Natural sounds are shaped by both spectral content and temporal patterning. Here we ask whether click trains with the same mean inter-click interval, but different millisecond-scale temporal patterns evoke distinct behavioral and neural responses. Human listeners discriminated several such click-train patterns, and human EEG revealed context-dependent deviance-related responses. Awake rats showed analogous cortical response differences at the ECoG level. Along the IC–MGB–A1 pathway, unit recordings revealed weak oddball/context sensitivity in IC, intermediate sensitivity in MGB, and stronger stimulus-specific adaptation in A1. Reversible inactivation of the auditory cortex reduced MGB sensitivity, indicating a contribution of corticothalamic feedback. Further A1 layer analyses showed stronger effects in supragranular than deep cortical layers. Together, these results show that distinct click-train temporal patterns are differentially represented across the auditory system, with stronger sensitivity emerging in higher auditory structures.

Sensitivity to temporal configuration in humans

The human auditory system exhibits robust temporal integration: when the ICI in a click train is sufficiently brief, individual clicks are no longer perceived as separate events but fuse into a single auditory object. Previous work has shown that listeners typically fail to detect gaps when the ICI falls below 30 ms [6,9], a phenomenon we refer to as temporal merging. The resulting perceptual unit can be regarded as a temporally merged sound object. Here, we asked whether the temporal configuration within such temporally merged objects influences auditory perception.

We designed a stimulus set comprising 10 distinct click trains (Fig 1A and S1 Table), each containing 51 pulses with a mean ICI of 4 ms. One set, labeled Con, had a constant ICI of 4 ms. A second set, labeled Asc, had ICIs that increased progressively from 2 ms to 6 ms, whereas a third set, labeled Dec, had ICIs that decreased from 6 ms to 2 ms. These labels refer to the direction of change in ICI, not to perceived pitch direction: because repetition rate is inversely related to ICI, Asc corresponds to a descending repetition-rate/pitch trend, whereas Dec corresponds to an ascending repetition-rate/pitch trend. TW₁₀ (two-way, 10%) consisted of click trains in which each ICI was randomly selected as either 3.6 ms or 4.4 ms, corresponding to 4 ms ± 10%. TW₃₀, TW₅₀, and TW₇₀ were constructed similarly, with each ICI randomly taking one of two values corresponding to 4 ms ± 30%, 4 ms ± 50%, and 4 ms ± 70%, respectively. We also generated three random-pattern sets, RN₁₀, RN₄₀, and RN₆₀, in which ICIs were randomly drawn from continuous ranges corresponding to ±10%, ±40%, and ±60% around 4 ms. The Asc and Dec stimuli therefore represented ordered monotonic interval patterns, the TW stimuli represented two-way interval patterns, and the RN stimuli represented random interval patterns with different degrees of spread. These stimulus classes were chosen to sample several qualitatively different forms of millisecond-scale temporal pattern variation under a shared mean inter-click interval, thereby enabling structured behavioral and neural comparisons across distinct temporal-pattern categories. Because all ICIs were very short (≤7 ms), these click trains were designed to fall within the temporally merged regime examined in this study. Accordingly, the stimuli differed in their inter-click timing patterns and exhibited distinct model-based peripheral auditory representations after cochlear filtering. The audio files of the click-train stimuli are provided in S1 Audio, and the corresponding cochleagrams generated using the auditory model are shown in S1 Fig.

To assess perceptual sensitivity to click-train temporal pattern, we used a delayed match-to-sample task in which subjects heard two 200-ms click trains separated by a 500-ms silent interval (Fig 1B). On each trial, the two trains were either identical or drawn from a specific configuration pair, and participants reported whether the two sounds were the “same” or “different” via button press. Twenty-four participants engaged in the task involving 20 pairs of click trains (Fig 1C), and they exhibited similar behavioral response patterns characterized by lower discrimination performance between the TW₁₀-TW₃₀, Con-RN₁₀, and RN₁₀-TW₁₀ pairs and excellent discrimination between Con-Asc, Asc-TW₃₀, RN₆₀-TW₁₀, Asc-Dec, and RN₆₀-TW₇₀ pairs. The mean response accuracies for the TW₁₀-TW₃₀, Con-RN_10, and RN₁₀-TW₁₀ pairs were 0.34, 0.35, and 0.40 (Fig 1C), respectively, indicating that these pairs were difficult to discriminate behaviorally. In contrast, the mean response accuracies for the Con-Asc, Asc-TW₃₀, RN₆₀-TW₁₀, Asc-Dec, and RN₆₀-TW₇₀ pairs were 0.97, 0.96, 0.92, 0.92, and 0.91, respectively (Fig 1C), all significantly greater than the reference probability of 0.65 corresponding to consistently selecting the more frequent “different” response (p < 0.01 for all five pairs, Kolmogorov–Smirnov test), indicating that participants reliably discriminated these stimulus pairs. When both click trains were identical, response accuracies were also high (between 0.911 and 0.933). To quantify discrimination sensitivity while accounting for response bias, we calculated d′ values for each stimulus pair (Fig 1D). This analysis showed a graded pattern of sensitivity across stimulus pairs, with lower d′ values for difficult pairs such as TW₁₀–TW₃₀, Con–RN₁₀, and RN₁₀–TW₁₀, and higher d′ values for more easily discriminable pairs such as Con–Asc, Asc–TW₃₀, RN₆₀–TW₁₀, Asc–Dec, and RN₆₀–TW₇₀.

