1. Neuroscience
Download icon

Striatal fast-spiking interneurons selectively modulate circuit output and are required for habitual behavior

  1. Justin K O'Hare
  2. Haofang Li
  3. Namsoo Kim
  4. Erin Gaidis
  5. Kristen Ade
  6. Jeff Beck
  7. Henry Yin
  8. Nicole Calakos Is a corresponding author
  1. Duke University Medical Center, United States
  2. Duke University, United States
Research Article
Cited
0
Views
1,985
Comments
0
Cite as: eLife 2017;6:e26231 doi: 10.7554/eLife.26231

Abstract

Habit formation is a behavioral adaptation that automates routine actions. Habitual behavior correlates with broad reconfigurations of dorsolateral striatal (DLS) circuit properties that increase gain and shift pathway timing. The mechanism(s) for these circuit adaptations are unknown and could be responsible for habitual behavior. Here we find that a single class of interneuron, fast-spiking interneurons (FSIs), modulates all of these habit-predictive properties. Consistent with a role in habits, FSIs are more excitable in habitual mice compared to goal-directed and acute chemogenetic inhibition of FSIs in DLS prevents the expression of habitual lever pressing. In vivo recordings further reveal a previously unappreciated selective modulation of SPNs based on their firing patterns; FSIs inhibit most SPNs but paradoxically promote the activity of a subset displaying high fractions of gamma-frequency spiking. These results establish a microcircuit mechanism for habits and provide a new example of how interneurons mediate experience-dependent behavior.

https://doi.org/10.7554/eLife.26231.001

eLife digest

From biting fingernails to the daily commute, habits are sets of actions that can be completed almost without thinking and that are difficult to change or stop. Behavioral neuroscientists refer to habits as “stimulus-response” behaviors, and know that forming a new habit requires a region deep within the brain called the dorsolateral striatum. Indeed, in this region, the outgoing neurons – which make up 95% of the cells - respond differently to incoming signals in mice that have learned habits compared to non-habitual mice. However a question remained: what exactly was producing these differences?

O’Hare et al. have now found, unexpectedly, that the answer resides not in the 95% of outgoing neurons, but rather in a rare type of cell known as the fast-spiking interneuron. This cell is connected to many others and it appears to act like a conductor, orchestrating the previously identified changes in the output neurons. These findings were made using mice that had been trained to press a lever for a sugar pellet reward. Habit was measured by how long mice kept pressing even if they had just been allowed to eat their fill of pellets and the test lever was no longer dispensing pellets. Habitual mice continue to press the lever in this circumstance, while other mice do not.

O’Hare et al. found that inactivating the “conductor” cell made the output neurons respond in the opposite way to how they normally respond in habitual mice. Further experiments showed that fast-spiking interneurons were also more easily activated in habitual mice. To test whether this putative “conductor” cell was necessary for habitual behaviors, a technique known as chemogenetics was used to turn down its activity in habitual mice. Indeed, reducing activity in the conductor cell blocked the habitual behavior.

While some habits are a helpful and economical way to get through daily life, habits are also thought to be corrupted in a number of diseases such as neurodegenerative diseases, addictions and compulsions. Identifying this specific, yet rare, cell as a critical part of maintaining habits points out a new target to consider for therapies. Further work is needed before such treatments might become available to treat habit-related disorders; though O'Hare et al. are now taking steps in this direction by trying to work out how the fast-spiking interneuron changes its own activity when a habit is formed.

https://doi.org/10.7554/eLife.26231.002

Introduction

Habit formation is an adaptive behavioral response to frequent and positively reinforcing experiences. Once established, habits allow routine actions to be triggered by external cues. This automation frees cognitive resources that would otherwise process action-outcome relationships underlying goal-directed behavior. The dorsolateral region of the striatum has been heavily implicated in the formation and expression of habits through lesion and inactivation studies (Yin et al., 2004; Yin et al., 2006), in vivo recordings (Tang et al., 2007; Jog et al., 1999), and changes in synaptic strength (Shan et al., 2015). More recently, properties of the dorsolateral striatum (DLS) input-output transformation of afferent activity to striatal projection neuron firing were found to predict the extent of habitual behavior in individual animals (O'Hare et al., 2016). Despite these observations, the cellular microcircuit mechanisms driving habitual behavior have not been identified.

DLS output arises from striatal projection neurons (SPNs), which comprise ~95% of striatal neurons and project to either the direct (dSPNs) or indirect (iSPNs) basal ganglia pathways. The properties of evoked SPN firing ex vivo linearly predict behavior across the goal-directed to habitual spectrum in an operant lever pressing task (O'Hare et al., 2016). Specifically, habitual responding correlates with larger evoked responses in both the direct and indirect pathways as well as a shorter latency to fire of dSPNs relative to iSPNs. To identify a microcircuit mechanism for habitual behavior, we manipulated the striatal microcircuitry to identify local circuit elements that modulated these habit-predictive SPN firing properties (Figure 1A,B).

Figure 1 with 2 supplements see all
Striatal output reconfiguration following pharmacological inhibition of FSIs directly opposes substrates for habitual behavior.

(A) Schematic of calcium imaging approach. Top: SPN activity was evoked by electrical stimulation of cortical afferent fibers in an acute parasaggital brain slice. Bottom: Evoked SPN firing was imaged in the direct and indirect pathways simultaneously using a transgenic direct pathway reporter mouse line (left), calcium indicator dye fura-2 (middle) and two-photon laser scanning microscopy (right, see scanning vector in overlay). (B) Experimental approach. Striatal microcircuitry was manipulated in tissue from untrained animals in order to reproduce the known circuit substrate for habitual behavior (described in O'Hare et al., 2016) and thereby identify a candidate microcircuit mechanism. (C) Representative heat maps of dSPN (x) and iSPN (●) calcium transient amplitudes before (left) and after (right) pharmacological inhibition of FSIs using IEM-1460 show a selective reduction in cells with the strongest (bright red) initial responses. (D) Left: Representative SPN calcium transient waveforms before and after wash-in of IEM-1460. SPNs were grouped into ‘high-firing’ (red) or ‘low-firing’ (blue) clusters based solely on their baseline response amplitudes using a Gaussian mixture model. SPNs with strong baseline responses (red, ‘high firing’) show weaker responses after wash-in whereas those with initially weak responses (blue, ‘low firing’) are unaffected. Right: Evoked calcium transient amplitudes for all imaged SPNs before (-) and after (+) wash-in of IEM-1460. For both cell types, high-firing SPNs showed decreased responses after IEM-1460 wash-in (dSPNs: t(22) = 6.43, p=0.0000018, n = 23 cells; iSPNs: t(17) = 3.43, p=0.0032, n = 18 cells) whereas low-firing SPNs did not (dSPNs: p=0.24, n = 64 cells; iSPNs: p=0.21, n = 34 cells). (E) Linear regression and correlational analyses show that the inhibitory effect of IEM-1460 on SPN responses (post – baseline difference) is a linear function of baseline response amplitudes for both dSPNs (red; r(86) = −0.87, p=2.20×10−28, n = 87 cells) and iSPNs (green; r(51) = −0.80, p=1.59×10−12, n = 52 cells). (F) Relative pathway timing, as measured by latency to peak detection, before and after inhibition of FSIs using IEM-1460. Indirect pathway activation precedes direct pathway activation by a greater margin after wash-in of IEM-1460 (t(102) = 2.42, p=0.017, n = 52 independent dSPN/iSPN pairs). *p<0.05. Dotted error bands indicate 95% confidence interval. Error bars indicate SEM. Effects of IEM-1460 on FSI and SPN spike probability are shown in Figure 1—figure supplement 1. Electrophysiological assessment of IEM-1460’s effect on evoked multi-AP SPN responses is included in Figure 1—figure supplement 2. GMM parameters and calcium transient amplitude source data can be found in Figure 1—source data 1.

https://doi.org/10.7554/eLife.26231.003

Glutamatergic corticostriatal synapses express dopamine-dependent forms of long-lasting synaptic potentiation and depression (Shen et al., 2008), making these connections a fitting site for experience-dependent adaptation of striatal output. Although such plasticity accompanies changes in behavior, including the formation of habits (Shan et al., 2015; Nazzaro et al., 2012), it does not readily explain the finding that increased gain in the direct and indirect SPNs in habitual mice was balanced (O'Hare et al., 2016) since synaptic strengthening would occur separately on the two SPN classes through dichotomous mechanisms (Shen et al., 2008). In addition, within the DLS, habit-predictive SPN firing properties were distributed uniformly rather than in discrete subpopulations of SPNs (O'Hare et al., 2016). Because interneurons are often anatomically suited to tune SPN activity in a similarly broad manner through extensive axonal arbors (Kawaguchi et al., 1995; Tepper et al., 2010), we hypothesized that plasticity of striatal interneurons might underlie the habit-associated changes in striatal output.

Among the various interneuron types resident to the striatum (Tepper et al., 2010), parvalbumin-positive, fast-spiking interneurons (FSIs) provide the strongest source of local modulation, exerting strong, feedforward inhibition of SPNs via perisomatic GABAergic contacts onto virtually all SPNs (Gittis et al., 2010; Koós and Tepper, 1999; Koos et al., 2004; Mallet, 2005; Taverna et al., 2007; Straub et al., 2016; Szydlowski et al., 2013). Notably, FSIs are expressed in the dorsal striatum on a mediolateral gradient with the most residing in DLS (Gerfen, 1985). FSIs also preferentially innervate dSPNs relative to iSPNs (Gittis et al., 2010), suggesting a potential mechanism by which FSI-mediated inhibition could allow iSPNs to fire before dSPNs in response to coincident excitatory input. Based on these considerations, we hypothesized that FSIs might drive the habit-predictive circuit features through a disinhibitory mechanism that would promote SPN firing and a preferentially earlier activation of the direct pathway. Striatal FSI plasticity has been demonstrated through experimenter-induced activity and genetic manipulations (Mathur et al., 2013; Winters et al., 2012; Orduz et al., 2013; Gittis et al., 2011a), but it remains unknown whether dorsal striatal FSIs undergo plasticity normally in the context of experience-dependent adaptive behavior.

Using pharmacological and optogenetic manipulations, we found that striatal FSIs modulate the pathway-specific properties of DLS output that predict habitual behavior. Surprisingly though, silencing FSIs produced the opposite directionality for each habit-predictive circuit feature, suggesting that an increase, rather than decrease, in FSI activity might drive habitual behavior. Indeed, when FSI firing was evoked ex vivo by stimulation of cortical afferents, FSIs from habitual mice fired more readily than FSIs from goal-directed mice. To test the significance of this plasticity for the expression of habitual behavior, we acutely inhibited FSIs in DLS chemogenetically. Inhibiting FSIs in habit-trained mice blocked habit expression, but not lever-pressing per se, while identically-trained control subjects displayed robust habitual behavior. In vivo recordings revealed that the effects of FSI activity on striatal output appear to be more selective than previously appreciated. While FSIs exert the expected strongly inhibitory influence over DLS output, they also promote activity in a subset of SPNs that can be identified a priori based upon individual SPN firing patterns. Our results identify a mechanism for habit by which FSI strengthening reconfigures DLS output and promotes the expression of habitual behavior.

Results

Inhibiting fast-spiking interneurons drives a striatal circuit endophenotype opposite that of habitual behavior

To manipulate FSI activity, the calcium-permeable AMPA receptor (CP-AMPAR) antagonist IEM-1460, which predominantly weakens excitatory synaptic inputs onto FSIs in striatum (Gittis et al., 2011b), was used. Striatal FSIs express AMPARs lacking the GluA2 subunit, rendering them permeable to calcium (Hollmann et al., 1991), whereas SPNs do not typically express CP-AMPARs. Consistent with this difference in AMPAR subunit expression, IEM-1460 does not affect excitatory synaptic currents in SPNs but strongly decreases excitatory transmission onto FSIs (Gittis et al., 2011b). Cell-attached FSI recordings before and after exposure to IEM-1460 (50 μM) confirmed the drug’s efficacy to reduce synaptically-evoked AP firing in our acute parasagittal DLS preparation (Figure 1—figure supplement 1). To first approximate how FSIs modulate the habit-predictive properties of evoked striatal output, the same ex vivo population calcium imaging approach that identified the behavior-predictive properties (O'Hare et al., 2016) was used on tissue prepared from untrained animals (Figure 1A,B). Firing responses evoked by electrical activation of cortical afferents were measured in dozens of pathway-defined SPNs of both types simultaneously using the calcium indicator dye fura-2AM, the Drd1a-tdTomato (Ade et al., 2011) reporter, and vector-mode two-photon laser scanning microscopy (2PLSM) (Figure 1A; see Materials and methods and O'Hare et al., 2016). Action potential responses were detected by cross-correlation analysis with a template waveform that was obtained from single-action potential responses during simultaneous cell-attached electrophysiological recordings for each SPN subtype (see Materials and methods). Contamination of dSPN and iSPN datasets by interneurons was minimized by selection criteria and monitoring datasets for outliers (See Materials and methods for further details).

Firing properties of SPNs were compared within-cell before and after wash-in of IEM-1460. IEM-1460 decreased the amplitude of evoked calcium transients in both dSPNs (t(86) = 3.42, p=0.001, n = 87) and iSPNs (t(51) = 2.11, p=0.040, n = 52). IEM-1460 also changed the relative latency to fire between direct and indirect pathway SPNs by increasing the pre-existing bias in relative pathway timing whereby iSPNs tend to respond to cortical excitation more quickly than dSPNs (Figure 1F) (mean absolute latency values for dSPNs: 144.03 ± 7.08 ms ACSF, 154.33 ± 7.92 ms IEM-1460, N = 87; iSPNs: 130.31 ± 7.87 ms ACSF, 134.43 ± 8.89 ms IEM-1460, N = 52).

Upon closer inspection, the decrease in calcium transient amplitude seen at the population level appeared to be dominated by the subset of SPNs with larger baseline responses (for example, see brightest red cells before wash-in in Figure 1C). To determine whether there was selectivity for IEM-1460’s effects on SPNs with large basal responses, calcium transient amplitude was used as a feature to classify SPNs as having large or small evoked calcium transients prior to drug wash-in. Rather than specifying an arbitrary cutoff value for the transient amplitude, we used an unsupervised clustering algorithm known as a Gaussian mixture model (GMM) to separate SPNs into two clusters. Based on calibration data in this preparation demonstrating the relationship between calcium transient amplitude and number of action potentials (O'Hare et al., 2016), the GMM separated SPNs into clusters corresponding to multi-action potential (larger transients; ‘high-firing’) and single-action potential (smaller transients; ‘low-firing’) responses (Figure 1D). Compared to the use of a physiologically-based 0.05 ΔF/F0 cutoff value, the unbiased GMM classification was in 90.5% agreement. According to this pre-IEM-1460 categorization, low-firing SPNs were unaffected whereas calcium transient amplitudes of high-firing SPNs were significantly reduced by IEM-1460 (Figure 1D).

This selective relationship was also borne out by examining the relationship between basal calcium transient amplitude and the magnitude of IEM-1460 effect. Consistent with a selective inhibition of multi-action potential responses, basal calcium transient amplitudes linearly predicted the inhibitory effect of IEM-1460 in both SPN subtypes (Figure 1E). Moreover, IEM-1460 did not affect spike probability in either SPN subtype (Figure 1—figure supplement 1). These pharmacological experiments in acute brain slices indicate that IEM-1460 promotes an indirect pathway timing advantage and selectively diminishes multi-action potential evoked SPN responses.

Because the within-cell experimental design of measuring effects before and after IEM-1460 application did not exclude the possibility that changes in calcium signals occurred during the 20 min wash-in period independently of IEM-1460, we performed a separate across-group study. Brain slices were incubated with either IEM-1460 or vehicle prior to and during imaging. Group mean calcium transient amplitudes were lower in IEM-1460 relative to vehicle in both dSPNs (vehicle: 0.043 ± 0.0011 ΔF/F0, N = 202 cells; IEM-1460: 0.037 ± 0.0021 ΔF/F0, N = 72 cells; t(272) = 2.62, p=0.0093) and iSPNs (vehicle: 0.040 ± 0.0014 ΔF/F0, N = 143 cells; IEM-1460: 0.033 ± 0.0011 ΔF/F0, N = 56 cells; t(197) = 2.93, p=0.0038) and IEM-1460-treated slices showed a preference for faster indirect pathway activation relative to vehicle-treated slices (t(197) = 3.83, p=1.41×10−7, N = 143 and 56 independent dSPN/iSPN pairs). These results are generally consistent with findings from within-cell pre-post measurements.

To further test whether IEM-1460 selectively inhibited multi-spike SPN responses using methodology that did not involve inferring action potentials through calcium imaging, we used conventional electrophysiological methods to record cortically-evoked SPN firing in cell-attached mode. Brief single-pulse electrical stimuli (300–600 μs) were calibrated to elicit a stable multi-action potential response in SPNs prior to taking a baseline measurement. Responses to the same stimulus were then recorded after wash-in of IEM-1460 or vehicle. Consistent with the calcium imaging results, IEM-1460 decreased evoked SPN firing (t(7) = 2.37, p=0.029, n = 8) while vehicle had no significant effect (p=0.76, n = 8). Moreover, the same selectivity for modulating multi-action potential responses was observed in that the magnitude of IEM-1460’s effect correlated with the size of baseline responses and there was no effect on single-action potential responses (Figure 1—figure supplement 2). This result confirms that IEM-1460, which inhibits FSI firing (Figure 1—figure supplement 1), selectively reduces multi-action potential SPN responses to afferent stimulation as suggested by calcium imaging experiments (Figure 1D,E).

Altogether, this series of experiments identifies a pharmacological agent that potently inhibits FSI activity and modulates all of the habit-predictive SPN firing properties. These results were surprising for two reasons. First, rather than a blockade of FSI activity causing disinhibition of SPNs as we had hypothesized, we found that when FSI activity was reduced, SPN activity was also reduced. This result suggests that FSI activity is capable of promoting, rather than inhibiting, SPN activity at least in the acute brain slice preparation. Secondly, although IEM-1460 strikingly affected the same features of DLS output that predict the expression of habitual behavior (calcium transient amplitude in both pathways and relative pathway timing) (O'Hare et al., 2016), the directionality of these effects was opposite in all measures. Therefore, these results revise the overall hypothesis to involve a gain, rather than loss, of FSI activity as a candidate mechanism for habitual behavior.

Parvalbumin-positive interneurons selectively promote multi-action potential SPN responses to cortical excitation ex vivo

While IEM-1460 has been shown to have selective effects on the firing of FSIs in striatum, its effect of inhibiting AMPAR-mediated excitatory postsynaptic currents (EPSCs) in cholinergic interneurons (CINs) (Gittis et al., 2011b) leaves open the possibility that CINs might contribute to our observed IEM-1460 effects. To isolate the effects of FSIs, the light-activated hyperpolarizing proton pump Archaerhodopsin-3 fused to green fluorescent protein (Arch-GFP) was Cre-dependently expressed in parvalbumin (PV)-expressing cells. Pvalb-Cre mice were crossed to a line which Cre-dependently expressed Arch-GFP (See Materials and methods). Control experiments showed that, as predicted, 532 nm light drove outward currents in FSIs but not SPNs (Figure 2—figure supplement 1). Additionally, Arch expressed in PV+ cells (PV-Arch) abolished high-frequency firing of FSIs in response to somatic current injection (Figure 2—figure supplement 1) and had no effect on SPN firing in the same recording configuration (Figure 2—figure supplement 1).

