Abstract
Automatic detection of a surprising change in the sensory input is a central element of exogenous attentional control. Stimulus-specific adaptation (SSA) is a potential neuronal mechanism detecting such changes and has been robustly described across sensory modalities and different instances of the ascending sensory pathways. However, little is known about the relationship of SSA to perception. To assess how deviating stimuli influence target signal detection, we used a behavioral cross-modal paradigm in mice and combined it with extracellular recordings from the primary somatosensory whisker cortex. In this paradigm, male mice performed a visual detection task while task-irrelevant whisker stimuli were either presented as repetitive “standard” or as rare deviant stimuli. We found a deviance distraction effect on the animals' performance: Faster reaction times but worsened target detection was observed in the presence of a deviant stimulus. Multiunit activity and local field potentials exhibited enhanced neuronal responses to deviant compared with standard whisker stimuli across all cortical layers, as a result of SSA. The deviant-triggered behavioral distraction correlated with these enhanced neuronal deviant responses only in the deeper cortical layers. However, the layer-specific effect of SSA on perception reduced with increasing task experience as a result of statistical distractor learning. These results demonstrate a layer-specific involvement of SSA on perception that is susceptible to modulation over time.
SIGNIFICANCE STATEMENT Detecting sudden changes in our immediate environment is behaviorally relevant and important for efficient perceptual processing. However, the connection between the underpinnings of cortical deviance detection and perception remains unknown. Here, we investigate how the cortical representation of deviant whisker stimuli impacts visual target detection by recording local field potential and multiunit activity in the primary somatosensory cortex of mice engaged in a cross-modal visual detection task. We find that deviant whisker stimuli distract animals in their task performance, which correlates with enhanced neuronal responses for deviants in a layer-specific manner. Interestingly, this effect reduces with the increased experience of the animal as a result of distractor learning on statistical regularities.
- attention
- barrel cortex
- cortical layers
- deviant distraction
- local field potentials
- stimulus-specific adaptation
Introduction
Detecting unexpected changes in the continuous flow of sensory information is essential for everyday activities in humans (e.g., driving a car) and animals (e.g., foraging). Behavioral responses depend on the ability to detect deviancy in ongoing stimuli, which can be either behaviorally relevant (e.g., approach of predators) or distracting (e.g., approach of a kin). The process of neuronal change detection is reflected by an additional negative wave in the event-related potentials in humans, mismatch negativity (MMN) (for review, see Näätänen et al., 2007). Typically, MMN is evoked by using an oddball paradigm, in which an unexpected, deviant stimulus is embedded in a sequence of expected standard stimuli. Distraction by the deviant stimulus is observed when subjects are asked to perform an attention-demanding primary task while task-irrelevant deviant and standard stimuli are presented concurrently (for review, see Escera and Corral, 2007). These studies postulate an involuntary attentional shift toward and subsequently from the unexpected deviant stimulus observed in human-EEG signals (P3a and RON, respectively) (Leiva et al., 2015). Attentional reorientation, from the deviant back to the task at hand, usually has a behavioral impact measured by increased response times in the primary task (Berti and Schröger, 2003; Parmentier et al., 2011; Leiva et al., 2015). While numerous studies have investigated MMN and deviance distraction using EEG and fMRI with comparatively low spatial resolution, much less is known about its behavioral relevance at the level of cortical layers and multiunit activity (MUA). Stimulus-specific adaptation (SSA) has been proposed as a single-neuron correlate of MMN and indicates how neuronal populations adapt to expected stimuli while remaining highly responsive to unexpected, deviant stimuli eliciting an early onset deviant response across layers of primary sensory cortices (Ulanovsky et al., 2004; Katz et al., 2006; Musall et al., 2017). Little is known about the relationship between SSA and perception and its underlying neuronal mechanisms. Specifically, to our knowledge, no study has investigated deviance distraction and the role of SSA on exogenous attention. In this study, we used a cueing paradigm, commonly used to study visual selective attention and recently observed to be exhibited in mice (Wang and Krauzlis, 2018), in combination with an oddball paradigm, commonly used to elicit MMN (Restuccia et al., 2009). We performed simultaneous extracellular recordings from the primary somatosensory cortex (S1) to assess whether SSA at the multiunit level influences perceptual decision-making. In our cross-modal paradigm, head-fixed mice were trained to perform a visual detection task while task-irrelevant deviant and standard whisker stimuli were presented. With a cross-modal design, we measured neuronal responses to the oddball stimulus presentation without confounds of target and deviant stimulus processing. As the mice reported signal detection by licking a central spout, and the visual target and deviant stimulus occurred either in the right or left visual field and whisker, respectively, lateralized effects on perceptual choice were measured without confounds by motor response biases. Importantly, the spatial deviant stimulus did not carry any relevant information about the following visual target location. We found that uninformative deviant stimuli yielded a distracting behavioral impact as measured in worsened perceptual sensitivity. This distracting effect was triggered by deviant stimuli and observed in decreased response times exclusively when the deviant was spatially congruent to the visual target location. These deviance distracting effects correlated with enhanced evoked responses for deviant stimuli compared with standard stimuli. Our results suggest that SSA plays a role on perception in a layer-specific manner. Correlations between our findings in neurophysiology and in behavior were predominantly observed in the infragranular layers of S1, indicating a tendency toward a more critical role of deep cortical layers on attention. We also demonstrate how SSA and its effect on perception is robustly modulated with experience potentially as a result of statistical learning, increasing the predictability of the distracting deviant (Chelazzi et al., 2019; van Moorselaar and Theeuwes, 2021).
Materials and Methods
Animals
All experimental and surgical procedures were approved by the local veterinary authorities of the Canton Zurich, Switzerland, and were conducted in accordance with the guidelines published in the European Communities Council Directives 2010/63/EU. Procedures were conducted on a total of 8 male mice (C57BL/6J; Charles River Laboratories), housed in enriched cages in a 12 h reversed light/dark cycle, with the dark phase starting at 8:00 A.M. All experimental procedures and behavioral training were done in the dark phase from 9:00 A.M. to 8:00 P.M., allowing the animals to adjust to their active phase first. All mice were implanted with a light weight head post at age 7-10 weeks (animal weight 23.9 ± 1.4 g), to allow head fixation during behavioral training and neuronal recording. Mice were in group housing (3 or 4 cage mates) before the initial surgical procedure and subsequently single housed, because of social incompatibility, until the end of the experiment. After finishing behavioral training, mice underwent electrode implantation for neurophysiological recordings during behavior. All mice had free access to food, but water was scheduled for training/experimental sessions. During all training, experimental and health monitoring procedures, mice were exclusively tunnel-handled.
Surgical procedures
For all surgical procedures, mice were briefly induced with 2.5% isoflurane in oxygen anesthesia and injected with 1 mg/kg meloxicam (Metacam, Boehringer Ingelheim) and placed in a stereotaxic apparatus. Subsequently, the level of isoflurane was decreased to 1.4% for remaining surgical procedures. Body temperature was maintained at 37°C using a homeothermic heating pad (Harvard Apparatus). Eyes were protected using ophthalmic gel (vitamin A, Bausch+Lomb). For 2 d after surgery, 0.5 mg/ml of oral meloxicam (Metacam, Boehringer Ingelheim) diluted in liquid-food was provided. All animals were healthy, and weight was monitored daily after surgery.
Head-post implant
After 1 week of habituation to the experimenter and tunnel handling, all mice underwent their first surgery for head-post implantation. A small incision was made in the scalp, the skull was cleaned, and a customized aluminum head post was implanted using dental cement (Tetric EvoFlow, Ivoclar Vivadent), aligned to bregma and midline. Mice recovered for at least 3 d before start of behavioral training.
Electrode implant
After successfully learning the behavioral task during training sessions, mice underwent a second surgery for electrode implantation. To guide placement of the electrode shanks into the left C1 and B1 barrel, intrinsic optical imaging was performed through the intact skull (Grinvald et al., 1986). After imaging, a small craniotomy was made over the ROI. A 4-shank linear silicon probe (impedances of 1-2 mΩ at 1 kHz; electrode contact surface of 177µm2, 8 contacts/shank with 100 µm distance between contacts, A4x8-5 mm-100-200-177-H32 21 mm, NeuroNexus) was slowly inserted into the cortex (∼100 µm/min) until the topmost electrode touched layer I. The craniotomy was first sealed with silicone elastomer (Kwik-Cast, WPI), and then with dental cement. Mice recovered for 3-5 d before start of experiment and were in experiment for 30.7 ± 2.6 sessions.
