Parallel mechanisms signal a hierarchy of sequence structure violations in the auditory cortex

  1. Sara Jamali
  2. Sophie Bagur
  3. Enora Bremont
  4. Timo Van Kerkoerle
  5. Stanislas Dehaene
  6. Brice Bathellier  Is a corresponding author
  1. Université Paris Cité, Institut Pasteur, AP-HP, Inserm, Fondation Pour l'Audition, Institut de l’Audition, IHU reConnect, France
  2. Université Paris Saclay, INSERM, CEA, Cognitive Neuroimaging Unit, NeuroSpin Center, France
  3. Collège de France, PSL University, France
6 figures and 1 table

Figures

Figure 1 with 2 supplements
Sequence violations are broadly signaled in the mouse auditory cortex.

(A) Sketch of the experimental setup and of the four different blocks of sound sequences. (B) Representative 1 × 1 mm two-photon imaging field-of-view with a magnification of the red rectangle. (C) Sample raw ΔF/F calcium traces (left, gray shadings=5 tone sequences) and trial-averaged activity during sequences (right black, raw ΔF/F; right red, temporally deconvolved ΔF/F=dΔF/F) for 4 neurons. (D) (left) Mean ± SEM population responses (dΔF/F) to rare or frequent AAAAB and AAAAA sequences. (right) Same for BBBBA and BBBBB sequences. (E) (left) dΔF/F population responses to the fifth tone averaged from 0 to 300 ms after tone offset for the AAAAX (left) and BBBBX (right) blocks. Mean + SEM & statistical tests: AAAAB rare vs common  0.36 ± 0.028% vs 0.25 ± 0.010% dΔF/F, p=5 × 10–14, AAAAB rare vs common 0.25 ± 0.022% vs  0.24 ± 0.008% dΔF/F, p=0.09; BBBBA rare vs common  0.62 ± 0.063% vs  0.34 ± 0.023% dΔF/F, p=8 × 10–18; BBBBB rare vs common 0.17 ± 0.023% vs 0.18 ± 0.008% dΔF/F, p=0.54; Mann-Whitney U test; n=30 and n=180 for common and rare repetitions. (F) Proportion of cells that respond significantly more to rare than common AAAAB (9 ± 2.33%), AAAAA (1.24 ± 0.15%), BBBBA (19.4 ± 3.49%), or BBBBB (1.09% ± 0.14%, mean ± SEM, n=12 sessions, significance in each cell assessed with Mann-Whitney U test p<0.01, n=30 and n=180 for common and rare repetitions). (G) Performance of a cross-validated classifier, trained on time-averaged responses of 90 ms bins, for predicting rare vs common sequences. p<0.01 for all points above 50% (shuffle test 100 repetitions). (H) Sketch of electrophysiological recordings. (I) Mean ± SEM of z-scored single unit activities averaged across all units. (J) Mean single unit activity in the time window indicated by lines in (I) (rare vs common BBBBA  0.04 ± 0.006 vs  0.03 ± 0.002, p=8 × 10–4; rare vs common BBBBB, 0.007 ± 0.004 vs  0.000 ± 0.001; p= 0.03; Mann-Whitney U test; n=30 for rare and n=200 for common repetitions). (K) Same as (F) for electrophysiological recordings (BBBBA, 6.90 ± 2.57%, and BBBBB, 5.05 ± 0.90%, Mann-Whitney U test for each cell, n=8 sessions). (L) Same as (G) for electrophysiological recordings. p<0.05 for all points above 50% (shuffle test 100 repetitions).

Figure 1—figure supplement 1
Responses of synapsin-based expression of GCAMP6s to the omitted sequences in the auditory cortex.

For the two-photon imaging recordings reported in Figure 1 (synapsin-based expression of GCAMP6s), mean population activity for omitted sequences (AAAA left and BBBB right) appearing as oddball sequences in blocks where AAAAA or BBBBB (red), AAAAB or BBBBA (pink), or AAAA (blue) is the common sequence. We observed no significant response to omissions in this dataset. (Wilcoxon ranksum tests; Respectively p=0.271; p=0.436; p=0.694; p=0.905; n=20 for rare and n=90 for common repetitions).

Figure 1—figure supplement 2
Probe localizations from Neuropixel recordings of the auditory cortex.

(A) Histological section showing the location of the probe (stained with diI, diO, or diD). (B) Reconstructed probe locations from five recordings in four mice.

Local violations selectively modulate the responses of neurons tuned to the violating sound.