To examine the contribution of FSI activity to SPN firing, cortically-evoked SPN action potentials were recorded in cell-attached mode, as in the cell-attached IEM-1460 experiments, while nearby PV+ interneurons (~0.5 mm radius from recorded SPN) were silenced in alternating trials with 532 nm light exposure (Figure 2A). In this configuration, PV-Arch effectively blocked evoked FSI firing (Figure 2B). We found that optical inhibition of PV+ interneurons reliably decreased evoked SPN firing (Figure 2C, left and middle panels). Given that IEM-1460 selectively reduced the probability of multi-action potential SPN responses, we examined whether optical inhibition of PV+ neurons had a similar selectivity. Analysis of SPN responses by trial (paired consecutive laser OFF/ON sweeps), rather than by cell, indicated that single-action potential events and failures were unaffected when FSIs were silenced (Figure 2C, right panel). Moreover, a single-exponential fit of all trial-by-trial data showed a selective contribution of FSIs to multi-spike SPN responses (Figure 2C, right panel). Consistent with the IEM-1460 results in 2PLSM calcium imaging (Figure 1D–E) and cell-attached recording (Figure 1—figure supplement 2) experiments, this optogenetic result indicates that FSIs promote multi-action potential SPN responses to cortical excitation in the brain slice and that the effects of IEM-1460 on striatal output occur primarily through a reduction of striatal FSI activity.

Figure 2 with 1 supplement see all
Ex vivo optogenetic inhibition of FSIs selectively reduces evoked multi-action potential SPN responses.

(A) Experimental setup to record cortically-evoked action potentials in cell-attached mode with interleaved optogenetic inhibition of striatal FSIs. (B) Example traces (left) and mean number of APs (right) for evoked FSI firing with laser off (grey) and on (green). 532 nm light strongly inhibits evoked FSI firing (t(5) = 15.54, p=0.000020, n = 6 cells). (C) Evoked SPN action potential firing with interleaved optical inhibition of striatal FSIs. Left: Example traces showing consecutive sweeps of evoked multi-AP SPN firing with laser off (grey) and on (green). Middle: Mean number of evoked SPN APs with laser off (grey) and on (green). Inhibition of striatal FSIs caused SPNs to fire fewer action potentials (t(12) = 3.33, p=0.0060, n = 13 cells). Right: Data in middle plot shown as individual laser ON-OFF paired trials instead of by cell. Black dashed line denotes hypothetical regression line if laser had no effect. Data were jittered in x and y with Gaussian N(0, 0.15) to visualize overlapping points. Single exponential fit consistent with specific laser effect on multi-AP SPN responses (τ=13.78, r2(127)=0.54, n = 130 paired trials from 13 cells). *p<0.05. Dotted error bands indicate 95% confidence interval. Cell type specificity of Arch expression is shown in Figure 2—figure supplement 1.

https://doi.org/10.7554/eLife.26231.007

FSIs undergo long-lasting plasticity to become strengthened with habit formation

While results thus far show that FSIs appear capable of specifically modulating habit-predictive properties of striatal output, we next examined whether FSI activity was different as a result of experience. We measured FSI synaptic and cellular electrophysiological properties in DLS brain slices prepared from habitual and goal-directed mice. Pvalb-Cre mice were bilaterally injected with AAV5-Ef1a-DIO-eYFP in the DLS to label PV+ interneurons and subsequently trained on an operant task in which they learned to press a lever for sucrose pellet rewards. Lever presses were reinforced on a random interval (RI) schedule to induce habit formation (Dickinson et al., 1983; Hilário et al., 2007) or on an abbreviated random ratio (RRshort) schedule to produce goal-directed behavior (O'Hare et al., 2016) (Figure 3—figure supplement 1). Habit was measured by evaluating the sensitivity of the learned lever press behavior to devaluation of the sucrose pellet reward. Goal-directed performance is known to be highly sensitive to outcome devaluation whereas habitual performance is less sensitive (Dickinson et al., 1983; Hilário et al., 2007; Dickinson, 1985). The sucrose pellet reward was devalued by inducing sensory-specific satiety. Specifically, mice were pre-fed with the reward pellets or, as a control for general satiety-related behavioral changes, identically-sized normal grain pellets. On separate but consecutive days, mice were alternately pre-fed 1.3 g of either the sucrose pellet reward (devalued condition) or the grain-only pellet (non-devalued condition), counterbalancing which pre-feed condition was tested first. Lever press rates were then measured during brief 3 min probe tests without reinforcement. Habitual behavior was quantified in individual mice as the log2 ratio of the devalued versus non-devalued lever press rates (normalized devalued lever press rate; NDLPr). RI-trained mice with an NDLPr ≥ 0, that is, insensitive to outcome devaluation, were considered to be habitual. RRshort-trained mice with an NDLPr < 0 were considered to be goal-directed (Figure 3—figure supplement 1, shaded regions). Mice not meeting either inclusion criterion were not used for the electrophysiological studies.

We first examined whether excitatory synaptic transmission onto FSIs was altered with habit formation. Spontaneous EPSCs (sEPSCs) were recorded in the presence of the GABAA receptor antagonist picrotoxin (50 μM). No difference was detected in sEPSC frequency or amplitude between goal-directed and habitual FSIs (Figure 3A). Additionally, paired-pulse ratios of evoked EPSCs measured at a 50 ms inter-stimulus interval were similar between groups (Figure 3B). During these recordings, we also did not observe any group differences in a number of passive membrane properties (Figure 3—figure supplement 1).

Figure 3 with 1 supplement see all
Habit formation enhances sustained high-frequency firing and cortically-evoked action potential firing in DLS FSIs ex vivo.

(A) sEPSCs in FSIs of goal-directed (orange) and habitual (purple) mice. Left: Example sEPSC traces. No effect of training was found in sEPSC frequency (middle, p=0.45, n = 12 and 10 cells) or amplitude (right, p=0.42, n = 12 and 10 cells). (B) Paired-pulse measurements in FSIs of goal-directed and habitual mice. Left: Example traces showing FSI responses to paired single-pulse stimuli spaced 50 ms apart. Right: Habitual behavior was not associated with a change in paired pulse ratio relative to goal-directed behavior (p=0.29, n = 13 and 10 cells). (C) Input-output curve showing mean FSI firing rate in response to a series of increasing current steps. Habitual FSIs fired at an overall higher rate relative to goal-directed FSIs (F(1, 22) = 5.84, p=0.024, n = 15 and 9 cells). (D) FSI response durations, i.e. the time over which FSIs sustain firing. Left: Representative traces show that goal-directed FSIs often are unable to sustain firing for the duration of a 500 ms current step whereas habitual FSIs are typically able to do so. Right: Goal-directed FSIs are less-able to sustain firing than habitual FSIs (U = 34.5, p=0.049, n = 15 and 9 cells). Goal-directed response durations were bimodally distributed (p=0.020, Hartigans’ dip test). (E) Firing rates as in (C) normalized to response duration. When accounting for response duration, no difference in firing rates is observed (p=0.25, n = 15 and 9 cells). (F) Input-output curve showing mean number of synaptically-evoked action potentials fired by goal-directed versus habitual FSIs in response to a series of increasingly strong single-pulse stimuli delivered to cortical afferent fibers. Responses recorded in cell-attached mode. Habitual FSIs fired more readily than goal-directed FSIs in response to afferent activation (F(1,22) = 4.77, p=0.040, n = 13 and 11 cells). *p<0.05. Data are represented as mean ± SEM. Additional behavioral and electrophysiological measures are included in Figure 3—figure supplement 1.

https://doi.org/10.7554/eLife.26231.009

Rather than changes in synaptic strength, we instead found robust differences in FSI firing responses to somatic current injection. FSIs from habitual mice displayed higher firing rates compared to FSIs from goal-directed mice (Figure 3C). Action potential kinetics did not appear to explain these group differences in firing rates as action potential waveforms were not appreciably different between groups (Figure 3—figure supplement 1). However, the duration over which firing could be sustained markedly differed between the two behavioral groups (Figure 3D). The majority of FSIs from goal-directed mice were unable to maintain high-frequency firing for the entire duration of the 500 ms current injection (<250 ms of firing in 10/15 cells) whereas nearly all FSIs from habitual mice maintained such activity (>450 ms firing in 7/9 cells). Interestingly, the distribution of goal-directed FSI response durations was strongly bimodal whereas that of habitual FSI response durations was not (Figure 3D). The group difference in response durations explained the difference in firing rates between FSIs of habitual and goal-directed mice since, when firing rates were normalized to the duration of firing instead of duration of the current step, there was no longer a group difference in firing rate (Figure 3E).

Habitual behavior was associated with increased FSI firing in response to somatic current injection. However, it was afferent activation that initially revealed habit-predictive striatal output properties (O'Hare et al., 2016). Therefore, in order for FSI plasticity to alter striatal output, it must be sufficient to differentially drive FSI firing in response to similar coincident synaptic excitation. FSI firing was monitored in cell-attached mode in response to electrical stimulation of excitatory afferents. We found that FSIs of habitual mice fired more readily than those from mice with goal-directed behavior (Figure 3F). This habit-related difference in FSI excitability was not readily explained by other aspects of lever pressing performance including the total number of lever presses or rewards delivered over the course of training (Figure 3—figure supplement 1). We noted the apparent bimodal distribution of total rewards delivered for goal-directed subjects (p=0.013, Hartigans’ dip test; Figure 3—figure supplement 1) and wondered if the number of rewards received by an animal was related to the similarly-distributed FSI response durations to current injection (Figure 3D). Instead, we found that response durations from both modes of the distribution were commonly found in FSIs from the same goal-directed mouse (for example, 494.7 and 180.9 ms). Together, these experiments show that FSIs undergo long-lasting, experience-dependent plasticity with habit formation and that this plasticity is sufficient to increase FSI firing.

FSI activity is required for the expression of a learned habit

Since photo-inhibiting FSIs produces striatal output properties that directly oppose those seen in habit (Figure 1), we inhibited FSIs after habit training to determine the necessity of FSI activity for expression of habitual behavior. Mice underwent habit-training protocols in the operant lever press task and then, prior to testing the degree of habitual responding, FSIs were inhibited chemogenetically. We selected a chemogenetic approach to allow for continuous modulation of activity during the 3 min probe tests which measure habitual behavior. Drd1a-tdTomato::Pvalb-Cre mice were bilaterally injected in DLS with AAV vectors Cre-dependently encoding either the inhibitory hM4D chemogenetic receptor (Armbruster et al., 2007) (PV-hM4D) or eYFP (PV-eYFP) (Figure 4A,B). Both groups underwent the same habit-promoting RI reinforcement protocol and learned similarly (Figure 4C). For both the devalued and non-devalued conditions, after each pre-feeding period and thirty minutes prior to the outcome devaluation probe tests, the hM4D agonist clozapine N-oxide (CNO, 5 mg/kg) was delivered intraperitoneally (Figure 4D).

Figure 4 with 1 supplement see all
Acute chemogenetic inhibition of FSIs in dorsolateral striatum prevents expression of a learned lever pressing habit.

(A) Diagram of coronal brain section showing tdTomato expression throughout striatum in dSPNs and expression of hM4D:2a:GFP construct in DLS. (B) Epifluorescent images of DLS showing tdTomato in dSPNs (left), GFP in PV+ cells (middle), and overlay (right). (C) Learning curves for hM4D and reporter construct-injected cohorts show that groups did not learn the task differently (p=0.70, n = 10 and 11 mice). (D) Experimental flow of devaluation testing to evaluate habit expression. Upon completion of multi-day training sessions, mice were pre-fed sucrose or grain pellets on alternating days, intraperitoneally administered CNO, and subjected to a 3 min extinction probe test 30 min later. Devalued (sucrose) and non-devalued (grain) lever press rates (LPr) are compared ratiometrically using the normalized devalued LPr (NDLPr) to assess habitual behavior:NDLPr=log2devalued LPrnondevalued LPr. (E) Quantification of habit expression in individual subjects using NDLPr. PV-hM4D mice showed less habit expression relative to PV-eYFP controls (t(19) = 2.66, p=0.016, n = 10 and 11 mice). *p<0.05. Data are represented as mean ± SEM. Effect of CNO on absolute LPr in the non-devalued condition is shown in Figure 4—figure supplement 1.

https://doi.org/10.7554/eLife.26231.011

Chemogenetic inhibition of PV+ interneurons did not affect operant behavior in general, as evidenced by indistinguishable lever press rates between groups in the non-devalued (grain pellets) condition (Figure 4—figure supplement 1). In contrast, a comparison of sensitivity to outcome devaluation between groups revealed that habit expression was suppressed in PV-hM4D mice relative to PV-eYFP controls (Figure 4E). Mean NDLPr for RI-trained PV-eYFP control mice measured at 0.46 ± 0.27, indicating that control mice were insensitive to outcome devaluation, i.e. habitual. By contrast, PV-hM4D mice, which received the same RI training schedule and showed comparable rates of lever pressing (Figure 4C), displayed a mean NDLPr of −0.60 ± 0.30. A negative NLDPr indicates sensitivity to outcome devaluation, i.e. goal-directed responding. These findings show that acute suppression of FSI activity in DLS causes habit-trained subjects to behave as though they were goal-directed.

FSIs exert an inhibitory net effect on striatal output in vivo while paradoxically promoting activity in subsets of high-bursting SPNs

To understand how chemogenetic suppression of FSI firing affects striatal activity in vivo, single unit recordings were performed in a cohort of Drd1a-tdTomato::Pvalb-Cre mice implanted in DLS with multi-electrode arrays and injected with the Cre-dependent hM4D inhibitory chemogenetic virus. Single units corresponding to both FSIs and SPNs were recorded in freely-moving mice (Figure 5A–D) for 30 min before intraperitoneal (i.p.) injection of CNO (5 mg/kg) or vehicle and during the period of 30–60 min after injection. As expected for the inhibitory hM4D receptor, CNO significantly decreased FSI firing rates compared to vehicle-injected controls (CNO: 59.61 ± 8.08% baseline; vehicle: 86.89 ± 11.66% baseline) (Figure 5E). In line with previous ex vivo (Koós and Tepper, 1999; Koos et al., 2004) and in vivo (Mallet, 2005; Gittis et al., 2011b) studies, we further found that suppressing FSI activity caused an overall increase in SPN firing (i.e. disinhibitory effect) relative to vehicle (CNO: 472.00 ± 149.12%; vehicle: 188.02 ± 45.94%; Figure 5F).

Figure 5 with 1 supplement see all
Chemogenetic inhibition of FSIs in DLS exerts a strongly disinhibitory net effect and selective excitatory effect on striatal output.

(A) Locomotion before and after CNO or vehicle administration. Left: Example 3D traces showing head position during 30 min recordings before and after i.p. injection of CNO (blue) or vehicle (orange). Right: group-wise quantification of distance travelled shows that CNO- and vehicle-treated subjects did not respond differently to i.p. injections (p=0.16 for interaction of time and treatment, n = 6 and 7 mice). Subjects non-specifically decreased locomotor activity following the i.p. injection procedure (F(11,1) = 49.01, p=2.27×10−5, n = 6 and 7 mice). (B) Representative single-unit waveforms classified as FSIs (top, green) and SPNs (bottom, purple). (C) Waveform properties used for cell type classification. Left: FSI waveforms display a shorter spike width relative to those of SPNs (t(64) = 30.67, p=5.53×10−40, n = 23 FSIs and 43 SPNs). Right: FSIs display higher firing rates than SPNs (t(64) = 4.32, p=0.000056, n = 23 FSIs and 43 SPNs). (D) Classification of single units as FSIs (green) or SPNs (purple) by spike width and firing rate. (E) Time course showing FSI firing rates before (white background) and after (tan background) i.p. injection of CNO (blue) or vehicle (orange). CNO injection decreased FSI firing rate relative to vehicle (interaction between drug and time: F(5,105) = 2.51, p=0.034, n = 13 and 10 FSIs). (F) SPN responses to CNO or vehicle as in (E). CNO injection increased SPN firing rate relative to vehicle (interaction between drug and time: F(5,205) = 2.63, p=0.025, n = 23 and 20 SPNs). (G) Linear regression of fold-change (log2 post/pre) in firing rate after CNO (left) or vehicle (right) injection against the baseline fraction of ISIs in the gamma frequency band. SPNs with higher fractions of gamma-frequency ISIs at baseline are more likely to decrease firing rate when FSIs are inhibited with CNO (r(22) = −0.59, p=0.0032, n = 23 cells) whereas vehicle caused no change in firing rate that could be predicted by baseline fraction of gamma ISIs (p=0.92, n = 20 cells). *p<0.05. Data are represented as mean ± SEM. Example units before and after CNO shown in Figure 5—figure supplement 1.

https://doi.org/10.7554/eLife.26231.013

In contrast to the straightforward effect of CNO on FSI activity, the effect of CNO injection on SPNs was far more variable. Post-CNO SPN firing rates ranged from 32.5% to 2511.1% of baseline (CV = 147%) with 26% of SPNs displaying negative modulation. In acute slice experiments, FSIs had displayed an unexpected and selective effect of promoting multi-action potential responses (Figures 1D,E and and 2C) but not otherwise affecting spike probability (Figure 1—figure supplement 1). To assess whether FSIs also promoted activity in identifiable subsets of SPNs in vivo, we analyzed the baseline firing patterns in single SPNs prior to CNO injection. SPN spiking was categorized into discrete frequency bands by deriving instantaneous firing rate from interspike intervals (ISIs) and was then normalized to total number of ISIs for each single unit. This analysis defined the fraction of ISIs corresponding to each frequency band for each SPN and was independent of local field potentials.

We found that the baseline (pre-CNO) fraction of ISIs falling within the highest rate frequency band, gamma-frequency (30–100 Hz), linearly predicted how firing rates in individual SPNs changed when FSI activity was suppressed (Figure 5G, left; see Figure 5—figure supplement 1 for example units). That is, the higher the fraction of gamma-frequency spikes an SPN fired, the more likely it was to fire less when FSIs were chemogenetically inhibited. No such relationship was observed in response to vehicle (Figure 5G, right).

