Globus pallidus dynamics reveal covert strategies for behavioral inhibition
Abstract
Flexible behavior requires restraint of actions that are no longer appropriate. This behavioral inhibition critically relies on frontal cortex - basal ganglia circuits. Within the basal ganglia, the globus pallidus pars externa (GPe) has been hypothesized to mediate selective proactive inhibition: being prepared to stop a specific action, if needed. Here we investigate population dynamics of rat GPe neurons during preparation-to-stop, stopping, and going. Rats selectively engaged proactive inhibition towards specific actions, as shown by slowed reaction times (RTs). Under proactive inhibition, GPe population activity occupied state-space locations farther from the trajectory followed during normal movement initiation. Furthermore, the state-space locations were predictive of distinct types of errors: failures-to-stop, failures-to-go, and incorrect choices. Slowed RTs on correct proactive trials reflected starting bias towards the alternative action, which was overcome before progressing towards action initiation. Our results demonstrate that rats can exert cognitive control via strategic adjustments to their GPe network state.
Introduction
Our capacity for self-restraint is critical for adaptive behavior. Dysfunctions in behavioral inhibition are involved in many human disorders, including drug addiction (Ersche et al., 2012). A standard test of behavioral inhibition is the stop-signal task (Logan et al., 1984; Verbruggen et al., 2019), in which subjects attempt to respond rapidly to a Go cue, but withhold responding if the Go cue is quickly followed by a Stop cue. The stop-signal task has been invaluable for revealing specific cortical-basal ganglia mechanisms involved in both movement initiation (‘Going’; e.g. Hanes and Schall, 1996) and inhibition (‘Stopping’; e.g. Aron and Poldrack, 2006; Eagle et al., 2008). ‘Reactive’ inhibition – making quick use of a Stop cue – appears to involve at least two distinct mechanisms (Schmidt and Berke, 2017): a rapid Pause process mediated via the subthalamic nucleus (STN; Aron and Poldrack, 2006; Schmidt et al., 2013) followed by a Cancel process achieved through pallidostriatal inhibition (Mallet et al., 2016).
Behavioral inhibition can also be ‘proactive’: restraint of actions, in advance of any Stop cue. Proactive inhibition may be particularly relevant to human life (Aron, 2011; Jahanshahi et al., 2015). Whereas reactive inhibition typically involves a global, transient arrest of actions and thoughts (Wessel and Aron, 2017), proactive inhibition can be selectively directed to a particular action (Cai et al., 2011). A key behavioral signature of proactive inhibition is slowing of reaction times (RTs) for that action, when the anticipated Stop cue does not actually occur (e.g. Verbruggen and Logan, 2009; Chikazoe et al., 2009; Zandbelt et al., 2013). This overt behavioral signature presumably relies on covert shifts in information processing, yet the nature of these shifts is unclear. In some studies, fitting of models to behavioral data has suggested that slowed RTs reflect raising of a decision ‘threshold’ (Verbruggen and Logan, 2009; Jahfari et al., 2012), but other studies have found evidence for a slower rate of progression toward threshold instead (Dunovan et al., 2015).
The neural circuit mechanisms by which proactive control is achieved are also not well understood. It has been proposed that proactive inhibition critically depends on the basal ganglia ‘indirect’ pathway via GPe (Aron, 2011; Jahanshahi et al., 2015; Dunovan et al., 2015). Yet clear support for this hypothesis is sparse (Majid et al., 2013). There have been few electrophysiological studies of proactive inhibition at the level of individual neurons (Chen et al., 2010; Pouget et al., 2011; Hardung et al., 2017; Yoshida et al., 2018), and to our knowledge none in GPe. We therefore targeted GPe (often called simply GP in rodents) for investigating neural mechanisms of proactive control.
We also wished to integrate a dynamical systems approach into the study of behavioral inhibition, and the basal ganglia. Analysis of the collective dynamics of motor cortex neurons has provided insights into various aspects of movement control, including how brain networks may prepare actions without prematurely triggering them (Kaufman et al., 2014), and the origins of RT variability (Afshar et al., 2011). We demonstrate below that the analysis of GPe population activity can reveal distinct covert strategies underlying overt manifestations of proactive control.
Results
Action initiation is slower when a stop cue is expected
We trained rats in a modified version of our stop-signal task (Figure 1A; Leventhal et al., 2012; Schmidt et al., 2013; Mallet et al., 2016). Freely-moving rats poked their noses into a hole and maintained that position for a variable delay (500–1250 ms) before presentation of one of two Go cues (1 kHz or 4 kHz tone), instructing leftward or rightward movements respectively into an adjacent hole. If initiated rapidly (RT limit <800 ms), correct movements triggered delivery of a sugar pellet reward from a separate food hopper. On some trials the Go cue was quickly followed by a Stop cue (white noise burst), indicating that the rat instead needed to maintain its nose in the starting hole (for a total of 800 ms after Go cue onset) to trigger reward delivery. The delay between Go and Stop cue onsets (100–250 ms) ensured that stopping was sometimes successful and sometimes not. As expected, Failed Stop (error) trials had similar RTs to the faster part of the Go trial RT distribution (Figure 1B). This is consistent with the basic ‘race’ conceptual model of reactive inhibition (Logan et al., 1984): failures-to-stop typically occur when an underlying Go process evolves more quickly than average (Schmidt et al., 2013), and thus wins the race against a separate Stop process.