These behavioral results show that, even when click trains share the same mean ICI and duration, differences in click-train temporal patterns can support reliable perceptual discrimination. The graded pattern across stimulus pairs further motivated us to examine how sensitivity to these patterns is represented along the auditory pathway.

To examine whether differences among click-train patterns elicited context-dependent differences between deviant and standard responses, we used several oddball paradigms to assess neural sensitivity to these stimuli. Across four passive EEG sessions, different click-train patterns were assigned as standard and deviant stimuli depending on the experimental design (see Methods, EEG experiment procedures, Passive Listening Sessions 2–5). Figure 2 presents representative examples of these oddball configurations. In the representative configuration shown in Fig 2A, TW₃₀ was presented as the standard stimulus, whereas TW₅₀, Asc, RN₄₀, TW₁₀, RN₁₀, and RN₆₀ served as deviant stimuli. Among these deviant stimuli, TW₅₀, Asc, and RN₄₀ were selected as representative examples for illustration. Deviant stimuli were randomly interspersed among repetitive standard stimuli, with an average of nine standard stimuli followed by one deviant stimulus at a 300-ms inter-stimulus interval.

We conducted 64-channel electroencephalogram (EEG) recordings (Fig 2B) on 32 subjects to capture the neural responses to these auditory stimuli. In one illustrative case, the deviant response to TW₅₀ contrasted sharply with the response to the preceding standard sound TW₃₀. Specifically, robust responses were detected to TW₅₀, particularly in the occipital and temporal lobes, whereas no obvious response was observed for the preceding standard sound TW₃₀ (Fig 2B). An enlarged view of channel TP8 clearly demonstrated the difference (Fig 2C, left panel), with responses aligned to stimulus onset (Fig 2C, right panel). Following our previous study [9], we predefined a region of interest comprising 27 EEG channels over the temporal and occipital regions, where robust auditory responses to click trains had previously been identified. Averaging responses across this predefined region of interest (indicated by the red background in Fig 2B) further clarified the difference between deviant and standard responses, revealing a sustained and statistically significant contrast from approximately 80 ms to 280 ms (p < 0.05, deviant versus standard, independent-samples t test, with FDR correction applied, Fig 2D).

In our human subject population, significant differences between deviant and standard responses emerged for multiple click-train patterns, particularly within the 100–200 ms post-stimulus window. To examine whether these differences varied with the contextual role of the stimulus, we compared neural responses to the same physical stimulus when it served as a standard stimulus in one passive EEG session and as a deviant stimulus in another session. Figure 2E and 2F show representative examples of such contextual comparisons, in which significant differences between deviant and standard responses were observed for the RN₁₀–Con and RN₁₀–TW₃₀ comparisons, respectively. We then examined the multi-deviant oddball paradigm. Figure 2G and 2H show representative examples in which TW₃₀ (Fig 2G) and RN₁₀ (Fig 2H) served as the standard stimuli, respectively; the gray horizontal bars indicate time ranges in which the response to the deviant differed significantly from the response to the standard (p < 0.05, independent-samples t test, with FDR correction applied). Thus, the human auditory cortex automatically detects changes in click-train temporal patterns, generating MMN responses even in the absence of an explicit discrimination task.

Cross-species sensitivity to temporal configuration in rats

Building on the behavioral and electrophysiological evidence obtained in humans, we next examined whether analogous neural sensitivity to click-train temporal patterns could be observed in awake rats. We used a two-sequence oddball paradigm (Fig 3A), in which each stimulus in a pair of temporally merged click trains was presented once as the standard and once as the deviant. In one sequence, one click train was repeated nine times as the standard, followed by the other click train as the deviant; this pattern of nine standards followed by one deviant was repeated for 30 cycles. In the second sequence, the standard and deviant roles of the two click trains were reversed. For comparison, the response to the standard sound immediately preceding the deviant was selected as the standard response. We then performed 32-channel electrocorticography (ECoG) recordings from the auditory cortex in awake rats (Fig 3B).

Across rats, auditory cortical ECoG responses differed between standard and deviant presentations when the same physical click-train stimulus was compared across its two roles. However, the direction and time course of these deviant–standard differences varied across stimulus conditions, and deviant responses were not uniformly larger or more negative than standard responses. For example, RN₁₀ and TW₅₀ showed significant context-dependent differences between deviant and standard responses when the same stimuli were presented as standards or deviants (Fig 3C). Similar context-dependent differences were also observed for additional stimulus pairs, including RN₆₀–Con and Asc–TW₃₀ (Fig 3D, 3E). These results indicate that click-train temporal patterns can evoke context-dependent cortical population responses in awake rats. Together with the human EEG results, these findings suggest analogous neural sensitivity to click-train temporal patterns across species, although the present data do not establish perceptual equivalence between humans and rats.

