Recurrent Inhibition or Predictive Coding: Competing Models of Sensory Gating & P50 Suppression

Recurrent Inhibition or Predictive Coding: Competing Models of Sensory Gating & P50 Suppression | Authorea try { document.documentElement.classList.add('js'); } catch (e) { } var _gaq = _gaq || []; _gaq.push(['_setAccount', 'G-8VDV14Y67G']); _gaq.push(['_trackPageview']); (function() { var ga = document.createElement('script'); ga.type = 'text/javascript'; ga.async = true; ga.src = ('https:' == document.location.protocol ? 'https://ssl' : 'http://www') + '.google-analytics.com/ga.js'; var s = document.getElementsByTagName('script')[0]; s.parentNode.insertBefore(ga, s); })(); Skip to main content Preprints Collections Wiley Open Research IET Open Research Ecological Society of Japan All Collections About About Authorea FAQs Contact Us Quick Search anywhere Search for preprint articles, keywords, etc. Search Search ADVANCED SEARCH SCROLL This is a preprint and has not been peer reviewed. Data may be preliminary. 28 August 2025 V1 Latest version Share on Recurrent Inhibition or Predictive Coding: Competing Models of Sensory Gating & P50 Suppression Authors : Paul Kieffaber 0000-0002-6517-7734 [email protected] , Kathryn Gour , Samantha Kline , and Sabrina Ehmann-Jones Authors Info & Affiliations https://doi.org/10.22541/au.175636275.52453313/v1 206 views 115 downloads Contents Abstract Information & Authors Metrics & Citations View Options References Figures Tables Media Share Abstract One of the most widely studied indices of sensory gating is the P50 auditory evoked potential. The P50 event-related potential is typically measured using the paired-click paradigm, in which two identical auditory stimuli (S1 and S2) are presented in close succession (~500ms). Suppression of the P50 response to S2 has been proposed as a biomarker of inhibitory control in early sensory processing. However, despite its widespread use and potential clinical relevance, the underlying neural mechanisms responsible for P50 suppression remain a topic of debate. The longstanding recurrent inhibition account posits that the initial stimulus (S1) activates local inhibitory interneurons, resulting in reduced neural responsiveness to a second stimulus (S2). In contrast, the predictive coding framework offers a top-down explanation for P50 suppression, suggesting that the brain minimizes surprise or error signals by continuously generating predictions about the onset of S2. According to the predictive coding framework, when a stimulus is predictable, the prediction error is smaller and the corresponding neural response is attenuated. The present study offers a test of these competing theories regarding P50 suppression by manipulating S2 expectancy and removing the confound of temporal predictability. The results are consistent with the recurrent inhibition theory of P50 suppression, which did not vary with expectation of the occurrence of S2. This result is discussed with respect to these competing theories and implications for future research regarding P50 suppression deficits in clinical populations. Recurrent Inhibition or Predictive Coding: Competing Models of Sensory Gating & P50 Suppression Paul D. Kieffaber, Ph.D.*, William & Mary Kathryn A. Gour, B.S., William & Mary Samantha J. Kline, B.S., William & Mary Sabrina Ehmann-Jones, William & Mary * Corresponding Author: [email protected] KEYWORDS: P50, sensory gating, recurrent inhibition, predictive coding The authors listed above certify that they have NO affiliations with or involvement in any organization or entity with any financial interest (such as honoraria; educational grants; participation in speakers’ bureaus; membership, employment, consultancies, stock ownership, or other equity interest; and expert testimony or patent-licensing arrangements), or non-financial interest (such as personal or professional relationships, affiliations, knowledge or beliefs) in the subject matter or materials discussed in this manuscript. Abstract One of the most widely studied indices of sensory gating is the P50 auditory evoked potential. The P50 event-related potential is typically measured using the paired-click paradigm, in which two identical auditory stimuli (S1 and S2) are presented in close succession (~500ms). Suppression of the P50 response to S2 has been proposed as a biomarker of inhibitory control in early sensory processing. However, despite its widespread use and potential clinical relevance, the underlying neural mechanisms responsible for P50 suppression remain a topic of debate. The longstanding recurrent inhibition account posits that the initial stimulus (S1) activates local inhibitory interneurons, resulting in reduced neural responsiveness to a second stimulus (S2). In contrast, the predictive coding framework offers a top-down explanation for P50 suppression, suggesting that the brain minimizes surprise or error signals by continuously generating predictions about the onset of S2. According to the predictive coding framework, when a stimulus is predictable, the prediction error is smaller and the corresponding neural response is attenuated. The present study offers a test of these competing theories regarding P50 suppression by manipulating S2 expectancy and removing the confound of temporal predictability. The results are consistent with the recurrent inhibition theory of P50 suppression, which did not vary with expectation of the occurrence of S2. This result is discussed with respect to these competing theories and implications for future research regarding P50 suppression deficits in clinical populations. Recurrent Inhibition or Predictive Coding: Competing Models of Sensory Gating & P50 Suppression The human brain is perpetually confronted with a barrage of sensory information, much of which is repetitive or irrelevant. To prevent cognitive overload and facilitate efficient perception, it is believed that the nervous system engages in sensory gating, a process that filters out redundant or predictable inputs. One of the most widely studied indices of sensory gating is the P50 auditory evoked potential, a positive deflection in the EEG signal occurring approximately 50 milliseconds after the onset of an auditory stimulus (Davis et al., 1966). The P50 event-related potential (ERP) is typically measured using the paired-click paradigm, in which two identical auditory stimuli (S1 and S2) are presented in close succession (~500ms). Under normal conditions, the neural response to S2 is reduced compared to S1, a phenomenon known as P50 suppression (Boutros et al., 1995; Freedman et al., 1987; Nagamoto et