Behavioral apparatus
The behavioral apparatus consisted of a custom-built light and soundproof (>30 dB attenuation, IAC) behavioral chamber (interior dimensions: 60 cm × 60 cm × 60 cm) in which the head-fixed mouse was placed (see Fig. 1A). Two white-light LEDs (Tru Components) with luminous intensity of 17,000 millicandela were positioned at 45° angle from the animal's midline such that each LED was centered on either the right or left eye by 55° horizontally of the visual hemifield, with a viewing distance of 26 cm (Wang and Krauzlis, 2018). A custom-made water-reward delivery spout was positioned near the snout of the mouse. Lick contacts were detected by a customized piezo sensor. Water flow from an elevated container was controlled via a solenoid valve. A correct lick response by the animal opened the solenoid valve for delivering 5 µl of water. A negative auditory feedback (pink noise) was delivered through a loudspeaker positioned above the head of the mouse. Whisker deflections were provided via piezo-electric elements onto which glass capillaries were glued. The left and right C1 and B1 whiskers were plugged into the glass capillaries. A single whisker deflection consisted of a 120 Hz raised cosine waveform (8.3 ms duration) with an amplitude of 300 µm and a peak velocity of 113.1 mms−1 (see Fig. 1B). Further, a second loudspeaker positioned above the head of the animal delivered constant white noise background to cover potential vibration noise from the piezo-electric elements on whisker stimulation (see Fig. 1A). Whisker and visual stimulation sequences as well as lick detection, auditory feedback, and behavioral control were implemented with custom LabVIEW code (National Instruments). The head-fixed mouse was constantly monitored via a USB camera (IDS Imaging Development Systems).
Cross-modal visual detection task
Head-fixed water-scheduled mice performed a cross-modal Go-NoGo visual detection task, where the animals had to detect a visual LED stimulus, which was preceded by task-irrelevant whisker stimuli (see Fig. 1C). Trials began with an 850 ms no-lick period (licks within this period delayed the trial), followed by a sequence of repetitive, simultaneous and pulsatile deflections (4 Hz) of corresponding right and left single whiskers for 1000 ms. These bilateral stimuli were defined as “standard” stimuli (see Fig. 1D). In 50% of the trials, the standard stimuli were followed by a 250-300 ms interstimulus interval and a single deflection of another whisker on only one side. This stimulus was defined as the “deviant” stimulus. In trials without a deviant stimulus (“standard-only” condition), standard-stimuli presentation was followed by another bilateral single deflection of the standard whiskers, which was done in a subset of animals (n = 5) while in the remaining animals (n = 3) no further whiskers were deflected. In a given session the whisker identity for deviant and standard was not changed. The stimulated whiskers were in all animals B1 and C1, but their role as deviant and standard was switched between sessions (flip-flop design); for example, when B1 was standard and C1 deviant in the first session then C1 was standard and B1 was deviant in the second session and so on. The sequence of whisker stimulus presentation was followed by a varying delay period (stimulus onset asynchrony [SOA] of 50, 150, or 250 ms). The visual target stimulus was always preceded by a hard 150 ms no-lick period to avoid anticipatory and impulsive responses. Licks within this 150 ms no-lick period aborted the trial and excluded it from the behavior analysis. The visual LED target stimulus was then presented for 1200 ms during which mice could provide a lick response (decision window). This visual target was either on the left or the right side. A new trial started after a varying intertrial interval of 2000-3000 ms. Mice had to give a lick response when the visual target was present, regardless of its spatial location (Go trial), and withhold licking when no visual target stimulus was presented (NoGo trial) (see Fig. 1C). If mice correctly licked during the Go trial (hit), they were provided with a small water reward (5 μl). If mice incorrectly licked during the NoGo trial (false alarm), a negative auditory feedback (pink noise) and a timeout (2000 ms) was given (see Fig. 1C,D). If mice correctly withheld licking in NoGo trials or missed to lick in Go trials, these trials were classified as correct rejection and miss trials, respectively; and these were neither punished nor rewarded. Vibrotactile whisker stimuli were irrelevant for detecting the target stimulus. However, based on the spatial location of the deviant whisker stimulus with respect to the visual target location, two different conditions were possible: “congruent,” where both, deviant and target occurred on the same side; and “incongruent,” where the deviant and target occurred on opposite sides. The occurrence of congruent and incongruent conditions was randomized; both had the same probability and were independent of the visual target stimulus (see Fig. 1C). The location of the deviant was therefore noninformative with respect to the target stimulus. All mice performed two sessions per day. Each session consisted of 250-300 trials.
Extracellular recordings were conducted as the animals were performing the task using a 64-channel electrophysiology system (Gain = 1000×, USB-ME64-FAI-System, Multi-Channel System). All electrophysiological signals were acquired at a sample rate of 32 kHz at 12-bit depth. The preamplifier (µPA32, 2xGain, Multichannel Systems) was connected to the implanted Neuronexus probe via an omnetics connector before each recording session.
Water scheduling
During behavioral sessions, mice had scheduled access to water but free access to food. Body weight was monitored and documented before and at the end of daily behavioral sessions, in which mice received water as reward. At the end of training/experiment days, mice received ad libitum water for 5 min. On days without behavior sessions, mice had free access to water. The condition of animals with scheduled access to water was assessed (i.e., grooming, nest building, weight, movement) and documented daily.
Training
Mice were trained to enter a transparent tube voluntarily, subsequently entering the head fixation device for neuronal recording. After habituating the animals to the head fixation (18.6 ± 1.2 sessions), they were trained to detect the visual target stimulus (see Fig. 1E). Training sessions consisted of 30% NoGo trials and 70% Go trials. Mice were rewarded with water when licking in Go trials while receiving negative auditory feedback (pink noise) and a timeout when licking in NoGo trials. Mice performed two sessions per day, each session gradually increasing in trial number starting from 100 to 300 trials. When the animals achieved a hit rate of ≥ 75% and a false alarm rate ≤ 30% on two consecutive sessions, a bilateral whisker stimulation was introduced. Here, both corresponding whiskers (left and right) were deflected at 4 Hz repetition rate, mimicking standard stimuli presentation in the experiment. A deviant stimulus was never presented during the training sessions. Only after electrode implantation, the deviant whisker stimulus was introduced and these initial sessions with a deviant stimulus were defined as “novice” (experimental sessions 1-4). The last six experimental sessions were defined as “expert” (see Fig. 1E). For both training and experiment, the animal's left and right C1 and B1 whiskers were used. Importantly, other whiskers were neither trimmed nor manipulated in any other way the entire time.
Histology
At the end of the experiments, mice were anesthetized with 2.5% isoflurane in oxygen. Electrolytic lesions were made to the deepest and third-from-the-top electrode sites of all four shanks. Briefly, ±10 µA current with a duration of 1500 ms was delivered to the electrode sites to reconstruct recording locations and layers. After the lesion, animals were killed using pentobarbital (200 mg/kg body weight) and transcardially perfused with PBS followed by 4% PFA. The brain was carefully extracted and stored for 24 h in 4% PFA at 4°C. The brain was transferred in PBS and incubated for 30 min at 4°C before tissue was sectioned in the coronal plane at 80 µm thickness with a vibratome (Leica Biosystems). Sections were stained with DAPI solution (Fisher Scientific) diluted in PBS (2 µg/ml DAPI) and mounted on glass slides using Vectashield (Sigma-Aldrich). Fluorescence images were acquired with a confocal microscope and processed using ImageJ (National Institutes of Health).
Quantification and statistical analysis
Behavioral data analysis
Behavioral data were read and processed using custom Python scripts. Performance was measured using four indices: hit rate, false alarm rate, sensitivity d′, and Criterion C. Hit rate was calculated by the following equation: [hit/(hit + miss)]. False alarm rate was calculated by: [false alarm/(false alarm + correct rejection)]. Sensitivity d′ was calculated with the following equation:
Sensitivity d′ illustrates the performance of the animal to discriminate between Go and NoGo trials, while criterion C provides information about the animal's liberal or conservative decision making. Values of sensitivity d′ > 1.0 and criterion C ± 0.5 were considered as good performance. Behavioral data that included a deviant stimulus (congruent or incongruent) were normalized to the standard-only control condition. Sensitivity d′ for congruent and incongruent trials included their individual hit rates and the false alarm rates for trials including a deviant in general. Here, sensitivity d′ illustrates the proportion of congruent versus incongruent hit rates to the overall false alarm rate when the deviant is present.