To probe selective proactive inhibition we used a ‘Maybe-Stop versus No-Stop’ approach (Aron and Verbruggen, 2008). The three possible starting holes were associated with different Stop cue probabilities (Figure 1C): no possibility of Stop cue; 50% probability that a left Go cue (only) will be followed by the Stop cue; or 50% probability that a right Go cue (only) will be followed by the Stop cue. Our index of proactive inhibition was a preferential increase in RT for the Maybe-Stop direction, compared to the No-Stop conditions. Among rats that began learning this task variant, approximately half acquired clear proactive inhibition within 3 months of training (see Materials and methods), and were thus considered eligible for electrode implantation. Here we report behavioral and neural results for six rats for which we were able to obtain high-quality GP recordings as rats engaged proactive control.
We selected for further analysis those behavioral sessions (n = 63) with a significant proactive inhibition effect (i.e. longer RT when a Stop cue might occur; one-tail Wilcoxon rank sum test, p<0.05) and distinct GP single units (n = 376 neurons included). Prior work has shown particular basal ganglia involvement in the control of contraversive orienting-type movements (i.e. directed towards the opposite side; Carli et al., 1985; Isoda and Hikosaka, 2008; Schmidt et al., 2013; Leventhal et al., 2014). We therefore focused on proactive control of movements contraversive (‘contra’) to the recorded cell locations; for example we included a left GPe cell only if the rat demonstrated proactive control for rightward movements during that recording session. For included sessions, median RT for correct contra movements was 251 ms when the Stop cue could not occur (No-Stop), and 385 ms when the Stop cue could occur (Maybe-Stop) but did not. Results from all sessions, and from individual animals, are shown in Figure 1—figure supplement 1.
RT slowing due to proactive inhibition was highly selective to the Maybe-Stop direction (Figure 1D; Figure 1—figure supplement 1; for Maybe-Stop-Contra trials without a Stop cue, median ipsiversive (‘ipsi’) RT was unslowed at 264 ms). The Maybe-Stop condition was also associated with an increase in errors (Figure 1D), in particular not responding quickly enough to the Go cue that might be followed by Stop (RT limit error; RT >800 ms) and making the wrong choice (incorrect action selection). These error types are examined further below.
GP firing rate changes related to movement onset and proactive inhibition
We recorded individual neurons (n = 376) from a wide range of GP locations (Figure 2—figure supplement 1A). As expected from prior studies (DeLong, 1971; Brotchie et al., 1991; Gardiner and Kitai, 1992; Turner and Anderson, 1997; Arkadir et al., 2004; Gage et al., 2010; Shin and Sommer, 2010; Schmidt et al., 2013; Yoshida and Tanaka, 2016; Mallet et al., 2016) GP neurons were tonically-active (mean session-wide firing rate, 28 Hz) with diverse, complex changes in firing patterns during task performance (Figure 2A). The majority of GP cells showed strongest firing rate changes (increases or decreases) when activity was aligned relative to movement onset, rather than to the Go or Stop cues (Figure 2C,D; see also Figure 2—figure supplement 1B for ipsi movement trials). Individual neurons showed greater changes for either contra or ipsi movements (Figure 2A,B), but these were about equally represented in the overall population (Figure 2B,E), and the average GP activity was similar for the two movement directions (at least until the movement was already underway; Figure 2B).
We next examined how the activity of individual GP neurons is affected by proactive inhibition. As rats waited for the (unpredictably-timed) Go cue, average firing was similar between Maybe-Stop and No-Stop conditions (Figure 2F), regardless of whether we examined cells that predominantly increase or decrease activity during movements (Figure 2—figure supplement 1C). We hypothesized that this average activity obscures a sizable GP subpopulation that consistently and persistently ‘encodes’ proactive control as rats wait. To search for this putative subpopulation we used a screening approach (similar to our prior work on reactive stopping; Schmidt et al., 2013; Mallet et al., 2016), comparing the Maybe-Stop-contra and No-Stop conditions. We did find that the fraction of GP cells that fired differently between these conditions was slightly greater than expected by chance (Figure 2F), consistent with GP involvement in proactive control. However, contrary to our hypothesis, we were not able to identify a clear subgroup of individual neurons that strongly and persistently distinguished between conditions (Figure 2G). Rather, proactive control was associated with altered activity in different subsets of GP neurons at various brief moments before the Go cue (Figure 2—figure supplement 1D).
Population trajectories during movement selection and initiation
We next hypothesized that these GP firing rate differences, though subtle and diverse at the single-cell level, are coordinated to produce clear, interpretable changes in population dynamics. To observe these dynamics we began by reducing the dimensionality of population activity (Cunningham and Yu, 2014), using principal component analysis (PCA). For each neuron we included normalized, averaged firing rates for a 500 ms epoch around movement onset (separately for contra and ipsi movements; Figure 3A). We used the first 10 principal components (PCs; Figure 3—figure supplement 1A) to define a 10-dimensional state-space, with GP population activity represented as a single point in this space. For visualization we display the first 3 PCs (which together account for 71% of total population variance; Figure 3B), although statistical analyses used all 10 PCs.
