Inhibition of a pre-potent response, usually when the response is no longer appropriate, is defined as behavioral inhibition (Wong, 2013). Action cancelation is a specific type of inhibition where an ongoing motor response is stopped while it is already being executed. The stop signal task (SST) was developed to measure this ability of action cancelation (Logan et al., 1984; Tannock et al., 1989). In this task, subjects are first trained to perform a fast reaction with a defined beginning and end, the so called go response following a go signal. In the stop signal task with mice this “go signal” corresponds to an initial nose poke which triggers a light signal and immediately continues into a rapid motor action that ends in a second nose poke to a neighboring nose poke hole.

After a go response has been established, a stop signal is introduced in some of the trials (e.g., 20%). It informs the subject to stop its ongoing motor response such that after the first nose poke the second nose poke is suppressed. The stop signal is given with some delay (stop signal delay, SSD) after the first nose poke initiates the motor response. The longer this delay the harder it is for a subject to respond to the stop signal and inhibit its ongoing motor response.

Deficits in action cancelation ability occur in many psychiatric and neurological conditions such as attention deficit and hyperactivity disorder (Schachar et al., 2000; Overtoom et al., 2002; Alderson et al., 2007), schizophrenia (Hughes et al., 2012; Yang et al., 2020), obsessive-compulsive disorder (McLaughlin et al., 2016; Mancini et al., 2018), Parkinson’s disease (Gauggel et al., 2004; Obeso et al., 2011) and substance abuse (Monterosso et al., 2005; Li et al., 2006; Zhang et al., 2018). To understand the contribution of action cancelation deficits to those pathological conditions and develop better treatment, animal models are crucial. Furthermore, the stop signal task is crucial for dissociating the neurobiological mechanisms of action cancelation and action restraint (Johnstone et al., 2007; Eagle et al., 2008; Dambacher et al., 2014).

The stop signal task, first introduced for rats two decades ago (Feola et al., 2000), has been used to investigate: (i) anatomical structures and neurotransmitter systems involved in action cancelation (Eagle and Robbins, 2003a,b; Eagle et al., 2007; Bari et al., 2009, 2011), (ii) neural activity correlates of action cancelation (Bryden et al., 2012; Leventhal et al., 2012; Schmidt et al., 2013; Mayse et al., 2015; Mallet et al., 2016), and (iii) cognitive aspects of action cancelation (Beuk et al., 2014; Mayse et al., 2014; Xu et al., 2019). The stop signal task for mice (Humby et al., 2013) has been used to identify genetic correlates of action cancelation (Davies et al., 2014, 2019; Dent et al., 2016). Considering that psychiatric conditions with an impairment in action cancelation such as attention deficit and hyperactivity disorder (Thapar and Cooper, 2016; Faraone and Larsson, 2019) and schizophrenia (Cardno and Gottesman, 2000; Sullivan et al., 2003) have a significant genetic component, studies using this task with genetic mouse models have remained surprisingly scarce.

One of the factors that have limited the use of genetic mouse models with the stop signal task is the labor requirement due to the task’s long training period (∼44 days, Humby et al., 2013). In general, labor requirement can be reduced by utilizing automated home-cage based experimentation, that allows continuous experimentation with minimal experimenter involvement (Endo et al., 2011; Schaefers and Winter, 2011; Balci et al., 2013; Remmelink et al., 2016, 2017). Moreover, such minimized experimenter involvement can also reduce data variability (Crabbe et al., 1999; Sorge et al., 2014), and lead to more consistent results across laboratories (Lipp et al., 2005; Krackow et al., 2010).

We previously developed an ID chip based gating mechanism for home-cage based experimentation (Winter and Schaefers, 2011; Caglayan et al., 2021). This allows continuous testing with self-initiated individual experimental sessions by letting only one individual at a time into the experimental compartment. Therefore, individuals are free from interference by cage mates during sessions. Moreover, this gating mechanism allows setting individual inter-session intervals so that a recently admitted individual cannot reenter. Longer intersession intervals (e.g., 1 h) ensure high engagement during a time-restricted session (e.g., 30 min). Rodents readily adapt to the gating mechanism and the sorting procedure and have been shown to perform various operant tasks (Schaefers and Winter, 2011; Winter and Schaefers, 2011; Rivalan et al., 2017; Qiao et al., 2018; Caglayan et al., 2021). In this study, we employed a gating mechanism as one component of a more efficient stop signal experimental procedure for mice.