Since neurons with higher firing rates would be expected to have shorter ISIs in general, we examined the possibility that the fraction of gamma ISIs in SPNs might simply relate to mean firing rate. However, we found that the proportion of gamma-frequency ISIs was unrelated to mean firing rate in baseline single unit SPN recordings before either CNO or vehicle administration (pre-CNO: p=0.25, n = 23; pre-vehicle: p=0.28, n = 20). Additionally, we found that SPNs fire significantly more gamma-frequency spikes than expected by Poisson processes matched to firing rate (pre-CNO: t(44) = 5.76, p=7.67×10−7, n = 23 SPNs and rate-matched simulations; pre-vehicle: t(38) = 8.24, p=5.59×10−10, n = 20 SPNs and rate-matched simulations). Whereas baseline firing rates non-specifically predict fold change in firing rate after both CNO and vehicle injection (CNO: r(22) = −0.61, p=0.0022, n = 23; vehicle: r(19) = −0.45, p=0.045, n = 20), the excess probability of gamma-frequency ISIs (observed – expected) specifically predicts rate modulation after CNO (r(22) = −0.52, p=0.011, n = 23) but not vehicle (r(19) = 0.045, p=0.85, n = 20). Directly comparing correlation coefficients using Fisher r-to-z transformations showed that baseline firing rates did not predict SPN rate modulation by CNO better than by vehicle (z = −0.68, p=0.25) whereas baseline excess gamma specifically predicted modulation by CNO (z = −1.88, p=0.030). Therefore, gamma-frequency spiking represents a feature of interest in SPNs that predicts whether these output neurons will fire more or less as a consequence of reducing FSI activity.

These results demonstrate that FSIs modulate SPN activity in a more complicated manner than previously appreciated. While FSIs can have an overall strongly inhibitory effect in vivo on SPN firing as traditionally assumed, we also found evidence that they potentiate activity in a select population of SPNs that displays higher fractions of gamma-frequency spiking. This selective potentiation may be akin to a winner-take-all ‘focusing’ mechanism that increases the signal-to-noise ratio in corticostriatal transmission. According to such a mechanism, the subset of recruited SPNs would be facilitated while the less-relevant SPNs with low fractions of gamma spiking would be suppressed.

Discussion

With the recent availability of tools to study specific, genetically-defined types of neurons, critical roles for interneurons in facilitating behavioral adaptations to experience are becoming increasingly apparent. In brain regions other than the striatum, interneuron activity appears to most commonly serve as a gate for the induction of long-lasting plasticity elsewhere in the local circuitry (Kuhlman et al., 2013; Kvitsiani et al., 2013; Wolff et al., 2014; Yazaki-Sugiyama et al., 2009). Although the potential for FSIs themselves to exhibit long-lasting activity-dependent plasticity is well-documented in acute brain slice experiments (Mathur et al., 2013; Orduz et al., 2013; Hainmüller et al., 2014; Sarihi et al., 2012; Dehorter et al., 2015), we are aware of only one report in which these interneurons were found to undergo experience-dependent plasticity and contribute to the expression of an adaptive behavior or memory (Donato et al., 2013). Here we provide the first such example for striatal interneurons. We find that FSIs are a site of adaptive plasticity that drives circuit and behavioral hallmarks of habit. The habit-associated changes in FSI excitability appear distinct (Figure 3, Figure 3—figure supplement 1) from previously reported plasticity processes which included activity-induced changes in FSI-SPN synapses selectively at direct pathway SPNs (Mathur et al., 2013) and changes in firing rate related to the modulation of afterhyperolarization currents by parvalbumin expression levels (Orduz et al., 2013). Further characterizing the plasticity mechanisms we find in habit represents an important area for future research as it may reveal a useful target for pharmacological modulation of FSI activity.

The approach we took to reveal the microcircuit mechanisms for habit was to identify a potential source for the broad local DLS circuit reorganizations of SPN firing properties that strongly correlate with habit (Figure 1A,B). To do this, we first examined how FSIs influenced striatal output using a pharmacological approach that inhibits excitatory synapses on striatal FSIs (and also CINs). In brain slices from untrained mice, IEM-1460 treatment showed striking specificity in that it modulated all of the previously described (O'Hare et al., 2016) habit-predictive properties of evoked SPN firing ex vivo: gain of dSPN and iSPN responses (Figure 1D,E), and the relative timing of firing between dSPNs and iSPNs (Figure 1F). IEM-1460 also showed specificity in that it did not affect properties such as spike probability (Figure 1—figure supplement 1) that are not predictive of habit.

Unexpectedly, we found that the directionality by which FSIs modulated these properties was opposite to our original hypothesis: instead of the expected disinhibition of SPNs, silencing FSIs reduced SPN output (Figure 1B–E). FSI inhibition also altered the timing of direct and indirect pathway neuron firing in a direction that opposed the habit circuit signature (Figure 1B,F) and closely resembled previous observations in lever-press trained, goal-directed mice (O'Hare et al., 2016). This suggests that, in DLS, relative pathway timing is altered with habit formation but not with requisite goal-directed learning. Thus, the modest nature of the timing shift after pharmacological FSI blockade in untrained mice is likely due to a floor effect. Altogether, the observed effects of FSIs on SPNs lead to the prediction that an increase in FSI activity with habit formation would generate the evoked SPN properties that correlate with habit behavior (Figure 1B) (O'Hare et al., 2016). Accordingly, in habitual mice, we found that FSI firing was increased, and under the same cortical afferent stimulation conditions that evoke habit-predictive SPN firing properties (Figure 3F). This series of observations leads to a model of the striatal circuit basis for habitual behavior whereby habit formation is accompanied by a long-lasting increase in FSI excitability. In this setting, incoming cortical activity would be predicted to recruit more FSI activity that would in turn drive more firing of SPNs and shift their latencies such that direct pathway SPNs would tend to fire relatively sooner.

While anatomical and electrophysiological studies have long supported that striatal FSIs are critical for striatal circuit function (Gittis et al., 2010; Koós and Tepper, 1999; Koos et al., 2004; Mallet, 2005; Taverna et al., 2007; Straub et al., 2016; Szydlowski et al., 2013), an understanding of their specific behavioral contributions is much less developed. Prior in vivo studies have identified correlations of FSI activity with behaviors involving choice and reward-related actions (Gage et al., 2010; Schmitzer-Torbert and Redish, 2008), while more recent correlations of FSI activity with head movement velocity suggest another mechanism (Kim et al., 2014). In the present study, by chemogenetically inhibiting PV+ interneurons in vivo, we found that FSI activity in DLS is required for the expression of a learned habit (Figure 4E); an automated, reward-insensitive behavior quite different from behaviors previously studied. Previous pharmacological inactivation studies have demonstrated a role for DLS in habit expression (Packard and McGaugh, 1996; Zapata et al., 2010), indicating that general disruption of DLS activity also impairs established habitual behavior. Interestingly, in the present study, chemogenetic inhibition of FSI activity drove an overall increase in projection neuron activity (Figure 5F) which suggests that reducing FSI activity specifically may impair habit expression differently than a general inactivation of the circuitry.

While the disruption of habit by chemogenetically inhibiting FSIs supports a critical role for FSIs in this behavior, this experiment does not identify FSI plasticity as a mechanism for the expression of habit since artificially manipulating the activity of any cell that plays an otherwise critical role in the function of an implicated brain region might similarly disrupt behavior. Rather, in this study, a specific role for FSI plasticity as a mechanism for habit expression is indicated by the observations that these interneurons modulate those specific striatal output properties that correlate with habit (Figure 1 and 2) and show long-lasting changes in excitability after habit learning (Figure 3D,F).

Using opto- and chemo-genetic manipulations, we further found that FSIs, which are GABAergic, enhance activity in subsets of SPNs both in the acute slice and in vivo. Although it is unclear what if any relationship exists between the SPN subpopulations identified ex vivo versus in vivo, there exist multiple intriguing parallels. In the acute slice, only activity of those SPNs which displayed burst-like, multi-action potential responses to single-pulse stimuli (‘high-firing’ SPNs) was suppressed when FSIs were silenced (Figures 1D,E and 2C). In vivo, the activity of SPNs showing the highest fractions of gamma-frequency spiking was suppressed, instead of disinhibited, when FSI activity was chemogenetically reduced (Figure 5G). In both cases, the SPNs were distinguished by a higher propensity for burst-like firing patterns. It was further notable that the fraction of SPNs negatively modulated by reduced FSI activity was similar in both preparations (29% ex vivo compared to 26% in vivo). Conversely, we also found that less-active SPNs were not significantly modulated in the slice (Figures 1D,E and 2C) and SPNs with less gamma-frequency spiking were disinhibited in vivo when FSI activity was reduced (Figure 5G). This finding is reminiscent of a previous in vivo report that SPNs with weaker responses to cortical microstimulation displayed the most marked disinhibition upon GABAA receptor blockade (Mallet, 2005). An important future direction will be to determine whether there are unique biological properties that distinguish the subset of SPNs whose activity is promoted, as opposed to inhibited, by FSIs.

Although an activity-promoting effect of GABAergic FSIs may appear counterintuitive, previous computational (Humphries et al., 2009) and biological (Bracci and Panzeri, 2006) studies describe such a phenomenon based in part on the ‘up’ and ‘down’ resting membrane potential states of SPNs that straddle the chloride reversal potential (ECl-). While a voltage-dependent excitatory effect of GABA would not necessarily affect spike probability due to a concurrent decrease in membrane resistance and the disparity between ECl- and spike threshold, such an effect could boost the glutamate-driven depolarization of an SPN in its down state (Humphries et al., 2009; Bracci and Panzeri, 2006). Although disynaptic interneuron microcircuitry is a more common mechanism for disinhibitory effects of interneurons in other brain regions (Wolff et al., 2014; Lovett-Barron et al., 2012), some of our observations such as the influence of FSIs on SPN initial latency to fire (Figure 1F) are not consistent with the time delay necessitated by a disynaptic microcircuitry. For this reason, we instead favor a monosynaptic mechanism whereby properties of SPN resting membrane potential and firing patterns interact to yield activity-promoting effects of FSIs on SPN subsets.

Based on the previous observation that habit-predictive striatal output properties are relatively uniformly distributed when elicited by strong bulk stimulation of cortical afferents (O'Hare et al., 2016), it became apparent that habit-related adaptations of DLS broadly augment the propagation of cortical excitation into the basal ganglia. To confer specificity for certain actions, additional circuit dynamics would ostensibly be required. We hypothesized that such specificity could arise from the activation of subsets of task-specific cortical neuron projections that would in turn activate task-specific SPNs (Rothwell et al., 2015; Carelli and West, 1991; Gremel et al., 2016). Indeed, recent evidence suggests that spatially-clustered SPN activity encodes information relevant to locomotor behavior (Barbera et al., 2016). In habits, one possible mechanism then is that task-specific cortical commands drive (Smith et al., 2012), or at least initiate (Berke et al., 2004), high-frequency firing in a cluster/subset of SPNs that would then be preferentially excited by FSIs. Additionally, in such a mechanism, feed-forward inhibition of less-active SPNs (Mallet, 2005) by FSIs might then serve as a selective filter to further enhance signal-to-noise ratio in corticostriatal transmission. One testable prediction of this model is that different behaviors would reveal different subsets of gamma-rich SPNs whose activity is promoted by FSIs.

Lastly, it is notable that FSIs are also implicated in some pathological settings associated with compulsive behavior. For example, fewer striatal FSIs, as determined by parvalbumin-immunopositivity, have been observed in human brains from individuals with Tourette’s syndrome (Kalanithi et al., 2005) and mouse brains in a model of OCD-like behavior (Burguière et al., 2013). OCD is highly comorbid in Tourette’s syndrome (Sheppard et al., 1999) and disrupted habit learning has been implicated in pathological compulsivity in a variety of settings (Graybiel, 2008; Everitt and Robbins, 2005; Gerdeman et al., 2003). Interestingly, since both of the above studies defined FSIs by parvalbumin immunoreactivity, an intriguing alternative view of those results is that parvalbumin levels are below detection threshold but cell number is not necessarily reduced. Lower parvalbumin levels are associated with a hyperexcitable FSI phenotype (Orduz et al., 2013), which is akin to the direction of FSI plasticity we associate with habit in the present study. Thus, the finding of increased FSI excitability as a plasticity mechanism driving habitual responding also yields new insights to the potential mechanistic relatedness of habit and compulsion.

Materials and methods

Animals

All experiments were carried out under approved animal protocols in accordance with Duke University Institutional Animal Care and Use Committee standards. Mice were 2–4 months of age, in C57Bl/6 genetic background, and were hemi-/heterozygous for all transgenes. Drd1a-tdTomato line 6 BAC transgenic mice were generated in our laboratory (RRID: IMSR_JAX:016204) (Ade et al., 2011). To optically inhibit PV+ interneurons, a mouse line expressing Cre under control of the Parvalbumin (Pvalb) promoter (RRID:IMSR_JAX:012358) was crossed to the Ai35D line from Jackson Laboratory which Cre-dependently expressed Arch3.0-GFP (RRID:IMSR_JAX:012735). To target PV+ interneurons with Cre-dependent viral vectors, the Drd1a-tdTomato mouse line was crossed to the Pvalb-Cre line to produce experimental progeny hemizygous for Drd1a-tdTomato and heterozygous for Pvalb-Cre. For identification of PV+ neurons in 2PLSM calcium imaging experiments, the Pvalb-Cre mouse line was crossed to the Ai9 line (RRID:IMSR_JAX:007909) which Cre-dependently expressed tdTomato.

Viral vectors

The CAG-FLEX-rev-hM4D:2a:GFP plasmid was provided by the Sternson laboratory at Janelia Farm (Addgene #52536). UNC Viral Vector Core packaged this plasmid into AAV 2/5 and also provided AAV2/5-EF1a-DIO-eYFP. All viral aliquots had titers above 1 × 1012 particles/mL.

Intracranial viral injections

Stereotaxic injections were carried out on 2–3 month old Drd1a-tdTomato::Pvalb-Cre mice under isoflurane anesthesia (4% induction, 0.5–1.0% maintenance). Meloxicam (2 mg/kg) was administered subcutaneously after anesthesia induction and prior to surgical procedures for postoperative pain relief. Small craniotomies were made over the injection sites and 1.0 μL virus was delivered bilaterally to dorsolateral striatum via a Nanoject II (Drummond Scientific) at a rate of 0.1 μL/min. The injection pipette was held in place for 5 min following injection and then slowly removed. Coordinates for all injections relative to bregma were as follows: A/P: +0.8 mm, M/L: ±2.7–2.8 mm, D/V: 3.2 mm. Mice were allowed a minimum of 14 days recovery before behavioral training. For experiments involving chemogenetic inhibition of FSIs specifically in DLS, mice showing no expression or poor targeting (misses were medial to DLS) were excluded from the study prior to behavioral analysis and data unblinding. Two AAV2/5-CAG-FLEX-rev-hM4D:2a:GFP-injected mice showed expression in only one hemisphere of DLS. These mice were included for behavioral analysis and behaved no differently from bilaterally-infected mice. We note that exclusion of these two subjects does not affect the statistical significance of the result.

Lever press training

Prior to training, animals were restricted to 85–90% baseline weight to motivate learning. Lever presses were rewarded with sucrose-containing pellets (Bio-serv, F05684) and grain-only pellets (Bio-serv, F05934) were used as a sensory-specific control for satiety. Mice were trained in Med Associates operant chambers housed within light-resistant, sound-attenuating cabinets (ENV-022MD). Lever presses and food cup entries were recorded by Med-PC-IV software. During RR reinforcement, pellets were delivered every X times on average for an RR-X schedule. RI reinforcement gave a 10% probability of reward every X seconds for an RI-X schedule. Following random reinforcement training, subjects underwent devaluation testing to measure habitual behavior as previously described (O'Hare et al., 2016). When training schedule was a variable, experiments were performed with experimenter blind to training schedule.

For electrophysiological assessment of FSI properties, acute brain slices were prepared 0–24 hr after the final training session. Mice were excluded from analysis if they did not display the behavior that was expected based on training schedule. Specifically, mice that were trained to be habitual (random interval reinforcement) yet showed sensitivity to outcome devaluation (NDLPr <0) were excluded.

Brain slice preparation

Animals were anesthetized using 2,2,2-tribromoethanol and transcardially perfused with ice-cold N-Methyl-D-glucamine (NMDG) solution (Ting et al., 2014). Brains were quickly removed and 300 μm thick parasaggital sections were cut in NMDG solution using a Leica VT1200S. For electrophysiological experiments, slices recovered at 32°C in NMDG solution for 10–12 min and were then transferred to room temperature HEPES-containing holding solution (Ting et al., 2014) where they remained for the rest of the experiment. Slices remained undisturbed in the HEPES holding solution for at least one hour prior to recording. For 2PLSM calcium imaging experiments, slices were allowed to recover for approximately 45 min in NMDG solution at room temperature. Slices were then transferred to room temperature HEPES holding solution (Ting et al., 2014) shortly before bulk-loading with fura-2, AM. Cutting and holding solutions were calibrated to 305 ± 1 mOsm/L. ACSF was calibrated to 305 ± 1 mOsm/L for 2PLSM calcium imaging and 315 ± 2 mOsm/L for electrophysiological recordings with internal solutions at 295 mOsm/L. Solutions were pH 7.3–7.4 and were carbogenated to saturation at all times.

Drugs

For electrophysiological recordings, IEM-1460 was dissolved in deionized, distilled water at 100 mM and added to carbogenated ACSF for a final concentration of 50 μM. Picrotoxin was dissolved at 200 mM in DMSO and added to ACSF at 50 μM. For behavioral experiments, CNO was dissolved to 10 mg/mL in DMSO and diluted in sterile 0.9% saline solution to administer 5 mg/kg per subject with a maximum injection volume of 0.5 mL. For in vivo electrophysiological recordings, CNO and vehicle were administered on different days and in counterbalanced order.

Electrophysiological recordings

Data were acquired using an Axopatch 200B amplifier (Molecular Devices) and a Digidata 1440A digitizer (Axon Instruments). Data were digitized at 10–20 kHz and low-pass filtered at 2 kHz. Borosilicate glass pipettes were pulled to 2–5 MΩ resistance. Slices were continuously perfused with carbogenated ACSF (124 mM NaCl, 4.5 mM KCl, 1 mM MgCl2·6 H2O, 26 mM NaHCO3, 1.2 mM NaH2PO4, 10 mM glucose, 4 mM CaCl2) at a temperature of 29–31°C.

Current clamp experiments

Fast-spiking interneurons were identified by Cre-dependent fluorescence as well as their characteristically narrow action potential half-width. Current clamp (and cell-attached) recordings were carried out using a potassium methansulfonate-based internal solution (140 mM KMeSO4, 7.5 mM NaCl, 10 mM NaCl, 10 mM HEPES, 0.2 mM EGTA, 4.2 mM ATP·Mg, 0.4 mM GTP·Na3).

Voltage clamp experiments

Fast-spiking interneurons were identified by Cre-dependent fluorescence as well as previously reported ranges for input resistance and whole cell capacitance (Gittis et al., 2010). Voltage clamp recordings were carried out using a cesium methansulfonate-based internal solution (120 mM CsOH, 120 mM MeSO4, 15 mM CsCl, 8 mM NaCl, 10 mM TEA-Cl, 10 mM HEPES, 2 mM QX-314, 4 mM ATP·Mg, 0.3 mM GTP·Na3).