al., 1989). Suppression of the P50 response has been extensively studied in both healthy and clinical populations. Reduced P50 suppression has been observed in disorders such as schizophrenia (Atagun et al., 2020; Freedman et al., 1987, 2020; Potter et al., 2006), post-traumatic stress disorder (Javanbakht et al., 2011; Neylan et al., 1999), and attention-deficit/hyperactivity disorder (Carvalho & Leite, 2023; Micoulaud-Franchi et al., 2015), indicating that impaired sensory gating may contribute to aberrant perceptual and cognitive processing in these populations. P50 suppression has even been proposed as a biomarker of inhibitory control in early sensory processing (Atagun et al., 2020; Boutros et al., 1995; Fuentes-Claramonte et al., 2024; Potter et al., 2006). However, despite its widespread use and clinical relevance, the underlying neural mechanisms responsible for P50 suppression remain a topic of debate. Two prominent theoretical accounts have been proposed to explain the neural mechanism underlying the P50 suppression effect: the recurrent inhibition hypothesis and the predictive coding hypothesis. The recurrent inhibition account posits that the initial stimulus (S1) activates, in addition to the principal sensory neurons, local inhibitory interneurons, which then provide feedback inhibition to the same or nearby principal neurons, resulting in reduced neural responsiveness to a second stimulus (S2) (Freedman et al., 1987; Nagamoto et al., 1989). In contrast, the predictive coding framework offers a top-down explanation for P50 suppression. According to this view, the brain uses prior information (i.e., Bayesian inference) to minimize surprise or error signals by continuously generating predictions about the nature of incoming sensory inputs (Huang & Rao, 2011; Rao & Ballard, 1999). According to the predictive coding framework, when a stimulus is predictable, the prediction error is smaller and the corresponding neural response is attenuated (Friston, 2005, 2010; Huang & Rao, 2011). Thus, in the context of the conventional paired-click paradigm, S2 is highly predictable following S1, and should therefore lead to a reduced neural response not because of inhibition, but due to its expected nature (Alink et al., 2010; Ouden et al., 2010). These competing accounts differ fundamentally in their assumptions about the sources and nature of P50 suppression. The recurrent inhibition model assumes a reflexive, stimulus-driven mechanism that is relatively inflexible to contextual modulation. In contrast, the predictive coding account suggests that sensory gating is dynamic and sensitive to higher-order expectations. Importantly, these models make diverging predictions about the roles of stimulus regularity and manipulations of expectation in shaping the P50 S2 response. For example, if suppression reflects inhibition, it should be relatively insensitive to whether an S2 is expected to occur. Conversely, if suppression reflects predictive coding, then modulating the predictability of S2 should have a significant influence on the magnitude of P50 suppression. This prediction is supported by research measuring other ERPs. For example, the observation that the auditory N1 component is reduced when elicited by speech sounds during vocalization compared with when those same recorded speech sounds are played back to participants has been attributed to the effects of predictive coding (Kort et al., 2017). Using visual stimuli, Baker et al. (2023) demonstrated that the amplitude of the N170 ERP is increased following the presentation of faces with unexpected emotional expressions. The mismatch negativity (MMN) and repetition suppression (aka “stimulus-specific adaptation”) effects have also been attributed to predictive coding mechanisms (Carbajal & Malmierca, 2018; Rentzsch et al., 2015). Importantly, research using a modified auditory oddball procedure in which rare (aka “deviant”) stimuli were either predictable (occurring at regular intervals) or unpredictable (occurring at random intervals) demonstrates that increased neural activity in response to unexpected stimuli is not related to mere stimulus probability (Dürschmid et al., 2016). Although both theories have garnered empirical support, relatively few studies have directly compared them using paradigms specifically designed to manipulate sensory expectations. As a result, it remains unclear whether P50 suppression is best understood as the outcome of local inhibitory dynamics or of top-down prediction-based processes. This gap in the literature presents an opportunity to clarify the cognitive and neural basis of early sensory gating. The present study was designed to address this question by testing whether P50 suppression is better explained by recurrent inhibition or by predictive coding mechanisms. We employed a modified paired-stimulus paradigm in which the predictability of S2 was systematically manipulated. By examining how P50 suppression varies as a function of S2 expectancy, we aimed to distinguish between these two theoretical accounts. If suppression is due to recurrent inhibition, it should remain consistent regardless of the predictability of S2. If, on the other hand, suppression reflects predictive coding, we expect reduced P50 responses to S2 only when its occurrence is predictable. Study 1 To optimize the signal-to-noise ratio while preserving the interpretability of the P50 component, we introduced a small degree of temporal jitter to the interstimulus interval (ISI) between S1 and S2 rather than using the conventional, fixed ISI of 500 ms (Mann et al., 2008; Oranje et al., 2006; Pinto et al., 2022). We varied the S1-S2 interval randomly between 490 and 510 ms across trials. This 0-50 ms range of variability falls within the average range of the auditory temporal just-noticeable difference (JND), which is 25-50 ms for a 500 ms reference duration (Quené, 2007) and is thus unlikely to be consciously perceived by participants. In addition to reducing the risk of stimulus phase-locking and rhythmic entrainment that can interfere with evoked potential averaging and confound comparisons of S1 and S2, this jitter serves a theoretical purpose in the context of our study. By introducing subtle variability in S2 timing, we reduce the precision of low-level temporal predictions while preserving the higher-level expectation that S2 will follow S1. This methodology creates a condition in which recurrent inhibition should remain intact, because it depends on local circuit dynamics and has been shown to tolerate variability in stimulus transmission delays without losing its