Another factor that limits the use of mice in the stop signal task is the requirement for a rather loud stop signal (100 dB, 300 ms white noise) to train mice (Humby et al., 2013). It necessitates placing the operant chamber in a sound-attenuating box, which requires considerable space. Furthermore, despite the use of these relatively loud stop signals in previous studies behavioral inhibition observed in mice has remained relatively low (∼85%) (Humby et al., 2013; Davies et al., 2014) on stop signal baseline trials, i.e., trials during which the stop signal is given simultaneously with the “go.” During these baseline trials (without stop signal delay), the task becomes a go/no-go task and measures action restraint rather than action cancelation (Eagle et al., 2008). As action restraint ability will also affect the outcomes during trials with stop signal delays, any inability for action restraint will confound action cancelation measurements. A lower level of baseline inhibition therefore limits the sensitivity of the test as it reduces the measurable outcome for action cancelation from 0–100% to for example 0–85% behavioral inhibition. Such ceiling or floor effects are shown to affect the observed effect sizes (Šimkovic and Träuble, 2019) and in turn can lead to decreased sensitivity. Moreover, variability in action restraint ability will add to the observed variability of the outcome measurements. Increased variability will decrease the sensitivity of a test. Furthermore, as stop signal reaction times are calculated based on the probability of behavioral inhibition, the resulting estimates will be more accurate the smaller the influence from action restrain. Hence, achieving relatively high baseline inhibition with low variance can increase the sensitivity and accuracy of the stop signal task. We were able to greatly improve baseline inhibition by introducing a mild air puff as a negative reinforcer, as described in previous go/no-go tasks (Harrison et al., 2016; Chen et al., 2019; Maor et al., 2020).

With this study we established an automated stop signal task for group housed mice that is based on a gating mechanism for individual separation and uses a mild air puff with escape option to enhance operant conditioning. We estimated individual stop signal reaction times (SSRT) and show that mice can be trained to a high level of baseline inhibition in the stop signal task.

The results of this study show that it is possible to greatly improve both the sensitivity and the animal throughput of the mouse stop signal task. This should greatly increase the usability of the SST procedure with mice. Our modifications aimed at improving two aspects of the experiment: Implementing automated procedures made the SST both less labor intensive for experimenters and faster to complete for mice. Introducing a negative reinforcer significantly improved the SST test procedure by greatly increasing the sensitivity of the test. In the following, we shall address this second point first.

The usefulness of any diagnostic procedure depends among others on its sensitivity to the parameter of interest. In the stop signal task this parameter is the stop signal reaction time that is the time required to process the stop signal and cancel the ongoing action. The stop signal is given before the expected time of the completion of the go response as “advance notice interval” and the shorter this becomes (equivalent to a longer delay since action onset) the more difficult it is to respond to it by stopping the action. Estimating SSRT from behavioral performance during tasks critically rests on the assumption that a subject will always stop after processing the stop signal. This is because analysis cannot easily distinguish between the inability of the animal to respond to the stop signal and an animal ignoring the stop signal. Therefore, it is especially important that subjects respond at a high rate to the stop signal in this task. This action restraint ability is trained in baseline trials.

Recognizing that the inability for action restraint can confound action cancelation measurements, initial studies with rats included a correction for baseline stopping errors (Eagle and Robbins, 2003a,b). However, later studies (Eagle et al., 2007; Bari et al., 2009, 2011, 2015) have not followed that example, maybe because rats showed a generally higher level of baseline inhibition. On the other hand, in mice, baseline behavioral inhibition only reached levels around 85% (even with a very loud stop signal) in previous studies (Humby et al., 2013; Davies et al., 2014). In the present study, we increased the baseline inhibition of mice in stop signal trials to an average of 94%, a significant increase over the 84% reported previously (especially considering that, in contrast to previous studies, the baseline inhibition was corrected for omission rate). The major factor responsible for this high degree of task compliance was our air puff negative reinforcer. While air puffs have been used commonly with auditory go/nogo procedures (Otazu et al., 2009; Hangya et al., 2015; Harrison et al., 2016; Chen et al., 2019; Christensen et al., 2019; Maor et al., 2020), our use of an air puff negative reinforcer is novel for the SST. Failing to inhibit the operant response on a stop signal trial commonly only results in to an unrewarded trial and a timeout, thereby reducing the overall probability of obtaining rewards. The introduction of a mild air puff adds an additional aversive dimension to the task procedure, potentially motivating animals to be more attentive to stop signals. In addition, the air puff might have improved the animal‘s ability to separate go trial omissions due to responses after LH from incorrect responses following a stop signal, as otherwise both would have been indicated by a timeout (signaled by a house light). The resulting high baseline inhibition levels result in more precise SSRT and inhibition curve estimates. Moreover, they increase the overall sensitivity of the test as it broadens the range of performances in which experimental groups can vary from each other and thus also prevents a floor effect which could otherwise occur with longer stop signal delays.