Sessions with >10% hit trials of the session's total trial number and an overall sensitivity d′ of >0.4 were considered as “animal participation,” consequently excluding sessions that did not meet these criteria. Reaction times (RTs) in milliseconds were extracted as the first lick within the animal's decision window which was aligned to visual target stimulus onset. Based on previous literature (Berditchevskaia et al., 2016), trials with a RT <50 ms were considered as impulsive licks by the animal and excluded from all analysis. To rule out that a 50 ms early-lick threshold is too short, the same behavior analysis was performed with a threshold of 150 ms, which revealed the same effects. For all analysis including the RT, only hit trials were considered. Behavioral data showed no significant difference between performance with B1 versus C1 whisker and were subsequently grouped. Analysis to provide evidence for congruency effects (comparison between congruent and incongruent trials) was separated into short and long SOAs. Delay periods between the last whisker stimulus and the visual target onset of 50 and 150 ms were summarized as short SOAs, while a delay period of 250 ms was categorized as long SOA (Goldstein et al., 2022). To compare the performance of the animals in time, the experimental phase of the animals was categorized in novice versus expert. For each animal, after the application of aforementioned exclusion criteria, the first four sessions were considered as novice sessions while the last six sessions were considered as expert sessions. It was made sure that at least 10 sessions separated the novice and expert phase. Group data are presented as mean ± SEM. Statistics were calculated using the software package R. In case of non-normal, paired data Wilcoxon signed rank tests and in case of normal, paired data the paired t test was used for statistical analysis. For comparisons between novice (4 sessions) versus expert (6 sessions), data were aggregated (over electrode sites or behavioral conditions) to meet criteria for paired data analysis. For comparison of more than two groups, a non-normality Kruskal–Wallis test was applied combined with Dunn post hoc test. Multiple testing for paired data was corrected with Benjamini–Hochberg correction. In all cases, differences were considered significant at p < 0.05. If not stated otherwise in Results, Wilcoxon signed rank tests were used for statistical testing.
Neural data analysis
Neural data were read and processed using custom Python scripts. Electrodes were assigned to cortical layers using δ-Source inverse current source density analysis (Pettersen et al., 2006) combined with histologic and local field potential (LFP) trace analysis. Typically, we had no probe contact inserted in layer I, so that recordings were primarily made from layer II/III, layer IV, layer V, and upper layer VI with 100 µm probe distance (8 probes per shank) and a total span of 700 µm recording depth. For the analysis of stimulus responses, data from the recording site contralateral to the stimulated C1 whisker were exclusively used.
LFP processing
Raw data were low-pass filtered <300 Hz (third-order Butterworth), then aligned to stimulus onset triggers to focus on signal window between 50 ms before and 200 ms after whisker stimulus presentation. Trigger-aligned signals were then downsampled from 32 to 2 kHz and normalized to 0 V at stimulus onset for further analysis. To eliminate electrophysiology data contaminated with motion artifacts, all trials containing responses exceeding ±1 mV in the prestimulus time window of –50 to 0 ms as well as in the poststimulus time window of 30-200 ms were excluded.
MUA processing
To extract the MUA, a window of 2200 ms, which included the complete whisker stimulus sequence, were selected from each trial, starting at 200 ms before the first standard stimulus onset. Raw data were first band-passed filtered between 300 and 6000 Hz using a third-order Butterworth filter along one dimension using cascaded second-order sections. Then the sample-by-sample RMS was computed as described by Stark and Abeles (2007). Briefly, for each element of the signal, the square, mean (low-pass filtering at 100 Hz with a first-order Butterworth filter along with one dimension), and square root were calculated. After extracting the MUA, the traces were aligned to the first standard stimulus onset. Only trials that were not excluded because of artifacts in the LFP processing, as described previously, were considered.
To ensure responsiveness to sensory stimulation, Glass's Δ was used as a measure of the response-to-baseline ratio. This was calculated as described previously (Musall et al., 2017) with slight modification, in which 70 ms before and after the first standard stimulus onset was considered. The formula to compute Glass's Δ is as follows:
Index and correlation analysis
For the calculation of SSA, data analysis was confined to responses elicited by the deviant C1 stimulus and the preceding standard C1 stimulus. SSA indices (SIs) were computed based on the averaged responses over all standard and deviants per session and along the laminar profile. For LFPs, the peak negative value of the deviant and its preceding standard response were used and for MUAs, the peak positive value of the deviant and its preceding standard response were used, when deviant and standard were stimulated at the C1 whisker. SIs were computed as: SI = (Deviant - Standard)/(Deviant + Standard), with positive SIs indicating enhanced evoked responses for the deviant compared with the standard, negative SIs indicating enhanced evoked responses for the standard compared with the deviant, and SIs close to zero indicating no difference between deviant and standard response. To ensure responsiveness to sensory stimulation, trials with an LFP amplitude of >−30 µV (= SIs > 1.0, SIs < −1.0) for the first standard and the deviant stimulus were considered as trials without evoked responses and excluded from further analysis. Adaptation indices (AIs) and Deviant-standard-only-indices (DSIs) were computed as: AI = (First Standard - Last Standard)/(First Standard + Last Standard); DSI = (Deviant - Last Standard-Only)/(Deviant + Last Standard-Only). Same exclusion criteria as for SIs were applied for AIs and DSIs.
Correlation analysis was performed with the Kendall's rank correlation in R for LFP responses of each standard stimulus (1-5) with the LFP response of the deviant stimulus. This analysis was performed for each layer and novice/expert separately. Correlation coefficient (Kendall's tau) differences between novice and expert standard-deviant correlations, including 95% CIs were calculated via bootstrap estimation based on methods from Wilcox (2016) and Rousselet et al. (2021) to confirm significance level.
Linear mixed model (LMM) analysis
For correlation analysis, trials were aligned between behavior and neurophysiology. Statistics were calculated in R using the lme4 package for mixed model analysis. To investigate the correlation between behavioral and neuronal responses, LMMs (Bates et al., 2015) were used. Residuals of a linear model were tested for normality by visualization methods using QQ plots and histograms and the Shapiro–Wilk test. In all cases, residuals provided evidence for normal distribution. Only for the MUA-related analysis and its effect on RT, the residuals followed a non-normal distribution. RT in that model was therefore log-transformed. The basic model was fitted as follows:
Model comparison was performed with ANOVA. p values > 0.05 indicated the more parsimonious model (basic model) to be better fitted. This was further confirmed by comparisons of AIC (Akaike, 1974) and BIC values (Schwarz, 1978) between the two to-be-compared models. In all instances, the basic model was the best fitted model.
Results
We designed a Go-NoGo paradigm where head-fixed, water-scheduled mice learned to lick in response to a spatial visual target stimulus after exposure to a sequence of task-irrelevant whisker stimuli. Concurrent to the paradigm, extracellular recordings were made throughout the layers of the left primary somatosensory cortex (Fig. 1A,B). Trials began with the task-irrelevant whisker stimulus presentation, consisting of standard and deviant stimuli, and subsequently ended with a visual target stimulus (Fig. 1C,D; for details, see Materials and Methods). Mice were trained to give a lick response with the presence of the visual target stimulus (Go trial) regardless of its spatial location, and withhold licking when no visual target stimulus (NoGo trial) was presented.
Illustration of cross-modal visual detection task in mice. A, Schematics of behavioral apparatus. The head-fixed mouse had access to a center waterspout, with eyes aligned to two LEDs and four whiskers attached to piezo stimulators. Two loudspeakers were placed above the mouse's head. The mouse was monitored with a camera. B, B1 whiskers were stimulated bilaterally at 4 Hz (standards, brown), C1 whisker was stimulated unilaterally (deviant, green). Whisker identities switch between sessions. Single whisker deflections consisted of a 120 Hz cosine waveform (8.3 ms duration) with an amplitude of 300 µm. Electrophysiological recordings were conducted at the left hemisphere barrel cortex. C, Bilateral standard whisker stimulation (blue) was followed either by the unilateral deviant whisker (red) or by a bilateral deflection of the standard whisker (blue). After whisker stimulation, a lateralized visual stimulus was presented. The deviant whisker and visual stimulus could appear on the same side (congruent) or on opposite sides (incongruent). Dashed lines indicate same outcomes for standard-only trials. Mice were rewarded when correctly licking in Go trials and not rewarded in NoGo trials. D, Schematic representation of the behavioral paradigm. Trials started with a no-lick period (purple), followed by whisker stimulation and a varying delay period (brown). Another no-lick period (purple) was followed by the visual target (yellow). Licks during the pretarget no-lick period aborted the trial and excluded it from analysis. During visual stimulation in Go trials, mice received a reward when licking (decision window, green). If the animals licked in NoGo trials, a timeout and auditory feedback (orange) was provided. Intertrial intervals (ITI) were jittered. E, Following a habituation phase, training for visual signal detection started (top). After electrode implantation, the experiment started (bottom). Here, the deviant stimulus was presented for the first time. The first sessions of the experiment were categorized as novice (orange) and the last as expert (purple) sessions. Recordings were made simultaneously.