Within state space, population activity was very similar for contra and ipsi movements at the Go cue (Figure 3C), and initially evolved in a common direction before progressively separating into distinct trajectories (Video 1). We used the common direction to define an ‘Initiation Axis’, scaled between 0 (mean location at Go cue) and 1 (mean location at movement onset, Center Out). This allows us to quantify progression towards (or away from) movement onset. We used the difference between trajectories to define a ‘Selection Axis’, scaled between −1 (mean of the ipsi trajectory) and +1 (mean of the contra trajectory). This allows us to quantify bias toward one movement direction or the other. Along both Initiation and Selection axes, change was not dominated by a small proportion of GP neurons. Instead, there were smaller contributions from many individual cells located throughout GP (Figure 3—figure supplement 1AC-E).
Failed stops reflect earlier evolution of GP activity
We then considered how GP population activity is evolving when Stop cues occur. As noted above, standard race models of reactive stopping (Logan et al., 1984), together with prior data (Schmidt et al., 2013), suggest that failures-to-Stop occur when an underlying Go process evolves more quickly than average, and thus the Stop cue arrives too late. GP population activity was consistent with these ideas (Figure 3D–F). On successful-Stop trials GP activity showed little or no movement before the Stop cue. By contrast, on failed-Stop trials GP activity was in a significantly different state by the time of the Stop cue, having already evolved a substantial distance along the Initiation Axis (Figure 3D; includes both contra- and ipsi-cued trials). Thus, our observations of neural dynamics support hypothesized internal dynamics that determine whether we can react to new information, or are already committed to a course of action.
When stop cues may occur, GP activity starts farther from movement initiation
Conceptually, the slowing of RT with proactive inhibition could reflect any of several distinct underlying changes (Figure 4A), that would manifest in GP dynamics in different ways. If slowing involves mechanisms ‘downstream’ of GP, we might observe no change in the GP population trajectory when aligned on the Go cue (hypothesis 1). Alternatively, the GP might be in a different state at the time the Go cue arrives. In particular, GP activity might start farther away from ‘threshold’ (in dynamical terms, farther from a subspace associated with movement initiation), and thus take longer to get there (hypothesis 2). Finally, proactive inhibition might cause GP activity to evolve differently after Go cue onset. Various, non-mutually-exclusive possibilities include a delayed start (hypothesis 3), slower progress along the same trajectory (hypothesis 4), and/or a threshold that is shifted further away from the starting point (hypothesis 5). Of note, only hypothesis two predicts a change in the trajectory start location at the time of the Go cue (Figure 4A).
We compared GP population activity between Maybe-Stop and No-Stop conditions, immediately before the Go cue (−100 ms - 0 ms; including all trial subtypes). When proactive inhibition was engaged, GP activity occupied a significantly shifted location within state-space (Figure 4B,C). When examined along the Initiation axis (Figure 4C), the direction of this shift was consistent with a longer trajectory required for movements to begin (hypothesis 2). In other words, the brain can restrain actions by placing key circuits into a state from which actions are slower to initiate.
Distinct state-space positions predict distinct types of errors
Proactive inhibition of contra movements also produced a significant shift along the Selection axis before the Go cue, in the direction associated with ipsi movements (Figure 4C). This suggests a preparatory bias against contra movements, when the contra-instructing Go cue may be followed by a Stop cue. To examine how starting position affects behavioral outcome, we examined how state-space location at the Go cue varies with distinct types of errors (Figure 4D). Failures to respond quickly enough to the Go cue (RT limit errors) were associated with starting farther away on the Initiation Axis (Figure 4E). By contrast, incorrect choices (ipsi movements despite contra cue) were associated with starting closer to movement initiation, together with a more-ipsiversive position on the Selection axis at Go cue (Figure 4E; Video 2). Thus, even while the animals are holding still, waiting for the Go cue, GP networks show distinctly-biased internal states that predict distinct subsequent behavioral outcomes.
Overcoming a selection bias delays movement initiation
The starting ipsiversive bias on the Selection axis when contra actions might have to be cancelled can be overcome, as even on contra Maybe-Stop trials the rats usually made the correct choice. To examine how this occurs we compared neural trajectories for correct, contra Maybe-Stop and No-Stop trials (Figure 5A; only correct trials without Stop cues are included). Just before the Go cue on Maybe-Stop trials, rats showed no difference on the Initiation Axis but were significantly shifted on the Selection axis, in the ipsiversive direction (Figure 5A,B). After the Go Cue, movement on the Initiation axis was delayed compared to No-Stop trials, but movement on the Selection Axis occurred earlier (Figure 5C; Video 3). Thus, on correctly-performed Maybe-Stop trials the GP network engaged a dynamical sequence that was not observed on No-Stop trials: they first overcame a proactive bias towards the alternative action, before proceeding to initiate the action that had been cued.