Cell-attached experiments

Stimuli were delivered to cortical afferent fibers at the cortical side of the internal capsule (Figure 2A) using a bipolar stimulating electrode (FHC, CBARC75). Responses in SPNs and FSIs were recorded in cell-attached configuration with voltage clamped at 0 mV. Leak current was continuously monitored to detect partial break-ins. In the event of a partial membrane rupture, leak currents increased significantly due to the voltage at which the membrane patch was clamped. In these events, data were discarded. The same potassium methansulfonate-based internal solution as in the current clamp experiments was used to enable break-in and cell type identification or further recordings after cell-attached experiments concluded. All stimuli were delivered with a 20 s inter-stimulus interval. For input-output experiments, 300 μs single-pulse stimuli were delivered with 5 sweeps per intensity, in order from weakest to strongest intensity, and cells were recorded at a consistent distance from the stimulating electrode (600–650 μm). For pre-post experiments with application of IEM-1460, 300–600 μs single-pulse stimuli were delivered to drive multi-action potential responses prior to drug wash-in. 10 sweeps were analyzed as baseline and another 10 sweeps, using the same stimulus parameters, following a 20 min wash-in period were analyzed to measure drug effect.

In vitro optical inhibition of FSIs

532 nm light was delivered from a diode-pumped solid state laser (Opto Engine) coupled to a 300 μm core, 0.39 NA patch cable which terminated into a 2.5 mm ferrule (Thorlabs Inc.). The ferrule was submerged in the perfusion chamber and positioned with a micromanipulator to illuminate a ~0.5 mm radius around the tip of the recording pipet. Laser onset coincided with electrical stimulation of cortical afferents. Laser stimulation lasted 500 ms in whole cell current clamp experiments and 1 s when monitoring synaptically-evoked responses in cell-attached mode.

In vivo single-unit recordings

Custom-made multi-electrode arrays were used for all recordings. The arrays consisted of fine-cut tungsten wires and a 6-cm-long silver grounding wire. Tungsten wires were 35 μm in diameter and 6 mm in length, arranged in a 4 × 4 configuration. The row spacing was 150 μm, and electrode spacing was 150 μm. All arrays were attached to the 16-channel Omnetics connector and fixed to the skull with dental acrylic. After hM4D viral injection into the dorsolateral striatum, the electrode arrays were lowered at the following stereotaxic coordinates in relation to bregma: 0.8 rostral, 2.75 lateral, and 2.6 mm below brain surface. Single-unit activity was recorded with miniaturized wireless headstages (Triangle BioSystems International) using the Cerebus data acquisition system (Blackrock Microsystems), as previously described (Fan et al., 2011). The chronically implanted electrode array was connected to a wireless transmitter cap (~3.8 g). During recording sessions, single units were selected using online sorting. Infrared reflective markers (6.35 mm diameter) were affixed to recording headstages to track mouse position as subjects moved freely on a raised platform. Marker position was monitored at 100 Hz sampling rate by eight Raptor-H Digital Cameras (MotionAnalysis Corp.). Before data analysis, the waveforms were sorted again using Offline Sorter (Plexon). Only single-unit activity with a clear separation from noise was used for the data analysis. In each case, a unit was only included if action potential amplitude was ≥5 times that of the noise band. FSIs and SPNs were classified on the basis of spike width and baseline firing rate (Figure 5B–D).

2PLSM calcium imaging of DLS output

Synaptically-evoked action potential firing was monitored in dozens of direct and indirect pathway SPNs simultaneously as previously described (O'Hare et al., 2016) in acute brain slices prepared from untrained Drd1a-tdTomato hemizygous mice (Ade et al., 2011) aged 2–4 months. Detailed methods are included below.

Bulk-loading of fura-2, AM

Fura-2, AM (Life Technologies, F-1221) was dissolved in a solution of 20% pluronic acid F-127 (Sigma) in DMSO by vortexing and sonication. The solution was then filtered through a microcentrifuge tube. Slices were transferred to small loading chambers with room temperature ACSF + 2.5 mM probenecid (osmolality and pH readjusted to 305 ± 1 and 7.3–7.4). 1.1 μL fura-2, AM solution was slowly painted directly onto the striatum of each slice. Additional fura-2 AM solution was added as needed to reach a final DMSO concentration of 0.1% by volume. Slices were incubated in a dark environment for 1 hr at 32–33° with continuous carbogenation of the loading chambers. The prolonged 1 hr incubation was found to be necessary for satisfactory loading in acute slices prepared from adult and aging animals.

Selecting field of view

After the incubation period, slices were moved to carbogenated HEPES holding solution. Slices remained in holding solution until used for an experiment, at which point they were moved to a recording chamber and continuously perfused with carbogenated ACSF (124 mM NaCl, 4.5 mM KCl, 1 mM MgCl2·6 H2O, 26 mM NaHCO3, 1.2 mM NaH2PO4, 10 mM glucose, 4 mM CaCl2) at a temperature of 29–31°C. To evoke SPN responses, cortical afferents were stimulated in bulk by a bipolar concentric electrode (FHC, CBARC75) placed at the dorsoanterior edge of the internal capsule (Figure 1A). A 410 × 410 μm field of view (FoV) was selected by following the cortical fibers along a diagonal ventroposterior path from the electrode at a distance of 600–650 μm. At this distance, SPN action potentials can be evoked without the cells being directly depolarized (O'Hare et al., 2016). Fura-2 and tdTomato (expressed in dSPNs) were excited simultaneously at 750 nm (fura-2 isosbestic wavelength) using a Ti:Sapphire laser (Chameleon Ultra 1, Coherent Inc.). Red and green photons were collected by separate photomultiplier tubes (PMTs) both above and below the microscope stage.

Classifying regions of interest

Regions of interest (ROIs) showing red and green were classified as dSPNs whereas green-only ROIs were classified as iSPNs (O'Hare et al., 2016). The small percentage of green-only cells which would have been striatal interneurons was partially mitigated by ignoring abnormally large ROIs which were likely to be cholinergic interneurons. FSIs comprise approximately 1% of all striatal neurons and are not present in the Drd1a-TdTomato labelled population (Ade et al., 2011). However, FSIs might be included in the ‘putative iSPN’ population. Empiric experiments using Pv-Cre x Ai9 reporter mice demonstrate that approximately 50% of FSIs are labelled with fura2-AM and pass data inclusion criteria in our experimental setting (see Figure 6). Therefore, we expect at most that approximately 1% of iSPN and none of dSPN data may represent FSIs. Because CIN and FSI firing properties are very different from SPNs, we reviewed our data sets for the presence of outliers using Robust regression and Outlier (ROUT) analysis (GraphPad Prism software) and a false discovery rate of Q = 1%. No outliers were detected in iSPN baseline amplitude, post-IEM-1460 amplitude, or latency modulation datasets of Figure 1.

Assessing potential contribution of FSIs to SPN 2PLSM datasets.

(A) Figure panel adapted from Ade et al. (2011) showing that the BAC transgenic Drd1a-tdTomato mouse that was used in 2PLSM experiments described in Figure 1 does not express tdTomato in striatal PV+ neurons. tdTomato signal (left) was not detected in neurons staining positive for the parvalbumin antigen (middle) as shown in overlay (right). Scale bars represent 25 μm. (B) Field of view for 2PLSM calcium imaging of PV+ neurons in Pvalb-Cre/Ai9 mice which Cre-dependently express tdTomato. Fura-2-loaded PV+ neurons were identified by transgenic expression of tdTomato (indicated by white arrows) and overlapping fura-2 signal (green) (C) Pie chart showing that only 50% of transgenically-labelled PV+ neurons are sufficiently loaded with fura-2 and pass other ROI criteria to be included in 2PLSM calcium imaging analysis (n = 18 cells). (D-F) Firing properties of PV+ neurons, as a function of stimulation intensity at cortical afferent fibers, including spike probability (D), amplitude of evoked calcium transients (E), and latency to fire (F) (n = 9 cells).

https://doi.org/10.7554/eLife.26231.015

Population imaging of evoked SPN calcium transients

ROIs were manually selected in ImageJ until no ROIs remained in the FoV. A matrix of ROI centroid coordinates was then imported to PrairieView to generate a linescan vector. If temporal resolution for the linescan vector fell below the minimum of 12 Hz due to an overabundance of ROIs (and thus lengthy vector), they were removed in order of increasing light intensity in the green channel. Centroid coordinates were permanently attributed to each ROI in order to retain spatial information along with SPN subtype and firing properties. Fura-2 signal was measured at each ROI along the scan path at 770 nm in response to 300 μs, 300 μA single-pulse stimulation of cortical afferents. Images were acquired at a frequency of 12–15 Hz. A diffractive optical element was used to increase signal-to-noise (SLH-505D-0.23–785, Coherent Inc.) (Watson et al., 2009). Photons were collected by a 40X, 0.8 NA LUMPlanFL water immersion objective lens and Aplanat/Achromatic 1.4 NA oil immersion condenser (Olympus). Red and green photons were directed to dedicated PMTs by a 575 nm dichroic mirror. Images were acquired via PrarieView image acquisition software and were time-locked to stimuli by Trigger Sync (Bruker Corp.). Stimuli were delivered 10 times with a 20 s interstimulis interval. For each ROI, firing properties were calculated at the single-trial level before being averaged across trials.

Data analysis

All experiments and data analyses were performed with experimenter blind to the experimental variable (e.g. viral construct, training schedule). A priori sample sizes were established based on power analyses. Data exclusion criteria and decisions were made prior to data unblinding.

Cell-attached experiments

Action potentials were detected in cell-attached mode by cross-correlating data to a template waveform. Template waveforms were composite action potentials recorded in cell-attached mode from single neurons that were positively identified as the corresponding cell type in a subsequent whole cell current clamp recording. The dot product representing a perfect fit was obtained by cross-correlating the template peak to itself. If the dot product of the data and the template peak was equal or greater than 25% of this perfect fit, then an action potential was called by a peak detection algorithm (Mathworks, Inc.). Stimulus artifacts and rare spontaneous action potentials were excluded by only analyzing data from 1 to 100 ms (FSIs) or 1–600 ms (SPNs) after stimulus delivery. Due to the sharp FSI cell-attached waveform, electrical noise was matched to the FSI template peak in some recordings. To exclude these false calls, two additional exclusion criteria were added: action potentials were excluded (1) if their amplitudes were less than 10 standard deviations of the recording minus the stimulus artifact, i.e. electrical noise and (2) if their cross-correlation peak amplitudes were less than 25% of the maximum peak in a given sweep.

Current-clamp experiments

Action potentials were detected by running a peak detection algorithm (Mathworks Inc.) on voltage velocity data with a peak threshold of 1 × 104 V/s and a minimum peak distance of 2 ms. Action potential onset and offset were defined at the intersections of the waveform with a sliding mean baseline voltage that constituted 10% of the length of the current injection. Action potential and after-hyperpolarization properties were measured up to the point when increasing current injection attenuated firing rate. Action potential half-width was defined as half the time between onset and peak voltage. Action potential amplitude was defined as the voltage difference between the sliding baseline and peak amplitude. AHP potential onset and offset were defined as the next two intersections with the sliding baseline after the action potential peak voltage. AHP amplitude was defined as the negative-most voltage between onset and offset and the AHP waveform was integrated over the sliding baseline for total voltage. AHP voltage measurements were converted to current using input resistance. Firing rates were measured in response to a series of increasing 500 ms current step amplitudes ranging from −0.4 to 2.0 nA in 200 pA intervals. Maximum response duration was defined as the longest period of sustained firing observed during this series of current injections. Rheobase was determined by identifying the 200 pA interval in which the first action potential was fired and subsequently interrogating this interval with 500 ms current injections at 10 pA resolution. Subthreshold test pulses were used to determine passive membrane properties. Input resistance was calculated as RI = dV/I. Whole cell capacitance was calculated by integrating the decay phase after current injection to measure discharged current and dividing by voltage of the current injection: ∫Vdecay/IR(Gittis et al., 2010). Series resistance was calculated by fitting a standard double-exponential function to the decay transient and deriving the time constant τ = 1/λfast to find τfast = Rs x Cwhole cell. Cells with Rs >30 megaOhms were excluded from analysis.

Voltage clamp experiments

Voltage clamp experiments assessing habit-related FSI physiology were carried out in the presence of picrotoxin (50 μM). Paired pulse ratio was calculated as log2(EPSC2/EPSC1) for first and second EPSC amplitudes. Paired stimuli were delivered 50 ms apart. Spontaneous EPSCs were recorded at Vm = −70 mV at 5X gain for 5 min per cell. Automated event detection was performed using MiniAnalysis (Synaptosoft). To validate the use of PV-Arch, 532 nm light-induced currents were recorded in FSI and SPNs in the presence of gabazine (10 μM), AP5 (50 μM), and NBQX (50 μM) to block GABAA, NMDA, and AMPA receptors, respectively.

In vivo single-unit recordings

Single unit activity was sorted into frequency bins by converting interspike intervals to instantaneous firing rates. Frequency bands were defined as Δ = 0–4 Hz, θ = 4–8 Hz, α = 8–13 Hz, β = 13–30 Hz, and γ = 30–100 Hz. The fraction of ISIs falling in a particular frequency band was calculated relative to the total number of ISIs. To compare frequency band distributions of single unit records to rate-matched Poisson processes, for each single unit with N ISIs, N points were randomly drawn from a Poisson distribution with λ set to the mean ISI (1/mean firing rate) for the corresponding single unit. This simulation was run 20 times per single unit. All 20 simulations were binned according to the described frequency band bounds and normalized counts were averaged across simulations. Since each simulated unit corresponded to a real recording with mean firing rate = 1/λ, observed and simulated data were compared via multiple paired t-tests and Bonferroni-Sidak correction for multiple comparisons. For behavioral analysis, 3D tracking data were transformed into Cartesian coordinates (x, y and z) by the Cortex software (MotionAnalysis Corp.) to allow distance calculations.

2PLSM calcium imaging

Raw frames were corrected using a drift correction algorithm (Li et al., 2008) to control for minor fluctuations in X and Y. Baseline fluorescence was measured over a 2 s sliding window to calculate change in fluorescence over baseline (ΔF/F0). Action potentials were detected using a cross-correlation approach as described for current clamp and cell-attached recordings above. The template peak was generated by simultaneous calcium imaging and cell-attached electrophysiological experiments and represented a single action potential (O'Hare et al., 2016). Detected peaks possessed dot-products at least 50% that of a perfect fit (cross-correlating template to itself). Although dSPN and iSPN calcium transients are similar in these experimental conditions (O'Hare et al., 2016), separate template peaks corresponding to the SPN subtype classification of each ROI were used.

Additional inclusion criteria beyond the cross-correlation threshold were used at the level of event detection, ROI inclusion, and slice inclusion to maximize data quality and reliability. Detected events were included as evoked responses only if they occurred within 375 ms of stimulation- any other events were excluded from analysis. Additionally, a lockout window was set in the peak detection algorithm to ensure that no event could occur within 1 s of the previously detected event. For an ROI to be included, a noise threshold was empirically determined to avoid excessive false event detection: the standard deviation of the ΔF/F0 signal could not equal or exceed 0.0575. Additionally, ROIs were excluded if fluorescence was saturating, if they had drifted from the scan path such that signal was no longer detected, if they did not respond at least once to a supra-threshold stimulus (1.5 mA) delivered 10 times at the end of the experiment, and if the ratio of non-evoked to evoked events detected at this suprathreshold stimulation intensity was greater than 4.5. These parameters were tested against multiple data sets from simultaneous calcium imaging and cell-attached recording experiments as previously reported (O'Hare et al., 2016) and were found to create an optimal balance of minimizing false detections and maximizing correct detections. Finally, slices which displayed poor loading, likely due to poor slice health or experimenter error during bulk-loading, were excluded from analysis. Each slice was required to have at least 12 SPNs of each subtype that passed all other exclusion criteria. This criterion was determined by finding the N at which coefficient of variation (CV) became a linear function of sample size, i.e. decreased only due to the CV denominator and not undersampling.

To analyze a pre-post effect within cell, such as wash-in of IEM-1460, only ROIs which were present and passed exclusion criteria both in pre and post recordings were included in analysis (See Figure 1C for matching ROIs before and after). Thus, drug effect was calculated for each individual cell using an internal baseline. Spike probability was calculated as the fraction of trials in which an evoked response was detected. Amplitude of an evoked calcium transient was calculated as the maximum ΔF/F0 in the transient waveform. Latency was calculated as the time between stimulus delivery and the time at which the cross-correlation dot-product reached half that of the perfect fit, i.e. the time of peak detection. When calculating dSPN/iSPN ratios, SEM was derived as: log2(dSPN¯iSPN¯)CVdSPN2+CViSPN2N . All analysis functions were custom-made in MATLAB unless otherwise noted.

To classify SPNs as ‘high-firing’ or ‘low-firing’ prior to application of IEM-1460 (Figure 1D), baseline calcium transient amplitudes for each SPN subtype were separated into two clusters according to a Gaussian mixture model (GMM). The effect of IEM-1460 was then calculated separately for ‘high-firing’ and ‘low-firing’ SPNs of each subtype and significance was determined using paired t-tests. In fitting the GMMs for dSPNs and iSPNs, the only user-specified input was the number of clusters (k = 2).

Statistics

F statistics were calculated using repeated measures analysis of variance. For within-cell comparisons, t statistics were calculated by paired, two-sided t-tests. Otherwise, unpaired, two-sided t-tests were used. For non-normal data sets, Mann-Whitney U tests were used. All r values were obtained using Pearson correlation analyses. Normality was measured using the Kolmogorov-Smirnoff test of the data against a hypothetical normal cumulative distribution function. Unless otherwise indicated (e.g. Figure 2C, right panel), N values denote number of replicates considered biologically distinct for statistical measures (Blainey et al., 2014) (see Table 1 for further detail). Technical replicates within a single biological sample were averaged to obtain a single value. For all statistical tests, confidence interval was set to α = 0.05.

Table 1
Details of sample sizes.