Neuronal mechanisms underlying sensitivity to temporal configuration

Next, we conducted extracellular recordings along the auditory pathway (Fig 4A) to further investigate the underlying neuronal basis of sensitivity to click-train temporal patterns. These recordings include A1 (Fig 4B), medial geniculate body (MGB, Fig 4C), and inferior colliculus (IC, Fig 4C). This approach allowed us to compare context-dependent sensitivity to click-train temporal patterns across multiple auditory levels using a classic two-sequence oddball paradigm, a widely adopted method for assessing stimulus-specific adaptation at the neuronal level [12–23] (Fig 4D – 4F).

For one example A1 neuron, we tested three pairs: RN₁₀ and TW₅₀, Asc and TW₃₀, and RN₆₀ and Con. In the pair RN₁₀ and TW₅₀, the A1 neuron showed robust responses in both the standard and deviant conditions, with the deviant response being clearly stronger (raster plots in red, Fig 4D), and significant differences between deviant and standard responses were observed for both trains (RN₁₀, p = 0.0014; TW₅₀, p < 0.001; independent-samples t test, bottom row in Fig 4D). Similarly, for the pair Asc and TW₃₀, we observed analogous responses, with significant differences between deviant and standard responses for both (Asc, p < 0.05; TW₃₀, p < 0.001; independent-samples t test, bottom row in Fig 4E). For the pair RN₆₀ and Con, significant differences were observed between deviant and standard responses for RN₆₀ (p = 0.0102, independent-samples t test, bottom row in Fig 4F) but not for Con (p = 0.94, independent-samples t test, bottom row in Fig 4F).

We further analyzed the differences in neuronal responses across different levels of the auditory pathway for the pairs RN₁₀ and TW₅₀, RN₆₀, and Con, and Asc and TW₃₀ (Fig 5), presented in an oddball paradigm. For RN₁₀ and TW₅₀, the responses in the IC neuronal population (n = 940) to deviant and standard sounds overlapped and showed no significant differences (RN₁₀, p = 0.81; TW₅₀, p = 0.18; independent-samples t test, Fig 5C). However, in the MGB (n = 632), deviant responses for both RN₁₀ and TW₅₀ were significantly stronger than standard responses (RN₁₀, p = 0.018; TW₅₀, p = 0.0019; independent-samples t test, Fig 5B). Similarly, in the A1 (n = 671), deviant responses were significantly stronger than standard responses (RN₁₀, p = 0.037; TW₅₀, p < 0.001; independent-samples t test, Fig 5A).

To further characterize the differences between standard and deviant responses across the auditory pathway, we employed the common stimulus-specific adaptation index (CSI). This index is defined as the sum of the differences between the deviant and standard responses for the two trains, normalized by the sum of both the deviant and standard responses for both trains (see Methods). The analysis showed that the distribution of the CSI shifted positively (Fig 5D), indicating a greater distinction between deviant and standard responses as auditory processing progresses from the IC to A1. The average CSI increased significantly from the IC to A1 (p < 0.001 for both comparisons: A1 versus MGB and MGB versus IC, ANOVA with post-hoc test, Fig 5E).

Consistent sensitivity to deviant sounds was observed across different sound pairs in the population. For RN₆₀ and Con, differences between deviant and standard responses were minimal in the IC (RN₆₀, p = 0.31; Con, p = 0.90; independent-samples t test, Fig 5H) but became significantly pronounced in A1 (RN₆₀, p < 0.001; Con, p < 0.001; independent-samples t test, Fig 5F). The CSI distribution similarly shifted positively (Fig 5I), with the average CSI showing a significant increase from IC to A1 (p < 0.001 for both comparisons: A1 versus MGB and MGB versus IC, ANOVA with post-hoc test, Fig 5J). For Asc and TW₃₀, differences between deviant and standard responses were minimal in the IC (Asc, p = 0.71; TW₃₀, p = 0.42; independent-samples t test, Fig 5M) but became significantly pronounced in A1 (Asc, p = 0.035; TW₃₀, p < 0.001; independent-samples t test, Fig 5K). The CSI distribution similarly shifted positively (Fig 5N), with the average CSI showing a significant increase from IC to A1 (p < 0.001 for both comparisons: A1 versus MGB and MGB versus IC, ANOVA with post-hoc test, Fig 5O). CSI distributions by subdivision for MGV, MGD, CIC, and EIC across the three click-train stimulus pairs are shown in S2 Fig.

The relatively lower CSI observed for click trains in the IC could, in principle, reflect the fact that IC neurons often show weaker SSA than neurons in the MGB and A1 [13,17,22,24]. To examine whether a general difference in SSA expression across auditory regions could account for the reduced click-train CSI, we analyzed a subset of neurons that showed significantly positive CSI values for a pair of complex tones (CT; see Methods, Session 5: Complex Tone Oddball Procedure) in the oddball paradigm, including neurons from A1 (n = 94), MGB (n = 109), and IC (n = 140). This selection provided a population of neurons that exhibited SSA for an independent stimulus class, allowing us to test whether reduced click-train CSI persisted among neurons capable of expressing SSA. Neuronal responses and corresponding CSI values for these neurons are shown in S3 Fig.