Together our results indicate that, when faced with the challenging Maybe-Stop condition, rats adopt multiple, distinct, covert strategies. They can position neural activity farther from movement onset (on the Initiation Axis), but this produces limited hold violations – essentially making this a bet that the Stop cue will in fact occur. Alternatively, they can bias neural activity in the ipsi direction (on the Selection Axis). This delays contra choices, but also increases the rate of incorrect ipsi choices.
Slower RTs can arise through multiple dynamic mechanisms
We considered the possibility that this apparent ‘strategy’ for proactive inhibition simply reflects the slower RT. In other words, is the distinct trajectory seen for correct Maybe-Stop trials also seen for slower No-Stop trials? Our data indicate that this is not the case. Comparing Maybe-Stop trials with No-Stop trials with the same RT (RT-matching) again showed different positions on the Selection Axis at Go cue (Figure 5—figure supplement 1). This difference was not seen when comparing slower and faster RTs within the No-Stop condition (Figure 5D,E). Rather, spontaneously-slower RTs appeared to arise through slower evolution along both Initiation and Selection Axes simultaneously (Figure 5F). Furthermore, on Maybe-Stop trials movement along the Selection axis overshot the level reached on No-Stop trials, as if overcompensating for the initial bias on this axis (Figure 5A, Figure 5—figure supplement 2). This overshoot was not seen for spontaneously-slower No-Stop trials (Figure 5—figure supplement 2). We conclude that variation in RT reflects multiple dynamic processes within basal ganglia circuits, with slowing due to proactive inhibition involving distinct internal control mechanisms to spontaneous RT variation.
Although reducing the dimensionality of data can be very useful for visualizing trajectories through state-space, we wished to ensure that our conclusions are not distorted by this procedure. We therefore repeated key analyses within the full 376-dimensional state space. Defining Initiation and Selection Axes in the same way as before, but without the PCA step, produced essentially identical trajectory differences between conditions (Figure 6).
Discussion
Stop-signal tasks are widely-used to test cognitive control (Lipszyc and Schachar, 2010), with proactive inhibition considered especially reliant on top-down, effortful, resource-demanding processes (Jahanshahi et al., 2015). Yet there have been extended debates about which psychological and neural mechanisms support proactive control (Verbruggen and Logan, 2009; Chatham et al., 2012; Aron et al., 2014; Leunissen et al., 2016). We have demonstrated here that a key behavioral signature of proactive control – selective slowing of RTs when a Stop signal is expected - can arise through multiple covert strategies. These are visible as changes to the dynamic state of GPe by the time of Go cue presentation, and include a bias towards an alternative action, and/or starting further from the ‘point-of-no-return’ in action initiation.
Which internal strategies are employed for proactive inhibition is likely influenced by the specific experimental conditions (Mayse et al., 2014; Yoshida et al., 2018). For example, we used a brief limited hold period (800 ms) to encourage subjects to respond rapidly to the Go cue rather than waiting to see if the Stop cue is presented. This time pressure may have led rats to sometimes make guesses as to which cues will be presented, and position their neural state accordingly. We also used a task design with asymmetric (ipsi/contra) stop probabilities, to probe the selectivity of proactive inhibition (Aron and Verbruggen, 2008). Motivational aspects are known to be important in proactive inhibition (Meyer and Bucci, 2016): the ipsi bias we observed on the Selection axis on Maybe-Stop (contra) trials may partly reflect asymmetric reward expectancy (Kawagoe et al., 1998), simply because ipsi movements are more consistently rewarded from that state. Unlike human subjects, we cannot verbally instruct rats to perform the task in a certain way (although human cognitive strategies do not always follow experimenter intentions either). It might seem simpler, and less error-prone, for the rats to just select from the slower portion of their regular RT distribution. We suggest that they are unable to consistently do so, given the high spontaneous variability in RTs. The degree to which specific neural strategies are employed may also vary between rats; we found some preliminary evidence for this (Figure 4—figure supplement 1), though investigating this further would require more animals and more recorded cells in each animal.
The term ‘proactive’ or ‘cognitive’ control has been used to refer both to stop-signal tasks like this one, in which subjects are cued about the upcoming stop probability (e.g. Cai et al., 2011; Jahfari et al., 2012; Zandbelt et al., 2013), and also to uncued behavioral adjustments that subjects make after each trial (e.g. longer RTs following trials in which Stop cues occurred; Chen et al., 2010; Pouget et al., 2011; Mayse et al., 2014). Although not the focus of this study, our rats did slow down slightly on average after Stop trials or errors (Figure 4—figure supplement 2A). This slowing was associated with a modest shift on the Initiation Axis in the same, movement-opposed direction as in our main results (Figure 4—figure supplement 2B), but this effect did not reach significance. Thus both behavioral and neural data suggest that the cued component of proactive inhibition was substantially greater than post-trial adjustments under our particular task conditions.
