sEPSPs peaked at times corresponding to the peak of the driving excitatory current ( Figures 4H and 4I). To explore whether this behavior was mediated by the activation of voltage-gated potassium channels, we constructed families of sEPSPs, recorded in the presence of TTX, and after the addition of the potassium channel blocker 4-amino pyridine (4-AP) (5 mM; n = 11; Figures 4H and 4I). The application of 4-AP transformed sEPSPs, removing both rectification and the voltage-dependent sculpting of sEPSP waveforms, to reveal families of sEPSPs that exhibited electrotonic charging patterns and peaked at times following the driving current ( Figures 4H and 4I). Consequently, the pharmacological blockade of potassium channels significantly reduced the driving current required to generate sEPSPs of a prescribed amplitude ( Figures 4H and 4I). Taken together, these data reveal that separable dendritic amplification and suppression mechanisms, mediated by dendritic sodium and potassium channels, respectively, are operational in the nAB-dendritic tree of CINs.
To directly examine how these opposing mechanisms regulate the efficacy of nAB-dendritic excitatory input, we generated families of sEPSPs under control conditions and then applied 4-AP (5 mM; Figures 4J, 4K, and S5). As 4-AP was washed into the recording chamber, the peak amplitude of regenerative activity generated by dendritic sEPSPs increased, until at steady-state dendritic sEPSPs were crowned by the firing of a large-amplitude dendritic spike that preceded the generation of AP firing ( Figures 4J and 4K). Analysis revealed the unmasking of powerful dendritic electrogenesis by 4-AP was mediated by the removal of the fast repolarization of sEPSP-evoked dendritic spikes (AP initiation threshold time calculated as the time at which the first derivative of the somatic voltage exceeded 50 V s À1 ; dendrite to soma voltage disparity: control = 0.67 ± 1.4 mV, 4-AP = 29.6 ± 4.1 mV, t test, p < 0.0001; Figures 4K and S5). Furthermore, the application of 4-AP reduced the threshold excitatory current required to generate AP firing, normalizing the ratio of somatic and dendritic simulated excitatory synaptic current required to generate neuronal output at hyperpolarized membrane potentials (control ratio = 2.00 ± 0.15, 4-AP ratio = 1.16 ± 0.07, p = 0.0006, n = 14; Figures 4M and S5). These results therefore demonstrate that 4-AP-sensitive potassium channels underlie the voltage-and compartment-specific control of dendritic sEPSP efficacy.

Properties and dendritic distribution of a transient potassium conductance
To directly explore the properties of the potassium conductance that heavily dampens the excitability of CINs, we excised nucle-ated and somatic and dendritic outside-out patches from rat CINs (Figures 5 and 6). Nucleated patch recordings, at physiological temperatures, revealed a large transient outward current in response to positive voltage test steps when sodium and mixed cation channels were pharmacologically blocked, which rapidly decayed to a fractional steady state (voltage step to 40 mV, holding potential À110 mV; outward current: peak amplitude = 2.19 ± 0.04 nA; 10%-90% rise-time = 0.42 ± 0.01 ms; t decay = 14.49 ± 0.31 ms; steady state = 0.35 ± 0.01 nA; n = 129). Families of positive voltage steps demonstrated that the transient outward current was first activated at $À50 mV, whereas the steady-state component was activated from more depolarized potentials, suggesting separable conductances ( Figures 5A and 5B). Furthermore, a steady-state inactivation voltage protocol demonstrated that the transient, but not the sustained, component was powerfully inactivated at potentials positive to À60 mV ( Figure 5A). Analysis of the activation and inactivation properties of the transient outward current revealed relationships well described by single Boltzmann functions, with a half-maximal voltage of activation of À9.1 ± 2.1 mV (n = 10) and inactivation of À72.7 ± 0.8 mV (n = 42) (Figure 5C). Notably, the transient outward current could be rapidly inactivated when the membrane was stepped to relatively depolarized potentials (t onset of inactivation at À40 mV = 5.49 ± 0.38 ms, n = 9) and recovered from inactivation following membrane repolarization with fast kinetics (repolarization to À110 mV, t recovery = 17.3 ± 2.2 ms, n = 10; Figure 5D). Consistent with our currentclamp recordings, the transient outward current was substantially attenuated by the application of 4-AP (61.3% ± 1.9% reduction, n = 5; Figure 5E). These data reveal that CINs express a prominent I A , a fractional I KD , and a voltage-gated potassium current, a finding that supports previous observations made from acutely dissociated striatal cholinergic neurons. 