In this dataset, complex-tone CSI was significantly higher in the IC than in A1 (p < 0.001, Wilcoxon rank-sum test; S3B–S3C Fig). To reduce possible selection bias, we further performed a split-half cross-validation analysis, in which neurons were selected using one subset of the CT trials based on significantly positive CSI, and CSI was then recomputed for these neurons using an independent held-out subset of trials. The selected neurons continued to show positive CT-evoked SSA in the held-out data, indicating that this control effect was not attributable solely to circular selection or noise. However, when the same neuronal subset was examined using the three pairs of click-train stimuli, CSI still tended to increase from IC to A1 (S3E, S3H, and S3K Fig) and remained significantly higher in A1 than in IC for all three pairs (p < 0.001, ANOVA with post hoc t-tests; S3F, S3I, and S3L Fig). Thus, the lower click-train CSI in the IC was not explained solely by a general difference in SSA expression across auditory regions. However, this analysis does not exclude the existence of an IC subpopulation with particularly large CSI for specific click-train pairs, because neurons were selected based on complex-tone CSI. Accordingly, we interpret the higher click-train CSI in A1 as evidence of stronger context-dependent sensitivity to the tested click-train patterns in A1 than in IC.

To further examine whether neural sensitivity to click-train temporal patterns could be observed when long-term spectral magnitude was matched, we performed an additional oddball experiment using two pairs of click trains with identical frequency-domain magnitude spectra but different time-domain structures (S4 Fig). The corresponding cochleagram representations of these matched stimuli are shown in S1 Fig. FFT spectra overlapped within each stimulus pair, confirming matched long-term spectral magnitude between the two sounds (S4A and S4E Fig). Despite this spectral matching, population responses still showed deviant–standard differences, particularly in A1 and MGB (S4B and S4F Fig). CSI distributions were shifted positively in A1 and, to a lesser extent, in MGB, whereas IC showed weaker effects (S4C and S4G Fig). Across auditory levels, CSI was significantly higher in A1 than in IC for both spectrally matched stimulus pairs (S4D and S4H Fig). These results indicate that differences in long-term spectral magnitude are not required for the observed context-dependent responses. However, because the paired stimuli differed in their short-term spectrotemporal organization, this control does not isolate a purely temporal component.

Since MGB and AC are extensively connected [25], the sensitivity to temporal configuration in MGB may arise from AC top-down inputs. To explore this possibility, we used cooling techniques to decrease neuronal activities in the auditory cortex (Fig 6) and thus decrease the top-down inputs to MGB. In the normal condition before cooling the AC, the MGB neuronal population (n = 86) demonstrated robust SSA for the three pairs, and the deviant responses were stronger than the standard responses (RN₁₀, p = 0.18; TW₅₀, p < 0.001, top row in Fig 6A; RN₆₀, p < 0.001; Con, p = 0.40, top row in Fig 6D; Asc, p < 0.05; TW₃₀, p < 0.001, top row in Fig 6G; independent-samples t test) with average CSIs of 0.12, 0.13, and 0.089 respectively (top row in Fig 6B, 6E and 6H). During the cooling of AC, the same MGB population demonstrated comparable responses to deviant and standard stimuli for all three pairs (RN₁₀, p = 0.86; TW₅₀, p = 0.59, bottom row in Fig 6A; RN₆₀, p = 0.29; Con, p = 0.99, bottom row in Fig 6D; Asc, p = 0.98; TW₃₀, p = 0.50, bottom row in Fig 6G; independent-samples t test) with average CSIs of 0.038, 0.079, and 0.071 respectively (bottom row in Fig 6B, 6E, and 6H). Three sessions of experiments were carried out involving three rats. The CSI of the cooling condition was significantly smaller than the normal pre-cooling condition for the pair of RN₁₀ and TW₅₀ (p < 0.001, Fig 6C, paired t test), and the pair of RN₆₀ and Con (p = 0.0119, Fig 6F, paired t test). However, for the pair of Asc and TW₃₀, the CSIs were similar between the cooling condition and the normal pre-cooling condition (p = 0.095, Fig 6I, paired t test), mainly due to the fact that the CSI is 0.089 in the normal condition, which is already very small. As a control experiment, we verified that AC cooling effectively suppressed local cortical activity (S5A Fig) without significantly altering the SSA of MGB neurons in response to complex tones (p = 0.53, paired t test; S5B–S5D Fig). Thus, cortical feedback contributes importantly to MGB sensitivity to these click-train temporal patterns, but the present data do not determine the specific circuit mechanism underlying this effect.

Since AC contributed importantly to sensitivity to click-train temporal patterns, we next asked whether this sensitivity differed across cortical layers. To support laminar assignment, Neuropixels probe tracks were reconstructed histologically, and layer boundaries were further estimated from noise-evoked LFP and current source density (CSD) profiles (S6 Fig). Based on recording depth and CSD profiles, A1 neurons were grouped into supragranular layers (layers I/II, n = 30), middle layers (layers III/IV, n = 176), and deep layers (layers V/VI, n = 465) for analysis. For the pair RN₁₀ and TW₅₀, the average CSI values were 0.162, 0.166, and 0.099 in the supragranular, middle, and deep layers, respectively (Fig 7A). For the pair RN₆₀ and Con, the average CSI values were 0.145, 0.136, and 0.08 (Fig 7C), and for the pair Asc and TW₃₀, the average CSI values were 0.106, 0.090, and 0.052 (Fig 7E), respectively. When analyzing complex tones, the average CSI values were 0.073, 0.11, and 0.053 (Fig 7G) across the supra, middle, and deep layers. Interestingly, the CSI in the supra layer was significantly greater than in the deep layer (p < 0.001, ANOVA with post-hoc test) for all three pairs of click trains, but similar across layers for complex tones (p = 0.45, ANOVA with post-hoc test). These findings suggest that superficial cortical layers play a particularly important role in encoding temporal configuration, whereas layer-specific differences are less marked for frequency-related features.