Mice successfully learn and maintain task performance over full experimental window
The training consisted of different steps (Fig. 1E) starting with visual target detection in a Go-NoGo task design. All mice (n = 8) learned to detect the visual target stimulus with time (Fig. 2A). Target detection performance was quantified as sensitivity d′ and criterion C (for details, see Materials and Methods). Sensitivity d′ increased during visual stimulus training from initial (0.754 ± 0.09, mean ± SEM) to final (1.349 ± 0.07, mean ± SEM; t value (−5.58), df (39), p = 1.989 × 10−6, paired t test) and middle (0.773 ± 0.05, mean ± SEM) to final phase (t value (−7.17), df (39), p = 1.267 × 10−8, paired t test). Criterion C was reduced from initial (0.554 ± 0.11, mean ± SEM) to final (0.142 ± 0.09, mean ± SEM; p = 0.003) and initial to middle phase (0.068 ± 0.06, mean ± SEM; p = 4.933 × 10−6), demonstrating a learning curve. Next, the animals were familiarized to the whisker stimulation by presenting the bilateral standard stimuli at 4 Hz, while the deviant was explicitly excluded from all training and introduced for the first time in the experiment (Fig. 1E). During the training with additional standard-whisker stimulation, the animals' performance initially increased to a sensitivity d′ of 1.451 ± 0.09 (mean ± SEM) and criterion C of 0.213 ± 0.09 (mean ± SEM) and remained steady in the middle phase (sensitivity d′: 1.291 ± 0.09, mean ± SEM, t value (1.28), df (21), p = 0.212, paired t test; criterion C: 0.247 ± 0.09, mean ± SEM, p = 0.679). The decrease in sensitivity d′ from initial to final (1.160 ± 0.10, mean ± SEM); t value (2.83), df (21), p = 0.010, paired t test), but the unaffected criterion C (0.296 ± 0.08, mean ± SEM; p = 0.503) indicated the effects of potential overtraining but otherwise a constant performance (Fig. 2A). After training, all animals underwent electrode implantation surgery followed by at least 3 d of recovery before starting with the experiment. The experiment consisted of standard, deviant whisker and visual target presentation, and unilateral extracellular recordings of the left-hemisphere barrel cortex (Fig. 1E, bottom). To demonstrate how deviant stimuli affect perception and neuronal activity over time, our analysis was restricted to two time windows (Fig. 1E, bottom): The first window included sessions when the whisker deviant was initially introduced (Novice from here on), while the second window included sessions at the end of the experiment when the animal was well experienced in receiving deviant whisker stimuli (Expert from here on). Of 8 animals trained initially, 7 were included in behavior analysis and 1 animal was excluded as it did not participate in the experiment (Fig. 2A). These 7 animals maintained their overall performance throughout the experiment [Sensitivity d′ novice (0.973 ± 0.07, mean ± SEM) vs expert (0.811 ± 0.04, mean ± SEM); p = 0.05, criterion C novice (0.610 ± 0.06, mean ± SEM) vs expert (0.420 ± 0.05, mean ± SEM); p = 0.01]. The slight tendency of worsened performance from training to experiment (overall training mean: d′ = 1.07, C = 0.26, overall experiment mean: d′ = 0.89, C = 0.48) was potentially triggered by implantation surgery.
Overview of the cross-modal visual detection performance. A, Sensitivity d′ (light green) and criterion C (dark green) at different training and experiment phases. The performance at the different phases during the visual stimulus (left, 5 sessions each, n = 8 mice), the whisker stimulus training (middle, 3 sessions each, n = 8 mice), and the experiment (right, novice = 4 sessions, expert = 6 sessions, n = 7 mice) are shown in mean ± SEM values. Asterisks indicate statistical significance for comparisons to initial/novice phase (dark gray) and middle phase (light gray) for sensitivity d′, paired t test and criterion C: ***p < 0.001; **p < 0.01; * p < 0.05; Wilcoxon signed rank test. B, Histogram of first lick probability at corresponding RTs for hit trials (dark green) and false alarm trials (black) for novice (top) and expert (bottom) during the experiment. Trials with licks faster than 50 ms (dashed yellow line) were excluded from analysis. Median values for hit (dark green arrow) and false alarm (black arrow) trials are shown (n = 7). **p < 0.01 (Kruskal–Wallis with Dunn post hoc test). C, Aborted trials because of licks in the no-lick period as percentage of all trials (left) for novice and expert animals. Right, Percentage of aborted trials as share of all trials either with whisker deviant or when the deviant was replaced with a standard stimulus. In novice animals (orange), a slightly higher percentage of aborted trials were preceded by a deviant stimulus. This difference reduced over time in the expert (purple) animals. Black cross indicates mean value. *p < 0.05 (Wilcoxon signed rank test). D, Response rates for different behavior conditions. Hit rates (as percentage of all Go trials) are shown for congruent, incongruent, and standard-only conditions of novice and expert animals (n = 7). False alarm rates (as percentage of all No-Go trials) are shown for trials with deviant and standard-only whisker stimuli. The conditions congruent and incongruent were only possible when a visual target stimulus (Go trial) was presented, either on the same or the opposite side of the whisker deviant, respectively. In the absence of a visual target stimulus, a whisker deviant stimulus was still possible, but the conditions congruent and incongruent not. Statistically significant difference between novice and expert could only be observed for the standard-only condition (p = 0.03). Error bars indicate ± SEM values.
The cross-modal experiment was designed as Posner-Paradigm (Posner et al., 1980) to demonstrate whether deviant stimuli can capture attention. Whisker stimuli were task-irrelevant, but based on the spatial location of the deviant whisker stimulus with respect to the visual target location this paradigm allowed the comparison of performance in conditions where the deviant may direct attention toward the target location (congruent trials) or away from that location (incongruent trials). Congruent, incongruent, and NoGo trials were presented at equal probability (33% each). Standard-only (no whisker deviant) trials served as control (Fig. 1C) to isolate the behavioral effect of the deviant stimulus. The presentation sequence of the different conditions was randomized. Behavioral performance was measured in RT and sensitivity d′. To rule out a general change in RT from novice to expert and response biases for certain conditions, lick responses over time (Fig. 2B) and response rates (Fig. 2D) were measured. First lick responses within the first 50 ms after stimulus onset (Fig. 2B, dashed line) were classified as impulsive licks (Berditchevskaia et al., 2016), and thus excluded from all behavior and neuronal analysis. No significant differences in RT, demonstrated as first lick probabilities in time (Fig. 2B), was observed between novice and expert conditions (novice hit mean: 327.02 ms, expert hit mean: 321.81 ms). In novice animals only, hit trials yielded higher RTs than false alarm trials (novice hit median: 278 ms, novice false alarm median: 174 ms, p = 0.003, Kruskal–Wallis with Dunn post hoc test). These faster false alarm responses by the novice animal may result from the more pronounced deviant impact on the animal's decision-making (Fig. 2C). All lick responses within the 150 ms no lick period (Fig. 1D) were aborted and not considered for further analysis. On average, 3.3% of the total trials were aborted in the novice phase and 4.5% were aborted in the expert phase (p = 0.09, Fig. 2C) for the cohort in which the deviant whisker was replaced with a standard stimulus. Out of those aborted trials and only in the novice animals, on average 58% trial-aborting licks occurred after deviant presentation while 49% of the aborted trials were led by standards in the standard-only condition (p = 0.03). For the cohort, in which the whisker deviant stimulus was omitted, 1.4% of the total trials in the novice and 1.0% in the expert phase were excluded because of licks in the no-lick period. Furthermore, response rates for congruent, incongruent, and standard-only conditions did not differ significantly in their hit and false alarm rates, showing no bias toward a specific behavioral condition in the novice or in the expert phase (Fig. 2D). Differences between novice and expert could only be observed for the standard-only condition with slightly higher false alarm rates in the expert animals (p = 0.03).