60,61 To examine the properties and distribution of the voltage-gated potassium channel underlying this conductance, we excised outside-out patches from identified CINs. The general properties of ensemble potassium channel activity were similar in patches excised from somatic and dendritic sites ( Figure 6). In each patch, a transient outward current that rapidly decayed to a fractional steady state was apparent, which could be inactivated by a brief voltage pre-pulse (peak amplitude = 237.0 ± 18.3 pA; 10%-90% rise-time = 0.39 ± 0.01 ms; t decay = 10.24 ± 0.35 ms; steady state = 24.9 ± 2.5 pA, n = 67 patches; with 50 ms pre-pulse to À40 mV: peak amplitude = 33.5 ± 3.2 pA, steady state = 18.3 ± 2.1 pA, n = 37; Figures 6A, 6E, and 6F). The activation process and the fractional contribution of delayed rectifier channel activity were found to be similar to those of macroscopic currents ( Figures 6B and 6E). Notably, however, the ensemble I A channel activity recorded from dendritic patches first activated at relatively hyperpolarized potentials and exhibited a half-maximal (M) Quantification of the somatic (black symbols) and dendritic (blue symbols) current required to generate suprathreshold sEPSPs under the indicated conditions. Note that 4-AP normalizes the disparity between somatic and nAB-dendritic sites. Pooled data represent the mean ± SEM. See also Figure S5. activation of À23.8 ± 3.1 mV (n = 7), significantly different from the currents recorded from somatic patches (À4.8 ± 4.6 mV, n = 8, Mann-Whitney test, p = 0.0093; Figure 6C). In contrast, the voltage-dependence of inactivation of ensemble I A channel activity was similar between somatic and dendritic patches but notably negatively shifted in comparison to nucleated patch recordings, suggesting that the voltage-dependence of inactivation is powerfully regulated by intracellular factors. Pooled data revealed that I A potassium channels were expressed at high densities, with >600 pA of peak outward current generated in some patches ( Figures 6D and 6E). We observed no somato-dendritic gradient in the distribution of potassium channels, finding high densities at dendritic (I A = 378.2 ± 35.9 and I KD = 36.6 ± 4.8 pS mm À2 ) and somatic sites (I A = 453.3 ± 59.2 and I KD = 53.1 ± 8.0 pS mm À2 ), calculated assuming a patch area of 4.5 mm 2 and a reversal potential of À90 mV. This dendritic density of I A channels is $5-fold larger than that observed from other central neurons, examined using similar recording methods. [62][63][64][65][66] Neuromodulation of a transient potassium conductance The defining role of potassium channels in the voltage-dependent control of the excitability of CINs suggests a pivotal site for neuromodulation. We therefore next used excised patch techniques to explore whether the transient potassium conductance was modulated by the exogenous application of neuromodulators. 48 As the synchronized activation of CINs has been shown to dramatically enhance the synaptic release of dopamine, through the direct activation of dopaminergic axons, 26 we first examined whether the properties of I A were modulated by dopamine. Nucleated patch recordings, however, revealed that the application of dopamine (100 mM) did not alter the amplitude of the transient outward current evoked by voltage test steps, delivered from a holding voltage at which most channels were available for activation or from a holding voltage set close to the half-maximum voltage of inactivation (V H = À110: control = 1.96 ± 0.18 nA, dopamine = 1.99 ± 0.18 nA, t test, p = 0.588; V H = À80 mV: control = 1.69 ± 0.18 nA, dopamine = 1.64 ± 0.17 nA, t test, p = 0.257; Figures S6A-S6C). Furthermore, the construction of activation and inactivation curves revealed that the bath application of dopamine had a minimal impact on the voltagedependent gating of the transient potassium conductance (voltage of half-maximal inactivation = À73.4 ± 1.7 mV, n = 8; activation = À10.6 ± 2.7 mV; Figures S6A-S6C). A defining feature of CINs is the expression of neurokinin-1 (NK1) receptors,   Article which are thought to be endogenously activated by the synaptic release of SP from the extensive intra-striatal axon collaterals of D1-MSNs. 21,47,67,68 To explore whether SP signaling