Our ability to reveal distinct strategies for proactive inhibition relies on a dynamical systems approach with single-cell resolution. This method may be especially important for deciphering structures like GP, where projection neurons show continuous, diverse activity patterns. As intermingled GP neurons increased and decreased firing at each moment, the resulting network state changes would likely be undetected using aggregate measures such as photometry or fMRI. Speculatively, we suggest that an enhanced ability to make subtle adjustments to dynamical state may be part of the reason why GP projection neurons show high spontaneous activity, in contrast to (for example) the near-silence of most striatal projection neurons, most of the time.
Prior examinations of motor/premotor cortical dynamics during reaching movements in non-human primates have demonstrated distinct neural dimensions for movement preparation and execution (‘What’ to do) and movement triggering (‘When’ to do it) (Elsayed et al., 2016; Kaufman et al., 2016). Our Selection and Initiation axes are analogous, although our task lacks an explicit preparation epoch and has only two action choices (left vs. right). One notable difference in the non-human primate studies is that movement preparation occurred in distinct, orthogonal dimensions to movement execution, whereas we saw preparatory ‘bias’ along the same Selection axis that differentiated ipsi and contra trajectories during movement itself. Nonetheless, our observation that on correct Maybe-Stop trials, GP state evolved first along the Selection axis is consistent with evidence that movement preparation and movement initiation can be independent processes (Haith et al., 2016; Thura and Cisek, 2017), and that these can be differentially modulated by the basal ganglia and dopamine (Leventhal et al., 2014; Manohar et al., 2015). It also appears consistent with recent observations that, following an unexpected late change in target location, preparation dimensions are rapidly re-engaged (Ames et al., 2019).
The distinction between What and When dimensions is not readily compatible with sequential-sampling mathematical models of decision-making (Smith and Ratcliff, 2004; Brown and Heathcote, 2008; Noorani and Carpenter, 2016), which typically assume that RTs (When) directly reflect sufficient accumulation of evidence for a particular choice (What). Furthermore, when sensory cues are unambiguous the selection process appears to be much faster than standard RTs (Stanford et al., 2010; Haith et al., 2016). Why RTs are typically so much slower and more variable than required for sensory processing or action selection is not fully clear, but this extra time provides opportunity for impulsive or inappropriate responses to be overruled, to increase behavioral flexibility.
The GPe is well positioned to contribute to such behavioral control. GPe has bidirectional connections with the subthalamic nucleus, a key component of the ‘hyperdirect’ pathway from frontal cortex that slows decision-making under conditions of conflict (Cavanagh et al., 2011). GPe itself is the target of the ‘indirect’ (striatopallidal) pathway, believed to discourage action initiation (‘NoGo’; Yoshida and Tanaka, 2016; Kravitz et al., 2010), possibly due to pessimistic predictions of reward (Collins and Frank, 2014; Kim et al., 2017). In standard, firing rate-based models of basal ganglia function, GPe activity restrains actions by preventing pauses in the firing of basal ganglia output, that are in turn required to disinhibit movement-related activity in the brainstem and elsewhere (Chevalier and Deniau, 1990; Roseberry et al., 2016).
However, it is well-recognized that this model is too simple (Gurney et al., 2001; Klaus et al., 2019), and it does not account for the complex activity patterns within GPe that we and others have observed. For example, a straightforward application of the rate model might predict a systematic decrease in GPe firing rate with proactive inhibition, but we did not observe this (Figure 2), with the possible exception of trials with RT limit errors (Figure 2—figure supplement 1). Based on the current results, examining dimension-reduced population dynamics is a promising alternative approach for deciphering how subtle modulations in the firing of many basal ganglia neurons are coordinated to achieve behavioral functions.
At the same time, our study has several noteworthy limitations. Our reduction of complex dynamics to movement along Initiation and Selection axes is obviously a simplification. We did not record large populations of neurons simultaneously, which precludes effective analysis of neural dynamics on individual trials (Afshar et al., 2011). We did not classify GPe neurons by projection target (Mallet et al., 2016; Abecassis et al., 2020) largely because we did not consistently record sleep data to enable that classification (Mallet et al., 2016). We do not yet know the extent to which these population dynamics are shared with upstream (e.g. striatum) and downstream (e.g. substantia nigra pars reticulata) structures, which will be essential for elucidating how these dynamic changes actually influence behavior. Finally, we have not yet determined how the population dynamics reported here relate (if at all) to oscillatory dynamics reported in cortical-basal ganglia circuits during movement suppression (Swann et al., 2009; Cavanagh et al., 2011; Leventhal et al., 2012) and in pathological states such as Parkinson’s Disease (Hammond et al., 2007). These are all worthy subjects for future investigation.
Key resources
Rat (adult, male, Long-Evans, bred in-house).
Materials and methods
All animal experiments were approved by the University of California, San Francisco Committee for the Use and Care of Animals. Adult male Long-Evans rats were housed on a 12 hr/12 hr reverse light-dark cycle, with training and testing performed during the dark phase.