In addition, training all mice to high baseline inhibition levels, decreases the variability in baseline inhibition within a group which would otherwise add to the variability of the response inhibition when assessing action cancelation ability. Before we introduced an air puff as a negative reinforcer, we observed high levels of variability in baseline inhibition during a pilot study (Supplementary Results, Supplementary Figure 2). Furthermore, addition of an air puff also decreased baseline inhibition variability compared to a previous study in mice. The standard deviation of baseline inhibition was 8.9 in the conventional setup (Humby et al., 2013, calculated by using the standard error of mean reported in Supplementary Material), while the standard deviation of baseline inhibition in the present study was 5. The introduction of a negative reinforcer thus improved action cancelation assessment in the stop signal task in mice.

In addition, we were able to reduce the sound intensity of the acoustic stop signal from 100 dB commonly used in previous studies (Humby et al., 2013; Davies et al., 2014) to 70 dB. Apart from animal welfare concerns this was especially important for us, as our home cage based systems did not include sound attenuating cabinets. While mice could hear the stop signal in their home cage, rodents are known to be capable of learning the context of auditory signals (Honey and Good, 1993; de Hoz and Nelken, 2014). Similarly, mice learnt auditory cues in an automated home-cage-based auditory discrimination cage, despite multiple such cages being present in the same room (Francis and Kanold, 2017).

After the mice had been trained on the stop signal task, they proceeded to the probe sessions. We used the probe session data to construct inhibition curves. Overall, successful behavioral inhibition decreased with increasing SSD as expected (Logan, 1994), indicating successful SST implementation. Different from earlier authors we analyzed data by fitting a logistic model (GLMM) instead of using ANOVA since the dependent variable was binary (inhibited vs. non-inhibited) and the two horse race model predicts an S-shaped inhibition curve.

Here, we briefly recount the two horse race model which underlies analysis. This model assumes that the go response and the inhibition of that response are two independent processes running concurrently. During a stop trial, a subject may either complete its go response or stop. Stopping occurs when the response inhibition process is completed before the ongoing go response. Both processes are assumed to have a normal distribution and the inhibition process is shorter than the go process. Therefore, by gradually increasing a stop signal delay this leaves less and less time for successful inhibition. The stop signal reaction time SSRT is then inferred from the go reaction time distribution of go trials and the probability of inhibition with a given stop signal delay (Logan et al., 1984; Band et al., 2003; Verbruggen and Logan, 2009).

After constructing the inhibition curve, we estimated SSRTs for each individual using the two horse race model. The median SSRT was 88 ms and therefore considerably lower than ∼350 ms (Humby et al., 2013). However, previously reported SSRTs for rodents have varied widely from ∼150 ms (Mayse et al., 2014) to ∼350 ms (Eagle and Robbins, 2003b), and even within the same study, SSRTs from one batch to another varied significantly (Eagle and Robbins, 2003b, 298 vs. 342 ms in different batches). Therefore, natural variability between batches might contribute to some of the difference between our and previous findings of SSRT values. In macaques, SSRTs estimated from saccades (eye movements) are in a range similar to our results (60–140 ms, Hanes et al., 1998).