modulated the transient potassium conductance, we applied SP (0.2-1 mM) to nucleated patches. When patches were held at the holding potential most of the channels were available for activation (V H = À110 mV), bath application of SP did not alter the amplitude of the transient potassium current (control = 2.47 ± 0.18 nA, SP (0.2 mM) = 2.51 ± 0.19 nA, Wilcoxon test, p = 0.424, n = 20; control = 2.10 ± 0.11 nA, SP (1 mM) = 2.18 ± 0.10 nA, Wilcoxon test, p = 0.497, n = 12). In contrast, when nucleated patches were held at a potential close to the half-maximal voltage of inactivation of the transient potassium current (V H = À80 mV), the application of SP significantly reduced the amplitude of the transient potassium current (control = 1.96 ± 0.17 nA, SP (0.2 mM) = 1.46 ± 0.17 nA, Wilcoxon test, p = 0.0005; control = 1.38 ± 0.12 nA, SP (1 mM) = 0.96 ± 0.14 nA, t test, p = 0.0018). These data suggest that SP modulates the voltage dependency of inactivation. To test this, we injected families of voltage command steps to generate steady-state inactivation relationships, under control and in the presence of SP ( Figures 7A  and 7B). The application of SP profoundly shifted the voltage dependency of inactivation, leading to a negative voltage shift of the inactivation process, without alteration of the slope of the Boltzmann relationship (control = À71.2 ± 0.9 mV, SP (0.2 mM)= À81.0 ± 1.4 mV, t test, p < 0.0001; control = À72.2 ± 1.1 mV, SP (1 mM) = À84.6 ± 1.6 mV, t test, p < 0.0001; Figures 7A, 7B, and S6D-S6F). Furthermore, analyses of the voltage dependency of activation revealed that SP also led to a negative shift of the activation process, when compared with recordings made under control conditions (control = À9.1 ± 2.1 mV, SP (1 mM) = À23.4 ± 1.5 mV, t test, p < 0.0001; Figure S7).
We next explored the physiological impact of this peptidergic modulation of the transient potassium current, by first testing in nucleated patches whether the I A current evoked in response to voltage waveforms mimicking the time course of EPSPs was modulated by SP (voltage waveform: t rise = 3 ms, t decay = 30 ms; Figures 7C and S8A-S8F). Under control conditions, EPSP-shaped voltage waveforms evoked overlapping, fractional, and inward and outward currents, consistent with our findings made under the current clamp (peak outward current = 253.4 ± 32.3 pA; TTX (1 mM) subtracted inward current = À45.6 ± 8.6 pA, n = 5; Figures S8A and S8B). When the sodium current was pharmacologically blocked, EPSP-shaped voltage waveforms generated a large transient outward current, which was substantially attenuated by the application of 4-AP (5 mM, con-trol = 239.0 ± 24.2, 4-AP = 44.5 ± 8.3 pA, t test, p < 0.0001; Figures 7C and 7D). Analyses revealed that the I A current evoked by EPSP-shaped voltage waveforms was highly voltage dependent, increasing in amplitude as the EPSP-shaped voltage command increased and was powerfully inactivated at depolarized holding potentials (À60 mV; Figures S8C-S8F). To test whether SP selectively modulated the transient potassium current evoked by EPSP-shaped voltage waveforms, we injected EPSP-shaped voltage waveforms from a holding potential of À80 mV or À60 mV in interleaved trials, generating a transient potassium current at À80 mV and a slower rising outward current at a holding potential of À60 mV, which exhibited properties consistent with the activation of I KD ( Figure 7E). The application of SP (1 mM) rapidly attenuated the transient potassium current generated from À80 mV but did not alter the amplitude of the slowly rising outward current evoked from À60 mV (n = 11; Figures 7E-7G). Taken together, these data reveal that SP powerfully and selectively decreases I A channel availability through neuromodulation of the inactivation process, thereby attenuating the transient potassium current evoked by EPSPshaped voltage waveforms. To confirm the receptor specificity of this effect, we explored whether specific neurokinin receptor antagonists could prevent the neuromodulation of the transient potassium current by SP. Analyses revealed that SP modulation of the inactivation process of the transient potassium conductance was attenuated by the co-application of a series of NK1receptor antagonists, aprepitant (AT), N-acetyl-L-tryptophan (NAT), and GR 203040 ( Figures 7H and 7I), with the water-soluble NK1-receptor antagonist GR 203040 (10 mM) fully antagonizing the SP modulation of the voltage-dependence of inactivation ( Figures 7H and 7I).