Behavior
Request a detailed protocolOperant chambers (Med Associates, Fairfax VT) had five nose-poke holes on one wall, a food dispenser on the opposite wall, and a speaker located above the food port. The basic rat stop signal task has been previously described (Leventhal et al., 2012.; Mallet et al., 2016; Schmidt et al., 2013). At the start of each trial, one of the three more-central ports was illuminated (‘Light On’) indicating that the rat should poke in that port (‘Center In’) and wait. After a variable delay (500–1250 ms), a higher (4 kHz) or lower (1 kHz) pitch tone was presented for 50 ms (‘Go Cue’), instructing a move to the adjacent port on the left or right side respectively. In Go trials (those without a Stop cue) if the rat left the initial center port (‘Center Out’) within 800 ms of Go cue onset, and then moved to the correct side port (‘Side In’) within 500 ms, a sugar pellet reward was delivered to the food dispenser with an audible click. As the rat left the center port, the center port light was turned off and both side port lights turned on. On Stop trials, the Go cue was followed by a Stop cue (white noise, 125 ms) with a short delay (the stop-signal delay, SSD). The SSD was randomly selected on each trial within a range (uniform distribution) of 100–200 ms (four rats) or 100–250 ms (two rats). Stop trials were rewarded if the rat maintained its nose continuously within the start hole for a total of 800 ms after Go cue onset. Stop trials in which the rat initiated movement before the Stop cue began were converted into Go trials (i.e. no Stop cue was presented). Failed-Stop trials with RT >500 ms were excluded from electrophysiological analyses, since these were presumed to reflect trials for which rats successfully responded to the Stop cue, but then failed to maintain holding until reward delivery (see Leventhal et al., 2012; Schmidt et al., 2013; Mayse et al., 2014). Inter-trial intervals were randomly selected between 5 and 7 s. For included sessions, the median number of Go trials was 266 (range, 167–361) and the median number of Stop trials was 57 (range, 27–95).
To vary proactive inhibition, we changed the Stop cue probabilities between starting holes (as shown in Figure 1). The spatial mapping of probabilities was constant for each rat across sessions, but varied between rats. Within each session, the same start hole (and thus proactive condition) was repeated for 10–15 trials at a time. After ~3 months of training, rats showing consistent reaction time differences between Maybe-Stop and No-Stop conditions were eligible for electrode implantation.
Electrophysiology
Request a detailed protocolWe report GP data from six rats (all animals in which we successfully recorded GP neurons during contraversive proactive inhibition). Each rat was implanted with 15 tetrodes (configured as independently-driveable bundles of 2–3 tetrodes, each within a polyimide tube with outer radius 140 µm), bilaterally targeting GP and substantia nigra reticulata (SNr). During task performance, wide-band (0.1–9000 Hz) electrophysiological data were recorded with a sampling rate of 30000/s using an Intan RHD2000 recording system (Intan Technologies). All signals were initially referenced to a skull screw (tip-flattened) on the midline 1 mm posterior to lambda. For spike detection we re-referenced to an electrode common average, and wavelet-filtered (Wiltschko et al., 2008) before thresholding. For spike sorting we performed automatic clustering units using MountainSort (Chung et al., 2017) followed by manual curation of clusters. Tetrodes were usually moved by 159 µm every 2–3 sessions. To avoid duplicate neurons we did not include data from the same tetrode across multiple sessions unless the tetrode had been moved by >100 µm between those sessions. Based on waveform and firing properties we further excluded an additional 25 units that appeared to be duplicates even though the tetrode had been moved. After recording was complete, we anesthetized rats and made small marker lesions by applying 10 µA current for 20 s for one or two wires of each tetrode. After perfusing the rats and slicing (at 40 µm) tissue sections were stained with cresyl violet and compared to the nearest atlas section (Paxinos, 2006).
Data analysis
Request a detailed protocolSmoothed firing rates were obtained by convolving each spike time with a Gaussian kernel (30 ms SD). Firing rates were normalized (Z-scored) using the neuron’s session-wide mean and SD. Normalized average time series for contra and ipsi actions (500 ms each, around Center Out) were concatenated and used to construct a population activity matrix R = TC by N, with T = 251 (timepoints, at 2 ms intervals), C = 2 (ipsi/contra conditions), and N = 376 (the number of neurons). We subtracted the mean of each of the N columns to make data zero-centered, then performed principal components analysis (PCA) over matrix R using the MATLAB ‘svd’ function. Using the right singular vectors (W), we can calculate the PC scores (S) as S = RW. For example, the first column of S contains the first principal component (PC1) over time, and the first column of W contains the weights for each of the N units for PC1. We used the first 10 PCs for analysis, and the Euclidean distance between conditions was compared in this 10-D space. The projections onto the Initiation or Selection Axes were calculated as the dot product of the state space position vector and the axis vector. State-space positions around the Go cue (or Stop cue) were calculated using the set of weights W to project the Go cue–aligned (or Stop cue-aligned) firing rates into the 10-D PC space. In other words, each neuron has a weight for each PC, and we calculate a net population position along each PC by multiplying each neuron’s instantaneous firing rate by its weight, and summing across all neurons.
To test if whether state-space positions for two conditions (e.g. Successful- and Failed-Stops) are significantly separated, we ran permutation tests by randomly shuffling the trial conditions for each neuron (10000 shuffles for each test). Then, the distance in the population state space at each time point was reconstructed using the firing rate differences between the shuffled trial averages for each condition. For example, if the mean FR of a unit (n) in surrogate Failed Stop trials (c1) and surrogate Successful Stop trials (c2) at Stop cue time (t) is and , respectively, the difference between two conditions in k-dimension, is:
Repeated shuffling produces a surrogate data distribution for differences at each time point, and the original difference between conditions is compared to this distribution to determine statistical significance.