(A) Sample cross-validated cluster with its d∆F/F response profile for the training (left) and test (right) sets (shading=SEM). Single-trial responses in test and training sets are shown below. (B) Same as (A) for a cluster that did not pass cross-validation. (C) AAAAX block responses of all cross-validated neuron clusters for the imaging dataset of Figure 1. The frame indicates clusters whose responses to AAAAX and BBBBX blocks are shown on the right. (D) Sample neuronal clusters for the electrophysiological recordings showing a local + global effect (cluster I and cluster II). (E.F) Responses of sample clusters with typical sound responses to both A and B tones and with a significant pure global effect (IV and V). All tests are Mann-Whitney U tests on the z-scored test set responses; n=10 rare and n=66 common sequence responses. Cluster I: offset (p=8 × 10–4) and onset (p=0.777) of rare vs common BBBBA, offset of rare vs common BBBBB (p=0.656). cluster IV: offset (p=0.038) and onset (p=0.913) of rare vs common BBBBB. cluster V: offset (p=0.383) and onset (p=0.907) of rare vs common BBBBA, offset (p=0.012) and onset (p=0.388) of rare vs common BBBBB.

Figure 3 with 3 supplements
Modulation of sound responses by unexpected local violations is only partially due to adaptation.

(A) Sketch of the experimental paradigm. (B) (left) Trial-averaged population responses (d∆F/F) to rare and common AAAAB or AAAAA sequences (right). Same for the BBBBA and BBBBB sequences. (C) Time-averaged (0–300 ms after 5th tone offset) population responses to the fifth tone, rare vs common. AAAAB: 0.29 ± 0.018% vs  0.18 ± 0.006% dΔF/F, p=2 × 10–4; AAAAA: 0.12 ± 0.018 vs  0.12 ± 0.007% dΔF/F, p=0.41; BBBBA:  0.22 ± 0.013% vs   0.17 ± 0.005% dΔF/F, p=6 × 10–3; BBBBB: 0.10 ± 0.005% vs  0.09 ± 0.002% dΔF/F, p=0.40; Mann-Whitney U test; n=5 for rare and n=40 for common sequences. (D) Proportion of sound responsive cells that are significantly (Mann-Whitney U-test, p<0.01) upward modulated at the end of rare sequences (see Materials and methods) for long (~30 s, right) and short (1.5, left) inter-sequence intervals (ISI). The number of sequence repeats was equalized across datasets (n=5 for rare and n=40 for common sequence). Long ISI (27 sessions), AAAAB: 10.61 ± 2.51% and AAAAA: 0.40 ± 0.26% BBBBA: 5.10 ± 1.40% and BBBBB: 1.44 ± 0.54%. Short ISI (12 sessions), AAAAB: 12.12 ± 3.51%, AAAAA: 0.60 ± 0.12%; BBBBB: 20.78 ± 2.04%, BBBBB: 1.16 ± 0.08%. (E) Same as (B) for z-scored electrophysiological recordings (for statistics see Figure 3—figure supplement 2). (F) Mean ± SEM population responses to rare (orange) and common (yellow) in classical stimulus-specific adaptation (SSA) paradigm using single tones instead of sound sequences. (*p<0.05, **p<0.01, ***p<0.001; Mann-Whitney U test; see Table 1 for detailed statistics.) (G) Mean ± SEM population responses to common BBBBA (yellow) and rare BBBBA (orange) in BBBBX blocks and to A tones played at the same time intervals as in the BBBBX blocks but excluding all B tones. Green: frequent. Brown: rare. Gray: blank sound probe. Statistics are computed on the time-averaged response, 0–500 ms after A tone onset: common BBBBA 0.08 ± 0.01% dΔF/F, rare BBBBA: 0.20 ± 0.01% dΔF/F, p=0.10, Mann-Whitney U test; n=6 for rare and n=43 for common sequences. A-only (frequent) 0.23 ± 0.01% dΔF/F, (rare)  0.27 ± 0.02; p=2 × 10–4; Mann-Whitney U test; n=5 for rare and n=42 for frequent repetitions.

Figure 3—figure supplement 1
Clustering of neuronal responses for the long inter-sequence interval, dataset 1.

(A) Plot of the response profile for the 18 clusters derived from the dataset presented in Figure 3B, for the local-global with an inter-sequence interval of ~30 s. Most clusters with specific responses to sound B (#9–18) display a larger response for rare than for common AAAAB sequences, except cluster #15. (B) For four clusters, all response profiles are magnified, and superimposed and the response profile for the BBBBX sequence is shown on the right.

