Neurons along the auditory pathway exhibit a hierarchical organization of prediction error

Abstract

Perception is characterized by a reciprocal exchange of predictions and prediction error signals between neural regions. However, the relationship between such sensory mismatch responses and hierarchical predictive processing has not yet been demonstrated at the neuronal level in the auditory pathway. We recorded single-neuron activity from different auditory centers in anaesthetized rats and awake mice while animals were played a sequence of sounds, designed to separate the responses due to prediction error from those due to adaptation effects. Here we report that prediction error is organized hierarchically along the central auditory pathway. These prediction error signals are detectable in subcortical regions and increase as the signals move towards auditory cortex, which in turn demonstrates a large-scale mismatch potential. Finally, the predictive activity of single auditory neurons underlies automatic deviance detection at subcortical levels of processing. These results demonstrate that prediction error is a fundamental component of singly auditory neuron responses.

Fig. 1

Experimental design. a Sketch of experimental setup. While stimulating with sequences of pure tones, isolated neurons were recorded from three auditory nuclei of anesthetized rats: IC, MGB and AC (colored). The schematic representation of the ascending auditory pathway information flow (orange) shows how lemniscal (green) and non-lemniscal (purple) subdivisions can be distinguished in the IC, MGB and AC. b Stimulation sequences. For each neuron, 10 tones of evenly-spaced frequencies were selected to construct the stimulation sequences. Each tone f_i (i = 1…10) lying inside the neuron’s receptive field could be presented in two experimental conditions (deviant and standard, in separated oddball sequences, left column), and two control conditions (cascade and many-standards, right column) for adaptation effects. Note that ascending and descending deviant tones will be compared to the control ascending or control descending sequences, respectively. They will also be compared to the many-standards sequence for both types of deviants (see Methods). c Decomposition of neuronal mismatch responses (DEV−STD) to the oddball sequence using either one of the control conditions. Under the assumption of predictive coding, CTR−STD (if positive) represents repetition suppression, and DEV−CTR (if positive) represents prediction error. d Two hypothetical scenarios according to two possible competing mechanisms accounting for the neuronal mismatch: SSA (top) and predictive coding (bottom). For SSA, there is response suppression to the standard (blue bars), which progressively increases from lower order to higher order. In addition, due to suppression of the deviant relative to control, the prediction error is increasingly negative (blue and orange bars) as one progresses to higher-order regions. For predictive coding, repetition suppression of the standard (blue bars) increases from lower to higher-order regions. Unlike SSA, responses to deviants are higher than controls, especially in higher-order regions, leading to a positive prediction error (orange bars)

Fig. 2

Prediction error in representative examples of neuronal responses in anaesthetized rat. a Examples of lemniscal neuronal responses in each recorded auditory station (columns). The first row contains schematics of the lemniscal subdivisions (green) within each nuclei. The second row shows the frequency-response area (representation of neuronal sensitivity to different frequency-intensity combinations) of representative lemniscal neurons from each nucleus. Ten grey dots within each frequency-response area represent the ten tones (f_i) selected to build the experimental sequences (see Methods). The third row displays the measured responses of the particular neuron to each f_i tone (baseline-corrected spike counts, averaged within 0–180 ms after tone onset) for all conditions tested. Note that measured conditions tend to overlap in the subcortical stations (IC_L and MGB_L), and only start differentiating from each other once auditory information reaches the cortex (AC_L). The fourth row contains sample peri-stimulus histograms comparing the neuronal responses to each condition tested for an indicated f_i tone. A thick horizontal line represents stimulus duration. A small inset within the upper right corner of each panel features the isolated spike (mean ± SEM) of that single neuron. b Examples of non-lemniscal neuronal responses in each recorded auditory nuclei, organized as in a. The first row highlights non-lemniscal divisions in purple. In the second row, note frequency-response areas tend to be more broadly tuned, as compared to lemniscal neurons. In the third row, responses to deviant conditions tend to relatively increase and distance themselves from their corresponding controls as information ascends in the auditory pathway. Also note that responses to last standards are feeble or even completely missing across all non-lemniscal stations (IC_NL, MGB_NL and AC_NL). In the last row, the strong influence of the experimental condition over the neuronal response to the same tone can be clearly appreciated in the three nuclei