Data availability
The neurophysiology data and analysis code used in this study are available at figshare: https://doi.org/10.6084/m9.figshare.12367541.
-
figshareGlobus pallidus dynamics reveal covert strategies for behavioral inhibition.https://doi.org/10.6084/m9.figshare.12367541
References
-
Npas1+-Nkx2.1+ neurons are an integral part of the Cortico-pallido-cortical loopThe Journal of Neuroscience 40:743–768.https://doi.org/10.1523/JNEUROSCI.1199-19.2019
-
Independent coding of movement direction and reward prediction by single pallidal neuronsJournal of Neuroscience 24:10047–10056.https://doi.org/10.1523/JNEUROSCI.2583-04.2004
-
Inhibition and the right inferior frontal cortex: one decade onTrends in Cognitive Sciences 18:177–185.https://doi.org/10.1016/j.tics.2013.12.003
-
Stop the presses: dissociating a selective from a global mechanism for stoppingPsychological Science 19:1146–1153.https://doi.org/10.1111/j.1467-9280.2008.02216.x
-
Motor function of the monkey globus pallidus. 2. cognitive aspects of movement and phasic neuronal activityBrain: A Journal of Neurology 114:1685–1702.https://doi.org/10.1093/brain/114.4.1685
-
A proactive mechanism for selective suppression of response tendenciesJournal of Neuroscience 31:5965–5969.https://doi.org/10.1523/JNEUROSCI.6292-10.2011
-
Subthalamic nucleus stimulation reverses mediofrontal influence over decision thresholdNature Neuroscience 14:1462–1467.https://doi.org/10.1038/nn.2925
-
Supplementary motor area exerts proactive and reactive control of arm movementsJournal of Neuroscience 30:14657–14675.https://doi.org/10.1523/JNEUROSCI.2669-10.2010
-
Disinhibition as a basic process in the expression of striatal functionsTrends in Neurosciences 13:277–280.https://doi.org/10.1016/0166-2236(90)90109-N
-
Dimensionality reduction for large-scale neural recordingsNature Neuroscience 17:1500–1509.https://doi.org/10.1038/nn.3776
-
Activity of pallidal neurons during movementJournal of Neurophysiology 34:414–427.https://doi.org/10.1152/jn.1971.34.3.414
-
Single-unit activity in the globus pallidus and neostriatum of the rat during performance of a trained head movementExperimental Brain Research 88:517–530.https://doi.org/10.1007/BF00228181
-
A computational model of action selection in the basal ganglia. I. A new functional anatomyBiological Cybernetics 84:401–410.https://doi.org/10.1007/PL00007984
-
Independence of movement preparation and movement initiationJournal of Neuroscience 36:3007–3015.https://doi.org/10.1523/JNEUROSCI.3245-15.2016
-
Pathological synchronization in Parkinson's disease: networks, models and treatmentsTrends in Neurosciences 30:357–364.https://doi.org/10.1016/j.tins.2007.05.004
-
Role for subthalamic nucleus neurons in switching from automatic to controlled eye movementJournal of Neuroscience 28:7209–7218.https://doi.org/10.1523/JNEUROSCI.0487-08.2008
-
A fronto-striato-subthalamic-pallidal network for goal-directed and habitual inhibitionNature Reviews Neuroscience 16:719–732.https://doi.org/10.1038/nrn4038
-
Cortical activity in the null space: permitting preparation without movementNature Neuroscience 17:440–448.https://doi.org/10.1038/nn.3643
-
Expectation of reward modulates cognitive signals in the basal gangliaNature Neuroscience 1:411–416.https://doi.org/10.1038/1625
-
What, if, and when to move: basal ganglia circuits and Self-Paced action initiationAnnual Review of Neuroscience 42:459–483.https://doi.org/10.1146/annurev-neuro-072116-031033
-
A proactive task set influences how response inhibition is implemented in the basal gangliaHuman Brain Mapping 37:4706–4717.https://doi.org/10.1002/hbm.23338
-
Inhibitory control and psychopathology: a meta-analysis of studies using the stop signal taskJournal of the International Neuropsychological Society 16:1064–1076.https://doi.org/10.1017/S1355617710000895
-
On the ability to inhibit simple and choice reaction time responses: a model and a methodJournal of Experimental Psychology: Human Perception and Performance 10:276–291.https://doi.org/10.1037/0096-1523.10.2.276
-
Proactive selective response suppression is implemented via the basal gangliaJournal of Neuroscience 33:13259–13269.https://doi.org/10.1523/JNEUROSCI.5651-12.2013
-
Reward pays the cost of noise reduction in motor and cognitive controlCurrent Biology 25:1707–1716.https://doi.org/10.1016/j.cub.2015.05.038
-
Proactive and reactive inhibitory control in ratsFrontiers in Neuroscience 8:104.https://doi.org/10.3389/fnins.2014.00104
-
Neural and behavioral mechanisms of proactive and reactive inhibitionLearning & Memory 23:504–514.https://doi.org/10.1101/lm.040501.115
-
The LATER model of reaction time and decisionNeuroscience & Biobehavioral Reviews 64:229–251.https://doi.org/10.1016/j.neubiorev.2016.02.018
-
Neural basis of adaptive response time adjustment during saccade countermandingJournal of Neuroscience 31:12604–12612.https://doi.org/10.1523/JNEUROSCI.1868-11.2011