SP modulation of dendritic excitability
To directly explore how peptidergic neuromodulation of the transient potassium conductance influences dendritic excitability, we made simultaneous somato-dendritic recordings from rat CINs. Under control conditions, threshold sEPSPs generated at the nAB-dendritic sites exhibited a characteristic shape, where the peak of sEPSPs was decorated by a dendritic spike that rapidly repolarized, followed by the generation of a highly attenuated BPAP ( Figure 8A). The bath application of SP (1 mM) transformed the pattern of dendritic electrogenesis evoked by sEPSPs, phenocopying the effects of the potassium channel antagonist 4-AP, to reveal sEPSPs that generated powerful dendritic spikes preceding large-amplitude BPAPs (compare Figures 4J and 8A). Notably, this effect of SP was (E) The application of SP modulates outward currents generated in response to EPSP-shaped voltage waveforms delivered from a holding potential of À80 mV (blue traces), but not À60 mV (gray traces). (F) The bath application of SP selectively modulates the outward current generated in response to EPSP-shaped voltage waveforms delivered from a holding potential of À80 mV. The application period of SP is indicated by the horizontal bar.

OPEN ACCESS
Article accompanied by a small membrane depolarization, consistent with previous findings 69 (depolarization: soma = 6.6 ± 0.5 mV, dendrite = 7.2 mV ± 0.5 mV, n = 18; Figures 8A and 8B). As membrane depolarization limits I A channel availability, due to the voltage-dependent properties of inactivation ( Figures 5 and 7), we investigated whether the SP-evoked membrane depolarization influenced the transformation of dendritic excitability. To do so, we constructed families of sEPSPs under control conditions, in the presence of SP, and when the membrane depolarization evoked by SP was repolarized by the injection of negative holding current through dendritic and somatic recording pipettes to restore the membrane potential to control values (Figures 8C-8E). The application of SP significantly decreased the excitatory current required to generate suprathreshold sEPSPs. In the presence of SP, however, repolarization of the membrane potential to control levels did not restore the threshold current required to generate AP firing, which remained significantly lower than under control conditions (Figures 8C-8E). Consequently, the dramatic disparity between the amplitude of excitatory current at nAB-dendritic and somatic sites required to generate suprathreshold sEPSPs observed under control conditions was significantly blunted by the application of SP, an action that phenocopied the pharmacological blockade of I A potassium channels (threshold current ratio: control = 1.74 ± 0.09, SP = 1.24 ± 0.04, t test, p < 0.0001; Figure 8F). Furthermore, in the presence of SP, when the membrane potential was restored to control values, the waveform of suprathreshold sEPSPs was significantly modulated, with the fast repolarization of sEPSP-evoked dendritic spikes characteristically diminished and the coupling between dendritic spike generation and AP firing enhanced (Figure 8C). This modulation of dendritic excitability was a receptorspecific effect, as the application of SP in the presence of the NK1-receptor antagonist GR 203040 (10 mM) failed to alter the membrane potential, the threshold current required to generate APs, or the waveform of sEPSPs (threshold current ratio: GR 203040 = 1.54 ± 0.12, SP + GR 203040 = 1.51 ± 0.12, t test, p = 0.533; membrane potential: GR 203040 = À75.0 ± 0.8 mV, SP + GR 203040 = À75.5 ± 0.9 mV, t test, p = 0.095; Figures 8G and 8H). Thus, we conclude that SP modulation of the voltage-dependent availability of potassium channels profoundly enhances the impact of dendritic excitatory input on the AP output of CINs.