functional effect (Edgley et al., 2021; Mackey & an der Heiden, 1984). In contrast, predictive coding mechanisms may show reduced suppression if expectation-driven attenuation of the neural response requires precise temporal prediction. However, it is important to note that the conventional articulation of predictive coding emphasizes the content (i.e., “what”) of the sensory input rather than its exact timing (Antonello & Huth, 2024; Huang & Rao, 2011; Rao & Ballard, 1999). The theory of predictive coding has been recently extended to include temporal prediction. Still, these newer models, sometimes referred to as “temporal predictive coding,” explicitly address the prediction of when events will happen, in addition to what those events are (Millidge et al., 2024). Thus, temporal variability in this context helps to attribute any observed effects of P50 suppression more specifically to the mechanisms of interest (predictive coding vs. recurrent inhibition), rather than to extraneous factors like neural entrainment or expectancy based on exact timing. Given that the effects of S1-S2 latency variability on P50 suppression have not been previously reported, the primary aim of Experiment 1 was to determine whether significant P50 suppression could be observed in a conventional paired-click paradigm with the introduction of 0-20 ms of temporal jitter in the ISI. Because paired-click stimuli were presented during the intertrial interval (ITI) of an unrelated Flanker task in the present study (see Materials and Procedure), a second aim of Experiment 1 was to demonstrate that the cognitive task in which the stimuli were embedded did not interfere with P50 suppression given that prior research has shown that increasing cognitive load concurrently with stimulus presentation can modify P50 suppression effects (Yee & White, 2001). Participants Data were collected from 24 young adult participants recruited through the research pool at a liberal arts university on the East Coast of the United States. Inclusion criteria for all participants consisted of an age of at least 18 years and normal or corrected-to-normal vision and hearing. Exclusion criteria for all participants included any history of neurological and/or psychiatric diagnosis. Three participants were excluded prior to data analysis due to the presence of excessive movement artifacts in the EEG recordings, leaving a total of 21 participants (54.5% Female). The mean age of participants was 19.1 years (SD = 0.990). 81.8% (N=18) of the participants were white, 4.5% (N=1) were Hispanic/Latino, 4.5% (N=1) were black/African American, and 9.1% (N=2) were Asian. Written informed consent was obtained from each participant in accordance with protocol # PHSC-2024-10-28-17305 approved by the Institutional Review Board at William & Mary. Materials and Procedure After consenting to the experimental procedures, participants were fitted with an EEG cap and seated in front of a computer monitor in an electrically shielded booth. Participants completed three blocks of 160 trials of a modified Flanker task (See Scrivano & Kieffaber, 2022) that was unrelated to the present study. Each trial of the task was separated by a 1000 ms intertrial interval (ITI) that followed corrective feedback regarding the participant’s response to the Flanker task on the previous trial, and the beginning of the next trial. The paired click stimuli were randomly presented in one of the three blocks. The first of the paired-click stimuli was presented at a random time between 0 and 200 ms from the start of the ITI, indicated by the appearance of a fixation cross at the center of the monitor for a total of 160 click pairs. Participants were informed that the click stimuli would be presented, but told that they were not relevant to the experiment. Auditory click stimuli consisted of digitally created, 1-ms duration square wave pulses presented binaurally at 75 dB through E-A-Rtone 3A insert earphones (3M Auditory Systems, Indianapolis, IN, USA) using foam-tipped tubing to ensure consistent acoustic delivery and ambient noise attenuation. Clicks within each pair were separated by a random inter-stimulus interval (ISI) between 475 ms and 525 ms (M = 500.23, SD = 2.54). Because click pairs were presented during the ITI of a Flanker task, the delay between the first click of each pair was, in part, determined by the response time of the participants on each trial. The average delay between click pairs was 3048 ms (SD = 85.4 ms). EEG Recordings EEG data were recorded from 32 sensors placed according to the standard 10-10 montage and referenced to channel Cz. The data were sampled continuously at 1000 Hz using a BrainProducts actiCHamp Plus amplifier (actiCHamp Plus, 2025). EEG data were pre-processed offline using EEGLAB v2024.0 (Delorme & Makeig, 2004) and ERPLABv10.1(Lopez-Calderon & Luck, 2014) in MATLAB 2024a (The Mathworks, Matick, MA). After removing DC offset, artifact-laden channels and segments (5 seconds) of the continuous data containing extreme artifacts were identified using the clean_rawdata plugin for EEGLAB. Channels were marked as “bad” if they (1) flatlined for more than five seconds, (2) contained line noise artifact higher than four standard deviations relative to the population of channels, and/or (3) their correlation with nearby channels was less than 0.75. Artifact Subspace Reconstruction (ASR) was applied to remove high-amplitude bursts with a conservative BurstCriterion of 30 standard deviations. These parameters were chosen to effectively mitigate transient artifacts while minimizing the loss of genuine EEG activity. On average, 11.4% of the continuous recordings were identified and corrected across participants. An IIR Butterworth band-pass filter of 1–75 Hz (12dB/oct) was applied only for the purpose of ocular artifact correction. The data, excluding bad channels and segments, were then submitted to Independent Components Analysis (ICA) (Jung et al., 2000). The ICA components were automatically classified using the ICLabel plugin for EEGlab (Pion-Tonachini et al., 2019), and components labeled as eye movements and muscle activity above 80% probability were automatically marked for rejection. Trained experimenters reviewed the results of the ICA decomposition and the selection of artifact components for accuracy and made the final judgment about which components to remove (M = 2.52 components, Range = 2-4). The retained independent components were then back-projected to the raw, unfiltered recordings. Bad channels identified prior to ICA decomposition were then reviewed by a trained experimenter and reinstated