We would like to address one other aspect of our study that may have facilitated shorter SSRTs. When compared to previous studies our overall mean reaction times were longer (921 ms compared to 837 ms in Eagle and Robbins, 2003a and 656 ms in Humby et al., 2013). Two factors are likely to have contributed to longer reaction times. First, our experimental apparatus was different. A previous study with mice used a 5-hole chamber where nose poke sensors are directly behind the hole opening and triggered by a shallow poke. Our chamber had three regular food receptacles which required full head insertion to trigger the photo gate. Also, food receptacles were farther apart than holes on a 5-hole wall. The different geometry led to different movement kinematics which needed more time. Equally important may be a second factor that was a consequence of not putting the mice on food restriction. Without food restriction our mice may have been less motivated to perform the task without omissions (Mai et al., 2012; Burnett et al., 2016). As a consequence, we as experimenters, had to adjust the limited hold time interval to a duration that kept experimental omission rate below 30%. With mice that were less motivated our limited hold time interval was therefore most likely longer than it would have been with mice under food restriction. This experimental condition gave mice more time, and allowed mice to move less fast. This chain of reasoning has a further consequence. Slower movement might have eliminated the so-called “ballistic component” of the go response. The “ballistic component” corresponds to the part of the go response that cannot be stopped (e.g., due to muscle activation) even though the stop signal is processed (Logan, 1994; Verbruggen and Logan, 2009). In our case, with slower movement, this “ballistic component” effect may have been reduced thus enabling the mice to stop even on short notice. A corollary of this effect is that we also obtained a relatively high standard deviation and coefficient of variation for pooled go reaction times (mean: 921 ms, SD: 200 ms, CV: 21.7%) compared to a previous rat study (mean: 837 ms, SD: 106 ms, CV: 12.7%) (Eagle and Robbins, 2003a). The estimated standard deviations for individual animals ranged between 140 and 213 ms in our study. In a future study a kinematic analysis of stop signal responding using e.g., DeepLabCut (Mathis and Mathis, 2020) may be useful to further improve the sensitivity and resolution of this diagnostic procedure.

Putting rodents on food restriction is common practice to increase performance levels in operant choice experiments with food rewards (Mai et al., 2012; Burnett et al., 2016). Although food restriction played no part in the present study, the potential effect of food restriction increasing motivation and subsequently resulting in more reliable estimates of SSRT should be further investigated. However, even if there is such an effect for probe trials, we show that there is no need to impose food restriction through all training stages. We demonstrated that mice reliably learn the go response and to stop upon a stop signal without any food restriction. If food restriction is to result in more precise SSRT estimates, this food restriction can be imposed at a very late stage of the training just before probe sessions.

In future, the sensitivity of our task can be further validated by using manipulations previously known to affect SSRTs. One commonly used manipulation with SST is lesioning the medial prefrontal cortex (mPFC, Eagle and Robbins, 2003a; Humby et al., 2013). Lesions in mPFC cause impairments in action cancelation similar to humans suffering from cortical dysfunction (Rubia et al., 1999; Robbins, 2007). One can also examine the effects of enhancement with ADHD drugs in both healthy and lesioned animals. Drugs used for ADHD medication (methylphenidate and atomoxetine) are known to improve action cancelation in rodents (Eagle and Robbins, 2003b; Humby et al., 2013) akin to the improvements observed in human patients (Aron et al., 2003; Chamberlain et al., 2007). With similar interventions the predictive and construct validity of our SST can be corroborated.

In addition to higher sensitivity through higher baseline inhibition, our home cage based procedure has the following benefits: (i) a higher level of both animal welfare and experimenter efficiency due to the omission of food restriction and minimized contact between animals and experimenter (ii) individual sessions free from cage mate interference due to the temporary separation of individuals via the sorter, and (iii) reduction in the duration of the training phase of the experiment %17 (median 36.5 days with our system vs ∼44 days in Humby et al., 2013 using the same strain of mice). Other benefits of using a sorter or gating system for home-cage experimentation have been discussed previously (Schaefers and Winter, 2011; Winter and Schaefers, 2011; Rivalan et al., 2017; Caglayan et al., 2021). The additional benefits make our procedure of performing the stop signal task a convenient and efficient tool for studying deficits in action cancelation. Our procedure should greatly facilitate such investigations in mice and genetic mouse models which will remain crucial to investigate the genetic correlates and etiology of action cancelation deficits under pathological conditions such as ADHD, schizophrenia, and OCD.

Our newly developed mouse stop signal task procedure led to a significantly higher level of compliance by the animals and therefore to higher baseline inhibition. We achieved this by introducing an air puff as a negative reinforcer during stop signal trials. This approach solved a problem of the SST in which mice show a tendency to not restrain their actions despite their general ability to do so and subsequently led to higher task sensitivity. The inhibition curves we obtained from mice confirm that our newly developed procedure yields the expected measure of action cancelation ability (with longer delays, the action is harder to cancel). Furthermore, our procedure of continuous, automated experimentation allowed high throughput under conditions without food restriction, without extensive experimenter involvement and without loud auditory signals. These improvements should facilitate a wider application of the mouse stop signal task, especially important for investigating correlates of action cancelation impairment in gene manipulated mouse models.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The animal study was reviewed and approved by the animal welfare officer of Humboldt University.

AC and YW conceived the project. AC conducted the experiment under the supervision of YW and KS. AC performed data analysis and prepared the figures (except Figure 2 was prepared by KS). All authors contributed to writing of the manuscript.

YW owns PhenoSys equity. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.