Whisker deviants influence visual detection performance
The effect of a whisker deviant on target detection performance was measured as sensitivity d′ and RT for the novice and expert animals and normalized to the standard-only condition (Fig. 3). Here, RT was restricted to hit trials to exclude confounds with deviant-related false positive rates. Furthermore, to avoid confounds with impulsive and whisker-triggered licks by the animal, two early-lick threshold (ELT) were set: 50 ms (Berditchevskaia et al., 2016) and 150 ms (Goldstein et al., 2022), excluding trials with a faster lick responses. In novice animals, spatially congruent deviants decreased RT compared with the standard-only condition regardless of the ELT (novice congruent median RT, ELT 50 ms: 0.930, p = 0.042, Fig. 3A, left; ELT 150 ms: 0.931, p = 0.045, Fig. 3A, right), indicating a deviant-specific effect. At the same time, spatially congruent deviants yielded faster RTs compared with incongruent deviants (novice incongruent median RT, ELT 50 ms: 1.00, p = 0.025, Fig. 3A, left; ELT 150 ms: 1.004, p = 0.005, Fig. 3A, right). RTs for incongruent deviants were statistically not different from standard-only conditions (ELT 50 ms: p = 0.614, ELT 150 ms: p = 0.218). In expert animals, congruent deviants did not affect RT (standard-only vs expert congruent median RT, ELT 50 ms: 1.008, p = 0.785; ELT 150 ms: 1.037, p = 0.379; expert congruent vs incongruent median RT, ELT 50 ms: 0.951, p = 0.372; ELT 150 ms: 0.980, p = 0.209). Additionally, normalized RT for congruent deviants increased from novice to expert (ELT 50 ms: p = 8.5 × 10−4, ELT 150 ms: p = 0.030), which speaks in favor of an attenuating effect of the deviant on the speed of decision-making with increasing experience. Furthermore, we measured the sensitivity of the animals to the presence of the target signal preceded by congruent or incongruent deviants with respect to the general presence of a deviant in the absence of the target. Deviant stimuli affected sensitivity d′ regardless of the ELT (Fig. 3B) in the novice animals. In contrast to the accelerating effect of the congruent deviant on RT in novice animals, congruent deviants yielded worsened perceptual sensitivity compared with standard-only conditions (novice congruent median d′, ELT 50 ms: 0.711, p = 0.002, Fig. 3B, left; ELT 150 ms: 0.625, p = 0.019, Fig. 3B, right). A tendency of performance decline was also shown for the incongruent deviant (novice incongruent median d′, ELT 50 ms: 0.832, p = 0.080; ELT 150 ms: 0.744, p = 0.188). Although sensitivity generally declined in expert animals (Fig. 2A), in deviant trials, the sensitivity significantly increased (median d′ for congruent novice vs congruent expert ELT 50 ms: 1.10, p = 0.01; ELT 150 ms: 1.07, p = 0.017; incongruent novice vs incongruent expert ELT 50 ms: 1.273, p = 6.1 × 10−4; ELT 150 ms: 1.175, p = 0.035). Interestingly, trials including a deviant compared with trials without a deviant and regardless of the ELT resulted in an improved perceptual sensitivity (incongruent expert vs standard-only: ELT 50 ms, p = 0.006; ELT 150 ms, p = 0.036), which speaks in favor of an attenuating effect of the deviant to negatively influence decision-making (Fig. 3B). Furthermore, the results in RT and sensitivity measured for different ELTs revealed the robustness of the effects so that further analysis was performed with a 50 ms ELT.
Effects of congruent and incongruent deviant whisker stimuli on visual target detection. A, Deviant effect on RT with different early-lick thresholds (left, ELT = 50 ms; right, ELT = 150 ms) in novice and expert mice. Only hit trials are considered. Measures for congruent and incongruent deviant trials are normalized to standard-only trials (dashed line). Each point indicates individual session of the different animals (n = 7). Each animal is represented with one color. Gray points indicate averaged performance per animal. Gray lines indicate the population trend from novice to expert. ***p < 0.001; **p < 0.01; *p < 0.05; Wilcoxon signed rank test. B, Deviant effect on sensitivity d′ with different ELTs (left, 50 ms; right, 150 ms) in novice and expert mice. Measures for congruent and incongruent deviant trials are normalized to standard-only trials (dashed line). Same conventions as in A apply here. ***p < 0.001; **p < 0.01; *p < 0.05; Wilcoxon signed rank test. C, Deviant effect on RT (left) and sensitivity d′ (right) with short and long SOAs in novice and expert mice. Measures for congruent and incongruent deviant trials are normalized to standard-only trials (dashed line). Same conventions as in A apply here. ***p < 0.001; **p < 0.01; *p < 0.05; Wilcoxon signed rank test.
Different SOA of 50, 150, and 250 ms between last the whisker stimulus and visual target onset were used to avoid anticipatory responses. We investigated behavioral differences given by short (50-150 ms) and long (250 ms) SOAs (Goldstein et al., 2022). A congruency effect in RT (i.e., difference between congruent and incongruent condition) was revealed for the novice animals only in the short SOAs (Fig. 3C, left). Here, spatially congruent deviants decreased RT compared with spatially incongruent deviants (novice congruent median RT, short SOA: 0.886 vs novice incongruent median RT, short SOA: 0.976, p = 0.018). The previously described deviant-specific effect (difference between deviant and standard-only condition) was confirmed with short SOAs in novice animals. Congruent deviants yielded faster RTs compared with the standard-only condition (p = 0.047). In expert animals, these deviant-triggered behavioral effects diminished under short SOAs (Fig. 3C, left), which was further evident by the comparison between novice and expert congruent deviants (novice vs expert congruent median RT, short SOA: 0.986, p = 0.048). Although no statistically significant impact of the deviant could be observed in long SOAs, the direction of potential effects was similar to that within short SOAs (novice congruent median RT, long SOA: 0.829, expert congruent median RT: 0.943; novice incongruent median RT, long SOA: 0.983, expert incongruent median RT: 1.034), indicating a robust distracting effect by the deviant in novice animals. The same analysis for different SOAs was performed to demonstrate deviant effects in perceptual sensitivity (Fig. 3C, right). Neither in short nor in long SOAs, statistically significant effects could be observed. A trend toward improved sensitivity in deviant conditions from novice to expert animals was, however, noticeable (short SOA: novice congruent median d′: 0.709, expert congruent median d′: 1.007; novice incongruent median d′: 0.851, expert incongruent median d′: 0.919; long SOA: novice congruent median d′: 0.841, expert congruent median d′: 1.084; novice incongruent median d′: 0.802, expert incongruent median d′: 1.191.), suggesting a facilitating deviant effect on perceptual sensitivity in expert animals.
Continuous extracellular recordings from C1 barrel
Simultaneous electrophysiological recordings were conducted as the animals were performing the behavioral task to investigate the role of deviant stimuli on perception. A multishank 32-channel silicon probe was implanted chronically into the left barrel cortex to cover the laminar profile (Fig. 4A). Histological and current source density analysis (Fig. 4B) was used for mapping the position of the electrode to the different layers. For analysis, the layers were then categorized into supragranular (SG, layer I-III), granular (G, layer IV), and infragranular (IG, layer V and upper layer VI) layers. Whisker assignment of deviant and standard was flipped between sessions (e.g., first session: C1 = standard and B1 = deviant; second session: C1 = deviant and B1 = standard; etc.). Exclusively, C1 barrel responses to contralateral C1 whisker deflections were used for further analysis, both for deviant and standard stimuli. The C1 barrel was successfully targeted with one electrode shank in 6 of 7 mice, excluding one mouse from further analysis. Over the whole recording period (30.7 ± 2.6 sessions), the amplitude of the evoked LFPs did not change significantly (p > 0.05, Fig. 4C), indicating stable recording conditions.
SSA across layers of the barrel cortex. A, The C1 barrel (green) in the left barrel cortex was targeted for electrode implantation with electrodes spanning the laminar profile. B, Histologic reconstruction of recording sites in 1 animal. The overview of the brain (right) shows the barrel cortex (bcx) and neighboring areas. Left, Inset, The barrels in layer IV, the electrolytic lesions in layer II/III and V/VI and superimposed the recording electrodes (white dots, red electrodes were used for lesions). C, Normalized mean peak negative LFP amplitude for C1 deviant and C1 first standard whisker stimulus at different experimental phases. Responses during the experimental phases middle and final were normalized to the initial phase for deviant and standard separately (each phase is the average of 5 sessions, n = 6 mice) across supragranular (SG), granular (G), and infragranular (IG) layer and show no statistically significant difference. Error bars indicate ± SEM. D, E, Grand mean LFP traces for responses to C1 deviant (red) and the preceding C1 standard (blue) (D), to C1 first (cyan) and last standard (blue) stimulus (E) and their difference traces (black) are shown for novice (left) and expert animals (right). Responses were normalized within each panel and for each animal to the deviant response. Shades represent ± SEM; n = 6. Gray bars represent statistically significant differences between the two stimuli type traces (Wilcoxon signed rank test combined with Benjamini–Hochberg correction p < 0.05).