by spherical spline interpolation (M = 1.25 channels, Range = 0-5), and the data were re-referenced to the common average, reinstating Cz. The continuous, artifact-corrected recordings were segmented between -50 and 300 ms with respect to the onset of each click stimulus. Artifacts were identified in each channel/segment as voltages exceeding the thresholds of -100 and 100 μV. Only those channels within each segment that contained an artifact were then interpolated using a spherical spline. EEG epochs were then averaged over trials separately for the S1 and S2 clicks and baseline corrected over the interval -50 to 0 ms. P50 Measurement & Analysis P50 amplitudes and latencies were measured at the position of the local maximum with a width of at least three ms and the highest amplitude in the window between 45 and 85 ms in channel Fz. P50 suppression was assessed using two complementary approaches: (1) a dependent-samples t -test comparing P50 amplitudes elicited by S1 and S2, and (2) a one-sample t -test evaluating the observed P50 suppression ratios (S2/S1) against a theoretical value of 1, indicating no suppression. All 21 participants included in the analysis evidenced a local maximum spanning at least three ms within the P50 window. The assumption of normality was evaluated using the Shapiro–Wilk test. Results The results of experiment 1 are summarized in Figure 1. A dependent-samples t -test comparing P50 amplitudes at S1 and S2 revealed that amplitudes elicited by S2 were significantly smaller (M = 0.72 μV, SD = 0.69) than P50 amplitudes elicited by S1 (M = 1.04, SD = 0.52), t (20) = 4.26, p < 0.001. The effect size, as measured by Cohen’s d , was d = 0.93, indicating a large effect. A second dependent-samples t -test comparing P50 peak latencies at S1 and S2 revealed that S1 latency (M = 59.9, SD = 7.02) did not differ significantly from the peak latency of S2 (M = 59.3, SD = 5.07), t (20) = 0.69, p > 0.05. A one-sample t -test was used to test the mean S2/S1 ratio against a theoretical mean of 1. This analysis revealed that the mean S2/S1 ratio of 0.73 (SD = 0.29) was significantly below the comparison value of 1, t (20) = 11.4, p < 0.001. The effect size, as measured by Cohen’s d , was d = 2.43, indicating a substantial effect. Figure 1. Grand average S1 and S2 waveforms and topographies. Note : Grand average topographical distribution of peak P50 amplitude at the peak P50 latencies (top left). Grand average S1 and S2 waveforms at channel Fz (bottom left). Bar and violin plots illustrating the P50 amplitudes (green) and the S2/S2 P50 ratio (gold; bottom right). Discussion The results of Study 1 indicate that valid measures of P50 suppression can be elicited by presenting click stimuli during the ITI of a relatively simple cognitive task (e.g., Flanker task) and that measures of P50 suppression are robust to the addition of 0-20 ms of random variability in the interstimulus interval (ISI). While most P50 paradigms use a fixed ISI of 500 ms, this study employed a range of 490–510 ms to avoid exact temporal expectation. Although this 20 ms variation is below the average JND for temporal discrimination and is unlikely to be perceptible to participants, it serves two important methodological purposes. First, jittering the ISI reduces the likelihood that stimulus timing becomes phase-locked with ongoing background EEG oscillations, which can improve the signal-to-noise ratio of the averaged P50 waveform. Second, in the context of theoretical models such as predictive coding, jittered timing reduces the precision of temporal predictions about S2, allowing emphasis on the prediction that an S2 will occur rather than its precise timing. Although it is noteworthy that the size of the P50 suppression effect was quite large in this study, the average suppression ratio (S2/S1) was 0.73, which is at the upper end of the 0.09 - 0.74 range of reported P50 ratios for healthy control participants in meta analysis of 84 studies comparing healthy control participants and participants with schizophrenia (Patterson et al., 2008). Whether or not the P50 suppression ratio is increased (even if the ratio is still significantly less than one) by the introduction of slight temporal jitter in the ISI remains a promising direction for future research. Study 2 The primary aim of Study 2 was to determine whether recurrent inhibition or predictive coding would better account for observed P50 suppression effects during blockwise manipulation of the predictability of S2. To dissociate the contributions of recurrent inhibition and predictive coding to P50 suppression, we introduced two key manipulations to disrupt temporal predictability. First, we applied a wide range of temporal jitter to the interstimulus interval (ISI) between S1 and S2, varying randomly between 450 and 850 ms. This range exceeds the just noticeable difference (JND) for temporal intervals in the auditory modality, ensuring that participants could not reliably anticipate the timing of the second click. By eliminating fixed timing cues, this manipulation reduces the likelihood that P50 suppression is driven by low-level temporal entrainment or rhythmic prediction. Second, we manipulated the probability of S2 occurrence, such that in some conditions, a second click followed S1 on only a subset of trials. This probabilistic omission further disrupted sensory expectation and allowed us to assess how P50 suppression varies as a function of stimulus expectation. Together, these manipulations were designed to test whether P50 suppression persists in the absence of temporal expectation—supporting a recurrent inhibition model—or whether suppression is attenuated when the occurrence of S2 is uncertain, consistent with predictive coding accounts. Method Participants Data were collected from 31 young adult participants recruited using the same protocol as described in Experiment 1. Inclusion criteria for all participants consisted of an age of at least 18 years and normal or corrected-to-normal vision and hearing. Exclusion criteria for all participants included any history of neurological and/or psychiatric diagnosis. Two participants were excluded prior to data analysis due to the presence of excessive movement artifacts in the EEG recordings, leaving a total of 2911Note that demographic information was not available for one participant. That participant was included in the reported results because their exclusion did not change any of the results. participants (53.6% female). The mean age of participants was 18.8 years (SD = 0.863). 