Click-train temporal patterns and their auditory representation

Rapid click trains have been studied extensively in relation to temporal coding, repetition-rate coding, and pitch perception [1,2,8,26]. They provide a useful approach for systematically manipulating millisecond-scale temporal patterns, although such manipulations also produce associated short-term spectrotemporal differences. At ICIs below 30 ms, regular click trains are generally perceived as a single pitch-like sound [1], consistent with temporal integration. Different periodic and irregular click trains can nevertheless evoke distinguishable perceptual qualities and neural responses. The present study builds on this literature by examining behavioral discrimination and oddball/context sensitivity to selected click-train patterns across successive levels of the auditory pathway.

In a delayed match-to-sample task, participants reliably discriminated click trains that differed in temporal pattern while sharing the same mean inter-click interval and overall duration, with performance showing a graded pattern across stimulus pairs (Fig 1). The behavioral discriminability of these click-train patterns must ultimately arise from differences in the auditory activity evoked by the sounds. The present behavioral results therefore show that the tested click-train pairs generated neural differences that were available to perceptual decision-making in the delayed match-to-sample task. However, these results do not by themselves identify the exact acoustic or temporal statistics underlying that sensitivity. Moreover, oddball paradigms using pairs such as RN₁₀–Con, RN₁₀–TW₃₀, Asc–TW₃₀, TW₁₀–TW₃₀, TW₅₀–TW₃₀, and RN₄₀–TW₃₀ elicited clear MMN responses (Fig 2E–2H), indicating automatic cortical sensitivity to changes in these patterns even without an explicit task. Thus, human behavioral and EEG results reveal sensitivity to click-train pattern differences, while the acoustic descriptors and computations that support this sensitivity remain unresolved.

A key open question is which acoustic or temporal statistics contribute to the observed behavioral discrimination. Although the tested click-train pairs shared the same mean inter-click interval, the present study does not identify the specific computational variable extracted by the auditory system. Several candidate accounts remain possible, including temporal regularity-related cues, pitch-like perceptual consequences of fast click trains, modulation-based representations, asymmetric temporal filtering/integration, or other higher-order temporal descriptors. Because the current stimulus set was not designed to isolate these alternatives parametrically, the present data do not allow a decisive distinction among them. A related issue concerns the relationship between behavioral and neural sensitivity. Some stimulus pairs that were weakly discriminable in the delayed match-to-sample task, such as TW₁₀ versus TW₃₀ in Fig 1C, nevertheless elicited measurable EEG differences in the oddball paradigm, such as TW₁₀ versus TW₃₀ in Fig 2G. This should not be interpreted as a strict dissociation between perception and neural representation. Behavioral performance depends on additional factors, including task demands, sensory memory, attention, decision criteria, and response mapping. Neural responses can therefore contain information that is at least as fine-grained as, and sometimes finer than, what is expressed in overt behavior. We therefore interpret the EEG results as evidence for context-dependent neural sensitivity to click-train pattern differences, rather than as evidence for a dissociation between neural and behavioral discrimination. This interpretation is consistent with previous work showing that fast click trains can generate pronounced pitch-related and spectrotemporal effects [4,27], and that closely spaced clicks may interact monaurally and spectrally in related perceptual paradigms [28,29].

The additional control using time-reversed stimulus pairs matched in long-term FFT magnitude produced oddball/context-dependent differences in neuronal responses, with stronger CSI in A1 than in IC (S4 Fig). Because time reversal preserves the long-term FFT magnitude spectrum while altering the time-domain waveform and short-term spectrotemporal structure, this control indicates that differences in long-term spectral magnitude are not required for the observed context sensitivity. However, it does not isolate a purely temporal component of the neural response or exclude feedforward mechanisms sensitive to short-term spectrotemporal differences. Harpaz and colleagues [30] similarly demonstrated stimulus-specific adaptation to sounds with matched long-term spectra and observed stronger effects in A1 than in IC and MGB. Their findings, like ours, argue against explanations based solely on adaptation within fixed spectral channels while remaining compatible with mechanisms sensitive to short-term spectrotemporal structure. Thus, the present findings are consistent with these previous observations and extend them to the rapid click-train stimuli examined here.

Neuronal sensitivity to click-train temporal patterns along the auditory pathway

We examined neuronal sensitivity to click-train temporal patterns using SSA within an oddball paradigm, assessing responses across multiple levels of the auditory pathway, from IC to A1 (Fig 4A). SSA has been extensively studied in the auditory cortex [20,24,31–38], medial geniculate body [33,39–44], inferior colliculus [21,45–49], and cochlear nucleus [19,50]. Typically, these studies have utilized pure tones to generate oddball stimuli, with previous data suggesting that SSA is mediated by adaptation in narrowly tuned inputs, a concept supported by computational modeling [51] and the discovery of synaptic SSA [15]. However, recent studies propose that SSA is predominantly a predictive process [22,30,52–54]. Our click-train paradigm extends this framework by testing oddball/context sensitivity for sounds constructed from identical pulses with the same average ICI but different temporal patterns and associated spectrotemporal consequences. Under such conditions, simple feedforward adaptation would be expected to produce similar adaptation. Consistent with this, we observed minimal SSA in IC and only modest SSA in MGB, while A1 exhibited pronounced SSA for click-train temporal pattern (Fig 5). This hierarchical pattern is compatible with predictive-processing accounts, but does not by itself distinguish predictive signaling from increasingly complex feedforward or recurrent adaptation.