Enhanced evoked responses for deviant compared with standard stimuli
The neuronal deviant response was characterized as amplitude of the evoked LFP for the deviant C1 and compared with the preceding standard C1 stimulus response (Fig. 4D). Across all cortical layers, deviant stimuli evoked larger LFP responses compared with repetitive standard stimuli (Fig. 4D, left) in novice animals, demonstrating the presence of a deviant effect and SSA (peak negative difference (black) trace ± SEM: SG: −0.438 ± 0.156, G: −0.316 ± 0.077, IG: −0.369 ± 0.088). This difference between deviant C1 and standard C1 responses declined with experience. In expert animals (Fig. 4D, right) a reduced difference between deviant and standard evoked responses was observed (peak negative difference (black) trace ± SEM: SG: −0.161 ± 0.047, G: −0.149 ± 0.054, IG: −0.248 ± 0.065). Starting roughly at 120 ms after stimulus onset, a second difference in deviant-standard LFP responses was observed across cortical layers for expert animals. This deviant-specific late response has been previously linked to a “True-Deviance Detection” signal and was suggested as a physiological substrate of MMN (Musall et al., 2017), resulting from thalamocortical feedback inputs. Whether enhanced deviant responses are because of adaptation to the standard stimuli was assessed by comparing evoked responses for the first standard C1 and the last standard C1 stimulus within a trial (Fig. 4E). In novice animals, a difference between first and last standard stimuli was observed across cortical layers (peak negative difference (black) trace ± SEM: SG: −0.392 ± 0.054, G: −0.320 ± 0.028, IG: −0.454 ± 0.069), indicating adaptation to the repetitive standards that per definition forms the basis of SSA. This difference was reduced in expert animals (peak negative difference (black) trace ± SEM: SG: −0.231 ± 0.049, G: −0.221 ± 0.032, IG: −0.292 ± 0.061). Further, a second negative peak difference was observed predominantly in the infragranular layer at ∼25 ms after stimulus onset in the novice animals, which sustained in the expert animals (second peak negative difference (black) trace ± SEM: novice IG: −0.300 ± 0.107, expert IG: −0.291 ± 0.084; p < 0.05, Wilcoxon signed rank test combined with Benjamini–Hochberg correction).
Magnitude of SSA reduces with increasing experience
Differences in evoked responses were quantified as SI and AI (for details, see Materials and Methods). Here, more positive indices indicated enhanced evoked responses for the deviant stimulus compared with the preceding standard stimulus (SI) or enhanced evoked responses for the first standard compared with the last standard stimulus (AI). SIs decreased from novice to experts (Fig. 5A) across cortical layers (mean index for: SG novice: 0.252, SG expert: 0.081, p = 0.022; G novice: 0.247, G expert: 0.061, p = 0.002; IG novice: 0.225, IG expert: 0.093, p = 0.016). SIs in the expert animals were reduced but did not diminish completely as they were still significantly different from SI = 0 (Fig. 4D). AIs also decreased (Fig. 5B) from novice to experts (mean index for: SG novice: 0.298, SG expert: 0.173, p = 0.004; G novice: 0.219, G expert: 0.150, p = 0.090; IG novice: 0.272, IG expert: 0.193, p = 0.004), confirming a reduced deviant response and adaptation to the standard. In order to rule out that the last whisker stimulus in the sequence was treated differently from the preceding sequence (e.g., that top-down stimulus anticipation was giving rise to the observed SSA), we compared the C1 responses when it was either stimulated as a deviant or as the last stimulus in the standard-only trials (Fig. 5C). DSIs indicated enhanced evoked responses for the deviant stimulus compared with the standard stimulus in standard-only trials which decreased with experience in supragranular and granular cortical layers (mean index for: SG novice: 0.370, SG expert: 0.207, p = 0.003; G novice: 0.296, G expert: 0.184, p = 0.042), confirming a deviant-specific effect. Interestingly, only a tendency of decreasing DSI from novice to expert could be observed in infragranular layers (DSI mean novice: 0.221, DSI mean expert: 0.162; IG: p = 0.111), suggesting layer-specific involvement in the preservation of deviance signal detection.
Experience-related reduction of SSA across cortical layers. A-C, Comparison of SSA (A), AI (B), and DSI (C) for the individual recording sites from 6 animals show reduced indices from novice to expert across supragranular (SG), granular (G) and infragranular (IG) layers. Black cross indicates mean. **p < 0.01; *p < 0.05; Wilcoxon signed rank test. D, Percentage change in evoked responses over time (novice vs expert) for standard stimuli (1-5) across cortical layers. Asterisk indicates a significant percentage change in response amplitude (n = 6 mice). Black cross indicates mean. ***p < 0.001; **p < 0.01; *p < 0.05; Wilcoxon signed rank test. E, Difference of the correlation coefficient (deviant with each standard stimulus) between novice and expert estimated with bootstrap across cortical layers (SG, G, IG). Error bars indicate 95% CIs.
The experience-related reduction of SSA may be explained by a habituation effect, which has been observed in decreasing MMN signals and deviance-distractor effects on behavior in humans (Sams et al., 1984; Debener et al., 2002; Rosburg et al., 2018; Littlefair et al., 2022) and guinea pigs (McGee et al., 2001). SSA as a precursor of MMN has the potential to exhibit long-term habituation as well. To confirm whether a generalized cortical adaptation for whisker stimulation, resulting in decreased whisker stimulus responses, explains the reduction of SSA as a consequence of the habituation effect, the percentage change of all five standard stimuli over time was measured. Here, more positive values indicated an increase in LFP amplitude (e.g., from novice: −200 µV to expert: −400 µV) and more negative values indicated a decrease in LFP amplitude (e.g., from novice: −200 µV to expert: −100 µV). Across cortical layers and standard stimulus types, an increase in mean LFP amplitude from novice to expert was observed (Fig. 5D), contradicting a generalized cortical adaptation and speaking in favor of an enhanced processing of the sequential standard stimuli. In supragranular layer, evoked response amplitude for standard 3 significantly increased by mean 74% (p = 0.0002) from novice to expert. Similarly, an increase in evoked response amplitude for standard 4 by mean 49% (p = 0.007) and for standard 5 by mean 76% (p = 0.0006) was measured. In infragranular layer, a significant increase in response amplitude by mean 34% (p = 0.005) for standard 3, by mean 26% (p = 0.02) for standard 4 and by mean 45% (p = 0.002) for standard 5 was observed as well. In granular layer, standard 5 response amplitude increased by 44% (p = 0.05). Responses for the first and second standard remained constant over time across all cortical layers, suggesting enhanced processing of the sequence of standard stimuli as such. Against this finding that the evoked standard responses increase with experience, the concept of distractor learning through statistical regularities (given by a fixed context) (Turatto et al., 2018; Slagter and van Moorselaar, 2021) was tested. Theoretically, the repetitive presentation of a distractor within a fixed context (in this case five standard stimuli) can increase its predictability and reduce its distracting effect. This change in predictability was assessed as the change in the correlation between deviant and the individual standard stimuli for novice and expert animals. The difference of the correlation coefficients between expert and novice was then used as a measure for predictability (Fig. 5E). Correlations between deviant and standard 3 (correlation coefficient difference SG: 0.21, 95% CI [0.021, 0.466]; G: 0.23, 95% CI [0.051, 0.44]; IG: 0.21, 95% CI [0.016, 0.458]), deviant and standard 4 (SG: 0.32, 95% CI [0.074, 0.622]; G: 0.18, 95% CI [−0.008, 0.450]; IG: 0.13, 95% CI [−0.102, 0.382]) and finally deviant and standard 5 (G: 0.14, 95% CI [−0.065, 0.369]; IG: 0.16, 95% CI [−0.065, 0.413]) increased notably across cortical layers (Fig. 5E), suggesting increased predictability of the deviant based on the sequential context of five standard stimuli.
SSA is robustly exhibited in the MUA across cortical layers
The neuronal deviant response was further characterized as amplitude of the MUA for the deviant C1 and standard C1 stimulus response (Fig. 6A). Because of a decreased signal-to-noise ratio when extracting multiunits, measured by Glass's Δ (for details, see Materials and Methods), 1 animal and one novice session of a second animal were excluded from the MUA analysis. Comparison of deviant C1 and its preceding standard C1 stimulus demonstrated a robust SSA signal in the MUA (Fig. 6A, left) in novice animals (peak positive difference (black) trace ± SEM: SG: 0.381 ± 0.132, G: 0.371 ± 0.124, IG: 0.334 ± 0.114). This difference between deviant C1 and standard C1 responses declined with experience. In expert animals (Fig. 6A, right), a reduced difference between deviant and standard responses was observed (peak positive difference (black) trace ± SEM: SG: 0.194 ± 0.060, G: 0.166 ± 0.062, IG: 0.171 ± 0.068). A second long-latency sensory response starting 120 ms after stimulus onset in the expert animals, as shown in the LFPs, could be confirmed at the multiunit level predominantly in the infragranular layer (gray bars: p < 0.05, Wilcoxon signed rank test combined with Benjamini–Hochberg correction; SG: mean p = 0.02; G: mean p = 0.03; IG: mean p = 0.005). To further confirm SSA on the basis of standard adaptation, responses for the first standard C1 stimulus and the last standard C1 stimulus within a trial were compared (Fig. 6B). In novice animals, a difference between first and last standard stimuli was observed across cortical layers (Fig. 6B; peak positive difference (black) trace ± SEM: SG: 0.506 ± 0.082, G: 0.442 ± 0.084, IG: 0.505 ± 0.060). This difference was reduced in expert animals (peak positive difference (black) trace ± SEM: SG: 0.301 ± 0.057, G: 0.291 ± 0.059, IG: 0.333 ± 0.058). A sustained second peak across cortical layers, as observed in the LFPs (Fig. 4E), could not be confirmed in the MUA leaving the origin of the LFP-specific peak unresolved.