Table showing source of sample sizes for each subfigure in the study. Ex: Fig. 2C shows N = 13 cells from 11 slices and 6 mice.

https://doi.org/10.7554/eLife.26231.016
FigureCellsSlicesMice
1D13952
1E13952
1F52 independent pairs52
2B653
2C13116
3A222112
3B232112
3C242012
3D242012
3E242012
3F242312
4CN/AN/A21
4EN/AN/A21
5AN/AN/A11
5C66N/A11
5E23N/A11
5F43N/A11
5G (CNO)23N/A5
5G (Vehicle)20N/A6
1 - figure supplement 1A661
1 - figure supplement 1B13952
1 - figure supplement 2 (IEM)886
1 - figure supplement 2 (Veh)885
2 - figure supplement 1B442
2 - figure supplement 1C664
2 - figure supplement 1E552
2 - figure supplement 1F871
3 - figure supplement 1AN/AN/A16
3 - figure supplement 1BN/AN/A16
3 - figure supplement 1CN/AN/A12
3 - figure supplement 1DN/AN/A12
3 - figure supplement 1E242012
3 - figure supplement 1F242012
3 - figure supplement 1G242012
3 - figure supplement 1H242012
3 - figure supplement 1I211912
4 - figure supplement 1N/AN/A21
5 - figure supplement 2A (Pre CNO)23N/A5
5 - figure supplement 2A (Pre Vehicle)20N/A6
5 - figure supplement 2B (Pre CNO)23 + 23 rate-matched simulationsN/A5
5 - figure supplement 2B (Pre Vehicle)20 + 20 rate-matched simulationsN/A6

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
    Actions and habits: the development of behavioural autonomy
    1. A Dickinson
    (1985)
    Philosophical Transactions of the Royal Society B: Biological Sciences 308:67–78.
    https://doi.org/10.1098/rstb.1985.0010
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
    Journal of Neuroscience
    1. N Mallet
    (2005)
    3857–3869, Feedforward Inhibition of Projection Neurons by Fast-Spiking GABA Interneurons in the Rat Striatum, Journal of Neuroscience, 04, 25, 10.1523/JNEUROSCI.5027-04.2005.
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
  61. 61
  62. 62
  63. 63

Decision letter

  1. Michael J Frank
    Reviewing Editor; Brown University, United States

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

[Editors’ note: this article was originally rejected after discussions between the reviewers, but the authors were invited to resubmit after an appeal against the decision.]

Thank you for submitting your work entitled "Striatal fast-spiking interneurons drive habitual behavior" for consideration by eLife. Your article has been reviewed by two peer reviewers, and the evaluation has been overseen by a Reviewing Editor and a Senior Editor.

Our decision has been reached after consultation between the reviewers. Based on these discussions and the individual reviews below, we regret to inform you that your work will not be considered further for publication in eLife. All involved expressed great interest in your work, but reviewers noted some serious limitations that would require new experiments and/or extensive reanalysis (which were amplified during the consultation session amongst reviewers). We believe would likely require more work than is typical for an invited resubmission to eLife.

Reviewer #1:

This manuscript aims to uncover the role of striatal FSIs in habit formation. An extensive set of findings from different experimental approaches is presented, including (1) ex vivo slice work using cell-attached recordings and 2-photon imaging to examine the physiological changes in FSIs in habitual vs. non-habitual animals, (2) a behavioral study examining the effects of FSI manipulations on behavior, and (3) in vivo recordings examining the effects of FSI manipulation on striatal activity. Combining the results from these different experiments, the authors suggest that increased FSI activity accompanies habit formation, which in turn enables a specific set of SPNs to become active.

An obvious strength of the work is the combination of techniques to provide a truly multimodal perspective on FSI changes. This integrative approach to FSI function is novel and innovative: the in vitro and modeling literature has identified FSIs as powerfully shaping striatal activity, yet not much is known about their functional role in vivo and habitual behavior in general. Thus, this work has the potential to provide a major step forward in uncovering the neural substrates of habits, an issue of widespread interest.

Unfortunately, I found serious issues with most of the individual experiments in the manuscript, most obviously with the in vivo recordings and analysis (with which I am most familiar) but also in the other components. I am supportive of the authors' integrative approach in principle, and the conclusions advanced in the manuscript may well turn out to be correct; however, as things stand, there are too many missing controls, confounded analyses, and alternative interpretations for me to have confidence in the conclusions.

Specifically:

1) I am most qualified to comment on the in vivo recording studies. There are some major flaws in the experimental design (missing controls) and analysis, and as a result there are several possible alternative interpretations. In particular:

1a) It appears the in vivo recording comparison between vehicle and CNO was always run in the same order (vehicle followed by CNO). This introduces an obvious confound that time, or variables correlated with it, mediate some of the observed differences. The authors should control for this by counterbalancing the order across days, doing only one session per day and alternating between CNO and vehicle, or having a washout comparison following the CNO. A less convincing approach that would at least demonstrate awareness of this issue is to fit statistical models that include time as a regressor, using multiple regression (e.g. for Figures 1 and 5) – do the drug effects still hold?

1b) There is no description of what the animals are doing during the recordings. Striatal firing patterns are known to be correlated with behaviors such as grooming, running, and sleep (Berke et al., Neuron 2004; Burguiere et al. Curr Op Neuro 2015). The authors need to make sure that their reported effects are not due to differences in the expression of these behaviors. Were video data collected to rule this out?

1c) The identification of rhythmic signatures in those SPNs that appear to be most affected by FSI changes is potentially exciting. However, the analysis used (proportion of ISIs) to identify rhythmic activity is flawed and should be replaced by something more appropriate (see below). What the authors are doing is confounded by firing rate: more active neurons are more likely to have ISIs in higher frequency ranges. Thus, the result reported that higher firing rate neurons have more gamma ISIs is expected based on statistical properties alone. To conclude that this reflects a physiological property of interest, the authors need to use either spike-field measures of phase locking (e.g. Bokil et al. J Neurosci 2010) or compare the observed ISI distribution in specific bands to that obtained from a firing rate matched Poisson process (see e.g. Kass et al. Analysis of Neural data book).

2a) The behavioral part of the paper, on which FSI inhibition with DREADDs results in less habitual behavior, is novel but invites a simpler explanation than that provided by the authors: any disruption of DLS activity would interfere with habitual behavior. This would be consistent with the substantial literature on this from lesion and local infusions (e.g. Packard & McGaugh 1996; Yin et al. 2004).

2b) Comparing the RRshort and RI groups, it seems there is a difference in total number of lever presses, and potentially in the number of rewards earned (Figure 3—figure supplement 1). The authors should examine if differences in these behavioral measures are correlated with their physiological variables; if they do, that would call into question the interpretation that habit is the key factor here.

3) I am not an expert on the slice and 2P components of the paper, but here too I have some concerns about controls and interpretation:

3a) I found it notable that there was no control for time (through counterbalancing or washout). As a result, how can the authors be confident that results like Figure 1 do not simply reflect nonspecific changes over time such as homeostasis?

3b) If I am understanding correctly how iSPNs were identified – through the absence of DrD1a-tdTomato – this means interneurons including FSIs are included in this category. If this is correct, I don't see how the shift in direct/indirect pathway latency, and other results that claim a difference between dSPN and iSPN groups, are supported by the data. I realize this method is commonplace when you want to make a statement about dSPNs vs iMSNs and have no reason to think anything is changing in FSIs, CINs etc. But in this study we are provided explicit evidence that there are systematic changes happening in FSIs, so there is going to be a clear bias in a measurement that has those in the pool.

3c) In the IEM-1460 part of the paper, the authors could better motivate their logic about selectivity to FSIs by documenting other work that AMPA receptors in striatal FSIs lack GLuA2 subunits, while SPNs AMPARs have GluA2, and that IEM-1460 does not affect glu signaling in MSNs (most clearly demonstrated by Gittis et al. 2011 JNsci).

Otherwise their conclusion would seem to require careful voltage-clamp experiments showing the effects of this agent on identified AMPA currents in FSIs and SPNs.

3d) In the PV-Arch experiments and Figure 2—figure supplement 1 in particular: given that there is an effect of laser on SPN firing, again, how can the authors claim that FSI changes are what underlies their results? I now understand that this is in fact consistent with the mechanism the authors are proposing, but was confused by their use of the term "off-target" which to me seemed to imply that they wanted to verify that their stim affected only FSIs and not SPNs directly – which would be useful to test under synaptic blockage conditions, but that isn't what they actually do. I would find an expanded description in the text helpful.

Reviewer #2:

The study by O'Hare et al. provides timely evidence for how PV+ FSI in the dorsal striatum influence the activity of direct and indirect SPNs. The authors nicely demonstrate that the key circuit modifications performed by FSI map onto previously described activity patters for habitual-vs-goal directed behaviors. They argue that FSIs undergo plastic changes during habit formation, and influence a subset of SPNs to promote the expression of habitual behavior. I am generally enthusiastic about this study. I thought the manuscript was presented with a logical flow, and easy to understand. The experimental methods and analyses used are OK, and I do not foresee the necessity of additional experimental work.

General comments:

The authors provide evidence for acute functional reorganization of the microcircuit upon application of IEM-1460. Specifically, a change in dSPN and iSPN response magnitude and latency (Figures 1D, F). It will be useful to report the actual (ms) change in latency (instead of a log-relative measure) because it will facilitate comparison with other work (including O'Hare et al. Neuron 2016), and to assess if these changes are a result of a delay in dSPN responding or speeding of iSPN response.

Secondly, it appears that most of the arguments the authors make are essentially boil down to FSIs affecting the activity of a subset of 'high firing' or very active d/ and iSPNs. Despite this recurring observation (in pharmacological, optogenetic, ex-vivo and in vivo experiments), the authors do not speculate about the significance of this particular aspect of their results. Concretely, naive slices contain SPN responses that have a distribution of amplitudes (1D) and slices from habitual animals have a large-amplitude, indirect-SPN-delay biased responses. How, exactly, do the authors envision FSI recruitment (or lack thereof) during habit formation or goal directed behaviors shape the naive SPN responses into what is cartoonishly depicted in Figure 1B. In other words, I am asking what the authors' framework is for striatal ensamble activity that underlies habit beyond the clear cell-by-cell results they provide.

It is odd that the authors focus their physiology experiment (Figure 3 onwards) to mice that behaved in stereotyped 'habit' or 'goal-directed' ways (shaded blocks in Figure 3—figure supplement 1B). I did not notice a justification for this in the text. Is it the case that goal-directed mice in RI group (excluded dots at bottom) exhibited physiological characteristics that were similar to RR group, or alternatively naive mice? How are the cellular physiology results changed if those animals were included? In a previous publication from the group, they combined all mice, and formed a continuous distribution of goal-vs-habit mice.

[Editors’ note: what now follows is the decision letter after the authors submitted for further consideration.]

Thank you for resubmitting your work entitled "Striatal fast-spiking interneurons drive habitual behavior" for further consideration at eLife. Your revised article has been favorably evaluated by a Senior editor, a Reviewing editor, and two reviewers.

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below:

1) In this revision the authors have made several essential clarifications and performed additional control experiments and additional analyses, which have strengthened the study in important respects. Despite these improvements, several experiments in my mind are still missing the correct analyses or controls; details follow below. As before, I am enthusiastic about the integrative approach taken by the authors in combining and synthesizing across multiple experimental approaches. However, I detect an overall tendency to constrain the interpretation of specific experiments by referring to an overall synthesis or working model derived from multiple experiments. As I elaborate on below, for me to have more confidence in the results, I expect the authors to first carefully interpret the results from specific individual experiments in the Results section, without reference to an overall favored working model. Then, they are encouraged (in the Discussion) to come up with a synthesis across experiments. What I think they should avoid is writing parts of the results as if this synthesis has already been shown to be correct. This overall issue is illustrated below, particularly in my response to points 3a and 3b.

1a) For the order of vehicle and CNO applications, I assumed the worst because there was no mention of session ordering in the original manuscript, and the Figure 1 experiments employed a fixed control followed by treatment order (see my comments on this below). The fact that vehicle/CNO treatments were given on separate days and counterbalanced for order is important information that should be included in the Materials and methods, ideally with the table shown in the rebuttal, so that readers can see that in some cases, there were 10+ weeks between the two experiment days.

1b) The new inclusion of locomotion data, which show similar distance traveled for both groups, is useful and increases confidence in the interpretation of the DREADD experiments.

1c) I apologize for any confusion that my original description may have caused. My point here is quite simple: what is the evidence that relative gamma ISI density, rather than firing rate, is the relevant variable here? In the rebuttal, the authors provide a figure showing a lack of significant correlation between firing rate and the fraction of gamma-band ISIs. But this does not exclude the possibility that if the new Figure 5G would test for a correlation between firing rate (fold change; the quantity plotted on the vertical axis) and baseline firing rate instead of gamma density, the reported correlation would persist. I agree with the points made by the authors that it's entirely possible that low overall firing rates are responsible for the lack of a significant correlation between firing rate and the fraction of gamma-band ISIs, and the Poisson analysis is interesting in its own right (my original suggestion would require a further step: asking if the excess gamma ISI probability predicts firing rate change following CNO). However, these are all indirect points. The most direct way of testing whether invoking gamma ISI density is necessary is to determine if using baseline firing rate yields the same overall result or not. If so, I don't see the need for a more complicated, gamma-ISI related explanation. In the rebuttal, the authors state, "fold change correlated with gamma density, not the rate itself" but I did not see evidence for the latter claim in the manuscript: this needs to be tested and stated explicitly.

2a) Acknowledging the interpretation that the FSI manipulations performed are in effect a disruption of DLS activity, as the authors do now in the Discussion, is appropriate. However, I believe the authors' suggestion that their work is the first to show an effect of DLS manipulation on the expression (rather than acquisition) of habits is overstated. Packard and McGaugh (1996) performed lidocaine injections in the caudate nucleus following acquisition of a response strategy on a plus maze. Post-training baclofen/muscimol infusions into the DLS reduce habitual lever-pressing for alcohol (2012 Biol Psychiatry, 2016 Frontiers Behav Neuro), and has similar effects on lever pressing for cocaine (Zapata et al. J Neurosci 2010). Thus, for different operationalizations of habit and reinforcer, post-training manipulations of DLS have been shown to affect habitual (but not goal-directed) behavior.

2b) Great, inclusion of the lever press and total rewards data strengthens the results.

3a) I appreciate the authors are combining multiple approaches and lines of evidence to converge on a proposed mechanism – this is a strength of the paper. But in this case it's hard to see beyond the most direct way to rule out nonspecific changes related to time, which is simply to counterbalance the order of the treatment and control groups, or to collect washout data. The authors second observation provided is the most relevant in that it does contain a treatment vs. control comparison without the potential time confound, but this is a separate experiment, but low n, where we are not told how its timescale maps on to the main experiment, while leaving open whether the effect shown here underlies the full Figure 1 results. The other observations are more indirect still. Fixing this would require re-doing the experiment. However, if the authors feel strongly that they can argue against this confound, this flaw may be outweighed by intriguing results in other parts of the paper that collectively contribute to an emerging new view of FSI-SPN interactions.

3b) As with 3a, I note that the authors' style is to make inferences from multiple experiments to converge on an overall explanation. This is admirable and important. However, this does not absolve them of responsibility in accurately describing what can and cannot be concluded from each experiment separately. I do not think it is appropriate to take their proposed overall mechanism and use it as a source of top-down constraints on the initial interpretation of specific, individual experiments. What I want to see is careful descriptions and conclusions for each individual experiment in the Results that appropriately take the limitations of each into account. Then, the authors are encouraged to integrate the results into an overall proposed mechanism, and if they wish, use the resulting synthesis to inform interpretation of specific pieces of data – but not where those experiments and data are first described.

For this particular point, while it may be true that about 1% of striatal neurons are FSIs, they tend to be highly active, rendering them more likely to be included in recording studies (and, depending on how cells are identified in imaging studies, in those too; see e.g. Harris et al. Nat Nat Neurosci 2016). I agree with the authors that results such as Figure 1E are unlikely to result from inadvertent inclusion of a few FSIs because we are shown the full distribution of data points. However for others such as Figure 1F which simply compare means, it seems possible that a few potential FSI outliers could shift the means significantly.

3c) This additional explanation is helpful.

3d) Great, this new figure fully addresses my concern.

4) The comments related to behavior-physiology studies (Figure 3) were circumnavigated. The authors DO report an exclusion criterion, namely; "RRshort-trained mice with NDLPr <0 were considered to be goal-directed…Mice not meeting inclusion criterion were not used…". But, as part of their conclusions they routinely refer to the task as " habit promoting reinforcement protocol", and "habit-trained subjects…". Figure 3—figure supplement 1B indicates that 4 of 9 mice trained in this task do not engage in a habitual behavior, an insignificant 44% of mice.

In Figure 4, the authors exposed mice to the behavioral protocol that already produces ~44% of goal directed behavior, to argue that DREADD manipulation produces robust goal directed behavior.

So, without a devaluation test before CNO day to assess degree of habit induced, it is hard to accept the claim that CNO promoted less habitual behavior. If the physiological phenotype of the intermediate animals (trained on task, but did not exhibit habitual behavior) were known, I would be more convinced that CNO inactivation of FSI would (at least) shape an incomplete phenotype described in Figure 3. I really think Figure 4 has compelling results, and hope the authors address this concern.

But as it stands now, I am not fully convinced that the hm4D manipulation produces less habitual behavior. The author could come up with a clever, but robust statistical approach to assess/argue more convincing statistical result. For example if they serially excluded 40% of subjects in both groups, and performed the test, what fraction of the comparisons are significant? Or another formulation of shuffle tests. As I pointed out, a within-subject experiment would have been most ideal; that is each animal (in both PV-hM4D and OV-GFP groups) getting two devaluation sessions with and without CNO. If the authors cannot address this weakness via reanalysis of existing data, I think they will have to demonstrate a more convincing behavioral result with a new cohort of mice.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Striatal fast-spiking interneurons drive habitual behavior" for further consideration at eLife. Your revised article has been favorably evaluated by a Senior editor, a Reviewing editor (Michael Frank), and one reviewer.

We have sent the manuscript back to the one reviewer who had more reservations (the other reviewer was largely satisfied in the last round and we determined that you have addressed their remaining issues). The manuscript has been improved but there are now just a few remaining issues that need to be addressed before acceptance, as outlined by the reviewer below:

This resubmission contains several additional control experiments and analyses, which collectively address three major issues identified as remaining after the last round of reviews, with mixed success. One case is convincing, the second exacerbates my concerns about the authors' interpretation, and the third appears reasonable but is not incorporated into the manuscript.

– My request to control for the effects of time in the IEM-4160 slice experiments (Figure 1) is fully satisfied by the addition of new, across-group experiments for slices pre-incubated with IEM and vehicle.

– The authors claim that gamma-band ISIs are predictive of firing rate changes in SPN firing following CNO. However, pre-CNO firing rate is a numerically better predictor of such changes than excess gamma-band ISIs (r = -0.61 vs -0.45), casting doubt on the relevance of this feature given that a simpler and more predictive measure is available. The authors argue that firing rate "non-specifically" predicts such changes (i.e. for both vehicle and CNO) and that therefore excess gamma-band ISIs (which only predict post-CNO firing rates, but not post-vehicle rates) is an interesting quantity; however, these results could also be interpreted to mean that gamma-band ISIs is simply a worse predictor overall. I could imagine additional analyses, such as a multiple regression, that could help clarify this issue; however, regardless of the specific outcome, any remaining variance explained will likely be low and of unclear relevance. In addition, the authors should be aware that a comparison between two t-tests (one significant, one non-significant) does not constitute sufficient evidence for an interaction (see Nieuwenhuis et al. Nat Neurosci 2011) and report the correct test instead.