Figure 3—figure supplement 2
The same neurons exhibit local + global violation responses for both short and long intersequence intervals in wakefulness but not under anesthesia (although it is present at the population level).

(A) Three examples of cluster of neuron responses were tracked across three different experiments. (first column) Local + global paradigm with a short inter-sequence interval (1.5 s), (second column) long inter-sequence interval (~30 s), and (third column) long inter-sequence interval (~30 s) in anesthetized animals. (B) Same as (A) electrophysiological recordings (for mean population responses see Figures 1I and 3E) (C) For electrophysiological recordings in long inter-sequence interval protocol (Figure 3E). (left) Mean population responses to the fifth A tone (within a 300 ms time window starting from the onset of the fifth tone) in BBBBA are compared between rare and common conditions mean rare A (in BBBBA, orange)=0.037 (± 0.006); mean common A (yellow)=0.016 (± 0.005); p=0.027; mean rare B (purple)=–0.010 (± 0.012); same for B tone in BBBBB, mean common B (blue)=0.009(± 0.003); p=0.95; Mann-Whitney U test; n=6 for rare and n=44 for common repetitions. (right) Proportion of cells across sessions that are significantly responding more to rare versus common sequences (local + global effect: 2.108% (± 0.759%); pure global effect: 0.759% (± 0.537%); Mann-Whitney U test, n=8 sessions).

Figure 3—figure supplement 3
The local + global effect is preserved across wakefulness and anesthesia.

(A) Responses of same neurons in the auditory cortex expressing GCAMP6s compared in three different experiments. (first column) Local + global paradigm with a short inter-sequence interval (1.5 s), (second column) long inter-sequence interval (~30 s), and (third column) long inter-sequence interval (~30 s) in anesthetized animals. (B) Same as (A) for suppressed neurons (C) Same as (A) for activated neurons. (D) Same as (A) for electrophysiological recordings (note that single unit activity (SUA) is not the same between anesthesia and awakening.).

Unexpected local violations produce a surprise response based on recently experienced sequence structure.

(A) (left) Heatmap of population responses of ordered single trials for two different blocks with short inter-sequence intervals (ISI) (arrows indicate rare stimuli). (right) Mean population responses to the fifth tone of each trial (0–300 ms from tone onset). (B) Same as A, for the long ~30 s ISI. (C) Time-averaged population responses to the fifth tone for common and rare sequences, comparing responses of the first and to the last trial of each condition, and measured for the first (left) and second repetition of each block. Statistics 1st block repetition: first rare: 1.466 ± 0.173%, last rare: 1.038 ± 0.117%, p=4 × 10–4; first common: 1.102 ± 0.151%, last common: 0.751 ± 0.095%, p=0.06 (first rare vs first common, p=0.003); Statistics 2ns block repetition: first rare: 1.351 ± 0.146%, last rare: 1.105 ± 0.196%, p=0.038; first common: 1.065 ± 0.177%; last common: 0.592% ± 0.089%; p=0.008 (first rare vs first common, p=0.010). Wilcoxon signed-rank test; n=12 sessions. (D) Same as (C) for the ~30 s ISI (Mean ± SEM): first rare: 0.724% ± 0.070%, last rare: 0.658 ± 0.077%, p=0.055; first common: 0.519 ± 0.044%, last common: 0.534 ± 0.055%, p=0.328 (first rare vs first common, p=0.003). Wilcoxon signed-rank test; n=27. (E) Short ISI - stimulus adaptation effect: 0.411 ± 0.128%; p=0.003; surprise: 0.325 ± 0.050%; p=4 × 10–4; surprise adaptation: 0.337 ± 0.105%, p=0.009; Wilcoxon signed-rank test; n=12. Long ISI; stimulus adaptation effect = 0.015 ± 0.066 %, p=0.648; surprise: 0.204 ± 0.072%; p=0.005; surprise adaptation: 0.065 ± 0.08%; p=0.107. Wilcoxon signed-rank test; n=27.

Vasoactive intestinal peptide positive (VIP) interneurons signal sequence termination and omission of expected local violations.