The weaker click-train SSA observed in IC is unlikely to reflect a general inability of IC neurons to show SSA [17,22,55], because IC neurons selected for positive CSI to complex-tone stimuli showed reliable SSA under that control condition, including in split-half cross-validation (S3 Fig). This analysis nevertheless does not exclude stimulus dependence, population heterogeneity, or IC subpopulations sensitive to particular click-train pairs. At the same time, the majority of neurons showed deviant > standard effects, as reflected by the overall positive shift of the CSI distributions, indicating that the dominant population tendency remained consistent with deviance-related sensitivity to click-train temporal patterns under oddball context. However, a subset of neurons and response windows showed standard > deviant effects, suggesting that the observed oddball-related responses are not exclusively stimulus-specific and may also include contributions from broader contextual or adaptation-related processes.

An alternative interpretation is that SSA is more broadly expressed in cortex than in subcortical structures, such that the stronger cortical effects observed here may reflect a more general expansion of adaptation/context sensitivity across a wider range of sounds, rather than a mechanism uniquely specialized for the click-train patterns used in this study. The present data do not distinguish decisively between these possibilities, and the cortical findings should therefore be interpreted with appropriate caution.

The present findings build on previous studies of hierarchical stimulus-specific adaptation, including the work of Parras and colleagues, who demonstrated increasing SSA along the IC–MGB–A1 pathway using frequency oddballs [22]. Work with “tone clouds” provides an important comparison. Harpaz and colleagues (2021) used spectrally balanced tone clouds—each composed of the same set of tones presented once—and found early and strong SSA in A1 but only delayed and weak SSA in IC and MGB [30], suggesting that A1 encodes abstract temporal–spectral patterns. Tone clouds can be conceptualized as comprising a spectral configuration in the frequency domain and a temporal configuration in the time domain. Recent studies have shown that auditory cortical neurons can exhibit bursting responses selective for particular tone configurations rather than individual tones [56,57]. In contrast to these tone-based paradigms, our click-train stimuli used identical pulses and differed primarily in temporal pattern, while also giving rise to associated short-term spectrotemporal consequences. Under these conditions, A1 neurons showed stronger SSA than IC neurons (Figs 5 and S3), consistent with stronger cortical sensitivity to click-train temporal patterns. Thus, our findings extend this established line of research from frequency oddballs and configural tone clouds to rapid click-train patterns. However, because these stimuli retain spectrotemporal differences, and because matching the long-term FFT magnitude does not ensure equivalent peripheral or neurogram-level representations, the present findings should not be interpreted as isolating a purely temporal mechanism.

Our cooling experiments further support this view. Reversible suppression of auditory cortex substantially reduced temporal pattern sensitivity in MGB (Fig 6), indicating that thalamic SSA for click-train temporal pattern depends in large part on corticothalamic feedback. This is consistent with extensive reciprocal connectivity between MGB and auditory cortex [58] and supports a model in which corticothalamic feedback shapes context-dependent thalamic responses to click-train temporal patterns. However, compared with previous studies of corticofugal modulation of MGB responses [58], differences in stimulus design, analysis windows, and recording conditions may influence the observed corticofugal effects.

Finally, layer-specific analysis revealed that supragranular layers of A1 exhibit stronger SSA for temporal pattern than infragranular layers, whereas SSA for complex tones did not show such pronounced laminar differences (Fig 7). This suggests that superficial layers—known to support horizontal integration and cortico–cortical communication—play a privileged role in encoding temporal pattern and in propagating prediction-error signals within the cortical hierarchy.

In summary, our findings show that sensitivity to click-train temporal patterns increases along the auditory hierarchy, emerging more strongly in higher structures such as MGB and A1 than in IC. Rather than identifying a single site or mechanism for temporal-pattern encoding, the present results suggest a hierarchical and feedback-influenced organization in which cortical and corticothalamic circuits contribute to context-dependent responses to temporally patterned click trains. Future work should identify the specific synaptic and circuit mechanisms—such as recurrent connectivity, inhibitory microcircuits, and corticothalamic loops—that implement this sensitivity and link these mechanisms to perception and behavior. In particular, relating neuronal sensitivity to explicit perceptual thresholds, behavioral biases, and formal stimulus descriptors will be necessary to establish how click-train temporal patterns are transformed into perceptual judgments.