– In order to address the concern that FSIs included in the iSPN dataset (which is defined by the absence of D1R label) may contribute to the effects in Figure 1, the authors provide new data suggesting this is unlikely. At a minimum the authors need to discuss this possible confound in the manuscript, and ideally include the data shown in a supplementary Figure.

https://doi.org/10.7554/eLife.26231.018

Author response

[Editors’ note: the author responses to the first round of peer review follow.]

Reviewer #1:

[…] 1) I am most qualified to comment on the in vivo recording studies. There are some major flaws in the experimental design (missing controls) and analysis, and as a result there are several possible alternative interpretations. In particular:

1a) It appears the in vivo recording comparison between vehicle and CNO was always run in the same order (vehicle followed by CNO). This introduces an obvious confound that time, or variables correlated with it, mediate some of the observed differences. The authors should control for this by counterbalancing the order across days, doing only one session per day and alternating between CNO and vehicle, or having a washout comparison following the CNO. A less convincing approach that would at least demonstrate awareness of this issue is to fit statistical models that include time as a regressor, using multiple regression (e.g. for Figures 1 and 5) – do the drug effects still hold?

When subjects received both vehicle and CNO injections, the order of these sessions was counterbalanced and the tests occurred on separate days. Below is a summary table for reviewers’ convenience. The authors were unable to locate the source of confusion in the manuscript – please advise if there is language explicitly indicating that Vehicle was always administered before CNO.

Mouse IDCNOVehicle
9695/4/16
9718/26/168/24/16
9728/24/16
15215/4/167/25/16
15225/4/167/25/16
15235/4/16
85118/24/16
85128/24/16
85138/26/168/24/16

1b) There is no description of what the animals are doing during the recordings. Striatal firing patterns are known to be correlated with behaviors such as grooming, running, and sleep (Berke et al., Neuron 2004; Burguiere et al. Curr Op Neuro 2015). The authors need to make sure that their reported effects are not due to differences in the expression of these behaviors. Were video data collected to rule this out?

Yes, 3D motion tracking data were recorded for all mice during these open field recordings and we now include these data as part of Figure 5. We observed no effect of treatment (CNO versus vehicle in PV‐hM4D mice) on locomotion.

Given similarities in overall locomotion, it is unlikely that systematic biases in sleep or grooming could account for group‐wise differences. Our lab routinely studies the overgrooming “OCD” model referenced above. We regularly observe an inverse relationship between grooming and distance travelled (as in Ade et al., 2016) because mice stop walking when they groom, as well as sleep.

1c) The identification of rhythmic signatures in those SPNs that appear to be most affected by FSI changes is potentially exciting. However, the analysis used (proportion of ISIs) to identify rhythmic activity is flawed and should be replaced by something more appropriate (see below). What the authors are doing is confounded by firing rate: more active neurons are more likely to have ISIs in higher frequency ranges. Thus, the result reported that higher firing rate neurons have more gamma ISIs is expected based on statistical properties alone. To conclude that this reflects a physiological property of interest, the authors need to use either spike-field measures of phase locking (e.g. Bokil et al. J Neurosci 2010) or compare the observed ISI distribution in specific bands to that obtained from a firing rate matched Poisson process (see e.g. Kass et al. Analysis of Neural data book).

We thank Reviewer #1 for their diligence in considering sources of potential caveats in analysis and interpretations – we are always appreciative of opportunities to avoid erroneous conclusions. On this particular point though, there appears to be some error in interpreting the manuscript.

Reviewer #1 re‐states our result as “Thus, the result reported that higher firing rate neurons have more gamma ISIs” but this is not our observation. Instead, we only found that the degree of modulation of firing, ie fold change, correlated with gamma density, not the rate itself. However, the reviewer’s interpretation has helped us recognize a better way of labeling the y‐axis. We have removed the phrase “firing rate (fold change)” and replaced it with “rate modulation (fold change)”, as the former may have been misleading. We have also relabeled the x‐axis from “Baseline relative gamma density” to “Fraction gamma ISIs, baseline” because a higher number does not mean “more gamma ISIs” in the absolute sense. Other than these changes to graph labels, statements in the legend and main manuscript were accurate.

We have also added two further analyses for the reviewer.

1) We find that there is no significant association between firing rate and fraction of gamma range ISIs in our data.

2) We perform the suggested Poisson process comparison, and find that the result continues to support the significance of gamma frequency ISIs as a distinct SPN activity feature above and beyond what would occur by chance due to overall firing rate.

Author response image 1 shows new analysis of firing rate (not rate modulation) vs. gamma ISI density. No association is noted (r = 0.25, p = 0.11), likely because the mean frequencies in our dataset are so low to begin with (0.01‐2.5 Hz) that the sort of temporal crowding that would drive such a relationship is not significantly present.

Author response image 1

Author response image 2 compares Poisson process‐simulated distribution of ISIs (blue) compared to actual observations (red). As we now describe in Materials and methods, each individual SPN was simulated by a Poisson process matched to its own firing rate. SPNs consistently displayed more gamma‐frequency ISIs than expected by a Poisson point process (for both CNO‐ and vehicle‐treated cohorts).

Author response image 2

2a) The behavioral part of the paper, on which FSI inhibition with DREADDs results in less habitual behavior, is novel but invites a simpler explanation than that provided by the authors: any disruption of DLS activity would interfere with habitual behavior. This would be consistent with the substantial literature on this from lesion and local infusions (e.g. Packard & McGaugh 1996; Yin et al. 2004).

We thank the reviewer for highlighting the alternative interpretation that non‐specifically disrupting DLS activity could affect habit expression and we now acknowledge this point in the main text Discussion section.

We do not think this observation lessens the impact of our findings for three major reasons.

1) This is a frequent unavoidable caveat for a large number of studies of this nature. Following this 3 point summary, we describe specific examples from two recent, high impact studies to make this point concretely.

2) More importantly, the support for FSIs as a mechanism for habit expression in this study does not rest solely on this in vivoperturbation of behavior experiment, but rather the larger body of findings throughout the manuscript prior to this key test of the hypothesis. In this regard, our findings are stronger than several other high impact studies that only show in vivobehavioral perturbation as evidence for a specific role. Briefly, in addition to eliminating habit by reducing FSI activity, we find that: (1) Pharmacological and optogenetic reduction of FSI activity modulates ex vivostriatal SPN firing properties that correlate with habitual behavior (Figure 1, Figure 1—figure supplement 2, and Figure 2) (correlations originally reported in O’Hare et al., 2016). (2) Long‐lasting increases in FSI excitability are found as a consequence of habit learning in mice, in comparison to goal‐directed mice (Figure 3). In the revised Discussion section, we discuss the supporting nature of these data. We think the revised Discussion benefits from discussing this important point, and thank the reviewer for raising it.

3) An aspect of the novelty of our findings was overlooked. To date, lesions of DLS and their impact on habit have only studied lesion effects on learning/acquiring the behavior. This study, therefore, contributes a first demonstration that a post‐learning DLS manipulation can prevent the expression of an acquired habit.

Copies of key data from manuscript referenced in Point 2 above appear in Figure 1D-F, Figure 2C and Figure 1—figure supplement 1:

Specific discussion of two published comparable studies:

a) Witten, et al. (2010, Science) found that silencing cholinergic interneurons (CINs) of the nucleus accumbens (NAc) disrupts cocaine conditioning in mice. However, an alternative interpretation exists that a general disruption of NAc activity would yield a similar result. Indeed, in the second‐to‐last paragraph of the main text, the authors note that their result is more similar to that seen from general, less‐specific NAc manipulations relative to studies which chronically ablated CINs:

“Together, these data demonstrate that selectively inhibiting ChAT interneurons in the NAc with high temporal precision has the overall effect of increasing MSN activity and blocking cocaine conditioning in freely moving mammals. These behavioral results do not support conclusions arising from chronic ablation of the cholinergic interneurons (20); instead they are more consistent with interpretations arising from faster but less cellularly targeted pharmacological modulation in the NAc”.

In addition, as mentioned in point (3), unlike the above study’s context in cocaine conditioning, our result of a role of DLS in habit expression has not been reported in the literature using less‐specific manipulations, to our knowledge.

b) A recent study also examining the role of PV+ interneurons in dorsolateral striatum by Lee, et al. (2017, Neuron) uses opto‐ and chemo‐genetic inhibition of FSI activity to show a role for these cells in early‐phase Pavlovian conditioning. For example, “F” in Author response image 3 shows the study’s positive result of a change in hit rate with chemogenetic inhibitory DREADD expression in DLS FSIs (light blue).

Author response image 3

A legitimate claim could be levied on this study that this result in hit rate might also occur with a less‐specific DLS manipulation. No general manipulation of DLS activity was made to assess this possibility.

The authors do provide some demonstration of behavioral specificity by showing that their manipulation has no effect on false alarm rate (licking behavior to a non‐conditioned stimulus) in subfigure G of Author response image 3. We essentially perform the analogous control in our task, showing that chemogenetically inhibiting PV cells in DLS does not alter general lever pressing behavior by evaluating behavior in the non‐devalued probe test (Figure 4—figure supplement 1).

2b) Comparing the RRshort and RI groups, it seems there is a difference in total number of lever presses, and potentially in the number of rewards earned (Figure 3—figure supplement 1). The authors should examine if differences in these behavioral measures are correlated with their physiological variables; if they do, that would call into question the interpretation that habit is the key factor here.

The reviewer raises an important concern. A key feature of our chosen training protocol is that it allows us to make training experiences as similar as is practically possible apart from the feature of habitualness. We present the requested data in the revised manuscript (Figure 3—figure supplement 1C,D). We also refer the reviewer to our prior analysis of this question in a separate cohort in which we also found that lever press rate was insufficient as a variable to predict striatal physiological properties (O’Hare et al., Neuron, 2016).

As a side point, we also asked whether the bimodal distribution of rewards for the goal-directed subjects corresponded to the bimodally‐distributed response durations displayed by FSIs from goal‐directed mice in response to somatic current injections shown in Figure 3. However, we found this not to be the case. In fact, FSIs falling into both modes were found within single goal‐directed subjects. For example, two FSIs were recorded from one goal-directed mouse with response durations of 494.7 ms and 180.9 ms.

3) I am not an expert on the slice and 2P components of the paper, but here too I have some concerns about controls and interpretation:

3a) I found it notable that there was no control for time (through counterbalancing or washout). As a result, how can the authors be confident that results like Figure 1 do not simply reflect nonspecific changes over time such as homeostasis?

We agree with this concern and would like to clarify how we approached obtaining additional evidence to assure ourselves that the result was not spurious due to a technical artifact related to time that passed between pre and post measures. Instead of a vehicle control to account for the variable of time, we chose to perform two additional ephys experiments because this choice offers the advantage of making the same observations multiple times using orthogonal techniques. When we found an effect of IEM drug, instead of repeating the experiment with vehicle, we opted to pursue more direct electrophysiological measures (Figure 1—figure supplement 2, Figure 2) and more specific manipulations (photoinhibition, Figure 2) to prospectively test the Figure 1 result.

We believe that these three results combined remove doubt for the potential of the original observation to be an artifact – all three experiments show that FSIs increase evoked SPN responses, and specifically multi‐AP responses, but not unitary.

We also note that if the above Arch data comparing laser on vs off are presented in a manner analogous to the Figure 1E 2PLSM data (laser ON – OFF) shown earlier, the result is strikingly similar – analysis shown in Author response image 4 for reviewer’s benefit:

Author response image 4

3b) If I am understanding correctly how iSPNs were identified – through the absence of DrD1a-tdTomato – this means interneurons including FSIs are included in this category. If this is correct, I don't see how the shift in direct/indirect pathway latency, and other results that claim a difference between dSPN and iSPN groups, are supported by the data. I realize this method is commonplace when you want to make a statement about dSPNs vs iMSNs and have no reason to think anything is changing in FSIs, CINs etc. But in this study we are provided explicit evidence that there are systematic changes happening in FSIs, so there is going to be a clear bias in a measurement that has those in the pool.

The reviewer again makes an astute point. We had considered this possibility and concluded that it was not a tenable explanation. We relay our thinking here. First, CINs and some FSIs get excluded a priori from analysis based on size exclusion criteria we set up in developing the imaging approach initially (O’Hare et al., 2016). In that development phase, we used reporter cell lines to define properties of non‐SPNs and developed exclusion criteria to limit their contamination of the data set. This size criterion is noted in the current manuscript’s Materials and methods section. Second, even without any exclusion criteria, FSIs comprise only 1% of the striatal neurons, therefore, even in the most optimistic setting, the data would have to derive from at most 2‐3 cells among the 52 imaged cells that generated the observation. Such a minor fraction of cells is insufficient to generate the correlations we find (see for example Figure 1E). Moreover, if the affected cells were FSIs, they should partition unequally between SPN types, predominantly in the “nonDrd1a” iSPN group (see Ade et al., Frontiers Neurosci. 2011), and this is also not the case (e.g. Figure 1E). Third, subsequent electrophysiological experiments in SPNs confirmed by their electrophysiological signatures also show the finding, and in both SPN subgroups (Figure 1—figure supplement 2).

3c) In the IEM-1460 part of the paper, the authors could better motivate their logic about selectivity to FSIs by documenting other work that AMPA receptors in striatal FSIs lack GLuA2 subunits, while SPNs AMPARs have GluA2, and that IEM-1460 does not affect glu signaling in MSNs (most clearly demonstrated by Gittis et al. 2011 JNsci).

Otherwise their conclusion would seem to require careful voltage-clamp experiments showing the effects of this agent on identified AMPA currents in FSIs and SPNs.

The paper that the reviewer alludes to (Gittis et al. 2011 JNsci) was referenced in the original manuscript. In the revision, we further expand this discussion to specifically mention GluA2 receptors.

Revised manuscript text: “To manipulate FSI activity, the calcium‐permeable AMPA receptor (CP‐AMPAR) antagonist IEM‐1460, which predominantly weakens excitatory synaptic inputs onto FSIs in striatum24, was used. Striatal FSIs express AMPARs lacking the GluA2 subunit, rendering them permeable to calcium 25, whereas SPNs do not typically express CP‐AMPARs. Consistent with this difference in AMPAR subunit expression, IEM‐1460 does not affect excitatory synaptic currents in SPNs but strongly decreases excitatory transmission onto FSIs24.”

3d) In the PV-Arch experiments and Figure 2—figure supplement 1 in particular: given that there is an effect of laser on SPN firing, again, how can the authors claim that FSI changes are what underlies their results? I now understand that this is in fact consistent with the mechanism the authors are proposing, but was confused by their use of the term "off-target" which to me seemed to imply that they wanted to verify that their stim affected only FSIs and not SPNs directly – which would be useful to test under synaptic blockage conditions, but that isn't what they actually do. I would find an expanded description in the text helpful.

The original manuscript did address this point (copied below). In addition, we have gone ahead and performed the straightforward suggested experiment under voltage‐clamp conditions. We again thank the reviewer for the overarching interest in ensuring our study is of the highest quality.

Original Legend of Figure 2—figure supplement 1:

”…(B) Left: recording configuration to assess off-target effects of 532 nm light on SPN firing. Middle and Right: SPN responses to somatic current injection with interposed 532 nm light as in (A). Although analysis of variance showed an effect of laser on SPN firing (F(1.04, 7.27) = 9.80, p = 0.015, n = 8), this effect was due to an early frequency adaptation which SPNs are known to display in response to suprathreshold excitation1. SPN firing rates during and after laser stimulation were indistinguishable (p = 0.31, n = 8).”

Revision, new data in Figure 2—figure supplement 1:

Under pharmacological block of NMDA, AMPA, and GABAA receptors, 532 nm light‐induced currents were measured in voltage‐clamp in SPNs and striatal FSIs of PV‐Arch mice. We observed large lightdriven currents in FSIs, but not SPNs.

Reviewer #2:

General comments:

The authors provide evidence for acute functional reorganization of the microcircuit upon application of IEM-1460. Specifically, a change in dSPN and iSPN response magnitude and latency (Figures 1D, F). It will be useful to report the actual (ms) change in latency (instead of a log-relative measure) because it will facilitate comparison with other work (including O'Hare et al. Neuron 2016), and to assess if these changes are a result of a delay in dSPN responding or speeding of iSPN response.

We thank reviewer #2 for this suggestion. We now include mention of the absolute latency values in the revised. They are primarily informative for showing that both cell types have a trend toward altered latencies in which the effect is slightly greater in the dSPNs. However, we would also like to provide further explanation as to why we find the relative difference in latency more valuable than absolute values. Firstly, absolute values are influenced by electrode position in the brain slice relative to the imaged area, whereas the difference between two populations distributed about the same electrode bypasses that concern. Secondly, the absolute latency values include the superimposition of not only distance from electrode to soma, but also kinetics of calcium dye transients, and time differences related to the location of the cell along the line scan vector. Since all of these factors are consistent between the test and control conditions, significant differences in relative latency can be observed, it is just that the absolute values are of a bit less importance. Nonetheless, we are happy to provide values closer to the primary data and thank you for your interest.

In the present study, we find that IEM‐1460 did not significantly affect the latency of calcium transients in either SPN subtype; only the relative latency between these two pathways. We observed a non‐significant trend for both pathways to exhibit slower latencies with dSPNs showing a more pronounced delay. These data are now reported in the text and copied below. The finding of a significant effect for relative differences in latency, but not absolute latencies is also consistent with the results from our prior analysis in habit versus goal mice (O’Hare et al., 2016).

“IEM-1460 also changed the relative latency to fire between direct and indirect pathway SPNs by increasing the pre-existing bias in relative pathway timing whereby iSPNs tend to respond to cortical excitation more quickly than dSPNs (Figure 1F) (mean absolute latency values for dSPNs: 144.03 ± 7.08 ms ACSF, 154.33 ± 7.92 ms IEM-1460, N = 87; iSPNs: 130.31 ± 7.87 ms ACSF, 134.43 ± 8.89 ms IEM- 1460, N = 52).”

Secondly, it appears that most of the arguments the authors make are essentially boil down to FSIs affecting the activity of a subset of 'high firing' or very active d/ and iSPNs. Despite this recurring observation (in pharmacological, optogenetic, ex-vivo and in vivo experiments), the authors do not speculate about the significance of this particular aspect of their results. Concretely, naive slices contain SPN responses that have a distribution of amplitudes (1D) and slices from habitual animals have a large-amplitude, indirect-SPN-delay biased responses. How, exactly, do the authors envision FSI recruitment (or lack thereof) during habit formation or goal directed behaviors shape the naive SPN responses into what is cartoonishly depicted in Figure 1B. In other words, I am asking what the authors' framework is for striatal ensamble activity that underlies habit beyond the clear cell-by-cell results they provide.

We are pleased to learn of the interest in further speculation to address this important point and have added this to Discussion. We initially included speculative interpretation of a number of aspects of how the cellular findings would impact the broader circuitry. Those points are copied below as a reminder.