85.7% (N=24) of the participants were white, 3.6% (N=1) were black/African American, and 10.7% (N=3) were Asian. Written informed consent was obtained from each participant in accordance with protocol PHSC-2024-10-28-17305 approved by the Institutional Review Board at William & Mary. Materials and Procedure All Materials and procedures were identical to those described in Study 1, with two important exceptions. First, as in Study 1, click stimuli were presented during the ITI between trials of the Simon-Flanker task that were unrelated to the current study; however, in Study 2, the probability that the click stimuli would occur in pairs was manipulated across the three blocks of 150 trials. In one block, paired click stimuli were presented during 100% of the ITIs (identical to Study 1). In a second block, the paired click stimuli were presented during 80% (N = 120 trials) of the ITIs. During the remaining 20% (N = 30) of the ITIs, only one of the two click stimuli was presented. Finally, in a third block of trials, the paired-click stimuli were presented during only 20% (N = 30) of the ITIs, with a single click stimulus presented during the remaining 80% (N = 120) of the ITIs in that block. The order in which each block was presented was randomized across participants. The second exception was that, while the paired click ISI varied within the range of the JND in Study 1, the paired click ISI in Study 2 was between 450 and 850 ms (M = 575.47, SD = 54.90). Because click pairs were presented during the ITI of a Flanker task, the delay between click pairs was, in part, determined by the response time of the participants on each trial. The average delay between click pairs was 4,376 ms (SD = 158.5 ms) EEG Recordings EEG recording and preprocessing methods were identical to those described in Study 1. On average, 11.7% of the continuous recordings were identified and corrected using ASR. The average number of ICA components removed during ocular artifact correction was 2.72 (Range = 2-5). The average number of channels reinstated by spherical spline interpolation was 0.97 (Range = 0-5). P50 Measurement & Analysis ERP Analysis and measurement methods were identical to those described in Study 1. A local maximum spanning at least three ms could not be measured for one subject following S1in the 20% Pair condition. For this subject and condition, the P50 amplitude was measured using the absolute maxima in the 45-85 ms window because removing the subject from the analysis did not change the statistical or qualitative nature of the results. As in Study 1, P50 suppression was assessed using complementary approaches: (1) a 2 (Stimulus: S1, S2) X 3 (Pair Probability: 100%, 80%, 20%) repeated measures ANOVA (Greenhouse-Geisser corrected degrees of freedom used when appropriate) with peak P50 amplitudes as the dependent measures, (2) a repeated measures ANOVA comparing the S2/S1 ratio across the three levels of Pair Probability (100%, 80%, & 20%), and (3) three one-sample t -tests evaluating the observed P50 suppression ratios (S2/S1) at each level of Pair Probability against a theoretical value of 1, indicating no suppression. The assumption of normality in model residuals was evaluated using Shapiro–Wilk tests. Parametric tests were supplemented with equivalent non-parametric tests (when possible) in cases of potential violations of the normality assumption. Results The grand average waveforms elicited by S1 and S2 in each of the three trial blocks are illustrated in Figure 2. Notably, the S2/S1 ratio was numerically less than 1 in all three Pair Probability conditions. One-sample t -tests were used to test the hypothesis that the P50 suppression ratio in each block would be less than a theoretical value of one. Shapiro-Wilk tests indicated that S2/S1 P50 ratios violated the normality assumption in the 80% ( W = 0.44, p < 0.01), and 20% ( W = 0.73, p < 0.01) Pair Probability conditions. Thus, one-sample, non-parametric Wilcoxon signed-rank tests were used. These tests indicated significant P50 suppression in the 100% Pair Probability block (M = 0.64, SE = 0.06), W (28) = 36.0, p < 0.001, and in the 80% Pair Probability block (M = 0.99, SE = 0.19), W = 109.0, p < 0.01, and in the 20% Pair Probability block (M = 0.77, SD = 2.01) W = 75.0, p < 0.001. Descriptive statistics for measures of P50 amplitude and the S2/S1 ratios are illustrated in Figure 3. The Stimulus (S1/S2) X Pair Probability (100%/80%/20%) ANOVA comparing S1 and S2 amplitudes across blocks revealed a significant main effect of Stimulus, F (1, 28) = 40.84, p < 0.001, \(\eta_{p}^{2}\) = 0.59, with peak P50 amplitudes significantly suppressed following S2 (M = 0.85, SE = 0.09) compared to S1(M = 1.22, SE = 0.12). The main effect of Pair Probability was also statistically significant, F (2, 56) = 5.35, p < 0.01, \(\eta_{p}^{2}\) = 0.16, indicating a trending increase in P50 amplitudes from the 100% Pair Probability (M = 0.91, SE = 0.09) to the 80% Pair Probability (M = 1.01, SE = 0.12) to the 20% Pair Probability (M = 1.17, SE = 0.11) condition. Critically, however, the interaction between Stimulus and Pair Probability was not statistically significant, F (2, 56) = 2.39, p > 0.05, \(\eta_{p}^{2}\) = 0.07, indicating that P50 suppression was not modulated by expectancy. Note that there were no significant violations of sphericity, and a Shapiro-Wilk test of the full model residuals suggested that the assumption of normality was met, W = 0.98, p > 0.05. A repeated measures ANOVA comparing the S2/S1 P50 peak amplitude ratio across the levels of Pair Probability was not statistically significant, F (1.42, 39.84) = 2.00, p > 0.05,\(\eta_{p}^{2}\) = 0.06. In addition to the violation of sphericity, a Shapiro-Wilk test of the model residuals indicated the potential for a violation of the normality assumption, W = 0.98, p < 0.01. Thus, the parametric repeated measures AVNOA was supplemented by a non-parametric, Friedman’s ANOVA, which indicated significant differences in the P50 ratio across Pair Probability conditions, X 2 (2) = 8.90, p < 0.05. Specifically, Durbin-Conover pairwise tests indicated that the P50 ratio was higher in the 80% Pair Probability condition (M = 0.99, SE = 0.19) than both the 100% (M = 0.64, SE = 0.06), p < 0.01 , and the 20% (M = 0.77, SE = 0.10), p < 0.05, conditions, which were not significantly different from one another. P50 peak latencies were also evaluated using a Stimulus (S1/S2) X Pair Probability (100%/80%/20%) repeated measures ANOVA. Neither the main effects of Stimulus, F (1, 28) = 2.73, p > 0.05, \(\eta_{p}^{2}\) = 0.09, and Pair Probability, F (2, 56) = 0.39, p > 0.05, \(\eta_{p}^{2}\) = 0.01, nor the interaction