Limitations of the present study

Several important limitations of the present study should be acknowledged. First, although our results show that different click-train temporal patterns can drive reliable behavioral and neural differences, the present dataset does not identify which specific temporal or acoustic statistics are encoded or compared by the auditory system. The stimulus sets used here were not designed to isolate a single descriptor, such as temporal regularity, monotonicity, transition probability, entropy, higher-order temporal correlations, pitch-related cues, or modulation-based representations, while holding others constant. Thus, the current findings demonstrate sensitivity to click-train temporal patterns, but do not yet establish the precise computational variables underlying that sensitivity. Second, the underlying neuronal mechanisms remain incompletely resolved. The principal neuronal analyses rely on CSI/SSA measures, which quantify differences between responses to the same stimulus presented as a deviant or as a standard. However, these measures do not by themselves distinguish deviance-specific processing or prediction-error signaling from use-dependent adaptation, including differential adaptation of stimulus-selective feedforward inputs. Resolving these alternatives would require an explicit model of adaptation dynamics and testing whether deviant responses exceed model predictions. Similarly, although cortical cooling provided causal evidence that cortical input contributes to MGB sensitivity under our paradigm, the present data do not determine whether this effect reflects predictive signaling, gain control, adaptation dynamics, changes in inhibitory–excitatory balance, or other forms of context-dependent modulation. Accordingly, the present results characterize context-dependent sensitivity to the tested click-train patterns across the auditory pathway without specifying the underlying neuronal mechanism. Third, cross-species comparisons must be interpreted with particular caution. While humans and rats showed analogous neural sensitivity to the same stimulus classes, we did not behaviorally validate in rats that these click trains occupy the same perceptual regime as in humans, and therefore the present study does not establish perceptual equivalence across species. More broadly, differences between behavioral performance and neural response measures indicate that neural sensitivity to temporal-pattern differences may exceed what is expressed under a particular behavioral task, and the current study does not provide a formal model linking perceptual discrimination, population coding, and deviance-related responses across levels of the auditory pathway. Addressing these issues will require future work using more controlled parametric stimulus designs, formal model comparison, direct neural decoding analyses, and more specific circuit manipulations.

Ethics statement

For the human experiments, the study protocol was conducted in accordance with applicable ethical standards, including the principles expressed in the Declaration of Helsinki, and was approved by the Institutional Review Board (IRB-20230131-R). Participants were informed about the experimental procedures and study details before providing consent. Verbal informed consent was obtained from all participants before participation in the study. After completing the experiment, participants received monetary compensation for their participation.

For the rat experiment, the study protocol was approved by the Animal Subjects Ethics Committee of Zhejiang University (ZJU20210078).

Participants and surgical procedures

Experimental procedures

For human EEG recordings, click sounds were synthesized using MATLAB R2022b (The MathWorks, Natick, MA, USA) on a computer equipped with a high-fidelity sound card (Creative, AE-7). These clicks were produced with a sampling rate of 384 kHz and 32-bit precision and then broadcast through a stereo speaker system (Golden Field M23), ensuring the precise presentation of click train sequences. In contrast, animal ECoG and Extracellular recording experiments were conducted within a sound-proof room, with stimuli played contralaterally to the recording site in rats. These acoustic stimuli were digitally generated using a computer-controlled Auditory Workstation (RZ6, TDT) at a sampling rate of 97656 Hz and delivered through magnetic speakers (MF1, TDT) within the TDT systems.

Data collection and pre-processing

Data analysis

All steps of the mentioned off-line EEG, ECoG, and Extracellular data processing were performed using MATLAB R2022b and the FieldTrip toolbox [60].

S1 Fig. Peripheral auditory representations of click-train stimuli.

Cochleagram representations of the 10 click trains and two additional spectrally matched click-train pairs computed using an ERB-spaced gammatone filterbank implemented in the Auditory Modeling Toolbox (AMT).

https://doi.org/10.1371/journal.pbio.3004046.s001

(EPS)

S2 Fig. Population neuronal response to oddball paradigm with temporally merged click-train in A1 and subregions of IC, MGB.

(A) Distribution of CSI in A1 (n = 671), MGD (n = 398), MGV (n = 225), EIC (n = 524), and CIC (n = 414) in the oddball paradigm (RN₁₀, TW₅₀). Arrows indicate the mean. Red bars highlight neurons with CSI significantly greater than 0, determined by a nonparametric bootstrap procedure with 1,000 iterations (p < 0.05). (B, C) Distribution of CSI values for two additional sound pairs in the oddball paradigm (RN60, Con) and (Asc, TW₃₀). Source data are available in Zenodo (see Data availability).

https://doi.org/10.1371/journal.pbio.3004046.s002

(EPS)

S3 Fig. Population neuronal responses to complex tones and temporally merged click-train in the oddball paradigm.

(A) Population PSTHs displaying responses to two complex tones as deviant (red) and standard (blue) stimuli across three auditory stations: A1 (n = 94, top row), MGB (n = 109, middle row), and IC (n = 140, bottom row). Only neurons with significant CSI values, determined by a nonparametric bootstrap procedure with 1,000 iterations (p < 0.05), were included in these datasets. (B) Distribution of CSI in A1 (top row), MGB (middle row), and IC (bottom row). Arrows point to the mean. Red bars highlight neurons with CSI significantly greater than 0, determined by a nonparametric bootstrap procedure with 1,000 iterations (p < 0.05). (C) Comparison of CSI among A1, MGB, and IC. Error bars represent standard error. ***p < 0.001, ANOVA with post-hoc test. (D–F) Population responses from the same datasets featured in panels (A-C) to the sound pair in the oddball paradigm (RN₁₀, TW₅₀): Adheres to the same conventions as panels (A-C). (G–I) Population responses from the same datasets featured in panels (A-C) to the sound pair in the oddball paradigm (RN₆₀, Con): Adheres to the same conventions as panels (A-C). (J–L) Population responses from the same datasets featured in panels (A-C) to the sound pair in the oddball paradigm (Asc, TW₃₀): Adheres to the same conventions as panels (A-C). Source data are available in Zenodo (see Data availability).

https://doi.org/10.1371/journal.pbio.3004046.s003

(EPS)

S4 Fig. Population neuronal responses to spectrally matched click-train pairs in the oddball paradigm recorded in IC, MGB, and A1.