-
Canceling actions involves a race between basal ganglia pathwaysNature Neuroscience 16:1118–1124.https://doi.org/10.1038/nn.3456
-
A Pause-then-Cancel model of stopping: evidence from basal ganglia neurophysiologyPhilosophical Transactions of the Royal Society B: Biological Sciences 372:20160202.https://doi.org/10.1098/rstb.2016.0202
-
Psychology and neurobiology of simple decisionsTrends in Neurosciences 27:161–168.https://doi.org/10.1016/j.tins.2004.01.006
-
Perceptual decision making in less than 30 millisecondsNature Neuroscience 13:379–385.https://doi.org/10.1038/nn.2485
-
Pallidal discharge related to the kinematics of reaching movements in two dimensionsJournal of Neurophysiology 77:1051–1074.https://doi.org/10.1152/jn.1997.77.3.1051
-
Proactive adjustments of response strategies in the stop-signal paradigmJournal of Experimental Psychology: Human Perception and Performance 35:835–854.https://doi.org/10.1037/a0012726
-
Wavelet filtering before spike detection preserves waveform shape and enhances single-unit discriminationJournal of Neuroscience Methods 173:34–40.https://doi.org/10.1016/j.jneumeth.2008.05.016
-
Expectations and violations: delineating the neural network of proactive inhibitory controlHuman Brain Mapping 34:2015–2024.https://doi.org/10.1002/hbm.22047
Article and author information
Author details
Funding
National Institute of Mental Health (R01 MH101697)
- Joshua D Berke
National Institute on Drug Abuse (R01 DA045783)
- Joshua D Berke
National Institute of Neurological Disorders and Stroke (R01NS078435)
- Joshua D Berke
University of California, San Francisco
- Joshua D Berke
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
Acknowledgements
We thank Michael Farries and Ali Mohebi for technical advice, Wei Wei and Vikaas Sohal for comments on the manuscript, and Alejandro Jimenez Rodriguez for discussions. This work was supported by NIH grants R01MH101697, R01NS078435 and R01DA045783, and the University of California, San Francisco.
Ethics
Animal experimentation: All animal experiments were approved by the University of California, San Francisco Committee for the Use and Care of Animals (approval number: AN181071).
Copyright
© 2020, Gu 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
-
- 1,861
- views
-
- 354
- downloads
-
- 23
- citations
Views, downloads and citations are aggregated across all versions of this paper published by eLife.
Download links
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Further reading
-
- Developmental Biology
- Neuroscience
Two major ligand-receptor signaling axes – endothelin Edn3 and its receptor Ednrb, and glial-derived neurotrophic factor (GDNF) and its receptor Ret – are required for migration of enteric nervous system (ENS) progenitors to the hindgut. Mutations in either component cause colonic aganglionosis, also called Hirschsprung disease. Here, we have used Wnt1Cre and Pax2Cre in mice to show that these driver lines label distinct ENS lineages during progenitor migration and in their terminal hindgut fates. Both Cre lines result in Hirschsprung disease when combined with conditional Ednrb or conditional Ret alleles. In vitro explant assays and analysis of lineage-labeled mutant embryos show that GDNF but not Edn3 is a migration cue for cells of both lineages. Instead, Edn3-Ednrb function is required in both for GDNF responsiveness albeit in different ways: by expanding the Ret+ population in the Pax2Cre lineage, and by supporting Ret function in Wnt1Cre-derived cells. Our results demonstrate that two distinct lineages of progenitors give rise to the ENS, and that these divergently utilize endothelin signaling to support migration to the hindgut.
-
- Neuroscience
The slow-intrinsic-pacemaker dopaminergic (DA) neurons originating in the ventral tegmental area (VTA) are implicated in various mood- and emotion-related disorders, such as anxiety, fear, stress and depression. Abnormal activity of projection-specific VTA DA neurons is the key factor in the development of these disorders. Here, we describe the crucial role of the NALCN and TRPC6, non-selective cation channels in mediating the subthreshold inward depolarizing current and driving the firing of action potentials of VTA DA neurons in physiological conditions. Furthermore, we demonstrate that down-regulation of TRPC6 protein expression in the VTA DA neurons likely contributes to the reduced activity of projection-specific VTA DA neurons in chronic mild unpredictable stress (CMUS) depressive mice. In consistent with these, selective knockdown of TRPC6 channels in the VTA DA neurons conferred mice with depression-like behavior. This current study suggests down-regulation of TRPC6 expression/function is involved in reduced VTA DA neuron firing and chronic stress-induced depression-like behavior of mice.