MUA exhibits reduced SSA over time and across cortical layers. A, B, Grand mean MUA traces for responses to C1 deviant (red) and the preceding C1 standard (blue) (A), to C1 first (cyan) and last standard (blue) stimulus (B) and their difference traces (black) are shown for novice (left) and expert animals (right). Responses were normalized within each panel and for each animal to the deviant/first standard response. Shades represent ± SEM (n = 5). Gray bars represent statistically significant differences between the two stimuli type traces. First and last standard are significantly different for the whole trace-window starting at ∼10 ms after stimulus onset for novice and expert across cortical layers (Wilcoxon signed rank test combined with Benjamini–Hochberg correction, p < 0.05). C, Comparison of SI (top) and AI (bottom) of MUA values for the individual recording sites from 5 animals show reduced indices from novice to expert across cortical layers. Black cross indicates mean. ***p < 0.001; **p < 0.01; *p < 0.05; Wilcoxon signed rank test.
Further, quantification of the observed differences in the evoked responses showed decreased SIs (Fig. 6C, top) from novice to expert across cortical layers (mean index for: SG novice: 0.299, SG expert: 0.080, p = 0.027; G novice: 0.288, G expert: 0.073, p = 0.014; IG novice: 0.225, IG expert: 0.069; p = 0.003). Decrease in AIs was also observed (Fig. 6C, bottom) across cortical layers (mean index for: SG novice: 0.350, SG expert: 0.141, p = 0.0005; G novice: 0.297, G expert: 0.147, p = 0.002; IG novice: 0.306, IG expert: 0.203; p = 0.006), confirming a robust reduced deviant response and adaptation to the standard as shown in the grand mean averages (Fig. 6A,B) and in the LFP (Fig. 4D,E).
Perceptual decisions and SSA correlate in the deeper cortical layers
The representation of deviant stimuli was significantly attenuated with increasing experience (novice vs expert) in the LFP and MUA responses, and this change was paralleled by significantly reduced negative influence of deviant stimuli on the target detection performance. Therefore, the correlation between these two outcomes was analyzed by using LMM (for details, see Materials and Methods).
We fitted a LMM (estimated using maximum likelihood) to predict RT (for hit trials only) and sensitivity d′ with experimental phase (novice vs expert) and SI on a session-by-session basis. The model included animals as random effect. For each cortical layer, a separate LMM was fitted. The RT model's total explanatory power was substantial across layers for SI(LFP) and SI(MUA) (conditional R2 = 0.39 ± 0.02, mean ± SEM). For the model including SI(MUA), RT was log-transformed to ensure that residuals of the linear model followed a normal distribution. In infragranular layers (Fig. 7A,B), the interaction effect of SI(LFP) on phase = expert was statistically significant and positive (95% CI[59.37, 585.64], df (56), t value (2.46), p = 0.017. Thus, the average RT increased by 218.86 ms per unit increase of SI(LFP) in expert animals. This positive slope was also observed in the MUA (Fig. 7C,D). The interaction effect of SI(MUA) on phase = expert was marginal and positive (95% CI[−0.20, 1.00], df (46), t value (1.35), p = 0.172. Thus, the average RT increased by a factor of 1.48 per unit increase of SI(MUA) in expert animals, demonstrating slow RTs with increasing SIs. Additionally, the effect of phase alone was statistically significant and negative (Fig. 7A,B; 95% CI[−173.27, −6.44], df (56), t value (−2.16), p = 0.035) decreasing RT by 89.85 ms in expert animals regardless of changes in the SI(LFP). In granular layers, the interaction effect of SI(LFP) on phase = expert was marginally significant and positive (95% CI[−75.99, 591.22], df (58), t value (1.55), p = 0.127) increasing the average RT by 221.30 ms per unit increase of SI(LFP) in expert animals. In the MUA, no such effect was observed in granular layers (Fig. 7C,D). In supragranular layers, no statistically significant effects on RT were observed in the LFP and MUA (Fig. 7C,D), suggesting a layer-specific effect of SSA on RT.
Model analysis of the effect of SSA on behavioral performance. A, C, E, G, LMM (fitted green line, gray shades represent ± SEM) shows SSA effect on RT (A,C) and on sensitivity d′ (E,G) for sessions of novice (orange) and expert (purple) mice (LFP, n = 6; MUA, n = 5) for granular and infragranular but not supragranular layers. Pink triangles represent mean SI values. B, D, F, H, Estimates of effects, including interactions, on RT (B,D) and sensitivity d′ (F,H) derived from LMM analysis after model fitting are shown across cortical layers (blue represents increased changes in behavior: green represents decreased changes in behavior). Error bars indicate 95% CIs. *p < 0.05; °p ≦ 0.1; t test based on Satterthwaite approximation.
Next, a LMM analysis was performed to estimate the effect of SSA on sensitivity d′. The d′ model's total explanatory power was substantial across layers (SI(LFP): conditional R2 = 0.38 ± 0.02, SI(MUA: conditional R2 = 0.28 ± 0.01, mean ± SEM). In infragranular layers (Fig. 7E,F), the interaction effect of SI(LFP) on phase = expert was significant and negative (95% CI[−1.86,−0.15], df (56), t value (−2.40), p = 0.021). The average sensitivity d′ value decreased by 0.77 per unit increase of SI(LFP) in expert animals. Similarly, in granular layers, the average sensitivity d′ value decreased by 0.43 per unit increase of SI(LFP) in expert animals, demonstrating higher sensitivity with lower SI(LFP) (95% CI[−1.89, 0.28], df (57), t value (−1.49), p = 0.140). No interaction effect of SI(MUA) and phase could be observed across cortical layers (Fig. 7G,H). However, across all cortical layers, the effect of phase alone was (marginally) significant and negative (IG: df (48), t value (−1.68), p = 0.09, G: df (49), t value (−1.70), p = 0.09, SG: df (49), t value (−2.35), p = 0.03) with decreasing average d′ by 0.22 ± 0.02 (mean ± SEM) at SI(MUA) = 0 in expert animals. No interaction effects between SI and phase nor SI effects alone in the supragranular layer were observed in the LFP and MUA (Fig. 7E–H).
These results demonstrate a correlation between SSA measured as SI and perceptual decision-making. SI interdependent to experimental phase correlated with the animal's performance exclusively in granular and/or infragranular cortical layers. In supragranular layers, no such effect was found, suggesting a layer-specific role of SSA on perception.
Discussion
In the present study, the changes in RT and sensitivity d′ in novice animals provide evidence for the distractive nature of deviating stimuli. Importantly, congruent deviants yielded faster RTs than incongruent deviants (Fig. 3A). Our finding is comparable to previous studies in both humans (Soto-Faraco et al., 2005; Feng et al., 2014, 2017) and mice (Wang and Krauzlis, 2018), which also reported faster RTs for congruent compared with incongruent trials. Wang and Krauzlis (2018) show RTs of 448.1 ± 12.8 ms for congruent (= validly cued trials, with a presentation probability of 37.5%) compared with RTs of 466.8 ± 15.8 ms, p = 0.045 for incongruent trials (= invalidly cued trials, with a presentation probability of 12.5%). Feng et al. (2014) reported in their cross-modal study, using auditory, uninformative cues and visual targets, that faster responses were given for validly cued targets (valid, 545 ± 65 ms; invalid, 553 ± 64 ms). Furthermore, we found deviant-specific distraction when comparing effects of the deviant to effects without a deviant on visual target detection. In the presence of deviant stimuli, sensitivity d′ was decreased in novice animals. This finding may result from a nonspecific arousal on the occurrence of a deviant leading to faster lick responses in false alarm compared with hit trials (Fig. 2B).