DISCUSSION
A defining anatomical feature of CINs is the large size and volume of their somata, 16,17,24,45,70 which is predicted to act as a large electrical load. Experimental and theoretical studies have demonstrated that the electrical load of the soma and dendritic tree has a significant impact on the initiation of AP firing. 71,72 In central neurons where the axon emerges from the soma, the site of the AP initiation typically occurs at an axonal site $20-40 mm from the soma, a relatively distal initiation site that arises partly because of the diminished impact of the electrical load of the soma and dendrites at the distal end of the AIS. [49][50][51][52][53]72,73 In contrast to the classical somatic emergence of the axon, our functional observation of the first occurrence of APs at the dendritic sites in striatal CINs, together with a high-resolution morphological analysis, reveals that the axon originates from a dendritic site, consistent with previous anatomical descriptions. 17,45,70 Our findings indicate that dendritic-emerging axons are typical given the frequency of first detecting APs at the dendritic sites in simultaneous somato-dendritic recordings and the number of primary dendrites of CINs. We suggest that the dendritic origin of the axon is required in CINs to allow the faithful initiation of APs due to the fact the axial resistance of the intervening dendritic cable reduces the impact of the electrical load of the soma at the axonal site of AP initiation. Indeed, we directly show that the soma acts as a significant sink for sodium channelmediated regenerative events by demonstrating the dramatic dendro-somatic attenuation of nAB-dendritic spikes, which were barely detectable at the soma. Furthermore, we find that the spread of the AIS spike to the soma is highly decremental in CINs and acts as a variable trigger for the initiation of the secondary, soma-dendritic AP spike component, resulting in a wide cell-to-cell range of AP base width.
A dendritic origin of the axon is not a unique feature of CINs, with recent work demonstrating that this anatomical arrange-ment is surprisingly common in other classes of mammalian central neurons. [55][56][57][58][59]74 Notably, in neocortical pyramidal neurons, a dendritic origin of the axon has been suggested to compensate for heterogeneity in somato-dendritic electrical load and normalize the properties of somatically recorded APs. 57 Our findings suggest such normalization does not occur in striatal CINs as the waveform of somatically recorded APs exhibited considerable cell-to-cell variability accompanying a wide heterogeneity of morphology. By contrast, we suggest that the dendritic origin of the axon facilitates the execution of dendriticcompartment-specific computations. Consistent with this idea, a dendritic origin of the axon in CA1 pyramidal neurons has been shown to allow excitatory input directed to this dendrite to be privileged for AP initiation. 58,59 In an analogous manner, our direct recordings demonstrate that the current threshold for AP initiation in CINs is lowest at AB-dendritic sites compared with the soma and greater at nAB-dendritic sites than the soma, thereby revealing that excitatory input received within the AB-dendritic tree is privileged over other dendrites for driving AP output.