From the last paragraph of Results after describing our in vivo findings:

“While FSIs can have an overall strongly inhibitory effect in vivo on SPN firing as traditionally assumed, we also found evidence that they potentiate activity in a select population of high-gamma SPNs. This selective potentiation may be akin to a winner-take-all “focusing” mechanism that increases the signal-to-noise ratio in corticostriatal transmission. According to such a mechanism, the subset of recruited SPNs would be facilitated while the less-relevant, low-gamma SPNs would be suppressed.”

From Discussion paragraph 3:

“This in vivo finding is also consistent with a previous in vivo report that SPNs with weaker responses to cortical activity displayed the most marked disinhibition upon GABAA receptor blockade16. While directly comparing the subsets of excited SPNs identified in the slice and in vivo would not be technically straightforward, an important future direction will be to determine whether there are unique biological properties that set the excited subset of SPNs apart.”

From Discussion paragraph 5:

“Indeed, recent evidence suggests that spatially-clustered SPN activity encodes information relevant to locomotor behavior45. In habits, one possible mechanism then is that task-specific cortical commands drive46, or at least initiate47, high-frequency firing in a cluster/subset of SPNs that would then be preferentially excited by FSIs. Additionally, in such a mechanism, feed-forward inhibition of less-active SPNs16 by FSIs might then serve as a selective filter to further enhance signal-to-noise ratio in corticostriatal transmission. One testable prediction of this model is that different behaviors would reveal different subsets of high-gamma SPNs that are excited by FSIs.”

Additional discussion of model for how FSIs drive the phenomena depicted in Figure 1B:

“Unexpectedly, we found that the directionality by which FSIs modulated these properties was opposite to our original hypothesis: instead of the expected disinhibition of SPNs, silencing FSIs reduced SPN output (Figure 1B‐E). […] In this setting, incoming cortical activity would be predicted to recruit more FSI activity that would in turn drive more firing of SPNs and shift their latencies such that direct pathway SPNs would tend to fire relatively sooner.”

It is odd that the authors focus their physiology experiment (Figure 3 onwards) to mice that behaved in stereotyped 'habit' or 'goal-directed' ways (shaded blocks in Figure 3—figure supplement 1B). I did not notice a justification for this in the text. Is it the case that goal-directed mice in RI group (excluded dots at bottom) exhibited physiological characteristics that were similar to RR group, or alternatively naive mice? How are the cellular physiology results changed if those animals were included? In a previous publication from the group, they combined all mice, and formed a continuous distribution of goal-vs-habit mice.

The experiments to which the reviewer refers are single cell patch clamp experiments, and as such, the technical and biological variation from 1‐2 whole cell recordings/animal simply do not permit the individual subject correlations that the population 2PLSM imaging data enabled. Rather we needed to perform group‐wise comparisons, as is routinely done. In this case, we used the two training protocols that generally but don’t completely bias to habit v goal behavior and determined a priori that we would take the habitual mice from the habit‐biased protocol and the goal‐directed subjects from the goal‐biased protocol.

We would also like to assure the reviewer that there was no selective data exclusion. The process for determining subjects to study was as follows: the experimenter (JOH) was given trained mice, blind to behavior and training schedule, by another scientist (EG). π was given the LPr data from the probe trials conducted by EG, she (NC) calculated the Normalized Devalued Lever Press rates and then identified the subject numbers along with group assignments that were to be used for recordings to JOH.

If the reviewer wishes, the authors will include a note regarding this nuance in the Materials and methods section.

[Editors' note: the author responses to the re-review follow.]

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below:

1) In this revision the authors have made several essential clarifications and performed additional control experiments and additional analyses, which have strengthened the study in important respects. Despite these improvements, several experiments in my mind are still missing the correct analyses or controls; details follow below. As before, I am enthusiastic about the integrative approach taken by the authors in combining and synthesizing across multiple experimental approaches. However, I detect an overall tendency to constrain the interpretation of specific experiments by referring to an overall synthesis or working model derived from multiple experiments. As I elaborate on below, for me to have more confidence in the results, I expect the authors to first carefully interpret the results from specific individual experiments in the Results section, without reference to an overall favored working model. Then, they are encouraged (in the Discussion) to come up with a synthesis across experiments. What I think they should avoid is writing parts of the results as if this synthesis has already been shown to be correct. This overall issue is illustrated below, particularly in my response to points 3a and 3b.

We thank the reviewer for bringing this stylistic concern to our attention. We have further modified the manuscript to minimize this. To facilitate a quick re-review, below we extracted the main introductory and conclusion sentences throughout the results, and indicated our edits.

Conclusion unmodified:

“These pharmacological experiments in acute brain slices indicate that IEM-1460 promotes an indirect pathway timing advantage and selectively diminishes multi-action potential evoked SPN responses.”

Conclusion added:

“This result confirms that IEM-1460, which inhibits FSI firing (Figure 1—figure supplement 1), selectively reduces multi-action potential SPN responses to afferent stimulation (Figure 1—figure supplement 2) as suggested by calcium imaging experiments (Figure 1F).”

Unmodified. In our view, this is an important point to discuss at this point in the paper because the results suggest that the sign of the hypothesis is inverted. In this case, we believe that some reference to the overall working hypothesis is necessary to communicate rationale behind subsequent experiments.

“Altogether, this series of experiments identifies a pharmacological agent that potently inhibits FSI activity and modulates all of the habit-predictive SPN firing properties. These results were surprising for two reasons. First, rather than a blockade of FSI activity causing disinhibition of SPNs as we had hypothesized, we found that when FSI activity was reduced, SPN activity was also reduced. This result suggests that FSI activity is capable of promoting, rather than inhibiting, SPN activity at least in the acute brain slice preparation. Secondly, although IEM-1460 strikingly affected the same features of DLS output that predict the expression of habitual behavior (calcium transient amplitude in both pathways and relative pathway timing) (Figure 1B), the directionality of these effects was opposite in all measures. Therefore, these results revise the overall hypothesis to involve a gain, rather than loss, of FSI activity as a candidate mechanism for habitual behavior.”

Unmodified:

“While IEM-1460 has been shown to have selective effects on the firing of FSIs in striatum, its effect of inhibiting AMPAR-mediated excitatory postsynaptic currents (EPSCs) in cholinergic interneurons (CINs) leaves open the possibility that CINs might contribute to our observed IEM-1460 effects.”

Modified. Original phrasing could have been misinterpreted to mean behavior, whereas we meant to refer only to the physiological “habit-predictive” findings at this point.

“Consistent with the IEM-1460 results in 2PLSM calcium imaging (Figure 1D-E) and cell-attached recording (Figure 1—figure supplement 2) experiments, this optogenetic result indicates that FSIs promote multi-action potential SPN responses to cortical excitation in the brain slice and that the effects of IEM-1460 on striatal output occur primarily through a reduction of striatal FSI activity.”

Original:

“This optogenetic result is consistent with the IEM-1460 results in 2PLSM calcium imaging (Figure 1D-E) and cell-attached recording (Figure 1—figure supplement 2) experiments. Taken together, these data indicate that FSIs promote multi-action potential SPN responses to cortical excitation in the brain slice and that the habit-opposing circuit output effects of IEM-1460 occur primarily through a reduction of striatal FSI activity.”

Modified. Although we prefer to remind reader of hypothesis and expectations, we have deleted this part to satisfy the reviewer’s stylistic concerns.

“While results thus far show that FSIs appear capable of specifically modulating habit-predictive properties of striatal output, we next examined whether FSI activity was different as a result of experience. We measured FSI synaptic and cellular electrophysiological properties in DLS brain slices prepared from habitual and goal-directed mice.”

Original:

“While results thus far show that FSIs appear capable of specifically modulating habit-predictive properties of striatal output, FSIs themselves would presumably need to undergo experience-dependent plasticity during the course of habit formation in order to alter their modulation of SPN firing and drive expression of habitual behavior. Because inhibiting striatal FSIs produced circuit effects opposite to those of habit expression, we hypothesized that FSIs would become strengthened with habit formation, for example through increases in synaptic strength and/or excitability. To test this hypothesis, FSI synaptic and cellular electrophysiological properties were measured in DLS brain slices prepared from habitual and goal-directed mice.”

Opening and closing, unmodified:

“Habitual behavior was associated with increased FSI firing in response to somatic current injection. However, it was afferent activation that initially revealed habit-predictive striatal output properties7. Therefore, in order for FSI plasticity to alter striatal output, it must be sufficient to differentially drive FSI firing in response to similar coincident synaptic excitation.”

“Together, these experiments show that FSIs undergo long-lasting, experience-dependent plasticity with habit formation and that this plasticity is sufficient to increase FSI firing.”

Modified:

“Since photo-inhibiting FSIs produces striatal output properties that directly oppose those seen in habit (Figure 1), we inhibited FSIs after habit training to determine the necessity of FSI activity for expression of habitual behavior.”

Original:

“Since photoinhibiting FSIs produces striatal output properties that directly oppose those seen in habit (Figure 1), the observed increase in FSI excitability would be predicted to drive the output-level striatal circuit signature for habit. Therefore, to test the necessity of FSI activity for the expression of habitual behavior, mice underwent habit-training protocols in the operant lever press task and then, prior to testing the degree of habitual responding, FSIs were inhibited chemogenetically.”

“These findings show that acute suppression of FSI activity in DLS causes habit-trained subjects to behave as though they were goal-directed.”

Unmodified:

“To understand how chemogenetic suppression of FSI firing affects striatal activity in vivo, single unit recordings were performed in a cohort of PV-Cre::Drd1a-tdTomato26 mice implanted in DLS with multi-electrode arrays and injected with the Cre-dependent hM4D inhibitory chemogenetic virus.”

Deleted:

“Notably, this trend was reminiscent of the linear relationship between basal calcium transient amplitude and the inhibitory effect of IEM-1460 observed ex vivo (Figure 1E).”

Final paragraph summarizing in vivo FSI experiments, somewhat interpretative, but an important point and immediately precedes Discussion section. We don’t believe this “constrains” thinking to fit our hypothesis, which is the main concern of the reviewer.

“These results demonstrate that FSIs modulate SPN activity in a more complicated manner than previously appreciated. While FSIs can have an overall strongly inhibitory effect in vivo on SPN firing as traditionally assumed, we also found evidence that they potentiate activity in a select population of SPNs that displays higher fractions of gamma-frequency spiking. This selective potentiation may be akin to a winner-take-all “focusing” mechanism that increases the signal-to-noise ratio in corticostriatal transmission. According to such a mechanism, the subset of recruited SPNs would be facilitated while the less-relevant, low-gamma SPNs would be suppressed.”

1a) For the order of vehicle and CNO applications, I assumed the worst because there was no mention of session ordering in the original manuscript, and the Figure 1 experiments employed a fixed control followed by treatment order (see my comments on this below). The fact that vehicle/CNO treatments were given on separate days and counterbalanced for order is important information that should be included in the Materials and methods, ideally with the table shown in the rebuttal, so that readers can see that in some cases, there were 10+ weeks between the two experiment days.

Table is included in the Review Response. A sentence describing this explicitly is included in the Materials and methods section of the manuscript.

“For in vivo electrophysiological recordings, CNO and vehicle were administered on different days and in counterbalanced order.”

1b) The new inclusion of locomotion data, which show similar distance traveled for both groups, is useful and increases confidence in the interpretation of the DREADD experiments.

1c) I apologize for any confusion that my original description may have caused. My point here is quite simple: what is the evidence that relative gamma ISI density, rather than firing rate, is the relevant variable here? In the rebuttal, the authors provide a figure showing a lack of significant correlation between firing rate and the fraction of gamma-band ISIs. But this does not exclude the possibility that if the new Figure 5G would test for a correlation between firing rate (fold change; the quantity plotted on the vertical axis) and baseline firing rate instead of gamma density, the reported correlation would persist. I agree with the points made by the authors that it's entirely possible that low overall firing rates are responsible for the lack of a significant correlation between firing rate and the fraction of gamma-band ISIs, and the Poisson analysis is interesting in its own right (my original suggestion would require a further step: asking if the excess gamma ISI probability predicts firing rate change following CNO). However, these are all indirect points. The most direct way of testing whether invoking gamma ISI density is necessary is to determine if using baseline firing rate yields the same overall result or not. If so, I don't see the need for a more complicated, gamma-ISI related explanation. In the rebuttal, the authors state, "fold change correlated with gamma density, not the rate itself" but I did not see evidence for the latter claim in the manuscript: this needs to be tested and stated explicitly.

We thank the reviewer for this point of clarification, and believe we now address it. As the reviewer suspects, in the analyses below, we do find that it is specifically the “excess gamma probability” that differentiates CNO response from vehicle and not a phenomenon related solely to firing rate. This additional analysis reinforces our original conclusions. We have accordingly expanded on this discussion in the manuscript and in Author response image 5 and 6 we include the new analyses related to this point.

Author response image 5

A similar analysis using firing rate alone indicates that it is the excess gamma specifically that predicts rate modulation by CNO. We find that, as you would imagine, increased rate increases modulation since this is a statistically improbable event in the setting of low basal firing rates; and accordingly, this effect does not show specificity for CNO over vehicle.

Author response image 6

We have modified the Results section to include these points as follows:

“Since neurons with higher firing rates would be expected to have shorter ISIs in general, we examined the possibility that the fraction of gamma ISIs in SPNs might simply relate to mean firing rate. However, we found that the proportion of gamma-frequency ISIs was unrelated to mean firing rate in baseline single unit SPN recordings before either CNO or vehicle administration (pre-CNO: p = 0.25, n = 23; pre-vehicle: p = 0.28, n = 20). Additionally, we found that SPNs fire significantly more gamma-frequency spikes than expected by Poisson processes matched to firing rate (pre-CNO: t(44) = 5.76, p = 7.67 x 10-7, n = 23 SPNs & rate-matched simulations; pre-vehicle: t(38) = 8.24, p = 5.59 x 10-10, n = 20 SPNs & rate-matched simulations). Whereas baseline firing rates non-specifically predict fold change in firing rate after both CNO and vehicle injection (CNO: r(22) = -0.61, p = 0.0022, n = 23; vehicle: r(19) = -0.45, p = 0.045, n = 20), the excess probability of gamma-frequency ISIs (observed – expected) specifically predicts rate modulation after CNO (r(22) = -0.52, p = 0.011, n = 23) but not vehicle (r(19) = 0.045, p = 0.85, n = 20). Therefore, gamma-frequency spiking represents a feature of interest in SPNs that predicts whether these output neurons will fire more or less as a consequence of reducing FSI activity.”

2a) Acknowledging the interpretation that the FSI manipulations performed are in effect a disruption of DLS activity, as the authors do now in the Discussion, is appropriate. However, I believe the authors' suggestion that their work is the first to show an effect of DLS manipulation on the expression (rather than acquisition) of habits is overstated. Packard and McGaugh (1996) performed lidocaine injections in the caudate nucleus following acquisition of a response strategy on a plus maze. Post-training baclofen/muscimol infusions into the DLS reduce habitual lever-pressing for alcohol (2012 Biol Psychiatry, 2016 Frontiers Behav Neuro), and has similar effects on lever pressing for cocaine (Zapata et al. J Neurosci 2010). Thus, for different operationalizations of habit and reinforcer, post-training manipulations of DLS have been shown to affect habitual (but not goal-directed) behavior.

We thank the reviewer for pointing these studies out. We very much wish to frame our findings accurately within the field and have modified our conclusions by deleting this point in the manuscript and including the earliest of the above-mentioned citations, copied below:

“…In the present study, by chemogenetically inhibiting PV+ interneurons in vivo, we found that FSI activity in DLS is required for the expression of a learned habit (Figure 4E); an automated, reward-insensitive behavior quite different from behaviors previously studied. Previous pharmacological inactivation studies have demonstrated a role for DLS in habit expression42, 43, indicating that general disruption of DLS activity also impairs established habitual behavior. Interestingly, in the present study, chemogenetic inhibition of FSI activity drove an overall increase in projection neuron activity (Figure 5F) which suggests that reducing FSI activity specifically may impair habit expression differently than a general inactivation of the circuitry.”

2b) Great, inclusion of the lever press and total rewards data strengthens the results.

3a) I appreciate the authors are combining multiple approaches and lines of evidence to converge on a proposed mechanism – this is a strength of the paper. But in this case it's hard to see beyond the most direct way to rule out nonspecific changes related to time, which is simply to counterbalance the order of the treatment and control groups, or to collect washout data. The authors second observation provided is the most relevant in that it does contain a treatment vs. control comparison without the potential time confound, but this is a separate experiment, but low n, where we are not told how its timescale maps on to the main experiment, while leaving open whether the effect shown here underlies the full Figure 1 results. The other observations are more indirect still. Fixing this would require re-doing the experiment. However, if the authors feel strongly that they can argue against this confound, this flaw may be outweighed by intriguing results in other parts of the paper that collectively contribute to an emerging new view of FSI-SPN interactions.

For technical reasons, a counter balance or washout design is not possible due to kinetics of drug action in the brain slice (approx. 20 minutes to take effect) and loss of Fura-2 signal with time. We have piloted a second approach to address this reasonable point, in which we perform group-wise comparisons to complement the original within-cell design. Slices are pre-incubated in vehicle or IEM drug prior to recording evoked calcium transients on the 2PLSM.

These new and confirmatory results are now described in the manuscript:

“Because the within-cell experimental design of measuring effects before and after IEM-1460 application did not exclude the possibility that changes in calcium signals occurred during the 20-minute wash-in period independently of IEM-1460, we performed a separate across-group study. Brain slices were incubated with either IEM-1460 or vehicle prior to and during imaging. Group mean calcium transient amplitudes were lower in IEM-1460 relative to vehicle in both dSPNs (vehicle: 0.043 ± 0.0011 ΔF/F0, N = 202 cells; IEM-1460: 0.037 ± 0.0021 ΔF/F0, N = 72 cells; t(272) = 2.62, p = 0.0093) and iSPNs (vehicle: 0.040 ± 0.0014 ΔF/F0, N = 143 cells; IEM-1460: 0.033 ± 0.0011 ΔF/F0, N = 56 cells; t(197) = 2.93, p = 0.0038) and IEM-1460-treated slices showed a preference for faster indirect pathway activation relative to vehicle-treated slices (t(197) = 3.83, p = 1.41 x 10-7, N = 143 & 56 independent dSPN/iSPN pairs). These results are generally consistent with findings from within-cell pre-post measurements.”

And corresponding data included here for reviewers’ reference:

Author response image 7

As to the other points, we clarify that the time scale of the vehicle versus IEM-1460 washin experiments in Figure 1—figure supplement 2 is approximately the same as the 2PLSM. Stable cell-attached recordings are obtained, drug is washed in for 20 minutes and then the post condition is recorded (approx. 30 min total).

We are unclear about the comment of other observations are more indirect still. We view Figure 2 as the most direct because we use a genetically defined targeting of FSI activity instead of the IEM drug, and because the laser pulsing on and off is temporally interleaved, controlling for that confound the best.

To emphasize the similarities of observations across experimental paradigms, we had re-analyzed the Figure 2 data in the same was as in Figure 1F and the supplement to Figure 1. This figure was included in the original reviewer response.