(A) Sketch of the experimental setup. (B) Mean ± SEM population responses to the rare and common AAAAB and AAAAA sequences. Statistics for the time-averaged fifth tone response: AAAAB rare: 0.003 ± 0.002% dΔF/F; AAAAB common: –0.003 ± 0.0007% dΔF/F; p=5 × 10–4; AAAAA rare 0.002 ± 0.002% dΔF/F, AAAAA common: 0.0008 ± 0.0008% dΔF/F, p=0.29, Mann-Whitney U test, n=30 for rare and n=180 for common repetitions. (C) Mean ± SEM population responses to AAAA sequence in the AAAAA (omission, red), AAAAB (omission, pink) or AAAA block (control, blue). Statistics comparing omission to control conditions: 0.003 ± 0.003% dΔF/F,  0.006 ± 0.002% dΔF/F, –0.0001% dΔF/F ± 0.0009%; omission of A in AAAAA, p=0.292; omission of B in AAAAB, p=0.008; Mann-Whitney U test; n=20 for rare and n=100 for common repetitions. (D) Same as (B) under isoflurane anesthesia. AAAAB rare: 0.010 ± 0.002% dΔF/F, AAAAB common 0.003 ± 0.0008% dΔF/F; p=0.0056, AAAAA rare 0.006 ± 0.001% dΔF/F, AAAAA common: 0.004 ± 0.0008% dΔF/F, p=0.123, Mann-Whitney U test; n=30 for rare and n=180 for common repetitions. (E) Performance of a fully cross-validated classifier for predicting the rare sequence against the common sequence. The classifier is trained on time-averaged responses of 160 ms bins. p<0.01 for all points above 50% (shuffle test 100 repetitions). (F) Cross-validated clusters of neurons. Mean ± SEM responses during wakefulness and under anesthesia.

Parvalbumin positive (PV) interneurons weakly signal pure global violations.

(A) Sketch of the experimental setup. (B) Mean ± SEM population responses (deconvolved calcium signals) to rare and common AAAAB and AAAAA sequences, AAAAB rare: 0.058 ± 0.009% dΔF/F, AAAAB common:  0.021 ± 0.003% dΔF/F, p=1 × 10–4; AAAAA rare: 0.012 ± 0.009%; AAAAA common: –0.007 ± 0.003% dΔF/F, p=0.025, Mann-Whitney U test; n=30 for rare and n=180 for common sequences. (C) Mean ± SEM population responses to AAAA sequence in the AAAAA (omission, red), AAAAB (omission, pink) or AAAA block (control, blue). Statistics comparing omission to control conditions: –0.009 ± 0.011% dΔF/F,  –0.008 ± 0.009% dΔF/F; –0.007 ± 0.004% dΔF/F, omission of A in AAAAA, p=0.6176; omission of B in AAAAB, p=0.523; Mann-Whitney U test; n=20 for rare and n=100 for common repetitions. (D) Same as (B) under isoflurane anesthesia/ AAAAB rare: 0.030 ± 0.003% dΔF/F, AAAAB common, 0.015 ± 0.001% dΔF/F, p=1× 10–4; AAAAA rare: 0.002 ± 0.003% dΔF/F, AAAAA common 0.002 ± 0.001% dΔF/F, p=0.535, Mann-Whitney U test, n=30 for rare and n=180 for common sequences. (E) Performance of a fully cross-validated classifier for predicting the rare sequence against the common sequence. The classifier is trained on time-averaged responses of 160 ms bins. p<0.01 for all points above 50% (shuffle test, 100 repetitions).

Tables

Table 1
Mean ± SEM population responses to rare (orange) and common (yellow) in classical stimulus-specific adaptation (SSA) paradigm in Figure 3.

p-values are from the statistical tests comparing between rare and common conditions (Mann-Whitney U test; n=10 for rare and n=47 for common repetitions).

Stimulus specific adaptation paradigm
Inter-stimulusintervals (second)Mean (+SEM) dAF/F rare (%)Mean (+SEM) dAF/F common (%)p-value
0.50.193 (=0.019)0.073 (+0.015)1×10–4
10.252 (+0.019)0.114 (+0.010)1×10–4
20.140 (+0.017)0.076 (+0.010)0.002
40.100 (+0.048)0.130 (+0.009)0.545
80.185 (+0.023)0.139 (+0.014)0.055
160.266 (+0.058)0.172 (+0.013)0.073

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Sara Jamali
  2. Sophie Bagur
  3. Enora Bremont
  4. Timo Van Kerkoerle
  5. Stanislas Dehaene
  6. Brice Bathellier
(2024)
Parallel mechanisms signal a hierarchy of sequence structure violations in the auditory cortex
eLife 13:RP102702.
https://doi.org/10.7554/eLife.102702.2