Surprisingly, we find that this dendritic-compartment specificity for excitatory input efficacy is not constant but regulated in a voltage-dependent manner. This is a property of particular relevance in CINs, as these neurons exhibit prolonged periods of membrane hyperpolarization in response to salient sensory and motivational cues in vivo. 11,24,[30][31][32][33] At the hyperpolarized potentials achieved by CINs during pauses of low-frequency AP firing, 38 our results demonstrate that the efficacy of excitatory input to nAB-dendritic sites is significantly attenuated, whereas somatic or AB-dendritic input is less diminished. Thus, at hyperpolarized potentials, the privileged impact of AB-dendritic excitatory input is further enhanced. We find this attenuation of dendritic excitatory input at nAB-dendritic sites is an active phenomenon, mediated by voltage-dependent recruitment of dendritic potassium channels. Although it may seem counterintuitive that membrane hyperpolarization leads to enhanced activation of potassium channels, our direct recordings demonstrate I A potassium channels are expressed at high densities in the dendrites of CINs and their voltage-dependent activation and inactivation properties ensure enhanced channel availability at hyperpolarized membrane potentials; a finding consistent with the voltage-dependence of the K V 4.2 channel-mediated potassium conductance recorded at the whole-cell level from acutely disassociated CINs. 60,61 Given our observation of a high somato-dendritic distribution of I A channels in CINs, these data suggest that membrane hyperpolarization predominately attenuates n-AB dendritic excitability because the increased electrical distance of nAB-dendrites from the AIS exposes excitatory input to greater attenuation en route to the AIS.
When recorded under conditions of ideal voltage-clamp in excised nucleated and outside-out patches, 48,66 the kinetics of activation, inactivation, and recovery from inactivation of the I A current were rapid and capable of dynamically sculpting the waveforms of EPSP-shaped patterns of the dendritic excitatory input. Accordingly, our current-clamp recordings revealed that the pharmacological blockade of I A potassium channels dramatically enhanced the efficacy of nAB-dendritic excitatory input, an effect augmented by the unmasking sodium-channel-mediated dendritic spikes at nAB-dendritic sites, which proceeded and contributed to driving AP firing. I A potassium channels therefore act to regulate the dendritic excitability of CINs by controlling the amplitude of dendritic sEPSPs, the current threshold for AP initiation, and the time course and propagation of dendritic spikes and BPAPs. In this way, the activation of I A potassium channels by sEPSPs generated from hyperpolarized membrane potentials augments the dendro-somatic electrical load imbalances imposed by morphology.
This voltage-dependent brake of active dendritic integration is mechanistically similar to the role of I A potassium channels in the control of dendritic excitability of CA1 hippocampal pyramidal neurons, where I A recruitment gates dendritic excitability, a function that is regulated in an activity-dependent manner and subject to neuromodulation. 63,75,76 The defining role of I A potassium channels in controlling dendritic excitability of CINs therefore suggests that this channel class may be an important target for neuromodulation. Consistent with this, the properties of I A have been shown to be modulated by the pharmacological activation of protein kinase C pathways in CINs. 77 Here, we demonstrate that the application of the neuropeptide SP modulates the voltage-dependent availability of I A channels through an NK-1 receptor-mediated pathway to powerfully control the efficacy of dendritic excitatory input. Surprisingly, we found the application of SP led to a negative shift in the voltage of half-maximal activation and inactivation of I A , in an analogous manner to the mAChR modulation of the voltage-dependent gating of I A channels in MSNs. 25 Notably, the modulation of the inactivation process by SP was sufficient to ensure that little I A current was available for activation even at the hyperpolarized membrane potentials achieved during pauses of AP firing of CINs in vivo. Consequently, the application of SP phenocopied the pharmacological blockade of I A channels to enhance the efficacy of nAB-dendritic excitatory input, an action that was found to operate in concert with the well-characterized membrane potential depolarization evoked by SP mediated by cation channel opening in CINs. 78 Thus, through membrane depolarization and the modulation of the voltage-dependency of I A inactivation, SP annuls the dramatic voltage-and dendritic-compartment-specific control of the efficacy of dendritic excitatory input in CINs. As SP is synaptically released from D1-MSN neurons during periods of high-frequency AP firing, 21 through an extensive network of axon collaterals which directly innervate CINs, 68 these data suggest that this mechanism will act to negate the voltagedependent control of dendritic excitability and so the relative privileged efficacy of dendritic excitatory input to AB-dendritic sites, thereby equalizing the integration scheme of synaptic input across dendritic trees.