3b) As with 3a, I note that the authors' style is to make inferences from multiple experiments to converge on an overall explanation. This is admirable and important. However, this does not absolve them of responsibility in accurately describing what can and cannot be concluded from each experiment separately. I do not think it is appropriate to take their proposed overall mechanism and use it as a source of top-down constraints on the initial interpretation of specific, individual experiments. What I want to see is careful descriptions and conclusions for each individual experiment in the Results that appropriately take the limitations of each into account. Then, the authors are encouraged to integrate the results into an overall proposed mechanism, and if they wish, use the resulting synthesis to inform interpretation of specific pieces of data – but not where those experiments and data are first described.

As above, we reviewed the manuscript for this stylistic concern and believe we have achieved this goal.

For this particular point, while it may be true that about 1% of striatal neurons are FSIs, they tend to be highly active, rendering them more likely to be included in recording studies (and, depending on how cells are identified in imaging studies, in those too; see e.g. Harris et al. Nat Nat Neurosci 2016). I agree with the authors that results such as Figure 1E are unlikely to result from inadvertent inclusion of a few FSIs because we are shown the full distribution of data points. However for others such as Figure 1F which simply compare means, it seems possible that a few potential FSI outliers could shift the means significantly.

We understand the reviewer’s continued concern and include new data explicitly testing the degree to which FSI activity is captured by our activity imaging approach. We conclude that it is not possible for unexpected contamination of “SPN” signal by FSIs to underlie any of our conclusions.

Using a Pv-Cre x Ai9 reporter mouse we monitored calcium transients in genetically identified FSIs. We found that, using our Fura cell loading and data analysis conditions, evoked firing activity was detected in 50% of the total transgenically reported FSIs. We further found that while increasing stimulation intensity can increase FSI spike probability, we did not see a significant increase of the calcium transient amplitude. This observation is consistent with our unpublished observations that FSIs rarely burst fire under our electrical afferent stimulation and recording conditions. Both of these empiric observations indicate that FSI contamination is not a source of our present study’s main findings (i.e. we find modulation of amplitude but not spike probability). We also looked for statistical outliers and did not find support that it was driving mean effects.

Author response image 8

Lastly, we have previously documented the sensitivity and specificity of the Drd1a-tdTomato reporter line and find that Pv-positive cells are not present in td-Tomato expressing cells (Ade et al., Frontiers, 2011, Figure 3) which specifically indicates that FSI contamination could not explain any findings in the putative dSPN population.

To detect outliers, we employed the Robust regression and Outlier removal (ROUT) method available in the GraphPad Prism software. This outlier detection analysis fits a robust nonlinear regression to the data and detects outliers by residuals between the data and the robust fit based on a preset false discovery rate (we used Q = 1%).

We detected zero outliers in our pre-IEM iSPN data set, indicating that a small fraction of misclassified cells does not drive our finding:

iSPNs ACSFiSPNs IEM-1460
Method
ROUT (Q = 1.000%)
Number of points
Analyzed5252
Outliers00

In case a subset of cells showed an outlying response to IEM-1460 rather than an outlying baseline latency, we also ran this analysis for absolute pre-post latency differences in the iSPN data:

iSPN latency differences
Method
ROUT (Q = 1.000%)
Number of points
Analyzed52
Outliers0

3c) This additional explanation is helpful.

3d) Great, this new figure fully addresses my concern.

4) The comments related to behavior-physiology studies (Figure 3) were circumnavigated. The authors DO report an exclusion criterion, namely; "RRshort-trained mice with NDLPr <0 were considered to be goal-directed…Mice not meeting inclusion criterion were not used…". But, as part of their conclusions they routinely refer to the task as " habit promoting reinforcement protocol", and "habit-trained subjects…". Figure 3—figure supplement 1B indicates that 4 of 9 mice trained in this task do not engage in a habitual behavior, an insignificant 44% of mice.

We have re-read our original reviewer response and do not understand what aspect we “circumnavigated”. We understood the point of the reviewer to be whether the mice in the habit promoting RI protocol that had values indicating goal-oriented behavior had physiological properties that were more goal-like, more habit-like, or other. We clarified by stating that we never recorded from these mice and excluded them a priori. We could not evaluate the relationship of physiology across a behavioral continuum in Figure 3 because single cell recording variation is underpowered relative to our population-based calcium transient data presented in O’Hare et al., 2016. Secondly, we did not exclude any mice in Figure 4, the chemogenetic behavior experiment except for those with no detectable viral expression.

We fully agree and believe that we acknowledge in the manuscript that RI training does not produce total insensitivity to outcome devaluation in 100% of subjects. It is for this reason that we refer to it as a “habit promoting” protocol rather than “determining”, for example.

In Figure 4, the authors exposed mice to the behavioral protocol that already produces ~44% of goal directed behavior, to argue that DREADD manipulation produces robust goal directed behavior.

So, without a devaluation test before CNO day to assess degree of habit induced, it is hard to accept the claim that CNO promoted less habitual behavior. If the physiological phenotype of the intermediate animals (trained on task, but did not exhibit habitual behavior) were known, I would be more convinced that CNO inactivation of FSI would (at least) shape an incomplete phenotype described in Figure 3. I really thing Figure 4 has compelling results, and hope the authors address this concern.

This is a very interesting point and one that we can agree with in principle but not totality. The 9 mice cohort in Figure 3 is not appropriate to generalize about the effectiveness of RI reinforcement in driving habitual behavior. To better determine the effects of RI reinforcement, we have now analyzed the data from every mouse that we have trained on this reinforcement schedule (N = 59) and found that RI-trained mice display a mean NDLPr of 0.052 ± 0.13.

More importantly, behavioral results vary between cohorts for a number of reasons having to do with animal housing conditions and subtle differences in training conditions and trainer (see Sorge, et al. 2014 Nature Methods for a recent example).

Author response image 9

The variance of the data, from 59 RI-trained mice across 8 different experiments, is 1.06. However, when variance is measured within-group, variance is significantly lower (0.86), consistent with our observations that within training cohort are the best comparators. In addition, the aforementioned concern of the poor rate of instantiating habitual behavior refers to one of our weakest habit-inducing cohorts. For these reasons, we remain convinced that it is most appropriate to compare an experimental group to a contemporaneous control group rather than one produced separately in time, as we have done in this study.

Nevertheless, for the reviewer’s benefit, even when we compare the PV-hM4D mice to all 59 historical data points, we see a nearly significant difference whereas control PV-eYFP mice are more similar to unmodified RI-trained mice:

Author response image 10

But as it stands now, I am not fully convinced that the hm4D manipulation produces less habitual behavior. The author could come up with a clever, but robust statistical approach to assess/argue more convincing statistical result. For example if they serially excluded 40% of subjects in both groups, and performed the test, what fraction of the comparisons are significant? Or another formulation of shuffle tests. As I pointed out, a within-subject experiment would have been most ideal; that is each animal (in both PV-hM4D and OV-GFP groups) getting two devaluation sessions with and without CNO. If the authors cannot address this weakness via reanalysis of existing data, I think they will have to demonstrate a more convincing behavioral result with a new cohort of mice.

We would like to note that the study in Figure 4 is powered to the standard 1-β = 0.80 and α = 0.05 parameters for effect detection. For us to remove data and rerun the analysis would be to remove power from our study, which does not seem appropriate.

However, we appreciate the reviewer’s point that behavioral outcomes can be variable and that we may have “gotten lucky” (the p-value suggests a 1.6% chance of this happening), so we ran the clever shuffling analysis suggested by the reviewer based on the success rate of RI reinforcement driving habitual behavior in the control cohort (73% success). We serially removed random data points from both experimental and control groups, repeating this shuffling test Nhm4D x NeYFP, or 110 times. In Author response image 11 is the distribution of resulting p-values from two-tailed, unpaired t-tests:

Author response image 11

Our mean p-value from this shuffling experiment was also within the 95% confidence range (p = 0.048).To promote confidence that our analysis was carried out as suggested, MATLAB code is pasted below for reviewers’ reference:

function p = leaveNout(exp,ctrl)

groups = {'exp','ctrl'};

threshold = 0;

Ns = [length(exp) length(ctrl)];

fracRemove = sum(ctrl < threshold)/length(ctrl);

numRemove = floor(fracRemove.*Ns);

numShuffles = prod(Ns);

p = nan(numShuffles,1);

for i = 1:numShuffles

for g = 1:length(groups)

removeInds = randperm(Ns(g),numRemove(g));

eval(sprintf('curr_%s =% s;', groups{g}, groups{g}))

eval(sprintf('curr_%s(removeInds) = [];', groups{g}))

end

[~, p(i)] = ttest2(curr_exp, curr_ctrl);

end

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Striatal fast-spiking interneurons drive habitual behavior" for further consideration at eLife. Your revised article has been favorably evaluated by a Senior editor, a Reviewing editor (Michael Frank), and one reviewer.

We have sent the manuscript back to the one reviewer who had more reservations (the other reviewer was largely satisfied in the last round and we determined that you have addressed their remaining issues). The manuscript has been improved but there are now just a few remaining issues that need to be addressed before acceptance, as outlined by the reviewer below:

This resubmission contains several additional control experiments and analyses, which collectively address three major issues identified as remaining after the last round of reviews, with mixed success. One case is convincing, the second exacerbates my concerns about the authors' interpretation, and the third appears reasonable but is not incorporated into the manuscript.

– My request to control for the effects of time in the IEM-4160 slice experiments (Figure 1) is fully satisfied by the addition of new, across-group experiments for slices pre-incubated with IEM and vehicle.

– The authors claim that gamma-band ISIs are predictive of firing rate changes in SPN firing following CNO. However, pre-CNO firing rate is a numerically better predictor of such changes than excess gamma-band ISIs (r = -0.61 vs -0.45), casting doubt on the relevance of this feature given that a simpler and more predictive measure is available. The authors argue that firing rate "non-specifically" predicts such changes (i.e. for both vehicle and CNO) and that therefore excess gamma-band ISIs (which only predict post-CNO firing rates, but not post-vehicle rates) is an interesting quantity; however, these results could also be interpreted to mean that gamma-band ISIs is simply a worse predictor overall. I could imagine additional analyses, such as a multiple regression, that could help clarify this issue; however, regardless of the specific outcome, any remaining variance explained will likely be low and of unclear relevance. In addition, the authors should be aware that a comparison between two t-tests (one significant, one non-significant) does not constitute sufficient evidence for an interaction (see Nieuwenhuis et al. Nat Neurosci 2011) and report the correct test instead.

Per Nieuwenhuis et al. “When making a comparison between two correlations, researchers should directly contrast the two correlations using an appropriate statistical method”. We would like to first note that this article specifically discusses situations when p-values are quite similar, but fall on each side of 0.05 (such as 0.05 and 0.06) and the sign of the observed effects are similar. This is not our particular situation. Our p values for predicting CNO versus vehicle modulation are 0.011 and 0.85, respectively.

In any case, we now include this additional analysis. We performed a Fisher r-to-z transformation to assess the significance of the difference between two correlation coefficients. The results are detailed below, but in summary, these analyses show that our main conclusions are supported and remove this remaining concern.

Using the Fisher r-to-z transformation we demonstrate three points related to the reviewer’s concerns.

1) Firing rate is not a “better predictor” than excess gamma band ISIs for the effects of CNO on rate modulation. The two r-values (r = -0.61 vs. -0.52) are not significantly different, p = 0.34, z = -0.42.

As a reminder of our values used to perform these calculations:

Correlation of CNO modulation with baseline firing rate: r = -0.61, p = 0.0022, n = 22.

Correlation of CNO modulation with baseline excess gamma: r = -0.52, p = 0.011, n = 22.

(The reviewer mistakenly cites -0.61 v -0.45, which are the correlation values for baseline firing rate vs rate modulation by CNO and VEH groups. These two correlations are also not significantly different from one another: p = 0.25, z = -0.68.)

2) By contrast, the correlation coefficients for baseline excess gamma predicting rate modulation DO significantly differ between VEH and CNO,

p = 0.030, z = -1.88.

Based on our findings of:

Correlation for CNO: r = -0.520, p = 0.011, n = 23

Correlation for VEH: r = 0.045, p = 0.850, n = 20

We now report these measures in revised manuscript:

“Since neurons with higher firing rates would be expected to have shorter ISIs in general, we examined the possibility that the fraction of gamma ISIs in SPNs might simply relate to mean firing rate. […] Therefore, gamma-frequency spiking represents a feature of interest in SPNs that predicts whether these output neurons will fire more or less as a consequence of reducing FSI activity.“

In summary, we adopted an appropriate statistical test for comparing correlational analyses and found that baseline excess gamma activity indeed specifically predicts how the firing rates of individual SPNs will be modulated by chemogenetic inhibition of FSI activity, in comparison to Vehicle. Secondly, we found that SPN modulation by CNO was not predicted better by baseline firing rates than baseline gamma density. And thirdly, we found that baseline firing rates also do not predict SPN modulation by CNO specifically, but rather baseline firing rate correlates with rate modulation equally well for both CNO and VEH.

(N.B. If interest in this last point remains, we would like to clarify that we believe these are the relationships you would expect to find by chance due to the law of large numbers and a tendency to regress to the mean. We have directly tested this using random samplings of our data shuffled and find that our observed values are within those predicted by the simulation.)

– In order to address the concern that FSIs included in the iSPN dataset (which is defined by the absence of D1R label) may contribute to the effects in Figure 1, the authors provide new data suggesting this is unlikely. At a minimum the authors need to discuss this possible confound in the manuscript, and ideally include the data shown in a supplementary Figure.

As requested, we now include discussion on interneuron contamination of datasets in the manuscript and the 2PLSM FSI data as a new supplementary figure.

“Contamination of dSPN and iSPN datasets by interneurons was minimized by selection criteria and monitoring datasets for outliers (See Materials and methods for further details).”

“Classifying regions of interest: Regions of interest (ROIs) showing red and green were classified as dSPNs whereas green-only ROIs were classified as iSPNs. The small percentage of green-only cells which would have been striatal interneurons was partially mitigated by ignoring abnormally large ROIs which were likely to be cholinergic interneurons. FSIs comprise approximately 1% of all striatal neurons and are not present in the Drd1a-TdTomato labelled population. However, FSIs might be included in the “putative iSPN” population. Empiric experiments using Pv-Cre x Ai9 reporter mice demonstrate that approximately 50% of FSIs are labelled with Fura2-AM and pass data inclusion criteria in our experimental setting (see Materials and methods and Figure 1—figure supplement 1). Therefore, we expect at most that approximately 1% of iSPN and none of dSPN data may represent FSIs. Because CIN and FSI firing properties are very different from SPNs, we reviewed our data sets for the presence of outliers using Robust regression and Outlier (ROUT) analysis (GraphPad Prism software) and a false discovery rate of Q = 1%. No outliers were detected in iSPN baseline amplitude, post-IEM-1460 amplitude, or latency modulation datasets of Figure 1.”

https://doi.org/10.7554/eLife.26231.019

Article and author information

Author details

  1. Justin K O'Hare

    1. Department of Neurobiology, Duke University Medical Center, Durham, United States
    2. Department of Neurology, Duke University Medical Center, Durham, United States
    Contribution
    Conceptualization, Data curation, Software, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon 0000-0002-7363-6064
  2. Haofang Li

    Department of Psychology and Neuroscience, Duke University, Durham, United States
    Contribution
    Resources, Investigation, Methodology
    Competing interests
    No competing interests declared
  3. Namsoo Kim

    Department of Psychology and Neuroscience, Duke University, Durham, United States
    Contribution
    Data curation, Formal analysis, Investigation, Methodology
    Competing interests
    No competing interests declared
  4. Erin Gaidis

    Department of Psychology and Neuroscience, Duke University, Durham, United States
    Contribution
    Data curation, Investigation, Methodology
    Competing interests
    No competing interests declared
  5. Kristen Ade

    1. Department of Neurobiology, Duke University Medical Center, Durham, United States
    2. Department of Neurology, Duke University Medical Center, Durham, United States
    Contribution
    Software, Validation, Methodology
    Competing interests
    No competing interests declared
  6. Jeff Beck

    Department of Neurobiology, Duke University Medical Center, Durham, United States
    Contribution
    Formal analysis, Methodology
    Competing interests
    No competing interests declared
  7. Henry Yin

    Department of Psychology and Neuroscience, Duke University, Durham, United States
    Contribution
    Conceptualization, Resources, Supervision, Funding acquisition, Methodology, Writing—original draft, Writing—review and editing
    Competing interests
    No competing interests declared
  8. Nicole Calakos

    1. Department of Neurobiology, Duke University Medical Center, Durham, United States
    2. Department of Neurology, Duke University Medical Center, Durham, United States
    Contribution
    Conceptualization, Resources, Supervision, Funding acquisition, Methodology, Writing—original draft, Project administration, Writing—review and editing
    For correspondence
    nicole.calakos@duke.edu
    Competing interests
    No competing interests declared
    ORCID icon 0000-0002-9918-3294

Funding

National Institute of Neurological Disorders and Stroke (NS064577)

  • Nicole Calakos

National Institutes of Health (ARRA supplement to NS064577)

  • Nicole Calakos

National Institute on Alcohol Abuse and Alcoholism (AA021075)

  • Henry Yin

National Institute on Drug Abuse (DA040701)

  • Henry Yin

McKnight Foundation

  • Henry Yin
  • Nicole Calakos

National Institute of Neurological Disorders and Stroke (NS051156)

  • Kristen Ade

National Institute of General Medical Sciences (GM008441)

  • Justin K O'Hare

Brain and Behavior Research Foundation

  • Kristen Ade

Tourette Association of America

  • Kristen Ade

Ruth K. Broad Biomedical Research Foundation

  • Justin K O'Hare

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

The authors thank L Glickfeld, M Rossi, and MB Branch for their productive discussions and comments on the manuscript. The authors thank S Sternson for providing the CAG-FLEX-rev-hM4D:2a:GFP plasmid and the UNC Viral Vector Core for production of viruses. We gratefully acknowledge the following sources of funding: NS064577 and ARRA supplement (NC), AA021075 (HY), DA040701 (HY), McKnight Endowment Fund for Neuroscience (NC, HY), GM008441-23 (JO), NS051156 (KA), The Brain and Behavior Foundation (KA), The Tourette Association of America (KA) and the Ruth K. Broad Foundation (JO).

Ethics

Animal experimentation: All experiments were carried out under approved animal protocols (A112-17-04 & A263-16-12) in accordance with Duke University Institutional Animal Care and Use Committee standards.

Reviewing Editor

  1. Michael J Frank, Reviewing Editor, Brown University, United States

Publication history

  1. Received: February 21, 2017
  2. Accepted: August 8, 2017
  3. Version of Record published: September 5, 2017 (version 1)

Copyright

© 2017, O'Hare et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 1,985
    Page views
  • 270
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: PubMed Central, Scopus, Crossref.

Comments

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)