A recent study in humans investigated the role of deviance distraction as a result of spatial attentional capture (Weise et al., 2020) with a cross-modal paradigm including bilateral standard, lateralized deviant and lateralized target stimuli presentation. Their results on the effects of spatial attentional shifts caused by congruent versus incongruent deviants are comparable to our observations, demonstrating robust deviant-triggered attentional reallocation across species. When comparing deviant versus standard trials, Weise et al. (2020) showed longer RTs, defined as distraction, and increased hit rates, which they linked to general arousal effects. Differences in the experimental design may explain these different behavioral results. The integration of a classic oddball-paradigm interleaved into target presentation results in that the to-be-ignored stimulus (standard or deviant) is always followed by the target stimulus (Berti et al., 2004; Parmentier et al., 2011; Weise et al., 2020). This can delay RT for low-probable deviants compared with high-probable standards but might be construed as an unspecific warning stimulus improving cognitive functions of target signal detection (Parmentier and Hebrero, 2013). The latter may also explain our findings in the expert animals. We found improved target signal detection with the presence of the deviant in the expert animals potentially as a consequence of top-down processes of the deviant as an unspecific warning stimulus.
Previous studies have demonstrated that novel and task-irrelevant stimuli may capture attention as a result of exogenous selection disrupting top-down detection for less salient but task-relevant stimuli (for review, see Corbetta and Shulman, 2002). However, spatial shifting of exogenous attention is usually accompanied by “inhibition of return” in human (Klein, 2000) and in some animal behavior (Lev-Ari et al., 2020). Our findings show that deviating stimuli are effective for a short time window of up to 150 ms SOAs during which spatially congruent deviants trigger faster RTs. This finding is in agreement with a recent study in mice (Goldstein et al., 2022), where spatially congruent stimuli reduced RT and sensitivity d′ only at SOA 150 ms. Similar to our observations, Goldstein et al. (2022) did not observe any congruency effects nor evidence for inhibition of return on SOAs ≥ 250 ms. However, at longer SOAs (>150) reported reduced RTs and increased d′, which indicates a general facilitating effect. Our results support a trend toward deviant-triggered distraction which reduced with increasing task experience at shorter SOAs (<250ms) only.
In addition to the behavioral findings, we recorded electrophysiological responses to whisker stimuli and found robust SSA of neurons across cortical layers of S1, which are comparable to a large body of literature demonstrating SSA in the auditory (Ulanovsky et al., 2004; von der Behrens et al., 2009; Taaseh et al., 2011), visual (Reches et al., 2010; Vinken et al., 2017), and somatosensory (Katz et al., 2006; Musall et al., 2017) cortices. Similar to our behavioral findings, we found deviant-specific effects with stronger evoked potentials for the deviant compared with standard stimuli in standard-only trials (Fig. 5C). Standard adaptation extended to forthcoming standard whisker stimulation, as in standard-only trials, that might have been sequentially organized and held in memory (Sussman, 2007) allowing neurons to remain highly sensitive for deviant stimuli, eliciting an attentional effect exclusively for the deviant.
A second deviant-specific sensory response was observed ∼120 ms after stimulus presentation predominantly in the infragranular layer (MUA) and interestingly only in the expert animals (Fig. 6A). Sachidhanandam et al. (2013) similarly observed late sensory responses only in trained compared with naive mice in layer II/III barrel cortex resulting from learned goal-directed motor output (i.e., lick response). They further state the absence of such a late response in miss compared with hit trials, which we could not confirm, most likely because our behavior paradigm required the animals to respond to the visual target rather than the whisker stimulus. The long delay, deep layer- and deviance-specific sensory response in our study may be driven by thalamocortical feedback inputs (Musall et al., 2017; Han et al., 2021, 2023). Rather than a physiological substrate for the generation of MMN (Garrido et al., 2009), we conclude that these signals reflect top-down attentional control for distractive deviants and therefore appear only in expert animals. However, further studies are needed to elucidate the exact origin of late sensory responses and their importance in sensory perception in behaving animals.
In humans, it has been shown that deviant EEG signals correlate with behavioral consequences, for example, measured as change in RT (Berti and Schröger, 2001) and decreased task performance (Berti et al., 2004). To our knowledge, we are the first to report a correlation between the behavioral deviance distraction and SSA on the level of multiunit and LFP activity. Enhanced neuronal responses in favor of the deviant, as a result of SSA, gave rise to spatial attentional shifts toward and subsequently from the deviant, consequently distracting from the task-relevant performance (i.e., visual target detection). Additionally, our results suggest a trend toward layer-specific involvement of SSA on perception. The correlation between SSA and behavioral performance could only be observed in the deeper cortical layers in the LFP and MUA (Fig. 7A–H). Previous studies have shown that layer V and VI neurons of primary sensory cortices affect perception and sensory stimulus detection in mice (Takahashi et al., 2016, 2020; Voigts et al., 2020). Takahashi et al. (2020) demonstrated that whisker detection is correlated with activity in the apical dendrites of layer V pyramidal tract neurons in S1, that predominantly project to deep brain structures, such as striatum, thalamus, superior colliculus, and brainstem. Chemogenetically silencing layer V pyramidal tract neurons in S1 significantly impaired whisker detection performance in the animals, which the authors could particularly link to the descending projections from layer V pyramidal tract neurons to the posteromedial complex of thalamus and superior colliculus.
Indeed, the corticofugal projections from layer V and VI of the primary sensory cortices to subcortical regions (Kim et al., 2015; Hoerder-Suabedissen et al., 2018; Prasad et al., 2020) support the crucial involvement of the infragranular layers in behavior (for review, see Sherman and Usrey, 2021). That these subcortical structures are substantially involved in visual selective attention has been demonstrated by many studies, including the superior colliculus (Lovejoy and Krauzlis, 2010; Krauzlis et al., 2013), striatum (Wang and Krauzlis, 2016, 2020), and thalamus (Halassa and Kastner, 2017).
Overall, the change of SSA was surprisingly similar in the LFPs and MUA as well as their correlation with the change of the behavioral performance, although statistically not significant presumably because of the comparatively small sample size. Only in the expert animals, a qualitative difference between LFPs and MUAs was observed in that the LFP SI correlated negatively with the sensitivity while the MUA SI correlated positively (Fig. 7E,G), which may reflect the more localized nature of the MUA signal compared with the LFP that integrates over a larger volume (Stark and Abeles, 2007; Dezfouli et al., 2018).
We also report a robust decreasing effect of SSA and the distracting deviant on perception with increasing experience. Our findings suggest that the difference in neuronal responses for deviant versus standard stimuli significantly reduced in magnitude from novice to expert animals because of the decreased standard whisker adaptation, which in turn did not allow an enhanced response to the deviant whisker stimulus according to the definition of SSA. Recent studies found that top-down control of distractors is driven by forming expectations of its spatial or temporal appearance, as a result of statistical learning of the distractor stimulus within a fixed context (Chelazzi et al., 2019; Li and Theeuwes, 2020; van Moorselaar et al., 2020; Slagter and van Moorselaar, 2021; van Moorselaar and Theeuwes, 2021). The same rationale can be applied to our paradigm design, as the distractor deviant always follows a fixed sequence of five standard stimuli. An expectation of the temporal appearance of the distractor may be formed so that deviant stimuli become an unspecific warning stimulus to prepare for the upcoming visual target task in the more experienced animals, improving signal target detection (Fig. 3B). We observed increased processing of the sequence of standard stimuli, shown in individually increased standard stimulus responses (Fig. 5D), combined with an increased correlation between the individual standards and the deviant stimulus (Fig. 5E) in the expert animals. This increased correlation may explain an enhanced predictability of deviant stimuli following the fixed five-standards-context as a result of distractor learning. Previous studies have also shown that deviants, embedded in a familiar context and thus becoming more predictable, do not exhibit a distractive impact on behavior (Sussman et al., 2003; Horváth et al., 2011; Parmentier and Hebrero, 2013). More targeted experiments are needed to clarify whether distractor learning is elicited by statistical regularities and thus involved in the modulation of SSA.
Additionally, the suprathreshold visual target stimulus allowed distraction by, but also top-down control of, the irrelevant peripheral stimuli. Perceptual load theory suggests that interference of irrelevant distractor stimuli is prevented from perception under situations when there is insufficient capacity for their processing (for review, see Lavie, 2005). In other words, the more difficult and attention-demanding the task is (i.e., by providing near-threshold visual stimuli), the less likely a deviant could capture attention and cause a distraction effect (Yucel et al., 2007).
In conclusion, our findings provide insights into the neurophysiological representation of distracting deviant stimuli, their impact on perception, and their potential relationship to spatial exogenous attentional processes at the level of cortical layers and MUA.
Footnotes
This work was supported by Swiss National Science Foundation Grant 310030_172962. We thank Reto Zihlmann for reviewing the statistical model analysis; and Reza Mazloum for technical advice and proofreading the manuscript.
The authors declare no competing financial interests.
- Correspondence should be addressed to Newsha Ghasemi Nejad at gnewsha{at}ethz.ch or Wolfger von der Behrens at wolfger{at}ini.uzh.ch