Fig. 3

Prediction error at population level for each station in anaesthetized rat. Distribution of normalized responses and related indices of neuronal mismatch (iMM), repetition suppression (iRS) and prediction error (iPE). a Each grey dot in these scatter plots represents the three normalized responses of a single neuron to the same tone played as deviant (DEV), as standard (STD) and as control (CTR). Indexes result from the difference between two of these normalized responses, represented in the axes surrounding the scatter plots, where the dotted black lines marks the absence of difference between conditions (index = 0). Solid black lines represent the mean of each index, corresponding their intersection to the center of gravity of the distribution of responses in the normalized space. Note how, while the intersection for lemniscal subcortical stations (IC_L and MGB_L) is skewed towards CTR, in their non-lemniscal counterparts (IC_NL and MGB_NL) as well as all over the cortex (AC_L and AC_NL) the center of gravity of the distribution shifts closer and closer to DEV as it moves up in the auditory pathway, increasing the iPE as auditory information reaches higher-order stations. b Histograms represent distributions within stations of the three indexes for each neuronal response. Solid black lines indicate medians. The noticeable overall tendency of the median indexes to shift towards more positive values, from IC through MGB to AC, and from lemniscal to non-lemniscal divisions, unveils a hierarchy of processing in the auditory pathway

Fig. 4

Emergence of prediction error along the auditory hierarchy. a Median normalized responses (lines indicate SEM) to the deviant, standard and control within each station. b Median indices of prediction error (orange) and repetition suppression (cyan), represented with respect to the baseline set by the control. Thereby, iPE is upwards-positive while iRS is downwards-positive. Each median index corresponds to differences between normalized responses in a. Asterisks denote statistical significance of iPE against zero median (*p = 0.05, **p = 0.01, ***p = 0.001, see Table 1). c Linear model fitted for the iPE using SPL and direction (ascending / descending) as predictors. Error bars denote mean and SEM for each SPL and direction. Note the model predicts greater iPE values for ascending conditions at low intensities. d Same as in b, but only representing ascending conditions at intensities equal or lower than 40 dB SPL

Fig. 5

Correlation of iPE and prediction error potential (PE-LFP). a Population grand-averages for different response measures, computed for each lemniscal station (in columns, represented in first row highlighted in green). The second row shows the average LFP across all tested tones and single neurons from each station for different conditions. The third row displays the average firing rate profiles for each station as normalized spike-density functions. The fourth row contains the prediction error potentials (PE-LFP, black trace), which is the difference wave of the deviant and the control LFP. Along PE-LFP, the time course of the average iPE is plotted in orange (mean ± SEM, asterisks indicating significant iPE for the corresponding time window; Wilcoxon signed-rank test for 12 comparisons, corrected for FDR = 0.1). Next row shows an instantaneous p value (white trace) of the corresponding PE-LFP (paired t test against equal means, corrected for FDR = 0.1, critical threshold for significance set at 0.05 represented as a horizontal bar). Thick black bars of the grey panel mark time intervals for which the average PE-LFP is significant. Note that only AC_L shows a significant prediction error signal. b Same as in a but computed for each non-lemniscal station (highlighted in purple in the first row). Note in the last row significant PE-LFPs appear in all three stations (IC_NL, MGB_NL and AC_NL), and prominently in AC_NL. Note also how highest iPE values are concurrent with the strongest PE-LFPs in time and location (auditory cortex, both AC_L and AC_NL)

Fig. 6

Prediction error in representative examples of neuronal responses in awake mouse. a Examples of lemniscal multiunit activity recorded in two auditory nuclei (columns). The first row contains schematics of the lemniscal subdivisions (green) within each station. The second row shows a frequency-response area of each nuclei. Ten grey dots within those frequency-response area represent the ten tones (f_i) selected to build the experimental sequences (see Methods). The third row displays the measured responses to each fi tone (baseline-corrected spike counts, averaged within 0–180 ms after tone onset) for all conditions tested. The fourth row contains sample peri-stimulus histagrams comparing the neuronal responses to each condition tested for an indicated f_i tone. Stimulus duration is represented by a thick horizontal line. b Examples of multiunit activity recorded in non-lemniscal divisions (first row, colored purple) of each auditory nuclei, organized as in a

Fig. 7

Population results for awake mouse. a Summary of population results for each recorded station (columns). The first row displays lemniscal (green) and non-lemniscal (purple) subdivisions of two recorded auditory nuclei of the mouse brain. The second row contains scatter plots featuring normalized responses of each multiunit recording to the same tone played as DEV, STD and CTR (grey dots) and the mean population values of each index (solid black bars). The third row contains the average LFP across all tested tones from each station for different conditions. Thick black bars at the bottom of the panels mark the time intervals were the difference between the deviant LFP and the control LFP is significant, thereby producing a prediction error potential. b Median normalized responses (bar indicate interquartile range) to the deviant, standard and control within each station. c Median indices of prediction error (orange) and repetition suppression (cyan), represented with respect to the baseline set by the control. Asterisks denote statistical significance of iPE against zero median (*p = 0.05, **p = 0.01, ***p = 0.001). Note the overall similarities with results in the anaesthetized rat (Figs. 3–5), confirming a hierarchical generation of prediction error also in awake preparations