between Stimulus and Pair Probability, F (2, 56) = 0.06, p > 0.05, \(\eta_{p}^{2}\) = 0.01, was statistically significant (See Figure 4). The Shapiro-Wilk test of the full model residuals indicated a potential violation of the normality assumption for this test, W = 0.97, p < 0.05. Figure 2. Grand average S1 and S2 topographies and ERP waveforms. Note: Grand average P50 ERP waveforms and peak amplitude topographical maps for the 100% (A), 80% (B), and 20% (C) Pair Probability conditions. Also shown are the S1-S2 difference wave and topographies for each of the three Pair Probability conditions (D). Figure 3. P50 amplitudes and S2/S1 ratios. Note: (A) Boxplots illustrating mean P50 S1 and S2 amplitudes across each of the three Pair Probability conditions. (B) P50 gating ratios across each of the three Pair Probability conditions. Figure 4. P50 S1 and S2 latencies across Pair Probability conditions. Note: Boxplots illustrating the average P50 S1 and S2 peak latencies across the 100%, 80%, and 20% Pair Probability conditions. General Discussion The present study aimed to determine whether P50 suppression is best explained by recurrent inhibition or predictive coding. In the first study, we demonstrated the feasibility of measuring P50 suppression with paired clicks embedded in the ITI of an unrelated flanker task. Study 1 also demonstrated that P50 suppression can be reliably measured with subtle variability (below the JND) of the ISI between paired clicks. In Study 2, we employed a modified paired-stimulus paradigm in which the predictability of the occurrence of a second auditory stimulus (S2) was systematically manipulated. By examining how P50 suppression varies as a function of S2 expectancy, we aimed to distinguish between these two theoretical accounts. Overall, the results supported the recurrent inhibition hypothesis, that P50 suppression is likely due to the activation of local inhibitory neurons by the initial stimulus (S1), which then provide feedback inhibition to the same or nearby principal neurons, resulting in reduced neural responsiveness to the second stimulus (S2) (Freedman et al., 1987; Nagamoto et al., 1989). Although the non-parametric test of the P50 ratios across Pair Probability conditions indicated significant differences, the observation that P50 suppression was reduced in the 80% but not the 20% Pair Probability conditions is in direct conflict with the predictions of predictive coding. Another important contribution of this research is that it demonstrates that robust P50 suppression to the second auditory stimulus (S2) can be observed even when the ISI between S1 and S2 varies unpredictably across trials. This finding challenges the conventional assumption that P50 gating depends on temporal predictability and supports research by Schwartze et al. (2013) suggesting that sensory gating mechanisms, at least in part, may operate on the formal properties (i.e., “what”) of a stimulus, independent of the precise timing expectations. While these findings support recurrent inhibition as the dominant mechanism underlying P50 suppression, particularly in the earliest stages of auditory sensory processing, this does not preclude the involvement of predictive coding in shaping neural responses to sensory input. The contribution of predictive coding may emerge more clearly in other ERP components or at later stages of processing. For example, ERP components such as the mismatch negativity (MMN) and the N170 have been extensively linked to predictive coding models, with robust evidence showing that these components are sensitive to violations of stimulus expectation and reflect the brain’s hierarchical inference processes (Baker et al., 2021, 2023; Carbajal & Malmierca, 2018). The contrast between the relative insensitivity of P50 suppression to manipulations of temporal and probabilistic predictability and the pronounced modulation of MMN and N170 by expectancy in prior research points to the possibility that sensory gating mechanisms are organized hierarchically within the brain. In this framework, rapid, stimulus-driven recurrent inhibition may serve as a foundational “first pass” filter for redundant or irrelevant inputs, while predictive coding processes operate at later stages, modulating neural activity based on top-down predictions and contextual information. Future research should consider how these mechanisms might interact dynamically, with inhibition setting the stage for prediction-based modulation of sensory processing. Such a hierarchical approach could account for both the robust automaticity of early recurrent inhibition and the flexible, context-sensitive nature of higher-order perceptual inference. That P50 suppression is robust to manipulations of temporal expectancy also has implications for clinical research, particularly research on populations with schizophrenia, in which impaired P50 suppression has been widely interpreted as evidence of disrupted early inhibitory processing (Freedman et al., 2020; Patterson et al., 2008). Variable ISI paradigms may offer a more sensitive approach to assessing gating deficits in at-risk or clinical populations, potentially revealing impairments not captured under traditional fixed-ISI designs. An avenue for future research may be to probe the flexibility and robustness of inhibitory processes in both fixed-ISI and variable-ISI conditions. While fixed-ISI designs inherently confound temporal expectancy with stimulus qualities, variable ISI designs can be used to determine whether sensory gating deficits in clinical populations reflect a fundamental deficit in automatic inhibitory processing related to the formal representation of a stimulus, or are due to disrupted timing mechanisms and/or an inability to form expectations. In summary, the present study provides strong evidence that P50 suppression in the auditory paired-click paradigm is best explained by recurrent inhibition rather than predictive coding. Across both experiments, P50 suppression remained robust even when temporal predictability and S2 occurrence probability were systematically disrupted, indicating that this neural mechanism operates automatically and is relatively insensitive to higher-order expectations. These findings refine our understanding of sensory gating by highlighting the role of local inhibitory processes in early auditory cortical responses, while suggesting that predictive coding may be more relevant for modulating later or integrative ERP components. Importantly, the variable-ISI paradigm may enhance future clinical assessments of inhibitory dysfunction, potentially offering