The functional implications of this modulation of dendritic integration will, however, depend on the pattern of excitatory inputs received in the dendritic tree of CINs during active behaviors in vivo. Our data reveal that the targeting of excitatory inputs to AB-and nAB-dendrites, as well as their temporal patterning, is critical for the synaptic control of CIN AP outputs. Previous observations have revealed that CINs receive powerful thalamostriatal and cortico-striatal excitatory synaptic inputs. 23,31,43,79 Although the dendritic specificity of these excitatory inputs is unknown and has been found to represent only a fraction of the $5,000 asymmetrical synapses formed within the dendritic tree of CINs, 31 it is tempting to speculate that the targeting of the excitatory input to distinct dendritic domains of CINs will have a significant role in neuronal computation. Notwithstanding, our data suggest that the efficacy of the excitatory synaptic input to CINs is regulated in a dendritic-domain-and voltage-dependent manner, a pattern of dendritic-compartment-specific computation powerfully modulated by NK1-receptor activation. Thus, through control of the local neuropeptide environment, D1-MSNs may modulate the CIN-mediated temporal framing of responsiveness and plasticity in the striatum.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

ACKNOWLEDGMENTS
We are grateful to J. Bertran-Gonzalez and Rowan Tweedale for comments on a previous draft of the manuscript and Rumelo Amor, Andrew Thompson, and Arnaud Gaudin for their help with procedures for the imaging and reconstruction of neurons. This work was supported by the National Health and Medical Research Council of Australia (GA59869 to S.R.W.). Imaging was performed at the Queensland Brain Institute's Advanced Microscopy Facility, generously supported by the Australian Government through the Australian Research Council LIEF grants LE100100074 and LE130100078.

AUTHOR CONTRIBUTIONS
Experimental: S.R.W., X.Z., and L.N.F. performed immunohistochemistry, confocal imaging, and neuronal reconstructions. Conceptualization and analysis: S.R.W. Writing: S.R.W. wrote the original draft manuscript, and all authors participated in writing, reviewing, and editing of the manuscript.

DECLARATION OF INTERESTS
The authors declare no competing interests.

INCLUSION AND DIVERSITY
We support inclusive, diverse, and equitable conduct of research. Article  STAR+METHODS   KEY RESOURCES TABLE   RESOURCE AVAILABILITY Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Stephen R. Williams (srw@uq.edu.au).

Materials availability
This study did not generate any unique reagents. REAGENT  Data and code availability d All data reported in this paper will be shared by the lead contact upon request. d This paper does not report original code. d Any additional information required to reanalyze the data reported in this paper is available from the lead contact on request.

EXPERIMENTAL MODEL AND SUBJECT DETAILS
Optogenetic activation Light stimuli (1.2 -5.2 mW / mm 2 ; 480 nm emission filter; light source: Xeon lamp (Olympus), or LED illuminator (Thor) were controlled by a computer-controlled fast shutter, or the trigger signal to the LED illuminator. In all experiments full-field illumination, centered at the recording site, was achieved using a 60 x (LumplanFL 1.0 NA) immersion lens (Olympus) using short duration light stimuli (2 -4 ms).

Experimental design
No strategy for randomization and/or stratification was adopted. It was not possible to perform this study blinded.

QUANTIFICATION AND STATISTICAL ANALYSIS
Values reported are mean ± standard error of the mean (SEM), unless otherwise stated, and n values represent the number of cells (individual n values are shown in the results section and in figures). Statistical significance was defined as when the probability of the data being observed under a null hypothesis was less than an alpha level of 0.05. The normality of the distribution of data was tested using a combination of Anderson-Darling, D'Agostino and Pearson, and the Shapiro-Wilk tests. Parametric data were tested with a two-tail Students' t test or ANOVA, Welch's correction was used for samples with unequal variance. The Mann-Whitney and Wilcoxon tests were used for non-parametric data. Details of statistical tests can be found in the results section. Statistical analysis was carried out using Prism 9 (Graphpad).
Somatic or dendritic electrophysiological recordings were rejected if the series resistance changed by >15 % during the experiment or exceeded 20 or 50 MU, respectively, if observed recordings were highly unstable, or if cells had a depolarized membrane potential relative to the norm.