(A) FFT magnitude spectra of the first stimulus pair used in (B–D). Black solid and orange dashed lines represent the two stimuli within the pair, which overlap because the stimuli were matched in long-term spectral magnitude. (B) Population PSTHs showing averaged neuronal responses to S-Pair1A (left) and S-Pair1B (right) as deviant (red) and standard (blue) stimuli in A1 (n = 466, top row), MGB (n = 348, middle row), and IC (n = 634, bottom row). (C) Distribution of CSI values calculated for neurons in A1, MGB, and IC within 5–200 ms from stimulus onset. Red bars indicate neurons with CSI significantly greater than 0, determined by a nonparametric bootstrap procedure with 1,000 iterations (p < 0.05). Arrows indicate the mean. (D) Comparison of CSI values among A1, MGB, and IC. Error bars represent SEM. ***p < 0.001, ANOVA with post hoc test. (E–H) Population responses to a second spectrally matched stimulus pair in the oddball paradigm, following the same conventions as panels (A–D). Source data are available in Zenodo (see Data availability).

https://doi.org/10.1371/journal.pbio.3004046.s004

(EPS)

S5 Fig. Effect of AC cooling on the response in A1 and oddball paradigm with complex tones in MGB.

(A) Neuronal responses in the AC before, during, and after AC cooling. Raster plots display the responses to the noise sound (duration = 100 ms) in each of the three conditions: before cooling (top row), during cooling (middle row), 2 hours after cooling (bottom row). (B–D) Population responses to two complex tones in the oddball paradigm, adhering to the same conventions as Fig 6, p = 0.53, paired t test. Source data are available in Zenodo (see Data availability).

https://doi.org/10.1371/journal.pbio.3004046.s005

(EPS)

S6 Fig. Histological reconstruction and laminar CSD profile of a Neuropixels penetration in A1.

(A) Histological reconstruction of the Neuropixels probe track in auditory cortex. The orange signal marks the probe track within AC, and the dashed line indicates the estimated trajectory of the probe. Scale bar, 1 mm. (B) Laminar LFP and current source density profiles evoked by noise stimulation from the same penetration. Left, LFP responses recorded along the probe from superficial to deep cortical depths. Middle, schematic of the 384-channel Neuropixels probe, with channels selected every 80 µm for CSD analysis. Right, CSD profile calculated from the selected LFP channels and smoothed for visualization. Sources are shown in blue and sinks in red. Horizontal dashed lines indicate approximate boundaries between cortical layers. Vertical dashed lines indicate noise onset. Source data are available in Zenodo (see Data availability).

https://doi.org/10.1371/journal.pbio.3004046.s006

(EPS)

S1 Audio. Audio files of click-train stimuli and spectrally matched control stimuli.

This file contains the 10 click trains used as auditory stimuli and two additional pairs of click trains with identical frequency-domain magnitude spectra but different time-domain structures. 001_Con.wav: click train with a constant inter-click interval (ICI) of 4 ms; 002_Ascend.wav: click train with the ICI progressively increasing from 2 to 6 ms; 003_Descend.wav: click train with the ICI progressively decreasing from 6 to 2 ms; 004_TW10.wav: click train with the ICI varying randomly between 3.6 and 4.4 ms (4 ± 10%); 005_TW30.wav: click train with the ICI varying randomly between 2.8 and 5.2 ms (4 ± 30%); 006_TW50.wav: click train with the ICI varying randomly between 2 and 6 ms (4 ± 50%); 007_TW70.wav: click train with the ICI varying randomly between 1.2 and 6.8 ms (4 ± 70%); 008_RN10.wav: click train with the ICI varying randomly within the range of 3.6–4.4 ms (4 ± 10%); 009_RN40.wav: click train with the ICI varying randomly within the range of 2.4–5.6 ms (4 ± 40%); 010_RN60.wav: RN60 click train with the ICI varying randomly within the range of 1.6–6.4 ms (4 ± 60%); 011_S-Pair1A.wav: stimulus A of spectrally matched click-train Pair 1, generated using a multi-sine modulation of the inter-click intervals; 012_S-Pair1B.wav: time-reversed counterpart of S-Pair1A, with the same frequency-domain magnitude spectrum but a different temporal structure; 013_S-Pair2A.wav: stimulus A of spectrally matched click-train Pair 2, generated using an exponentially varying inter-click interval pattern with added random noise; 014_S-Pair2B.wav: time-reversed counterpart of S-Pair2A, with the same frequency-domain magnitude spectrum but a different temporal structure.

https://doi.org/10.1371/journal.pbio.3004046.s008

(RAR)