more sensitive tools for investigating sensory gating deficits in psychiatric populations. Continued research is needed to map the hierarchical interplay between inhibition and prediction in sensory processing and to determine how these mechanisms contribute to both typical cognition and neuropsychiatric disorders. References Alink, A., Schwiedrzik, C. M., Kohler, A., Singer, W., & Muckli, L. (2010). Stimulus Predictability Reduces Responses in Primary Visual Cortex. Journal of Neuroscience , 30 (8), 2960–2966. https://doi.org/10.1523/JNEUROSCI.3730-10.2010 Antonello, R., & Huth, A. (2024). Predictive Coding or Just Feature Discovery? An Alternative Account of Why Language Models Fit Brain Data. Neurobiology of Language , 5 (1), 64–79. https://doi.org/10.1162/nol_a_00087 Atagun, M. I., Drukker, M., Hall, M. H., Altun, I. K., Tatli, S. Z., Guloksuz, S., van Os, J., & van Amelsvoort, T. (2020). Meta-analysis of auditory P50 sensory gating in schizophrenia and bipolar disorder. Psychiatry Research. Neuroimaging , 300 , 111078. https://doi.org/10.1016/j.pscychresns.2020.111078 Baker, K. S., Johnston, P., Yamamoto, N., & Pegna, A. J. (2023). Event-Related Potentials Index Prediction Error Signalling During Perceptual Processing of Emotional Facial Expressions. Brain Topography , 36 (3), 419–432. https://doi.org/10.1007/s10548-023-00951-2 Baker, K. S., Pegna, A. J., Yamamoto, N., & Johnston, P. (2021). Attention and prediction modulations in expected and unexpected visuospatial trajectories. PLOS ONE , 16 (10), e0242753. https://doi.org/10.1371/journal.pone.0242753 Boutros, N. N., Torello, M. W., Barker, B. A., Tueting, P. A., Wu, S.-C., & Nasrallah, H. A. (1995). The P50 evoked potential component and mismatch detection in normal volunteers: Implications for the study of sensory gating. Psychiatry Research , 57 (1), 83–88. https://doi.org/10.1016/0165-1781(95)02637-C Carbajal, G. V., & Malmierca, M. S. (2018). The Neuronal Basis of Predictive Coding Along the Auditory Pathway: From the Subcortical Roots to Cortical Deviance Detection. Trends in Hearing , 22 , 2331216518784822. https://doi.org/10.1177/2331216518784822 Carvalho, S., & Leite, J. (2023). Cognitive Event-Related Potentials and ADHD Across the Lifespan. In J. L. Matson (Ed.), Clinical Handbook of ADHD Assessment and Treatment Across the Lifespan (pp. 139–159). Springer International Publishing. https://doi.org/10.1007/978-3-031-41709-2_6 Davis, H., Mast, T., Yoshie, N., & Zerlin, S. (1966). The slow response of the human cortex to auditory stimuli: Recovery process. Electroencephalography and Clinical Neurophysiology , 21 (2), 105–113. https://doi.org/10.1016/0013-4694(66)90118-0 Delorme, A., & Makeig, S. (2004). EEGLAB: An open source toolbox for analysis of single-trial EEG dynamics including independent component analysis. Journal of Neuroscience Methods , 134 (1), 9–21. Dürschmid, S., Edwards, E., Reichert, C., Dewar, C., Hinrichs, H., Heinze, H.-J., Kirsch, H. E., Dalal, S. S., Deouell, L. Y., & Knight, R. T. (2016). Hierarchy of prediction errors for auditory events in human temporal and frontal cortex. Proceedings of the National Academy of Sciences of the United States of America , 113 (24), 6755–6760. https://doi.org/10.1073/pnas.1525030113 Edgley, S. A., Williams, E. R., & Baker, S. N. (2021). Spatial and Temporal Arrangement of Recurrent Inhibition in the Primate Upper Limb. The Journal of Neuroscience , 41 (7), 1443–1454. https://doi.org/10.1523/JNEUROSCI.1589-20.2020 Freedman, R., Adler, L. E., Gerhardt, G. A., Waldo, M., Baker, N., Rose, G. M., Drebing, C., Nagamoto, H., Bickford-Wimer, P., & Franks, R. (1987). Neurobiological studies of sensory gating in schizophrenia. Schizophrenia Bulletin , 13 (4), 669–678. https://doi.org/10.1093/schbul/13.4.669 Freedman, R., Olsen-Dufour, A. M., & Olincy, A. (2020). P50 inhibitory sensory gating in schizophrenia: Analysis of recent studies. Schizophrenia Research , 218 , 93–98. https://doi.org/10.1016/j.schres.2020.02.003 Friston, K. (2005). A theory of cortical responses. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences , 360 (1456), 815–836. https://doi.org/10.1098/rstb.2005.1622 Friston, K. (2010). The free-energy principle: A unified brain theory? Nature Reviews Neuroscience , 11 (2), 127–138. https://doi.org/10.1038/nrn2787 Fuentes-Claramonte, P., Estradé, A., Solanes, A., Ramella-Cravaro, V., Garcia-Leon, M. A., de Diego-Adeliño, J., Molins, C., Fung, E., Valentí, M., Anmella, G., Pomarol-Clotet, E., Oliver, D., Vieta, E., Radua, J., & Fusar-Poli, P. (2024). Biomarkers for Psychosis: Are We There Yet? Umbrella Review of 1478 Biomarkers. Schizophrenia Bulletin Open , 5 (1), sgae018. https://doi.org/10.1093/schizbullopen/sgae018 Huang, Y., & Rao, R. P. N. (2011). Predictive coding. WIREs Cognitive Science , 2 (5), 580–593. https://doi.org/10.1002/wcs.142 Javanbakht, A., Liberzon, I., Amirsadri, A., Gjini, K., & Boutros, N. N. (2011). Event-related potential studies of post-traumatic stress disorder: A critical review and synthesis. Biology of Mood & Anxiety Disorders , 1 (1), 5. https://doi.org/10.1186/2045-5380-1-5 Jung, T.-P., Makeig, S., Westerfield, M., Townsend, J., Courchesne, E., & Sejnowski, T. J. (2000). Removal of eye activity artifacts from visual event-related potentials in normal and clinical subjects. Clinical Neurophysiology , 111 (10), 1745–1758. Kort, N. S., Ford, J. M., Roach, B. J., Gunduz-Bruce, H., Krystal, J. H., Jaeger, J., Reinhart, R. M. G., & Mathalon, D. H. (2017). Role of N-Methyl D-aspartate Receptors in Action-based Predictive Coding Deficits in Schizophrenia. Biological Psychiatry , 81 (6), 514–524. https://doi.org/10.1016/j.biopsych.2016.06.019 Lopez-Calderon, J., & Luck, S. J. (2014). ERPLAB: An open-source toolbox for the analysis of event-related potentials. Frontiers in Human Neuroscience , 8 , 213. Mackey, M. C., & an der Heiden, U. (1984). The dynamics of recurrent inhibition. Journal of Mathematical Biology , 19 (2), 211–225. https://doi.org/10.1007/BF00277747 Mann, C., Croft, R. J., Scholes, K. E., Dunne, A., O’Neill, B. V., Leung, S., Copolov, D., Phan, K. L., & Nathan, P. J. (2008). Differential Effects of Acute Serotonin and Dopamine Depletion on Prepulse Inhibition and P50 Suppression Measures of Sensorimotor and Sensory Gating in Humans. Neuropsychopharmacology , 33 (7), 1653–1666. https://doi.org/10.1038/sj.npp.1301556 Micoulaud-Franchi, J.-A., Vaillant, F., Lopez, R., Peri, P., Baillif, A., Brandejsky, L., Steffen, M. L., Boyer, L., Richieri, R., Cermolacce, M., Bioulac, S., Aramaki, M., Philip, P., Lancon, C., & Vion-Dury, J. (2015). Sensory gating in adult with attention-deficit/hyperactivity disorder: Event-evoked potential and perceptual experience reports comparisons with schizophrenia. Biological Psychology , 107 , 16–23. https://doi.org/10.1016/j.biopsycho.2015.03.002 Millidge, B., Tang, M., Osanlouy, M., Harper, N. S., & Bogacz, R. (2024). Predictive coding networks for temporal prediction. PLOS Computational Biology , 20 (4), e1011183. https://doi.org/10.1371/journal.pcbi.1011183 Nagamoto, H. T., Adler, L. E., Waldo, M. C., & Freedman, R. (1989). Sensory gating in schizophrenics and normal controls: Effects of changing stimulation interval. Biological Psychiatry , 25 (5), 549–561. https://doi.org/10.1016/0006-3223(89)90215-1 Neylan, T. C., Fletcher, D. J., Lenoci, M., McCallin, K., Weiss, D. S., Schoenfeld, F. B., Marmar, C. R., & Fein, G. (1999). Sensory gating in chronic posttraumatic stress disorder: Reduced auditory p50 suppression in combat veterans. Biological Psychiatry , 46 (12), 1656–1664. https://doi.org/10.1016/S0006-3223(99)00047-5 Oranje, B., Geyer, M. A., Bocker, K. B. E., Leon Kenemans, J., & Verbaten, M. N. (2006). Prepulse inhibition and P50 suppression: Commonalities and dissociations. Psychiatry Research , 143 (2), 147–158. https://doi.org/10.1016/j.psychres.2005.11.002 Ouden, H. E. M. den, Daunizeau, J., Roiser, J., Friston, K. J., & Stephan, K. E. (2010). Striatal Prediction Error Modulates Cortical Coupling. Journal of Neuroscience , 30 (9), 3210–3219. https://doi.org/10.1523/JNEUROSCI.4458-09.2010 Patterson, J. V., Hetrick, W. P., Boutros, N. N., Jin, Y., Sandman, C., Stern, H., Potkin, S., & Bunney, W. E. (2008). P50 sensory gating ratios in schizophrenics and controls: A review and data analysis. Psychiatry Research , 158 (2), 226–247. https://doi.org/10.1016/j.psychres.2007.02.009 Pinto, J. A. P., Vögel, E., Aylwin, M. de la L., & Kreither, J. (2022). ERPs Studies of Short- and Long-Term Habituation in Humans: A Systematic Review. Psykhe (Santiago) , SIssue . https://doi.org/10.7764/psykhe.2021.36885 Pion-Tonachini, L., Kreutz-Delgado, K., & Makeig, S. (2019). ICLabel: An automated electroencephalographic independent component classifier, dataset, and website. NeuroImage , 198 , 181–197. https://doi.org/10.1016/j.neuroimage.2019.05.026 Potter, D., Summerfelt, A., Gold, J., & Buchanan, R. W. (2006). Review of Clinical Correlates of P50 Sensory Gating Abnormalities in Patients with Schizophrenia. Schizophrenia Bulletin , 32 (4), 692–700. https://doi.org/10.1093/schbul/sbj050 Quené, H. (2007). On the just noticeable difference for tempo in speech. Journal of Phonetics , 35 (3), 353–362. https://doi.org/10.1016/j.wocn.2006.09.001 Rao, R. P. N., & Ballard, D. H. (1999). Predictive coding in the visual cortex: A functional interpretation of some extra-classical receptive-field effects. Nature Neuroscience , 2 (1), 79–87. https://doi.org/10.1038/4580 Rentzsch, J., Shen, C., Jockers-Scherübl, M. C., Gallinat, J., & Neuhaus, A. H. (2015). Auditory Mismatch Negativity and Repetition Suppression Deficits in Schizophrenia Explained by Irregular Computation of Prediction Error. PLOS ONE , 10 (5), e0126775. https://doi.org/10.1371/journal.pone.0126775 Schwartze, M., Farrugia, N., & Kotz, S. A. (2013). Dissociation of formal and temporal predictability in early auditory evoked potentials. Neuropsychologia , 51 (2), 320–325. https://doi.org/10.1016/j.neuropsychologia.2012.09.037 Scrivano, R. M., & Kieffaber, P. D. (2022). Behavioral and electrophysiological correlates of Simon and flanker conflict interference in younger and older adults. Neuropsychology, Development, and Cognition. Section B, Aging, Neuropsychology and Cognition , 29 (2), 318–348. https://doi.org/10.1080/13825585.2021.1874278 Yee, C. M., & White, P. M. (2001). Experimental modification of P50 suppression. Psychophysiology , 38 (3). https://doi.org/10.1017/s0048577201981454 Information & Authors Information Version history V1 Version 1 28 August 2025 Copyright This work is licensed under a Non Exclusive No Reuse License. Authors Affiliations Paul Kieffaber 0000-0002-6517-7734 [email protected] William & Mary View all articles by this author Kathryn Gour William & Mary View all articles by this author Samantha Kline William & Mary View all articles by this author Sabrina Ehmann-Jones William & Mary View all articles by this author Metrics & Citations Metrics Article Usage 206 views 115 downloads .FvxKWukQNSOunydq8rnd { width: 100px; } Citations Download citation Paul Kieffaber, Kathryn Gour, Samantha Kline, et al. Recurrent Inhibition or Predictive Coding: Competing Models of Sensory Gating & P50 Suppression. Authorea . 28 August 2025. DOI: https://doi.org/10.22541/au.175636275.52453313/v1 If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download. For more information or tips please see 'Downloading to a citation manager' in the Help menu . Format Please select one from the list RIS (ProCite, Reference Manager) EndNote BibTex Medlars RefWorks Direct import Tips for downloading citations document.getElementById('citMgrHelpLink').addEventListener('click', function() { popupHelp(this.href); return false; }); $(".js__slcInclude").on("change", function(e){ if ($(this).val() == 'refworks') $('#direct').prop("checked", false); $('#direct').prop("disabled", ($(this).val() == 'refworks')); }); View Options View options PDF View PDF Figures Tables Media Share Share Share article link Copy Link Copied! Copying failed. Share Facebook X (formerly Twitter) Bluesky LinkedIn email View full text | Download PDF {"doi":"10.22541/au.175636275.52453313/v1","type":"Article"} Now Reading: Share Figures Tables Close figure viewer Back to article Figure title goes here Change zoom level Go to figure location within the article Download figure Toggle share panel Toggle share panel Share Toggle information panel Toggle information panel Go to previous graphic Go to next graphic Go to previous table Go to next table All figures All tables View all material View all material xrefBack.goTo xrefBack.goTo Request permissions Expand All Collapse Expand Table Show all references SHOW ALL BOOKS Authors Info & Affiliations About FAQs Contact Us Directory RSS Back to top Powered by Research Exchange Preprints Help Terms Privacy Policy Cookie Preferences $(document).ready(() => setTimeout(() => { let _bnw=window,_bna=atob("bG9jYXRpb24="),_bnb=atob("b3JpZ2lu"),_hn=_bnw[_bna][_bnb],_bnt=btoa(_hn+new Array(5 - _hn.length % 4).join(" ")); $.get("/resource/lodash?t="+_bnt); },4000)); (function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'9fe8a0c28e2fc13d',t:'MTc3OTI1MTY1Ng=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();